**Renewable Energies for Sustainable Development**

Printed Edition of the Special Issue Published in *Sustainability* M. Dolores Esteban, José-Santos López-Gutiérrez and Vicente Negro Edited by

www.mdpi.com/journal/sustainability

## **Renewable Energies for Sustainable Development**

## **Renewable Energies for Sustainable Development**

Editors

**M. Dolores Esteban Jos ´e-Santos L ´opez-Guti´errez Vicente Negro**

MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade • Manchester • Tokyo • Cluj • Tianjin

*Editors* M. Dolores Esteban Universidad Europea de Madrid Spain Jose-Santos L ´ opez-Guti ´ errez ´ Universidad Politecnica de ´ Madrid (UPM) Spain

Vicente Negro Universidad Politecnica de ´ Madrid (UPM) Spain

*Editorial Office* MDPI St. Alban-Anlage 66 4052 Basel, Switzerland

This is a reprint of articles from the Special Issue published online in the open access journal *Sustainability* (ISSN 2071-1050) (available at: www.mdpi.com/journal/sustainability/special issues/ Renewable Energies Sustainable Development).

For citation purposes, cite each article independently as indicated on the article page online and as indicated below:

LastName, A.A.; LastName, B.B.; LastName, C.C. Article Title. *Journal Name* **Year**, *Volume Number*, Page Range.

**ISBN 978-3-0365-1382-9 (Hbk) ISBN 978-3-0365-1381-2 (PDF)**

© 2021 by the authors. Articles in this book are Open Access and distributed under the Creative Commons Attribution (CC BY) license, which allows users to download, copy and build upon published articles, as long as the author and publisher are properly credited, which ensures maximum dissemination and a wider impact of our publications.

The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commons license CC BY-NC-ND.

## **Contents**



## **About the Editors**

#### **Mar´ıa Dolores Esteban**

Mar´ıa Dolores Esteban (Ph.D.) is an associate professor of coastal eEngineering at the Universidad Politecnica de Madrid and of coastal engineering and renewable energies at the ´ Universidad Europea. She has more than 15 years of experience in the fields of maritime engineering and renewable energies as an engineer at a private company and as a lecturer and researcher at universities. She has published more than 50 technical papers in SCI indexed journals and has presented at numerous international conferences. Her main fields of research are coastal and maritime engineering; harbor and ports; beaches; loads actions and effects in maritime structures; and renewable energies such as offshore wind, waves, and currents. She is very active in R&D projects and is an important asset to her research group the Environmental, Coast, and Ocean Research Laboratory (ECOREL) at the Universidad Politecnica de Madrid. ´

#### **Jos ´e-Santos L ´opez-Guti´errez**

Jose-Santos L ´ opez-Guti ´ errez (Ph.D.) is an associate professor of coastal engineering at the ´ Universidad Politecnica de Madrid. He has more than 25 years of experience in the fields of maritime ´ engineering and renewable energies as a lecturer and researcher at the university. He has supervised five Ph.D. dissertations. He has more than 60 technical papers published in SCI indexed journals and has presented at more than 60 national and international conferences. He is responsible for the research line on marine energies within the group of researchers of the Environmental, Coast, and Ocean Research Laboratory (ECOREL) at Universidad Politecnica de Madrid. Apart from research ´ fields related to marine energies, such as load and effect actions on marine structures for offshore wind installations, and wave and current exploitation, other research fields in which he is involved include coastal and maritime engineering, ports, and beaches.

#### **Vicente Negro**

Vicente Negro (Ph.D.) is a full professor of coastal engineering and the head of the Environmental, Coast, and Ocean Research Laboratory (ECOREL) at Universidad Politecnica de ´ Madrid. He has supervised more than 10 Ph.D. dissertations. He has published more than 70 technical papers in SCI indexed journals and more than 40 scientific and dissemination articles at the national level. He has more than 40 contributions at national and international conferences with peer arbitration. He is an expert in marine sciences for the Program of the Directorate General XII of Research in Brussels, Drafting Commission of the ROM (Puertos del Estado) and is an editorial member of the ROM0.3/91, Environmental Actions I: Waves; ROM 0.0/2002, General procedure and calculation bases in the maritime and port works project; ROM 1.0/09, Recommendations for the design of Breakwaters; and ROM 2.0/11, Recommendations for the project and execution of Docking and Mooring Works.

## **Preface to "Renewable Energies for Sustainable Development"**

In 2019, we started working together with the editorial team of the journal *Sustainability* (MDPI editorial) as guest editors of a Special Issue related to renewable energies. This opportunity was welcomed with great enthusiasm by the Guest Editorial Team. As soon as possible, we started working on the project, with the great support of the main editor, who guided us in publishing a Special Issue that exceeded our initial expectations. This Special Issue was entitled "Renewable Energies for Sustainable Development". It focused on the seventh Sustainable Development Goal (SDG), which is to ensure access to affordable, reliable, sustainable, and modern energy for all. In the current scenario in which climate change dominates our lives and in which we all need to combat and drastically reduce the emission of greenhouse gases, renewable energies play key roles as present and future energy sources. Renewable energies vary across a wide range, and therefore, there are related studies for each type of energy. The Special Issue called for studies integrating the latest research innovations and knowledge focused on all types of renewable energy: onshore and offshore wind, photovoltaic, solar, biomass, geothermal, waves, tides, hydro, etc. Authors were invited submit review and research papers focused on energy resource estimation, all types of TRL converters, civil infrastructure, electrical connection, environmental studies, licensing and development of facilities, construction, operation and maintenance, mechanical and structural analysis, new materials for these facilities, etc. Analyses of a combination of several renewable energies as well as storage systems to progress the development of these sustainable energies were welcomed. In the end, this Special Issue published 20 papers. The papers are of very good quality and are titled as follows: "An Intelligent Approach to Active and Reactive Power Control in a Grid-Connected Solar Photovoltaic System"; "Optimizing Wave Overtopping Energy Converters by ANN Modelling: Evaluating the Overtopping Rate Forecasting as the First Step"; "Emergy and Sustainability Ternary Diagrams of Energy Systems: Application to Solar Updraft Tower"; "Effect of Spatial and Temporal Resolution Data on Design and Power Capture of a Heaving Point Absorber"; "Hybrid Wind–PV Frequency Control Strategy under Variable Weather Conditions in Isolated Power Systems"; "From NEDC to WLTP: Effect on the Energy Consumption, NEV Credits, and Subsidies Policies of PHEV in the Chinese Market"; "A New Robust Energy Management and Control Strategy for a Hybrid Microgrid System Based on Green Energy"; "Revisiting the Relation between Renewable Electricity and Economic Growth: A Renewable–Growth Hypothesis"; "On the Design of an Integrated System for Wave Energy Conversion Purpose with the Reaction Mass on Board"; "Probabilistic Assessment of Hybrid Wind-PV Hosting Capacity in Distribution Systems"; "Residential Energy-Related CO<sup>2</sup> Emissions in China's Less Developed Regions: A Case Study of Jiangxi"; "Feasibility and Cost Analysis of Photovoltaic-Biomass Hybrid Energy System in Off-Grid Areas of Bangladesh"; "Accurate Sizing of Residential Stand-Alone Photovoltaic Systems Considering System Reliability"; "Strategic Planning of Offshore Wind Farms in Greece"; "A Compact Pigeon-Inspired Optimization for Maximum Short-Term Generation Mode in Cascade Hydroelectric Power Station"; "Evaluation of the Operating Efficiency of a Hybrid Wind–Hydro Powerplant"; "A Residential Load Scheduling with the Integration of On-Site PV and Energy Storage Systems in Micro-Grid"; "Thermo-Poroelastic Analysis of Induced Seismicity at the Basel Enhanced Geothermal System"; "Renewable Energy for Sustainable Growth and Development: An Evaluation of Law and Policy of Bangladesh"; and "Assessment on Global Urban Photovoltaic Carrying Capacity and Adjustment of Photovoltaic Spatial Planning". It was a pleasure for the Guest Editorial Team to put all of these interesting manuscripts together. We thank the authors of these articles who allowed this Special Issue to be of such a high quality.

> **M. Dolores Esteban, Jos´e-Santos L ´opez-Guti´errez, Vicente Negro** *Editors*

*Article*

## **A Compact Pigeon-Inspired Optimization for Maximum Short-Term Generation Mode in Cascade Hydroelectric Power Station**

#### **Ai-Qing Tian <sup>1</sup> , Shu-Chuan Chu 1,2, Jeng-Shyang Pan 1,\*, Huanqing Cui <sup>1</sup> and Wei-Min Zheng <sup>1</sup>**


Received: 19 November 2019; Accepted: 4 January 2020; Published: 21 January 2020

**Abstract:** Pigeon-inspired optimization (PIO) is a new type of intelligent algorithm. It is proposed that the algorithm simulates the movement of pigeons going home. In this paper, a new pigeon herding algorithm called compact pigeon-inspired optimization (CPIO) is proposed. The challenging task for multiple algorithms is not only combining operations, but also constraining existing devices. The proposed algorithm aims to solve complex scientific and industrial problems with many data packets, including the use of classical optimization problems and the ability to find optimal solutions in many solution spaces with limited hardware resources. A real-valued prototype vector performs probability and statistical calculations, and then generates optimal candidate solutions for CPIO optimization algorithms. The CPIO algorithm was used to evaluate a variety of continuous multi-model functions and the largest model of hydropower short-term generation. The experimental results show that the proposed algorithm is a more effective way to produce competitive results in the case of limited memory devices.

**Keywords:** compact pigeon-inspired optimization; maximum short-term generation; swarm intelligence; hydroelectric power station

### **1. Introduction**

The metaheuristic algorithm [1] has emerged as a very promising tool to solve complex optimization problems. Original pigeon-inspired optimization (OPIO) is a new type of metaheuristic search algorithm [2]. The algorithm simulates the behavior of pigeons going home. Preliminary studies indicate that it is a very promising optimization algorithm and can outperform excellent existing algorithms [3]. OPIO exploits a population of pigeons as candidate solutions by setting boundaries and optimizing the problem by moving the candidate solutions to approach the best solutions based on a given measure of quality. The general steps of the algorithm are described below.

OPIO can solve continuous solution space problems. In addition, many versions of OPIO in the literature are proposed to solve the problem of continuous and discrete solution spaces in recent years. An improved Gaussian pigeon inspired optimization algorithm preserves the diversity of early evolution to avoid premature convergence. The entire algorithm shows excellent performance in global optimization and is effective for solving multimodal and non-convex problems with higher dimensions. Multi-objective pigeon-inspired optimization (MPIO) is used for multi-objective optimization in designing the parameters of brushless direct current motors [4]. The multimodal multi-objective pigeon-inspired optimization algorithm (MMPIO) was proposed to figure out the multimodal multi-objective optimization problems [5,6].

With the continuous development of metaheuristic algorithms, intelligent group optimization has become an emerging technology to solve many engineering problems. Metaheuristic algorithms perform well on wireless sensor networks [7,8]. Since 2000, many scholars have designed many ant colony optimization algorithms, particle swarm optimization algorithms (PSO) [9], gray wolf optimization algorithms (GWO) [10,11], bat inspired algorithms (BA) [12,13], flower pollination algorithms (FPA) [14,15], cat swarm optimization (CSO) [16,17], differential evolution algorithm (DE) [18,19], quasi-affine transformation evolution algorithms (QUATRE) [20,21], genetic algorithms (GA) [22,23], etc. Based on the simulation of the above-mentioned functional mechanisms through an in-depth study, it is easy to observe that the adaptive phenomenon can widely exist in nature. Among them, OPIO was proposed by Duan and other scholars in 2014 [24]. It is a new intelligent optimization algorithm based on the homing behavior of pigeons. While it has not been long since its introduction, this algorithm has been used in model improvement and application, obtaining many research results. Because the algorithm has good adaptability and high calculation accuracy, various optimizations have been carried out in the fields of unmanned aerial vehicle (UAV) formation [25], control parameter optimization [26], and image processing [27].

A country's development and social progress are inseparable from its demand for energy [28,29]. Electric energy is a very flexible form of energy that is increasingly obtained from the sun. Electrical energy can be converted into heat, chemical energy, and mechanical energy. Its power is also convenient to use, easy to control, safe, and clean. More importantly, most of the development of today's society relies on the development of science and technology. The main ways to generate electricity are thermal, wind, hydropower, and nuclear power generation. Hydropower is a renewable energy source that can continuously generate and deliver electricity. The main advantage of hydropower generation is that it can eliminate fuel cost. The cost of operating a hydropower station is not affected by rising fossil fuel prices such as oil, natural gas, and coal. Hydroelectric power stations do not require fuel. The economic life of a hydroelectric power station is longer than those of fuel-fired power plants. In addition, hydropower stations are mostly operated automatically and normally. In addition, such power plants have low operating costs [30,31].

The short-term optimal dispatching of cascade hydropower stations refers to the maximum value of the objective function in the case of meeting the various constraints of the cascade hydropower stations on one or several days [32]. In general, this mainly refers to the following three mathematical models: The model with the shortest power generation, the short-term water consumption minimum model [33], and the short-term peak power maximum model [34]. These three mathematical models have the same properties, i.e., under certain constraints, the nonlinear multi-stage optimization problem is obtained. This paper only analyzes the model with the maximization of the short-term power generation.

In this paper, we combine the compact technique with the pigeon-inspired optimization to propose the compact pigeon-inspired optimization algorithm. The proposed CPIO not only improves the time efficiency but also reduces the hardware memory. The algorithm proposed in this paper has very good spatial complexity. The algorithm only has one particle to update, and the original algorithm uses the population to update. After expanding our work, our goal is to solve the problem of reducing memory usage and parameter selection in optimizing the short-term power generation of cascade hydropower stations. The reasons for expanding our work include adding sample probability functions that must control the perturbation vector and comparing them with other compact algorithms in this article. The probability function operates to solve the optimal value of the compact pigeon-inspired optimization (CPIO) algorithm, and uses a real-valued prototype vector to generate each candidate solution. The algorithm has been tested on multiple continuous multi-modal functions as well as the short-term power generation of cascade hydropower stations [35–37].

#### **2. Related Work**

#### *2.1. Principle of Electricity Generation of Cascade Hydropower Station*

The hydropower process is actually a process of energy conversion. By constructing a hydraulic structure on a natural river, concentrating the water head, and then guiding the high water to the low-position turbine through the water channel, the water energy is converted into rotational mechanical energy, and the generator coaxial with the turbine is used to generate electricity. It is pivotal to note the conversion from water energy to electricity. The electricity generated by the generator is sent to the user through the transmission line to form the entire process of for generating electricity, as listed in Figure 1. The water body in the high-altitude reservoir has a large potential energy. When the water body flows into the downstream of the hydropower station through the hydraulic pipe installed in the hydropower station, the water flow drives the runner of the water turbine to rotate, so that the hydrodynamic energy is converted into the rotating mechanical energy. The turbine drives the coaxial generator rotor to cut the magnetic lines of force, and generates an induced electromotive force on the stator winding of the generator. When the stator winding is connected to the external circuit, the generator supplies power to the outside. This way, the selected mechanical energy of the turbine is converted into electrical energy by the generator.

**Figure 1.** The principle of hydroelectricity.

Water energy resources are potential energy and kinetic energy existing in rivers and are a part of renewable resources, but the reserves of water energy are related to factors such as river flow, evaporation, precipitation, etc. Rivers vary greatly from region to region and climate varies. There is also a large difference in the amount of water energy resources in the region. Due to the current technical conditions, the water volume and the drop of the river part will not be utilized, and the mutual transition between energy also has a certain loss. Therefore, the technically developable hydropower resources are usually lower than their theoretical reserves. Taking the amount of technology developable resources as the basis, the economically available hydropower resources obtained by considering factors such as transmission distance, cost, and flooding loss are smaller than the technically exploitable amount.

Hydropower station reservoirs are generally divided into two categories, one is conventional scheduling and the other is optimized scheduling. Conventional scheduling is a commonly used scheduling method. It can also be called traditional scheduling. It is the most basic adjustment method. It only bases on historical data, no longer considers any other relevant factors, and then uses classical hydraulics and runoff regulation. The scheduling diagram and scheduling rules are used to guide the operation of the reservoir, and the water level of the reservoir is calculated and then expressed as the objective function, that is, the most basic scheduling method is used to ensure the operation of the hydropower station. These schedules are drawn based on past hydrological data and tasks when the

reservoir is at different water storage levels, that is, the reservoir scheduling rules when the reservoir is in different states.

The optimal dispatching model of the hydropower station reservoir is based on the optimal theory to establish a mathematical model based on the actual conditions of the hydropower station reservoir, and then using modern computer technology to find the optimal scheduling method that meets the scheduling principle in the process of establishing the optimal scheduling model of the hydropower station. In addition, there are many factors to be considered, including the connection between water and electricity as well as numerous constraints on cascade water inventory, This will minimize pollution in terms of ecological environment, and maximize profits and social benefits in terms of economic benefits. In order to meet the above requirements to the maximum extent and to minimize the water abandonment of hydropower stations, this paper establishes a short-term power generation model suitable for the optimal operation of cascade hydropower stations.

#### *2.2. Maximum Short-Term Generation Model*

With the completion and operation of a large number of hydropower stations, the effective solution of large-scale reservoir group optimization scheduling models has become an urgent problem to be solved. The reservoir optimization problem proposes a higher solution quality and running speed for the meta heuristic algorithm. Therefore, this paper proposes a new algorithm called CPIO. This aspect of research not only improves the speed of the operation, but also ensures that the quality of the solution is not worse than the original algorithm and can suppress the premature phenomenon.

Under the given inner diameter flow of the control period, the objective function of the long-term optimal scheduling model of the cascade hydropower station group is defined as: The maximum amount of cascade power generation under the condition of ensuring the output of the cascade is considered during the control period. Under the premise of satisfying the actual situation, this paper selects the maximum benefit of cascade power generation as one of the objective functions. At the same time, it is necessary in the medium and long-term optimization scheduling, it is necessary to consider the output of the period with the least output during the year as much as possible. Medium and long-term optimized dispatching provides the largest possible uniform and reliable output for the power grid, giving full play to the capacity benefits of hydropower generation which can replace thermal power.

$$\mathbf{E} = \max \mathbf{F} = \max \sum\_{i=1}^{N} \sum\_{j=1}^{T} A \times q \times \Delta h \times \Delta t \tag{1}$$

In the formula, *A*, *q*, ∆*h*, ∆*t* represents the output coefficient, the outflow rate, the upstream and downstream water level difference respectively. In addition, unit time *E* is the maximum annual power generation benefit of the cascade hydropower station, *N* is the total number of cascade hydropower stations, and *T* is within one year. Calculate the total number of time slots (*T* = 12).

In the process of optimizing the power generation of cascade hydropower stations, it is necessary to understand the water reservoir data in order to first calculate the reservoir upstream capacity and the outflow flow value, and then calculate the downstream water level value based on the outflow flow value.

$$V\_{i,j} = \frac{c\_1 + h\_{i,j}^1 - c\_2}{c\_3 - c\_4} \* (c\_5 - c\_6) \* 10^8 \tag{2}$$

In the formula *c*1, *c*2, *c*3, *c*4, *c*5, *c*<sup>6</sup> are a set of variable constants. This set of constants has different values according to the water level in each interval, and *h* 1 *i*,*j* is the upstream water level value of the *j*-th time period of the *i*-th hydropower station.

$$q\_{i,j} = h\_j^o - \frac{V\_{i,j+1} - V\_{i,j}}{t\_j \* 3600} \tag{3}$$

In the Equation (3), *qi*,*<sup>j</sup>* is the outflow of the *j*-th time period of the *i*-th hydropower station, *h o j* is the initial flow of the *j*-th hydropower station, and *t<sup>j</sup>* is the number of hours of the *j*-th time period.

$$h\_{i,j}^2 = \frac{c\_7 + (q\_{i,j} - c\_8)}{c\_9 - c\_{10}} \* (c\_{11} - c\_{12}) \tag{4}$$

In this Equation (4), *c*7, *c*8, *c*9, *c*10, *c*11, *c*<sup>12</sup> are a set of variable constants. This set of constants has different values according to the water level in each interval, and *h* 2 *i*,*j* is the downstream water level value of the *j*-th time period of the *i*-th hydropower station.

$$
\Delta h\_{\rm i} = \frac{h\_{\rm i,j}^1 + h\_{\rm i+1,j}^1}{2} - h\_{\rm i,j}^2 \tag{5}
$$

In the Equation (5), ∆*h<sup>i</sup>* is the upstream and downstream water level difference of the *i*-th hydropower station. When the upstream and downstream water level difference is obtained, since the upstream water level is greatly affected, the average value of the upstream water level is selected for calculation in this paper.

Let us introduce the constraints of the objective function:

$$h\_{i, \min} \le h\_{i, j} \le h\_{i, \max} \tag{6}$$

The level of the water level needs to be limited between *hmin* and *hmax*.

$$q\_{i,min} \le q\_{i,j} \le q\_{i,max} \tag{7}$$

The outflow of the reservoir must fluctuate between *qmin* and *qmax*. In order to ensure the stable operation of the power generation of the entire cascade hydropower station, the output of the power has to be relatively stable, so the outflow of the reservoir cannot be lower than the minimum flow. In addition, in order to stabilize the life of the turbine and generator, the outflow of the reservoir cannot be higher than the maximum flow.

$$V\_{i,min} \le V\_{i,j} \le V\_{i,max} \tag{8}$$

The capacity of the reservoir should fluctuate between the *Vmin* and *Vmax*. In order to ensure that the reservoir will continue to work under special circumstances, the reservoir's capacity cannot be lower than the originally set value to ensure the safe operation and that the downstream organisms are safe, so the capacity cannot be higher than the maximum reservoir capacity.

#### **3. Pigeon-Inspired Optimization**

Without prejudice, the minimization problem of the objective function *f*(*x*) is discussed in this paper, where *x* is the vector that defines the *n* design variables in the domain *D* in the decision space.

The pigeon-inspired optimization is a meta-heuristic algorithm that is inspired by the behavior of the pigeons returning home and is widely used in most continuous or discrete optimization problems. This article mainly introduces continuity problems. Referring to extensive literature reviews, a group of pigeons move in decision space *D* according to the update rules to find the optimal value when looking for the solution of the problem. More formally, in order to gain the satisfactory value of the objective function *f*(*x*), the population of the pigeons is randomly sprinkled in the previously set search space. The objective function judges the equivalent quality of solution based on the position information of each pigeon. At any stage *t*, the *i*-th pigeon has its own position vector *x t k* and velocity vector *v t k* . For each pigeon, the best solution is the value of the objective function. The best position of the position where the pigeon has passed will be stored. The global optimal solution is continuously

updated. To transition from the *t* step to the *t* + 1 step, a more competitive solution will be taken, and each particle is perturbed according to the following formula:

$$
\omega\_k^{t+1} = e^{-R \ast t} \ast V\_k^t + \phi\_1 \ast (\mathbf{x}\_{\S^{best}} - \mathbf{x}\_k) \tag{9}
$$

and:

$$\mathbf{x}\_{k}^{t+1} = \boldsymbol{\phi}\_{2} \* \mathbf{x}\_{k}^{t} + \boldsymbol{\phi}\_{3} \* \boldsymbol{v}\_{k}^{t+1} \tag{10}$$

As the formula above suggests, *x t k* refers to the current position of the *k*-th pigeon, and *xgbest* is the best position ever found in the entire herd, and the vector *v t k* is a perturbation vector, namely velocity. Finally, *φ*<sup>1</sup> is a variable constant, is a variable amount limited to 0–1, and *φ*2, *φ*<sup>3</sup> are two weight factors can be constants or variables. This stage belongs to the map and the compass operator. When the pigeon approaches the destination, the dependence on the sun and the magnetic object is reduced, and then the landmark operator is entered.

$$\boldsymbol{\mathfrak{x}}\_{center}^{t} = \frac{\sum\_{k=1}^{N^t} \boldsymbol{\mathfrak{x}}\_{k}^{t} \* F(\boldsymbol{\mathfrak{x}}\_{k}^{t})}{\mathbf{N}^{t} \* \sum\_{k=1}^{N^t} F(\boldsymbol{\mathfrak{x}}\_{k}^{t})} \tag{11}$$

From here on, the landmark operator is entered. In this operator, the pigeons continue to iterate according to the pigeons or landmarks of the roads understood by the population. In the above formula, the purpose of this operation is to find out the pigeons with a high fitness value in the flock. This pigeon is then considered to be the pigeon that knows the road, and the pigeons are iterated according to the pigeon. *N<sup>t</sup>* is the population number at the *t*-th iteration, and *F*(*X t k* ) is the fitness function value of the *k*-th pigeon position.

$$N^t = \frac{N^{t-1}}{2} \tag{12}$$

The significance of this operation is to halve the pigeons and discard the pigeons that do not have the way to know, to prevent such pigeons from misleading the population into local optimum.

$$\mathbf{x}\_{i}^{t+1} = \mathbf{x}\_{i}^{t} + \boldsymbol{\phi}\_{4} \* \left(\mathbf{x}\_{\text{center}}^{t} - \mathbf{x}\_{i}^{t}\right) \tag{13}$$

In the formula, *φ*<sup>4</sup> is a variable constant that value is a randomly generated value from (0, 1). In this operation, all pigeons that do not know the road will be iterated according to the pigeons that know the road.

$$F(\mathbf{x}\_i^t) = \frac{1}{Fitness(\mathbf{x}\_k^t) + \chi} \quad \text{For minimization problem} \tag{14}$$

$$F(\mathbf{x}\_i^t) = Fitness(\mathbf{x}\_k^t) \quad \text{For maximization problem} \tag{15}$$

For maximizing the problem, OPIO uses Equation (14) to calculate the value of *F*(*x t k* ) to find the pigeon with the ability to identify the function. For the minimization problem, OPIO uses Equation (15) to calculate the value of *F*(*x t k* ) to find the pigeon with the function of identifying. In Equation (14), *χ* is a non-zero constant whose purpose is to prevent the denominator from being zero.

#### **4. Compact Pigeon-Inspired Optimization**

The compact approach replicates the operation of the population-based algorithm by building the probability of a total solution. The optimal process encodes the probability representation of the actual population as a virtual counterpart. Compact pigeon-inspired optimization is a model built on a pigeon-inspired optimization-based framework. In the OPIO algorithm, the concept and design of the CPIO algorithm will be explored in more detail.

The purpose of CPIO is to simulate the operation of OPIO underlying overall algorithm in a smaller version of memory variable memory. By constructing a distributed data structure, the actual solution of the OPIO is transformed into a compact algorithm, the perturbation vector. The PV vector is a probabilistic model for the solution of the population.

$$PV^t = [\mu^t, \delta^t] \tag{16}$$

In the formula, *µ*, *δ* are two parameters of the standard deviation and the average of the vector *PV*, and t is the current number of iterations. The value of *µ*, *δ* is limited to probability density functions (*PDF*) [38] and is changed within [–1, 1]. The magnitude of the PDF is normalized by keeping the area to 1, because by obtaining approximately sufficient in well it is the uniform distribution with a full shape.

The initialization of the virtual population is performed as follows. For each design variable *µ* = 0 and *δ* = *λ* , where *λ* is a large constant (*λ* = 10). This value is initialized to initially obtain a normal distribution of truncated wide shapes.

The sampling mechanism of the design variable *x t k* associated with the generic candidate solution *x* in *PV* is not a simple process and requires extensive interpretation. For each design variable indexed by *k*, a truncated Gaussian *PDF* with the mean *µ* and standard deviation *δ* is associated, The *PDF* is described by the following formula:

$$PDF = \frac{e^{-\frac{(x-\mu[k])^2}{2\times\sigma[k]}} \times \sqrt{\frac{2}{\pi}}}{\sigma[k] \times (erf(\frac{\mu[k]+1}{\sqrt{2}\times\sigma[k]} - erf(\frac{\mu[k]-1}{\sqrt{2}\times\sigma[k]})))} \tag{17}$$

*PDF* is the probability distribution function of *PV*, and a truncated Gaussian PDF-related *µ*, and *δ* are formulated. A new candidate solution is generated by iteratively biasing towards a promising region of the optimal solution. Every component of the probability vector may be acquired by learning the previous generations. *er f* is the error function established by [39]. *PDF* corresponds to the cumulative distribution function (*CDF*) by constructing a Chebyshev polynomial [39], and the upper domain of the *CDF* is randomly changed between 0 and 1. *CDF* can be described as a real-valued random variable *x* with a probability distribution, and the value that can be obtained can be less than or equal to *x t k* .

The relationship between *CDF* and *PDF* can be defined as *CDF* = R 1 0 PDF(*x*)*dx*, and *PV* operations can sample the design variable *x t k* by randomly generating values within the range of (0, 1).

In the iterative process of the compact algorithm, in order to find a better individual, a function that can be compared by two parameters is proposed in this paper. The two variable pigeon parameters are two sample individuals of the *PV* operation. The vector represented by the winner is the value of the fitness function. This value is higher than other virtual members, and the vector represented by the loser is that the individual fitness value is lower than the fitness evaluation standard. Two variables with return values, the winner and the loser are obtained from the calculation of the objective function, and a new candidate solution is generated to compare with the original global optimal solution to generate new winners and losers. For updating *PV* operations, *µ* and *σ* can be considered for updates according to the rules below. If the mean value of *µ* is 1, Then the update rule becomes *µ <sup>t</sup>* and *σ t* for each of its elements *µ <sup>t</sup>*+<sup>1</sup> and *σ t*+1 [40] as described in the following:

$$
\mu\_i^{t+1} = \mu\_i^t + \frac{winner\_i - laser\_i}{N},
\tag{18}
$$

where *N* is the virtual population size and the value of *δ* is described below. The update rule for each element is given in the formula below.

$$
\sigma\_i^{t+1} = \sqrt{(\sigma\_i^t)^2 + (\mu^{t+1})^2 + \frac{winner\_i^2 - lower\_i^2}{N}} \tag{19}
$$

Mathematical details about construction Equations (18) and (19) have been given. The persistent and non-persistent structures of rcGA have been tested and can be seen in [41]. Seeing the virtual population size *N* as a parameter of a compression algorithm is not a true population-based algorithm. The virtual population size, in the real-valued compression algorithm, is an algorithm that depends on the convergence speed.

In general, a probabilistic model for compact OPIO is hired to represent all of the set of solutions for the pigeon group, neither storing location information nor storing speed information; however, storing newly generated candidate solutions. Therefore, the limited storage space is required to achieve the algorithm requirements which saves a lot of time and hardware resources for the cascade hydropower station to optimize the short-term power generation model.

CPIO uses a perturbation vector *PV* that has the same structure as the one shown in Equation (16), at the beginning of the optimization algorithm, just like the process described in Equation (17). The *PV* initialization is designed as (∀*k*, *µ*[*k*] = 0, *σ*[*k*] = *λ*, *λ* = 10) and the variables of each design are limited to one continuous space [−1, 1], and in addition, the position *x* and the velocity *v* are randomly initialized within a certain range.

Update velocity vector and position vector by slightly revised pigeon-inspired algorithm:

$$
\omega\_k^{t+1} = \omega\_1 \ast e^{-R \ast t} \ast V\_k^t + \omega\_2 \ast (\mathbf{x}\_{gbest} - \mathbf{x}\_k) \tag{20}
$$

and:

$$\mathbf{x}\_{k}^{t+1} = \mathfrak{f}\_{1} \* \mathfrak{x}\_{k}^{t} + \mathfrak{f}\_{2} \* v\_{k}^{t+1} \tag{21}$$

*ω*<sup>1</sup> is an inertia weight, *ω*<sup>2</sup> is a random variable between 0 and 1, and *ξ*<sup>1</sup> and *ξ*<sup>2</sup> are weighting factors that control the position update of the pigeon.

It can be seen that the equation updating of speed (Equation (20)) and position (Equation (21)) is similar to the OPIO algorithm. In the original version, pigeon *k* was closely related to pigeon group *N*. In the compact version, there was no real population, but the relevance of a virtual population pigeon to the virtual population was not that great. It is easy to see that compact OPIO is just a pigeon that uses the update formula to update it, so updating it once produces a solution that saves a lot of memory.

In the landmark operator entering the second stage of CPIO, the original algorithm uses the Equation (11) to determine the pigeon with the function of identifying the function based on the fitness solution of each pigeon position, so that it becomes the center point and continues to update. Since the CPIO has only one particle to update, it is not suitable when selecting a pigeon with a path function. In this paper, a center point suitable for CPIO is proposed. By setting a virtual center position point, the guiding pigeon is updated. When the virtual center position is established, it is based on the historical fitness value of the pigeon. The number selected is also based on the size of the virtual population.

$$\mathbf{x}\_{center}^{t+1} = \frac{\sum\_{i=l-N+1}^{N} F(\mathbf{x}\_k^t)}{N} \tag{22}$$

*l* is the number of iterations until now, *N* is the number of virtual populations. According to the Equation (22), the historical virtual center points of the pigeons can be selected, and it is known that they continue to iterate.

$$\mathbf{x}\_{k}^{t+1} = \mathbf{x}\_{k}^{t} + \omega\_{3} \* (\mathbf{x}\_{center}^{t} - \mathbf{x}\_{k}^{t}) \tag{23}$$

It is easy to see that CPIO saves a lot of memory space, so this approach can be applied to other variations of OPIO.

### **5. Numerical Results**

The test results of the CPIO that have been tested by 29 test functions, and these test functions come from [42]. Each test function has a very detailed introduction in Tables 1 and 2. Among these groups of questions, they have different search range and different expressions.

In Equations (20)–(23) and Algorithm 1, the parameters of the CPIO proposed herein are: *N* = 120, *R* = 0.2. The values of these parameters are referred to [43] and have a slight change. More specifically, in order to make CPIO work better, we modeled the virtual population size proposed by OPIO. In this article, CPIO is compared to the OPIO. In all test functions, CPIO is run 30 times and averaged. Take the minimum value of CPIO in all test functions.

**Algorithm 1** Compact pigeon-inspired optimization (CPIO) pseudo-code.

1: Map and compass factors *R*, The maximum number of iterations in the first stage *MaxDT*<sup>1</sup> and

The maximum number of iterations in the second stage *MaxDT*2, dimension is *dim*;

2: **for** *k* = 1 to *dim* **do**

3: // *PV* operation initialization; *N* is the total number of pigeons

4: initialize *µ*[*k*] = 0

5: initialize *σ*[*k*] = 10

6: **end for**


10: Generate the local best solution *x<sup>k</sup>* by *PV*

11: **for** *t* = 1 to *MaxDt*<sup>1</sup> **do**

12: *x t <sup>k</sup>* = Generate from *PV* operation

13: // Update position and velocity

$$\begin{array}{ll} \text{14:} & v\_{k}^{t+1} = \omega\_{1} \ast e^{-R\ast t} \ast V\_{k}^{t} + \omega\_{2} \ast (\mathfrak{x}\_{g\text{best}} - \mathfrak{x}\_{k})\\ \text{15:} & v^{t+1} - \mathfrak{x}\_{\ast} \ast \mathfrak{x}^{t} + \mathfrak{x}\_{\ast} \ast \mathfrak{x}^{t+1} \end{array}$$

*k*

15: *x t*+1 *<sup>k</sup>* = *ξ*<sup>1</sup> ∗ *x t <sup>k</sup>* + *ξ*<sup>2</sup> ∗ *v*

16: // Best Selection

17: [*winner*, *loser*]=compete(*x t*+1 *k* , *xgbest*)

*t*


```
20: µ
              [k] = µ
                      [k] + winner[k]−loser[k]
              [k] = q
```

$$\begin{array}{ll} \text{21:} & \sigma^{t+1}[k] = \mu^t[k] + \frac{N}{\sqrt{(\sigma^t[k])^2 + (\mu^t[k])^2 - (\mu^{t+1}[k])^2 + \frac{winner\_k^2 - 1oser\_k^2}{N}}}\\ \text{22:} & \mathbf{end} \text{ for} \end{array}$$

23: **end for**

24: **for** *t* = *MaxDt*<sup>1</sup> + 1 to *MaxDt*<sup>1</sup> + *MaxDt*<sup>2</sup> **do**

25: // Enter the second stage

26: // Select the virtual center pigeon from the historical best points )

27: *x t*+1 *<sup>k</sup>* = *x t <sup>k</sup>* + *ω*<sup>3</sup> ∗ (*x t center* − *x t k*

28: // Best Selection

29: [*winner*, *loser*]=compete(*x t*+1 *k* , *xgbest*)

30: // Update *PV* operation

31: **for** *i* = 1 : *dim* **do**

32: *µ t*+1 [*k*] = *µ t* [*k*] + *winner*[*k*]−*loser*[*k*]

*N* 33: *σ t*+1 [*k*] = <sup>q</sup> (*σ t* [*k*])<sup>2</sup> + (*µ<sup>t</sup>* [*k*])<sup>2</sup> − (*µt*+<sup>1</sup> [*k*])<sup>2</sup> + *winner*<sup>2</sup> *k*−*loser*<sup>2</sup> *k N*

34: **end for**

```
35: end for
```


**Table 1.** Details of 29 test functions.

When initializing the two algorithms CPIO and original pigeon-inspired optimization (OPIO) , the map and compass factor *R* are set to 0.2, and the result is to compare CPIO and OPIO. The quality of solution and the number of runs of CPIO and OPIO optimal solutions are compared as below described. The CPIO and OPIO data results are the average of 30 runs. All algorithms operate 500 times, including 300 in the first phase and 200 in the second phase.

In Table 3, CPIO performs better than OPIO in many test functions, and most of the values perform well. In terms of the time cost comparison, it is easy to see that CPIO time spent is much better than PIO, especially in several of them, and the time spent is more than a hundred times more.

According to the comparison of the two algorithms, it can be concluded that the running time of CPIO is much lower than that of the original algorithm. This is because the number of population used in the process of iteration is different. In the new algorithm, it uses an example to keep iterating, constantly adjusting the probability distribution according to the path that has been iterated, and the greater the possibility of generating particles where the function values are superior. However, this method also has a big problem, since in the search process of a single particle, randomness is often large, and it is thus easy to fall into the local optimal. It is also relatively simple to achieve the optimal, in the case of small dimension settings, the advantages of the algorithm are not obvious. Because of this characteristic of the new algorithm, it is easy to save time and reduce the time complexity of the algorithm.


**Table 2.** Details of 29 test functions.

Figure 2 shows the convergence trend of CPIO and OPIO. Best score obtained so far refers to the optimal value obtained by the algorithm during the iteration process. While the convergence speed of OPIO and the algebra needed to achieve optimal are small, the optimal value of CPIO is better or nearly equal to the value of OPIO. Here, CPIO uses one particle for updating and iteration, while OPIO uses the entire population for optimization. CPIO is far less than OPIO search capability, but CPIO can save a lot of memory and time to find excellence.

**Figure 2.** Compact pigeon-inspired optimization (CPIO) and original pigeon-inspired optimization (OPIO) performance in test functions. (**a**) Ackley; (**b**) Crossit; (**c**) Drop; (**d**) Griewank.


**Table 3.** Comparison, evaluation and speed of quality performance between CPIO and OPIO.

Among the four selected functions, Figure 3 shows the time trend of the four functions running 30 times. In general, the time spent by the CPIO and PIO algorithms does not change much, but the two algorithms compare. It is easy to see that CPIO runs much faster than the PIO.

Table 4 shows the comparison of CPIO and PIO mentioned above in the memory variables, which makes it very convenient to implement the calculation algorithm. The number of variables of the two algorithms of CPIO and PIO proposed in this paper is calculated by the equation used in the computational optimization. In Table 4, it is easy to see that in the same computing situation, CPIO uses less memory than PIO. For example, during an iteration, CPIO uses an iteration Equations (16)–(23); the formula for PIO update iteration is Equations (9)–(13).

**Table 4.** The space complexity of the two algorithms.


**Figure 3.** CPIO and OPIO performance in test functions. (**a**) Ackley; (**b**) Crossit; (**c**) Drop; (**d**) Griewank.

As can be seen from Table 4, the actual population size of the PIO is *N*, but the actual population size in the CPIO is 1, and the virtual population number is *N*. In the case where the number of iterations *l* and the running time *t* are the same, the memory usage of the variables of OPIO and CPIO is iterated by 4 × *t* × *N* and 8 × *t*, respectively. Here, it is seen that the memory occupancy of the PIO is larger than the memory usage of the CPIO.

In Figure 4, the consequence of the presented algorithm and the else three meta-heuristics are shown. According to Table 5, the trend and optimal value of CPIO are fundamentally better than the other three algorithms, and have a superior performance. Table 5 shows the comparison of CPIO, OPIO and other algorithms, such as CPSO and PSO algorithms. Among the four meta-heuristic algorithms, the performance is as follows in 29 test functions. In the process of algorithm simulation, as part of the images are not so obvious, four relatively obvious images are extracted for display.

**Figure 4.** *Cont*.

**Figure 4.** CPIO and three other meta-heuristic algorithms for testing function performance. (**a**) Ackley; (**b**) Crossit; (**c**) Drop; (**d**) Griewank.


**Table 5.** The optimal value and time cost of the four algorithms.

#### **6. Experiments of Short-Term Power Generation Model for Cascade Hydropower Stations**

Wanjiazhai Water Conservancy Project: The Wanjiazhai Water Conservancy Project is located in the canyon of the Tuoketuo to Longkou section of the Yellow River in the north of the Yellow River. It is the first of the eight cascades planned for the development of the middle reaches of the Yellow River. and also the Shanxi Yellow River Diversion Project. The starting point of the project the left bank is affiliated to the Pianguan County of Shanxi Province, and the right bank is subordinate to the Zhungeer Banner of Inner Mongolia Autonomous Region. The dam site controls a drainage area of 395,000 square kilometers, with a total storage capacity of 896 million cubic meters and a storage capacity of 445 million cubic meters. It has comprehensive benefits such as water supply, power generation, flood control and anti-icing.

Longkou Hydropower Station is located at the junction of two provinces, Hequ County, Shanxi Province and Zhungeer Banner, Inner Mongolia. It is 25.6 km from the upstream Wanjiazhai Water Control Project and 70 km from the downstream Tianqiao Hydropower Station. It is the regional center of energy and chemical bases in Shanxi Province and Inner Mongolia Autonomous Region, and controls the drainage area of 397, 406 square kilometers.

Table 6 shows the monthly inflow values of the two cascade hydropower stations in the wet years, the flat water years and the dry years. ASP is Annual scheduling period.


**Table 6.** Cascade hydropower station monthly water supply.

The short-term power generation model of cascade hydropower stations has been introduced above. Figure 5 showcases the main flow of the algorithm. In this paper, the three periods of the two cascade hydropower stations are scheduled and modeled by Equations (1)–(8) and the sum of the power generation of the two cascade hydropower stations is the largest. As shown in Figure 6, at any stage, CPIO has the largest scheduling capacity for the two cascade hydropower stations, and the total power generation is also relatively huge. CPIO dispatched the two cascade hydropower stations. The final result has the power generation at 3.968 <sup>×</sup> <sup>10</sup><sup>17</sup> KWH in the high flow year, and the total power generation at 3.108 <sup>×</sup> <sup>10</sup><sup>17</sup> KWH in the year of the median water. The power generation at 2.396 <sup>×</sup> <sup>10</sup><sup>17</sup> KWH in the low year.

**Figure 5.** The main process of optimizing hydropower station.

**Figure 6.** *Cont*.

**Figure 6.** Comparison of four meta-heuristic algorithms in cascade hydropower stations: (**a**) Wet water years schedule; (**b**) flat water years schedule; and (**c**) dry water years schedule;

## **7. Conclusions**

A novel optimization approach called compact pigeon-inspired optimization (CPIO) is proposed. The proposed CPIO was tested on 29 classical test functions to demonstrate the usefulness of the proposed optimization method. A compact method is successfully used in the pigeon-inspired optimization algorithm to reduce the usage of the memory size. The proposed CPIO was also applied to cascade hydroelectric power generation. Simulation results show the CPIO may reach better results compared with some existing algorithms for the cascade hydroelectric power station.

**Author Contributions:** Conceptualization, J.-S.P. and W.-M.Z.; Data curation, A.-Q.T. and H.C.; Formal analysis, A.-Q.T., S.-C.C., J.-S.P., H.C., and W.-M.Z.; Investigation, A.-Q.T.; Methodology, A.-Q.T., S.-C.C., J.-S.P., H.C., and W.-M.Z.; Software, A.-Q.T.; Validation, J.-S.P.; Visualization, A.-Q.T. and S.-C.C.; Writing—original draft, A.-Q.T.; and Writing—review and editing, S.-C.C. and J.-S.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** We wish to confirm that there are no known conflicts of interest and there has been no significant financial support for this work that could have influenced its outcome. We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed.

## **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **A New Robust Energy Management and Control Strategy for a Hybrid Microgrid System Based on Green Energy**

## **Bilal Naji Alhasnawi 1,\*, Basil H. Jasim <sup>1</sup> and M. Dolores Esteban <sup>2</sup>**


Received: 18 June 2020; Accepted: 14 July 2020; Published: 16 July 2020

**Abstract:** The recent few years have seen renewable energy becoming immensely popular. Renewable energy generation capacity has risen in both standalone and grid-connected systems. The chief reason is the ability to produce clean energy, which is both environmentally friendly and cost effective. This paper presents a new control algorithm along with a flexible energy management system to minimize the cost of operating a hybrid microgrid. The microgrid comprises fuel cells, photovoltaic cells, super capacitors, and other energy storage systems. There are three stages in the control system: an energy management system, supervisory control, and local control. The energy management system allows the control system to create an optimal day-ahead power flow schedule between the hybrid microgrid components, loads, batteries, and the electrical grid by using inputs from economic analysis. The discrepancy between the scheduled power and the real power delivered by the hybrid microgrid is adjusted for by the supervisory control stage. Additionally, this paper provides a design for the local control system to manage local power, DC voltage, and current in the hybrid microgrid. The operation strategy of energy storage systems is proposed to solve the power changes from photovoltaics and houses load fluctuations locally, instead of reflecting those disturbances to the utility grid. Furthermore, the energy storage systems energy management scheme will help to achieve the peak reduction of the houses' daily electrical load demand. Also, the control of the studied hybrid microgrid is designed as a method to improve hybrid microgrid resilience and incorporate renewable power generation and storage into the grid. The simulation results verified the effectiveness and feasibility of the introduced strategy and the capability of proposed controller for a hybrid microgrid operating in different modes. The results showed that (1) energy management and energy interchange were effective and contributed to cost reductions, CO<sup>2</sup> mitigation, and reduction of primary energy consumption, and (2) the newly developed energy management system proved to provide more robust and high performance control than conventional energy management systems. Also, the results demonstrate the effectiveness of the proposed robust model for microgrid energy management.

**Keywords:** inverters; converters; distributed generators; utility grid; hierarchical control

### **1. Introduction**

During the previous decade, distributed energy systems have assumed supreme importance in the electrical power infrastructure because they lead to a reduction in greenhouse gas emissions and produce quality power with increased efficiency and service reliability. Deploying distributed energy resources facilitates a modification in the way energy is transmitted through the electrical power grid and provides flexibility to the consumer concerning energy use.

To accomplish an adequately set up distributed energy ecosystem, there is a necessity to convert the power system into smaller distributed-energy systems [1]. These distributed systems may incorporate several power sources like photovoltaic systems, wind turbines, or fuel cells. These fuel-cell-based energy sources are essential since power generation based on such a system may be set up near or at the consumer premises. Recent research [2] indicates that such resources help decrease household emissions, operating costs, and primary energy required to produce electricity [3].

The integration of various distributed energy resources, distributed loads, and energy storage systems with a renewable energy ecosystem is called a microgrid [4]. In the recent past, microgrids have been explored with immense interest as power systems that offer resilience and help set up a smart active electrical grid. Additionally, it has the potential to enhance system safety, reliability, efficiency, and boost connectivity with renewable energy sources [5]. The microgrid may not be connected to the utility grid since it can be used as a standalone system supplying energy to controllable loads using several distributed energy sources. The introduction of controllable loads and distributed energy systems causes several challenges for energy management systems. The primary role of an energy management system is to calculate the adequate energy transmission from the microgrid and the primary energy network independently and meet load requirements on an hourly basis. The literature contains several proposals regarding microgrid energy management systems using varying algorithms and different microgrids [6]. The rest of this paper is organized as follows. Section 2 presents a description of the related works. Section 3 presents a description of the proposed system described in this paper. Section 4 presents the mathematical model of the distributed hybrid generation system, Section 5 presents the control methods for photovoltaics and fuel cell system, Section 6 presents the supervisory control method of the proposed system. Section 7 presents the energy management systems for houses and Section 8 presents the results of the proposed system. Finally, Section 9 concludes the paper.

#### **2. Related Work**

This section contains a review of the pertinent research regarding energy management strategies related to the topics connected to this work. In Lujano-Rojas et al. [7], the photovoltaic system with battery backup is operated in the grid-connected and islanded modes of operation. In this paper, separate control algorithms are implemented for the inverter, battery, and photovoltaic array for maximum utilization of available sources for meeting the energy requirement of the consumer. In Delgado and Navarro [8], the authors propose a linear programming-based algorithm to manage microgrid power. This algorithm facilitates adequate generator use for controllable and uncontrollable electrical loads. One example of an optimization challenge is the adequate use of diesel generators while maintaining operational and economic limitations due to energy demand and supply for each microgrid component (storage system, load, and generator). In the study conducted by Helal et al. [9], the power management system for an AC/DC hybrid microgrid was explored in an isolated area that uses a photovoltaic setup for desalination. The optimization algorithm suggested uses mixed-integer nonlinear programming, and the objective function works towards reducing the daily operational cost. In Correa et al. [10], the authors formulated a virtual power plant (VPP)-based power management system. The microgrid investigated consists of a solar panel and an energy storage system working in tandem. Linear programming methods are used to design the techniques required to decrease operating expenditure. Renewable energy is accounted for in the energy models like the one in Colombia and is typically based on hydropower resources. In Dufo and Agustín [11], the authors present a generic algorithm-based control system to optimally manage energy in a hybrid system. The system comprises renewable energy sources such as hydropower, wind turbines, and photovoltaics in conjunction with AC generators, fuel cells, and an electrolyzer. Energy management optimization leads to decreased operating costs. This facilitates the surplus produced by the renewable source to be stored in batteries and use of the electrolyzer to produce hydrogen gas. Loads that cannot be powered using a renewable energy source may be operated using a fuel cell, or by using energy from

the batteries. In Das et al. [12], the authors investigate the impact of supplementing a standalone hybrid microgrid based on photoelectric sources with gas turbines and an internal combustion engine. This multi-objective generic technique was implemented to achieve a system optimized considering energy cost and efficiency. Load evaluation was conducted using thermal and electrical measurements. Combined with heating or cooling, all the systems studied were able to meet the electrical demand. In Luna et al. [13], the author introduced a real-time power management system. The three cases were investigated with full, incomplete, and accurate forecasts. The optimization model was tested on connected and isolated microgrids, and there was a significant imbalance between loads and generations. In Abedini et al. [14], the author used the particle swarm algorithm with Gaussian boom and applied the energy management algorithm to a microgrid consisting of wind turbines, diesel generators, or photovoltaics. This study suggested a reduction in both capital and operational (fuel) costs. In Marzband et al. [15], energy management in the microgrid was tested using the artificial bee colony (ABC) algorithm. Provided the transient nature of wind and solar energy resources, an approach is required to evaluate the economic distribution of energy generating units in the microgrid. The results indicated a 30% reduction in cost. The non-dispatchable generation and load uncertainty are managed using Markov chain and neural networks. In Rouholamini and Mohammadian [16], the study proposes appropriate power management in the case of grid-tied hybrid energy generation systems, which include wind turbines, photovoltaics, electrolysis, and fuel cells. Electricity is traded with the local network using the real-time pricing determined over a 24 h time slot using simulation results. This energy management optimization case used the interior search technique. In Almada et al. [17], the research proposed a centralized microgrid energy management system in both grid-connected and island mode. In the island mode, the fuel cells start supplying energy only if the battery state of charge reaches less than 80%, whereas in the grid-connected mode, a 60% state of charge is required to ensure reliability. In Merabet et al. [18], the authors recommended an energy management system to handle a hybrid microgrid system having battery power, photovoltaics, and wind turbines. The data acquisition and control systems operate in real-time. A set of rules governs the power management platform, and those enhance microgrid performance by monitoring and manage load, power generation, and storage devices. Farzin et al. [19] proposed an energy management system to handle an isolated microgrid. An isolated failure event was considered as being a natural probability distribution failure scenario in the utility grid. The objective was to decrease operating expenditure, which considers the expenditure for running wind turbines, small turbines, batteries, and electrical load. In Battistelli et al. [20], the authors suggested a management system for a hybrid DC to AC microgrid that promises economic electrical transmission despite the skepticism about the economics of renewable energy. The system has on-demand load control, and factors in generators, battery charge/discharge thresholds (electric or thermal automobiles), and controllable loads.

The method presented has the following demerits:


To address these challenges, this paper proposes a newly designed and flexible energy management and control method applicable to a hybrid microgrid that uses green energy. The proposed technique has specific advantages which are listed below:

1. To the best of the authors knowledge, this is the first research paper that proposes an energy control system for homes, which regulates the frequency and restoration of the voltage for an islanded or a grid-connected microgrid, which is based on the use of a completely nonlinear framework, without considering uncertainties and parametric disturbances.


#### **3. Proposed System Description**

The hybrid microgrid evaluated comprises photovoltaic cells and fuel cells, along with super capacitors and batteries that are used for energy storage, an electrolyzer cell, and AC loads. These systems, when connected, create a small hybrid microgrid, which is connected to a low-voltage AC microgrid for distribution or islanded mode. Figure 1 depicts the configuration of such a hybrid system for the distribution microgrid case. The DC bus is used to interconnect the fuel cells to the boost converter. The battery is connected to the charging regulator using a two-way DC to DC converter, while the super capacitor has a direct connection to the DC bus. The one-way DC to DC converter is used to interconnect the FC and the DC buses, which maintains system stability in spite of varying electrical loads. Batteries are connected to the DC bus using a two-way DC to DC converter.

**Figure 1.** The proposed hybrid microgrid system structure.

#### **4. Distributed Hybrid Energy Generation System**

#### *4.1. Photovoltaic Cell Modeling*

Figure 2 shows an equivalent circuit based on the diode of the solar cell, which can be represented as a parallel resistor, current source, diode, and serial resistance. The current and voltage properties of solar cells are described by the standard mathematical equation (Equation (1)) [21]:

$$I = I\_{\rm ph,cell} - \underbrace{I\_{o,cell}[\exp\left(\left(q(V + IR\_{\rm \\_s}(s, cell))/akT\right) - 1\right)}\_{\rm L,cell} - \frac{V + IR\_{\rm s,cell}}{R\_{p,cell}}\tag{1}$$

where *Io*,*cell* is reversed leakage current, *Iph*,*cell* is photocurrent (A) of photovoltaic, *k* is Boltzmann's constant (1.38 <sup>×</sup> <sup>10</sup>−<sup>23</sup> J/K), *<sup>q</sup>* is the electron charge (1.602 <sup>×</sup> <sup>10</sup>−<sup>19</sup> C), and *<sup>R</sup>p*,*cell* is parallel resistance (Ω). *Rs*,*cell* is the series resistance (Ω) and *T* is the diode temperature.

**Figure 2.** The circuit of photovoltaic cell based on single diode mode.

The solar cell model is ideal if the parallel and series resistance of the solar cell are not taken into account. Figure 3 illustrates the ideal electrical current and voltage curves from Equation (1) [22,23].

**Figure 3.** Typical photovoltaic current and voltage curves.

#### *4.2. Modeling of Photovoltaic Module*

As mentioned earlier, the photovoltaic modules are made up of connected photovoltaic cells in parallel form. Therefore, the standard mathematical equation is derived from Equation (2) and the characteristic description of the I–V photovoltaic unit [23] is:

$$I = I\_{ph,cell} - I\_O \left[ \exp\left(\frac{V + IR\_S}{a\,Vt}\right) - 1\right] - \frac{V + IR\_S}{R\_p} \tag{2}$$

where *Iph*,*cell* is the photocurrent (A) of the photovoltaic, *I<sup>o</sup>* is photovoltaic reverse leakage current, *V<sup>t</sup>* is photovoltaic thermal voltage, *R<sup>p</sup>* is parallel resistance, *R<sup>s</sup>* is photovoltaic series resistance, Equation (2) generates the current and voltage curve as illustrated in Figure 4.

**Figure 4.** (**a**) A typical current–voltage curve of the photovoltaic (PV), (**b**) current–voltage curves of a photovoltaic at different temperature levels and constant irradiation, (**c**) Power–voltage curves of a photovoltaic at constant irradiation and different temperature levels.

The current of a photovoltaic (*IPH*) based on the amount of the photovoltaic irradiance falling on the photovoltaic and the photovoltaic cell temperature corresponds to Equation (3) [24]:

$$I\text{-}h\text{\textquotedblleft} = \frac{G}{Gn}(I\_{pl\text{\textquotedblleft}n} + Ki\Delta T) \tag{3}$$

where *Iph*;*<sup>n</sup>* is the photocurrent, *G* is the photovoltaic irradiance measured in W/m<sup>2</sup> , *G<sup>n</sup>* is the nominal irradiance (1000 W/m<sup>2</sup> ), and *Ki* is the temperature coefficient.

*Voc* is determined by the Equations (4) and (5):

$$V\_{\alpha\varepsilon} = V\_{\alpha\varepsilon;\mathfrak{n}} + \mathcal{K}\_{\overline{\mathfrak{v}}} \Delta T \tag{4}$$

where *K<sup>v</sup>* is coefficient of temperature and *Voc*;*<sup>n</sup>* is the open circuit voltage under the nominal conditions.

$$I\_o = \frac{I\_{sc;n} + K\_i \Delta T}{\exp\left(\frac{V\_{oc;n} + K\_v \Delta T}{a' V\_t}\right) - 1} \tag{5}$$

where *Isc*;*<sup>n</sup>* is the short-circuit current under the nominal conditions [25].

In this paper, the electrical parameters of the SPR-305E-WHT-D solar photovoltaic are extracted and used to simulate this modular model. These parameters are listed in Table 1.


**Table 1.** Electrical parameters of the SPR-305E-WHT-D photovoltaic.

#### *4.3. Battery Storage System*

Battery storage systems store additional energy produced by renewable energy systems. However, if there is insufficient energy from the renewable energy generation system, the battery will be discharged to meet the load demand. It consists of a 4800 (Ah) Li-ion battery unit with a rated voltage of 96 VDC. Equations (6)–(9) govern the battery charge/discharge process [26]:

$$E\_{\rm disch}^{\rm Ni-MH} = Eo - k \frac{\mathcal{Q}}{\mathcal{Q} - it} i^\* - k \frac{\mathcal{Q}}{\mathcal{Q} - it} it + e^t \tag{6}$$

$$E\_{ch}^{\text{Ni}-\text{MH}} = Eo - k \frac{\mathcal{Q}}{|it| - 0.1 \mathcal{Q}} i^\* - k \frac{\mathcal{Q}}{\mathcal{Q} - it} it + e^t \tag{7}$$

The equations that govern the discharging process of Li-ion batteries are:

$$E\_{\rm disch}^{\rm Li-Ion} = Eo - k \frac{\mathcal{Q}}{\mathcal{Q} - it} i^\* - k \frac{\mathcal{Q}}{\mathcal{Q} - it} it + Ae^{-Bit} \tag{8}$$

$$E\_{ch}^{\text{Li-Ion}} = Eo - k \frac{\mathcal{Q}}{|it| - 0.1 \mathcal{Q}} i^\* - k \frac{\mathcal{Q}}{\mathcal{Q} - it} it + Ae^{-Bit} \tag{9}$$

The discharge–charge conditions are the same as those of the Ni–MH battery type. In the equations above, the variables and parameters are: *k*, polarization constant (*Ah*−<sup>1</sup> ); *Eo*, battery voltage; *Q*, maximum battery capacity, in Ah; *i* ∗ , filtered low-frequency current dynamics, in A. In this paper,

the electrical parameters of the lithium-ion battery module are extracted and used to simulate this modular model. These parameters are listed in Table 2. Figure 5 illustrates the battery current curves at 0.083333C.

**Table 2.** Electrical parameter of lithium-ion battery module.

**Figure 5.** The battery current curves at 0.083333C (400A).

#### *4.4. Loads*

The loads consist of residential and commercial loads. Commercial loads appear on asynchronous devices to show the effect of commercial inductive loads, such as air conditioning systems, on the microgrids. Residential loads are designed according to the daily nonseasonal consumption profile of the resort island. Residential loads are simulated according to the actual difference in the specific load profile for the specified resort island.

#### *4.5. Fuel Cell Model*

Chemical energy is transformed into electrical energy using fuel cells based on a proton exchange membrane. This hybrid system, consisting of fuel cells and photovoltaics, offers better efficiency along with several other advantages of fuel cells and photovoltaics, which are reduction in carbon dioxide emission, water management, cell stack heating, and the development of cost-effective material. One cell consists of two variants of porous gas diffusion electrodes that separate the electrolyte, thus defining the cell type. The electrolyte consists of a thin conductive film, which conducts cations but restricts the flow of electrons and gases. The fuel cell is fed hydrogen gas from the electrolyzer. At the same time, oxygen enters the cell from the other side and causes a chemical reaction, where hydrogen dissociates into a proton and an electron. The chemical reaction is specified as (Equations (10) and (11)):

$$\text{H}\_2 \leftrightarrow 2\text{H}^+ + 2e^- \tag{10}$$

$$\frac{1}{2}O\_2 + 2H^+ + 2e^- \rightarrow H\_2O\tag{11}$$

Hydrogen dissociation occurs at the left anode, causing a concentration gradient between various electrodes across the membrane. The gradient leads to the protons spreading across the membrane, while the electrons are left behind. The protons drifting towards the cathode provide it a positive charge compared to the anode. Electrons are attracted to the cathode; however, their movement is restricted by the membrane, and the only way for the electrons to reach the cathode is through an external circuit, which causes a current flow. The output voltage *VFC*(*t*) for the fuel cell is specified as (Equation (12)):

$$V\_{F\overline{C}}(t) = E(t) - V\_{\text{act}} - V\_{\text{oh}} - V\_{\text{con}} - V\_{\text{trans}} \tag{12}$$

where *Voh*, *Vact*, *Vtrans*, and *Vcon* are ohmic voltage, activation voltage, mass transport loss, and concentration voltage, respectively. The voltage created by a fuel cell is specified as (Equation (12)):

$$E(t) = -\frac{\Delta G^{\circ}}{nF} + \frac{\Delta S}{nF}(T - T\_{ref}) + \frac{R \times T}{nF} \ln\left(\frac{P\_{O\_2}^{0.5} \times P\_{H\_2}^{1}}{P\_{H\_2O}^{1}}\right) \tag{13}$$

where ∆*G o* refers to electric work, also called Gibbs free energy; *F* is the Faraday constant; *n* refers to number of moles; *S* is the specific entropy; *T* refers to the operational absolute temperature; *Tre f* is set at 25 ◦C; *P* 0.5 *O*2 , *P* 1 *H*2 , and *P* 1 *H*2*O* are the respective pressures of pure oxygen, hydrogen, and water as fuel; *R* refers to the gas constant. For the fuel cell stack, the total power output *PFC*(*t*) is specified as (Equation (14)):

$$P\_{\rm FC}(t) \;=\; P\_{\rm FC} \times V\_{\rm FC}(t) \times I\_{\rm FC} \tag{14}$$

where *PFC* and *IFC* refer to the stacked fuel cell rating and the fuel cell current, respectively. For modeling fuel cell power generation, the primary aspect is the hydrogen mass flow rate. The flow rate, *QH*<sup>2</sup> *c* , required by a 1 kW fuel cell, expressed in kg/hour, is specified as (Equation (15)):

$$\left(Q\_{H\_2}\right)^{\varepsilon} = \alpha\_1 \times P\_{F\mathbb{C}}^r + \alpha\_2 \times P\_{F\mathbb{C}}^1(t) \tag{15}$$

where ∝1, ∝2, and *P r FC* refer to the fuel cell intercept coefficent in kh = g = kW rated, fuel cell curve slope in kg/h/W, and the rated fuel cell capacity in kW. For a fuel cell rated at 1 kW, assuming ∝1 and ∝2 at 0.00031 and 0.0581 kg/h/kW, respectively, *P* 1 *FC* is determined to be 0.059 kg/h, which indicates that the fuel cell requires this quantity of hydrogen per hour to produce its rated power [27].

#### *4.6. Electrolyzer Model*

The electrolyzer uses surplus electricity from the photovoltaic system to produce hydrogen for use in fuel cells, which is stored in a hydrogen tank until it is required. The aspect that requires attention is the quantity of hydrogen produced by the electrolyzer. For a 1 kW electrolyzer, the hydrogen mass-flow rate *mH*<sup>2</sup> *p* 1 in kg/hour is specified as (Equation (16)):

$$\left(m\_{H\_2}\right)\_1^p = \frac{3600P\_{el}^r \times n\_{el}}{HV\_{H\_2}}\tag{16}$$

where *P r el* is 1 kW, *HVH*<sup>2</sup> refers to the heating value of hydrogen (in MJ/kg), and *nel* represents the efficency of the electrolyte. Electrolyzer efficiency is assumed to be 90% [28]. Given the heating value of hydrogen at 142 MJ/kg, the hydrogen mass flow would be 0.02268 kg/h/kW. Therefore, the 1 kW electrolyzer will output 0.02268 kg of hydrogen per hour [29].

#### **5. Control Method for Photovoltaics and Full Cell System**

#### *5.1. Control of the Converter Interfaced Fuel Cell*

For the boost converter, general control comprising voltage stabilization and output current limitation is used (Figure 6). The current limit is fixed; however, it may be modified for varying sources (for instance, the current should be proportional to the amount of hydrogen injected in a fuel cell, depending on the attributes of the fuel cell). For the test considered in the study, the current is capped at 30 A. The DC reference voltage is aligned to the fixed value needed by the inverter connecting the energy source with the load or the grid. A deviation from the set values is assumed to be an energy deficit or surplus, and is indicated to the power management system. The reference voltage for this control system defines the maximum. Any DC voltage exceeding the maximum causes a decrease in inductor current. When State of Charge (SoC) falls below the reference, the load is powered using the fuel cells. Compared to previous schemes, the current scheme may be implemented with relative ease, while PI (Proportional Integral) gains are regulated online to provide a better response, PWM is the pulse width modulation signal, C is the capacitor, *Vdc* is the voltage of DC bus, L is the inductance.

**Figure 6.** The control of the fuel cell converter.

#### *5.2. Control of the Converter Interfaced Photovoltaics*

A single-phase boost is used to up the voltage from the panel and control maximum power point tracking. Input current (*IPV*) sensing happens before the measurement of input capacitance (*C<sup>i</sup>* ) in addition to the panel voltage (*VPV*). The maximum power point tracking algorithm uses these two values. The MPPT algorithm determines a reference point, which, when maintained at the panel, provides maximum power to be extracted from the PV system. As depicted in Figure 7, MPPT is achieved using the voltage of the outer loop and the current flowing in the inner loop. Hence, the signs for the reference outer voltage compensator and feedback are reversed. The converter output is not adjusted. A voltage feedback mechanism provides input to the internal voltage comparators, which use Pulse-Width Modulation (PWM) in an overvoltage situation to control the voltage from rising beyond the components' rated voltage.

**Figure 7.** Control of the converter with maximum power point tracking.

#### *5.3. Control of Bidirectional-Converter Interfacing Battery*

In the proposed setup, as depicted in Figure 8, the bidirectional converter has an output filter capacitor (*Cdc*), a high-frequency inductor (L), and two switches (S1 and S2) that permit bidirectional flow of current. The power management system consists of a two-voltage controller with appropriate limitation blocks to provide the necessary power flow under varying conditions. The controllers output a reference current for the storage of energy. The first aspect of control is DC bus voltage regulation, while the other controllers regulate battery voltage. To facilitate better power management in the microgrid, backup energy storage is a part of the system, which comprises a battery connected to the DC bus using a bidirectional converter. The converter serves multiple purposes: it serves as a battery charge regulator in grid-connected operation, and a boost converter to deliver energy from the batteries to the microgrid when the PV and fuel cell sources have insufficient power to feed the local loads in islanded operation. During island mode operation, the optimal operating scenario is the equivalence between load energy requirement and photoelectric power generation, where there is no need for the converter to process energy. Figure 8 depicts the bidirectional control structure and a simplified phase of the power from the converter.

#### *5.4. Control of Converter Interfaced Electrolyzer*

The buck converter is a DC to DC power converter which steps down voltage from its input (source) to its output (load). For the buck converter, general control comprising voltage stabilization and output current limitation is used (Figure 9). The input of the buck converter in this system interfaces with the DC bus in the hybrid microgrid, and the output coupling with electrolyzer.

**Figure 9.** The control of the electrolyzer buck converter.

The control system consists of one PI controller, electrolyzer voltage controller, current controller, and DC bus voltage controller. The current control loop should be faster than the voltage control loop to minimize interaction between the two loops and therefore prevent instability. The pulse width

modulation stage produces the duty cycle d that is proportional to the control voltage. The pulse width modulator makes a comparison between the control voltage and a reference waveform with peak to peak amplitude Vm. The value for Vm is selected by the designer. The frequency of the waveform corresponds to the desired converter switching frequency fs. This comparison is used to determine the switching on/off of the converter switch. The buck converter purpose, among others, is to act as an electrolyzer charge regulator during the islanded mode and grid-connected mode. Figure 9 depicts the unidirectional control structure and a simplified phase of the power from the buck converter.

#### *5.5. Method of Hybrid Microgrid Control*

The proposed microgrid has a control system implemented to stabilize the bus voltage and to regulate the power supply to serve the electrical demand during grid-connected as well as islanded mode. Table 3 lists the specifications for the design of the power management system. In such a control scheme, one unit is the master controller that controls the entire system, whereas the other units act as sources of current (slave). This case would not have a voltage differential between the outputs from the sources since the master unit is responsible for voltage regulation of all outputs. Hence, no current may circulate between the sources.


**Table 3.** The requirements of energy-management design.

In the microgrid, the DC bus voltage is quantified and matched against the reference voltage (300 V). Error processing is done using a compensator (PI block) to get the required reference impedance current for the loop. This compensation may be exhibited as the following Equation (17):

$$I\_L = K\_P \left(V\_{ref} - V\_{MG}\right) + K\_I \int \left(V\_{ref} - V\_{MG}\right) dt\tag{17}$$

The current controller regulates power flow using a comparison of the impedance current flowing through the master unit to the reference value required for system stability. Error processing is conducted using another PI block to get the appropriate duty cycle for the master converter. The PI block can be depicted as Equation (18):

$$d\_{-}=K\_{P}\left(I\_{L}^{\*}-I\_{L}\right)+K\_{I}\int\left(I\_{L}^{\*}-I\_{L}\right)dt\tag{18}$$

This topology presents the challenge of the master unit requirement. If a fault occurs, the control will not function correctly [30]. To enhance the reliability of the system, three individual sources may act as a master unit, thereby decreasing the chances of a fault in the microgrid control. The model studied has a voltage source converter (VSC) in the master role while the microgrid has a connection with the grid, where voltage and current loops are set up to regulate the voltage levels, so the VSC cannot change the power flow. The energy storage system uses a bidirectional converter to regulate voltage and power levels. During an islanded mode operation of the microgrid, this would be the master control where the voltage shall be maintained at 300 V while meeting the required load. Challenges include faults in the storage system or inappropriate SoC level for the system to operate in the island mode. There is a voltage control system having a voltage and current loop, as depicted in Figure 10.

**Figure 10.** The droop curve: (**a**) droop curve of two distributed generators and (**b**) the droop for the entire DC subgrid.

Drooping-based power sharing is displayed in Figure 10a. Terminal voltage of the DC link for every DC–DC converter is drooped with the DC power generated (*Pdcj*), using the droop coefficient *mdcj*, as specified below (Equation (19)):

$$V\_{oj}^{\*} = \left. V\_{oj}^{n} - m\_{dcj} \mathbf{P}\_{dcj} \right. \tag{19}$$

where *V* ∗ *oj*, *V n oj* are the reference and no-load DC link voltage of the converter, whereas the *j* represents a distributed generator unit. The injected DC power from each distributed generator unit (*Pdcj*) is determined to supply the common DC load (*Pdc*<sup>1</sup> ) (Equation (20)):

$$
\hbar \, m\_{dc1} \mathbf{P}\_{dc1} = \, ^\ast m\_{dc2} \mathbf{P}\_{dc2} \tag{20}
$$

#### **6. Supervisory Control Method of the Proposed System**

Calculations for the energy produced using the photovoltaic generator are done using temperature and solar radiation as input data points. The photovoltaic output (*PPV*) and the power demanded (*PL*) are compared to evaluate the energy flow to the storage system and the load. The electrolyzer (*Pel*) produces hydrogen using surplus power generated by the photovoltaic system. During an energy deficit scenario, the backup fuel cell generator (*PFC*) may supply energy using hydrogen. Figure 11 depicts the flow of in a photovoltaic-fuel-cell hybrid system control method. According to Figure 11, the control strategy is based on the following three different cases:


**Figure 11.** The proposed energy management control system.

Meaning, if the PV does not provide power for the load and the fuel cell does not start, the fuel cell and the load are separated, the electrolyzer is connected directly to the PV cells, then the load is connected according to the conditions in 1 or 2. Therefore, super capacitor units, which represent short-term volumes, provide a more stable energy response to transient load changes, or to stabilize fuel cell operation, are not considered here. Also, the electrolyzer of the proposed system is used to generate hydrogen by photovoltaic, so fuel cells can be turned on later when sufficient hydrogen is produced.

There are two levels in the higher-level controller. The first level uses individual component states to establish the operating modes. The second level is integrated with smart systems, which ascertain the way individual components behave in that operational mode. Operating modes consist of the normal mode, super capacitor charging mode, standby mode, battery charging mode, and transient mode. The standby mode has the DC–DC converter operational, while the others are disabled. When the power demand crosses a specific threshold, the controllers switch the mode from standby to transient, which sets up the console to include all controllers. A deficiency in the energy derived from the fuel cells and the battery is corrected using the super capacitor; when fuel is available, the controls switch the super capacitors to the charging mode. The super capacitor may have any magnitude of charge current until it reaches the fully charged state. Both the load and the super capacitor are powered using the fuel cells, while any shortage is supplemented using the batteries. After the super capacitor reaches 100% SoC, the controller transitions to normal load, where only the fuel cell provides power to the loads. This mode consists of the battery being disabled, while the super capacitor is still on. With a spike in load, the controller changes operation to the transient mode.

Battery charging mode may be set in two ways. The first situation, causing a switchover, is the battery state of charge falling below the minimum value. In contrast, the second situation comprises a sudden drop in load, and the batteries SoC is lesser than the maximum. Hence, with decreasing load power, the surplus flows to the battery if it is not fully charged. Splitting the complex control mechanism into several levels provides ease in modeling and controlling the individual components. Those are lower-level controllers, which include a converter controller along with the plant controller balance. The management system is examined for the proper operation of a microgrid in both grid-connected and island modes. Figure 10 depicts the general control.

In the case of grid connections, the master control is handled by the VSC inverter while the sources in the microgrid operate as slaves or sources of current. Hence, the VSC can regulate the DC voltage bus. When an unplanned event occurs, like a fault, a microgrid is required to work in the island mode and should be isolated from the grid. During such a situation, energy storage systems take master control and regulate voltage and power levels in the microgrid. Under all circumstances, the fuel cells and the photovoltaics behave as slave units. Any imbalance between the load and the power produced from the distributed generation sources is corrected by the energy management controls. Such imbalance is addressed by using battery power, given that those have an appropriate state of charge; otherwise, fuel cells are used. It must be noted that fuel cells, because of chemical reactions, respond slowly compared to an energy storage system. If the storage system is malfunctioning or has an unacceptable charge level, the fuel cell becomes the master. In case the microgrid controls are unable to balance the energy flow, and the load exceeds the power generation, the measure of last resort is load rejection. In case there is a power surplus in the grid, the DG (Distributed Generator) systems are disconnected from the microgrid, as required.

#### **7. Energy Management Systems in Houses**

In the case of houses, energy management systems incorporate intelligence to decrease energy consumption, enhance uptime, and provide added safety. Control and modeling framework for fuel cells and photovoltaics is designed. This technique uses primary performance attributes like super capacitor voltage, SoC, and DC bus voltage and subject to a proportional integral (PI). Proportional-integral control is straightforward to tune online for improved tracking. The load power is divided so as to permit the fuel cells to provide the steady load. An integrated system comprising these power sources may be considered by adding the current values where the system connects with the microgrid. The real power traded between the hybrid microgrid and the utility network *Pgrid*(*t*) is addition of power generated by the fuel cells (*PFC*(*t*)), batteries (*PBat*(*t*)), and the photovoltaic system (*PPV*), and the electrical load (Figure 1) (Equation (21).

$$P\_{grid}(t) = P\_{FC}(t) + P\_{PV}(t) + P\_{Bat}(t) + P\_{SCB}(t) - P\_{Load}(t) \tag{21}$$

*Sustainability* **2020**, *12*, 5724

During extended periods, rapid power variations in regard to the super capacitor *PSCB*(*t*) may be disregarded.

Equation (22) is specified for an extended time frame:

$$\left\{P\_{\mathcal{S}^{\mathrm{mid}}}\right\}T\;=\;\left\{P\_{\mathrm{FC}}+P\_{PV}+P\_{\mathrm{Bat}}(t)-P\_{\mathrm{Load}}\right\}T\tag{22}$$

For such a short duration, the super capacitor, due to its rapid response time, takes control of power flow. Using the inverse of Equation (21), reference power for the super capacitor *PSCB*(*t*) may be expressed as (Equation (23)):

$$P\_{\rm SCB}(t) = P\_{\rm grid}(t) - P\_{\rm FC}(t) - P\_{\rm PV}(t) - P\_{\rm Bat}(t) + P\_{\rmLoad}(t) \tag{23}$$

Super capacitors have a very swift charge/discharge response, and, therefore, they are employed in this study for voltage control on the DC bus for the hybrid microgrid. The following constraints should be factored in when determining the energy storage capacity of super capacitors (Equation (24)):

$$E\_{\text{SC\\_min}} \le E\_{\text{SC}}(t) \le E\_{\text{SC\\_max}} \tag{24}$$

where *ESC*\_*max* and *ESC*\_*min* are, respectively, the maximum and minimum permissible storage capacity for the supercapacitor. *ESC*\_*min* may be calculated as per Equation (25):

$$E\_{\text{SC\\_min}} = \text{SOC}\_{\text{SC}} \times E\_{\text{SC\\_max}} \tag{25}$$

Therefore, the larger the power drawn from the super capacitor, the stored energy reduces likewise (Equations (26) and (27)).

$$E\_{\rm SC} = \int \Delta P\_{\rm FC}(t)dt\tag{26}$$

$$
\Delta P\_{\rm FC}(t) = K\_{\rm p\varepsilon}(e\_{\rm sc\_{ref}}(t) - \ell\_{\rm sc}(t)) \tag{27}
$$

Battery storage capacity is subject to the following constraints (Equations (28)–(35)):

$$E\_{\text{Batt\\_max}} \ge E\_{\text{Batt}}(t) \ge E\_{\text{Batt\\_min}} \tag{28}$$

$$0 \le P\_{\text{bat},\text{cl}}(t) \le Y\_{\text{cl}}(t) \cdot P\_{\text{bat},\text{cap}} \cdot (1 - \text{SoC}(t-1)) \tag{29}$$

$$0 \le P\_{\text{bat}, \text{disch}}(t) \le \mathcal{Y}\_{\text{bat}, \text{disch}}(t) \cdot P\_{\text{bat}, \text{cap}} \cdot \text{SoC}(t-1) \tag{30}$$

$$\text{Y}\_{\text{bat},\text{ch}}(t) + \text{Y}\_{\text{bat},\text{disch}}(t) \le \text{1}, \text{Y}\_{\text{bat},\text{ch}}(t), \text{Y}\_{\text{bat},\text{disch}}(t) \text{ } \vert 0, 1 \rangle \tag{31}$$

$$\text{SoC}(t) = \text{SoC}(t-1) - \frac{1}{P\_{\text{but,cap}}} \times \left(P\_{\text{bat,disch}}(t) - P\_{\text{bat,ch}}(t)\right) \tag{32}$$

$$0 \le \text{SoC}(t) \le 1\tag{33}$$

$$\text{SoC}(t\_0) = \text{SoC}\_{\text{initial}} \tag{34}$$

$$\text{SoC}(t\_{end}) = \text{SoC}\_{final} \tag{35}$$

Equations (29) and (30) stand for the power capacity constraints of the battery for charging and discharging status. Equation (31) keeps the simultaneous occurrence of discharging and charging of the battery. Finally, Equations (32)–(35) are related to *SoC* or the level of the stored energy in the battery. In the above equations, *Pbat*,*ch*, *Pbat*,*disch*, and *Pbat*,*cap* stand in the charging power, discharging power, and the maximum available power from the battery, respectively. Also, *Ybat*,*ch* and *Ybat*,*disch* are ancillary binary variables indicating the charging or discharging status of the battery.

House-based microgrids may trade power with the utility grid. Any energy surplus produced by the microgrid after the batteries are fully charged could be sold to the grid. If the total production of the microgrid is insufficient to meet the load, power should be bought from the utility grid (Equation (36)).

$$P\_{\rm FC}(t) = P\_{\rm grid}(t) - \hat{P}\_{\rm PV}(t) + \hat{P}\_{\rm load}(t) + \Delta P\_{\rm FC}(t) \tag{36}$$

Energy management and power control systems are very important for hybrid networks based on intermittent sources. The photoelectric control provides reference voltage (V\_MPP) for the unidirectional converter. With the converter controlling the PV output, PV works at the voltage that uses the maximum power. Super capacitor discharge and charge circuit control maintains a DC voltage with a constant value of 270 V. Even with DC voltage ripples, controlling the transformer at the end of the load makes the load voltage as smooth as possible. Fuel cells can instantly provide maximum energy. The boost transformer sets the performance limit. Low capacitance and induction values accelerate transients and accelerate overall response, but require faster switches.

#### **8. Results of Proposed System**

Hybrid systems are corroborated by applying a thorough model in the environment of Simulink. This model represents a different emergency power system which is based on super capacitors, fuel cells, and Li-ion batteries. The proposed model also includes different power handling systems for fuel cell fusion power supplies. The findings are attained based on the standard profile of daily load for scrutinized homes, as displayed in Figure 12. Table 4 presents the details of the power system under study.

**Figure 12.** The curve of demand.



The 25 kV utility grid is connected to houses via a "25 KV/270 V line to line (V\_rms)" transformer to step down voltage to the required grid voltage, which means the output of the transformer equal to 380 V line to line peak voltage (*VPeak* (*L*−*L*) ), meaning the transformer output voltage is equal 220 V line to ground.

$$\text{V}\_{\text{Peak (L-L)}} = \sqrt{2} \times V\_{\text{rms (L-L)}} = \sqrt{2} \times 270 = 381.83766 \text{ V} \tag{37}$$

$$V\_{\text{Peak (L-N)}} = \frac{V\_{\text{Peak (L-L)}}}{\sqrt{3}} = \frac{381.837}{\sqrt{3}} = 220\,\text{V} \tag{38}$$

where *VPeak* (*L*−*L*) is the peak line to line voltage, *Vrms* (*L*−*L*) is the RMS (Root Mean Square) line to line voltage, *VPeak* (*L*−*N*) is the peak line to neutral voltage.

The three-phase AC loads with variable power factor and apparent power are used to simulate a residential load profile. The energy management system distributes energy between sources according to the given energy management strategies. The peak electrical load is 160 kW at 3 p.m., while the lowest load occurs at 29 kW from 10:30 p.m. to 8:30 a.m. Figure 13 shows the photoelectric voltage and photoelectric current. Photovoltaic power generation is set to track the maximum power point proportional to radiation. The meteorological data for the hybrid power generation system of photovoltaic/fuel cells are the amount of solar radiation (W/m<sup>2</sup> ) and ambient temperature (◦C). These typical weather data are collected every hour. The microgrid reference generated is set to one real energy step (25 s, Figure 13) and the energy detected between the microgrid and the main line is very close to the power reference. Super capacitors and batteries compensate for the differences in the reference power of the microgrid and all the differences of the negative energy of the microgrid (total load and photoelectric cells).

**Figure 13.** (**a**) Photovoltaics voltage and (**b**) photovoltaics current.

It can be seen that the fast power fluctuation due to the slow response time of the fuel cell is reduced in the short-term energy storage system, as shown in Figure 14. The fuel cell will try to charge the super capacitor to increase the SoC if the load is not too high. Figure 15 shows the fuel cell voltage and fuel cell current. Note that a positive super capacitor current means that the super capacitor will supply the load, and a negative current means that the super capacitor will be charged. In this case, the carrier voltage fluctuates in the permissible range from 226 to 270 V. The super capacitor voltage is properly controlled between 226 and 270 V, as shown in Figure 16c. SoC range for super capacitors is 85–100%. Figure 17 shows the load voltage and load current. The load current increases in 8 s and decreases in 20 s. As shown in Figure 18, the voltage decreases due to the sudden increase in the load current.

**Figure 15.** *Cont.*

**Figure 15.** (**a**) Fuel cells current and (**b**) fuel cells voltage.

In this study, the fuel cell and photovoltaic hybrid power generation structure is either the island or grid-connected mode. The load demand regions are determined by a preset power. Nonetheless, most resident energy consumption of power, like lighting and apparatus, is always varying. Based on the scheme of energy management, the energy regulation system manages the power of every power supply apparatus through the reference signals of the photovoltaic and fuel cell converter, together with super capacitors, batteries, and fuel cells. A description of what happens in a virtual emergency situation follows. At t = 0 s, the primary load is provided by the chief generator, which switches on the fuel cell hybrid system, to plan for an emergency scenario. Now, the fuel cell starts recharging the battery as well as the super capacitor using its optimal power. At time t = 8 s, every load is connected and the loads soar to 160 kW. At this moment, the extra power needed is instantly provided by the super capacitor because of its fast dynamics, although the power in the fuel cell increases gradually. At time t = 13 s, the super capacitor gets discharged and its DC bus voltage goes below the required amount (270 V) and the battery starts supplying power to control the DC bus voltage to 270 V. At time t = 20 s, the DC bus or super capacitor voltage gets to 270 V and gradually the power of the battery drops down to zero. The fuel cell supplies complete load power and keep on recharging the super capacitor. At t = 13 s time, the battery also gets to its highest power and the super capacitor supplies additional load power. At time t = 20 s, the load power reduces below the fuel cell maximum power. Due to the slow fuel cell dynamics, the extra fuel cell power during transients is transferred to the supercapacitor. Owing to the fuel cells' slow dynamics, power during transients is shifted towards the super capacitor. At time t = 20 s, the power of the load is reduced and falls below the fuel cell highest power, and the additional fuel cell power is shifted towards both the super capacitor and battery. At time t = 23 s, the load power gets reduced rapidly. The additional fuel cell energy gets stored in the batteries and super capacitor. The fuel cell provides nearly all the total load power needed and decreases its power gradually to its optimum and the battery gets recharged.

**Figure 16.** (**a**) State of charge of the super capacitor, (**b**) super capacitor current, (**c**) super capacitor voltage.

**Figure 17.** (**a**) Line to neutral loads voltage, (**b**) loads current, (**c**) line to line load voltage.

**Figure 18.** (**a**) Step change in load line to neutral voltage at 8 a.m., (**b**) step change in load current at 8 a.m., (**c**) load line to line voltage at 8 a.m.

#### **9. Conclusions**

In this study, the new home management system demonstrates the effectiveness of this method to deliver constant power to the customer through different controller designs. Moreover, this paper recommends an effective control scheme for the effortless shift from grid-connected mode to island mode because of unintentional islanding. Furthermore, a technique for smart management and regulation was employed to ensure optimum functioning of a hybrid microgrid formed using different renewable energy sources, as well as different energy storage devices. Using this energy management strategy as component of the control design of a grid-linked microgrid can reduce the total operational cost. The regulatory control was included in the control model to correct any variation among the primary grid power and the intended reference power by altering the reference fixed power of the battery provided by this energy management system. This study put forward a full local regulatory design for the recommended system. The local controller regulates the local power, DC bus voltage, and the current of the microgrid where the local controller's aim is to follow the fixed point of the monitoring controller. The outcomes show that the suggested robust energy management and regulation scheme in the hybrid microgrid system, which is based on ecofriendly energy, supports nominal use of the utility grid power. The proposed scheme is able to distribute power amongst the distributed generator components even under uneven conditions. Moreover, from the outcomes, it is apparent that the output voltage of the unit terminals of the distributed generator gets higher when the unit of fuel cell is attached along with the super capacitor and the photovoltaic. The loads of the houses for the hybrid microgrid were taken into account and the hybrid microgrid can swap power with the transmission grid. The hybrid microgrid used in this research is highly reliable.

The outcomes suggested that the alternating integration of solar energy sources and fuel cell in the microgrid should be devised carefully in individual operation. The recommended control schemes provide exceptional performance under various operating conditions. Batteries improve the system reliability since they store additional renewable energy when the demand is low and supply energy when the demand is high.

For future work, these results suggest (1) implementing the strategy presented on a real network and comparing the results, and (2) investigating the effects of other uncertain parameters, such as fuel cost, on microgrid planning.

**Author Contributions:** B.N.A.: writing—original draft, software, methodology, and validation; M.D.E.: formal analysis, investigation, resources, writing—review and editing; B.H.J.: supervision, funding, writing—review and editing. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **A Residential Load Scheduling with the Integration of On-Site PV and Energy Storage Systems in Micro-Grid**

#### **Ihsan Ullah <sup>1</sup> , Muhammad Babar Rasheed 2,**<sup>∗</sup> **, Thamer Alquthami 3,**<sup>∗</sup> **and Shahzadi Tayyaba <sup>1</sup>**


Received: 24 October 2019; Accepted: 19 December 2019; Published: 25 December 2019

**Abstract:** The smart grid (SG) has emerged as a key enabling technology facilitating the integration of variable energy resources with the objective of load management and reduced carbon-dioxide (CO2) emissions. However, dynamic load consumption trends and inherent intermittent nature of renewable generations may cause uncertainty in active resource management. Eventually, these uncertainties pose serious challenges to the energy management system. To address these challenges, this work establishes an efficient load scheduling scheme by jointly considering an on-site photo-voltaic (PV) system and an energy storage system (ESS). An optimum PV-site matching technique was used to optimally select the highest capacity and lowest cost PV module. Furthermore, the best-fit of PV array in regard with load is anticipated using least square method (LSM). Initially, the mathematical models of PV energy generation, consumption and ESS are presented along with load categorization through Zero and Finite shift methods. Then, the final problem is formulated as a multiobjective optimization problem which is solved by using the proposed Dijkstra algorithm (DA). The proposed algorithm quantifies day-ahead electricity market consumption cost, used energy mixes, curtailed load, and grid imbalances. However, to further analyse and compare the performance of proposed model, the results of the proposed algorithm are compared with the genetic algorithm (GA), binary particle swarm optimization (BPSO), and optimal pattern recognition algorithm (OPRA), respectively. Simulation results show that DA achieved 51.72% cost reduction when grid and renewable sources are used. Similarly, DA outperforms other algorithms in terms of maximum peak to average ratio (PAR) reduction, which is 10.22%.

**Keywords:** HEM; PV sizing; Load scheduling; Dijkstra Algorithm; BPSO; GA; optimization

## **1. Introduction and Motivation**

With the rapid increase in population and economic growth, global energy consumption is also increasing drastically. High reliance on fossil fuel and increased use of thermal power resources contributed to an increase in CO<sup>2</sup> emissions that causes an increase in global warming [1]. While adopting the ways to tackle the increased energy demand, a reduction in greenhouse gas emissions is a major challenge that must be met on a global scale in order to promote energy sustainability. In recent years, while reducing the gap between energy generation and demand through a combination of thermal and nuclear energy resources, the pursuit of power generation through renewable energy sources, such as solar, wind, biomass and lowering environmental load, has been promoted with the aim of reducing greenhouse gas emissions, significantly. Renewable energy is becoming widely accepted as a new source of energy to provide power to residential and commercial buildings around

the world. On the other hand, energy is one of the basic necessities for sustainable development of society that must surely be delivered in an efficient way in order to promote a green environment with reduced CO<sup>2</sup> emissions [1]. However, the great paradigm shift due to the involvement of information and communication technologies (ICTs) the traditional ways of energy generation, transmission and distribution need to be upgraded [2–8]. The author in [9] presented a hybrid opinion network containing of continuous and discrete valued agents. This model discussed a communication behavior for social dynamical systems. The scaled consensus of switching topologies with continuous and discrete time subsystems is investigated. The traditional electricity infrastructure is not fully capable of handling and managing the distributed energy resources with high quality of communication service and grid stability. This is due to the fact that increasing energy demand needs extra costlier generation or back-up reserve capacity units that lead to high tariff rates and CO<sup>2</sup> emissions.

Therefore, to cope with such types of situations, a smart grid (SG) concept has recently been introduced which has the capability to fulfil the load demand that benefices both the end users and utility market [10]. The users can enjoy an uninterruptible power supply with economical tariff rate and utility can avail the opportunity to improve power system stability through supply–demand balance. The former can be achieved by exploiting efficient load management through distributed as well centralized control strategies [11]. For this purpose, different researchers introduced residential and commercial load scheduling techniques considering both end user and utility objectives [12,13]. Among these solutions, the major emphasis is given to balance the load demand through market supply without heavily relying on peak power plants and reserved capacity units [14]. However, some researchers focussed on balancing the load demand through load scheduling and energy management using optimization technique. Although these kinds of techniques are efficient in managing the load demand with partial or full control on residential load [15]. The peaks and troughs in the aggregated user energy demand profile are caused by the temporal variations in energy demand. The user's participation is required in balancing the load by allowing the grid to reschedule the load. However, the end users are affected due to unplanned load scheduling by the grid [16]. For example, it is assumed that the user suffers the same degree of inconvenience if base and/or non-flexible load is shifted in either direction (before or after) of its most preferred time slot. In other words, delaying a base and/or non-flexible load by 2 time slots or advance scheduling the same load by 2 time slots will result in same amount of user inconvenience. To handle this situation, some researchers put their efforts into devising the load control strategies with major focus on load scheduling without disturbing end users' objectives [17]. Furthermore, it is also equally important to understand that world is moving from a centralized energy control system to a distributed control and management system with the objective of promoting the environment [18]. In addition, this is also due to the European vision of the year 2030 and 2050 for achieving the objective of green energy and green economy, incorporation of renewable energy resources such as PV, wind, geothermal, biomass and electric vehicles into existing electrical systems is required [19]. In achieving these objectives, numerous researchers have done some work in managing the integration of distributed energy resources through autonomous control strategies [20]. Some researchers focussed on developing the autonomous control algorithms that provide benefits to only end users [21], while others focussed on maximizing utility and energy retailers benefits [22]. Some work also tackled the utility and end user objectives as a joint objective and tried to find the solutions to provide the benefits to both parties [23].

In recent years, the resources of water for producing energy are depleting very quickly due to the effects of climate change on hydrology patterns [24,25]. It is estimated that only 200,000 km3—1% of water is available for agriculture, residential, industrial and power sectors in order to meet everyday needs [26]. Furthermore, the per capita energy demand is expected to rise significantly due to the rapid growth of population and industries in developed and developing countries [27]. As a result of these great challenges, the efficient management of the available capacity is imperative to prevent the over-exploitation [28]. On one hand, these important challenges are considered for managing the energy resources efficiently, whether these are used used for the end user comfort or upgrading

the existing standards or infrastructure [29]. On the other hand, better management of fuel intensive resources and end user consumption behaviours can alleviate the strain on these commodities [30]. Therefore, flexible control of the energy supply and use system [19,31] is crucial within the context of renewable energy and storage integration [32]. In addition, a recent study shows the water usage for carbon-based and nuclear fuel-based energy as follows: (i) oil from oil reservoirs (70–1800 L/GJ), (ii) shale gas (36–54 L/GJ), coal (5–70 L/GJ), (iii) uranium (4–22 L/GJ), and (iv) traditional oil (3–7 L/GJ) [31]. Another study reveals that 76.9% of electricity is produced from oil, natural gas, coal, and nuclear fuels in the year 2015, while only 16% is generated from hydrothermal power plants. The energy generated from hydrothermal power plants led to high emission of carbon and causes serious environment concerns [33]. Similarly, a significant amount of power is required to support industrial and residential sectors and with this level of coupling, significant synergy could be observed while analysing these types of coupled system holistically. In the meantime, the SG concept was introduced and gained significant attention is given to the integration of renewable energy resources, especially into the demand side load management as a mean of de-carbonizing the environment. The importance of producing energy from coal, natural gas, oil and nuclear commodities is decreasing day by day, thus results in moving towards integration of green energy concept. The integration of green energy (solar and wind) and storage systems are at present gaining popularity [34]. Recent studies regarding variable energy resources and energy management through electrical load scheduling show that a dynamic control mechanism is required that integrates the intermittent nature of variable energy resources and dynamic energy consumption trends. It is a complex task, however, studies showed that due to high penetration and intermittent nature of variable resources, operator and energy retailers are restricted to rely on the combined dispatch of all energy resources. Operators need to encourage the end users to install stand-alone renewable energy systems to improve the system stability through load management. Otherwise, the mismatch between demand and supply may lead to infeasible dispatch of energy and increase the marginal cost production [35–37], while the challenges of utility and residential premises in managing the residential load demand with the integration of renewable energy resources may seem unrelated. However, these are actually interlinked entities and their resolution is potentially synergistic. Renewable energy technologies are a cheap source of energy and provide with low CO<sup>2</sup> emissions. Since, renewable energy can easily be stored in back-up storage systems and can act as a flexible source of energy regarding supply and demand side of electricity systems [38]. Thus, demand response (DR) programs have an eminent effect on SG infrastructure in managing the end user load via market clearing prices and incentives. Because, residential sector is consuming 40% of energy and diversity in energy consumption plans poses major challenge to energy retailers to encourage them in participating energy management programs. The effective outcomes of SG concept are difficult to achieve without active participation of end users in DR-based energy management programs.

The review of relevant literature shows that homes equipped with Home energy management (HEM) architectures significantly reduce the electricity cost. HEMs equipped with renewable energy generation and storage architecture are given importance but the load matching with the renewable energy generation is ignored. Several HEM architectures efficiently shift the load from grid to an on-site energy generation and storage system, but the design of PV-based energy generation and storage setup has not yet been reported. An efficient and autonomous HEM architecture is required with twofold objectives, i.e., benefices the end user and utility. With the end user perspective, HEM needs to incorporate the on-site PV generation and storage system that optimally uses the grid and stored energy to maximize the comfort level. With the utility perspective, HEM should limit the energy consumption to a certain threshold level to avoid the load shedding, blackouts and use of peak power plants. Thus, our objectives are to design a demand side energy management (DSEM) model that (i) Categorizes, monitors and controls the household different loads according to end user preference and priority, (ii) Maximizes the end user comfort level by scheduling the load with minimum delay, (iii) Anticipates the optimum number of PV modules for an on-site load, i.e., smart home in our case, (iv) Efficiently integrates the PV generation and storage system with energy management controller

(EMC), (v) Optimally manages the grid energy and on-site stored energy, (vi) Minimizes the end user electricity cost, (vii) Minimizes the PAR to increase the grid lifetime, avoid the blackout, load shedding and use of Peak Power Plants (PPPs), and (viii) Efficiently formulates and solves the optimization problem considering the constraints.

#### **2. Literature Review**

With rapid increase in energy demand and CO<sup>2</sup> emissions due to massive use of fossil fuels, a significance attention needs to be given to consider renewable energy resources. Among all renewable energy resources, the most promising resources with nearly zero carbon emissions are the windmills and PV systems [39]. In comparison to both of the resources, energy obtained from PV system is highly attractive due to its abundant nature. Therefore, several incentive policies are designed especially for the end users to encourage them installing PV systems to alleviate uncertain energy demand during off peak hours. The integration of PV systems at end user side will reduce the energy cost, PAR, blackouts and carbon emissions, particularly. Consequently, PV systems tied with grid are expected to sustain popularity for the next generations [40]. One of the major challenges during the design of energy management systems (EMS) with the PV systems is its intermittent nature and thus leads to an increased design complexity. Other major challenges of PV system integration with existing electricity supply system are the variable PV generation and dynamic energy demand trends. The work reported in [41] used the on-site battery energy storage and PV system to solve the supply–demand mismatch problem. The work discussed in [42] focussed on efficient energy use for cost saving through an optimization algorithm with the integration of battery storage systems. The end user achieved the reduced energy cost by compromising on their comfort level. Thus, cost reduction and comfort maximization are contradictory objectives, hence difficult to achieve simultaneously.

Due to the battery-based ESS, autonomous control algorithms considering variable energy resources and dynamic energy demand are sought to optimally manage the available resources [43]. Therefore, combined dispatch and use of power resources considering a utility tariff scheme leads towards a complex optimization problem. Thus, finding the optimal charging and discharging patterns of battery storage system would ultimately minimize the end user electricity cost and maximize the power system stability. Numerous studies are available dealing with the complex nature of such types of problems and their possible solutions [44]. For nonlinear problems, it is less feasible to use mixed integer linear programming (MILP) and linear programming (LP) approaches without transforming the problem first [45]. The authors in [46] formulate the objective function using a binary and discrete decision variables with a low computing resources [47,48]. The authors in [49] used a hybrid computational intelligence approach for energy optimization using GA and Evolutionary algorithms. In this study, load is scheduled in such a way to minimize the electricity cost rebound peaks. Another study used ant colony optimization (ACO) algorithm to solve predefined load scheduling problem [50]. It is also concluded from the study that heuristic and metaheuristic algorithms such as; PSO, GA, ACO, are comparatively efficient in achieving the optimized solutions when uncertain decision variables are involved [51,52].

The extensive literature review regarding load scheduling has come up with individual technologies, policy recommendations, system analysis and control techniques, energy management, PV-based generation and storage integration issues [53]. Policy-based system analysis techniques sometimes tend to take statistical, qualitative and qualitative approaches while focusing on efficient resource management especially at residential premises [54]. Similarly, quantitative approaches have been case study driven and thus focusing on realistic solutions by taking into consideration utility and end user objectives. Some studies have a particular focus on devising customer centric solutions, while others tend to consider utilities concerns as well [55]. In the meantime, some researchers being able to devise autonomous energy management solutions considering utility and variable energy resources at the same time to facilitate both parties [56]. However, it is found that due to contradictory objectives, the solutions still have some gaps to be handled efficiently [57]. Another problem associated with some

load management techniques is the single layer optimizations, such as; residential load scheduling [58], generation scheduling with distributed resources [59], optimal storage integrations [60], end user comfort management with particular focus on cost reduction [61]. Thus, due to lack of generic techniques considering respective objectives, most of these techniques neither feasible nor generally extensible to other case study geographies [62].

To complete the discussion, several researchers used different algorithms and energy management techniques to achieve energy saving and optimization objectives. Statistical results show that the building equipped with EMSs have energy saving factor from 11.39% to 16.22% each year. The EMS used for artificial lighting system reported the highest energy saving factor, of up to 39.5%. Similarly, the energy saving factor for HVAC and other equipments are reported as 14.07% and 16.66%, respectively. Furthermore, the published work also reported that EMS based on different optimization algorithms, such as the harmony search algorithm, enhanced differential evolution, and harmony search differential evolution, efficiently reduced the end user electricity cost by a factor of 17.84%, 11.12%, and 13.2%, respectively. Homes and buildings equipped with EMS architectures significantly reduce the electricity cost. However, EMS equipped with renewable energy generation and storage system and load matching with the PV generation system is not yet reported in the residential sector.

#### *2.1. Contributions*

The review of relevant literature shows that homes equipped with HEM architectures significantly reduce the electricity cost, although HEM equipped with renewable energy generation and storage units are given importance. However, the estimation of installed capacity of renewable and storage units as per load requirements is yet to be explored and significant work is required in this area. The literature reveals that HEM architectures efficiently manage the load considering grid and on-site energy generation and storage systems. However, the design of PV-based energy generation and storage systems need to be explored in order to get the full benefits of SG technology. Thus, an efficient and autonomous HEM architecture is required to provide benefits to both end user and utility. With the end user perspective, HEM needs to incorporate the on-site PV generation and storage system that optimally uses the grid and stored capacity to maximize the comfort level. With the utility perspective, HEM should be able to minimize the energy consumption to a certain level to avoid the load shedding, blackouts and use of expensive peak power plants. Thus, on the basis of aforementioned discussion, this work proposes a new home energy management (HEM) mechanism to address the limitations and drawbacks. The proposed work has the following capabilities; it facilitates the end users to decide whether they want to maximize comfort or cost (through a waiting time parameter), efficient load management with the integration of RES and user involvement significantly improved the power system stability through balanced supply–demand profile and it helps in reducing CO<sup>2</sup> emissions as renewable energy and storage units act as "first choice" in the proposed work. The key contributions of this paper are discussed as follows:


#### *2.2. Paper Outline*

The rest of the paper is organized as: Section 3 discusses the DSEM model describing the mathematical description of different loads, on-site PV generation and storage systems, hybrid energy system and cost calculation mechanism. Section 4 presents the formulation of multi-objective optimization problem. Section 5 provides the description of different algorithms which are used to solve the proposed problem. Section 6 presents the results obtained from different algorithms focusing on load scheduling, cost and PAR reduction, optimal PV energy and storage integration and use. Finally, the paper is concluded in Section 7.

#### **3. DSEM Model**

The energy of a residential grid network is supplied by the utility and shared by several users through the power line. DSEM is equipped with EMC which is connected to household load on one side and smart meter on other side. The working of EMC is to: (i) collect the user preferences and demand information, (ii) send the control signal to each load via wireless network controller (WNC) for optimal consumption of energy, (iii) gather the energy generation and storage information from the on-site energy generation and ESS, (iv) optimally use the grid energy and ESS, and (v) receive the time variant electricity price signal from the utility. Based on the collected information, EMC sends an on-site energy demand to the smart meter. The smart meter bi-directionally communicates through communication line with the utility. The DSEM architecture is shown in Figure 1.

**Figure 1.** DSEM Architecture.

#### *3.1. Modelling Consumption Behaviour*

This section discusses the load consumption behaviour of the considered load. We also discuss the mathematical models of all load types and their respective constraints and limits. Let **A** denotes a set of automatic and manual operated residential loads such that:

$$\mathbf{A} = \mathbf{B} \cup \mathbf{C} \cup \mathbf{D} \cup \mathbf{E} \tag{1}$$

where, **B**, **C**, **D** and **E** represent sets of base, delay tolerant, semi delay tolerant and critical load, respectively. It is considered that **B** and **E** are zero-delay tolerant load, since the end user is not willing to give-up on the comfort level. The category **C** includes elastic loads in such that their operation time can be interrupted or shifted to any other time slot over the given time horizon H. The category **D** comprises semi-delay tolerant loads that are assumed to be operated within a specific time window. The load in category **E** can be added any time as per user needs. The loads in **B** and **E** have the highest priority and zero tolerance towards delay. Moreover, the loads in **D** has the second highest priority and can bear some delay on the basis of predefined intervals [*αa*, *βa*]. The category **C** is maximum delay tolerant and has the minimum priority. For each load *a* ∈ **A**, the power consumption vector over the given time horizon *t* ∈ H is written as;

$$P\_{a,t} \stackrel{\alpha}{=} [p\_{a\_1,t1}, p\_{a\_2,t\_{2'}}, \dots, p\_{a\_n,\forall t}].\tag{2}$$

The loads in **B**, **C**, **D** and **E** consume certain amount of power during time *t*, can be written as:

$$\sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} P\_{a,t} \times \sigma\_{a,t\_{\mathcal{I}}} \tag{3}$$

subject to:

$$
\underline{P\_a} \le P\_{a,t} \le \overline{P\_a} \text{ } \forall a \in \mathbf{A} \text{ }\tag{3a}
$$

$$
\sigma\_{a,t} = 1, \forall t \in \{t\_1, t\_2, t\_3..., t\_n\}, \forall a \in \mathbf{B}, \tag{3b}
$$

$$
\sigma\_{a,t} = 0, \forall t \in \mathcal{H} \backslash t, \forall a \in \mathbf{B}, \tag{3c}
$$

$$
\sigma\_{a,t} = 1, \forall t \in \mathcal{H}', \forall a \in \mathbb{C}, \tag{3d}
$$

$$
\sigma\_{a,t} = 0, \forall t \in \mathcal{H} \backslash \mathcal{H}', \forall a \in \mathbf{C}, \tag{3e}
$$

$$
\sigma\_{a,t} = 1, \forall \gamma\_a \le t \le \delta\_{a\prime} \forall a \in \mathbf{E}, \tag{3f}
$$

$$
\sigma\_{a,t} = 0, \forall t = \mathcal{H} - (\gamma\_a + l\_a), \forall a \in \mathbf{E}, \tag{3g}
$$

where, *σa*,*<sup>t</sup>* is the status of load *a* during time slot *t*, *P<sup>a</sup>* and *P<sup>a</sup>* is the minimum and maximum power consumption, *γ<sup>a</sup>* is the operation start time, *δ<sup>a</sup>* is the operation finish time, and *l<sup>a</sup>* is the length of operation time of load *a*. Equation (3a) gives upper and lower limits on power, while Equations (3b)–(3f) denote power ON/OFF status of various loads according to the respective limits and conditions. The category **E** includes critical load, that can be plugged-in to the HEM architecture any time H. The semi-delay tolerant load has a specific operation window e.g., for load *z*1, the lower and upper bound are z<sup>1</sup> and z1, respectively, in which load has to complete its task. The duration between z<sup>1</sup> and z<sup>1</sup> is divided as [z<sup>1</sup> , z<sup>1</sup> + *t*], [z<sup>1</sup> + *t*, z<sup>2</sup> ], [z<sup>2</sup> , z<sup>2</sup> + *t*], and [z<sup>2</sup> + *t*, z<sup>1</sup> − *t*], and [z<sup>1</sup> − *t*, z1]. During the interval [z<sup>1</sup> , z<sup>1</sup> + *t*], [z<sup>1</sup> + *t*, z<sup>2</sup> ], [z<sup>2</sup> , z<sup>2</sup> + *t*], and [z<sup>2</sup> + *t*, z<sup>1</sup> − *t*], and [z<sup>1</sup> − *t*, z1], the load may be either in ON or OFF state. After the time z1, the load must be in OFF state. The power scheduling of load category **D** over the specified time range [*αa*, *βa*] is given as:

$$
\alpha\_a \le t \le \alpha\_a - \beta\_a - \rho\_a, \quad \forall a \in \mathbf{D}, t \in \mathcal{H}. \tag{4}
$$

where, *ρ<sup>a</sup>* gives the amount of time delay in regard when load is rescheduled to other feasible time slot during peak hours or overload conditions. The Equation (4) denotes that the load *a* ∈ **D** is restricted in between *α* and *β* time intervals to complete its task. The allowed time horizon for any load *a* ∈ **A** should not be less than the length of its operation time and is expressed as:

$$
\mathbb{T}\_a \le \beta\_a - \mathfrak{a}\_{a\prime} \,\,\forall \mathcal{H}. \tag{5}
$$

Then based on the aforementioned limits and constraints associated with respective load types, the information is transmitted to EMC via home area network as shown in Figure 2.

**Figure 2.** Input set by the user.

The operation start and finish time of load *a* ranges between *γ* and *δ* and is calculated as:

$$
\gamma\_a \le r\_a \le \delta\_{a\nu} \; \forall \; a \in \mathbf{A} \tag{6}
$$

subject to:

$$
\omega\_a \le \gamma\_a \le \beta\_a - l\_{a\prime} \tag{6a}
$$

$$
\delta\_a \le \beta\_{a\nu} \tag{6b}
$$

where, Equation (6a) gives the operation start time of *γ<sup>a</sup>* and Equation (6b) shows that each load must complete its task during allowed time horizon. Our objective is to optimize the power consumption vector of each load type over the given time interval. Therefore, in order to achieve this objective, the major emphasis is given to obtain optimized scheduling patterns of each load, especially elastic and semi elastic loads.

The category **E** has the highest priority and EMC needs to operate immediately whenever any service is required. EMC manages, monitors and controls the household loads and its working operation is shown in Figure 3.

**Figure 3.** EMC Load Management and Control Architecture.

Load Delay

It is usually expected that residential load can finish their working within the permissible time limits in order to avoid extra delay, which may disturb end user comfort level. Keeping this objective in mind, the maximum acceptable delay (*ρ*) that any load can bear is calculated as:

$$\overline{\rho\_{a,t}} = 0, \forall a \in \{\mathbf{B}, \mathbf{E}\}, t \in \mathcal{H} \tag{7}$$

$$
\overline{\rho\_{a,t}} = \mathcal{H} - l\_{a\prime} \,\forall a \in \mathbb{C}, t \in \mathcal{H} \tag{8}
$$

$$\overline{\rho\_{a,t}} = (\beta\_a - a\_a) - l\_{a\nu} \,\forall \, a \in \mathbf{D}, t \in \mathcal{H} \tag{9}$$

whereas, the minimum delay (*ρ*) for each load over the given time interval is:

$$\underline{\rho\_{a,t}} = 0, \forall \; a \in \mathbf{A}, t \in \mathcal{H}, \tag{10}$$

where, Equation (7) shows that load *a* belongs to the load category **B** and **E** and thus have zero tolerance towards delay, Equation (8) represents the load category **C** that has the minimum priority and can be scheduled anytime in the entire time horizon and Equation (9) shows that load *a* belongs to the semi-delay tolerant category and can only be scheduled in the pre-set allowed time slots. Where, Γ*req*, *<sup>a</sup>* denotes the required operation time of load *a*, and Γ*EMC<sup>a</sup>* provides the scheduled time interval. Hence, Equation (10) elucidates that scheduled time interval is equal to the required amount of time needed to perform a specific task.

$$
\Gamma\_{req,a} = \Gamma\_{EMG\_a} \tag{11}
$$

Therefore, in order to achieve this objective, EMC seeks to minimize the mismatch between required and optimized time intervals in association with limits and constraints.

#### *3.2. On-Site Energy Generation and Storage System*

In communication and networks, the coexistence of both discrete and continuous valued agents is interpreted as hybrid system [9]. Whereas in SG, a hybrid energy system is defined as the existence of more than one energy system to accommodate the end users energy demand. As was discussed, total load demand is fulfilled by the grid, renewable energy and storage systems and this complete architecture is considered to be a hybrid energy system, wherein renewable energy and battery storage systems act as "primary choice", while the grid source acts as secondary choice in fulfilling the energy demand. In doing so, the end user can enjoy the reduced electricity cost, while utility gains the benefits in terms of power system stability as high peaks are eliminated during demanding hours. This proposed system architecture is shown in Figure 4.

**Figure 4.** A conceptual diagram of proposed hybrid energy system.

#### 3.2.1. PV Generation

The average power obtained from installed PV modules is calculated as:

$$\kappa\_{\text{avg}} = P\_{\text{i}} \times \varrho\_{\text{i}\nu} \tag{12}$$

where, *P<sup>i</sup>* denotes the power produced by PV cell at irradiance level *i* and *ς<sup>i</sup>* gives the probability of irradiance. The Equation (12) for irradiance spectrum can be written as [46]:

$$
\kappa\_{\text{avg},t} = \int P\_{i,t} \times \varrho\_{i,t} \, dt, \forall t \in \mathcal{H}, \tag{13}
$$

The Equation (13) denotes the average power output of a PV module for all the possible irradiance spectrums in the integral form. Where, *Pi*,*<sup>t</sup>* can be written in terms of voltage *V* and current *I* as under:

$$P\_{\mathbf{i},\mathbf{t}} = V\_{\mathbf{i},\mathbf{t}} \times I\_{\mathbf{i},\mathbf{t}'} \forall \mathbf{t} \in \mathcal{H}. \tag{14}$$

The Equation (14) gives average power obtained form PV array against specific temperature and irradiance over the given time *t*. Thus, the total obtained power as per dynamic temperature and irradiance can be calculated using [63–65]. Whereas, the rate of change in temperature with respect to the reference temperature is calculated as:

$$
\delta T = T - T\_r \tag{15}
$$

Equation (16) gives the rate of change in current due to variation in PV module tilt and temperature:

$$
\delta I = a \times \left(\frac{\mathcal{S}}{\mathcal{S}\_r}\right) \times \delta T + \left(\frac{\mathcal{S} - \mathcal{S}\_r}{\mathcal{S}\_r}\right) \times I\_{\text{sc}} \tag{16}
$$

Equation (17) is the change in voltage due to change in current during different time horizon:

$$
\delta V = -b \times \delta T - \mathcal{R}\_s \times \delta I \tag{17}
$$

This rate of change results in changing the value of *V* and *I*, which are given as:

$$V\_{new} = V\_r + \delta V \tag{18}$$

$$I\_{new} = I\_r + \delta I \tag{19}$$

The insolation and cell temperature level relation is defined as the sum of ambient temperature and two percent of the irradiance level and is given as under:

$$T = \frac{s}{50} + T\_A \tag{20}$$

The optimum number of PV modules required for the household maximum load shifted from grid to ESS at any time slot *t* is anticipated as: if *nκ<sup>t</sup>* < *PD*,*<sup>t</sup>* , increase the value of *n*. The quality of load matching depends on PV arrays, load characteristics, and isolation profile. Based on PV supply/demand match evaluation criteria, an optimum PV-site matching technique is used to select the module type with the highest average capacity and lowest cost. The current *I* of the PV module as a function of *V* can be expressed as [66]:

$$I = I\_{\rm sc} \left\{ 1 - \mathcal{g}\_1 \left[ e^{\left( \mathcal{g}\_2 V^n \right)} - 1 \right] \right\} \tag{21}$$

The values of constants, *g*1, *g*<sup>3</sup> and *n* are obtained from [63]. Finally, the total renewable energy *κtot*,*<sup>t</sup>* obtained from PV system is used in two ways; some amount *κa*,*<sup>t</sup>* is directly provided to fulfil the residential load demand, while, the remaining energy (*κsts*,*<sup>t</sup>* = *κtot*,*<sup>t</sup>* − *κa*,*t*) is stored in backup storage system. The relationship between these two amounts is given as:

$$\sum\_{t=1}^{\mathcal{H}} \left( \sum\_{a \in \mathbf{A}} \kappa\_{a,t} + \sum\_{s=1}^{N} \kappa\_{st\_{s,t}} \right) \le \kappa\_{tot,t} \tag{22}$$

In contrast to renewable energy, the net amount of energy obtained from grid at any time slot *t* ∈ H cannot exceed a certain level *P grid max* , which is given as:

$$\mathbb{P}\left(\sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} P\_{a,t} + P\_{c\_{\mathbf{s},t}}\right) - \kappa\_{\mathbf{s}l\_{\mathbf{s},t}} \le P\_{\max}^{grid} \quad \forall \ \sigma\_{a,t} = 1 \tag{23}$$

where, *Pcs*,*<sup>t</sup>* denotes the electrical charge in the battery power. Furthermore, the ESS must satisfy Equation (24) constraint for any type of load that is to be shifted from the grid to ESS:

$$
\kappa\_{\rm st\_{s,t}} - (P\_{a,t} \times \sigma\_{a,t}) \ge 0, \quad \forall a \in \mathbf{A}, \quad \forall 1 \le t \le \mathcal{H}. \tag{24}
$$

where, the remaining power obtained from PV generation must not exceed the ESS capacity:

$$P\_{\mathfrak{c}\_{\mathfrak{s},t}} - \kappa\_{\mathfrak{s}t\_{\mathfrak{s},t}} \ge 0, \quad \forall \, \mathbf{s}, \, t \in \mathcal{H}. \tag{25}$$

#### 3.2.2. ESS

As it is understood, ESS has the upper and lower levels to store the surplus energy, which is expressed as:

$$
\underline{\kappa L\_{s,t}} \le \kappa L\_{s,t} \le \overline{\kappa L\_{s,t}} \quad \forall \, 1 \le t \le \mathcal{H}. \tag{26}
$$

The total amount of surplus energy which can be stored in ESS at time *t*, which is greater than *κLs*,*<sup>I</sup>* is given as:

$$\mathbb{P}\left(\mathbb{x}\mathcal{L}\_{\mathsf{s},\mathsf{f}} = \mathsf{x}\mathcal{L}\_{\mathsf{s},\mathsf{f}-1} + \left(P\_{\mathsf{c}\_{\mathsf{s},\mathsf{f}}} \times \eta\_{\mathsf{s}}\right) - \frac{P\_{\mathsf{d}\_{\mathsf{s},\mathsf{f}}}}{\eta\_{\mathsf{s}}}\right) \geq \mathsf{x}\mathcal{L}\_{\mathsf{s},\mathsf{f}} \quad \forall \mathsf{f} \in \mathsf{\mathcal{H}}.\tag{27}$$

where, *η<sup>s</sup>* , *Pds*,*<sup>t</sup>* and *I* denote efficiency of battery *s*, discharge in power of the battery and pre-defined storage quantity, respectively. To prolong the battery life time, it is recommended to avoid deep discharge of a battery. Therefore, we defined upper and lower limits on battery charge and discharge rate, which are given as:

$$\underline{P\_{c\_{s,t}}} \le P\_{c\_{s,t}} \le \overline{P\_{c\_{s,t'}}} \forall t \in \mathcal{H}. \tag{28}$$

$$\underline{P\_{d\_{s,t}}} \le P\_{d\_{s,t}} \le \overline{P\_{d\_{s,t'}}} \,\forall t \in \mathcal{H}.\tag{29}$$

During the time when load demand exceeds a certain level and electricity tariff is also high, then ESS acts as "first choice" to fulfil the load demand. Therefore, ESS starts providing the energy as per schedules given by EMC and is given as:

$$\kappa\_{\mathfrak{sl}\_{s,t+1}} = \kappa\_{\mathfrak{sl}\_{s,t}} - P\_{d\_{s,t'}} \quad \forall \quad t \in \mathcal{H} \tag{30}$$

whereas,

$$P\_{d\_{s,t}} = \sum\_{a \in \mathbf{A}} (P\_{a,t} \times \sigma\_{a,t})\_{\prime} \quad \forall \quad t \in \mathcal{H} \tag{31}$$

The Equation (30) denotes the remaining stored energy in the ESS system. Whereas, the Equation (31) represents the amount of energy dissipated during time slot *t* for the load *a* ∈ **A** shifted from grid to ESS system.

#### 3.2.3. Evaluation of Hybrid Energy System

The daily load demand of the end user as a function of *t* and user preferences are shown in Table 2. End users are hoping to fulfil their demand using PV system, ESS and grid resources. Whereas, the choices to use these resources are mentioned above. Therefore, in order to manage these energy resources efficiently, with the objective of overall cost reduction and power system stability, autonomous energy management algorithm is needed. In this regards, this work considers all the objectives and limitations and designs an autonomous load scheduling algorithm(s). As the primary source of energy is PV and ESS system, therefore, the total PV modules required to fulfil the load demand and the energy obtained from the grid are calculated as:

$$j\_{pv} \times P\_{o,t} \times \phi\_{pv} = P\_{D,t} \tag{32}$$

$$P\_{\mathbb{S},t} \times \phi\_{\mathbb{S}} = P\_{\mathbb{D},t} \tag{33}$$

where, *jpv* denotes the total number of PV modules required for the energy demand *D*, *φpv* and *φ<sup>g</sup>* define efficiency of PV modules and grid system, and *Po*,*<sup>t</sup>* and *Pg*,*<sup>t</sup>* are power imported from single PV module and grid at time slot *t*, respectively. The equation formulated for a hybrid energy system is given as:

$$z \times P\_{o,t} \times \phi\_{pv} + (1 - z)P\_{\mathcal{g},t} \times \phi\_{\mathcal{g}} = P\_{\mathcal{D},t\prime} \tag{34}$$

where, *z* denotes the ratio which is used to define the amount of PV and grid energy. The best fit ratio *z* is anticipated by using LSM and is given as [64]:

$$\varpi\_{bf} = \left[ (\boldsymbol{\phi}\_{\mathcal{V}^{\mathcal{U}}} \times \boldsymbol{P}\_{\boldsymbol{o},\mathcal{I}} - \boldsymbol{\phi}\_{\mathcal{S}} \times \boldsymbol{P}\_{\boldsymbol{\mathcal{S}},\mathcal{I}})^{\boldsymbol{t}} (\boldsymbol{\phi}\_{\mathcal{V}^{\mathcal{U}}} \times \boldsymbol{P}\_{\boldsymbol{o},\mathcal{I}} - \boldsymbol{\phi}\_{\mathcal{S}} \boldsymbol{P}\_{\boldsymbol{\mathcal{S}},\mathcal{I}}) \right]^{-1} \times (\boldsymbol{\phi}\_{\mathcal{V}^{\mathcal{U}}} \times \boldsymbol{P}\_{\boldsymbol{o},\mathcal{I}} - \boldsymbol{\phi}\_{\mathcal{S}} \boldsymbol{P}\_{\boldsymbol{\mathcal{S}},\mathcal{I}})^{\boldsymbol{t}} (\boldsymbol{P}\_{\mathcal{D},\mathcal{I}} - (\boldsymbol{\phi}\_{\mathcal{S}} \times \boldsymbol{P}\_{\boldsymbol{\mathcal{S}},\mathcal{I}})) \tag{35}$$

It reduces the sum of squares of deviations of generated power and household load, which is expressed as [64–66]:

$$(\phi\_{\mathcal{V}^{\mathcal{V}}} \times P\_{o,t} - \phi\_{\mathcal{S}} \times P\_{\mathcal{S},t}) \times z = P\_{\mathcal{D},t} - (\phi\_{\mathcal{S}} \times P\_{\mathcal{S},t}) \tag{36}$$

Based on Equations (35) and (36), EMC takes decision about when to use ESS along with grid energy. For example, if 0 < *z* < 1, then it would be beneficial to usde ESS along with grid energy.

#### *3.3. Energy Consumption and PAR Model*

For each load *a* ∈ **A**, the energy consumed by load is given as:

$$E\_{a,t} = \sum\_{t=\gamma\_a}^{\delta\_a - \rho} \sum\_{a \in \mathbf{A}} P\_{a,t} \times \sigma\_{a,t} \times l\_{a\prime} \quad \forall \sigma\_{a,t} = 1. \tag{37}$$

Let the house have *N* number of loads, then the total daily energy demand *Etot*,*<sup>t</sup>* of a single household is written as:

$$E\_{tot,t} = \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} E\_{a,t} \times \theta\_{a,t} \,. \tag{38}$$

where, *ϑa*,*<sup>t</sup>* is the operation time of load *a* during time slot *t*. Generally, if we consider only the users' benefits, then the load scheduling has to be performed in a way that most of the load during high peak hours will be shifted to off-peak hours leading towards reduced billing cost. However, it may create rebound peaks which ultimately disturb power system stability regarding high PAR. In contrast, the effective load management mechanism has to reschedule the load to avoid the rebound peaks. Otherwise, the energy retailers and service providers keep backup generation capacity to protect electrical network. Eventually, this may lead costlier generation and hence increase the tariff price which would ultimately disturb end user benefits. Mathematically, the PAR is calculated as follows:

$$PAR = \frac{\overline{P}}{P\_{avg}}\tag{39}$$

where,

$$\overline{P} = \max\_{\mathbf{t} \in \mathcal{H}} \left( \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} P\_{a,t} \right) \tag{40}$$

and

$$P\_{avg} = \frac{\sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} P\_{a,t}}{\mathcal{H}}.\tag{41}$$

Equation (39) shows the ratio between the maximum power and average power consumed by household load *a* at a given time horizon H. An efficient energy consumption leads to a minimized PAR and thus our third objective is to minimize the maximum PAR over the given time slot.

#### *3.4. Cost Model*

The daily energy consumption cost of *N* household loads during time slot *t* ∈ H is calculated by EMC as:

$$\mathbf{C}\_{1,t} \stackrel{\scriptstyle \Delta}{=} \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} (\left[P\_{a,t} \times l\_{a,t}\right] \times \sigma\_{a,t}) \times \boldsymbol{\psi}\_{t\prime} \quad \forall \ \sigma\_{a,t} = 1. \tag{42}$$

where, *ψ<sup>t</sup>* is the electricity cost charged by the utility for the amount of energy consumed at time slot *t*. It is, therefore, reasonable to assume that the total cost anticipated by utility for load *a* ∈ **A** would become:

$$\mathbf{C}\_{2,t} = \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} \mathbf{E}\_{\mathrm{tot},t} \times \boldsymbol{\psi}\_t. \tag{43}$$

Now, the fair billing B is defined as:

$$\mathcal{B} \triangleq \frac{\sum\_{t=1}^{\mathcal{H}} \mathbb{C}\_{2,t}}{\sum\_{t=1}^{\mathcal{H}} \mathbb{C}\_{1,t}} \ge 1. \tag{44}$$

The fair billing is validated by the end user if B = 1, and if B > 1, then user's bill includes the cost E*cost* due the losses that occurred in between power distribution and smart meter at any time *t* and is calculated by:

$$\mathcal{E}\_{\text{cost},t} = \mathbb{C}\_{2,t} - \mathbb{C}\_{1,t}, \forall t \in \mathcal{H} \tag{45}$$

#### **4. Formulation of Multi-Objective Optimization Problem**

The objective of proposed model is to optimally schedule the household load over the given time period by taking into consideration respective constraints and limits. In this regard, the proposed algorithm is designed to optimally integrate PV system, ESS, and grid energy resources with the objective of cost and PAR reduction with increased comfort level. Hence, the primary objective is to fulfil the load demand through PV and ESS system without heavily relying on grid energy source.

$$P1 = \underset{t=1}{\text{minimize}} \{ \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} \{ [P\_{a,t} \times \theta\_{a,t}] - \kappa\_{tot,t} \} \}. \tag{46}$$

subject to:

$$0 \le \overline{\rho\_a} \le \delta\_a - \gamma\_a - l\_a \,\forall \; a \in \mathbf{A} \tag{47}$$

$$
\mathfrak{a}\_a \le \mathfrak{d}\_a \le \mathfrak{f}\_{a\prime} \,\,\forall \, a \in \mathbf{A} \tag{48}
$$

$$E\_{a,t} = \begin{cases} P\_{a,t\_\prime} & t\_a \in \delta\_a - \gamma\_a - \rho\_{a\prime} \,\,\forall \, a \in \mathbf{A}, \, t \in \mathcal{H}\_\prime \\\ 0, \quad \mathcal{H} \nmid t\_a \end{cases} \tag{49}$$

$$
\sigma\_{t,a} \in \{0, 1\} \tag{50}
$$

The constraints shown in Equations (47)–(50) define: (i) maximum load delay between starting and finish time while completing its task, (ii) load *a* must complete its assigned task within the allowed time (*α* and *β*), (iii) load *a* consumes certain amount of power during its operation time, and (iv) status of the load *a* (0 for idle mode, 1 for operation mode).

Therefore, the optimal management and use of the energy resources lead to effective cost minimization, because the electricity prices are dynamic in nature and usually known in advance to customers in a day-ahead market. However, prices are unknown to users in a real time pricing (RTP) environment, where it is more difficult to optimally manage the loads in conjunction with the integration of distributed energy resources (PV and ESS). Therefore, our next objective is to reduce the electricity cost through scheduling the load in optimal time slots without curtailing the load demand and is given as:

$$P2 = \underset{t=1}{\text{minimize}} \{ \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} \left( [P\_{a,t} \times \theta\_{a,t}] - \kappa\_{tot,t} \right) \times \psi\_t \}. \tag{51}$$

where, Equation (51) denotes the electricity cost charged by the utility against the amount of energy consumed by load *a* during time slot *t*. However, it is worth mentioning here that in order to avoid the situations when high peak can be generated due to turning-on high load during low pricing hours, a threshold is imposed to limit the maximum upper limit of energy consumption. Thus, an efficient energy consumption leads to a minimized PAR and thus our next objective is to minimize the maximum PAR over the given time slot *t* and is formulated as:

$$\text{P3} = \text{minimize}(\text{PAR}) \tag{52}$$

subject to:

$$\sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} P\_{a,t} \le \tau\_{\text{max}}.\tag{53}$$

where, *τmax* is the threshold value used to avoid different risks, such as blackout, use of expensive backup generation plants, and load shedding. Generally, each user has a desire to complete the tasks of each load with minimum delay subject to minimized electricity bill. Thus, our next objective is to minimize the load operation delay in order to achieve maximum end user comfort and is described as:

$$\text{AP4} = \underset{t = 1}{\text{minimize}} \{ \sum\_{t=1}^{\mathcal{H}} \sum\_{a \in \mathbf{A}} \rho\_{a,t} \}. \tag{54}$$

subject to:

$$
\rho\_{a,t} = \mathcal{H} - l\_{a\prime} \,\forall a \in \mathbb{C}, t \in \mathcal{H} \tag{55}
$$

$$
\rho\_{a,t} = (\beta\_a - a\_a) - l\_{a\nu} \forall a \in \mathbf{D}, t \in \mathcal{H} \tag{56}
$$

$$\rho\_{a,t} = 0, \forall \; a \in \{\mathbf{B}, \mathbf{E}\}, t \in \mathcal{H} \tag{57}$$

The constraints in Equations (55)–(57) define the allowed scheduling horizon for each load *a* ∈ **A**. Mathematically, the cost minimization objective function in conjunction with PV and ESS integration is written as:

$$P5 = \text{minimize}(P1 + P2 + P3 + P4).
\tag{58}$$

$$\text{subject to : Equations (7)-(11), (28), (29), (47)-(50)}$$

#### **5. Proposed Solutions**

In the following sections, we describe the working and coping of the algorithms used to solve the multi-objective optimization problem. For detailed analysis, we used heuristic and robust optimization algorithms and provided the merits and demerits on the basis of input and outputs.

#### *5.1. Optimal Load Scheduling Using DA*

DA is a greedy graph search algorithm that is used to find the shortest path for load scheduling problems in micro grid [56,67,68]. It calculates the shortest path (with lowest cost in our case) for a start point (source node; operation start time of load) to the end point (destination node; finish time of load) in the graph. The DA anticipates the shortest path, i.e., minimum cost from any specific source node *s*<sup>1</sup> to any node in a parent weighted graph such as; *G<sup>w</sup>* = (**F**, **e**), where **F** and **e** are the sets of vertices and edges, respectively. Initially, the algorithm calculates weights as it traverses from the source node to all the connected nodes and saves the best values. Then it moves through all other connected nodes to find the next best value. Meanwhile, it also avoids saving all the previously saved best value and calculates the cost by comparing with previously saved values. The weighting factor *x* is the sum of edges weights from the node *s*<sup>1</sup> to *d* and is given as:

$$\begin{aligned} \text{if} (u, f) \subset \mathbf{e} \to \mathbf{x} (u, f) > 0 \,\forall f, u \in \mathbf{F} \\ \mathbf{x} (f\_1, f\_2, \dots, f\_n) = \sum\_{i=0}^{i-1} \mathbf{x} (f\_i, f\_{i+1}) \end{aligned} \tag{59}$$

where, *x*(*f<sup>i</sup>* , *fi*+1) is the weight (cost) between *f<sup>i</sup>* and *fi*+<sup>1</sup> . The shortest path from *u* to *f*, *d*(*u*, *f*), in *G* has the minimum weight among all nodes. If there is no path between *u* and *f*, then *d*(*u*, *f*) = ∞.

$$\mathbf{S} = \{u | u \in \mathbf{F}\}$$

$$\{f \in \mathbf{F} \to f(u, f) \, is \, defined, \, \forall \, u \in \mathbf{S}\}\tag{60}$$

where, **S** is a subset of **F**, made-up gradually in an iterative process as follows: (*a*) start node *y* is selected from set **F** and inserted in **S**, (*b*) the shortest path from *y* to all available nodes in set **F** − **S** are determined as:

$$d(y, f) = d(y, u) + \min[\mathbf{x}(u, f)], \forall f \in \mathbf{F} \; \vert \; \mathbf{S} \tag{61}$$

where, *u* is the node between *f* and *y*. In the next step, the marked best fit solution *u* and the related edges to this node are added to set **S** and **e**, respectively. The algorithm continues until it updates all the nodes in set **S**, which means the final result is achieved. The time complexity of DA is (*O*(*N* + *N*(*N* − 1) + *N*(*N* − 1)(*N* − 2) + ... + *N*(*N* − 1)(*N* − 2)...(*N* − *N* − 1))). The pseudo-code of DA is shown in Algorithm 1.

#### **Algorithm 1** DSEM model based on DA.

**Require:** *a*, *b*,*s*, *d*, *g* 1: *d<sup>a</sup>* = 0, *g* ≥ 0, *P<sup>a</sup>* = ∞, ∀ *a* ∈ {1, 2, 3, ..., *N*} 2: *T<sup>c</sup>* = *mini*(*g*, 1), % *T<sup>c</sup>* → *currentTier* 3: *T<sup>c</sup>* + + 4: *T<sup>n</sup>* = *mini*(*g*, *Tc*), *TN<sup>c</sup>* = *d*1, *d*2, *d*3, ..., *dN*; % *TN<sup>c</sup>* → current TierNode 5: *TN<sup>n</sup>* = ∞ 6: Find *e* = *arg mini<sup>a</sup>* (*d da <sup>e</sup>* ) 7: Update P = P ∪ *d<sup>e</sup>* ; % *P* → *Path* 8: *TN<sup>n</sup>* = *TN<sup>c</sup>* − {*de*}; 9: **while** *TN<sup>c</sup>* 6= ∞ **do** 10: *TN<sup>c</sup>* = *TNn*; 11: *s* = *d<sup>e</sup>* ; %*d* → Destination 12: *e* = *arg mini*(*d da <sup>s</sup>* ); ∀ *d<sup>a</sup>* ∈ *TN<sup>n</sup>* 13: Update P = P ∪ *d<sup>e</sup>* ; 14: *TN<sup>n</sup>* = *TN<sup>c</sup>* − {*de*} 15: **end while** 16: *P<sup>r</sup>* is obtained; % *P<sup>r</sup>* → resultant Path

#### *5.2. Optimal Load Scheduling Using BPSO*

BPSO is a population-based heuristic optimization technique used in micro grid for load scheduling [69–71] in which each possible solution in entire search space is represented by a particle. Initially, each particle is assigned a random position and velocity. Each particle represents the status of the electrical load. In this work, a pattern of *n* number of particles is selected that represents the *n* number of household electrical loads. For example, if the household has 10 electrical loads then a pattern of 10 number of particles is selected to solve the problem. These particles then fly and move towards the optimal solution in the entire search space. Their flight of movement can be affected by local and global best positions.

$$l\_{\text{best}}(a, t) = \arg\_{t \in \mathcal{H}} \min \left[ f\_a(t) \right], \forall a \in \{1, 2, 3, \dots, M\}, \tag{62}$$

$$\log\_{best}(t) = \arg\_{\substack{a \in M \\ t \in \mathcal{H}}} \min \quad l\_{best}(a, t). \tag{63}$$

The velocity *g* of each particle is updated as [32]:

$$\mathcal{G}\_{i,t+1}(\boldsymbol{j}) = \sum\_{t=1}^{\mathcal{H}} \sum\_{i=1}^{M} \left( w g\_{i,t}(\boldsymbol{j}) + c\_1 r\_1(l\_{\text{best},i,t}(\boldsymbol{j}) - \sigma\_{i,t}(\boldsymbol{j})) + c\_2 r\_2(g\_{\text{best},i,t}(\boldsymbol{j}) - \sigma\_{i,t}(\boldsymbol{j})) \right), \forall \, r\_1, r\_2 \in [0, 1] \tag{64}$$

where, *σi*,*t*(*j*) is the *j*th element of *i*th particle in *t*th iteration of the algorithm and *r*<sup>1</sup> and *r*<sup>2</sup> are random variables. The pulls of the local and global best position are represented by constants *c*<sup>1</sup> and *c*2. The weight *w* of the particle momentum is mathematically defined as:

$$w = w\_i + \frac{\mathfrak{F}\_t}{\overline{\mathfrak{F}}} \times (w\_f - w\_i). \tag{65}$$

where,

$$\mathbf{g}\_{i,t+1} = \begin{cases} \overline{\mathbf{g}}\_{\prime} & \text{if } \; \overline{\mathbf{g}} < \mathbf{g}\_{i,t+1} \\ \underline{\mathbf{g}}\_{\prime} & \text{if } \; \mathbf{g}\_{i,t+1} \le \underline{\mathbf{g}}. \end{cases} \tag{66}$$

The position *λ* of each particle is updated as:

$$
\lambda\_{i,t+1}(j) = \begin{cases} 1, & \text{if } \text{sig}\left(g\_{i,t+1}(j)\right) > r\_{ij} \\ 0, & \text{otherwise}. \end{cases} \tag{67}
$$

where, sigmoid *sig* function is defined as:

$$\text{sig}\left(\mathbf{g}\_{i,t+1}(j)\right) = \frac{1}{(1 + \exp(-\mathbf{g}\_{i,t+1}(j)))}.\tag{68}$$

Equation (68) shows the sigmoid function and is used to map velocity of a particle coordinate with probability. The resulting probability determines whether the coordinates take values of 1's or 0's. Each particle is evaluated and ranked based on their fitness value. After *n*-iterations, the *gbest* is selected, which is known as best optimal solution. The *gbest* is the pattern of *N* bits representing the status of *N* number of household loads. The pseudo-code of BPSO is presented in Algorithm 2.

**Algorithm 2** Load management mechanism based on BPSO.

```
Require: p, i, s, l, ep, pos, g
 1: gbest = ∞ && lbest = zero
 2: for i = 1 to s do
 3: Generate population
 4: g ← rand(g, v)
 5: pos ← rand(s)
 6: lbest ← pos
 7: end for
 8: update g using Equation (64)
 9: Update pos using Equation (67)
10: for k = 1 to i do
11: gbest,i ← argmini (lbest,i
                              )
12: l − −
13: end for
14: gbest is obtained; % gbest → resultant best solution
```
## *5.3. Optimal Load Scheduling Using OPRA*

To develop an understanding of the performance of different load scheduling algorithms, we implement and solve load scheduling problem using OPRA. OPRA consists of *N* number of binary bits and each bit represents the status of the load *a* ∈ **A**, which are evaluated in each iteration using fitness function. Where, each binary bit represents the status of household load (ON or OFF) and 2 *N* number of possible combinations of *N* bits are generated to find best optimal solution. To find and verify best optimal solution, which is also known a global optimum, the fitness function against all the constraints is evaluated. The very initial pattern with minimum electricity cost is selected as the best pattern. This selection of patterns is shown in Figure 5. The algorithm stops working when the required optimal number of patterns from entire search space are obtained. Alternatively, we can also define total number of iterations as a stopping criteria. The pseudo-code of OPRA is shown in Algorithm 3.

### **Algorithm 3** Load management mechanism based on OPRA.

```
Require: ln, imax, ep, l and pa
 1: Generate the N bit pattern
 2: for h = 1 to H do
 3: Gbest ← ∞
 4: for k = 1 to 2
                  N do
 5: σ ← σ
             2
              N
 6: end for
 7: for t = 1 to T do
 8: Validate all constraints
 9: end for
10: for i = 1 to 2
                 M do
11: if f(σi) < f(Gbest) then
12: Gbest ← σi
13: else
14: Gbest,i ← Gbest,i
15: end if
16: la ← la − 1
17: end for
18: end for
```
**Figure 5.** Schematic diagram showing the selecting of optimal load patterns using OPRA algorithm.

#### *5.4. Optimal Load Scheduling Using GA*

GA algorithm is used to optimally schedule the household load [72–74] and generates the optimal patterns using objective function without violating the constraints. The algorithm usually adapts heuristics in the problem to provide the cost effective solutions. Initially, a population is randomly generated that consists of M number of chromosomes. Each chromosome represents a solution to the problem and is constructed as an array of bits (combination of genes). Each gene represents the ON/OFF status of the household electrical load. The length of chromosome is equal to the required number of bits that represent the household total electrical loads. These chromosomes are evaluated using fitness evaluation function and fitness of each chromosome is ranked, accordingly. After calculating the fitness value of each chromosome, the reproduction, crossover, and mutation operators are applied to form new off-springs from the previous population. The population of new off-springs from the existing ones have possibly higher fitness and this operation is known as crossover. The generation of new off-springs from the parent chromosomes is shown in Figure 6.

**Figure 6.** Generation of new off-springs from parent chromosomes through crossover operation.

Selection allocates more copies of those solutions with higher fitness values and thus imposes the survival of the fittest mechanism on the candidate solutions. To avoid same off-springs in the next population and pre-mature convergence, a little bit randomness needs to be added to avoid repetition and is demonstrated in Figure 7.

**Figure 7.** Single point mutation.

In this work, a single point is considered for crossover, the Roulette Wheel selection method for reproduction, and the single bit is flipped during mutation operation. After crossover and mutation, all the chromosomes are screened with the pre-defined constraints to filter out the best chromosomes among the population. The chromosomes with best fitness during *n* number of generations are selected as an optimal solution. The GA process is stopped after reaching *n* × *n* number of iterations. The pseudo-code of GA-based DSEM is shown in Algorithm 4.

**Algorithm 4** Load management mechanism based on GA.

```
Require: p, g, m, n, t, c, l, e, em, ec
 1: Generate initial population
 2: for i = 1 to t do
 3: best = arg mini(e)
 4: if best == 1 then
 5: e − −
 6: end if
 7: for j = 1 to l do
 8: m = arg mini(e, t)
 9: end for
10: if c(i) < cmax then
11: RE = RE − c(i)
12: else
13: e(i) = eg
14: end if
15: do selection process
16: if rand < ei
                 then
17: do crossover
18: end if
19: if rand < em then
20: do mutation
21: end if
22: end for
```
#### *5.5. Feasibility Region*

A set of optimal points that contains all the possible solutions satisfying all the constraints, equalities and in-equalities in a given scenario is known as feasible region. The main objective of the DSEM is to maximize the end user comfort with minimized electricity cost and PAR. Thus, electricity cost is comprised of two components, i.e., amount of energy consumed during certain time slot *t* and electricity price at that time. Four scenarios are considered to find the feasible region for energy

consumption and electricity cost and are given as: (i) S1: Minimum load, minimum electricity price, (ii) S2: Minimum load, maximum electricity price, (iii) S3: Maximum load, minimum electricity price and (iv) S4: Maximum load, maximum electricity price. The per hour electricity cost charged by the utility against any time slot is anticipated as:

$$\mathbb{C}\_{t} \triangleq \sum\_{a \in \mathbf{A}} P\_{a,t} \times \psi\_{t\prime} \quad \forall \; \sigma\_{a,t} = 1 \; \; h \; \in \; \mathcal{H} \,. \tag{69}$$

subject to:

$$\mathcal{C}1: 0 \le \mathcal{C}\_{\mathfrak{l}} \le 64 \quad \forall \mathfrak{t} \in \{1, 2, 3, \dots, \mathcal{H}\} \tag{70}$$

The daily electricity cost is calculated using Equation (42) and minimized with subject to the following constraints:

$$\mathsf{C2} \colon \mathsf{C}\_{1,t} \le \mathsf{232} \tag{71}$$

$$\mathcal{C}3: 0 \le E\_{a,t} \le 9\tag{72}$$

The feasible region of the electricity cost against the consumed energy subject to the constraints (C1, C2 and C3) is demonstrated in Figure 8. In Figure 8, the overall region for the load operation is an area covered by points *p*1(2, 4), *p*2(2, 16), *p*3(9, 18), *p*4(9, 72), *p*5(9, 64) and *p*6(8, 64). The points *p*<sup>1</sup> and *p*<sup>2</sup> represent the scenario *S1* and *S2* and *p*<sup>3</sup> and *p*<sup>4</sup> represent the scenario *S*3 and *S*4. This leads to the result that the scheduling horizon must not: (i) exceed the maximum unscheduled cost (232 cents) and (ii) violate the constraints given in Equations (70)–(72). After implementing this threshold level, feasible region of the load scheduling is shown by the pentagon of points *p*1, *p*2, *p*3, *p*5, and *p*6.

**Figure 8.** Feasible region of Energy consumption.

#### **6. Simulation Results and Discussion**

In this section, we analyse the performance of proposed model using different algorithms in terms of managing grid energy, electricity cost, comfort and PAR minimization. The performance analysis is done in software tool MATLAB version 2014b on Sony VAIO VPC-EB47GM machine with Intel(R) Core(TM) i5 CPU M 480 2.67 GHz and 6 GB of memory on Windows platform. Furthermore, the scalability of the algorithms in terms of total number of time slots and computation time is also measured, so as to analyze the optimality. We verify the applicability of proposed algorithms through optimally managing the energy consumption by finding the tuple of time shifts of various loads. In a result, minimum cost and PAR objectives are achieved along with high user comfort. For experiments, the electricity price data is obtained from a day-ahead market showing real time consumption trends, which is shown in Figure 9.

**Figure 9.** RTP scheme obtained from a day-ahead electricity market.

#### *6.1. Simulation Setup*

Table 1 provides the parameters of a single household, which are used for experimentation purpose. Generally, each user may have a variable load pattern, on the basis of their comfort level.


**Table 1.** Simulation parameters.

For simulation, we considered four different type of loads that are used in different seasons. In the proposed work, each user has 11 fixed loads with different power rating, operation time and comfort requirements. The total household load is managed and operated to complete the assigned task as under:



The power rating and load demand are summarized in Table 2.

**Table 2.** Load categorization and user comfort level.

#### *6.2. User Comfort*

The comfort level is defined as the services required by the end user to fulfil their load demand with reduced electricity cost and operating delay time, and is demonstrated in Table 2. Based on end user comfort requirement, each load has a predefined start and finish operation time horizon. In this work, the end user defines its comfort level by defining certain delay level in the operation of household load. EMC needs to schedule the load in the pre-defined window to complete the desired task while achieving the aforementioned objectives.

From Table 2, the *l* of load 4, 5 and 6 are 3 time slots each, which lies in semi-delay tolerant category. In the absence of DSEM architecture, each load is assigned *t*5, *t*<sup>6</sup> and *t*<sup>7</sup> time slots for operation and is shown in Figure 10a. The user comfort is also a major objective of our proposed work and is demonstrated in Figure 10b–d. It is evident from Figure 10b–d that our proposed model efficiently scheduled the end user loads in the allowed time horizon and achieved maximum user comfort. The operation time assigned by each optimization algorithm is tabulated in Table 3.

**Figure 10.** User comfort profiles of different loads over the given time period.


**Table 3.** DSEM based semi-delay tolerant load scheduling.

#### *6.3. Energy Consumption Profile*

Regarding energy consumption, three sets of evaluations are carried out to examine the performance of the proposed DSEM and optimization algorithm.


In the first case, the household loads are operated using grid energy only while ignoring the electricity tariff. It is clear from Figure 9 that the electricity tariff is at peak from 10:00 to 20:00. For simulation, we considered 10 time slots, i.e., *H* = *t*1, *t*2, *t*3, ..., *t*8, *t*9, *t*<sup>10</sup> to schedule the loads. The load demand in different time slots under the ZSM is tabulated in Table 4. It is observed in Table 4 that the household selected the loads to operate during some time slots (*t*3–*t*8) rather than scheduling in the entire allowed time horizon (*t*1–*t*10). From Table 4, it is also clear that most of the high loads are operated during high peak hours (*t*5–*t*7) which will ultimately result in high electricity cost.


**Table 4.** Load demand of different loads for ZSM.

However, in the second case, the loads are operated under DSEM and optimization algorithm. The energy consumed by ZSM and FSM is shown in Figure 11. BPSO meets the DA at some most costly time slots while the other two schemes exhibit different scheduling plans.

**Figure 11.** Energy Consumption profiles using different optimization techniques over the given scheduling time period.

For the highest tariff slots, i.e., 4, 5, 6 and 7, DA performed significantly better in comparison with other optimization algorithms. It can be seen in Figure 11 that DA managed the entire load in the low and off peak hours. From Figure 9, there are 10 peak hours starting from 10:00 to 20:00, i.e., electricity price is high during these time slots. It is clear from Figure 11 that FSM has optimally shifted the load from high peak hours to low peak hours. In case of FSM equipped with PV generation and storage system, the load during high peak hours are shifted from grid to ESS and draws energy from the ESS.

In this work, the storage capacity ranges from 3 KWh to 4 KWh depending upon the solar irradiance level and status of the battery cells. The health of the battery is good when it is kept at near 100% of its capacity or to be charged immediately as it discharges completely or partially. The output power from a PV array is directly proportional to the solar irradiance received from the sun. The output of the PV module starts from zero during sunrise, increases till noon and declines to zero till evening. The rated maximum output of the PV module depends on the actual site condition and is achieved only occasionally. In our scenario, smart home is connected to gird and PV, daily sun peak hours can be used for calculation and the number of sun peak hours is thus numerically equal to the daily solar irradiance measured in KWh/m<sup>2</sup> . The amount of usable energy harvested from PV modules is lower than the output of the PV modules due to the energy losses in the system components. The solar radiation observed greater than the 250 W/m<sup>2</sup> is taken in simulation. The solar irradiance and the ambient temperature for a typical day of June is shown in Figure 12. In our case, PV modules are selected using Module-site matching technique with an assumption of 80% efficiency, and parameters of the PV modules are tabulated in Table 5.

The ESS is exploited by the proposed DA and the effectiveness of the model is shown in Figure 13. It is evident from Figure 13 that the proposed model optimally used the grid and stored energy. The load that needs to be operated during high peak hours on grid energy is shifted to ESS by EMC thus results in optimal use of both energy resources.

**Figure 12.** Solar irradiance and ambient temperature profiles over the period of 24 h.


**Table 5.** Parameters of PV module at standard test conditions.

**Figure 13.** Energy consumption profile against different optimization techniques with the integration of ESS.

#### *6.4. Cost Profile*

The cost against the amount of energy consumed by the household loads and the performance of each optimization technique is shown in Figure 14. According to the electricity tariff, the highest cost needs to be paid against the time slots 5 and 7. It is evident from the results that DA outperformed all the traditional techniques by shifting all the loads to other time slots except the peak cost slots. Similarly, BPSO exhibits better results in comparison with OPRA and GA and is demonstrated in Figure 14.

**Figure 14.** Total cost incurred over the given time period using different optimization algorithms.

#### *6.5. Scalability Factor and Computation Time*

The scalability factor of the proposed schemes are evaluated in terms of:


In the first case, the performance of each optimization algorithm is evaluated based on increasing the number of time slots for the load given in Table 2. By increasing the number of time slots and following the RTP pricing scheme, more scheduling opportunities are available for each algorithm. This higher degree of freedom for each optimization algorithm results in better scheduling the loads as shown in Figure 15. It is clear from Figure 15 that DA-based DSEM performed better under wide range of scheduling horizons.

**Figure 15.** A comparison between energy consumption cost and total number of time slots against optimization algorithms (i.e., scalability w.r.t number of time slots).

In the second case, simulation is done for a fixed number of scheduling slots, i.e., 10 while the number of loads vary from 27 to 100. The total cost increases as the number of loads increases (greater the load, greater the energy consumption) and the result is shown in Figure 16.

**Figure 16.** A comparison between energy consumption cost and total number of time slots against optimization algorithms (i.e., scalability w.r.t number of time loads).

The results demonstrated in Figure 16 shows that the DA outperformed all the candidate solutions followed by BPSO. Thus, we summarize that our optimization schemes provide the low cost solutions irrespective of the number of scheduling time slots and loads.

Simulation is run for two different scenarios in order to calculate the computational time for each algorithm. In the first scenario, a fixed number of loads (30) with varying time slots (50 to 300) is considered and the computation time taken by each optimization technique is shown in Figure 17.

**Figure 17.** A comparison between computation time and total time slots.

In the second case, iteration number is fixed to 100 and number of load varies from 30 to 65. The computation time taken by each algorithm is depicted in Figure 18.

**Figure 18.** A comparison between computation time and total number of loads.

It is clear from Figure 17 that the OPRA outperformed all the candidate solution techniques and scheduled the load with minimum execution time. From Figures 17 and 18, BPSO took less computation time in comparison with DA due to its low complexity factor, i.e., *O*(*I* × *N*). From Figure 18, the complexity of DA increases as the number of nodes (loads) increases. The entire network has N different costs, i.e., each node in different tier has a different cost. The distance to reach a node in any tier does not depend on the previous starting node. One node will be selected in one tier that finds the next and current distance from the starting node. Thus, DA takes a bit high execution time due to its highest complexity nature (*O*(*N* + *N*(*N* − 1) + *N*(*N* − 1)(*N* − 2) + ... + *N*(*N* − 1)(*N* − 2)...(*N* − *N* − 1))).

#### *6.6. PAR Discussion*

The relationship between the ZSM and FSM with respect to the PAR is shown in Figure 19. Simulation is run for a fixed time slots of 24 hours with total number of loads varying from 11 to 20 each day.

**Figure 19.** PAR profiles of different algorithms over the period of one week.

The purpose of minimizing PAR is a two-way concept, i.e., for utility and end user. With utility prospects, the lifetime of the grid is increased as the demand curve always remains under the supply curve. The amount of energy produced by utility during any time slot is sufficient for the users demand thus resulting in no activation of peak power plants. On the other hand, the end user consumption is limited to a certain threshold level. This threshold level helps the users in getting the desired energy level from the utility with no load shedding and blackouts issues. With the addition of ESS, it is clear from Figure 19 that PAR is minimized as compared to the case where there is no ESS.

#### *6.7. Overall Analysis of the Proposed Schemes*

To complete the discussion, the proposed schemes are analysed based on: (i) maximizing user comfort, (ii) optimally consuming grid energy, (iii) minimizing energy cost, (iv) minimizing PAR and (v) optimally usage of ESS. In the first scenario, cost comparison is done by scheduling the load (*l*<sup>1</sup> − *l*11) during high peak hours, i.e., 10:00–20:00. The results obtained from the simulation are tabulated in Table 6.


**Table 6.** A comparison of the proposed algorithm with conventional algorithms in terms of performance parameters.

The results show that DA has considerably better results for all the aforementioned objectives. DA optimally managed the grid and ESS energy in comparison with ZSM and thus leads to low electricity cost and maximum user comfort level. Keeping the trend, BPSO meets the DA in some case (FSM with DSEM) while OPRA and GA provide different patterns each time. It is evident from Table 6 that DSEM model based on DA has efficiently scheduled the load and reduced the end user electricity cost by 44.61% and 51.72% in the presence of only DSEM architecture and DSEM along with ESS, respectively. The cost reduction between the DA and the next best optimization technique, i.e., BPSO, is 9.7% and 6.89% in case of only DSEM architecture and DSEM along with ESS, respectively.

In the second scenario, simulation is run for 24 h, 7 days per week to analyse the PAR response and results obtained are tabulated in Table 6. It is evident from the Table 6 that DA efficiently reduced the PAR by a factor of 7.49% and 10.22% in case of DSEM and DSEM with ESS, respectively. The overall trade-off among all optimization schemes are given in Table 6. Thus, we summarize that our proposed DSEM based on DA provides low cost and PAR solutions independent of the number of scheduling time slots and loads.

#### **7. Conclusions and Future Work**

A novel DSEM model and EMC based on DA were proposed and implemented for the optimal energy management of residential user in this paper. The novel aspect of this work is the optimal selection of PV arrays and ESS, based on which the optimal load scheduling algorithms have been designed. Prior to this, we categorized the load into base, schedulable, semi-schedulable and critical according to user preferences and assigned predefined priorities in order to provide operation flexibility. Then, these loads were mathematically modelled in such a way that optimization algorithm can have

better control. EMC is supported by optimum PV module selection and use scheme that efficiently shifts the high cost load from grid to an on-site PV and ESS. The cost reduction objective is achieved through optimal use of grid, PV and ESS sources, where PV units and ESS act as a "primary source" of energy during peak hours. The intuition behind our model is to address the optimal management of base, semi-schedulable, schedulable, and critical energy demands, time varying price signal, optimal selection and integration of PV modules, energy harvesting from PV modules and use of stored energy to maximize comfort level and minimize the end user electricity cost and PAR. Simulation results show that EMC based on DA achieved the maximum comfort level and also optimally scheduled the load, minimized the electricity cost and PAR in comparison with BPSO, OPRA and GA. An evaluation of the proposed algorithm in comparison with other algorithms is done based on an increased number of time slots and household loads and the presented results validates that the proposed algorithm leads to better solution. In addition, incorporating PV-based energy generation and storage system in the HEM architecture leads to better synergy between electricity consumption and PV production. The result is that the end user electricity cost is reduced by 52% and grid stability is increased by minimizing the PAR to 10.22% in comparison with ZSM. In future, we will update this model with other modules such as: (i) integration of windmill energy generation system, (ii) an energy exchange model among several micro smart grids, (iii) cloud-based infrastructure to connect micro smart grids, (iv) join the grid using Blockchain technology, (v) efficient storage system and (vi) facilitating electrical vehicles.

**Author Contributions:** Conceptualization, I.U. and M.B.R.; formal analysis, M.B.R. and T.A.; funding acquisition, M.B.R. and T.A.; investigation, I.U., M.B.R. and T.A.; methodology, I.U., M.B.R. and T.A.; project administration, M.B.R. and T.A.; software, I.U. and M.B.R.; supervision, M.B.R., T.A. and S.T.; validation, I.U., M.B.R. and T.A.; writing—original draft, I.U. and M.B.R.; writing—review and editing, I.U., M.B.R. and S.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This Project was funded by the Deanship of Scientific Research (DSR), King Abdulaziz University, Jeddah, under grant No. (DF-397-135-1441). The authors, therefore, gratefully acknowledge DSR technical and financial support.

**Conflicts of Interest:** The authors have no conflict of interest.

#### **Abbreviations**


#### **Nomenclature**


### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **An Intelligent Approach to Active and Reactive Power Control in a Grid-Connected Solar Photovoltaic System**

**Ibrahim Alsaidan <sup>1</sup> , Priyanka Chaudhary 2,3,\*, Muhannad Alaraj <sup>1</sup> and Mohammad Rizwan 1,2**


**Abstract:** The increasing demand of electrical energy and environmental concerns are invigorating the use of renewable energy resources for power generation. Renewable energy resources can provide an attractive solution for present and future energy requirements. In this scenario, solar photovoltaic systems are becoming prominent and sustainable solutions with numerous advantages. However, the utilization of solar photovoltaic systems in distribution generation makes it mandatory to deploy efficient and organized control measures for integrating solar photovoltaic plants with the grid. In this paper, the control of grid-tied solar photovoltaic systems using a Kalman filterbased generalized neural network is presented with a variable step size perturb and observe-based maximum power point tracking controller to extract the maximum power from a solar photovoltaic plant. The presented system provides power-quality enhancement and supports a three-phase AC grid. The proposed approach extracts the load currents' primary components for efficient harmonics elimination, synchronizes the system with the grid and provides a fast response during rapidly changing conditions. The results of the proposed control technique are also compared with the artificial neural network-based control technique for validation purposes. The proposed algorithm is found more suitable for using a smaller number of unknown weights and training patterns with reduced computational time.

**Keywords:** renewable energy resources; grid integrated solar PV systems; sustainable power generation; maximum power point tracking; grid reliability and voltage source converter

### **1. Introduction**

The increasing concern regarding greenhouse gas emissions and the depleting nature of conventional fuels has invigorated the use of renewable energy resources as an alternative and sustainable solutions for the power sector. Solar, wind, biomass and small hydro power are the main renewable energy-based resources which can fulfill the future energy requirements. The solar photovoltaic system (SPV) is more encouraged due to various advantages such as abundance of availability, less environmental pollution, reduced cost and many others. The progress of an SPV system in the existing power system is witnessed for its progress [1]. The integration of an SPV system into the grid causes positive effects, i.e., generating more power, along with some negative effects, i.e., the violation of voltage limitations at common coupling, frequencies disruption and grid stabilization problems, etc. There are set guidelines, codes and standards for grid-connected SPV systems. These standards include IEEE 1547, IEC 61727 and VDE-AR-N4105 [2]. The use of these standards is encouraged and imposed to maintain the stability and power quality related to the grid. The large-scale exploitation of distributed generation make it mandatory to deploy efficiency and organized control measures for integrating and measuring problems. The control algorithms are helpful for the accommodation and facilitation of the integration of an SPV in the distribution grid. Single-stage and two-stage systems have been used for

**Citation:** Alsaidan, I.; Chaudhary, P.; Alaraj, M.; Rizwan, M. An Intelligent Approach to Active and Reactive Power Control in a Grid-Connected Solar Photovoltaic System. *Sustainability* **2021**, *13*, 4219. https://doi.org/10.3390/su13084219

Academic Editors: M. Dolores Esteban, José-Santos López-Gutiérrez and Vicente Negro

Received: 6 March 2021 Accepted: 30 March 2021 Published: 10 April 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

the integration of SPV systems into the grid, as shown in previous works. It has also been shown by various researchers that for a three-phase system, a single-stage configuration is more advantageous [3]. However, conventional single-stage SPV systems suffer from the drawback of converters being idle during the unavailability of power from the SPV system. To overcome the abovementioned problems, the proposed system is designed in such a manner so that it operates as a distribution static compensator (DSTATCOM) while SPV power is not available and restores the operation after recovering the SPV power generation.

The electrical power distribution system has been designed to distribute active power pertaining a smaller number of harmonics to the consumers. DSTATCOM integrated as a shunt compensator with a distribution system effectively controls harmonics, fulfils the demand of reactive power of consumers and improves voltage regulation with balanced and unbalanced loads [4]. The researchers have reported various configurations of DSTAT-COM in the literature [5–8]. For efficiently integrating distributed a generation system that compensates for the reactive power and harmonics for utility grids at common coupling points, several converter topologies [9–11] and control strategies [12–18] were reported in the previous work. Additionally, Jain et al. [19] presented a grid-integrated solar energy conversion setup with a movable direct current link voltage to vary the voltages at common pairing points with two-stage circuitry methodology and an associated double-frequency second-order generalized integrator (DFSOGI)-dependent control strategy for controlling the multi-functional voltage source converter (VSC) in abrupt load modification at common connecting points.

Varma et al. [19] have proposed a solar photovoltaic static compensator for improving the power transfer capabilities of the system to transmit real power to utilities with an existing converter system. Humid et al. [20] proposed a strategy that depends on a power conditioning unit located parallel to the plant that works in feed-forward mode to compensate the distorted current photovoltaic output to decrease the harmonic of current from a PV system. Kannan et al. [21] presented icos Φ control technique for a distribution static compensator that gives uninterrupted harmonic degradation, compensates re-active power and compensates the loads along with comparing performances of fuzzy-logic controllers to ordinary proportional integral (PI) controllers. Mishra et al. [22] implemented an optimistic control technique which optimizes the proportional integral's coefficients and photovoltaic fed distribution static compensator's filtration parameters.

To enhance the performance of grid the integrated SPV plant, various control techniques were adopted by the researchers. There are some performance indicators such as less computational time, high accuracy and less complexity, etc. A huge number of techniques depending upon artificial neural networks (ANN) have been widely adopted by a large number of journalists for distribution static compensators implementing dynamic loading [23–27], whereas photovoltaic fed distribution static compensator systems have been adopted by very few. Singh et al. [28] presented a grid interfaced SPV generation plant implementing a neural network-based control technique that utilizes the least mean square (LMS) algorithm, termed adaline (adaptive linear element), for estimating the reference source current. Artificial neural network-based approaches are complex and need large computational times. Further, generalized neural network (GNN)-based control strategies can be used to overcome the abovementioned issues. Applications of GNN are widely available in load forecasting, solar irradiance/energy forecasting, load frequency control in power systems, power system stabilizers, electrical machines and control system-related problems [29–36]. There are few works are reported in the literature related to the application of GNN as a control strategy for grid-tied SPV systems. GNN-based models reduce the training time as well as improve the performance of the system. Considering this, a GNN-based control strategy has been developed for a PV DSTATCOM system in the present work.

The converting capability of a solar photovoltaic plant is comparatively small; thus, to increase the capability of solar photovoltaic plants, it is essential to keep a track of the maximum

power point. Maximum Power Point Tracking (MPPT) is complicated because of the non-linear character of SPV plants due to changing meteorological factors, i.e., solar irradiance, ambient temperature, wind velocity and relative humidity, etc. A lot work has been reported in the literature related to MPPT, including techniques such as perturb and observe, incremental conductance, constant voltage, open-circuit voltage, short-circuit current, extremum seeking control and hybrid, etc. Further, there are a number of intelligent techniques also available for MPPT such as artificial neural networks (ANN), fuzzy logic, genetic algorithms, etc. [37–39]. A changeable step size perturb and observe (P&O) MPPT technique is utilized in this research to track the MPP of a solar photovoltaic plant.

Further, to improve the performance of the system, an extended Kalman filter (EKF) [40–42]-based method is adopted to estimate and update the weights of the GNN model. The main contributions of this work are:


The developed system functions very well with an EKF GNN-based approach and gives a very fast response. Moreover, single-stage topology is able to reduce the losses in semiconductor devices and increase the overall efficiency. The proposed control approach performs with more flexibility in training the network under dynamic conditions. The proposed technique possesses various advantages such as its speed, using a single layer, less mathematical calculations and easy implementation on hardware.

The organization of paper is as follows: Section 2 provides details about the system configuration, followed by Section 3 which explains the developed control technique. The simulation and hardware results are provided in Section 4. The conclusion is given in Section 5, followed by acknowledgements and references.

#### **2. System Description**

The proposed system deploys a single-stage circuit topology and consists of a solar PV array, voltage source converter (VSC), ripple filter, loads and a three-phase utility grid, as shown in Figure 1. A solar PV array is a combination of several PV modules in a series and parallel according to the requirements. A PV cell serves as the basic unit for a solar PV module. A single diode solar PV cell is considered for this system, and mathematical modeling is given [43]. The developed system has been implemented with a minimum number of sensors with a variable step size perturb and observe (P&O) MPPT algorithm to achieve maximum power point operation of the solar PV system for different meteorological parameters. An extended Kalman filter (EKF)-based GNN methodology is used for controlling the VSC switching to synchronize the system with the grid and the main task of separating the weights of the primary real and reactive parts of three-phase load currents to obtain reference grid currents. The developed system design specifications are given below:

capacitor (CDC) = 10 mF, DC link voltage (VDC) = 700 V.

**3. Extended Kalman Filter-Based GNN Control Algorithm**  *3.1. Maximum Power Point Tracking Control*  A variable step size perturb and observe (P&O)-based MPPT algorithm is employed in this research for the extraction of maximum power from the SPV system. The main Grid description = three-phase, 415 Volts, 50 Hz, solar PV array ratings = (VMPP) = 700 Volts, current (IMPP) = 13.5 Amp, solar photovoltaic power at maximum power point (PMPP) = 10 kW, rating of interfacing inductors (Lfa = Lfb = Lfc) = 2.6 mH, rating of DC link capacitor (CDC) = 10 mF, DC link voltage (VDC) = 700 V.

Grid description = three-phase, 415 Volts, 50 Hz, solar PV array ratings = (VMPP) = 700 Volts, current (IMPP) = 13.5 Amp, solar photovoltaic power at maximum power point (PMPP) = 10 kW, rating of interfacing inductors (Lfa = Lfb = Lfc) = 2.6 mH, rating of DC link

#### operation of the given algorithm is dividing the SPV system's dPpv/dVpv curve into three **3. Extended Kalman Filter-Based GNN Control Algorithm**

#### independent regions. Region 0 denotes the SPV system power's closeness to the highest *3.1. Maximum Power Point Tracking Control*

power point. Region 0 uses the normal value of tracking step size, whereas region 1 and region 2 needed a larger value of step size as compared to region 0 for achieving the high tracking speed. Figure 2 explains the performed functions of the proposed technique. ∆Vref0, ∆Vref1 and ∆Vref2 represent the step size of tracking for region 0, region 1 and region 2, respectively. A variable step size perturb and observe (P&O)-based MPPT algorithm is employed in this research for the extraction of maximum power from the SPV system. The main operation of the given algorithm is dividing the SPV system's dPpv/dVpv curve into three independent regions. Region 0 denotes the SPV system power's closeness to the highest power point. Region 0 uses the normal value of tracking step size, whereas region 1 and region 2 needed a larger value of step size as compared to region 0 for achieving the high tracking speed. Figure 2 explains the performed functions of the proposed technique. ∆Vref0, ∆Vref1 and ∆Vref2 represent the step size of tracking for region 0, region 1 and region 2, respectively.

*Sustainability* **2021**, *13*, 4219 5 of 24

**Figure 2.** Flowchart of variable step size perturb and observe (P&O) algorithm. **Figure 2.** Flowchart of variable step size perturb and observe (P&O) algorithm. A generalized neural network works like a multi-layer feed-forward network, where

#### *3.2. Extended Kalman Filter-Based GNN Control Algorithm 3.2. Extended Kalman Filter-Based GNN Control Algorithm* every node implements a precise function on all signals that come to the node, and the

A generalized neural network works like a multi-layer feed-forward network, where every node implements a precise function on all signals that come to the node, and the parameters are set referring to the node. Figure 3 shows an aggregate-type format of a GNN model, with Σ and Π being the aggregated functions. Σ is an aggregated function, which is adopted with the sigmoidal characteristic function f1, whereas the Π aggregation function is adopted with the Gaussian function f2. The derivation of active and re-active load current components is performed by assuming load current (iLa, iLb, iLc) as the input to the summation (ΣA) and product (Π) neurons having undefined weights of (WΣi) and (WΠi), respectively. A generalized neural network works like a multi-layer feed-forward network, where every node implements a precise function on all signals that come to the node, and the parameters are set referring to the node. Figure 3 shows an aggregate-type format of a GNN model, with Σ and Π being the aggregated functions. Σ is an aggregated function, which is adopted with the sigmoidal characteristic function *f* <sup>1</sup>, whereas the Π aggregation function is adopted with the Gaussian function *f* <sup>2</sup>. The derivation of active and re-active load current components is performed by assuming load current (iLa, iLb, iLc) as the input to the summation (ΣA) and product (Π) neurons having undefined weights of (WΣ<sup>i</sup> ) and (WΠ<sup>i</sup> ), respectively. parameters are set referring to the node. Figure 3 shows an aggregate-type format of a GNN model, with Σ and Π being the aggregated functions. Σ is an aggregated function, which is adopted with the sigmoidal characteristic function f1, whereas the Π aggregation function is adopted with the Gaussian function f2. The derivation of active and re-active load current components is performed by assuming load current (iLa, iLb, iLc) as the input to the summation (ΣA) and product (Π) neurons having undefined weights of (WΣi) and (WΠi), respectively.

(**a**)

(**a**)

**Figure 3.** *Cont.*

**Figure 3.** (**a**) Summation-type generalized neural network (GNN) model, (**b**) a summation-type GNN structure to determine the fundamental active element of a load current. **Figure 3.** (**a**) Summation-type generalized neural network (GNN) model, (**b**) a summation-type GNN structure to determine the fundamental active element of a load current.

#### 3.2.1. Estimation of Amplitude of Terminal Voltage and Unit Templates 3.2.1. Estimation of Amplitude of Terminal Voltage and Unit Templates

With the help of sensed line voltages (*vab*, *vbc*, *vca*) at the point of common coupling (PCC), the amplitude of phase voltages (*va*, *vb* and *vc*) are calculated as follows: With the help of sensed line voltages (*vab*, *vbc*, *vca*) at the point of common coupling (PCC), the amplitude of phase voltages (*va*, *v<sup>b</sup>* and *vc*) are calculated as follows:

$$v\_a = \frac{2v\_{ab} + v\_{bc}}{3}, \ v\_b = \frac{-v\_{ab} + v\_{bc}}{3}, \ v\_c = \frac{-v\_{ab} - 2v\_{bc}}{3} \ \Sigma \tag{1}$$

The terminal voltage amplitude at PCC can be estimated as The terminal voltage amplitude at PCC can be estimated as

$$W\_t = \sqrt{\left[\frac{2(v\_a^2 + v\_b^2 + v\_c^2)}{3}\right]}\tag{2}$$

$$
\mu\_{pa} = \frac{v\_a}{V\_t}, \mu\_{pb} = \frac{v\_b}{V\_t}, \mu\_{pc} = \frac{v\_c}{V\_t} \tag{3}
$$

(3)

<sup>2</sup> (4)

Further, quadrature unit templates can be calculated using in-phase unit templates as Further, quadrature unit templates can be calculated using in-phase unit templates as

$$u\_{qa} = \frac{u\_{pc}}{\sqrt{3}} - \frac{u\_{pb}}{\sqrt{3}}, u\_{qb} = \frac{\sqrt{3}u\_{pa}}{2} + \frac{u\_{pb} - u\_{pc}}{2\sqrt{3}}, u\_{qc} = \frac{u\_{pb} - u\_{pc}}{2\sqrt{3}} - \frac{\sqrt{3}u\_{pa}}{2} \tag{4}$$

3.2.2. Terminal Voltage Amplitude and Unit Templates

(Vୢୡ

3.2.2. Terminal Voltage Amplitude and Unit Templates The DC link voltage (Vୢୡ) is sensed and compared with the reference DC link voltage The DC link voltage (*Vdc*) is sensed and compared with the reference DC link voltage (*V* ∗ *dc*) in order to determine the active loss component.

$$V\_{dcc}(k) = V\_{dc}^\*(k) - V\_{dc}(k)\tag{5}$$

k is the number of iterations. Further, the error (ௗ ) of reference Vୢୡ <sup>∗</sup> and sensed Vୢୡ are processed through a PI controller, the output of which is an active current component (୮ୢୡ) used to regulate the DC link. The controller output at k୲୦ iteration is estimated as *k* is the number of iterations. Further, the error (*Vdce* ) of reference *V* ∗ *dc* and sensed *Vdc* are processed through a PI controller, the output of which is an active current component *<sup>w</sup>pdc* used to regulate the DC link. The controller output at *k* th iteration is estimated as

$$w\_{pd}(k) = w\_{pd}(k-1) + k\_{pd} \{ V\_{dce}(k) - V\_{dce}(k-1) \} + k\_{\rm id} V\_{dce}(k). \tag{6}$$

During no sunshine conditions ( = 0), the set point DC link voltage is set to reference the DC link voltage of DSTATCOM so that the system operates in the power quality improvement mode (as a DSTATCOM). During no sunshine conditions (*PPV* = 0), the set point DC link voltage is set to reference the DC link voltage of DSTATCOM so that the system operates in the power quality improvement mode (as a DSTATCOM).

The error ௧ has been considered between actual and set terminal voltage, ௧and ௧ <sup>∗</sup>, at PCC to calculate the reactive loss component. Further, this error is passed through a PI controller for minimization. The calculated reactive loss component is considered to maintain AC terminal voltage constant and close to its reference value. The error *Vte* has been considered between actual and set terminal voltage, *V<sup>t</sup>* and *V* ∗ *t* , at PCC to calculate the reactive loss component. Further, this error is passed through a PI controller for minimization. The calculated reactive loss component is considered to maintain AC terminal voltage constant and close to its reference value.

$$V\_{\rm ft}(k) = V\_{\rm t}^\*(k) - V\_{\rm t}(k)\tag{7}$$

୯୲() = ୯୲(−1) + ௧ሼ௧() − ௧(−1)ሽ + ௧௧() (8)

The controller output at *k* th instant is

$$w\_{qt}(k) = w\_{qt}(k-1) + k\_{pt} \{ V\_{tc}(k) - V\_{tc}(k-1) \} + k\_{it} V\_{tc}(k) \tag{8}$$

where *wqt* is a part of the reactive loss current component, and *kpt* and *kit* are proportional and integral gains, respectively.

A feed-forward weight is calculated and incorporated into the controller to achieve a fast dynamic response, which can be written as:

$$w\_{PV}(k) = \frac{2P\_{PV}(k)}{\Im V\_t} \tag{9}$$

where *PPV* is solar power.

3.2.3. Fundamental Active and Reactive Component of Load Current

A generalized neural network (GNN) of a summation-type network is used in the present work for the estimation of the fundamental active and reactive components of the load current. The output calculations have been divided into two sections: forward calculations and reverse calculations.

With the help of Figure 3, the below mentioned equation is used to obtain the output of the summation part of the GNN:

$$O\_{\Sigma} = f\_1(\Sigma W\_{\Sigma i} X\_i + X\_{o\Sigma}) \tag{10}$$

Similarly, the output of the product part of the GNN can be calculated as

$$O\_{\rm II} = f\_2(\rm IIIW\_{\rm III}X\_i + X\_{o\rm II}) \tag{11}$$

The output of the final GNN will be the sum of the summation part and product part which can be expressed as

$$\mathbf{O}\_{i} = \mathbf{O}\_{\Sigma} \ast \mathbf{W}\_{\Sigma} + \mathbf{O}\_{\Pi}(\mathbf{1} - \mathbf{W}\_{\Sigma}) \tag{12}$$

In the above equation, *O*<sup>Σ</sup> denotes the output of the summation part of the neuron, *O*<sup>Π</sup> shows the output of the product part of the neuron, and *W* depicts the weights. The proposed control strategy using GNN utilizes a GNN model to determine the fundamental active current components of the sensed load currents (*iLa*, *iLb*, *iLc*). The sensed load currents are the input to the proposed model which are further multiplied with weights (*w*Σ*pa*, *w*Σ*pb*, *w*Σ*pc*) and (*w*Π*pa*, *w*Π*pb*, *w*Π*pc*) at summation (Σ*A*) and product (Π) neurons, respectively. At the initial level, these weights are referred as unknown weights and calculated by considering the in-phase unit templates (*upa*, *upb*, *upc*) as the reference weights. A summation-type GNN structure for the calculation of fundamental active current component of a phase is given in Figure 3b.

The output calculations can be divided into two parts: forward calculations and reverse calculations. The aggregation function of Σ*<sup>A</sup>* and Π of the forward mode calculations for phase "*a*" is shown in Equations (8) and (9), respectively. Similarly, the aggregation function of Σ*<sup>A</sup>* and Π of the forward mode calculations of another two phases, "*b*" and "*c*", can be calculated.

$$
\Sigma\_{A\_{Pa}} = i\_{La} w\_{\Sigma pa} + i\_{Lb} w\_{\Sigma pb} + i\_{Lc} w\_{\Sigma pc} + \Sigma\_{\text{bias}} \tag{13}
$$

$$
\Pi\_{pa} = \left. i\_{La} w\_{\Pi p a} \ast i\_{Lb} w\_{\Pi p b} \ast i\_{Lc} w\_{\Pi p c} \ast \Pi\_{bias} \right. \tag{14}
$$

Σ*bias* and Π*bias* represent the initial bias of the Σ*<sup>A</sup>* and Π part of the structure, respectively. The proposed developed neuron has both Σ and Π aggregation functions. The Σ*<sup>A</sup>* aggregation function has been used with the sigmoidal characteristic function *f* <sup>1</sup>, while the Π aggregation function has been used with the Gaussian function *f* <sup>2</sup> as a characteristic

function. The output of the Σ*<sup>A</sup>* part with a sigmoidal characteristic transfer function after threshold can be calculated as

$$O\_{\Sigma\_{Apa}} = f\_1 \left( \Sigma\_{Apa} \right) = \frac{1}{1 + e^{-\lambda\_{\Sigma p} \* \Sigma\_{Ap}}} \tag{15}$$

The output of the Π part is threshold by using Gaussian transfer function and can be written as

$$O\_{\Pi \text{pa}} = f\_2 \left( \Pi\_{\text{pa}} \right) = e^{-\lambda\_{\Pi \text{p}} \ast \Pi\_{\text{pa}}^{-2}} \tag{16}$$

where *λ*Σ*<sup>p</sup>* and *λ*Π*<sup>p</sup>* are the gain scaling parameters of the Σ*<sup>A</sup>* and Π part of the network, respectively, and are considered as unity here to avoid complexity. Similarly, output of the Σ*<sup>A</sup>* and Π part of the forward mode calculations of other phases, "*b*" and "*c*", are calculated as

$$O\_{\Sigma\_{Apb}} = f\_1 \left( \Sigma\_{Apb} \right) = \frac{1}{1 + e^{-\lambda\_{\Sigma p} \* \Sigma\_{Apb}}} \tag{17}$$

$$O\_{\Pi p b} = f\_2 \left( \Pi\_{p b} \right) = e^{-\lambda\_{\Pi p} \ast \Pi\_{p b}^2} \tag{18}$$

$$O\_{\Sigma\_{Apc}} = f\_1(\Sigma\_{Apc}) = \frac{1}{1 + e^{-\lambda\_{\Sigma p} \ast \Sigma\_{Ap\_c}}} \tag{19}$$

$$O\_{\rm IIpc} = f\_2(\Pi\_{\rm pc}) = e^{-\lambda\_{\rm IIp} \ast \Pi\_{\rm pc}^2} \tag{20}$$

The final output of the GNN will be the function of two outputs and related to the weights *W* and (1 − *W*), respectively, through the linear transfer function which can be written as

$$O\_{pa} = O\_{\Pi \mu a} (1 - W\_a) + O\_{\Sigma A \mu a} W\_a \tag{21}$$

$$\mathcal{O}\_{pb} = \mathcal{O}\_{\Pi pb}(1 - \mathcal{W}\_b) + \mathcal{O}\_{\Sigma Apb} \mathcal{W}\_b \tag{22}$$

$$O\_{pc} = O\_{\text{IIpc}}(1 - \mathcal{W}\_{\text{c}}) + O\_{\Sigma \text{Apc}} \mathcal{W}\_{\text{c}} \tag{23}$$

where *Wa*, *W<sup>b</sup>* , *W<sup>c</sup>* are weights associated with the phase "*a*", "*b*" and "*c*", respectively. The mean active component of load currents (*wLp*) is obtained by averaging the final output of the GNN for each phase and is given as

$$w\_{Lp} = \frac{\left(O\_{pa} + O\_{pb} + O\_{pc}\right)}{3} \tag{24}$$

However, the estimation of the fundamental reactive current component under the load current has been calculated by implementing the proposed GNN model. The sensed load currents (*iLa*, *<sup>i</sup>Lb*, *<sup>i</sup>Lc*) with their unknown weights, *<sup>w</sup>Aqa*, *<sup>w</sup>Aqb*, *<sup>w</sup>Aqc* and *<sup>w</sup>*Π*qa*, *<sup>w</sup>*Π*qb*, *<sup>w</sup>*Π*qc* , are processed as inputs to Σ*<sup>A</sup>* and Π neurons. The aggregation function of Σ*<sup>A</sup>* and Π of the forward mode calculations for phase "*a*" is given in Equations (23) and (24), respectively. Similarly, the aggregation function of Σ*<sup>A</sup>* and Π of the forward mode calculations of the other remaining phases "*b*" and "*c*" are calculated.

$$
\Sigma\_{\text{A}\_{qa}} = i\_{\text{La}} w\_{\Sigma qa} + i\_{\text{Lb}} w\_{\Sigma qb} + i\_{\text{Lc}} w\_{\Sigma qc} + \Sigma\_{\text{bias}} \tag{25}
$$

$$
\Pi\_{qa} = \left. i\_{\rm Ld} w\_{\rm IIqa} \ast i\_{\rm Lb} w\_{\rm IIqb} \ast i\_{\rm Lc} w\_{\rm IIqc} \ast \Pi\_{bia} \right. \tag{26}
$$

where Σ*bias* and Π*bias* denote the initial bias of the Σ*<sup>A</sup>* and Π part of the network, respectively. Similarly, as explained above while calculating the fundamental active load current component, here also the Σ*<sup>A</sup>* aggregation function has been adopted along with the sigmoidal characteristic function *f* <sup>1</sup>, whereas the Π aggregation function has been adopted

along with the Gaussian function *f* <sup>2</sup>, taken as a characteristic function. The output of the Σ*<sup>A</sup>* part with a sigmoidal characteristic transfer function after threshold can be obtained as

$$O\_{\Sigma\_{Aqa}} = f\_1 \left( \Sigma\_{Aqa} \right) = \frac{1}{1 + e^{-\lambda\_{\Sigma\_q \* \Sigma\_{Aqa}}}} \tag{27}$$

The output of the Π part becomes threshold by using the Gaussian transfer function and can be written as given:

$$O\_{\rm IIqa} = f\_{\rm 2} \left( \Pi\_{qa} \right) = e^{-\lambda\_{\rm IIq} \ast \Pi\_{qa}^{-2}} \tag{28}$$

where *λ*Σ*<sup>q</sup>* and *λ*Π*<sup>q</sup>* are the gain scaling parameters of the Σ*<sup>A</sup>* and Π part of the network, respectively, and are considered as unity here to avoid complexity. Similarly, the output of the Σ*<sup>A</sup>* and Π part of the forward mode calculations of the other two phases "*b*" and "*c*" are calculated as

$$O\_{\Sigma\_{Aqb}} = f\_1 \left( \Sigma\_{Aqb} \right) = \frac{1}{1 + e^{-\lambda\_{\Sigma q} \* \Sigma\_{A\_{qb}}}} \tag{29}$$

$$O\_{\Pi qb} = f\_2 \left( \Pi\_{qb} \right) = e^{-\lambda\_{\Pi q} \ast \Pi\_{qb}^2} \tag{30}$$

$$O\_{\Sigma\_{Aqc}} = f\_1\left(\Sigma\_{Aqc}\right) = \frac{1}{1 + e^{-\lambda\_{\Sigma q} \* \Sigma\_{Aqc}}}\tag{31}$$

$$O\_{\rm IIqc} = f\_2(\rm II\_{qc}) = e^{-\lambda\_{\rm IIq} \ast \rm II\_{qc}^2} \tag{32}$$

The output of the GNN model as a function of weights *W* and (1 − *W*) can be written

as

$$O\_{qa} = O\_{\text{II}qa}(1 - \mathcal{W}\_{d1}) + O\_{\Sigma Apa} \mathcal{W}\_{d1} \tag{33}$$

$$O\_{qb} = O\_{\text{II} qb} (1 - \mathcal{W}\_{b1}) + O\_{\Sigma A qb} \mathcal{W}\_{b1} \tag{34}$$

$$O\_{q\circledast} = O\_{\text{II} q\circledast} (1 - W\_{\text{c1}}) + O\_{\Sigma A q\circledast} W\_{\text{c1}} \tag{35}$$

where *Wa*1, *Wb*<sup>1</sup> and *Wc*<sup>1</sup> are the weights associated with the Σ*<sup>A</sup>* and Π part of the developed GNN model for the estimation of the fundamental reactive load current components. The fundamental reactive current component of the load current (*wLq*) is determined by taking the average of output of the GNN model for the reactive current component of the load currents and given as:

$$w\_{Lq} = \frac{\left(O\_{qa} + O\_{qb} + O\_{qc}\right)}{3} \tag{36}$$

#### 3.2.4. GNN Weight Prediction and Updating Using Extended Kalman Filter (EKF)

The Kalman Filter (KF) is a recursive algorithm used for estimating the state of a dynamic system in the case of less availability of data because of the presence of noise, etc. The KF algorithm utilizes the prior knowledge to predict the past, present and future state of the given system. The main advantage of the KF-based approach is less memory space requirements because the data are updated in each and every iteration. The basic KF theory is based on the probability of the hypothesis of the predicted state of the system under consideration by hypothesis of prior state and then using the available data from measurement sensors to correct the hypothesis to obtain the best estimation for each iteration. Two basic assumptions are made to derive the basic equations for KF to be optimal in the sense of mean square error, which should be described by a model of linear state space; the noises are white and Gaussian with zero mean, uncorrelated with each other.

$$\mathbf{x}\_{k+1} = \mathbf{F}\_{k+1,k}\mathbf{x}\_k + q\_k \tag{37}$$

$$y\_{k+1} = H\_{k+1} \ x\_{k+1} + r\_{k+1} \tag{38}$$

Equation (37) is known as a process equation. *x<sup>k</sup>* is a system state vector, a minimal set of data that uniquely defines the behavior of a system, and *k* depicts discrete time. *Fk*+1, *<sup>k</sup>* is the transition matrix to take the state *x<sup>k</sup>* from time *k* to *k* + 1, and *q<sup>k</sup>* is the additive process noise, white and Gaussian, with a zero mean and possessing a covariance matrix *Q<sup>k</sup>* . The measurement step is shown in Equation (38), where *Hk*+<sup>1</sup> is the measurement matrix, *yk*+<sup>1</sup> is observable at time *k* + 1, and *rk*+<sup>1</sup> is the additive process noise, white and Gaussian, with a zero mean and possessing a covariance matrix *R<sup>k</sup>* . Both noises are uncorrelated with each other. The KF algorithm works in two repeated functional steps:

The prediction step (time update): This step is to compute the estimation of state and error covariance.

$$\pounds^{-}\!\_{k+1} = F\_{k+1,\:k} \pounds\_k \tag{39}$$

$$P\_{k+1} = F\_{k+1,k} \, P\_k \, \, \, F\_{k+1,k}^T \, \, \, \, \, Q\_k \, \, \tag{40}$$

where *P <sup>k</sup>* is there error covariance matrix.

Correction step (measurement update): This step is to correct the estimated state according to the previous step with the help of *yk*+<sup>1</sup>

$$K\_{k+1} = P^{-}\_{k+1} \ H^{T}\_{k+1} \left[ H\_{k+1} P^{-}\_{k+1} \ H^{T}\_{k+1} + R\_{k+1} \right]^{-1} \tag{41}$$

$$\pounds\_{k+1} = \pounds\_k ^- \K\_{k+1} \left( y\_{k+1} - H\_{k+1} \: \pounds^- \_k \right) \tag{42}$$

$$P\_{k+1} = (I - K\_{k+1} \ H\_{k+1} \ ) P\_{k+1} ^ {-} \tag{43}$$

where *K <sup>k</sup>*+<sup>1</sup> is the Kalman gain matrix.

The GNN model is a nonlinear system, so the basic KF approach should be extended by using the linearization process and is known as the extended Kalman Filter (EKF). The basic difference between KF and EKF is the linearization of the nonlinear system function performed by using the Jacobian matrix in EKF, and then rest of the KF steps can be applied. The nonlinear dynamic system can be defined by using the following equations:

$$\mathfrak{x}\_{k+1} = \mathfrak{x}\_k + q\_k \tag{44}$$

$$y\_{k+1} = h(x\_{k+1}, u\_{k+1}) + r\_{k+1} \tag{45}$$

Equation (44) depicts the state of a stationary process corrupted with process noise *q<sup>k</sup>* , and state *x<sup>k</sup>* consists of network weights. *xk*+<sup>1</sup> is the weight vector matrix and input vector depicted by *uk*+<sup>1</sup> . The noise covariance matrix can be written as *Rk*+<sup>1</sup> = *E* - *rk*+<sup>1</sup> *r T k*+1 and *Qk*+<sup>1</sup> = *E* - *qk*+1*q T k*+1 

In EKF, the linearization of measurement equation is carried out at each time step around the newest state estimation by using the first-order Taylor approximation. The GNN training problem is taken as a problem to find the state estimation of *xk*+<sup>1</sup> for minimizing the least square errors by implementing the previous calculations. *Kk*+<sup>1</sup> Kalman gain can be written as

$$\mathbf{K}\_{k+1} = \,^p\_{k+1} \boldsymbol{H}^T \boldsymbol{k}\_{k+1} \left[ \boldsymbol{H}\_{k+1} \, \boldsymbol{P}\_{k+1} \boldsymbol{H}^T \boldsymbol{k}\_{k+1} + \boldsymbol{R}\_{k+1} \right]^{-1} \tag{46}$$

$$\pounds\_{k+1} = \pounds\_{k+1} + \pounds\_{k+1} \left[ \y\_{k+1} - h(\pounds\_{k+1} + \mu\_{k+1}) \right] \tag{47}$$

$$P\_{k+1} = P\_k - \mathcal{K}\_k H\_k \ P\_k + Q\_k \tag{48}$$

3.2.5. Reference Current Calculation

The fundamental real power current component of the load current can be obtained as

$$I\_{Lp} = w\_{pd} + w\_{Lp} - w\_{PV} \tag{49}$$

The reference active components of the grid currents can be written as

$$i\_{psa} = I\_{Lp} \* \mu\_{pa\prime} \ i\_{psb} = I\_{Lp} \* \mu\_{pb\prime} \ i\_{psc} = I\_{Lp} \* \mu\_{pc} \tag{50}$$

Similarly, the fundamental reactive current component of a load current can also be obtained as

$$I\_{Lq} = w\_{qt} + w\_{Lq} \tag{51}$$

The grid currents' active reference components can be written as

$$\dot{\mathbf{u}}\_{q\mathbf{s}a} = \mathbf{I}\_{\mathbf{L}\boldsymbol{\eta}} \ast \mathbf{u}\_{q\mathbf{a}\prime} \, \dot{\mathbf{u}}\_{q\mathbf{s}b} = \mathbf{I}\_{\mathbf{L}\boldsymbol{\eta}} \ast \mathbf{u}\_{q\mathbf{b}\prime} \, \dot{\mathbf{u}}\_{q\mathbf{s}c} = \mathbf{I}\_{\mathbf{L}\boldsymbol{\eta}} \ast \mathbf{u}\_{q\mathbf{c}} \tag{52}$$

The net fundamental reference grid currents (*i* ∗ *sa*, *i* ∗ *sb*, *i* ∗ *sc*) can be obtained as

$$\dot{\mathbf{i}}\_{\rm sa}^{\*} = \dot{\mathbf{i}}\_{\rm psa} + \dot{\mathbf{i}}\_{\rm qsa} \cdot \dot{\mathbf{i}}\_{\rm sb}^{\*} = \dot{\mathbf{i}}\_{\rm psb} + \dot{\mathbf{i}}\_{\rm qsb} \cdot \dot{\mathbf{i}}\_{\rm sc}^{\*} = \dot{\mathbf{i}}\_{\rm psc} + \dot{\mathbf{i}}\_{\rm qsc} \tag{53}$$

VSC gating signals have been generated by comparing reference (*i* ∗ *sa*, *i* ∗ *sb*, *i* ∗ *sc*) and actual (*isa*, *isb*, *isc*) grid currents for an individual phase, and the error is processed through the PI current regulator. A hysteresis current regulator is used for indirect current control.

#### **4. Results**

The SPV array is designed for a maximum power rating of 10 kW. The given system is designed, developed and simulated using the MATLAB/Simulink platform. The performance evaluation of the proposed system is carried out under a linear and nonlinear load with dynamic changes and a changing solar irradiance scenario.

#### *4.1. Performance Analysis of Proposed Controller for Linear Load for PFC*

The performance of the proposed EKF-based GNN controller is evaluated with a combined structure of the summation and product neurons collectively in single layer. In the proposed GNN network, each processed neuron consists of different weights with the inputs as measured load currents (*iLa*, *iLb*, *iLc*). With the help of this, the summation weights (*w*Σ*pa*, *w*Σ*pb*, *w*Σ*pc*) and product weights (*w*Π*pa*, *w*Π*pb*, *w*Π*pc*) are estimated. The proposed GNN is trained to calculate the unspecified weights and EKF filter which is used to predict and update the weights of the GNN network. The output of the proposed system is evaluated under the influence of several performance parameters, i.e., grid voltage (*vs*), grid current (*is*), load current (*iL*), voltage source converter current (*ivsc*), grid real power (*Pg*), grid re-active power (*Qg*), PCC terminal voltage (*Vt*), SPV power (*Ppv*), SPV array current (*Ipv*) and voltage at direct current link (*Vdc*), provided in Figure 4a,b in steady-state linear load (at *t* = 0.2 to 0.3 s) conditions. Intermediate signals are given in Figure 4b, which depicts the performance parameters and corresponding weights with a variation in error between the active fundamental components of the load currents. The SPV system remains in operation for feeding the maximum value of power onto the load and grid at unity power factor (UPF).

**Figure 4.** (**a**) Performance parameters, (**b**) estimated weights under steady-state linear load condi-**Figure 4.** (**a**) Performance parameters, (**b**) estimated weights under steady-state linear load conditions in PFC mode.

tions in PFC mode.

#### *4.2. Performance Analysis of Proposed Controller Considering Dynamic Linear Load for Zero Voltage Regulation (ZVR) 4.2. Performance Analysis of Proposed Controller Considering Dynamic Linear Load for Zero Voltage Regulation (ZVR)*

If the load is drawing reactive power with unbalanced behavior, then a change in the terminal voltage (Vt) occurs at the point of common coupling. Figure 5a,b shows the behavior of the given system for dynamic linear loading conditions under a zero voltage regulation (ZVR) mode. It can be observed from the results after phase "c" of the load is taken out from the supply at 1.1 s, the grid currents (௦) remain sinusoidal with the help of a voltage source converter. The PCC voltage (௧) and voltage at the DC junction are If the load is drawing reactive power with unbalanced behavior, then a change in the terminal voltage (*Vt*) occurs at the point of common coupling. Figure 5a,b shows the behavior of the given system for dynamic linear loading conditions under a zero voltage regulation (ZVR) mode. It can be observed from the results after phase "*c*" of the load is taken out from the supply at 1.1 s, the grid currents (*is*) remain sinusoidal with the help of a voltage source converter. The PCC voltage (*Vt*) and voltage at the DC junction are maintained at set values, which are 415 V and 700 V, respectively, without any fluctuations. Intermediate signals are provided in Figure 5b, which shows the performance parameters and corresponding weights with a variation in error between the active fundamental components of the load currents. The reactive power (Qg) taken from the utility grid is approximately equivalent to 0, for compensating the linear loads using reactive power. tions. Intermediate signals are provided in Figure 5b, which shows the performance parameters and corresponding weights with a variation in error between the active fundamental components of the load currents. The reactive power (Qg) taken from the utility grid is approximately equivalent to 0, for compensating the linear loads using reactive power.

maintained at set values, which are 415 V and 700 V, respectively, without any fluctua-

*Sustainability* **2021**, *13*, 4219 13 of 24

**Figure 5.** *Cont.*

**Figure 5.** (**a**) Performance parameters, (**b**) estimated weights under dynamic linear load conditions for zero voltage regulation (ZVR). **Figure 5.** (**a**) Performance parameters, (**b**) estimated weights under dynamic linear load conditions for zero voltage regulation (ZVR).

#### *4.3. Performance under a Nonlinear Load 4.3. Performance under a Nonlinear Load*

Figure 6a,b show the behavior of a developed system for dynamic nonlinear loads under the zero voltage regulation (ZVR) mode. By using the given control algorithm, the grid current () is sinusoidal in nature, when the load of phase "c" is extracted at 1.1 s. Intermediate signals are given in Figure 6b for corresponding weight signals with a variation in error between the active fundamental components of the load currents. Wpa of the linear transfer function Σ and the estimated GNN output of the active current component of the load current are represented by Opa, and its actual value is achieved by scaling and depicted by ILpa. This indicates that the calculated weights change according to the load condition. Figure 7a,c show the total harmonic distortion (THD) at the point of common coupling for phase "a" of a load current, grid current and grid voltage, respectively, and obtained as 37.48%, 1.04% and 1.07%, respectively, which is within acceptable limits of harmonics in utility currents and voltage fluctuations according to the IEEE-519 and IEEE-1547 standards. THD analysis of the given system is provided in Table 1 for a nonlinear load. Figure 6a,b show the behavior of a developed system for dynamic nonlinear loads under the zero voltage regulation (ZVR) mode. By using the given control algorithm, the grid current (*is*) is sinusoidal in nature, when the load of phase "*c*" is extracted at 1.1 s. Intermediate signals are given in Figure 6b for corresponding weight signals with a variation in error between the active fundamental components of the load currents. *Wpa* of the linear transfer function Σ and the estimated GNN output of the active current component of the load current are represented by *Opa*, and its actual value is achieved by scaling and depicted by *ILpa*. This indicates that the calculated weights change according to the load condition. Figure 7a,c show the total harmonic distortion (THD) at the point of common coupling for phase "*a*" of a load current, grid current and grid voltage, respectively, and obtained as 37.48%, 1.04% and 1.07%, respectively, which is within acceptable limits of harmonics in utility currents and voltage fluctuations according to the IEEE-519 and IEEE-1547 standards. THD analysis of the given system is provided in Table 1 for a nonlinear load.

**Figure 6.** (**a**) Performance indices, (**b**) weights with training parameters for phase "a" with steady-**Figure 6.** (**a**) Performance indices, (**b**) weights with training parameters for phase "*a*" with steady-state loading in PFC mode.

state loading in PFC mode.

**Figure 7.** Total harmonic distortion (THD) for (**a**) load current of phase "a", (**b**) grid current and (**c**) grid voltage in ZVR mode. **Figure 7.** Total harmonic distortion (THD) for (**a**) load current of phase "*a*", (**b**) grid current and (**c**) grid voltage in ZVR mode.

**Table 1.** Parameters of the developed system for the proposed controller. **Table 1.** Parameters of the developed system for the proposed controller.


#### *4.4. Performance Analysis at Varying Solar Irradiance*

the voltage at DC link is also maintained at a set point.

*4.4. Performance Analysis at Varying Solar Irradiance*  Dynamic behavior of the developed system is also observed under varying solar irradiance conditions. The solar irradiance (S) is raised to 1000 W/m2 from 600 W/m2 at 0.5 s. The SPV feeds the connected load and excess amount of power output provided to the utility grid. As shown in the results, the grid power (Pg) decreases after 0.5 s because of an increase in power (Ppv) output from SPV, given in Figure 8. The proposed system under variable solar irradiance remains operating at the maximum power point of the SPV array and works at unity power factor. The grid side current is maintained to be sinusoidal, and Dynamic behavior of the developed system is also observed under varying solar irradiance conditions. The solar irradiance (S) is raised to 1000 W/m<sup>2</sup> from 600 W/m<sup>2</sup> at 0.5 s. The SPV feeds the connected load and excess amount of power output provided to the utility grid. As shown in the results, the grid power (*Pg*) decreases after 0.5 s because of an increase in power (*Ppv*) output from SPV, given in Figure 8. The proposed system under variable solar irradiance remains operating at the maximum power point of the SPV array and works at unity power factor. The grid side current is maintained to be sinusoidal, and the voltage at DC link is also maintained at a set point.

#### *4.5. Comparative Study of Developed Algorithm with Other Conventional Approaches*

The proposed algorithm is analyzed for different scenarios and found accurate as per the desired performance. The parameters of proposed EKF GNN-based controller is presented in Table 1 and compared with the GNN approach. The developed EKF GNN based control approach is found better than ANN conventional approaches such as ADALINE and multilayer perceptron neural network (MLPN). A comparative analysis is made and presented in Table 2 under dynamic nonlinear load conditions on the basis of various performance parameters. The smaller number of unknown weights required and a single layer enhanced the performance of the controller. Additionally, the weight update time, settling period and maximum change in DC voltage is less for the proposed technique.

**Figure 8.** Analysis of performance for a non-linear load with variable solar irradiance. **Figure 8.** Analysis of performance for a non-linear load with variable solar irradiance.


*4.5. Comparative Study of Developed Algorithm with other Conventional Approaches*  **Table 2.** Comparative performance analysis of the proposed technique.

Learning Gradient decent (GD) GD/GDM GDM Layers Two Three One Training pattern Online Training Offline stochastic training Online training Estimation nature Linear Nonlinear Both types Figure 9 shows the performances by experimenting the given system in power factor correction mode by implementing linear and nonlinear loads. It depicts the voltage source current (*Ivsc*), load current (*ILa*), grid current (*Ig*) and direct current link voltage (*Vdc*). Phase "*a*" gid voltage and grid current are shown in Figure 10; it also shows the performances of the controller under PFC mode.

Transfer function Linear Sigmoidal All

Training Least Mean Square Back Propagation EKF

Settling period of DC

Settling period of DC

Max change in DC

Max change in DC

of the controller under PFC mode.

of the controller under PFC mode.

**Figure 9.** Performance parameters under linear load for the extended Kalman filter (EKF) GNN. **Figure 9.** Performance parameters under linear load for the extended Kalman filter (EKF) GNN. **Figure 9.** Performance parameters under linear load for the extended Kalman filter (EKF) GNN.

Weight update Time 15.7 µs 82 µs 6 µs

Weight update Time 15.7 µs 82 µs 6 µs

*Sustainability* **2021**, *13*, 4219 18 of 24

link 1 cycle 2 ½ cycle 1 cycle

link 1 cycle 2 ½ cycle 1 cycle

Figure 9 shows the performances by experimenting the given system in power factor correction mode by implementing linear and nonlinear loads. It depicts the voltage source current (I**vsc**), load current (I**La**), grid current (Ig) and direct current link voltage (Vdc). Phase "a" gid voltage and grid current are shown in Figure 10; it also shows the performances

Figure 9 shows the performances by experimenting the given system in power factor correction mode by implementing linear and nonlinear loads. It depicts the voltage source current (I**vsc**), load current (I**La**), grid current (Ig) and direct current link voltage (Vdc). Phase "a" gid voltage and grid current are shown in Figure 10; it also shows the performances

link voltage 4.5 V 10 V 3.7 V

link voltage 4.5 V 10 V 3.7 V

**Figure 10.** Source side voltage and current of phase "a" under the power factor correction mode. **Figure 10.** Source side voltage and current of phase "*a*" under the power factor correction mode.

Figure 11 provides the performance of the SPV plant for nonlinear loads, whereas Figure 12a depicts non-linear current and Figure 12b depicts the source current and voltages for phase "a". The THD of the load current is given in Figure 13. Figure 11 provides the performance of the SPV plant for nonlinear loads, whereas Figure 12a depicts non-linear current and Figure 12b depicts the source current and voltages for phase "a". The THD of the load current is given in Figure 13. Figure 11 provides the performance of the SPV plant for nonlinear loads, whereas Figure 12a depicts non-linear current and Figure 12b depicts the source current and voltages for phase "*a*". The THD of the load current is given in Figure 13.

**Figure 11.** Performance analysis considering non-linear load conditions for EKF GNN.

**Figure 11.** Performance analysis considering non-linear load conditions for EKF GNN.

**Figure 11.** Performance analysis considering non-linear load conditions for EKF GNN.

*Sustainability* **2021**, *13*, 4219 19 of 24

**Figure 12.** (**a**) Non-linear current, (**b**) current and voltage of phase "a". **Figure 12.** (**a**) Non-linear current, (**b**) current and voltage of phase "*a*". **Figure 12.** (**a**) Non-linear current, (**b**) current and voltage of phase "a".

**Figure 13.** (**a**) THD spectrum of nonlinear load, (**b**) improved THD for grid current under nonlinear load. GNN. **Figure 13.** (**a**) THD spectrum of nonlinear load, (**b**) improved THD for grid current under nonlinear load.

in Figure 14; when load of phase "b" is removed, the grid current is reduced. Further, the performance of intermediate signals for different weights is also shown in Figure 15. Fig-

ures 16 and 17 show the controller performance under variable solar irradiance.

**Figure 14.** Performance parameters under dynamic nonlinear load for EKF GNN.

**Figure 15.** Performance of intermediate weight signals under dynamic nonlinear load for EKF

The performance of the proposed controller under nonlinear dynamic load is shown in Figure 14; when load of phase "*b*" is removed, the grid current is reduced. Further, the performance of intermediate signals for different weights is also shown in Figure 15. Figures 16 and 17 show the controller performance under variable solar irradiance.

**Figure 14.** Performance parameters under dynamic nonlinear load for EKF GNN.

**Figure 15.** Performance of intermediate weight signals under dynamic nonlinear load for EKF GNN.

**Figure 16.** Performance parameters under nonlinear load with solar PV ON for EKF GNN. **Figure 16.** Performance parameters under nonlinear load with solar PV ON for EKF GNN.

**Figure 17.** Performance parameters under variable solar irradiance for EKF GNN. **Figure 17.** Performance parameters under variable solar irradiance for EKF GNN.

#### **5. Conclusions 5. Conclusions**

The proposed system was developed and tested using MATLAB simulation environment, and the performance of the EKF GNN-based VSC control approach was validated under linear and nonlinear loads for static and dynamic scenarios. The proposed EKF GNN approach has successfully improved the function of the developed SPV system. The performance of the developed system was found satisfactory under load unbalancing and varying solar irradiance. Additionally, it continues to operate in a UPF mode of operation, providing reactive power compensation, load balancing, MPP extraction and harmonics compensation. A changeable step size perturb and observe (P&O) MPPT approach was utilized in this research and enabled fast tracking of MPP and convergence. Moreover, single-stage topology was able to reduce the losses in semiconductor devices and increase The proposed system was developed and tested using MATLAB simulation environment, and the performance of the EKF GNN-based VSC control approach was validated under linear and nonlinear loads for static and dynamic scenarios. The proposed EKF GNN approach has successfully improved the function of the developed SPV system. The performance of the developed system was found satisfactory under load unbalancing and varying solar irradiance. Additionally, it continues to operate in a UPF mode of operation, providing reactive power compensation, load balancing, MPP extraction and harmonics compensation. A changeable step size perturb and observe (P&O) MPPT approach was utilized in this research and enabled fast tracking of MPP and convergence. Moreover, single-stage topology was able to reduce the losses in semiconductor devices and increase the overall efficiency.

the overall efficiency. The proposed control approach performs with more flexibly in training the network under dynamic conditions and provides improved online learning. This control technique needs a smaller number of training patterns and unknown weights and hence reduces the complexity and the computational time. The developed system obtains acceptable limits of harmonics in utility currents and voltage fluctuations according to the IEEE-519 and IEEE-1547 standards. The proposed control approach performs with more flexibly in training the network under dynamic conditions and provides improved online learning. This control technique needs a smaller number of training patterns and unknown weights and hence reduces the complexity and the computational time. The developed system obtains acceptable limits of harmonics in utility currents and voltage fluctuations according to the IEEE-519 and IEEE-1547 standards.

**Author Contributions:** Conceptualization, P.C., M.R. and I.A.; methodology, P.C., M.A. and M.R.; validation, P.C. and I.A.; formal analysis, P.C., I.A., M.A. and M.R.; resources, M.R. and I.A.; writing original draft, P.C. and M.R.; writing—review and editing, P.C., I.A., M.R. and M.A.; visualization, P.C.; supervision, M.R.; funding acquisition, I.A., M.A. and M.R. All authors have read and agreed to the published version of the manuscript.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** No data are used here.

**Acknowledgments:** Researchers would like to thank the Deanship of Scientific Research, Qassim University for funding publication of this project.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Nomenclature**



#### **References**


## *Article* **Accurate Sizing of Residential Stand-Alone Photovoltaic Systems Considering System Reliability**

**Eduardo Quiles 1,\*, Carlos Roldán-Blay <sup>2</sup> , Guillermo Escrivá-Escrivá <sup>2</sup> and Carlos Roldán-Porta <sup>2</sup>**


Received: 10 January 2020; Accepted: 6 February 2020; Published: 10 February 2020

**Abstract:** In rural areas or in isolated communities in developing countries it is increasingly common to install micro-renewable sources, such as photovoltaic (PV) systems, by residential consumers without access to the utility distribution network. The reliability of the supply provided by these stand-alone generators is a key issue when designing the PV system. The proper system sizing for a minimum level of reliability avoids unacceptable continuity of supply (undersized system) and unnecessary costs (oversized system). This paper presents a method for the accurate sizing of stand-alone photovoltaic (SAPV) residential generation systems for a pre-established reliability level. The proposed method is based on the application of a sequential random Monte Carlo simulation to the system model. Uncertainties of solar radiation, energy demand, and component failures are simultaneously considered. The results of the case study facilitate the sizing of the main energy elements (solar panels and battery) depending on the required level of reliability, taking into account the uncertainties that affect this type of facility. The analysis carried out demonstrates that deterministic designs of SAPV systems based on average demand and radiation values or the average number of consecutive cloudy days can lead to inadequate levels of continuity of supply.

**Keywords:** renewable energy; photovoltaic generation; battery storage; reliability evaluation; Monte Carlo Simulation

#### **1. Introduction**

The need to reduce dependency in fossil fuels has promoted the use of renewable energy sources. The appearance of renewable energy sources such as wind power, solar power, or small hydro plants in the electrical market is increasing every day. Many of these renewable energy plants are customer owned and have small unitary power.

These renewable microgeneration systems can be isolated from the distribution network (off-grid) or connected to it exporting their energy surpluses [1–3]. In rural areas or in isolated communities in developing countries it is increasingly common to install micro-renewables sources by residential consumers without access to the utility distribution network (DN). In Spain these micro-renewables are mainly off-grid photovoltaic (PV) systems for electrification of single residential households. These stand-alone photovoltaic (SAPV) energy systems generally include batteries for energy storage [4–6].

The Spanish Administration has recently regulated self-consumption of electricity [7]. This regulation was issued as an urgent measure for renewable energy promotion and consumer protection. It has reduced the administrative hurdles suffered by small-scale energy plants and it allows collective self-consumption for the first time.

This new regulation defines the concept of "neighbor facility" in order to regulate collective self-consumption and confirms the elimination of the charge to self-consumed energy. It creates a simplified compensation mechanism to compensate self-consumers with surplus energy that export to the network and facilitates the installation of energy storage elements without more requirements than to comply with safety and industrial quality regulations. This regulation is expected to incentive the installation of more SAPV generation systems for residential customers.

To design a SAPV system it is necessary to determine the PV panels' rated power and battery storage capacity. PV generation has uncertainty associated to its energy output which depends on the irradiation level [8,9]. This irradiation level depends on the weather condition, varying widely between sunny and cloudy days, in the short term (hour to hour) and in the long term (seasonal variations) [10,11]. Even for a typical clear-sky day, fluctuations of PV power are caused by passing clouds. Getting a good prediction of the reliability of a SAPV is mandatory in order to improve its sustainability. A detailed analysis of the characteristics of solar radiation for the area where a SAPV will be installed is convenient. Uncertainty associated to PV generation must be modelled to predict average performance in the future. Time-series measurements of solar radiation data from near-site weather stations are required to estimate expected generation [12–14].

Simultaneity between PV generation and energy consumption in residential households is limited. PV generation is maximum at noon while for a typical residential customer the peak period occurs in the evening. The battery system allows to storage PV energy surpluses produced during the day. As long as the battery does not reach its maximum state of charge (SOC), the PV energy surpluses will be used to charge the battery. Another option to improve self-sufficiency of PV systems is shifting the consumption of deferrable loads by demand-side management to periods with PV-surpluses [15,16]. The creation of cooperative microgrids with different SAPV generators has also been proposed [17,18] to address this problem.

To make a realistic design of the SAPV system an accurate model of load demand is required. This is especially difficult for individual residential customers that usually have a variable load profile. Residential load has a time-varying nature and it changes depending on the time of the day, day of the week and season of the year. To simulate the energy flows between PV unit, battery and load, time series data of PV generation and load demand with a high temporal resolution are required.

Many studies have been carried out to determine the feasibility of SAPV systems [19]. There are some studies [20–23] that evaluate residential PV plants connected to the DN. In this way they can sell energy surpluses to the utility and import energy from the DN when needed. Some other studies [24,25] consider SAPV systems, autonomous and isolated from the DN. This is usually the only available option in many rural areas. Different solutions have been proposed for the design of such SAPV systems. The main objective is to determine the most reliable and cost-effective configuration of PV units for energy generation and batteries for energy storage.

Some authors propose analytical solutions based on energy balance equations [26]. Some of them include statistical approaches to consider solar radiation fluctuations [12]. A review of SAPV systems sizing methodologies can be found in [1,27].

The SAPV design considering reliability of involved systems requires a realistic model of the energy resource, energy demand, and system components faults. Several studies usually consider average solar irradiance levels for a broad area. These values are monthly averages obtained from databases of meteorological services [28,29] In this study local hourly irradiance levels are measured and included in the design process.

As previously stated, residential demand is highly variable between different users and different days. The simplified approach of considering an average daily peak demand clearly underestimates demand fluctuation [1]. This work considers actual demand data measured in an hourly basis from average customers.

Another factor that influences the applicability of simulation results is the temporal resolution of the PV generation and load demand. At least an hourly resolution is required to reflect the power balance between PV generation and energy demand [30–32]. Some authors [1] propose smaller temporal resolutions (10-min sampled data) in order to evaluate energy flows between PV units, batteries, and loads.

Several studies conclude that SAPV systems are an economic and profitable solution for residential customers without access to the DN [33]. These studies show that the initial investments in PV panels, batteries, and installation costs have an amortization period smaller than the PV plant useful life [20,34]. Therefore, for many residential customers the reliability of the electrical supply is the key factor when planning the installation of a SAPV system, rather than the investment costs. Many individual residential investors are willing to afford the investment costs of the PV generation system as far as a satisfactory electrical supply reliability level is warranted [35].

SAPV design must take into account faults that unexpectedly occur in the system. Assuming no component faults can result in an over-optimistic performance prediction and in the subsequent infra-sizing of PV units and batteries [36].

Monte Carlo simulation (MCS) can be performed in a sequential or non-sequential manner [37]. In the sequential MCS the states of the components are sequentially sampled simulating the chronology of the stochastic process of the system operation. For renewable-energy systems with energy storage, the state of the system depends on previous states, i.e., battery SOC level. PV generation and residential demand are not usually correlated. The complexity of this reliability analysis can better be dealt with a sequential MCS. Other approaches as reliability evaluation based on analytical models or Markov models require modeling simplifications that are not suitable for a realistic assessment of SAPV systems [38].

This study addresses the evaluation of component faults using a sequential Monte Carlo simulation methodology. The objective of the paper is to obtain an optimal sizing of the SAPV system from the economical point of view, but imposing some constraints related to the desired reliability of the system. The novelty of this work compared with existing studies is the consideration of actual PV generation and load demand time series data, and the simultaneous evaluation of uncertainties associated to PV generation, load demand, and system component faults. To the authors' best knowledge this is the first work where these three sources of uncertainty are simultaneously taken into account in the design of SAPV systems. The objective is to guarantee a desired reliability in the continuity of the supply in the design of a SAPV generation system.

To validate the results of the proposed method, the sizing of a SAPV system is firstly approached using a deterministic worst case procedure. Then, a reliability evaluation method is used, considering PV generation uncertainty, demand uncertainty and unexpected faults performing a sequential Monte Carlo simulation. The results of both approaches are compared to extract relevant conclusions about the design process of SAPV systems.

#### **2. Materials and Methods**

#### *2.1. Stand-Alone Photovoltaic Energy System*

Many residential households have installed PV renewable generation to satisfy its own energy requirements. This generation is accompanied by batteries to storage energy when surpluses are available and to supply energy when the PV output is insufficient. Figure 1 shows a scheme of the installation of PV panels and batteries to supply a residential load without connection to the grid. The battery has a regulator or battery controller (BC) to control the SOC and its maximum current, both in charge and discharge operations. The BC decides the power flows between the PV panel array and the battery.

The SAPV system has been divided into two sections:


The reliability analysis performed in this work will take into account the possible failures in the PV panel array and the BC considering the failure rate per year of these elements (λc). These failures can be covered by the batteries, that also present a specific failure rate (λb). The possible failures in the inverter (λ<sup>i</sup> ) cannot be supported by other elements in the proposed scheme. Therefore, to include these failures in the analysis, this element should be considered in series with the generation section and the overall reliability would be the product of the reliability of both systems.

**Figure 1.** Stand-alone photovoltaic (SAPV) system electric scheme.

To model a typical household energy demand, the authors of this study have registered time series of data with a temporal resolution of an hour. Figure 2 shows typical values of the daily demand from a residential customer in different seasons.

**Figure 2.** Daily load profile for residential customers and hourly generation curve in different seasons.

PV generation used in this work is based in the actual measurements obtained in the laboratory of Distributed Energy Resources (LABDER) at the Institute for Energy Engineering of the Universitat Politècnica de València, Spain [39]. This laboratory enables the assembly of Hybrid Renewable Energy Systems (HRES) combining different renewable sources: photovoltaic, biogas, wind power and hydrogen fuel cells, interconnected by a controlled microgrid that supplies a specific load. Additionally, the laboratory includes the capability to store energy, both in batteries and hydrogen, to cover most of the possible HRES configurations [40].

All the systems are working in the 10 kW range. Specifically, the photovoltaic generator in the LABDER is made up of monocrystalline and polycrystalline silicon modules mounted on the roof of the laboratory, facing south with tilt angle of 30 degrees to produce maximum annual energy. The total power of the photovoltaic generator now installed in this lab is 2.1 kWp and the panels are connected to a single phase grid inverter.

The operating point of the panels and the inverter, currents, voltages, power and energy injected to the grid are also monitored using a power meter installed at the AC output of the inverter and a data logger to storage the data. This information enables the management system to check the correct operation of the system and to know the energy produced at any time.

The registered power produced during a year has been used to obtain the PV power generation in this work, as depicted in Figure 2. This register allows to use actual data and to obtain reliable values in the different simulations performed later.

To provide a deterministic design (DD) of the size of the SAPV system, the average demand recorded in this study, *Pd(t)*, at the most unfavorable time of the year (winter) is considered. The average daily energy consumption in this period *E<sup>d</sup>* (Wh/day) will be:

$$E\_d = \int\_0^{24} P\_d(t)dt. \tag{1}$$

This energy is obtained from the PV array. The output power of the PV system depends on the irradiance level. This fact introduces uncertainty in the energy resource.

The battery must be designed to secure the energy supply to the loads when the PV output is not enough (at nights and in cloudy days). From the values of the instantaneous PV energy generation *PS(t)* and the energy demand *Pd(t)* the useful energy contributed by the battery in one day is obtained, as depicted in Figure 3.

**Figure 3.** Useful energy contributed by the battery in one day.

As shown in Figure 3, the power produced by the array on a sunny day, *PS(t)*, must exceed the demand during the time interval t1–t2. Then, the energy available to be stored in the battery, *E1*, will be:

$$E\_1 = \int\_{t\_1}^{t\_2} (P\_s(t) - P\_d(t))dt \text{ when } P\_S(t) > P\_d(t). \tag{2}$$

To keep the system stable, the energy that the battery can return in a daily discharge cycle, *E<sup>b</sup>* will be computed according to Equation (3), where η*<sup>c</sup>* and η*<sup>d</sup>* are the battery charge and discharge performance and η*<sup>b</sup>* the overall efficiency of the energy storage and recovery process.

$$E\_b = (\eta\_\mathbf{c} \cdot \eta\_d) \cdot E\_1 = \eta\_b \cdot E\_1. \tag{3}$$

From the adjustment to the balance of the previous data, Equation (4) must be fulfilled:

$$\int\_0^{24} P\_s(t)dt = E\_d + E\_1 \cdot (1 - \eta\_b). \tag{4}$$

The minimum rated power required for the PV array (PVpeak) is then obtained from Equation (4).

For a set of *n<sup>c</sup>* cloudy days, the energy production would be much less than the theoretical one with good weather. For example, admitting that 15% of the theoretical value is obtained, since the irradiance can be between 10% and 20% of a sunny day [41], it would result:

$$n\_c \cdot \left(E\_d - 0.15 \cdot \int\_0^{24} P\_s(t)dt\right) = E \cdot \nu\_{b\prime} \tag{5}$$

$$Q\_b = \frac{E\nu\_b \cdot 100}{\eta\_d \cdot (100 - \text{SOC}\_{\text{min}})} \tag{6}$$

where *E*0*<sup>b</sup>* is the energy contributed by the battery the *n<sup>c</sup>* cloudy days (it has been assumed that *Pd(t)* > 0.15·*PS(t)* during all the time those days), *SOCmin* is the minimum admissible value of SOC after those *n<sup>c</sup>* days (for example 15%) and *Q<sup>b</sup>* is the value of the necessary battery capacity.

After these days it is necessary to recover the normal state of charge in the batteries in a not very long period, for this reason it is necessary to oversize the value PVpeak (for example 20%).

In this way, the batteries are designed for a sufficient capacity to supply the load for *n<sup>c</sup>* cloudy days.

#### *2.2. Reliability Assessment Methodology*

To evaluate the reliability of the SAPV system two questions must be addressed. An interruption of supply will occur if the instant power provided by the PV arrays plus the energy stored in the batteries is not enough to satisfy the demand. The second cause for an interruption of supply is when unexpected faults occur in the system.

The reliability model of a PV plant is a complex issue [42–44]. For the objectives of this paper an aggregated reliability model of the PV plant (panels and BC) is used (λc). The reliability model of the PV plant is assumed to be a two state model, with total power output in the up state and zero power output in the down state (Figure 4). Faults in the batteries are not considered as they are assumed to be maintained and substituted before ending its useful life period (so it is considered λ<sup>b</sup> = 0).

**Figure 4.** Up and down sequence for the photovoltaic (PV) generation unit.

Time to failure (TTF) of the PV plant is modeled using an exponential distribution and the mean time to repair (TTR) is modeled following a Rayleigh distribution [45]. TTF and TTR are randomly generated using the inverse transform method [46].

The PV generation curve is combined with the sequence of failures obtained, making it zero when there is a PV panel array or BC failure. In this way the generating capacity sequence (GCS) is obtained. The SOC of the battery is then calculated by combining the GCS with the demand sequence.

Considering the instantaneous power demand *Pd(t)* and *GCS(t)*, SOC (t) is obtained from Equation (7):

$$\text{SOC(t)} = \text{SOC(t-1)} + \alpha (\text{GCS}(\mathbf{t}) - \text{P}\_{\text{d}}(\mathbf{t})) \cdot \Delta \mathbf{t},\tag{7}$$

where α = η*c*, if *GCS(t)* − *Pd(t)* > 0 and α = 1/η*d*, if *GCS(t)* − *Pd(t)* < 0.

An interruption of supply due to generation inadequacy occurs when:

$$\text{SOC}(\mathbf{t}) \le \text{SOC}\_{\text{min}} \text{ and } \text{CCS}(\mathbf{t}) < \text{Pd}(\mathbf{t}). \tag{8}$$

With Equations (7) and (8) during PV failure periods, the energy provided by the battery is evaluated to reduce the failure time totally or partially until the battery is discharged to *SOCmin*. If a state in which P<sup>d</sup> cannot be supplied is reached, the frequency of interruptions (FOI in number of interruptions/yr) is increased and the loss of energy expectation (LOEE in *Wh*/*yr*), i.e., energy not supplied expectation, is evaluated as indicated in the description of the reliability analysis method.

If SOC(t) = *SOCmax* and *GCS(t)* > *Pd(t)* there is an energy production capacity that cannot be used. This energy will correspond to energy not used (ENU).

To apply the analysis method numerically, variables *Ps(t)*, and *Pd(t)* must be discretized. For each hour h of the year, the energy produced by the PV array is calculated and its value is assigned to the discrete variable *Ps(h).* Since its value corresponds to the average hourly power it can be expressed both in kW or in kWh equivalently. The *GCS(t)* is automatically discretized as *GCS(h)* if it is calculated with the discrete values of *Ps(h).* Similarly, the energy demanded in each hour is assigned to the discrete variable *Pd(h).*

## Description of the Reliability Analysis Method

A sequential Monte Carlo Simulation (MCS) is performed where the hourly behavior of the system for a series of years is simulated (400 years in this work). The randomized irradiance level for every hour of the day for every month of the year is considered, together with the hourly demand to construct chronological generation and demand random sequences.

The reliability evaluation method is performed as follows (Figure 5):

	- Initialize counters: h = 1 (number of simulated hours of the year); i = 0 (counter of interruptions); H = 0 (hours of interruption); LOEE = 0; ENU = 0, SOC = 80% (battery state of charge).
	- Simulate TTF and TTR consecutively to generate the annual failure sequence.
	- Obtain randomized hourly PV generation time series data Ps(h) from the historical record.
	- Generate the hourly chronological curve of annual demand Pd(h) from the historical record.
	- Combine Ps(h) and the annual failure sequence to get the generating capacity sequence GCS(h) for the simulated year.
	- FOR h = 1 to 8760 DO:
		- Using GCS(h) and Pd(h), obtain SOC(h) with Equation (7).
		- Update the number of interruptions i and evaluate the duration in hours of each interruption H<sup>i</sup>
		- If SOC = SOCmin and GCS(h)<Pd(h), update LOEE: LOEE = LOEE+Pd(h)-GCS(h)
		- If SOC = SOCmax and GCS(h)>Pd(h), update ENU: ENU = ENU+GCS(h)−Pd(h)
	- Evaluate FOI index: FOI = i.
	- Evaluate the loss of load expectation (LOLE) index: LOLE = P Hi (*h*/*yr*).
	- Evaluate the loss of load probability (LOLP) index: LOLP = 100·LOLE/8760.
	- Calculate average values of the indices for the n simulated years//Consider a possible convergence criterion (\*).

(\*) A possible convergence criterion is to stop the process if for 10 consecutive years, for example, the average values of the indices vary less than a fixed value.

**Figure 5.** Flowchart of the reliability evaluation method for the SAPV system.

#### **3. Results**

In this section, the reliability of supply is evaluated for a real SAPV system, using the proposed methodology.

On the one hand, using the registered demand data of a real consumer during a whole year, a deterministic design (DD) using equations (1) to (6) provides a design of PVpeak = 4 kW. Assuming *n<sup>c</sup>* = 4 days, it results in Q<sup>b</sup> = 35 kWh (Table 1). The selected case study developed corresponds to a dwelling, but the proposed method can be applied to larger SAPV systems.


**Table 1.** Generation and demand data.

On the other hand, using the MCS proposed several situations have been evaluated and different cases are obtained for the considered system.

First, depending on the load demand, PV generation and battery capacity, some days present a generation surplus, as represented in Figure 6a. The PV generation capacity exceeds the load demand and the SOC of the battery has reached SOCmax. This results in surplus power not used (PNU), due to a lack of energy storage capacity in the batteries, that would lead to surplus ENU.

Secondly, there are days when PV generation is low and the battery capacity is exhausted resulting in load curtailments and interruptions of supply to the user, as depicted in Figure 6b. These events result in a loss of energy expectation, or its instantaneous equivalent loss of power expectation (LOPE).

Lastly, in addition to the variability in PV generation, unexpected faults occur in the system that cause PV energy generation interruptions. These faults may result in interruptions of supply depending on the duration of the power interruption, the energy demand during that time and the energy stored in the battery (SOC). Figure 6c shows an example where the fault is successfully covered by the batteries and Figure 6d shows an example where the fault is not covered by the batteries.

The proposed MCS method enables the evaluation of reliability indices taking into account the uncertainty associated to variable PV generation, variable load demand and unexpected failures. The computed LOLE is shown in Figure 7 with an average value of 791.6 h/yr. This continuity of supply corresponds to an average LOLP of 9% and supposes that LOEE equals 450 kWh/yr (Figure 8). ENU is 1634 kWh/yr occurring during 1065 h when there is surplus non-used PV generation.

**Figure 6.** *Cont*.

**Figure 6.** (**a**) (15/4/2018) Day without failures and surplus energy not used (ENU) (ENU > 0); (**b**) (12/11/2018) Day without failures with an interruption due to a deficit of generation which is not supplied by the battery (state of charge (SOC) < SOCmin and LOEE > 0); (**c**) (20/4/2018) day with a failure that is supplied by the battery (SOC > SOCmin and LOEE = 0); (**d**) (11/11/2018) day with a failure which is not covered by the battery (SOC < SOCmin and LOEE > 0).

**Figure 7.** Evolution of average loss of load expectation (LOLE) along 400 simulations.

**Figure 8.** Evolution of average loss of energy expectation (LOEE) along 400 simulations.

This MCS method provides not only long-term average values but also the reliability indices distribution, useful to determine possible extreme values. Figure 9—(a) shows the yearly histograms of the number of failures in the generation system, (b) FOI due to a deficit of PV generation and energy storage, including the effect of the system failures, (c) LOLE and (d) ENU.

**Figure 9.** (**a**) Histograms of the number of failures in the generation system, (**b**) number of supply interruptions, (**c**) LOLE and (**d**) ENU.

Although average values of 2 failures per year, 83 energy deficit events, 790 h of LOLE and 1634 kWh of ENU are estimated, more extreme values are not unlikely.

#### **4. Discussion**

Results displayed in Figure 6 show that several factors interplay to determine supply interruptions in the system. Insufficient PV generation can lead to power interruptions when the battery energy is exhausted (Figure 6b). Component faults unexpectedly occur that can lead to an interruption of supply depending in the outage duration and the SOC of the battery (Figure 6c,d). The uncertainty associated to solar radiation, demand and component faults and the dependency on previous states of the system (battery SOC) to determine power outages make the evaluation of the system's expected reliability indices complex.

A method based on sequential MCS is then convenient as proposed in this work. Figures 7 and 8 show that when the number of simulations increases, the reliability indices converge to a steady state value.

One advantage of the proposed method is that the distribution of the reliability indices can be easily obtained (Figure 9). These distributions allow considering extreme values in the design of the SAPV system that are unlikely to happen.

A sensitivity analysis is performed to evaluate the impact of different parameters of the SAPV generation system in the reliability indices. The average failure tax and repair time will affect the number and duration of the interruptions. The capacity of the batteries and the installed peak power of the PV panels will affect the reliability indices experienced by the user in stand-alone installations.

The continuity of supply depends on the failure tax of the system. Figure 10a,b show, respectively, how the LOLP and LOEE indices increase linearly when the failure rate λ<sup>c</sup> increases from 0.1 f/yr to 20 f/yr for a fixed average repair time of 24 h.

**Figure 10.** (**a**) loss of load probability (LOLP) and (**b**) LOEE evolution when λ**c** varies from 0.1 to 20 f/yr.

In the proposed method system failures are just one of the three different sources of uncertainty considered (PV generation, power demand, and component faults). In fact, the main contribution to the outage time is due to deficits in the energy balance. As shown in Figure 10 for negligible failure rates, resulting in years with no faults in the PV components, power interruptions occur due to insufficient PV generation and stored energy in the batteries. Reliability indices for years with no component faults are LOLP = 9% and LOLE = 440 kWh for the installed PV power and battery storage resulting from the deterministic design. These results show that the design method based on worst case scenarios can lead to inadequate continuity of supply levels.

The average repair time after any failure also affects the reliability indices. Figure 11a,b show, respectively, how the LOLP and LOEE indices increase linearly when the average TTR increases from 6 h to 7 days for a fixed failure tax of 2 f/yr.

**Figure 11.** LOLP (**a**) and LOEE (**b**) evolution when time to repair (TTR) varies from 6 h to 7 days.

Higher installed peak power (PVpeak) in the PV panels will suppose increased self-sufficiency of the SAPV system. A sensitivity analysis is performed for variable PVpeak with fixed failure rate (2 f/yr), repair time (24 h) and for several values of Qb, from 10 to 50 kWh. Figure 12 shows how increasing PVpeak from 1 kWp to 8 kWp improves exponentially the LOLP, while further increases result in marginal improvements. The same can be said about the LOEE.

**Figure 12.** LOLP evolution when PVpeak varies from 1 to 8 kWp for several values of Qb.

Another sensitivity analysis is performed for variable Q<sup>b</sup> and fixed λ<sup>c</sup> (2 f/yr), repair time (24 h) and varying PVpeak from 4 to 10 kW. Figure 13 shows how increasing Q<sup>b</sup> from 1 kWh to 25 kWh improves exponentially the LOLP, while further increases result in a much slower decrease. The same can be said about the LOEE. Similarly, the ENU is reduced as Q<sup>b</sup> increases up to the same limit of Q<sup>b</sup> = 25 kWh where further reductions are marginal.

**Figure 13.** LOLP evolution when Q<sup>b</sup> varies from 1 to 95 kWh for several values of PVpeak.

The results shown demonstrate the usefulness of reliability analysis based on MCS to improve the design of SAPV systems. It is observed that the optimal values of Q<sup>b</sup> are between 20 and 30 kWh, while higher values provide very little improvements to the system. Likewise, it is observed that the installed power PVpeak must be increased to values between 6 and 8 kW in order to obtain a LOLP around 2%. Comparing to the initial deterministic design (PVpeak = 4 kW and Q<sup>b</sup> = 35 kWh are shown as circled points in Figures 12 and 13), this means that it is more convenient to increase PVpeak and slightly reduce Q<sup>b</sup> (to a value of 25–30 kWh). This type of design is the only one that can guarantee continuity in the supply, based on statistical values.

The main contributions of this work are:


#### **5. Conclusions**

The application of the developed probabilistic method enables the evaluation of reliability indices according to the system parameters. From the experimental results it can be concluded that, as one would expect, the values of loss of load probability (LOLP) or loss of energy expectation (LOEE) indices increase as the average failure rate per year and the average repair time increase. However, this increase is very small for moderate increases in the failure rate in the panels and battery controller (between 1 failure every 10 years and 3 failures per year) and in the average time to repair (between 6 and 24 h). This analysis makes it possible to evaluate the utilization of solar panels with greater robustness and an assistance service with a limited response time.

From the sensitivity analysis carried out in the presented case study, it is concluded that using a standard deterministic method based in worst case scenarios, as stated in Section 2, for the initial design of the installed power of photovoltaic panels and the battery capacity criteria results in inadequate levels of reliability. The developed method allows the determination of the expected continuity of supply values based on the energy parameters of the installation and the probability distributions associated with failure and repair times. It is concluded that the deterministic design results in an undersized installed power in PV panels and an oversized battery capacity.

The reliability assessment method proposed by the authors in Section 3 enables an optimal design of the installed power and the energy storage for the desired reliability in the system supply (a LOLP around 2%).

The use of renewable energy sources, such as solar energy, is one of the keys to improving energy sustainability. However, the intermittent nature of these sources and their difficult management are great disadvantages for their use. To improve these aspects, energy storage systems are used. As demonstrated in this paper, if the design of renewable energy facilities is not carried out with appropriate criteria, supply interruption periods may be inadmissible. Very few consumers would be willing to rely on energy systems that cause frequent interruptions. The article demonstrates that a design based on reliability parameters is possible, providing higher security to users and improving their confidence in these sustainable energy systems.

The obtained results can also be useful to guide the design of new installations in the same geographical area. In addition, the presented analysis method enables to draw conclusions about the modifications needed in a facility in which the admissible reliability criteria are not met. The results prove which measures will be more effective to improve reliability: increase batteries capacity, installed peak power or both at once. Without this kind of analysis, some extensions of facilities could be ineffective, have a very low impact or represent a very high and unjustified cost.

The presented case study and the discussion of the results show that in this type of stand-alone systems there are situations of lack of energy as well as moments with ENU due to an excess of generation. The diversity of consumers and their energy systems suggests studying the creation of small microgrids formed by several facilities such as the one described in this paper, interconnected by the grid, so that these facilities can be supported by energy transfers between them. Reliability analysis of these small communities linked in microgrids is now an objective for future research of the authors.

**Author Contributions:** Conceptualization and methodology, E.Q. simulation software design C.R.-B. experiments design and validation, all authors; formal analysis, E.Q. and C.R.-P. data curation C.R.-B., G.E.-E., and E.Q. writing-original draft preparation, E.Q., C.R.-B., and G.E.-E. writing-review and editing, all authors; supervision E.Q. and C.R.-P. funding acquisition, all authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work has been supported by research funds of the Universitat Politècnica de València.

**Acknowledgments:** This work has been supported by the Universitat Politècnica de València.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Abbreviations**


## **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Perspective* **Assessment on Global Urban Photovoltaic Carrying Capacity and Adjustment of Photovoltaic Spatial Planning**

**Siyuan Chen, Yukun Zhang and Jie Zheng \***

School of Architecture of Tianjin University, Tianjin 300000, China; miracle1987@126.com (S.C.); zyk.tj@163.com (Y.Z.)

**\*** Correspondence: tjuzhengjie@163.com

**Abstract:** To promote the effective combination of photovoltaic (PV) utilization and urban development, this study proposes that solar PV generation should be taken as an important resource and environmental carrying capacity factor, which is defined as "Photovoltaic Carrying Capacity (PVCC)", to be integrated into future urban planning. According to the PVCC assessment on global cities, the sensitivity of PVCC to different influence factors is analyzed and the benefits of different optimization strategies on PVCC are also discussed. Additionally, in an equilibrium analysis of PVCC distribution in six sample countries, we discuss both the causes and impact of their distribution gap. That analysis shows that the average PVCC in global urban areas can reach 23.13%, which approaches the anticipated needs of PV development in cities by 2050.Though different optimization strategies may be feasible to promote urban PVCC, they might only make significant changes in medium or large cities. When considering the high-efficient utility of local energy, dispersed layout of cities and population is necessary to implement PV spatial planning. According to the assessment and analysis results, the adjustment suggestions of PV spatial planning in each sample country are also discussed. Finally, it is pointed out that the proactive PV spatial may be of great significance to achieve higher solar energy supply and PVCC will be an available cognition in guiding this planning in the future.

**Keywords:** photovoltaic spatial planning; photovoltaic carrying capacity; influence factors; optimization strategies; carrying capacity distribution; planning adjustment

#### **1. Introduction**

With the depletion of fossil energy, countries all over the world have put forward their own energy transition plans. Renewable energy, mainly solar photovoltaic (PV) have increasingly become the mainstream development trend of global energy transition due to its more general applicability and sustainability. By the end of 2018, the world's total installed photovoltaic capacity had reached to about 480 GW, which had contributed 2% of the world's total energy consumption [1]. Meanwhile, the cost of PV generation per kilowatt hour has also shown a trend of rapid decline. Taking utility scale PV as an example, the average cost per kilowatt hour of the electricity has been reduced from 371 USD/MWh in 2010 to 85 USD/MWh in 2018, dropping by 77% in only eight years [2]. Although there is still a cost gap compared with the traditional energy, such as thermal power generation, solar PV is catching up with traditional energy to achieve connection to grid at an equal price and its competitive advantage constantly improves. An estimate from the International Renewable Energy Agency (IRENA) shows that, the total global installed PV capacity will reach 8519 GW by 2050, including about 60% utility scale and 40% distributed rooftop and the power output will take up 25% of the world's total electrical energy consumption [1], becoming a significant part of the energy consumption structure.

The energy transition to PV (and many other renewable energies) is not in doubt, but compared with traditional fossil energy, solar PV has particular characteristics, such as intermittence and lower energy density, which will necessitate a significantly different planning mode. Currently, comprehensive assessments of its sustainability (including

**Citation:** Chen, S.; Zhang, Y.; Zheng, J. Assessment on Global Urban Photovoltaic Carrying Capacity and Adjustment of Photovoltaic Spatial Planning. *Sustainability* **2021**, *13*, 3149. https://doi.org/10.3390/su13063149

Academic Editors: M. Dolores Esteban and Maria Malvoni

Received: 20 January 2021 Accepted: 10 March 2021 Published: 12 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

energy production, transportation and consumption) are not generally carried out [3]. To date, for the most part, spatial planning and energy planning have been treated as two separate domains. So-called "Energy Planning" is usually just some target-based guidelines which are largely concerned with the security of supply and energy demand. However, without resource distribution analysis and spatial construction planning, there will be obvious deficiencies in the guidance of local investment in the construction of renewable energy facilities. As a result, renewable energy planning is still carried out in the same way as for traditional fossil energy, becoming thereby subordinated to urban planning in the form of power distribution only. With regard to this background, many researchers have proposed the application of "Landscape: into energy planning, putting forward the concept of "Energy Landscape" [3–5]. According to the spatio-temporal modeling of renewable energy carriers' distribution, a comprehensive optimized utilization strategy could be achievable through methods of planning adjustment, policy incentive and financial support, etc. In that case, the goal of future renewable energy planning will no longer be simply limited to the development expectations of a single increasing of energy consumption proportion, but will be more inclined to the deployment of both regional energy distribution and urban development in a view of comprehensive spatial planning.

Energy is of great significance in the process of urban development. It is not only an indispensable natural resource, but also a necessary condition for the transformation of other natural resources into economic output. From the economic perspective, the essence of urban development is just a process of population accumulation during which natural resources are exhausted for economic production by converting their economic mode [6]. So, like many other natural resources, energy adequacy is usually reflected as an important "Resource and Environment Carrying Capacity (RECC)", which is initially used to assess the environmental constraints on population growth [7]. RECC has been defined by FAO and UNESCO as the maximum population that a country or region can afford by using local natural resources, energy, technology and technology [8]. However, with the development of worldwide research, its measurement has been gradually extended to include many other criteria, such as "Ecological Footprint", "Net Primary Productivity" and "Emergy" [9–11]. Although there is not yet a universal definition, the connotation of RECC allows the measurement of the resource endowment and environmental capacity of a certain regional space and the population and economic scale it could afford so as to maintain a good ecosystem without doing any harm to the natural ecological environment [12].

A great quantity of researches of RECC can be found on fossil energies like coal or gas-fired power generation [13–16], but there is as yet none on solar PV. The obvious reason is that these energy forms are generally considered as non-renewable. They are also affected to the stability and security of their energy supply. Compared with fossil energy forms, solar PV has not yet had any serious impact on either urban or economic development. That said, the proportion of PV power consumption will increase significantly by 2050. Solar PV will play the same role as other power generation technologies, providing important energy guarantees in the process of urban development in the future. Furthermore, according to the goal of "Energy Landscape" planning, solar PV will also become a key part of future urban spatial planning. We therefore propose, in this study, to take solar PV as a latent but necessary resource and environment carrying capacity factor that should be integrated into the comprehensive carrying capacity index system of future urban planning. On this basis, the distribution of PVCC in cities around the world is evaluated, then the deployment and further optimization strategy of future PV energy in urban spatial planning are also discussed in this study. These will help to make a tentative discussion on how PVCC would affect the strategy determination of solar PV spatial planning in a range that from urban planning to national geospatial planning.

#### **2. Method**

#### *2.1. Photovoltaic Carring Capacity*

According to the connotation of RECC, the "Photovoltaic Carrying Capacity (PVCC)" here in this paper is defined as the capacity of a city to use solar energy resource in its region for PV power generation to meet its own demand, as is shown in Equation (1). By the differences of PV system construction forms, the evaluation objects include two parts of power output: distributed rooftop PV system in urban built-up areas and utility scale PV stations in non-built-up areas.

$$PV\text{CC} = \frac{E}{\text{C}}\tag{1}$$

where *PVCC* is the photovoltaic carrying capacity of a city, measured in %; *E* is the annual power output that can be achieved by a city after PV installation, measured in kWh/year; *C* is the annual power consumption of a city, measured in kWh/year.

#### *2.2. Influence Factors*

According to Equation (1), the influence factors of PVCC can be generally divided into two main aspects: power generation and power consumption. On the basis of some existing researches, the generation aspect includes natural conditions such as solar irradiation, effective sunshine time, latitude, slope, geology, natural disasters, land usage, protected area, temperature, snow, dust, etc. and other spatial morphology factors like building density, spatial layout, building style, urban built up area, population density, population distribution, infrastructure (roads, transmission lines, substations) distribution, etc. [17–20]. The consumption aspect is often related to the living standard or urban development condition of a city, such as urbanization rate, industrial structure, power consumption level, national economy and spot price [21–24]. The assessment results of PVCC are closely related to the influencing factors selected. However, this does not mean that the PVCC of a city is always immutable. For instance, the change of urban spatial morphology by adjusting its planning strategy can significantly affect the city's final PVCC. In addition, considering the convenience and feasibility of the assessment, it is also necessary to define the range of influence factors selected according to different spatial assessment scales. Since the objective discussing spatial scale of this study is set as the range from urban planning to national geospatial planning, considering the replication of these assessments on such geographical scope (worldwide), we take only six main factors into consideration in this study, each of which is supposed to make more significant impact on final PVCC result than the other detailed factors. They are solar irradiation, urban population density, urban built-up area, annual electricity consumption per capita, available land area for PV installation and transmission distance. Moreover, each of these six factors could correspond to at least one independent optimization strategy (without causing any change of other influence factors) directly or indirectly, as shown in Table 1. In addition, these strategies will be later used to underpin a theoretical discussion on the effect that it would make through different methods of planning adjustment. As for the assessment of power consumption, considering the real situation of urban economic production, built-up area will be the main part where energy is exhausted (Almost 80% of total social electricity consumption) [25], so we only take this part of consumption into account for the PVCC assessment.

#### *2.3. Urban Geographical Division*

A key quality of future energy systems is low transmission distance to realize a highefficient utility of local energy [3]. Therefore, it is necessary to delimit the boundary of "local" region first to determine to which city the PV power output (Especially for utility scale) will be mainly supplied. On consideration of transmission line loss and infrastructure investment, PV power output is usually preferentially transmitted to its nearest urban load area to minimize extra consumption, therefore, the urban region division here in this study is determined by the geographical distance between the land location and its nearest urban center by using the cost allocation function in Arcgis, instead of a traditional

administrative division. According to the quality of low transmission distance, this method can ensure that the distance between any land pixel in a certain geographical division and its subordinate urban center is shorter than that from any other urban centers. Then, by extracting the average value of each influence factor in a certain geographical division (by zonal statistics function in Arcgis), the local condition of a city could be further described in brief.

**Influence Factors Corresponding Optimization Strategies** Solar irradiation Choosing PV panels of higher conversion efficiency, PV installation on building facades or using inclined roof to gain more irradiation in the same land area Urban population density When the urban built-up area is constant, population density will change with the variation of building density Built-up area Urbanization in spatial dimension, urbanization or counter-urbanization, make more land into built-up area or in the opposite Electricity consumption per capita Reduce building energy consumption, change people's electricity consumption habits or develop low-carbon economy Available land area Exploit more land to achieve more installation area (for utility scale photovoltaic stations) Transmission distance Closer site selection to urban load to decrease transmission line loss

**Table 1.** Influence factors of PVCC and its corresponding optimization strategies.

#### *2.4. Assessment of PV Generation Potential*

Different assessment methods of generation potential are adopted in two forms of PV system. As for distributed rooftop, the key point is to evaluate the PV installation area on the roof of a city. In this regard, the International Energy Agency (IEA)has given out an exponential function to describe the relationship between the installation area per capita and the urban population density [26], in order to meet the needs of PV potential assessment in a large range of geographical area, as is shown in Equation (2). Applying this, the annual power generation of distributed rooftop PV can be further calculated by Equation (3).

$$A\_{\mathfrak{a}} = \mathfrak{a} \cdot \mathfrak{p}^{-\beta} \tag{2}$$

$$E\_d = A\_d \cdot A\_b \cdot p \cdot GTI \cdot i \cdot K \tag{3}$$

where *A<sup>a</sup>* is the rooftop PV installation area per capita, measured in m2/person; *p* is the urban population density, measured in persons/km<sup>2</sup> ; *α* and *β* are theadjustment coefficients, unless otherwise specified, *α* = 172.3, *β* = 0.352; *E<sup>d</sup>* is the annual power generation of distributed rooftop PV, measured in kWh/year; *A<sup>b</sup>* is the urban built-up area, measured in km<sup>2</sup> ; *GTI* is annual solar irradiation on tilted surface, measured in kWh/m2/year; *i* is the conversion efficiency of PV panel, measured in %; *K* is the total efficiency of PV system, *K* = 75% in general.

As for utility scale PV power stations, the key factors include two main parts, the evaluations of the available land area and the transmission distance. According to a standard regulation of China [27] and our field investigation, four kinds of land usage are selected to be available for PV installation and each these four can be further reflected as a corresponding land cover type on basis of the "Land Cover Classification System (LCCS)" [28], including bare areas (LCCS-200), sparse vegetation areas (LCCS-150), mosaic vegetation/cropland areas (LCCS-30) and closed to open herbaceous vegetation areas (LCCS-140). The annual power output can be then further estimated after considering an average 40% ratio of panels area to land occupation and the adjustment coefficient of each available land type (to avoid vegetation occupation, see appendix II in [28]). Meanwhile, considering the lack of or absence of feasibility of installing PV over steep terrains, land with an average slope larger than 5◦ is excluded by referring to the value in [29] and the pixel size of the DEM data used in this study (2397 × 2397 m). The assessment of

transmission line loss is derived by extracting an average value of the distances between each available land pixel and its subordinate urban center. Moreover, an average line loss rate of 0.4%/km is considered in assumption that all the transmission lines are constructed in the system form of 10 kV three-phase four wires. Then, the annual PV power received by each urban built-up area from its surrounding utility scale PV stations could be further estimated as follows:

$$E\_{ll} = 40\% \cdot \left(\sum\_{i}^{n} A u\_{i} \cdot R\_{i}\right) \cdot GTI \cdot i \cdot K \cdot \left(1 - 0.4\% \cdot D\right) \tag{4}$$

where *E<sup>u</sup>* is the annual power received from utility scale PV, measured in kWh/year; *Au<sup>i</sup>* is the total area of each type of land usage (*I* = 30, 140, 150, 200), measured in m<sup>2</sup> ; *R<sup>i</sup>* is the adjustment coefficient of each available land type (R<sup>30</sup> = 50%, R<sup>140</sup> = 70%, R<sup>150</sup> = 90%, R<sup>200</sup> = 100%); *D* is the average transmission distance, measured in km. *GTI*, *i* and *K* see Equation (2).

#### *2.5. PVCC Assessment*

Through methods above, it can be estimated that the global PV installed capacity could reach a level of 8.58 TW for distributed rooftop and 3704 TW for utility scale, which has far exceeded the development goal in 2050 (3470.6 GW for distributed rooftop, 40%; 5111.4 GW for utility scale, 60% [1]). This just reflects a theoretical installed capacity. While considering the feasibilities of technical condition, development cost and industry capacity, such amount of installation would be hardly achieved, therefore, a hypothetical restrict of maximum installed capacity is further needed to reflect the realistic situation. By comparing the ratio of development goal to theoretical installation, on average, merely 0.14% of available land area and 39% of roof area are needed to meet 25% proportion of PV power in 2050. Therefore, the PVCC of each urban geographical division can be further evaluated as following equation:

$$PV\text{CC} = \frac{39\% \cdot E\_d + 0.14\% \cdot E\_u}{A\_b \cdot p \cdot \text{C}\_a} \tag{5}$$

where *C<sup>a</sup>* is the annual electricity consumption per capita; The other symbols see Equations (1)–(4).

#### *2.6. Data Sources*

The data used in this study include global digital elevation model (DEM) [30], Globcover 2009 [31], Gridded Population of the World, Version 4 (GPWv4) [32], Global irradiation for optimally tilted surface [33] and Urban population gathering points (urban centers). The power consumption is estimated based on the statistics data of "Electricity consumption per capita" from [34,35]. Although they only reflect the electricity consumption of the overall level of a country, while taking into consideration that the citizens of the same country would have certain similarities in their electricity consumption behaviors, habits or customs, these data are still considered as valid. Meanwhile, to ensure the consistency of assessment results, the conversion efficiency of PV panel is unified as 16.54% by referring to Jkm270pp-60-dv poly-silicon PV module [36].

#### **3. Results and Analysis**

#### *3.1. Results of PVCC Assessment*

The assessment result of each urban division worldwide is shown in Figure 1. In addition, the annual PV power generation from all urban divisions would take up in a total ratio of 43.22% to the global electricity demand. Considering that over 30% of this power output might not be used directly or even wasted because of intermittence and electrical-surplus in off-peak period, this assessment result is relatively reasonable. In addition, it can be seen from Figure 1 that the vast majority of PVCC are provided by deserts, tundra and other sparsely populated areas in the form of ground utility scale PV

station. Unless absolutely necessary, these sparsely populated areas are always suggested to better remain unexploited to protect their ecology and biodiversity. Generally, it is not essential to discuss the PVCC of these areas in priority. Therefore, by utilizing a regulation applicable in China [37], the standard of urban scale with a population of 50,000 is adopted in this study to distinguish urban and non-urban areas. On this basis, the average level of PVCC in urban areas worldwide could still reach up to 23.13%, nearly approaching the target of 25% by 2050. In other words, the global PV target of 2050 could be almost achieved only by PV installation in urban areas themselves together with methods such as low-priced energy storage and electric vehicles recharging (to avoid electrical-surplus), though this is just on the basis of global average level. and other sparsely populated areas in the form of ground utility scale PV station. Unless absolutely necessary, these sparsely populated areas are always suggested to better remain unexploited to protect their ecology and biodiversity. Generally, it is not essential to discuss the PVCC of these areas in priority. Therefore, by utilizing a regulation applicable in China [37], the standard of urban scale with a population of 50,000 is adopted in this study to distinguish urban and non-urban areas. On this basis, the average level of PVCC in urban areas worldwide could still reach up to 23.13%, nearly approaching the target of 25% by 2050. In other words, the global PV target of 2050 could be almost achieved only by PV installation in urban areas themselves together with methods such as low-priced energy storage and electric vehicles recharging (to avoid electrical-surplus), though this is just on the basis of global average level.

habits or customs, these data are still considered as valid. Meanwhile, to ensure the consistency of assessment results, the conversion efficiency of PV panel is unified as 16.54%

The assessment result of each urban division worldwide is shown in Figure1. In addition, the annual PV power generation from all urban divisions would take up in a total ratio of 43.22% to the global electricity demand. Considering that over 30% of this power output might not be used directly or even wasted because of intermittence and electricalsurplus in off-peak period, this assessment result is relatively reasonable. In addition, it can be seen from Figure 1 that the vast majority of PVCC are provided by deserts, tundra

**Figure 1.** Assessment on PVCC of global urban divisions.

**Figure 1.** Assessment on PVCC of global urban divisions.

*Sustainability* **2021**, *13*, 3149 6 of 13

by referring to Jkm270pp-60-dv poly-silicon PV module [36].

**3. Results and Analysis** *3.1. Results of PVCC Assessment*

#### *3.2. Sensitivity Analysis of PVCC*

*3.2. Sensitivity Analysis of PVCC* As mentioned above, the PVCC of a city will change with its influence factors. This section analyses the sensitivity of PVCC result to changes in the influence factors inputs. While considering the effects of different urban types to analysis accuracy, these analyses are conducted separately by different population scales based on a classification standard of China [37], specifically including seven scales that range from miniature city to mega city. For each type of urban scale, the median value of each influence factor is extracted as an assumed global average standard input into the PVCC equation. It is then adjusted independently of the others by ±50% (theoretical maximum and minimum) to observe its sensitivity and the results are shown in Figure 2. These analyses will further help to illustrate how different optimization strategies could make changes on the PVCC result, according to Table 1. As mentioned above, the PVCC of a city will change with its influence factors. This section analyses the sensitivity of PVCC result to changes in the influence factors inputs. While considering the effects of different urban types to analysis accuracy, these analyses are conducted separately by different population scales based on a classification standard of China [37], specifically including seven scales that range from miniature city to mega city. For each type of urban scale, the median value of each influence factor is extracted as an assumed global average standard input into the PVCC equation. It is then adjusted independently of the others by ±50% (theoretical maximum and minimum) to observe its sensitivity and the results are shown in Figure 2. These analyses will further help to illustrate how different optimization strategies could make changes on the PVCC result, according to Table 1. *Sustainability* **2021**, *13*, 3149 7 of 13

> The total PVCC of 23.13% in urban areas worldwide would approximately meet the development needs in 2050, but this is just from the view of total average. As is shown in Figure 1, the PVCC of different urban divisions are actually changing in varying degrees. That is also to say, in order to achieve the appropriate proportion of PV energy consumption in each urban division, long distance energy allocations are inevitable, but this would definitely break the key quality of low transmission distance and high-efficient local utility of future energy systems. In order to describe this imbalance distribution of PVCC, six countries (or regions) are extracted as the contrasting analyses in this study, being: China, USA, Europe (partial), Australia, Brazil and India. The equilibrium of the PVCC distribution can be quantified as a "Gini index" which is commonly used as an indicator to measure the income gap of residents in a country or region. As shown in Equation (6), the Gini index is a value between 0 and 1, where 0 indicates the absolute equilibrium distribution of PVCC and 1 indicates the totally unequal distribution and 0.4 is generally used as the warning line to make a distinction between equilibrium and inequilibrium according to the definition of United Nations [38].It is worth noting, however, that the Gini index will change with the selected objects that are taken into calculation and that, according to the results of global assessment, the PVCC would usually decrease with the increase of urban population. Therefore, the equilibrium analysis of each county is conducted in the range of different urban scale sections, such as miniature cites and above and small cities and above (shown in Figure 3), and this would be used to discover in which section the largest distribution gap will occur. By combination then with the average level of PVCC and annual energy acquisition per capita in each urban scale (shown in Table 2), this would help to make further discussions on the adjustment strategies of the PV spatial planning in each

**Figure 2.** Sensitivity analysis of PVCC to influence factors in different urban scales. Note: The length of each colored bar indicates the range of percentage of PVCC change by adjusting each influence factor to ±50%. In addition, the urban types are classified by its population scale, Miniature city: 50,000–200,000, Small city: 200,000–500,000, Medium city: 500,000– 1,000,000, Large city I: 1,000,000–3,000,000, Large city II: 3,000,000–5,000,000, Super city: 5,000,000–10,000,000, Mega city: >10,000,000. **Figure 2.** Sensitivity analysis of PVCC to influence factors in different urban scales. Note: The length of each colored bar indicates the range of percentage of PVCC change by adjusting each influence factor to ±50%. In addition, the urban types are classified by its population scale, Miniature city: 50,000–200,000, Small city: 200,000–500,000, Medium city: 500,000–1,000,000, Large city I: 1,000,000–3,000,000, Large city II: 3,000,000–5,000,000, Super city: 5,000,000–10,000,000, Mega city: >10,000,000.

=

2

1

<sup>2</sup> · ̅̅̅̅̅̅̅̅∑∑| <sup>−</sup>

is the PVCC of the jth city;

=1

where is the Gini index; is the number of cities selected; ̅̅̅̅̅̅̅̅ is the average value

=1


(6)

is the PVCC of the ith

country or region.

city.

of PVCC in cities selected;

**Urban Scale**

Miniature cities

Small

Medium

Large

Large

Super

Mega

#### *3.3. Equilibrium Analysis of PVCC Distribution*

The total PVCC of 23.13% in urban areas worldwide would approximately meet the development needs in 2050, but this is just from the view of total average. As is shown in Figure 1, the PVCC of different urban divisions are actually changing in varying degrees. That is also to say, in order to achieve the appropriate proportion of PV energy consumption in each urban division, long distance energy allocations are inevitable, but this would definitely break the key quality of low transmission distance and high-efficient local utility of future energy systems. In order to describe this imbalance distribution of PVCC, six countries (or regions) are extracted as the contrasting analyses in this study, being: China, USA, Europe (partial), Australia, Brazil and India. The equilibrium of the PVCC distribution can be quantified as a "Gini index" which is commonly used as an indicator to measure the income gap of residents in a country or region. As shown in Equation (6), the Gini index is a value between 0 and 1, where 0 indicates the absolute equilibrium distribution of PVCC and 1 indicates the totally unequal distribution and 0.4 is generally used as the warning line to make a distinction between equilibrium and inequilibrium according to the definition of United Nations [38].It is worth noting, however, that the Gini index will change with the selected objects that are taken into calculation and that, according to the results of global assessment, the PVCC would usually decrease with the increase of urban population. Therefore, the equilibrium analysis of each county is conducted in the range of different urban scale sections, such as miniature cites and above and small cities and above (shown in Figure 3), and this would be used to discover in which section the largest distribution gap will occur. By combination then with the average level of PVCC and annual energy acquisition per capita in each urban scale (shown in Table 2), this would help to make further discussions on the adjustment strategies of the PV spatial planning in each country or region.

$$G = \frac{1}{2n^2 \cdot \overline{PVCC}} \sum\_{j=1}^{n} \sum\_{i=1}^{n} |PVCC\_j - PVCC\_i| \tag{6}$$

where *G* is the Gini index; *n* is the number of cities selected; *PVCC* is the average value of PVCC in cities selected; *PVCC<sup>j</sup>* is the PVCC of the *j*th city; *PVCC<sup>i</sup>* is the PVCC of the *i*th city. *Sustainability* **2021**, *13*, 3149 8 of 13

**Figure 3.** Gini indices of PVCC distribution in each urban scale section of different countries. Note: The definition of Gini index value: 0.0–0.2 Absolutely average, 0.2–0.3 Comparative average, 0.3–0.4 Relatively reasonable, 0.4–0.5 Large disparity, 0.5–1.0 Great disparity. The value of 0.4 is generally considered as the warning line. **Figure 3.** Gini indices of PVCC distribution in each urban scale section of different countries. Note: The definition of Gini index value: 0.0–0.2 Absolutely average, 0.2–0.3 Comparative average, 0.3–0.4 Relatively reasonable, 0.4–0.5 Large disparity, 0.5–1.0 Great disparity. The value of 0.4 is generally considered as the warning line.

**Table 2.** Average level of PVCC and annual energy acquisition per capita in each urban scale.

cities 62.1% 18.8% 34.8% 16.2% 64.1% 248.0% 203.2% 2568.3 2287.9 1678.3 1556.2 1513.2 1794.1 2132.2

cities 26.7% 12.3% 29.3% 11.3% 44.8% 257.1% 82.7% 1193.7 1597.9 1302.7 1083.7 1082.3 1690.3 1389.1

cities I 20.8% 10.9% 21.4% 14.2% 36.9% 175.0% 88.8% 929.5 1376.6 1088.0 1370.6 891.5 1209.9 1141.4

cities II 18.8% 9.0% 17.3% 10.1% 32.3% 99.2% 40.5% 820.1 1198.9 881.6 972.7 780.1 798.1 891.8

cities 17.0% 7.3% 19.7% - 26.3% 99.9% 65.1% 732.4 988.9 936.3 - 635.0 694.5 782.6

cities 13.9% 5.1% 8.9% - 27.8% 109.9% 34.9% 600.2 670.9 519.5 - 671.9 652.7 611.6 Note: The data in this table are extracted from the global PVCC assessment results of this study, by calculating the average

*4.1. How Could Different Optimization Strategies Make Changes on the PVCC?*

Necessarily, as shown in Figure 2, the PVCC will be sensitive to each influence factor input in different degrees. Therefore, the beneficial effects from different methods of planning adjustment (Table 1) will change as well. For instance, reducing energy consumption will always be the most direct way to improve PVCC; methods of building morphology optimization and decreasing building density will also be positive in promoting irradiation acquiring; and it might not be the most efficient way to improve PVCC by building utility scale stations closer to urban load, because this merely reduces the transmission loss. Different optimization strategies will all help to improve PVCC more or less, but it is also worth noting that the beneficial effects brought by these optimization strategies will decrease with the upgrading of urban scale. In fact, their actual benefits might not always

level of each urban scale, according to the scope of each sample country.

**4. Discussions**

146.8% 38.8% 52.9% 27.1% 118.7% 411.7% 428.6% 5887.2 4948.7 2599.4 2603.4 2879.5 3236.7 4776.9


**Table 2.** Average level of PVCC and annual energy acquisition per capita in each urban scale.

Note: The data in this table are extracted from the global PVCC assessment results of this study, by calculating the average level of each urban scale, according to the scope of each sample country.

#### **4. Discussions**

#### *4.1. How Could Different Optimization Strategies Make Changes on the PVCC?*

Necessarily, as shown in Figure 2, the PVCC will be sensitive to each influence factor input in different degrees. Therefore, the beneficial effects from different methods of planning adjustment (Table 1) will change as well. For instance, reducing energy consumption will always be the most direct way to improve PVCC; methods of building morphology optimization and decreasing building density will also be positive in promoting irradiation acquiring; and it might not be the most efficient way to improve PVCC by building utility scale stations closer to urban load, because this merely reduces the transmission loss. Different optimization strategies will all help to improve PVCC more or less, but it is also worth noting that the beneficial effects brought by these optimization strategies will decrease with the upgrading of urban scale. In fact, their actual benefits might not always be as optimistic or even be limited. All optimization strategies show greater benefits in smaller urban scale such as miniature cites and small cities, but it can be seen from Table 2 that the original PVCC of these cities are already in a high level and that they are capable of meeting the PV consumption proportion even without any further adjustment. That said, the benefits of such optimization strategies in larger urban scales would probably become very limited. Taking the assumed global standard in this study as an example, by using all these optimization strategies, the PVCC in mega cities and super cities are likely to be improved on average, from 14.4% to 19.9% and 17.6% to 24.5%, respectively. That will, in most cases, fall short of meeting the desired 25%. Considering the difference of original PVCC in each country and the actual implementation difficulty, it could be roughly concluded that these optimization strategies might be effective only in urban scales that range from medium to large cities.

#### *4.2. In Which Section Will the Largest PVCC Distribution Gap Occur?*

As shown in Figure 3, the largest gap of PVCC distribution occurs between urban and non-urban areas. This reflects the fact that most of the PVCC are provided by sparsely populated areas and this situation is almost the same for all countries. In addition, this distribution gap will appear obviously between the medium cities and small cities in China and in miniature cities and small cities in USA and India. To illustrate the cause of these differences in PVCC distribution, we also conducted an equilibrium analysis of the distribution of influence factors in each country applying the same method as in Equation (6). If the distribution of certain influence factor tends to be even, then it cannot be the main reason to cause the gap of PVCC distribution. As shown in Table 3, the distribution of solar irradiation in each country does not show obvious imbalance sufficiently to effect the distribution of PVCC. The gap of PVCC distribution is mainly affected by the distribution differences of urban population density, built-up area and available land area. In fact, the gap of PVCC is closely related to the urban spatial layout of a country. As shown in Figure 4, the distribution of cities in the USA, Europe and India is more even, which makes the solar energy acquisition of each urban division relatively equal. Cities in Australia and Brazil are more closely around the harbors, but the only difference is that Brazil's urban distribution is more dispersed, which leads to a higher

average PVCC than that in Australia. Compared with countries above, however, the urban layout of China is gradually spreading and gathering from west to east, resulting in a huge gap in the distribution of PVCC. In order to achieve the overall PV development goal, China has to rely on long-distance power transmission, which would definitely break the key quality of high-efficient local utility. So, it can be concluded that the dispersed layout of cities and population is also of great significance for the future PV spatial planning.

**Table 3.** Gini indices of influence factors distribution in each urban division of different countries.


Note: The influence factors are arranged in descending order by the sensitivity of PVCC to each. The data of electricity consumption per capita is the overall level of a country, considering the similarity of electricity consumption behavior, the specific impact of its distribution is therefore ignored in this study.

**Figure 4.** Comparison of urban spatial layout and PVCC distribution in each country or region. **Figure 4.** Comparison of urban spatial layout and PVCC distribution in each country or region.

#### *4.3. What Should Be Done to Adjust the PV Spatial Planning? 4.3. What Should Be Done to Adjust the PV Spatial Planning?*

The key point of PV spatial planning adjustment is the crucial means by which to determine the most effective way to improve PVCC. Optimization strategies by transformations of electricity consumption mode, urban planning or building morphology will be still feasible, but as mentioned above, these methods might only be effective in certain urban scales. Realistically, the adjustment of PV spatial planning needs to pay close attention to the actual condition of each country. In the case of Brazil and India, unless higher development target is proposed, it is unnecessary to contemplate their adjustment, because all the PVCC in each urban scale in either country has exceeded the proportion of global average PV consumption (refer to Table 2). For Australia, the average PVCC in urban areas is relatively low, because Australia has relatively few cities, which are each concentrated but mostly distributed some distance from each other, it will be able to meet The key point of PV spatial planning adjustment is the crucial means by which to determine the most effective way to improve PVCC. Optimization strategies by transformations of electricity consumption mode, urban planning or building morphology will be still feasible, but as mentioned above, these methods might only be effective in certain urban scales. Realistically, the adjustment of PV spatial planning needs to pay close attention to the actual condition of each country. In the case of Brazil and India, unless higher development target is proposed, it is unnecessary to contemplate their adjustment, because all the PVCC in each urban scale in either country has exceeded the proportion of global average PV consumption (refer to Table 2). For Australia, the average PVCC in urban areas is relatively low, because Australia has relatively few cities, which are each concentrated but mostly distributed some distance from each other, it will be able to meet the needs of

the needs of future PV planning and development by just establishing a few utility scale

and Europe is also at a low level, but, in fact, their energy acquisition per capita is even higher than other countries (refer to Table 2). Therefore, the method of controlling energy consumption would be more effective for these countries. As for China, the excessive urban agglomeration would serve to aggravate the gap in PVCC distribution, which might seriously impact the high-efficient utility of local energy and expand its overall transmission grid coverage. In this regard, the equilibrium development of different urban scales (especially scales below small cities and above medium cities) by adjustments of population distribution, urban layout or even industrial deployment, might become a more chal-

The increasing utilization of renewable energy is essential to sustainable urban development in the future. In 2020, China put forward its guiding task for the development of renewable energy in the next five years [39]. To respond this, 30 provinces had also given out their own energy planning measures, which can be generally summarized into five aspects, including: (1) Increase the construction of renewable energy infrastructures

*4.4. What Is the Essence of Solar PV Spatial Planning?*

lenging problem currently faced by China in its overall PV planning.

future PV planning and development by just establishing a few utility scale PV stations and sufficient lengths of transmission lines. The PVCC in urban areas of USA and Europe is also at a low level, but, in fact, their energy acquisition per capita is even higher than other countries (refer to Table 2). Therefore, the method of controlling energy consumption would be more effective for these countries. As for China, the excessive urban agglomeration would serve to aggravate the gap in PVCC distribution, which might seriously impact the high-efficient utility of local energy and expand its overall transmission grid coverage. In this regard, the equilibrium development of different urban scales (especially scales below small cities and above medium cities) by adjustments of population distribution, urban layout or even industrial deployment, might become a more challenging problem currently faced by China in its overall PV planning.

#### *4.4. What Is the Essence of Solar PV Spatial Planning?*

The increasing utilization of renewable energy is essential to sustainable urban development in the future. In 2020, China put forward its guiding task for the development of renewable energy in the next five years [39]. To respond this, 30 provinces had also given out their own energy planning measures, which can be generally summarized into five aspects, including: (1) Increase the construction of renewable energy infrastructures to enhance the power generation capacity. (2) Deepen the reform of smart grid to realize the flexible allocation of power in urban area. (3) Expand the scale and capacity of trans-provincial power transmission to balance the difference of renewable energy power proportion in different regions. (4) Improve the manufacturing capacity and technology of renewable energy facilities to reduce the cost of power generation. (5) Transform the planning of energy storage facilities and energy utilization mode (electric vehicles, electric heating, etc.) to improve local renewable energy consumption. Apparently, all these measures are the continuation of previous cognition of "Power distribution and allocation", in the working mode of which, the city (energy demand side) is still considered as the principal part of planning and the power planning has to be adjusted passively with the change of urban energy demand. An inevitable result of this is that, still we have to make large scale of investment into building thousands miles of transmission lines, merely to deliver solar energy, the ubiquitous resource on earth, from one place to another. However, as repeatedly stated before, the main obstacle to the large-scale utilization of solar PV is more likely to be the inadaptability of existing spatial planning, such as population distribution, urban layout or industrial deployment. Therefore, the PV spatial planning proposed in this study is essentially a kind of urban planning or national geographic spatial planning, which could balance the relation between energy demand and supply, coordinated with regional (or urban) economy, resources, social structure and other comprehensive planning elements. Such "PV spatial planning" requires the active adjustment of urban (and national geographic) planning to adapt the needs of large-scale PV utilization in the future. However, it is not meant that solar PV would become the dominant factor to determine the morphology or deployment of the city. Instead, it is just one factor related to many other planning elements (e.g., water resources, food, transportation capacity, natural disasters, ecological or historical protection). In addition, PV spatial planning does not deny any current attempt of optimization measures, but at least these measures need to be integrated into a hierarchical process to cope with the transformation of geographic and urban spatial planning. For instance: (1) At the very first hierarchy of national geospatial planning, it needs the reasonable deployment of urban location and population scale to reduce the gap of PVCC distribution and maximize the efficient use of local energy (2) Through the assessment of regional solar irradiation, available land conditions and other planning elements, the function target and PV development anticipate of the city (or region) should be then formulated. (3) Furthermore, at the hierarchy of urban planning, it determines the scope of urban land use, building density control standards, building function layout and the optimization strategy of building morphology, etc. Compare the adjustment strategies of each planning hierarchy, to explore the most effective way to achieve the

optimal utilization of solar PV energy. Of course, such "PV spatial planning" is a kind of forward-thinking and might be regarded as a dispensable work for now, because, at least at present, the proportion of solar PV utilization is still at a low level, but with the rapid increasing utilization of renewable energy, such proactive adjustment of urban (or even geographic) spatial planning may be of great significance for achieving higher solar energy supply in the future.

#### **5. Conclusions**

Out of the demand for energy sustainability, solar PV will gradually play an important role in future energy consumption. PV utilization focuses on the improvement of power consumption proportion, meanwhile, it also needs the reasonable deployment of energy and urban development. Both of these factors are the key elements in the construction of energy systems in the future. As discussed herein, the spatial planning of PV might not only be the planning of solar PV itself, the cognition of PV spatial planning needs to be transformed into a kind of urban or even national geographic spatial planning, in the progress of which solar PV utilization should be taken as an important planning element into consideration. The concept of PVCC described in this study established a link between the PV development anticipation and its close relationship to other spatial planning on the one hand and make people think about PV utilization from a view of resource and environment carrying capacity on the other hand. Such carrying capacity may in turn become a valid intuitive concept for future PV spatial planning, providing analysis capabilities and methods, with which the spatial planners could make comparisons and decisions to plan future courses of action. At the very least, it is suggested that the framework in this study can be considered as a starting point, aiming to stimulate interdisciplinary discussions between renewable energy developers, urban planners and policy makers. The methods mentioned in this study may be still not precise or comprehensive enough to make implementable planning adjustments independently, after all, only six factors were fit into the comparison, but they do provide the logic to figure out optimal solutions in the process of PV spatial planning by analyzing PVCC distribution and comparing the beneficial effects from each optimization strategy. Therefore, we can conclude that these tentative methods are available, but still need to be connected with other planning elements (or detailed influence factors) to give both planners and policy-makers the ability to guide the feasible PV spatial planning in each hierarchy, for the time being and in the foreseeable future.

**Author Contributions:** Conceptualization, Y.Z. and S.C.; methodology, S.C.; software, S.C.; validation, Y.Z., J.Z. and S.C.; formal analysis, S.C.; investigation, S.C.; resources, S.C.; data curation, S.C.; writing—original draft preparation, S.C.; writing—review and editing, Y.Z., J.Z. and S.C.; visualization, S.C.; supervision, J.Z.; project administration, J.Z.; funding acquisition, Y.Z., J.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by National Natural Science Foundation of China, grant number 51978443, National Natural Science Foundation of China, grant number 52078322, National Natural Science Foundation of China, grant number 51708395 and The APC was funded by National Natural Science Foundation of China, grant number 51708395.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is contained within the article or supplementary.

**Acknowledgments:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

### **References**


## *Article* **E**ff**ect of Spatial and Temporal Resolution Data on Design and Power Capture of a Heaving Point Absorber**

## **Tunde Aderinto <sup>1</sup> and Hua Li 2,**<sup>∗</sup>


Received: 23 October 2020; Accepted: 13 November 2020; Published: 16 November 2020

**Abstract:** For a heaving point absorber to perform optimally, it has to be designed to resonate to the prevailing ocean wave period. Hence, it is important to make the ocean wave data analysis to be as accurate as possible. In this study, existing wave condition data is used to investigate the effect of the temporal resolution (daily vs. hourly) of wave data on the design of the device and power capture. The temporal resolution effect on the estimation of ocean wave resource theoretical potential is also investigated. Results show that the temporal resolution variation of the ocean wave data affects the design of the device and its power capture, but the theoretical power resource assessment is not significantly affected. The device designed for the Gulf of Mexico is also analyzed with wave condition in Oregon, which has about 40 times the wave resource theoretical potential compared to the Gulf of Mexico. The results confirmed that a device should be designed for a specific location as the device performed better in the Gulf of Mexico, which has much less ocean wave resource theoretical potential. At last, the effect of the design, diameter and season (summer and winter) on the power output of the device is also investigated using statistical hypothesis testing methods. The results show that the power capture of a device is significantly affected by these parameters.

**Keywords:** wave energy converters; heaving point absorber; design and performance; spatial and temporal variation

### **1. Introduction**

Ocean wave energy has continued to see increase in the level of awareness in recent years. Moreover, the last three to five years have seen a lot of research and development efforts into the ocean wave energy industry [1–3]. Others studies focused on specific aspects of the ocean wave energy such as resource characterization have been performed at global [4,5], regional [6–8] and local levels [9,10]. Serious exploration and exploitation of ocean wave energy resources are currently being investigated in U.S. [3], Europe [11–13], China [14], India [15], etc. Other aspects of ocean wave energy such as economics [16], environmental [17], design [18] and efficiency and performance [19] have all being studied by different researchers. Apart from the general aim for ocean wave energy to supply energy into the traditional grid systems, the work done by [20] investigated the potential of using ocean wave energy to supply power for offshore oil rigs and other offshore structures which can broaden the application of this vast but underutilized energy resource.

Waveenergyconverters(WECs)canbeclassifiedbasedontheirworkingprinciples. Anotherclassification method is based on the water depth of the WEC's site location (shoreline, nearshore and offshore). WECs are also classified based on the ratio of the wavelength magnitude to the interacting part of the

WEC. For example, WECs can be classified into oscillating water columns, oscillating body systems and overtopping devices (Figure 1) based on their working principles. Under this classification, a point absorber, which is when the WEC interacting part dimensions is considerably smaller than that of the interacting ocean wavelength [21,22], is considered as an oscillating body-based WEC. It is a terminator if the dominant wave direction is perpendicular to the structural extension of the WEC [23], while it is an attenuator if its structural extension is parallel to the interacting wave direction [24]. part of the WEC. For example, WECs can be classified into oscillating water columns, oscillating body systems and overtopping devices (Figure 1) based on their working principles. Under this classification, a point absorber, which is when the WEC interacting part dimensions is considerably smaller than that of the interacting ocean wavelength [21,22], is considered as an oscillating body‐ based WEC. It is a terminator if the dominant wave direction is perpendicular to the structural extension of the WEC [23], while it is an attenuator if its structural extension is parallel to the interacting wave direction [24].

*Sustainability* **2020**, *12*, x 2 of 17

offshore). WECs are also classified based on the ratio of the wavelength magnitude to the interacting

**Figure 1.** Classification of wave energy converter (WEC) extraction technologies: (**a**) oscillating water column, (**b**) overtopping devices and (**c**) oscillating bodies [2]. **Figure 1.** Classification of wave energy converter (WEC) extraction technologies: (**a**) oscillating water column, (**b**) overtopping devices and (**c**) oscillating bodies [2].

One of the promising methods of wave energy capture is the oscillating body system. One major thing that makes the use of these types of converters attractive is because the amount of the energy absorbed by the body can be improved upon significantly under the same wave conditions when the body is at resonance with the incoming waves as illustrated in Figure 2. In fact, the team that won the prestigious ocean wave energy prize offered by the United States Department of Energy developed their concept and design based on a heavingf oscillating buoy [25]. The Pelamis [26], which is one of the most studied converters, is a pitching (rotating) oscillating converter and is also another type of a wave activated body system. One of the promising methods of wave energy capture is the oscillating body system. One major thing that makes the use of these types of converters attractive is because the amount of the energy absorbed by the body can be improved upon significantly under the same wave conditions when the body is at resonance with the incoming waves as illustrated in Figure 2. In fact, the team that won the prestigious ocean wave energy prize offered by the United States Department of Energy developed their concept and design based on a heavingf oscillating buoy [25]. The Pelamis [26], which is one of the most studied converters, is a pitching (rotating) oscillating converter and is also another type of a wave activated body system.

*Sustainability* **2020**, *12*, x 3 of 17

**Figure 2.** Illustration of theoretical power capture of oscillating body system converters [1,2]. **Figure 2.** Illustration of theoretical power capture of oscillating body system converters [1,2].

The hydrodynamics of oscillating body systems including heaving systems were independently solved by [27–29], which show the theoretical maximum energy to be captured by an oscillating body system‐based wave energy converter. The results confirm that the highest possible capture occurs when the body is at resonance with the incoming waves. This behavior of a floating oscillating body poses a challenge for WEC designers because a typical body has a narrow resonance frequency band, and the body performs poorly outside this band. Hence, one of the many characteristics of a good oscillating WEC design is to make the buoy resonate to the prevailing ocean wave properties [2]. It should be noted that other factors, including survivability and profitability, need to be considered as well when designing a WEC. In order to capture considerable power outside the resonance frequency band, different optimization methods have been proposed and investigated. Some optimization methods include changes made to the shape and dimensions of the buoy [30], while latching and declutching methods [31] are also used in some existing studies. Latching control is achieved by holding the heaving WEC in a fixed position when the velocity is zero and releasing it at the right time so that its velocity can be in phase with the excitation force to achieve resonance [32]. On the other hand, declutching works by alternatively switching the power take off system on and off [31]. Another method is the model predictive control. It is an advanced control strategy [33] compared to the passive control methods, which may employ complex algorithms and simulations to achieve the optimization of power absorption by the WECs. While these methods have theoretical possibilities, The hydrodynamics of oscillating body systems including heaving systems were independently solved by [27–29], which show the theoretical maximum energy to be captured by an oscillating body system-based wave energy converter. The results confirm that the highest possible capture occurs when the body is at resonance with the incoming waves. This behavior of a floating oscillating body poses a challenge for WEC designers because a typical body has a narrow resonance frequency band, and the body performs poorly outside this band. Hence, one of the many characteristics of a good oscillating WEC design is to make the buoy resonate to the prevailing ocean wave properties [2]. It should be noted that other factors, including survivability and profitability, need to be considered as well when designing a WEC. In order to capture considerable power outside the resonance frequency band, different optimization methods have been proposed and investigated. Some optimization methods include changes made to the shape and dimensions of the buoy [30], while latching and declutching methods [31] are also used in some existing studies. Latching control is achieved by holding the heaving WEC in a fixed position when the velocity is zero and releasing it at the right time so that its velocity can be in phase with the excitation force to achieve resonance [32]. On the other hand, declutching works by alternatively switching the power take off system on and off [31]. Another method is the model predictive control. It is an advanced control strategy [33] compared to the passive control methods, which may employ complex algorithms and simulations to achieve the optimization of power absorption by the WECs. While these methods have theoretical possibilities, there exists very little information reporting their applications in real ocean conditions.

there exists very little information reporting their applications in real ocean conditions. In all these designs and optimization methods, the ocean wave properties have to be properly characterized first in order to have a good and effective WEC design. Although two ocean wave properties (wave height and wave period) guide the estimation of ocean wave resource potential and power capture of a WEC, it is the wave period that determines how the resonance behavior of a floating body will be engineered so it will operate near resonant level with the desired ocean wave period. There are existing studies that focus on general guidelines for designing a WEC. Meanwhile, most available wave condition database or monitoring/forecasting systems are designed for other marine systems instead of WEC design [8,9], so it is important to investigate the possibility of using existing wave condition data on designing a WEC. Instead of optimizing the size of a heaving point absorber WEC, this paper focuses on conducting detailed quantitative analysis on the changes on the power output of a heaving point absorber due to the variation of temporal and spatial resolutions of existing wave condition data. This paper uses available wave condition data from an existing database instead of collecting new wave condition data in the analysis. Section 2 introduces the methodology used including the ocean wave data analysis in the studied regions, the method for estimating the average yearly energy resource potential as well as the design process for the WEC which led to two different WEC designs in terms of dimensions. In Section 3, the annual energy resource potential is estimated under three different scenarios based on existing wave condition data: (1) wave data based on hourly resolution at the selected location in the Gulf of Mexico, (2) wave data based on daily resolution at the selected location in the Gulf of Mexico, and (3) wave data based on hourly resolution in an offshore location in Oregon. The power and annual energy matrices of the In all these designs and optimization methods, the ocean wave properties have to be properly characterized first in order to have a good and effective WEC design. Although two ocean wave properties (wave height and wave period) guide the estimation of ocean wave resource potential and power capture of a WEC, it is the wave period that determines how the resonance behavior of a floating body will be engineered so it will operate near resonant level with the desired ocean wave period. There are existing studies that focus on general guidelines for designing a WEC. Meanwhile, most available wave condition database or monitoring/forecasting systems are designed for other marine systems instead of WEC design [8,9], so it is important to investigate the possibility of using existing wave condition data on designing a WEC. Instead of optimizing the size of a heaving point absorber WEC, this paper focuses on conducting detailed quantitative analysis on the changes on the power output of a heaving point absorber due to the variation of temporal and spatial resolutions of existing wave condition data. This paper uses available wave condition data from an existing database instead of collecting new wave condition data in the analysis. Section 2 introduces the methodology used including the ocean wave data analysis in the studied regions, the method for estimating the average yearly energy resource potential as well as the design process for the WEC which led to two different WEC designs in terms of dimensions. In Section 3, the annual energy resource potential is estimated under three different scenarios based on existing wave condition data: (1) wave data based on hourly resolution at the selected location in the Gulf of Mexico, (2) wave data based on daily resolution at the selected location in the Gulf of Mexico, and (3) wave data based on hourly resolution in an offshore location in Oregon. The power and annual energy matrices of the two WEC designs based on the hourly and

daily resolution data are analyzed in Section 4. The design tailored to the hourly resolution data in the Gulf of Mexico is tested with the hourly data of an offshore location in Oregon, and the results are compared. Section 5 shows the results of a series of statistical hypothesis analysis such as *t*-test using data obtained from the Gulf of Mexico to determine the significance on the power output by different parameters such as diameter, design and seasonal variation (winter and summer). Section 6 is for discussions and conclusions. *Sustainability* **2020**, *12*, x 4 of 17 tailored to the hourly resolution data in the Gulf of Mexico is tested with the hourly data of an offshore location in Oregon, and the results are compared. Section 5 shows the results of a series of statistical hypothesis analysis such as t‐test using data obtained from the Gulf of Mexico to determine the significance on the power output by different parameters such as diameter, design and seasonal variation (winter and summer). Section 6 is for discussions and conclusions.

#### **2. Methods 2. Methods**

#### *2.1. Ocean Wave Data 2.1. Ocean Wave Data*

Existing default data of significant wave height and dominant wave period over a 9-year period was obtained from a buoy operated by National Data Buoy Center at a location in the Gulf of Mexico (GoM) at coordinates 26.968◦ N and 96.693◦ W with sea depth of 84 m in a watch circle of radius 138 m (Figure 3a) [34]. The percentage of occurrences were analyzed for two scenarios in terms of the temporal variation of the collected data: data obtained hourly (Table 1) and data obtained daily (Table 2). The daily and hourly significant wave height data represent the average of the highest one-third of waves in the given period of capture. For both datasets, the wave heights ranged from 1 m to 5 m and wave periods ranging from 3 s to 12 s captured about 99% of all data points. Another set of ocean wave data was obtained from a location in Oregon which is close to the PacWave test site at coordinates 44.667◦ N and 124.515◦ W with sea depth of 140 m and a watch circle of radius of 230 m (Figure 3b) [34]. Another reason to choose the Oregon site for comparison is that the Oregon site is considered as high wave energy potential site (Table 3) while the GoM site is normally considered as low wave energy potential site. It should be noted that extreme wave conditions exist with a very small occurrence, such that there are 0.0001% of waves with significant wave height larger than 12 m and dominant wave period higher than 22 s in the GoM. However, we didn't include these extreme wave conditions in the data tables and power generation estimation, because the focus of this paper is on the power generation, while the proposed WEC will not generate power during the extreme conditions. The extreme conditions will be considered when investigating the structural reliability of the proposed WEC. Existing default data of significant wave height and dominant wave period over a 9‐year period was obtained from a buoy operated by National Data Buoy Center at a location in the Gulf of Mexico (GoM) at coordinates 26.968N and 96.693W with sea depth of 84 m in a watch circle of radius 138 m (Figure 3a) [34]. The percentage of occurrences were analyzed for two scenarios in terms of the temporal variation of the collected data: data obtained hourly (Table 1) and data obtained daily (Table 2). The daily and hourly significant wave height data represent the average of the highest one‐third of waves in the given period of capture. For both datasets, the wave heights ranged from 1 m to 5 m and wave periods ranging from 3 s to 12 s captured about 99% of all data points. Another set of ocean wave data was obtained from a location in Oregon which is close to the PacWave test site at coordinates 44.667N and 124.515W with sea depth of 140 m and a watch circle of radius of 230 m (Figure 3b) [34]. Another reason to choose the Oregon site for comparison is that the Oregon site is considered as high wave energy potential site (Table 3) while the GoM site is normally considered as low wave energy potential site. It should be noted that extreme wave conditions exist with a very small occurrence, such that there are 0.0001% of waves with significant wave height larger than 12 m and dominant wave period higher than 22 s in the GoM. However, we didn't include these extreme wave conditions in the data tables and power generation estimation, because the focus of this paper is on the power generation, while the proposed WEC will not generate power during the extreme conditions. The extreme conditions will be considered when investigating the structural reliability of the proposed WEC.

**Figure 3. Figure 3.** Studying locations: ( Studying locations: (**aa**) location 1: Gulf of Mexico, and (**b**) location 2: Oregon [34]. ) location 1: Gulf of Mexico, and (**b**) location 2: Oregon [34].


**Table 1.** Percentage of ocean wave height and period occurrence based on hourly wave data (location 1: Gulf of Mexico).

**Table 2.** Percentage of ocean wave height and period occurrence based on daily resolution data (location 1: Gulf of Mexico).


**Table 3.** Percentage of ocean wave height and period occurrence based on hourly resolution data (location 2: Oregon).


#### *2.2. Design of Heaving WEC Dimensions 2.2. Design of Heaving WEC Dimensions* The initial dimensions of the proposed WEC (Figure 4) were estimated based on the theoretical

The initial dimensions of the proposed WEC (Figure 4) were estimated based on the theoretical hydrodynamics of floating bodies. The hydrodynamics describe the motion of a floating body under the action of external forces. The external forces in this case were mainly generated by the ocean waves. A combination of wave data analysis and the theoretical wave hydrodynamics were used to estimate the diameter of the cylindrical buoy, which is the shape of the proposed WEC device. From the wave data analysis shown in Tables 1 and 2, the dominant wave period falls between 5 and 6 s with 26.38% occurrence for the hourly data and between 6 and 7 s with 31.38% occurrence for the daily data. The following sets of equations were used to estimate the initial dimensions of the proposed WEC. hydrodynamics of floating bodies. The hydrodynamics describe the motion of a floating body under the action of external forces. The external forces in this case were mainly generated by the ocean waves. A combination of wave data analysis and the theoretical wave hydrodynamics were used to estimate the diameter of the cylindrical buoy, which is the shape of the proposed WEC device. From the wave data analysis shown in Tables 1 and 2, the dominant wave period falls between 5 and 6 s with 26.38% occurrence for the hourly data and between 6 and 7 s with 31.38% occurrence for the daily data. The following sets of equations were used to estimate the initial dimensions of the proposed WEC.

*Sustainability* **2020**, *12*, x 6 of 17

**12–13** 0.46% 4.28% 4.65% 2.78% 1.32% 0.44% **13–14** 0.38% 1.94% 1.80% 1.33% 0.71% 0.27% **14–15** 0.60% 1.77% 1.25% 1.02% 0.66% 0.18% **15–16** 0.00% 0.00% 0.00% 0.00% 0.00% 0.00% **16–17** 0.33% 1.24% 0.81% 0.81% 0.48% 0.16%

**Figure 4.** Schematic design of the proposed WEC buoy. **Figure 4.** Schematic design of the proposed WEC buoy.

From previous literature [35], From previous literature [35],

$$L\_{\text{max}} = \frac{\lambda}{2\pi} \tag{1}$$

$$
\lambda = \frac{gT^2}{2\pi} \tag{2}
$$

and as recommended by [36], and as recommended by [36],

absorber system.

$$\mathcal{C} = \frac{L\_{\text{max}}}{D} \ge 3 \tag{3}$$

 ൌ 3 (3) where *Lmax* is Maximum capture width, is wave length, *T* is dominant wave period, *D* is buoy diameter, *C* is captured width ratio and *g* is gravitational constant. These equations act as a guide to determine the size of a point absorber at the initial stage. It should be noted that Equation (2) is based where *Lmax* is Maximum capture width, λ is wave length, *T* is dominant wave period, *D* is buoy diameter, *C* is captured width ratio and *g* is gravitational constant. These equations act as a guide to determine the size of a point absorber at the initial stage. It should be noted that Equation (2) is based on approximation made for deep water condition.

on approximation made for deep water condition. Theoretically, the resonance frequency of a submerged body is given by Equation (4) below. This simplified equation can be used at an initial stage to estimate the natural frequency of the point Theoretically, the resonance frequency ω*<sup>n</sup>* of a submerged body is given by Equation (4) below. This simplified equation can be used at an initial stage to estimate the natural frequency of the point absorber system.

$$
\omega\_n = \sqrt{\frac{\rho g A\_w}{M\_w + a}} \tag{4}
$$

 ൌ ඨ ௪ ௪ (4) where *Aw* = water plane area, *Mw* is mass of displaced water, *a* is added mass and *ω<sup>n</sup>* is natural period. Added mass at this preliminary stage is given as 0.167 ρD3 [36]. The effective drafts for different where *A<sup>w</sup>* = water plane area, *M<sup>w</sup>* is mass of displaced water, *a* is added mass and ω*<sup>n</sup>* is natural period. Added mass at this preliminary stage is given as 0.167 ρD<sup>3</sup> [36]. The effective drafts for different diameters of a buoy whose density is equal to the density of water is given in Table 4. This application is for floating bodies alone. Steel is chosen as the material for the buoy in this paper. For the buoy to oscillate freely in seawater, the buoy must be hollow as the density of steel is greater than that of water. The mass of the steel buoy should be equal to the mass of the displaced water such that

$$M\_{\rm w} = M\_{\rm s};\tag{5}$$

where, *M<sup>s</sup>* is the mass of the steel cylindrical buoy.


**Table 4.** Dimensions of initial designs of the proposed WEC. GoM = Gulf of Mexico.

For any selected diameter and wave period, the corresponding draft and depth under 0.15 m thickness can be obtained in Tables 5 and 6, respectively. It should be noted that the depth of the buoy (Table 6) must be greater than or equal to the draft (Table 5) for a floating cylinder. Thus, the feasible diameter of the buoy is 5 m and above. According to Equation (4), the resonance period is a function of mass of displaced water, water plain area, etc. Thus, for any thickness, the mass of water displaced and the mass of the cylindrical buoy should be constant when choosing the diameter of the buoy, but the depth of the cylinder will change. Therefore, the impact of the buoy thickness on its performance may be minimal in this research. For the initial design of the buoy in this paper, 8 m was chosen as its diameter with a steel of density 7850 kg/m<sup>3</sup> . For design 1, whose dimensions were tailored to the most probable wave period of the hourly resolution data in the GoM, the most prevalent wave period lies between 5 and 6 s, so the initial draft and depth of design 1 were 7.3 m and 12.9 m, respectively. Similarly, for design 2, whose dimensions were tailored to the most prevalent wave period between 6 and 7 s of the daily resolution data in the GoM, the initial draft and depth of design 2 were 10.5 m and 18.6 m, respectively. These initial dimensions were further analyzed in ANSYS diffraction module to give more accurate values of the resonance period corresponding to the initial dimensions, which decide the final dimensions used to estimate the power capture of the WEC.



**Table 6.** Depth under different diameters and wave period to achieve resonance (density of material = density of steel).


Colored cells represent feasible depth of buoy.

A buoy diameter and associated draft used to form initial dimensions were tested using ANSYS/AQWA 18.1 [37]. ANSYS/AQWA is capable of simulating linearized hydrodynamic fluid wave loading on either floating or fixed rigid structures based on potential flow theory. It employs a three-dimensional diffraction theory in regular waves in the frequency domain. Furthermore, real time motion and force responses of bodies operating in regular or irregular waves can be studied. For each prevailing wave period based on the two temporal resolution data, the optimal dimensions (depth and draft to make buoy resonate with the wave period) for the selected diameters were decided after the analysis in ANSYS. AQWA The process is summarized in Figure 5.

decided after the analysis in ANSYS. AQWA The process is summarized in Figure 5.

A buoy diameter and associated draft used to form initial dimensions were tested using ANSYS/AQWA 18.1 [37]. ANSYS/AQWA is capable of simulating linearized hydrodynamic fluid wave loading on either floating or fixed rigid structures based on potential flow theory. It employs a three‐dimensional diffraction theory in regular waves in the frequency domain. Furthermore, real time motion and force responses of bodies operating in regular or irregular waves can be studied. For each prevailing wave period based on the two temporal resolution data, the optimal dimensions

**Figure 5.** Design process for arriving at each design. **Figure 5.** Design process for arriving at each design.

#### **3. Annual Energy Resource Assessment and Estimation**

**3. Annual Energy Resource Assessment and Estimation** The ocean wave power density in a location is a function of wave height and wave period given The ocean wave power density in a location is a function of wave height and wave period given by the relation in Equation (6) based on approximation made for deep water condition.

by the relation in Equation (6) based on approximation made for deep water condition. The annual power density can be estimated if information about the percentage occurrence of the wave height/period in the location is available [2]. The annual power density can be estimated if information about the percentage occurrence of the wave height/period in the location is available [2].

$$P = \frac{\rho g^2 T\_\epsilon H^2}{64\pi} \tag{6}$$

where *P* (KW/m) is power density (power per unit width of wave front), *ρ* (Kg/m3) is seawater density, *g* (m/s2) is gravitational acceleration, *H* (m) is significant wave height and *Te* (s) is energy wave period. Since the wave period date provided by National Oceanic and Atmospheric Administration (NOAA) is dominant wave period *Tp*, the energy wave period (*Te*) is calculated by multiplying a wave period conversion factor α to dominant wave period (*Tp*). In this study, the wave conversion factor was considered as 0.9, which is the equivalent of JONSWAP software [38] and has been used in the past by different studies [9,20,38]. where *P* (KW/m) is power density (power per unit width of wave front), ρ (Kg/m<sup>3</sup> ) is seawater density, *g* (m/s 2 ) is gravitational acceleration, *H* (m) is significant wave height and *T<sup>e</sup>* (s) is energy wave period. Since the wave period date provided by National Oceanic and Atmospheric Administration (NOAA) is dominant wave period *Tp*, the energy wave period (*Te*) is calculated by multiplying a wave period conversion factor α to dominant wave period (*Tp*). In this study, the wave conversion factor was considered as 0.9, which is the equivalent of JONSWAP software [38] and has been used in the past by different studies [9,20,38].

The potential energy density per year is calculated by multiplying the percentage occurrence in a year, and the results for the three different scenarios are provided in the Tables 7–9 below. From Tables 7 and 8, the hourly resolution data provides annual energy density potential of 105.3 MWh/m, and the daily resolution data provides annual energy density potential of 102 MWh/m. This shows about 3% difference in terms of energy density potential in the same location with different temporal resolution data. This shows that different temporal resolution of the data in same location may not significantly affect the estimation of the total wave energy resource potential present in the location. The potential energy density per year is calculated by multiplying the percentage occurrence in a year, and the results for the three different scenarios are provided in the Tables 7–9 below. From Tables 7 and 8, the hourly resolution data provides annual energy density potential of 105.3 MWh/m, and the daily resolution data provides annual energy density potential of 102 MWh/m. This shows about 3% difference in terms of energy density potential in the same location with different temporal resolution data. This shows that different temporal resolution of the data in same location may not significantly affect the estimation of the total wave energy resource potential present in the location.


**Table 7.** Annual energy density potential (kWh/(m∙yr)) based on hourly resolution data (location‐1: Gulf of Mexico). **Table 7.**Annual energy density potential (kWh/(m·yr)) based on hourly resolution data (location-1: Gulf of Mexico).


**Table 8.** Annual energy density potential (kWh/(m·yr)) based on daily resolution data (location-1: Gulf of Mexico).

**Table 9.** Annual energy resource potential (kWh/(m·yr)) based on daily resolution data (location-2: Oregon).


Similarly, the resource potential for the location in offshore Oregon is shown in Table 9 above. With a yearly total of about 43,374 MWhr/m of wave energy resource potential, it is confirmed that the location is a high wave energy resource area with resources about 40 times higher than that of the location in the Gulf of Mexico.

#### **4. Power Capture of the WEC**

Three scenarios of power capture are examined and compared in this section. After a series of computational fluid dynamics (CFD) diffraction analysis using ANSYS/AQWA suite version 18.1, which is based on potential flow theory, the response amplitude operator (RAO) which gives the motion response of the buoy at different periods is given in Figure 6. The three scenarios and the WEC dimensions are provided in Table 10. Design 1 was tailored to the dominant wave period based on the hourly resolution data while Design 2 dimensions was engineered to capture power optimally at the dominant wave period based on the daily resolution data from location 1 in the Gulf of Mexico. The power capture of design 1 in location 2 in Oregon is also estimated to investigate the impact of different wave conditions on the same WEC design in terms of power capture.

= response amplitude operator.

motion response of the buoy at different periods is given in Figure 6. The three scenarios and the WEC dimensions are provided in Table 10. Design 1 was tailored to the dominant wave period based on the hourly resolution data while Design 2 dimensions was engineered to capture power optimally at the dominant wave period based on the daily resolution data from location 1 in the Gulf of Mexico. The power capture of design 1 in location 2 in Oregon is also estimated to investigate the impact of

**Table 10.** Dimension of the final designs used in three scenarios for estimating power capture. RAO

**Scenario Design Location Heave RAO (s) Depth(m) Draft (m) Thickness (m) Diameter (m)** 1 1 GoM 5.5 8 4.5 0.15 8

different wave conditions on the same WEC design in terms of power capture.

**Figure 6.** Heave response amplitude for design 1 and design 2. **Figure 6.** Heave response amplitude for design 1 and design 2.

The power take‐off in this study was modeled as pure damper which is assumed to be frequency **Table 10.** Dimension of the final designs used in three scenarios for estimating power capture. RAO = response amplitude operator.


respect to time; *B* is damping coefficient and *C* is hydrostatic force coefficient; *F*(*t*) is the external force acting on the buoy while FPTO is the PTO force; and *DPTO* is the PTO damping coefficient. The maximum amount of energy captured by the buoy occurs when the PTO damping is equal The power take-off in this study was modeled as pure damper which is assumed to be frequency dependent. The motion equation of the heaving point absorber buoy with the power take off (PTO) can be described by Equation (7) below

where ்ை ൌ ்ைሶ, *M* is mass, *A* is added mass, *x* is heave displacements and its derivatives with

$$(M+A)\ddot{\mathbf{x}} + B\dot{\mathbf{x}} + \mathbf{C}\mathbf{x} = F(t) + F\_{\text{PTO}}\tag{7}$$

1 <sup>2</sup>்ைଶଶ (8) where, *ω* is the angular frequency at resonance. where *FPTO* = *DPTO* . *x*, *M* is mass, *A* is added mass, *x* is heave displacements and its derivatives with respect to time; *B* is damping coefficient and *C* is hydrostatic force coefficient; *F*(*t*) is the external force acting on the buoy while FPTO is the PTO force; and *DPTO* is the PTO damping coefficient.

Using the premises highlighted above, different values of PTO damping were tested on the buoy and the power capture and are shown in Figure 7. The maximum occurred when the damping coefficient was 50 kNm/s for both design 1 and design 2 when the diameter of the buoy was 8m and The maximum amount of energy captured by the buoy occurs when the PTO damping is equal to the radiation damping of the buoy [39]. Hence the PTO damping will be equal to that of the buoy at resonance. The mean absorbed power by the PTO is given by Equation (8)

$$\frac{1}{2}D\_{\rm TTO}u^2x^2\tag{8}$$

where, ω is the angular frequency at resonance.

Using the premises highlighted above, different values of PTO damping were tested on the buoy and the power capture and are shown in Figure 7. The maximum occurred when the damping coefficient was 50 kNm/s for both design 1 and design 2 when the diameter of the buoy was 8 m and 40 kNm/s when the diameter of the buoy was 7 m for both design scenarios.

*Sustainability* **2020**, *12*, x 11 of 17

**Figure 7.** Power take off (PTO) damping vs. power capture. **Figure 7.** Power take off (PTO) damping vs. power capture. **Figure 7.** Power take off (PTO) damping vs. power capture.

For both designs, the heaving oscillatory motion of the buoy was converted to power through a power take off (PTO). For the purpose of this design, a PTO damping of 40 kNm/s and 50 kNm/s was used for the WEC devices when the diameter of the buoy was 7 m and 8 m, respectively. Figures 8 and 9 show the change of the power capture with respect to the wave period and wave height, respectively. The figures show, as expected, the power capture peaks when the wave period is at the resonance period of the buoy. However, the power capture increases almost linearly with wave height increase. Hence, while both wave period and wave height affect the power capture, the relationship between the wave period and power capture of a heaving point absorber is more significant due to the resonant behavior pattern. For both designs, the heaving oscillatory motion of the buoy was converted to power through a power take off (PTO). For the purpose of this design, a PTO damping of 40 kNm/s and 50 kNm/s was used for the WEC devices when the diameter of the buoy was 7 m and 8 m, respectively. Figures 8 and 9 show the change of the power capture with respect to the wave period and wave height, respectively. The figures show, as expected, the power capture peaks when the wave period is at the resonance period of the buoy. However, the power capture increases almost linearly with wave height increase. Hence, while both wave period and wave height affect the power capture, the relationship between the wave period and power capture of a heaving point absorber is more significant due to the resonant behavior pattern. For both designs, the heaving oscillatory motion of the buoy was converted to power through a power take off(PTO). For the purpose ofthis design,aPTOdamping of 40kNm/s and 50kNm/swasused for the WEC devices when the diameter of the buoy was <sup>7</sup> <sup>m</sup> and <sup>8</sup> m, respectively. Figures <sup>8</sup> and9showthe changeofthepowercapturewithrespectto thewaveperiodandwave height, respectively. The figures show, as expected, the power capture peaks when the wave period is at the resonance period ofthe buoy.However,thepowercaptureincreasesalmost linearly with wave height increase. Hence, while both wave period and wave height affect the power capture, the relationshipbetweenthe waveperiod andpower capture ofa heavingpointabsorber ismoresignificant due to the resonant behavior pattern.

**Figure 8.** Period vs. power: (**a**) hourly resolution power (**b**) daily resolution power. **Figure 8.** Period vs. power: (**a**) hourly resolution power (**b**) daily resolution power. **Figure 8.** Period vs. power: (**a**) hourly resolution power (**b**) daily resolution power.

The annual energy captured in each bivariate wave height-wave period combinations under different scenarios are given in Tables 11–13. Table 11 shows the yearly energy that can be harvested using design 1 tailored using the hourly data from the GoM, while Table 12 shows the comparative annual energy based on design 2 tailored using daily resolution data. Table 13 shows the annual energy by design 1 when operated in the Oregon location, an offshore region with over 40 times the wave energy potential compared to location 1 in the GoM. When the wave height-wave period combination did not exist due to physics of wave formation, "NA" is input in the tables, while a "0" in the table means the occurrence of that specific combination was zero in the specific time and location.

significant due to the resonant behavior pattern.

**Figure 8.** Period vs. power: (**a**) hourly resolution power (**b**) daily resolution power.

**Figure 7.** Power take off (PTO) damping vs. power capture.

For both designs, the heaving oscillatory motion of the buoy was converted to power through a power take off (PTO). For the purpose of this design, a PTO damping of 40 kNm/s and 50 kNm/s was used for the WEC devices when the diameter of the buoy was 7 m and 8 m, respectively. Figures 8 and 9 show the change of the power capture with respect to the wave period and wave height, respectively. The figures show, as expected, the power capture peaks when the wave period is at the resonance period of the buoy. However, the power capture increases almost linearly with wave height increase. Hence, while both wave period and wave height affect the power capture, the relationship between the wave period and power capture of a heaving point absorber is more

.

**Figure 9. Figure 9.** Wave height vs. power: ( Wave height vs. power: (**aa**) ) hourly resolution power ( hourly resolution power ( **b b** ) daily resolution power. ) daily resolution power.


**Table 11.** Annual energy matrix (Scenario-1: Gulf of Mexico).


**Table 12.** Annual energy matrix (Scenario-2: Oregon).

The result shows that the total annual energy capture under Scenario 1 was about 212 MWh/yr while Scenario 2 had about 178 MWh/yr. Scenario 3 had about 150 MWh/yr. The results show about 16% difference between Scenarios 1 and 2. It is also interesting to note that the annual energy performance estimation of design 2 was lower than that of that of design 1 even though the dimensions of design 2 were larger. It can be inferred that the temporal resolution of the ocean wave data also significantly affects the overall energy performance of the device in addition to significantly affecting the dimensions of the WEC buoy. When comparing Scenario 1 with Scenario 3, despite the buoys in both scenarios having the same dimensions, the total annual estimated energy capture in Scenario 1 was about 41% greater than that of the Scenario 3 even though the total annual energy theoretical potential in location 2 is over 40 times greater than that of location 1. This reinforces the idea that heaving buoys should

be designed and tailored to specific locations. In the next section, the factors involved in the power capture estimation in this paper are further analyzed by conducting detailed statistical analysis.


**Table 13.** Annual energy matrix (Scenario 3).

#### **5. Design Parameters and Their E**ff**ects on WEC's Power Output**

Different parameters and their effects on the WEC's power production are tested using statistical hypothesis testing methods such as *t*-tests and factorial analysis. These tests have given more insights into the complexities that exist between some of different parameters that contribute to the energy produced by a WEC. Real ocean data of a 7-day period randomly picked from summer (July, 13–19) and winter (Jan, 1–7) in the GoM. One summer week and one winter week wave data are extracted including both hourly resolution (168 data points) and daily resolution (7 data points). The power output of WECs with different design parameters including design types (design 1& design 2) and diameters (7 m & 8 m) are estimated under the two weeks wave data. These estimated power output values are used as samples for the hypothesis testing. For the two-sample two-tailed *t*-test under hourly and daily data, each sample has 168 and 7 data points, respectively. For the factorial analysis, the estimated power output values are considered as the response/variable, and design type, diameter, and season are considered as factors with each factor having two different levels.

For the hourly resolution data within the week used for design 1, the standard deviations of the wave height and wave period are 0.32 m and 1.03 s for the summer and 0.65 m and 1.05 s for the winter. Similarly, for the daily resolution data used for design 2, the standard deviations of the wave height and wave period are 0.25 m and 0.69 s for the summer and 0.54 m and 0.78 s for the winter. As described in previous sections, design 1 is tailored to capture highest energy between 5–6 s (RAO = 5.5 s), and design 2 is tailored to capture most energy between 6–7 s (RAO = 6 s). So the distribution of wave period within the selected period does influence the results. For the winter season, the wave period falls below 6 s for 95 times out of 168 for the hourly resolution data and 4 times out of 7 times for the daily resolution data. Similarly for the summer season, the wave period falls below 6 s for 23 times for the hourly data and zero time for the daily data. The results of the two-sample two-tailed t-tests are shown in Table 14, where the estimated power outputs are the samples.


**Table 14.** Results of Series of *t*-tests.

From the results, a p-value less than or equal to 5% indicates that the variables (parameters) significantly influenced the power output. First, different designs (design 1 versus design 2) significantly affected the power output irrespective of the season and diameter. Second, for the seasonal variations, most results were significant except for when the diameter was 7 m in design 2. Last, the size of the diameters in most cases significantly affected the power output of the device, except for design 1 during the winter time.

The factorial analysis was performed on the power output with design, diameter and season as the factors. Each factor had two different levels, so a 2<sup>3</sup> factorial design was formed and tested. The results from the factorial analysis are shown in Table 15. The results show a *p*-value of less than 5% for all the main effects and interaction between design and season, while the remaining two-way and three-way interactions were not significant.

**Table 15.** Results (*p*-value) of the 2<sup>3</sup> factorial design analysis.


#### **6. Conclusions**

The effects of the spatial and temporal resolution of the ocean wave data on the design of heaving point absorber and its power capture have been analyzed in this study. The effects have been analyzed quantitatively by comparing power capture performance of different designs of heaving point absorber based on different temporal and spatial wave conditions. By applying the normal convention to design the WEC device to resonate, it would capture energy in the most prevalent ocean wave period. The results show that different temporal resolution data lead to different designs, which may capture different amount of energy. However, the difference in temporal resolution of data did not significantly affect the estimation of the ocean wave theoretical power present in a location.

The results confirm the importance of designing a WEC device specific to a location. The WEC device designed to operate in a region in the Gulf of Mexico may generate much less energy in an offshore location in Oregon, even though Oregon location has almost 40 times more ocean wave energy potential than the location in the Gulf of Mexico. Results of these analysis show that the power captured by the same device in the Gulf of Mexico was higher despite having lesser wave energy potential. This is because the WEC was designed to resonate with the prevalent ocean wave condition in the Gulf of Mexico, so its performance was poor in the Oregon location. The influence of the WEC design, diameter and seasonal changes were also examined on the power output of the device. These effects were analyzed using *t*-tests and factorial analysis. The results show that all the parameters to varying degrees have significant effect on the power output. However, only the interaction between design and season shows significant effect. Furthermore, a linear damper of 40 kNm/s and 50 kNm/s was used

in this paper to represent the power take off for the 7 m and 8 m buoy. Future studies will perform sensitivity analysis to determine the optimal damper and its nonlinear behavior.

Studying the complexities that exists in the interaction of WEC devices with the ocean waves will increase the knowledge in the design of more effective devices, which will make the penetration of ocean wave energy increase. The scope of this study deals with only the primary capture of mechanical energy from the ocean waves through the hydrodynamic interaction of the WEC device and is most useful at the feasibility stage of the project. This aspect is only a part of the wave energy process, which includes the hydrodynamic conversion, conversion to electrical energy, and transmission. The wave condition data used in this paper was obtained from NOAA directly, which provides hourly and daily resolution. During the feasibility analysis stage of wave energy harvesting and WEC design, it was possible to directly use existing wave condition data with default temporal resolution as many existing studies did [4,6,8,40–45]. Meanwhile, IEC TC114 technical specification recommends 30 min temporal resolution when designing WECs for commercial use. Decision makers and WEC designers should choose the right temporal resolution based on different needs and budget limitations. As the commercialization of WECs draws more attention, it is expected more wave condition data with 30 min temporal resolution may become available, which could reduce the additional cost for collecting wave energy data for WEC designs. In addition, the structural reliability of any WEC devices should also be considered in the design stage, especially their reliability under extreme and hash ocean wave conditions.

**Author Contributions:** T.A. conducted data collection, product design and data analysis with the suggestions and guidance from H.L. T.A. wrote the initial draft paper under the supervision of H.L. H.L. made major revision on the initial draft paper, and approved the final version to be published. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors are thankful to the support from Texas A&M University-Kingsville and National Science Foundation (award # EEC-1757812).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Emergy and Sustainability Ternary Diagrams of Energy Systems: Application to Solar Updraft Tower**

### **Islam Elsayed \* and Yoshiki Nishi \***

Department of Systems Design for Ocean-Space, Graduate School of Engineering Science, Yokohama National University, 79-5 Tokiwadai, Hodogaya, Yokohama, Kanagawa 240-8501, Japan

**\*** Correspondence: islam-elsayed-bz@ynu.jp (I.E.); nishi-yoshiki-rg@ynu.ac.jp (Y.N.)

Received: 26 November 2020; Accepted: 14 December 2020; Published: 16 December 2020

**Abstract:** To facilitate sustainable energy development, one has to understand the limited availability of nonrenewable energy resources, and the ability of the earth to renew or recover. Emergy is an instrument that measures environmental loading, ecological economics, and regional sustainable development. In this study, emergy indicators are calculated to investigate the sustainability of solar updraft tower (SUT). SUT produces energy from the hot air, utilizing a combination of a solar collector, central tower, and air turbines. The results demonstrate that the sustainability of SUT grew as the size of the plant increased. Further, emergetic ternary diagrams are drawn to facilitate the comparison between SUT and various technologies. The resources-use efficiency of wind energy and SUT, 200 MW is found to be the lowest among all energy technologies presented in this research. Scenario analysis is performed to explore the future optimization directions. The results demonstrate that the development direction of SUT systems should mainly focus on reducing the materials demanded by the manufacturing and construction of its solar collectors. This study aims to demonstrate the value of emergy as a powerful instrument for drawing long-term sustainable strategies in energy markets for a greener tomorrow.

**Keywords:** emergy; emergetic ternary diagrams; sustainability; environmental loading; energy systems assessment; solar updraft tower

#### **1. Introduction**

The World Commission on Environment and Development in 1987 developed the official definition of the sustainable development concept for the first time, describing it as "the development which meets the needs of current generations without compromising the ability of future generations to meet their own needs" [1,2]. The realization of such development necessitates awareness of the limited availability of nonrenewable resources, and the ability of the earth to renew or recover [3]. The greening of energy emerged as a key element in any region to facilitate sustainable socioeconomic development. Total global carbon emissions from the power sector were about 13.4 billion tons in 2016, about 42% of total carbon emissions from fuel combustion. Coal-fired power plants are one of the largest sources of air pollution. High CO<sup>2</sup> levels accelerated the negative impact on the environment, including, but not limited to, global warming and the rising sea level. An inefficient relationship with the environment will lead to a dramatically negative impact on societies and economies [4]. Therefore, clean energy infrastructure is essential economically, politically, environmentally, and socially for a greener tomorrow.

While the public emphasizes the imperative of green energy infrastructures, engineers and researchers doubt the investment in greener generation technologies like wind and solar unless new evaluation mechanisms are introduced to support renewables and decarbonatization strategies [5]. On the one hand, the de-risking of investments and the idea of reliable sources of energy limit

investment strategies in renewables worldwide. On the other hand, some countries around the globe (Denmark, 2019) had nearly half of their total energy consumption powered by renewable resources (wind energy) [4]. Thus, there is a body of academic knowledge that claims that the efficient management of renewable resources and the diversified use of green energy technologies can offer a reliable energy infrastructure. To accelerate and strengthen the greening of the energy industry, one has to understand the complexity of the energy markets and their abundant considerations.

This paper addresses "emergy", proposed by Howard T. Odum [6,7], as a powerful instrument for measuring sustainable development. Emergy is able to facilitate regional sustainable development, ecological management, and environmental conservation for policymakers [8]. Emergy analysis considers that the formation of any system on earth derives from solar energy. The emergy value of any resource is its "solar energy memory" or, in other words, the amount of solar energy directly and/or indirectly used in its formation process. Different types of resources (material, energy, labor, etc.) possess different properties and are expressed in different units. The concept of solar transformity in emergy analysis quantifies the emergy value of any resource in one unified unit. Consequently, the framework of emergy analysis in the environmental accounting of any economic system requires ecological and economic emergy transformation chains to determine the total emergy input that drives the system (system empower). As a result, all systems are evaluated on an ecocentrism basis [3,7,9].

M. T. Brown and S. Ulgiati [3] presented emergy in the evaluation of electricity production systems and compared six power plants based on their emergy indicators. Consequently, many researchers reported the evaluation of various power generation technologies based on an emergy analysis. These studies include, but are not limited to, hydropower stations [10,11], geothermal and biomass [12,13], wind energy [14–16], solar photovoltaics and solar thermal collectors [17], and solar tower power plant systems [18]. Furthermore, emergetic ternary diagrams were presented by B. F. Giannetti et al. [19] as graphic tools that offer not only prompt visualization of emergy analysis but also readily recognize system improvements. C. M. V. B. Almeida et al. [20] discussed emergetic ternary diagrams in the application of electricity production systems.

Solar updraft tower (SUT) is one of the green technologies for acquiring renewable energy [21–24]. In nature, the updraft effect causes wind and hurricanes. The geometry of SUT mimics the wind cycle in nature to produce the thermal wind that drives turbines to generate electricity. It generates power using the sun as the only source of energy and using only air as working fluid without phase change; there is no water demand or working mediums, and no cooling mechanisms are needed [25]. While many areas worldwide suffer from severe water shortages, SUT saves gallons of cooling water associated with conventional and solar thermal generation plants [25]. Together with its low-cost thermal energy storage potential, SUT energy can be generated around the clock, eliminating environmental threats associated with the use of batteries in photovoltaic and wind power generation plants. Numerous studies reported cost–benefit models to assess the economic feasibility of SUT power plants [26–28]. Further, other researchers reported the ecological analysis in the lifecycle of SUT power plants [22,29]. I. Elsayed and Y. Nishi examined the sustainability of SUT power generation technology using the Inclusive Impact Index concerning both ecological and economic costs [30]. Despite that, emergy analysis of an SUT power plant has not been reported.

The main contribution of the present study is the evaluation of large-scale SUT power plants using emergy as a benchmark (Figure 1). Emergy accounting and emergy-based indicators are estimated to investigate the relative performance and sustainability of SUT systems. Comparisons of emergy indicators with various conventional and renewable power-generation technologies are then performed. To facilitate the latter comparison, emergetic ternary diagrams are drawn to provide a better interpretation of emergy accounting results and to visualize the interactions between power plants and the environment. Additionally, based on emergetic ternary diagrams, scenario analysis is performed to explore future optimization directions of large-scale SUT power plants. This paper offers a way forward to draw the sustainability pathways of large-scale SUT power systems for optimal socioeconomic decision-making.

**Figure 1.** Graphical abstract of emergy analysis for the lifecycle of solar updraft tower power plants.

#### **2. Methodology and Data**

#### *2.1. Solar Updraft Tower: Design Parameters*

SUT produces energy from the hot air, utilizing a combination of a flat plate solar collector, a central tower, and air turbines. The air is heated by the greenhouse effect in the collector; this hot air produced is lighter than the ambient cold air at the top of the tower, and, consequently, the hot air rises up the tower. In other words, the density difference of the air caused by the temperature rise in the collector is converted to a pressure difference, which generates a fluid flow (thermal wind) that drives the turbines to generate electricity [21–25]. In 1983, W. Haaf et al. presented the first large-scale SUT pilot plant, which was in Manzanares, Spain [21]. The plant operated with a tower (steel), height of 194.6 m; a collector (plastic), diameter of 244 m; and a capacity of 50 kW. Following the measurements and investigations from the Manzanares pilot plant, researchers put forward detailed calculations and structural design proposals for large-scale SUT plants. J. Schlaich et al. [22,23] established a design model for commercial large-scale SUT plants of different capacities and calculated the costs and GHG emissions per kWh produced.

The technical constraints of building 1000 m high towers are a challenge; however, they can be built today, and several SUT projects are in development [22]. EnviroMission, an Australian company, began moving forward to build a 200 MW SUT power plant in Arizona, USA [31]. The project proposal includes a concrete tower of 800 m in height, a glass collector of 4800 m in diameter, and 32 horizontal axis turbines (6.25 MW each). Shimizu, a Japanese corporation, is in the process of developing a residential project known as "The Green Float" [32]. With a circular section extending to a diameter of 3000 m, and a central tower section of 1000 m in height, the latter project has the same configurations as a large-scale SUT plant. Shimizu shows that such construction can be done soon and provides insights for the implementation of SUT plants in a desert landscape, as well as nearshore or offshore.

Therefore, it is necessary to perform an impact assessment of SUT power plants and investigate their sustainability. The J. Schlaich et al. [22] model of commercial SUT power plants of different capacities (Table 1) is used in the present study to establish emergy analysis for SUT systems.


**Table 1.** The design parameters of J. Schlaich et al. [22] for SUT commercial models used in the calculations of the present study.

<sup>1</sup> The SUT power plant site is assumed to have an annual global solar radiation of 2300 kWh/m<sup>2</sup> /y. Besides, there is no additional thermal storage.

#### *2.2. Materials and Resources in the Lifecycle of SUT Power Plants*

Since SUT power plants demand massive structures and materials, questions of technical constraints and energies expended to construct the plant are recurrent [21]. Estimating from the Manzanares pilot, W. Haaf et al. reported the building materials and energies to construct a 100 MW SUT (concrete tower and glass collector) [21]. Subsequently, J. Schlaich and other researchers reported the lifecycle analysis of SUT power plants [22,30]. The turbines are at the periphery of the transitional area between the collector and tower. J. Schlaich et al. reported an approximate number of 24 to 36 turbines for large-scale SUT power plants [22]. Thus, this study considers the use of horizontal-axis turbines with rated capacities of 4.1 and 5 MW. The design theory of the SUT turbines is usually adapted from horizontal-axis wind turbines [22,26,28]. Henceforth, the materials of 4.1 and 5 MW wind turbines are quoted from [33,34].

The construction period of a large-scale SUT power plant is approximately two to three years [22,29,31]. This study assumes that the materials require transportation from a region 1000 km away from the plant's location. The present study adapted the materials, weights, and energy for the construction of the tower and collector from W. Haaf et al. [21]. Table 2 illustrates the materials, component weights, and machinery for construction of the study's SUT power plants.


**Table 2.** Materials and energy consumption for the construction of commercial SUT power plants of different capacities.

In the present analysis, there is no additional thermal storage, thus no salt or water usage. Additionally, no replacement tower or collector is in place [29]. Therefore, work for the operation and maintenance during the lifetime of the plant is necessary mainly for the turbines. A great deal of data from wind power stations can be used to estimate turbine substitutions' materials in the lifecycle. The intended service life of commercial SUT power plant ranges from 80 to 120 years [27]; this study considers a lifetime of 30 years.

#### *2.3. Emergy Analysis*

The sun is the primary source of energy that fuels the earth's biological and economic growth. Emergy theory, proposed by H. T. Odum, traces the origin of the products and services formation. Considering that solar energy is the main energy input to the Earth, emergy measures how much energy would be needed to form a resource if solar radiation were the only input. Thus, emergy describes the amount of one form of energy invested in the system or, in other words, the "energy memory" of any system. Emergy converts different forms of energy, materials, goods, and human labor and services to equivalents of solar energy in a unified unit, solar embodied joules, abbreviated sej. The emergy required to make one unit of product is called the "solar transformity (sej/unit)" [6,7,9,35].

Emergy assumes all living systems sustain one another by participating in a network of energy flows; even the economy can be incorporated into this energy flow network. Hence, emergy analysis for any technology measures all input and output quantities, considers it as a network of emergy flows, and then determines the total emergy value. This approach helps to analyze the effect of each material and process, and the contribution of the ecosystem (environmental services); thus, it guides the future development of the technology. In emergy community, price does not reflect the real value of the environmental resources. The money is paid only to humans for their services, not to the environmental contributions and resources [35]. The emergy evaluation of any system can be divided into three consecutive steps. The first determines the analysis scale and boundary (analyzing the input and output data of the system). The second computes an emergy diagram and evaluation table. Finally, the third calculates emergy-based indices.

#### 2.3.1. Analysis Scale and System Boundary

Emergy analysis is used in this study to evaluate the sustainability of power generation systems. The research scale is a power generation plant, the solar updraft tower. In the case of power plants, the types of inputs include energy resources (oil, coal, wind, solar radiation, etc.), materials (concrete, steel, glass, etc.), and labor and services necessary to construct, operate, and maintain the plant. The boundary of the analysis covers all phases of the SUT including manufacturing, installation (construction and transportation), and operation and maintenance during its lifetime. The main output of the system is the electricity generated.

#### 2.3.2. Emergy Diagram and Evaluation Table

A typical emergy flow network is shown in Figure 2. The engineering systems comprise renewable and nonrenewable resources powered by both the ecosystem and human work. Therefore, the framework inputs of any engineering system are divided into three types: renewable resources (*R*), such as solar radiation, wind, rain, etc.; nonrenewable resources (*N*), such as land area, groundwater, etc. (limited availability); and purchased resources (*F*), which refers to economic transactions such as electricity, machinery, and human labor, etc. Renewable and nonrenewable resources are considered local resources. The emergy value of an input resource (*i*) is expressed as follows:

$$\text{Energy}\_{i} = \text{Quantity}\_{i} \times \text{Solar transfer} \text{(mity}\_{i} \tag{1}$$

**Figure 2.** Typical emergy flow diagram of SUT power plant. The diagram shows the environmental renewable (R) and nonrenewable (N) inputs supplied by the local ecosystem, together with purchased inputs (F) from the economic systems. The electricity produced by the plant is the output from the SUT system. Local air storage represents the atmosphere, which receives pollutants (W) and heat released by the power plant (H). Local sea storage represents the marine environment [3,19].

The emergy baseline is a basic parameter for the equation, as it is the base of solar transformities calculations. The emergy transformity data adopted from references should all have the same baseline. The emergy baseline of the solar transformities used in this study is "15.83 <sup>×</sup> <sup>10</sup><sup>24</sup> sej/y". The total emergy input into the system is described as the emergy yield driving the system which equals the sum of emergy inflows (*R* + *N* + *F*). The annual electricity production (AEP) of the power plant is converted to joules using the standard conversion of 3.6 <sup>×</sup> <sup>10</sup><sup>6</sup> J/kWh.

#### 2.3.3. Emergy indicators

Based on the above characteristics of emergy flows, emergy-based indices were established to investigate the efficiency and sustainability of various systems. In the present work, the most common emergy-based indicators were used to evaluate power generation from SUT power plants [3].

The transformity (*Tr*) of the system is the emergy invested to generate one unit of electricity. It measures the environmental and economic resources' use efficiency (emergy conversion efficiency) of the system as a whole (Equation (2)) (*Eout* is the total electricity production).

$$Tr = \frac{R + N + F}{E\_{\text{out}}} \tag{2}$$

Emergy yield ratio (*EYR*) is the total emergy yield driving the system divided by the outside emergy sources (goods and services) from the economy. It measures the ability of the system to rely on local resources. The higher the value of this index, the larger the contribution to the economy per unit of emergy invested (Equation (3)).

$$EYR = \frac{R + N + F}{F} \tag{3}$$

Environmental loading ratio (*ELR*) is the local nonrenewable resources and total purchased resources from the economy divided by local renewable resources. It measures the stress on the environment and the potential impact of the system on the ecosystem (Equation (4)).

$$ELR = \frac{N + F}{R} \tag{4}$$

Emergy index of sustainability (*EIS*) is the emergy yield ratio per unit of environmental loading. It measures the overall sustainability of the system (Equation (5)). A high resources utilization efficiency (*EYR*) and low stress on the ecosystem means more sustainable technology. A power plant with *EIS* higher than "1" is considered sustainable [3].

$$EIS = \frac{EYR}{ELR} \tag{5}$$

#### *2.4. Emergetic Ternary Diagrams*

A ternary diagram is a three-dimensional graph but is illustrated in two dimensions in a triangular coordinate system. This graph type is proposed for the analysis of mixed components. It is commonly used to visualize phase diagrams in the physical sciences [19,20]. Emergetic ternary diagrams were presented by B. F. Giannetti et al. [19] to represent graphically emergy indicators. The authors presented emergetic ternary diagrams as graphic tools that offer not only prompt visualization of emergy data but also readily illustrate system improvements [19].

The emergy driving the system (100%) equals the sum of the percentages of the three types of Emergy resources' inflows (*R*%, *N*%, and *F*%) which are the elements of the emergetic ternary diagram. The diagram is an equilateral triangle where each corner represents an element (resources type). The triangular coordinates are chosen with renewable resources (*R*) on the top apex, nonrenewable resources (*N*) on the left apex, and purchased resources (*F*) on the right apex. The total emergy input into the system is represented by points within the triangle, the percentage of each element being given by the length of the perpendicular line from the given point to the side of the triangle opposite to the appropriate element. Moreover, the emergetic ternary diagram has auxiliary lines to facilitate emergy analysis: resource flow lines, sustainability lines, and sensitivity lines. Resource flow lines define emergy indicators *EYR* and *ELR*. Sustainability lines define the emergy index of sustainability, *EIS*. The sensitivity lines point out the system's behavior with changes in elements (resource fluxes) associated with the apex of each sensitivity line. The latter lines are very useful for identifying areas of improvement to enhance the sustainability of the systems.

In the present work, the emergetic ternary diagrams were depicted to provide a better understanding of emergy flows for SUT systems and to visualize the interactions between the power plant and the environment. Moreover, the graphical analysis facilitates the comparison between SUT systems and various energy technologies presented in this research.

#### *2.5. Data Sources*

The studied SUT power plants consist of concrete towers and glass solar collectors. The data in this study are mainly from the feasibility and lifecycle assessment reports of large-scale SUT. The total area of the canopy is considered the plant area, at a site with an annual global solar radiation of 2300 kWh/m<sup>2</sup> /year [22]. The total masses and the electricity consumption for the construction of the tower and collector are adapted from the sample estimates of W. Haaf et al. [21]. Turbines' materials are calculated by analogy with similar wind turbines [34]. During the operation and maintenance phase, it is assumed no replacement of the tower or collector is taking place [29]. Meanwhile, one-third of the turbine's materials should be substituted during the average useful life [33,34]. Labor and services costs for installation (engineering, tests, etc.) and operation and maintenance are derived from [22]. In the calculation, considering the energy and materials loss in the production process, and due to the difficulty of getting data, some calculations are still not included [16]. The solar transformity coefficients of each resource are quoted from relevant references.

#### **3. Results: Emergy Accounting**

All the input flows during manufacturing, installation, and operation and maintenance of the SUT, 200 MW power generation plant are analyzed and described in Table 3. Energy sources, materials, land loss, and human labor and services expressed in their common units (J, kg, m<sup>3</sup> , \$, etc.) were converted to emergy flows (sej) by means of appropriate emergy transformation coefficients (called solar transformity, sej/unit). The total emergy of the SUT system accounts for all environmental contributions, energies, and human services used directly and indirectly to produce and maintain the power plant. The fractions of emergy inputs for the SUT, 200 MW plant are depicted in Figure 3. The manufacturing materials of the components are the dominant inputs, constituting 34.9% of total emergy. This is followed by construction works (materials and energy consumption), constituting 29.0% of total emergy. The latter results are the output of the massive structure and materials demanded by a large-scale SUT power plant. Meanwhile, the glass of the canopy makes up the largest proportions of components' manufacturing materials, constituting 20.8% of the total emergy inputs. The emergy invested by local environmental resources like solar radiation and land area contribute small fractions of the total emergy driving the system; this is because of the small solar transformity value of the latter resources. The other generation capacities of the SUT systems are evaluated using emergy tables like Table 3, and the results of emergy flows are listed in Table 4.


**Table 3.** Emergy evaluation table of SUT, 200 MW power generation plant during lifetime.

**Figure 3.** Fractions of emergy input flows for SUT, 200 MW.

**Table 4.** Emergy input flows and annual energy production of commercial SUT power plants of different capacities.


#### **4. Evaluation and Discussion**

## *4.1. Emergy-Based Indicators*

In order to investigate the relative performance and sustainability of SUT generation systems, their emergy indicators are presented in Table 5, together with comparisons of various conventional and renewable power generation technologies. Although the emergy input flows increased as the capacity of SUT enlarged, the SUT, 100 and 200 MW performed the best with regard to emergy-based indicators. This confirms that the sustainability of SUT grew as the size of the plant increased. Similar conclusions were achieved using a different method to assess SUT power plants [30].


**Table 5.** Comparison of emergy resources' inflows and emergy indicators for various power generation technologies.

Transformity (*Tr*) is an important indicator to measure the overall efficiency of the production system at the biosphere scale. Power plants with greater transformities demand more environmental and economic resources to meet the same electricity demand. Large-scale SUT power plants are excellent in terms of emergy conversion, except for very small generation capacities. The transformity of onshore wind energy and SUT, 200 MW is found to be the lowest among all conventional and renewable generation technologies presented in this research (Figure 4). Further, it is the closest in value to the thermodynamic minimum transformity for electricity production cycles [3]. This is an outcome of the simplicity of this technology as its materials are very conveniently based on environmentally sound production from renewable or recyclable materials.

**Figure 4.** Emergy transformity of various power generation technologies.

The emergy yield ratio (*EYR*) of SUT, 200 MW (1.1) is higher than solar photovoltaics. However, the value is much smaller than those of the other power generation plants. In general, *EYR* evaluates the protentional contribution of the power plant to the economy, and the smaller *EYR* of SUTs indicates a relatively high amount of emergy resources invested by the economy.

The environmental loading ratio (*ELR*) measures the environmental stress of the power plant. The *ELR* of SUT, 200 MW (12.27) is small compared to solar photovoltaics and oil-fired power generation plants. However, the latter value is larger than those of the other renewable power plants. The larger

environmental stress is the negative effect of the enormous land area (nonrenewable resources) demanded by the SUT power plant compared to other renewable technologies.

The sustainability of power generation plants is interpreted by the emergy Index of Sustainability (*EIS)*, which is the ratio of *EYR* to *ELR*. It should be considered that an *EIS* less than one indicates an unsustainable process from the viewpoint of emergy-based indices [3]. The *EIS* of large-scale SUT (0.09) is higher than the solar photovoltaics. However, the latter value is lower than other power plants and lower than one. This is because of the high value of *ELR*.

#### *4.2. Emergetic Ternary Diagram*

The graphical analysis assists the interpretation of emergy accounting results and the discussion. Therefore, emergetic ternary diagrams were drawn to facilitate the comparison between SUT and various energy technologies. In Figure 5, points #1 to #10 represent SUT and various power generation technologies which are presented in this paper (Table 5). The position of any point is plotted based on the percentages of the three emergy resources inflows *R*, *N*, and *F*. To clarify, in Figure 5a, point #10 (coal-fired, 1280 MW) is composed of 9% of *R*, 73% of *N*, and 18% of *F*. The resource flow lines are parallel to the triangle side and indicate a constant value for each resource flow (Figure 5b). They also identify emergy indicators *EYR* (yellow lines) and *ELR* (green lines). The resource flow lines are very useful for comparing the use of resources by each power plant [19,20]. By way of illustration, it can be observed that geothermal (#7) and coal-fired (#10) had almost the same quantity of purchased resources (*F*), but the quantities of renewable and nonrenewable renewable resources (*R* and *N*) are extremely different. Moreover, it can be observed that SUT, solar photovoltaics, and wind-powered plants demanded higher quantities of purchased resources (*F*) than other power generation plants presented in this research.

**Figure 5.** Emergetic ternary diagrams representing SUT and various power generation systems shown in Table 5. (**a**) represents the composition of point #10 (coal-fired, 1280 MW) and (**b**) the use of resource flow lines (*EYR*: emergy yield ratio; *ELR*: environmental loading ratio).

The emergy index of sustainability *EIS* is presented in emergetic ternary diagrams through the sustainability lines (orange lines) which indicate constant values of *EIS* (Figure 6a). The sustainability lines depart from the *N* apex in the direction of the *RF* side. According to M. T. Brown and S. Ulgiati [3], power plants with an *EIS* higher than one denote sustainable power generation in a long term. Consequently, the sustainability line where *EIS* = 1 divides the triangle into sustainability areas. The presentation of the sensitivity lines (Figure 6b) is very significant for identifying critical elements that may be changed to improve the sustainability of power plant systems. The sensitivity lines are straight lines that depart from an apex towards the opposite side. They can be drawn for any point (system) inside the triangle and trace the impact of changes in a given resource flux (*R*, *N*,

or *F*), while the other two fluxes remain constant. By way of illustration, it can be observed that the sustainability of SUT, 200 MW (#4) can be improved by changing the quantities of purchased resources. In the same way, an increase in the renewable services of a coal-fired (#10) plant may enhance its sustainability. As such, the sensitivity lines are a powerful tool to draw the sustainability pathways for decision-makers [19,20].

**Figure 6.** Representation of the auxiliary lines of emergetic ternary diagrams: (**a**) the use of sustainability lines, and (**b**) the use of sensitivity lines for SUT and various power generation systems shown in Table 5 (*EIS*: emergy index of sustainability).

#### *4.3. Scenario Analysis*

To further investigate the above results and explore future optimization directions of the SUT power plant, scenario analysis was performed. Based on the emergy index of sustainability, the sustainability of the SUT power generation system is less than coal-fired and oil-fired power generation systems. Further, its value is less than one which means unsustainable generation in the long term. Among the other capacities of SUT, large-scale SUT, 200 MW performs the best with regard to emergy-based indicators. It is worth studying SUT, 200 MW in terms of *EIS* change.

Derived from the sensitivity lines of emergetic ternary diagrams, the sustainability of SUT is strongly relative to purchased resources. The scenario analysis focus on dominant purchased resources includes the materials of the tower, collector, and turbines (manufacturing and construction materials). Additionally, the transportation distance of materials to the plant's location (1000 km in this study) was accounted for. The latter resources are taken as variables and the relationship between *EIS* is shown in Figure 7. The *EIS* of the SUT, 200 MW is mostly affected by the emergy of collector materials, followed by tower materials. On the other side, the turbines' materials and transportation distance have weak impacts. Therefore, the development direction of large-scale SUT power generation systems should mainly focus on reducing materials demanded by the manufacturing and construction of solar collectors.

The service life of a commercial SUT power plant ranges from 80 to 120 years [27]. Figure 8a shows the *ESI* of SUT, 200 MW changes with the length of lifetime (30 years in this study). It can be seen that the length of the SUT lifecycle has a slight impact on its sustainability.

As a final point, the emergy indicators are relative measures and depend on the characteristics of the fractions of emergy inputs. To compare different systems by the emergy index of sustainability, one has to follow the perspective of the evaluation [3]. To clarify further, the renewable fraction of human labor and services changes with the economic level of different regions and time periods, which directly affects *EIS* [3,14]. In fact, numerical variations of emergy indicators can be found in the literature for the same power generation technology. In the present analysis, the renewable fraction of human labor and services is considered as 26% [36]. In Figure 8b, the change of the *EIS* with the

percentage of the renewable fraction of human labor and services is observed. Notably, the change in renewable fraction has a relatively high impact on *EIS*. Therefore, in the emergy evaluation, the region and economic level should be unified to facilitate a successful comparison of different power generation systems.

**Figure 7.** Percentage increase in *EIS* of SUT, 200 MW with the reduction of the tower, collector, and turbines' materials, as well as transportation distance of these materials to the plant's location.

**Figure 8.** Changes in *EIS* of SUT, 200 MW with: (**a**) lifetime length of the power plant and (**b**) renewable fractions of human labor and services.

#### **5. Conclusions**

Clean energy infrastructure is essential economically, politically, environmentally, and socially for a greener tomorrow. To accelerate and strengthen the greening of the energy industry, the sustainable evaluation mechanisms of energy systems should be empowered. This paper addresses "emergy" as a powerful instrument to facilitate efficient energy management for policymakers as they choose

alternatives concerning the environment. The framework of emergy analysis for the power plant system involves ecological and economic emergy transformation chains to determine the total emergy input that drives the system. Thereupon, all energy systems are evaluated on an ecocentrism basis. Solar updraft tower (SUT) is one of the green technologies for acquiring renewable energy. SUT produces energy from the hot air, utilizing a combination of a flat plate solar collector, a central tower, and air turbines. Emergy accounting and emergy-based indicators are estimated to investigate the relative performance and sustainability of four power-generation capacities of SUT power plants.

Energy sources, materials, land loss, and human labor and services were converted to emergy flows. The total emergy of the SUT system accounts for all environmental contributions, energies, and human services used directly and indirectly to produce and maintain the power plant. The SUT, 100 and 200 MW performed the best with regard to emergy-based indicators. The results demonstrate that the sustainability of SUT grew as the size of the plant increased. For SUT, 200 MW, the manufacturing materials of the components are the dominant inputs, constituting 34.9% of total emergy. This was followed by the construction works (materials and energy consumption), constituting 29.0% of total emergy. The latter results are the output of the massive structure and materials demanded by a large-scale SUT power plant. Large-scale SUT power plants (100 and 200 MW) are excellent in terms of resources-use efficiency (emergy conversion). The emergy transformity of onshore wind energy and SUT, 200 MW was found to be the lowest among all conventional and renewable generation technologies presented in this research. This is the outcome of the simplicity of this technology as its materials are very conveniently based on environmentally sound production from renewable or recyclable materials.

Emergetic ternary diagrams were drawn to facilitate the comparison between SUT and various energy technologies. The sensitivity lines of emergetic ternary diagrams are very significant in identifying critical elements that may be changed to improve the sustainability of power plant systems. Derived from sensitivity lines, the sustainability of SUT is strongly relative to purchased resources. Scenario analysis was performed to explore future optimization directions of the SUT power plant. The scenario analysis focus on dominant purchased resources included the materials of the tower, collector, and turbines (manufacturing and construction materials). The transportation distance of materials to the plant's location was also taken into account. The *EIS* of the SUT, 200 MW was mostly affected by the emergy of collector materials, followed by that of tower materials. On the other side, the turbines' materials and transportation distance had weak impacts. Therefore, the development direction of large-scale SUT power generation systems should mainly focus on reducing materials demanded by the manufacturing and construction of solar collectors. Moreover, it can be seen from the analysis that the length of the SUT lifecycle had a slight impact on its sustainability. Notably, the change in the percentages of the renewable fraction of human labor and services had a relatively high impact on *EIS*. The renewable fraction changes with the economic level of different regions and time periods. Therefore, in the emergy evaluation, the region and economic level should be unified to facilitate a successful comparison of different power generation systems.

**Author Contributions:** I.E.: conceptualization, methodology; analysis, investigation, visualization, writing—original draft preparation. Y.N.: methodology, writing—review and editing, supervision, funding acquisition. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by Grant-in-Aid for Scientific Research (B) (No. 18H01636) from the Japan Society for the Promotion of Science.

**Acknowledgments:** The first author, I.E., is indebted to the Japanese Ministry of Education, Culture, Sports, Science and Technology (MEXT) for financial support. Islam is also grateful to Camila Gonzalez Lopez for her support to refine the manuscript.

**Conflicts of Interest:** The authors declare no conflicts of interest.

### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Evaluation of the Operating E**ffi**ciency of a Hybrid Wind–Hydro Powerplant**

#### **Francisco Briongos <sup>1</sup> , Carlos A. Platero <sup>1</sup> , José A. Sánchez-Fernández 2,\* and Christophe Nicolet <sup>3</sup>**


Received: 17 December 2019; Accepted: 14 January 2020; Published: 16 January 2020

**Abstract:** This paper analyzes the operating efficiency of a hybrid wind–hydro power plant located in El Hierro Island. This plant combines a wind farm (11.48 MW) and a pumped storage power plant (11.32 MW). It was built with the aim of supplying the island demand from renewable energy instead of using existing diesel units. The paper discusses several operational strategies and proposes an efficiency metric. Using 10 min data, the operation of this power plant has been simulated. From these simulations (more than 50,000 for a year), the operating efficiency and the percentage of demand covered from renewable energy is obtained. The difference between the worst and the best strategy is a twofold increase in efficiency. Moreover, the results of the simulations are compared with the system operational history since June 2015 (when the wind–hydro power plant started operation) until 2018. These comparisons show a reasonable agreement between simulations and operational history.

**Keywords:** wind farm; pumped storage; isolated systems; power plant efficiency

#### **1. Introduction**

Since the beginning of electrification, fossil fuel burning has been the main source of primary energy on which electricity generation is based [1]. However, there is growing evidence that fossil fuel burning for electricity generation is one of the main causes of global warming because of greenhouse gas emissions [2]. For this reason, renewable energy resources have had an important development. Among these renewable energy resources, wind energy has the largest installed capacity in the world [3]. Notwithstanding, wind speed variations and its prediction difficulties create some problems in power system operation. Therefore, when wind penetration increases, power system operators have to take measures to maintain system integrity [4] or consider complementary renewable energy sources [5].

Small islands are of particular interest because, on them, fossil fuel costs are higher [6]. Therefore, renewable energies can be cost competitive. In fact, many islands are installing some renewable energy supplies [7]. To this aim, the support of the local community is an important contribution to success, as the case of Samsø has proven [8]. However, island power systems usually lack the support of a strong interconnected system. So, maintaining system integrity is a harder problem on islands [9]. Notwithstanding, there are several examples of island power systems that have largely increased their share of renewables. To this aim, one successful approach is combining several renewable energy sources [10]. Another option is adding an energy storage system [11]. Furthermore, there are islands that combine both possibilities, such as the Caribbean island of St. Eustatius [12]. In addition, the possible contribution of electric vehicles smart charging to solar and wind integration has been proposed in The Barbados [13].

One of those islands is El Hierro, an island located in the Canary archipelago (Spain). UNESCO declared it a biosphere reserve in 2000. As such, it aims to become entirely free from carbon dioxide emissions [14]. To fulfill this aim, a combined wind–hydro power plant was built. This plant started operation in 2014 but was not fully operational until 2015 [15]. wind integration has been proposed in The Barbados [13]. One of those islands is El Hierro, an island located in the Canary archipelago (Spain). UNESCO declared it a biosphere reserve in 2000. As such, it aims to become entirely free from carbon dioxide emissions [14]. To fulfill this aim, a combined wind–hydro power plant was built. This plant started

Eustatius [12]. In addition, the possible contribution of electric vehicles smart charging to solar and

According to the Merriam-Webster dictionary, efficiency is "the ratio of the useful energy delivered by a dynamic system to the energy supplied to it" [16]. Therefore, each energy conversion technology has its own efficiency. However, the evaluation of energy efficiency in power systems has been based on economic metrics [17]. In [18], Maheshwari and Ramakumar described a procedure to design an optimal size microgrid that supplies energy to a rural area, but do not analyze nor define its operating efficiency. In [19], several procedures to calculate the optimum number, size and location of photovoltaic power plants is applied to two different test networks. The optimality condition tries to minimize power line losses and harmonic distortion. So, a technological metric for evaluating power systems efficiency is needed. This paper intends to do so, using an isolated power system that has a wind–hydro hybrid power plant. operation in 2014 but was not fully operational until 2015 [15]. According to the Merriam-Webster dictionary, efficiency is "the ratio of the useful energy delivered by a dynamic system to the energy supplied to it" [16]. Therefore, each energy conversion technology has its own efficiency. However, the evaluation of energy efficiency in power systems has been based on economic metrics [17]. In [18], Maheshwari and Ramakumar described a procedure to design an optimal size microgrid that supplies energy to a rural area, but do not analyze nor define its operating efficiency. In [19], several procedures to calculate the optimum number, size and location of photovoltaic power plants is applied to two different test networks. The optimality condition tries to minimize power line losses and harmonic distortion. So, a technological metric for evaluating power systems efficiency is needed. This paper intends to do so, using an isolated power system that has a wind–hydro hybrid power plant.

This paper is organized as follows: Section 2 describes the El Hierro power system. Section 3 analyzes the operational strategies that were considered. Section 4 presents the simulation model of the power system. Section 5 presents the results of the simulations done and discuss the results comparing them with the operating experience. Finally, Section 6 summarizes the conclusions. This paper is organized as follows: Section 2 describes the El Hierro power system. Section 3 analyzes the operational strategies that were considered. Section 4 presents the simulation model of the power system. Section 5 presents the results of the simulations done and discuss the results comparing them with the operating experience. Finally, Section 6 summarizes the conclusions.

#### **2. Power System Description 2. Power System Description**

The power system considered for the case study is located on El Hierro Island. This Island is placed in the Canary Islands archipelago, Spain. Its electric power supply was based on diesel engine driven generators, which is a common solution in small islands. There are 10 diesel engines, with rated power from 775 kW to 1870 kW, installed in several phases. The generators are connected to the main busbar through transformers. Electric energy is supplied to the island through four 20 kV power lines [20]. Figure 1 shows a simplified one-line diagram of the El Hierro power system. The power system considered for the case study is located on El Hierro Island. This Island is placed in the Canary Islands archipelago, Spain. Its electric power supply was based on diesel engine driven generators, which is a common solution in small islands. There are 10 diesel engines, with rated power from 775 kW to 1870 kW, installed in several phases. The generators are connected to the main busbar through transformers. Electric energy is supplied to the island through four 20 kV power lines [20]. Figure 1 shows a simplified one-line diagram of the El Hierro power system.

**Figure 1.** El Hierro power system simplified one-line diagram. **Figure 1.** El Hierro power system simplified one-line diagram.

El Hierro Island's orography makes it possible to build an upper and a lower reservoir at a reasonable cost and with a gross height difference larger than 700 m. On the other hand, this island has high wind during long periods. After several studies discussing using fresh or salt water for the El Hierro Island's orography makes it possible to build an upper and a lower reservoir at a reasonable cost and with a gross height difference larger than 700 m. On the other hand, this island has high wind during long periods. After several studies discussing using fresh or salt water for the hydro power plant, in 2015, a wind farm combined with a pumped storage power plant started commercial operation (using fresh water).

The wind farm comprises five Enercon E-70 wind turbines connected to the 20 kV grid, through separate transformers. The pumped storage power plant has two separate facilities, a hydropower plant and a pumping station. This arrangement is known as 4-machine-type [21]. The Appendix A provides some tables with detailed data on this wind–hydro power plant.

The hydro power plant comprises four Pelton turbines (Table A1), driving synchronous generators (Table A2) with a rated power of 2830 kW, connected to the grid by four transformers (Table A3). The pumping station has eight centrifugal multistage pumps driven by induction motors. There are two variable speed pumps of 1500 kW (Table A6) and six fixed speed pumps of 500 kW (Table A9) [20].

#### **3. Operating e**ffi**ciency and operational strategies**

The efficiency of a system is the ratio of the energy obtained and the energy used. In the case of a power plant is the ratio of energy generated and primary energy used.

The traditional definition of efficiency in case of windmills is the ratio of the wind power converted into electric power and the available wind power. The average efficiency of a windmill is between 32%–43%.

#### *3.1. Operating E*ffi*ciency*

However, in case of the operating efficiency, the definition is slightly different. It is the ratio between the generated energy and the maximum theoretical electric energy available using the best operational strategy.

In the case of a wind farm, the operating efficiency is the relationship of the produced energy (*EWind Produced*) and the maximum possible energy that can be generated in particular wind conditions (*EWind*∞). The maximum possible energy corresponds to the operation of the windmills at maximum power mode; so, the pitch should be regulated to get the maximum energy at any wind speed condition and connected to an infinite power grid.

$$
\eta = \frac{E\_{\text{Wind Produced}}}{E\_{\text{Wind}}} \tag{1}
$$

In order to get a 100% operating efficiency, the produced energy (*EWind Produced*) should be the maximum possible energy (*EWind*∞). This is only possible to reach in a large power system, where the power system is able to absorb all the produced energy. In the case of a hybrid wind–hydro power plant installed in a small power system, depending on consumers demand and wind speed, the hydro power plant should store in some periods and generate in others. Therefore, the maximum possible energy cannot always be put on line. Both process, energy storage (pumping mode) and the posterior hydraulic energy generation (turbine mode), have losses. The operating efficiency is defined as the direct wind energy (*EWind Grid*) put on line plus the hydro energy divided by the maximum energy that the windfarm would have produced in a large power system at maximum power mode (*EWind*∞).

$$\eta = \frac{E\_{\text{Wind Grid}} + E\_{\text{Hydro}}}{E\_{\text{Wind } \infty}} \tag{2}$$

The operational strategy of the power system has a large influence on its efficiency. In order to analyze this efficiency, all possible strategies should be considered. They have to take into account the non-dispatchable nature of wind energy. Therefore, these strategies are defined by how they manage wind power generation. In what follows, these operational strategies are discussed.

#### *3.2. Isolated Wind*

One of the possible operational strategies of this power system consists of splitting the 20 kV bus bar in two. One bar connects the wind farm and the pumping station. The other bar connects the remaining power system.

All the electric energy produced by the wind farm feeds the pumps through the first bus. Through the other bus, the conventional hydro power plant—in combination with the Diesel power plant—feeds the consumers. This is a well-known procedure; the operation is very safe, as the consumers are not exposed to wind power fluctuations. Another important advantage is that the frequency regulation in the windfarm and pumps 20 kV busbar is not critical, as the pumps have no stringent frequency requirements. All the electric energy produced by the wind farm feeds the pumps through the first bus. Through the other bus, the conventional hydro power plant—in combination with the Diesel power plant—feeds the consumers. This is a well-known procedure; the operation is very safe, as the consumers are not exposed to wind power fluctuations. Another important advantage is that the frequency regulation in the windfarm and pumps 20 kV busbar is not critical, as the pumps have no stringent frequency requirements.

The biggest drawback of this operational strategy is the efficiency. This operation mode has a low efficiency, as all renewable energy has been previously pumped and turbined. From the power system point of view, this case is equivalent to a 0% wind power penetration. As wind generators are not electrically connected to the power system, there is no limitation in the wind power. For the above reasons, the operating efficiency of this strategy is around 40%. The biggest drawback of this operational strategy is the efficiency. This operation mode has a low efficiency, as all renewable energy has been previously pumped and turbined. From the power system point of view, this case is equivalent to a 0% wind power penetration. As wind generators are not electrically connected to the power system, there is no limitation in the wind power. For the above reasons, the operating efficiency of this strategy is around 40%.

#### *3.3. Restricted Wind Power 3.3. Restricted Wind Power*

As wind energy is not fully controllable, the insertion of this type of energy in a small autonomous power system is not an easy task. Close attention should be paid to frequency regulation and wind gusts. A safe strategy is limiting the power injected by the wind farm. In this way, it is possible to get experience of the hybrid wind–hydro power plant operation and, afterwards, slowly increase this limit. When this limit is applied, the wind turbines operate at constant power and, consequently, some wind energy is not harvested. As wind energy is not fully controllable, the insertion of this type of energy in a small autonomous power system is not an easy task. Close attention should be paid to frequency regulation and wind gusts. A safe strategy is limiting the power injected by the wind farm. In this way, it is possible to get experience of the hybrid wind–hydro power plant operation and, afterwards, slowly increase this limit. When this limit is applied, the wind turbines operate at constant power and, consequently, some wind energy is not harvested.

This was the first strategy followed by El Hierro wind–hydro power plant when its operation started. Figure 2 shows wind generation limited to 4.4 MW. This was the first strategy followed by El Hierro wind–hydro power plant when its operation started. Figure 2 shows wind generation limited to 4.4 MW.

**Figure 2.** El Hierro power system daily generation and demand 25/11/2015. Source REE data. **Figure 2.** El Hierro power system daily generation and demand 25/11/2015. Source REE data.

#### *3.4. Restricted Wind Penetration 3.4. Restricted Wind Penetration*

One other safety-minded strategy is limiting the percentage of system demand covered directly by the wind farm. In this way, not all the energy demanded comes from the wind farm. In case of a sudden wind change or a wind generator trip, there are some backup generators in the power system than can supply the missed energy. However, using this strategy, sometimes there is wind energy that is failed to be harvested. One other safety-minded strategy is limiting the percentage of system demand covered directly by the wind farm. In this way, not all the energy demanded comes from the wind farm. In case of a sudden wind change or a wind generator trip, there are some backup generators in the power system than can supply the missed energy. However, using this strategy, sometimes there is wind energy that is failed to be harvested.

*Sustainability* **2020**, *12*, 668

This was the second strategy used by El Hierro wind–hydro power plant. Figure 3 shows wind penetration limited to 70% and wind power limited to 7.2 MW. This was the second strategy used by El Hierro wind–hydro power plant. Figure 3 shows wind penetration limited to 70% and wind power limited to 7.2 MW.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 5 of 17

**Figure 3.** El Hierro power system daily generation and demand 19/02/2016. Source REE data. **Figure 3.** El Hierro power system daily generation and demand 19/02/2016. Source REE data.

#### *3.5. Non-Restricted Wind 3.5. Non-Restricted Wind*

On this strategy, wind generators operate as if they were connected to an infinite bus. Therefore, they extract maximum power from the wind, with no limits apart from their own safety. This is only possible if the power system can absorb it. To this aim, the remaining parts of the power plant should adapt their point of operation accordingly. This means increasing or decreasing, in a fast way, the On this strategy, wind generators operate as if they were connected to an infinite bus. Therefore, they extract maximum power from the wind, with no limits apart from their own safety. This is only possible if the power system can absorb it. To this aim, the remaining parts of the power plant should adapt their point of operation accordingly. This means increasing or decreasing, in a fast way, the turbines generated, or the pumps demanded, power.

turbines generated, or the pumps demanded, power. From an efficiency point of view, this is the best strategy. However, integrating all this wind power in El Hierro is a complex problem. The power system should be able to absorb the energy from the wind farm. In case of a wind generator trip, the remaining generators should modify fast enough their generation to maintain frequency. Additionally, in case of wind generation larger than the consumers load, the pumping station should consume the exceeding power. In this way, the From an efficiency point of view, this is the best strategy. However, integrating all this wind power in El Hierro is a complex problem. The power system should be able to absorb the energy from the wind farm. In case of a wind generator trip, the remaining generators should modify fast enough their generation to maintain frequency. Additionally, in case of wind generation larger than the consumers load, the pumping station should consume the exceeding power. In this way, the frequency could be maintained.

frequency could be maintained. Therefore, there are two different problems in this strategy. The first one is increasing or decreasing, in a fast way, the non-wind generated power in cases of wind generation decreasing Therefore, there are two different problems in this strategy. The first one is increasing or decreasing, in a fast way, the non-wind generated power in cases of wind generation decreasing below the load [22,23]. The second is increasing or decreasing, also in a fast way, the pumping power demanded [24].

below the load [22,23]. The second is increasing or decreasing, also in a fast way, the pumping power demanded [24]. As can be seen in Figure 4, this operational strategy is used nowadays in El Hierro wind–hydro As can be seen in Figure 4, this operational strategy is used nowadays in El Hierro wind–hydro power plant.

#### power plant. **4. Simulation Model**

**4. Simulation Model**  The power system has been modelled in MS Excel using Visual Basic for Applications (VBA). The models use data measured every 10 min (wind speed and power system load).

The models use data measured every 10 min (wind speed and power system load).

The power system has been modelled in MS Excel using Visual Basic for Applications (VBA).

#### *4.1. General Power System Model*

In order to analyze how different operational strategies affect the power system, a simulation model has been developed. This model takes into account system restrictions and uses actual wind and load data to obtain the percentage of load covered through renewable energy. *4.1. General Power System Model*  In order to analyze how different operational strategies affect the power system, a simulation model has been developed. This model takes into account system restrictions and uses actual wind

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 17

To this aim, all machines of the hybrid wind–hydro powerplant have been modeled. So, the performance of each machine for a given power level can be obtained. Thus, knowing this power for each machine, the overall performance of the system can be calculated. and load data to obtain the percentage of load covered through renewable energy. To this aim, all machines of the hybrid wind–hydro powerplant have been modeled. So, the performance of each machine for a given power level can be obtained. Thus, knowing this power for

Each machine will operate at a power level determined by the load share assigned to each generation technology. As the preferred energy source is the wind farm, determining its power output is the first step. Therefore, considering the power available to the wind farm, the hydraulic storage, the diesel power plant and the consumers load, the contribution of each technology can be determined every ten minutes. Adding each of these contributions over a year, the performance of the hybrid wind–hydro power plant and the percentage of renewable energy used in this period can be determined. each machine, the overall performance of the system can be calculated. Each machine will operate at a power level determined by the load share assigned to each generation technology. As the preferred energy source is the wind farm, determining its power output is the first step. Therefore, considering the power available to the wind farm, the hydraulic storage, the diesel power plant and the consumers load, the contribution of each technology can be determined every ten minutes. Adding each of these contributions over a year, the performance of the hybrid wind–hydro power plant and the percentage of renewable energy used in this period can

When making the power dispatch, first wind farm-available power is compared with load demand. Then, hydraulic storage is checked. Finally, if the load cannot be supplied by wind or hydro, then diesel generators are assigned. As wind and load data are available every ten minutes, this process allows us to evaluate the power assigned to each technology and the variation in hydraulic stored energy every ten minutes. be determined. When making the power dispatch, first wind farm-available power is compared with load demand. Then, hydraulic storage is checked. Finally, if the load cannot be supplied by wind or hydro, then diesel generators are assigned. As wind and load data are available every ten minutes, this process allows us to evaluate the power assigned to each technology and the variation in hydraulic stored energy every ten minutes.

**Figure 4.** El Hierro power system daily generation and demand 06/08/2019. Source REE data. **Figure 4.** El Hierro power system daily generation and demand 06/08/2019. Source REE data.

#### Power Assigning Rules Power Assigning Rules

Every 10-min period, the power is assigned according to the following steps: Every 10-min period, the power is assigned according to the following steps:

and in case of missing power, turbine conditions will be evaluated.


3a) If wind power is larger than load demand, pumping conditions are evaluated:

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 7 of 17

3a) If wind power is larger than load demand, pumping conditions are evaluated:


If there is not enough water in the lower reservoir, pumping will not be carried out and the power supplied by the wind farm will be limited. On the other hand, if the wind farm power surplus is higher than maximum power of the pumps, wind power will be limited to the sum of consumer demand plus the maximum power of the pumps. pumps? If there is not enough water in the lower reservoir, pumping will not be carried out and the power supplied by the wind farm will be limited. On the other hand, if the wind farm power surplus is higher than maximum power of the pumps, wind power will be limited to the sum of consumer demand plus the maximum power of the pumps.

3b) If the wind power sent to the grid is not enough to cover consumers demand turbine conditions are evaluated. When the water level of the upper reservoir is not high enough, the turbines will not be activated. So, the diesel power plant will be activated to cover the demand. Additionally, if the wind power plus the turbines power were not enough to cover consumers demand, the diesel plant would also be activated to completely cover the demand. 3b) If the wind power sent to the grid is not enough to cover consumers demand turbine conditions are evaluated. When the water level of the upper reservoir is not high enough, the turbines will not be activated. So, the diesel power plant will be activated to cover the demand. Additionally, if the wind power plus the turbines power were not enough to cover consumers demand, the diesel plant would also be activated to completely cover the demand.

#### *4.2. Wind Farm Model 4.2. Wind Farm Model*

For modeling the wind farm in MS Excel, the manufacturer's curve that relates wind speed to the power generated by the wind turbine has been taken as a reference. For modeling the wind farm in MS Excel, the manufacturer's curve that relates wind speed to the power generated by the wind turbine has been taken as a reference.

Since the known parameter is the wind speed every 10 min, using the manufacturer's curve (shown in Figure 5), the power delivered by each wind turbine is calculated. To make these calculations simpler, a polinomic aproximation of this curve (for wind speeds between 0 and 16 m/s) is used. For wind speeds beteeen 16 and 25 m/s the result is rated power and for wind speeds greater than 25 m/s, zero power is output. These aproximations are programmed in Microsoft Visual Basic for Applications (VBA) Since the known parameter is the wind speed every 10 min, using the manufacturer's curve (shown in Figure 5), the power delivered by each wind turbine is calculated. To make these calculations simpler, a polinomic aproximation of this curve (for wind speeds between 0 and 16 m/s) is used. For wind speeds beteeen 16 and 25 m/s the result is rated power and for wind speeds greater than 25 m/s, zero power is output. These aproximations are programmed in Microsoft Visual Basic for Applications (VBA)

Assuming that all wind turbines in the park receive the same wind speed, the total power produced by the wind farm is obtained multiplying the power generated by one turbine by the number of wind turbines installed in the farm. Assuming that all wind turbines in the park receive the same wind speed, the total power produced by the wind farm is obtained multiplying the power generated by one turbine by the number of wind turbines installed in the farm.

**Figure 5.** Wind turbine Enercon E-70 power curve. **Figure 5.** Wind turbine Enercon E-70 power curve.

#### *4.3. Hydropower Plant Model*

can be obtained.

*4.3. Hydropower Plant Model*  A similar aproach has been used for making both the turbining and pumping model. The starting data are the power needed and the storage state. From them and the performance of each machine involved, the power generated/consumed by the pumped storage power plant every 10 min A similar aproach has been used for making both the turbining and pumping model. The starting data are the power needed and the storage state. From them and the performance of each machine involved, the power generated/consumed by the pumped storage power plant every 10 min can be obtained.

#### 4.3.1. Pelton Turbines Model 4.3.1. Pelton Turbines Model For turbining, the performance of a generic Pelton turbine with the same maximum efficiency of 4.3.1. Pelton Turbines Model For turbining, the performance of a generic Pelton turbine with the same maximum efficiency of

For turbining, the performance of a generic Pelton turbine with the same maximum efficiency of the turbines installed in the plant is used (shown in Figure 6). The number of turbines in operation is determined using the criterion of maximum combined performance. Using this criterion and applying the previously mentioned performance curve, a curve for the plant—shown in Figure 7—is obtained. the turbines installed in the plant is used (shown in Figure 6). The number of turbines in operation is determined using the criterion of maximum combined performance. Using this criterion and applying the previously mentioned performance curve, a curve for the plant—shown in Figure 7—is obtained. the turbines installed in the plant is used (shown in Figure 6). The number of turbines in operation is determined using the criterion of maximum combined performance. Using this criterion and applying the previously mentioned performance curve, a curve for the plant—shown in Figure 7—is obtained.

**Figure 6.** Pelton turbine efficiency curve. **Figure 6.** Pelton turbine efficiency curve.

**Figure 7.** Power plant turbines efficiency curve. **Figure 7.** Power plant turbines efficiency curve. **Figure 7.** Power plant turbines efficiency curve.

#### 4.3.2. Electric Generators Model 4.3.2. Electric Generators Model 4.3.2. Electric Generators Model

For modeling the generators, generic performance Equations (3)–(5) have been used. The parameters of this Equation are taken from the generators installed in the Gorona plant datasheet (shown in Table A2). For modeling the generators, generic performance Equations (3)–(5) have been used. The parameters of this Equation are taken from the generators installed in the Gorona plant datasheet (shown in Table A2). For modeling the generators, generic performance Equations (3)–(5) have been used. The parameters of this Equation are taken from the generators installed in the Gorona plant datasheet (shown in Table A2).

$$P\_{Out} = P\_{In} - \left[P\_{Fix} + P\_{Var} \left(\frac{P\_{Out}}{FdP \ S\_N}\right)^2\right] \tag{3}$$

$$P\_{\rm Fix} = P\_{\rm Fe} + P\_{f/w} \tag{4}$$

$$P\_{Var} = P\_{\text{Cu1}} + P\_{\text{Cu2}} + P\_{\text{Add}} \tag{5}$$

#### where:


Since each generator is in the same shaft of its turbine, the generators' activating sequence is the same one as of the turbines. Therefore, the overall performance of the generators depend on the turbines operating point.

#### 4.3.3. Main Transformers Model

The last component of the turbining process is the main transformer that connects each generator to the bars of the electrical substation. To calculate the performance of this machine, the well-known Equation (6) is used. Its parameters are in Table A3.

$$P\_{Out} = P\_{In} - \left[P\_0 + P\_{Sc} \left(\frac{P\_{Out}}{FdP \ S\_N}\right)^2\right] \tag{6}$$

where:


The activation sequence of the transformers is the same as the turbines and generators.

#### 4.3.4. Power Plant Model

Finally, the overall performance of turbining, depending on the power required in the substation bars, can be calculated by applying sequentially the performance of all machines involved.

With this overall performance, it is also possible to calculate the water volume from the upper reservoir turbined and transferred to the lower one. In this way, the evolution of the hydraulic storage for the next 10 min can be calculated.

Therefore, in order to implement the whole model of the power plant, a relation between the water volume turbined and the power demanded is applied.

#### *4.4. Pumping Station*

#### 4.4.1. Pumps Model

In the Gorona station, there are two types of pumps: fixed speed 500 kW-rated power pumps and variable speed 1500 kW-rated power pumps.

For fixed pumps, as they always work at rated speed and power, their performance is the specified in the manufacturer's data sheet. While for variable speed pumps, taking the operating points from the diagrams provided by the manufacturer, their performance, for each operating point, could be

obtained. In this way, taking values from the pump manufacturer's curve, a performance curve for the whole pump set can be calculated (Figure 8). could be obtained. In this way, taking values from the pump manufacturer's curve, a performance curve for the whole pump set can be calculated (Figure 8). could be obtained. In this way, taking values from the pump manufacturer's curve, a performance curve for the whole pump set can be calculated (Figure 8).

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 10 of 17

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 10 of 17

**Figure 8.** Pumps efficiency curves. (**a**) Fix speed pumps (**b**) Variable speed pumps. **Figure 8.** Pumps efficiency curves. (**a**) Fix speed pumps (**b**) Variable speed pumps.**Figure 8.** Pumps efficiency curves. (**a**) Fix speed pumps (**b**) Variable speed pumps.

As in the case of turbining, pump operation is only possible if there is enough water stored in the lower reservoir. If this is the case, the activation sequence of the pumps has to take into account that fixed pumps can only be off (zero power consumption) or on (rated power consumption), while variable speed ones can consume any level of power between a minimum and a maximum. As in the case of turbining, pump operation is only possible if there is enough water stored in the lower reservoir. If this is the case, the activation sequence of the pumps has to take into account that fixed pumps can only be off (zero power consumption) or on (rated power consumption), while variable speed ones can consume any level of power between a minimum and a maximum. As in the case of turbining, pump operation is only possible if there is enough water stored in the lower reservoir. If this is the case, the activation sequence of the pumps has to take into account that fixed pumps can only be off (zero power consumption) or on (rated power consumption), while variable speed ones can consume any level of power between a minimum and a maximum.

Therefore, if there are enough water for pump operation, the sequence of activation will start with a variable speed pump up to its rated power is reached. Then, a fixed speed pump is started while the variable one reduces its consumption. This process continues until all pumps are in operation at rated power, or the required power consumption is reached. Figure 9 shows the efficiency evolution of the pump station depending on the power required for frequency regulation Therefore, if there are enough water for pump operation, the sequence of activation will start with a variable speed pump up to its rated power is reached. Then, a fixed speed pump is started while the variable one reduces its consumption. This process continues until all pumps are in operation at rated power, or the required power consumption is reached. Figure 9 shows the efficiency evolution of the pump station depending on the power required for frequency regulation [24]. Therefore, if there are enough water for pump operation, the sequence of activation will start with a variable speed pump up to its rated power is reached. Then, a fixed speed pump is started while the variable one reduces its consumption. This process continues until all pumps are in operation at rated power, or the required power consumption is reached. Figure 9 shows the efficiency evolution of the pump station depending on the power required for frequency regulation

**Figure 9.** Pumping station efficiency curve. **Figure 9.** Pumping station efficiency curve. **Figure 9.** Pumping station efficiency curve.

#### 4.4.2. Electric Motors Model 4.4.2. Electric Motors Model

motors will work at constant efficiency.

[24].

[24].

4.4.2. Electric Motors Model 600 kW motors (Table A7) drive the fixed speed pumps and 1600 kW motors (Table A4) drive the variable speed ones. Both types of motors have a known performance for their rated power according to the manufacturer's specifications. As fixed speed motors will work at constant load, they will exhibit constant efficiency. Regarding variable speed motors, as they are driven by a converter (Table A5) that varies frequency and voltage they will work at constant efficiency too. Therefore, all 600 kW motors (Table A7) drive the fixed speed pumps and 1600 kW motors (Table A4) drive the variable speed ones. Both types of motors have a known performance for their rated power according to the manufacturer's specifications. As fixed speed motors will work at constant load, they will exhibit constant efficiency. Regarding variable speed motors, as they are driven by a converter (Table A5) that varies frequency and voltage they will work at constant efficiency too. Therefore, all motors will work at constant efficiency. 600 kW motors (Table A7) drive the fixed speed pumps and 1600 kW motors (Table A4) drive the variable speed ones. Both types of motors have a known performance for their rated power according to the manufacturer's specifications. As fixed speed motors will work at constant load, they will exhibit constant efficiency. Regarding variable speed motors, as they are driven by a converter (Table A5) that varies frequency and voltage they will work at constant efficiency too. Therefore, all motors will work at constant efficiency.

As was the case with turbines and generators, each motor is in the same shaft as its pump. Therefore, they have the same activation sequence as that of the pumps.

#### 4.4.3. Main Transformer Model

The pump station transformer is modelled as the hydropower transformers. The transformer data are in Table A8.

#### 4.4.4. Pumping Station Model

As in the turbining process, pump power implies that some water volume will be pumped from the lower reservoir to the upper reservoir for each 10 min period.

The performance of the different pumping machines is applied along the range of operating power to obtain the overall performance of the entire process. In the same way as in turbining, there is a curve that relates pumping power with hydraulic storage variation.

#### *4.5. Diesel Power Plant Model*

Due to the aim of this study, the operation of the hybrid wind–hydro power plant, the diesel power plant is taken as a generation source that helps in covering the electrical demand. This is so because the performance of the diesel power plant as a whole is not relevant to evaluate the performance of the hybrid wind–hydro power plant. Only the energy values of the wind farm and the hydroelectric power plant in a specified time period is analyzed. The energy delivered by the diesel power plant is only used to provide the power needed to cover the demand.

#### **5. Results and Discussion**

Using the aforementioned simulation models, the performance of the wind–hydro El Hierro power plant has been simulated. To this aim, wind data corresponding to years 2016 and 2017 have been used. As wind data is available every 10 min, and a year has 6 × 24 × 365 = 52,560 10-min periods, simulating an entire year requires running the models 52,560 times. In Figures 10 and 11, the percentage of load demand covered by renewable energy in 2016 and 2017, respectively, is shown. This percentage increases when there is no limit on the power that the wind farm can generate, and also increases when wind penetration is not limited. In both years, the maximum possible renewable load coverage is slightly over 70%.

In those years, the percentage of load demand actually covered by renewable energy has increased. In 2016, it was 40.62%, while it was 46.26% in 2017. Figure 3 shows that in 2016, wind penetration was limited to around 70% and the maximum wind power was around 7.2 MW. Applying this limit to Figure 10 the maximum percentage of demand covered by renewable energy is around 50%, which is in reasonable agreement with the actual value. It should be noted that some factors were not considered in the calculations as penstock losses or evaporation in the upper reservoir. In 2018, this percentage increases again up to 56.62%. Therefore, the power system operation is increasingly approaching theoretical performance.

**6. Conclusions** 

**6. Conclusions** 

As was the case with turbines and generators, each motor is in the same shaft as its pump.

The pump station transformer is modelled as the hydropower transformers. The transformer

As in the turbining process, pump power implies that some water volume will be pumped from

The performance of the different pumping machines is applied along the range of operating power to obtain the overall performance of the entire process. In the same way as in turbining, there

Due to the aim of this study, the operation of the hybrid wind–hydro power plant, the diesel power plant is taken as a generation source that helps in covering the electrical demand. This is so because the performance of the diesel power plant as a whole is not relevant to evaluate the performance of the hybrid wind–hydro power plant. Only the energy values of the wind farm and the hydroelectric power plant in a specified time period is analyzed. The energy delivered by the

Using the aforementioned simulation models, the performance of the wind–hydro El Hierro power plant has been simulated. To this aim, wind data corresponding to years 2016 and 2017 have been used. As wind data is available every 10 min, and a year has 6 × 24 × 365 = 52,560 10-min periods, simulating an entire year requires running the models 52,560 times. In Figures 10 and 11, the percentage of load demand covered by renewable energy in 2016 and 2017, respectively, is shown. This percentage increases when there is no limit on the power that the wind farm can generate, and

Therefore, they have the same activation sequence as that of the pumps.

the lower reservoir to the upper reservoir for each 10 min period.

is a curve that relates pumping power with hydraulic storage variation.

diesel power plant is only used to provide the power needed to cover the demand.

4.4.3. Main Transformer Model

4.4.4. Pumping Station Model

*4.5. Diesel Power Plant Model* 

**5. Results and Discussion** 

load coverage is slightly over 70%.

data are in Table A8.

**Figure 10.** Annual demand covered by renewable energy for 2016. **Figure 10.** Annual demand covered by renewable energy for 2016. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 12 of 17

**Figure 11.** Annual demand covered by renewable energy for 2017. **Figure 11.** Annual demand covered by renewable energy for 2017. factors were not considered in the calculations as penstock losses or evaporation in the upper reservoir. In 2018, this percentage increases again up to 56.62%. Therefore, the power system

In those years, the percentage of load demand actually covered by renewable energy has increased. In 2016, it was 40.62%, while it was 46.26% in 2017. Figure 3 shows that in 2016, wind penetration was limited to around 70% and the maximum wind power was around 7.2 MW. Applying this limit to Figure 10 the maximum percentage of demand covered by renewable energy is around 50%, which is in reasonable agreement with the actual value. It should be noted that some factors were not considered in the calculations as penstock losses or evaporation in the upper Regarding the operating efficiency, its value for 2017 wind data is shown in Figure 12. As it was the case in load demand coverage, the efficiency increases when the wind power limit increases and, also, when wind penetration increases. Maximum operating efficiency is achieved when none of these limits apply. In that case, operating efficiency reaches 80%. In Section 3.2, a 40% operating efficiency using the isolated wind strategy was estimated. This means that an unrestricted wind operation achieves a twofold efficiency increase over an isolated wind operation. operation is increasingly approaching theoretical performance. Regarding the operating efficiency, its value for 2017 wind data is shown in Figure 12. As it was the case in load demand coverage, the efficiency increases when the wind power limit increases and, also, when wind penetration increases. Maximum operating efficiency is achieved when none of these limits apply. In that case, operating efficiency reaches 80%. In Section 3.2, a 40% operating efficiency using the isolated wind strategy was estimated. This means that an unrestricted wind operation achieves a twofold efficiency increase over an isolated wind operation.

reservoir. In 2018, this percentage increases again up to 56.62%. Therefore, the power system

**Figure 12.** Operational efficiency for 2017. **Figure 12.** Operational efficiency for 2017.

As shown in Figure 4, nowadays this power system is close to achieving this operating efficiency. As shown in Figure 4, nowadays this power system is close to achieving this operating efficiency.

This paper presents a model of an isolated power system located in El Hierro Island—a UNESCO biosphere reserve. This model is based on the analysis of energy flows with the aim of evaluating the

**Figure 12.** Operational efficiency for 2017.

As shown in Figure 4, nowadays this power system is close to achieving this operating efficiency.

This paper presents a model of an isolated power system located in El Hierro Island—a UNESCO biosphere reserve. This model is based on the analysis of energy flows with the aim of evaluating the

#### **6. Conclusions**

This paper presents a model of an isolated power system located in El Hierro Island—a UNESCO biosphere reserve. This model is based on the analysis of energy flows with the aim of evaluating the operating efficiency of a renewable power plant. This power plant is composed of a wind farm and a pumped storage hydropower plant. For every component of the power system, a model that takes into account its energy performance has been presented.

In addition, a definition of operating efficiency adapted to this renewable power plant has been proposed. Operating efficiency is defined through the comparison of the net energy put on the power system by the hybrid power plant with the electrical energy that its wind farm will put in an infinite bus. Therefore, if the wind farm was connected to an infinite bus, its operating efficiency would be 100%. In a small isolated power system, such as El Hierro Island, a storage system is needed and, so, the operating efficiency is always less than 100%.

In order to analyze this operating efficiency, several power system operational strategies have been presented. Several tens of thousands of simulations have been conducted to evaluate the yearly operating efficiency of the renewable power plant depending on the strategy followed. In addition, the percentage of demand covered by the renewable power plant has been also evaluated.

The results of the simulations show that the operating efficiency of the wind–hydro power plant is heavily dependent on the power system operational strategy followed. This efficiency increases when lighter restrictions are imposed on the wind generation. In fact, a twofold increase in efficiency can be achieved between the best (80%) and worst (40%) strategies.

Consequently, the non-restricted wind operational strategy is recommended. Along the operational history of the plant, several of these strategies were used. As the operator's confidence increased, the wind generation restrictions have been reduced. Nowadays, non-restricted wind generation is mostly applied. In fact, there are some periods when the energy supply is drawn 100% from renewable sources.

The percentage of demand covered by this power plant along its operational history has been compared with simulation results showing a reasonable agreement. In addition, as wind generation restrictions have been reduced, the percentage of demand covered by this power plant has increased. The yearly demand coverage by renewable energy was 40.6% in 2016. It has increased to 46.26% in 2017 and to 56.6% in 2018. Nowadays, there are some periods of 100% demand covered by this renewable power plant.

**Author Contributions:** Conceptualization, C.A.P. and J.A.S.-F.; methodology, F.B.; software, F.B.; validation, C.A.P., J.A.S.-F. and C.N.; formal analysis, F.B.; investigation, C.A.P. and J.A.S.-F.; resources, C.N.; data curation, F.B.; writing—original draft preparation, C.A.P. and J.A.S.-F.; writing—review and editing, C.A.P., J.A.S.-F. and C.N.; visualization, F.B.; supervision, C.N. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Universidad Politécnica de Madrid, grant number RP1604330010.

**Conflicts of Interest:** The authors declare no conflict of interest

**Appendix A**




**Table A2.** Hydro generator performance data.

**Table A3.** Hydro generator transformers performance data.


**Table A4.** Variable speed drive motors performance data.


**Table A5.** Variable speed driver performance data.


**Table A6.** Variable speed pumps performance data.



**Table A7.** Fixed speed motors performance data.

**Table A8.** Pumping station main transformer performance data.


**Table A9.** Fix speed pumps performance data.


#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Feasibility and Cost Analysis of Photovoltaic-Biomass Hybrid Energy System in O**ff**-Grid Areas of Bangladesh**

#### **Nusrat Chowdhury <sup>1</sup> , Chowdhury Akram Hossain <sup>2</sup> , Michela Longo <sup>3</sup> and Wahiba Yaïci 4,\***


Received: 27 January 2020; Accepted: 17 February 2020; Published: 19 February 2020

**Abstract:** In this progressing technological advancement world, hybrid systems for power generation is one of the most promising fields for any researcher. In this context, photovoltaic-biomass hybrid systems with off-grid applications have become extremely popular with both Governments and individual users in rural areas of any part of the world. This system has gained popularity because of low cost, sustainability and very effective outcome with the use of natural resources at the rural areas. In this paper a proposed hybrid system which contains photovoltaics (PV) and biomass along with an additional storage has been considered to find the different aspects from an end user point of view. It also discusses the feasibility of the proposed model for an off-grid power system located in the remote areas of Ashuganj, Bangladesh. In order to analyse the pollutant emissions and calculate the cost parameters of the proposed system, RETScreen simulation software was deployed. This research also carries out a brief financial analysis considering the annual income of the end user and the payback periods for the installed system. It endeavours to provide complete information about different parameters which also includes the environmental impacts involved in establishing the proposed system. The conventional system in the pilot area is a kerosene-based system, hence in this research, a comparison between the proposed and the conventional system has been analysed using simulated results. The simple payback of the project was estimated to be 6.9 years and this model will be able to reduce the CO<sup>2</sup> emissions by approximately 3.81 tonnes per year. The results have significantly supported the proposed system to be more reliable, environmentally-friendly and less costly than the conventional kerosene-based system.

**Keywords:** photovoltaics (PV); biomass; battery storage; off-grid electrification; feasibility analysis; cost analysis; simple payback period; CO<sup>2</sup> emissions; renewable energy

### **1. Introduction**

Electricity is the major source of energy in most urban systems worldwide. Since the first high-voltage Alternating Current (AC) coal power station was commissioned in London in 1890, electrification of residential and industrial installations has grown exponentially expanding toe 83% of all urban areas by 2010 [1]. Even with a projection of increased electricity use for future energy systems the electrification of rural areas still represents a relevant issue. Such expanded use will likely seriously affect different sectors of developing economies, ranging from industrial to transportation uses [2]. In particular, the augmenting of current electricity distribution grids in rural areas, which

are located far from the main national grid, may result in excessive costs in terms of installation, transmission, distribution, and maintenance [3,4]. Indeed, the electrification of rural areas through the extension of grid connections may raise the overall generation costs, which could reach up to seven times the normal price obtainable in urban areas [4]. Some studies have been focussed on operational performance of the Bangladesh rural electrification program and its determinants with a focus on political interference [5–8]. Per [5] for instance, complete electrification through Bangladesh will take many years, and thus the diminishing returns to scale of incremental investment for further rural electrification will be faced in the long run. The authors suggested that both Bangladesh and international donors revisit the original principle of the Rural Electrification Program, eradicate political hindrances from the program, and make sustained efforts to develop more efficient infrastructure of delivering electricity to the rural poor and encouraging local economic development.

Another issue of concern is the unreliability in frequency, blackouts, power losses, and fluctuations of the grid voltage. In such cases, the proper use of Renewable Energy Sources (RESs) in remote areas could represent a viable and economical alternative to the extension of electricity grids [9–11].

Authoritative studies have shown that hybrid stand-alone electricity-generating systems are more economically feasible for off-grid consumers located in distant areas [12,13]. In addition, RES installations can also reduce the amount of CO<sup>2</sup> emissions emanating from electrical energy generation. Studies in Bangladesh have demonstrated, in fact, that 1kWh of electricity generated by solar photovoltaic (PV) systems can reduce the amount of CO<sup>2</sup> emissions by approximately 660 tonnes per year [14]. However, the use of stand-alone PV systems in off-grid applications also presents certain drawbacks, mainly related to the intrinsic intermittency and stochasticity of the solar source. To overcome these drawbacks, the use of electrical Energy Storage System (ESS) solutions is usually implemented [15,16], along with the adoption of hybrid configurations, i.e., by adding traditional controllable power generators (such as diesel-electric motors), to support the intermittency and unreliability of PV systems.

Currently, in off-grid PV systems, the use of large ESS solutions is usually considered economically unviable due to the high investment costs, that is, high costs of storage devices [17,18], whereas the combined adoption of PV with ESS and diesel-electric generators has been widely adopted [19]. Nevertheless, increasing concern about global warming and environmental pollution is propelling the replacement of generators that were traditionally powered by fossil fuels, while there is more interest in adopting greener solutions. To this end, the use of biomass generators could indeed represent an interesting alternative, due both to its low carbon impact and its lower investment and developmental costs.

As indicated in the literature, numerous feasibility and techno-economic studies were performed on micro-grid projects in different countries, specifically on stand-alone hybrid energy systems for applications in remote areas [20–45]. The inference drawn is that it appears that no feasibility study or other similar work has been conducted on such a system in the remote communities of Bangladesh; thus, the present study is an original research initiative and a firm contribution to knowledge.

Considering the environmental and cost concerns described earlier, the purpose of this paper is a feasibility study of the potentials and likely impact of a hybrid PV-biomass system as a possible option for the provision of power in a rural area of Bangladesh. The primary objective of the study is to present a preliminary design of a hybrid PV-biomass system that is capable of satisfying the energy needs of the selected off-grid application, and to compare its economic performance in respect of existing solutions, by making use of discounted cash flow and payback period analyses. The paper examines the technical, economic and environmental feasibility, both the integration and sizing of a hybrid PV-biomass system, and the energy storage of microgrid for remote electrification within Ashuganaj, Bangladesh. In addition, the impact of related CO<sup>2</sup> emissions is also analysed, and compared to traditional solutions. The study is essentially performed employing the simulation tool referred to as RETScreen [46].

The paper is organised as follows: Section 2 gives a full description of the proposed project location as well as the simulation methods; also provides the load profiles and climate data related to the project location and Section 3 presents details on the proposed PV-biomass hybrid system. Simulation results are discussed in Section 4, including the sizing of the system, the cost and emissions analyses, while conclusions are drawn in Section 5. The paper is organised as follows: Section 2 gives a full description of the proposed project location as well as the simulation methods; also provides the load profiles and climate data related to the project location and Section 3 presents details on the proposed PV-biomass hybrid system. Simulation results are discussed in Section 4, including the sizing of the system, the cost and emissions analyses, while conclusions are drawn in Section 5.

#### **2. Project Location and Simulation Methods 2. Project Location and Simulation Methods**

#### *2.1. Simulation Method 2.1. Simulation Method*

Many studies have analysed renewable energy systems using RETScreen [47–51]. RETScreen software was employed in this design to achieve the least energy costs that people will find affordable [52]. The RET Screen Clean Energy Management Software (usually shortened to RET Screen) is a clean energy software package developed by Ministry of Natural Resources Canada (Government of Canada) for evaluating both financial and environmental costs and benefits of different renewable energy technologies for any location in the world. This software uses visual basic and C language as working platform. RETScreen PV model also covers off-grid PV applications and include stand-alone, hybrid and water pumping systems also. It has a global climate data database of more than 6000 ground stations (month wise solar irradiation and temperature data for the year), energy resource maps (i.e., wind maps), hydrology data, product data like solar photovoltaic panel details and wind turbine power curves. It also provides link to NASA climate database. It enables comprehensive identification, assessment and optimization of the technical and financial viability of potential renewable energy and energy-efficient projects. It also allows measurement and verification of the actual performance of facilities and identification of energy savings/production opportunities. The software tool can determine the technical and financial viability of renewable energy, energy efficiency and cogeneration projects. The area number of worksheets for performing detailed project analysis includes Energy Modelling, Cost Analysis, Emissions Analysis, Financial Analysis and Sensitivity and Risk Analyses sheets. The analysis of different types of energy efficient and renewable technologies (RETS) covers mainly energy production, life-cycle costs and greenhouse gas emission reduction. In summary, RETScreen Plus is a Windows-based energy management software tool to study the energy performance [51,52]. Many studies have analysed renewable energy systems using RETScreen [47–51]. RETScreen software was employed in this design to achieve the least energy costs that people will find affordable [52]. The RET Screen Clean Energy Management Software (usually shortened to RET Screen) is a clean energy software package developed by Ministry of Natural Resources Canada (Government of Canada) for evaluating both financial and environmental costs and benefits of different renewable energy technologies for any location in the world. This software uses visual basic and C language as working platform. RETScreen PV model also covers off-grid PV applications and include stand-alone, hybrid and water pumping systems also. It has a global climate data database of more than 6000 ground stations (month wise solar irradiation and temperature data for the year), energy resource maps (i.e., wind maps), hydrology data, product data like solar photovoltaic panel details and wind turbine power curves. It also provides link to NASA climate database. It enables comprehensive identification, assessment and optimization of the technical and financial viability of potential renewable energy and energy-efficient projects. It also allows measurement and verification of the actual performance of facilities and identification of energy savings/production opportunities. The software tool can determine the technical and financial viability of renewable energy, energy efficiency and cogeneration projects. The area number of worksheets for performing detailed project analysis includes Energy Modelling, Cost Analysis, Emissions Analysis, Financial Analysis and Sensitivity and Risk Analyses sheets. The analysis of different types of energy efficient and renewable technologies (RETS) covers mainly energy production, life-cycle costs and greenhouse gas emission reduction. In summary, RETScreen Plus is a Windows-based energy management software tool to study the energy performance [51,52]. This simulation process empowers experts and decision-makers to identify and evaluate the

This simulation process empowers experts and decision-makers to identify and evaluate the technical and financial viability of potential clean energy projects and also to measure and verify the actual and ongoing energy performance of energy-efficiency projects. The process also enables the evaluation of energy production, life-cycle costs and greenhouse gas emission reductions for the proposed Hybrid system [46]. The flowchart used in the simulation tool is shown in Figure 1. technical and financial viability of potential clean energy projects and also to measure and verify the actual and ongoing energy performance of energy-efficiency projects. The process also enables the evaluation of energy production, life-cycle costs and greenhouse gas emission reductions for the proposed Hybrid system [46]. The flowchart used in the simulation tool is shown in Figure 1.

**Figure 1.** RETScreen technical evaluation structure flowchart [53]. **Figure 1.** RETScreen technical evaluation structure flowchart [53].

Ashuganj city located in the Brahmanbaria District of Chittagong Division in Bangladesh, was

Ashuganj city located in the Brahmanbaria District of Chittagong Division in Bangladesh, was selected as the reference site for the application and evaluation of the proposed hybrid PV-biomass system. The city was selected because of its residents' pervasive dependence on fossil fuels to meet their energy needs, due to the absence of a power grid extension. Residents situated close to the city depend mainly on horticulture and animal husbandry for their livelihood. Therefore, in rural areas, it is economical to use hybrid systems comprising solar and biomass, provided the biomass supply is consistently available. In most of the areas, manure, crop wastes and cooking wastes are accessible free of charge.

### *2.2. Load Profile and Climate Data of Project Location*

The load demand of the proposed system is estimated by reference to a household load (4 LED light, 2 DC fan and 1 DC TV) for a typical lower middle-class family. The study evaluated fifty houses for provision of electricity for the proposed hybrid system (Table 1). The actual daily average energy demand for fifty houses is calculated as 45.6 kWh.



The load characteristics are calculated on the basis of three seasons: summer (March–June), spring (July–October) and winter (November–February). The RETScreen software calculates the monthly and yearly energy consumptions for the load based on the following indices: appliance used, season, weather condition based on location, number of people per house and number of total houses. Considering the system loss, the daily average energy demand is estimated as 47.2 kWh which is higher than the actual demand and annual energy demand is 14.161 MWh for both the base case and the proposed case. The detailed data of monthly and yearly energy consumption is shown in Table 2. In winter season the load demand is comparatively lower than the other two seasons. Therefore, it is considered 54%. In spring season considering the weather condition, the percentage of energy use is 93%. Due to the extreme hot weather in summer the percentage of energy is considered 99%.

As mentioned earlier, this paper discusses the feasibility study of the proposed hybrid system with RETScreen. The financial analysis has been carried out for systematic progression towards a PV-Biomass-based hybrid system. We need to determine the latitude and longitude of the study area and analysis is done to derive the climate data. The RETScreen software gives complete weather details based on climate location. Table 3 indicates the climate data location and the project location which has been collected from NASA by RETScreen. The daily solar radiation, air temperature, humidity and earth temperature data are also collected for the reference location to check the feasibility of solar project implementation. The detailed data are shown in Table 4.


**Table 2.** Load characteristics of base and proposed cases.

**Table 3.** Ashuganj site reference data.


**Table 4.** Site reference conditions.


#### **3. The Proposed PV-Biomass Hybrid System**

The proposed system consists of electric DC loads, solar PV, biomass generator, battery, and converter. Figure 2 demonstrates the block diagram of the proposed system. The system is fed by PV arrays and a biomass generator. There is no grid connection between the systems. Biomass is

**3. The Proposed PV-Biomass Hybrid System**

an abundant source of energy around the world, which is composed of organic matter including agricultural residues, and wood, animal and human wastes. Use of biomass for the purpose of power generation has become very popular, especially since it is an easily obtainable source of energy in the rural parts of Bangladesh. Additionally, it is a cleaner source of energy than fossils throughout the world. Its relative abundance makes it a viable option for use as a potential source of energy for electricity-generation in the country where it comprises animal manure that can either be converted through the absorption process or its residues extracted through the combustion process. In this system hybrid solar and biomass system was chosen as biomass is accessible effectively in the form of manure throughout the year. This hybrid system using biomass and solar with a battery as a storage system for electricity generation is more economical, because it can generate electricity during cloudy days also. agricultural residues, and wood, animal and human wastes. Use of biomass for the purpose of power generation has become very popular, especially since it is an easily obtainable source of energy in the rural parts of Bangladesh. Additionally, it is a cleaner source of energy than fossils throughout the world. Its relative abundance makes it a viable option for use as a potential source of energy for electricity-generation in the country where it comprises animal manure that can either be converted through the absorption process or its residues extracted through the combustion process. In this system hybrid solar and biomass system was chosen as biomass is accessible effectively in the form of manure throughout the year. This hybrid system using biomass and solar with a battery as a storage system for electricity generation is more economical, because it can generate electricity during cloudy days also.

abundant source of energy around the world, which is composed of organic matter including

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 16

The proposed system consists of electric DC loads, solar PV, biomass generator, battery, and

**Figure 2.** Block diagram of proposed photovoltaics (PV)-biomass hybrid energy system. **Figure 2.** Block diagram of proposed photovoltaics (PV)-biomass hybrid energy system.

In the proposed system, the AC generator is used. Here, the output of the generator is converted into 24 V DC using the AC to DC converter and then connected to 24 V DC bus bar. Finally, DC to DC converter is used in every house to convert 24 V DC to 12 V DC to operate the DC house load. In the proposed system, the AC generator is used. Here, the output of the generator is converted into 24 V DC using the AC to DC converter and then connected to 24 V DC bus bar. Finally, DC to DC converter is used in every house to convert 24 V DC to 12 V DC to operate the DC house load.

The total capital cost of 1 kW biogas fuel-based generator considered as BDT (Bangladeshi Taka) 60,000/kW (USD \$714.29/kW) and the lifetime of the generator is specified in hours of operation. The lifetime of the generator is considered as 15,000 hr. The efficiency of the generator is considered as 80%. We estimate the cost per tonne of biogas at BDT 70 (USD \$0.833) based on biomass resources being obtainable almost free of charge. The total capital cost of 1 kW biogas fuel-based generator considered as BDT (Bangladeshi Taka) 60,000/kW (USD \$714.29/kW) and the lifetime of the generator is specified in hours of operation. The lifetime of the generator is considered as 15,000 hr. The efficiency of the generator is considered as 80%. We estimate the cost per tonne of biogas at BDT 70 (USD \$0.833) based on biomass resources being obtainable almost free of charge.

#### **4. Results and Discussion 4. Results and Discussion**

RETScreen software has been used to analyse the different parameters of the proposed case and finally, the proposed case was compared with the base case. After the comparison, based on the financial viability, annual savings and evaluation of GHG emissions it can be easily deduced that the proposed case is more beneficial than the existing one. Analysis types and methods selected in RETScreen are mini-grid and method two respectively. RETScreen software has been used to analyse the different parameters of the proposed case and finally, the proposed case was compared with the base case. After the comparison, based on the financial viability, annual savings and evaluation of GHG emissions it can be easily deduced that the proposed case is more beneficial than the existing one. Analysis types and methods selected in RETScreen are mini-grid and method two respectively.

#### *4.1. Base Case Power Study 4.1. Base Case Power Study*

Having chosen an off-grid area, the kerosene lamp was evaluated as a power source in the base case. Estimated total power capacity for base case is 4.70 kW. We estimated the cost of a litre of Having chosen an off-grid area, the kerosene lamp was evaluated as a power source in the base case. Estimated total power capacity for base case is 4.70 kW. We estimated the cost of a litre of kerosene to be BDT 65 (USD \$0.77) and so the total cost of electricity is calculated as BDT 474,702 (USD \$5651.21) for the existing kerosene-based system by RETScreen. Table 5 indicates the unit cost and total electricity cost for the base case.


**Table 5.** Base case power system.

From the simulation, it was evident that the unit cost of electricity for the existing system is very high and also harmful to the environment. A new system has been proposed for this reason.

#### *4.2. Proposed Case Power Study*

*4.3. Cost Analysis*

the project.

In the proposed case analysis, mono-crystalline silicon PV solar cell with a power capacity of 12.9 kWp, and an efficiency of 13.1% was used. In this model maximum power point tracker is used as a control method and miscellaneous losses are considered as 5%. To fulfil the peak time energy demand biomass generator is used where the biomass rate is BDT 70/tonne (USD \$0.83/tonne) and capacity of the generator is considered as 1 kW. As a storage system and an emergency backup, battery has been used. In this system, a total of 9 batteries are considered where each battery has a capacity of 24 V, 200 Ah. A one-day autonomy has been estimated for reliable power supply. The total capacity of the battery bank of 1800 Ah and 43 kWh is considered. Figure 3 presents the battery, PV and biomass generator specification that is given as an input in RETScreen and also shows that for the given combination approximately 81.2% of total energy comes from PV while the remainder of the energy comes from biomass generator. Thus, the total annual energy delivered to the load from PV and biomass generator are 13.98 MWh and 3.2 MWh respectively. Therefore, from the proposed system yearly 17.185 MWh energy can be produced which can easily fulfil the required load demand of 14.161 MWh. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 16

**Figure 3.** The specification of different components and energy produced by each source. Notation— Mono-Si: Monocrystalline silicon; CS4A: Multi-contact connector type 4. **Figure 3.** The specification of different components and energy produced by each source. Notation—Mono-Si: Monocrystalline silicon; CS4A: Multi-contact connector type 4.

implementing the project in Ashuganj it is necessary to test the feasibility of the project to determine its suitability for the selected area. To check the climate feasibility, some testing is necessary which comes at a cost. For a "Feasibility analysis," more detailed and more accurate information is usually required. The calculations performed by the RETScreen Software for this step are straightforward and relatively simple (addition and multiplication). Figure 4 shows the detailed cost calculation for

In the proposed system, since most of the energy comes from PV, it produces clean energy and reduces CO<sup>2</sup> emissions compared to the existing system.

## *4.3. Cost Analysis*

In order to analyse the costs of the proposed system, the initial, annual, and periodic costs as well as credits for any base case costs that are avoided in the proposed system are analysed. Before implementing the project in Ashuganj it is necessary to test the feasibility of the project to determine its suitability for the selected area. To check the climate feasibility, some testing is necessary which comes at a cost. For a "Feasibility analysis," more detailed and more accurate information is usually required. The calculations performed by the RETScreen Software for this step are straightforward and relatively simple (addition and multiplication). Figure 4 shows the detailed cost calculation for the project. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 9 of 16


**Figure 4.** Initial cost of proposed system. Notation—BDT: Bangladeshi Taka. **Figure 4.** Initial cost of proposed system. Notation—BDT: Bangladeshi Taka.

In cost analysis, it has been observed that the total initial cost is BDT 2,190,089 (USD \$26,072.49) where 89.1% cost comes from power system sources such as PV, battery, biomass generator while the remaining cost components are from feasibility study and system miscellaneous. In the proposed system the lifetime of the PV, battery and converter have been based on 25 years, 5 years and 10 years respectively. Therefore, at full project life in 25 years the battery will require replacement 4 times, while the converter will be replaced twice. For the proposed system the total annual cost is BDT 167,696 (USD \$1996.38). In periodic cost, it is seen that the battery replacement cost in full project life is BDT 403,200 (USD \$4800), and converter cost is BDT 89,600 (USD \$1066.67). After all the expenses the annual savings from the project is BDT 474,702 (USD \$5,651.21) which is provided in Table 6. In cost analysis, it has been observed that the total initial cost is BDT 2,190,089 (USD \$26,072.49) where 89.1% cost comes from power system sources such as PV, battery, biomass generator while the remaining cost components are from feasibility study and system miscellaneous. In the proposed system the lifetime of the PV, battery and converter have been based on 25 years, 5 years and 10 years respectively. Therefore, at full project life in 25 years the battery will require replacement 4 times, while the converter will be replaced twice. For the proposed system the total annual cost is BDT 167,696 (USD \$1996.38). In periodic cost, it is seen that the battery replacement cost in full project life is BDT 403,200 (USD \$4800), and converter cost is BDT 89,600 (USD \$1066.67). After all the expenses the annual savings from the project is BDT 474,702 (USD \$5,651.21) which is provided in Table 6.


**Table 6.** Annual saving of proposed system.

#### *4.4. Financial Viability & Cumulative Cash Flow Analysis*

The RETScreen Software enables a user to input various forms of financial data such as discount rates, etc., which it automatically calculates to produce key financial feasibility indicators such as simple payback, equity payback, and net present value. Based on the data entered by the user, financial indicators for the project being analysed are provided, thus deriving vital information which facilitates the project evaluation process for planners and decision-makers [54].

The simple payback SP is the number of years it takes for the cash flow (excluding debt payments) to equal the total investment (which is equal to the sum of debt and equity):

$$SP = \frac{\mathbb{C} - IG}{\left(\mathbb{C}\_{energy} + \mathbb{C}\_{capacity} + \mathbb{C}\_{RE} + \mathbb{C}\_{GHG}\right) - \left(\mathbb{C}\_{O \&M} + \mathbb{C}\_{fuel}\right)} \tag{1}$$

where, C is the total initial cost of the project and IG is the value of incentives and grants. *Cenergy*, *Ccapacity*, *CRE*, and *CGHG* are annual energy saving or income, annual capacity saving or income, annual renewable energy production credit income and greenhouse gas reduction income respectively. *CO*&*M*, *Cf uel* represent the yearly operation and maintenance cost and yearly cost of fuel or electricity respectively.

Similarly, the year-to-positive cash flow (also equity payback), NPCF is the first year that the cumulative cash flows for the project are positive. It is calculated by solving the following equation for NPCF:

$$0 = \sum\_{n=0}^{N\_{\text{PCF}}} \stackrel{>}{\mathcal{C}}\_n \tag{2}$$

*4.5. Emissions Analysis*

where, > *C<sup>n</sup>* is the after tax cash flow in year n.

The net present value NPV of a project is calculated by discounting all cash flows as given in the following equation: =0 where, N is the project life in years and r is the discount rate.

= ∑

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 11 of 16

̌ (1 + ) 

$$NPV = \sum\_{n=0}^{N} \frac{\stackrel{\text{>}}{\mathcal{C}\_n}}{(1+r)^N} \tag{3}$$

(3)

(5)

where, N is the project life in years and r is the discount rate.

Discounted payback period, DPBP can be calculated that can be calculated that the discounted cash flow method discounts each inflow considering the time value of money until NPV equals zero at the certain year n of the system operation as indicated in Equation (4) [55,56]. ∑ − (1 + ) =0 = 0 (4)

$$\sum\_{n=0}^{DPBT} \frac{Cln - COn}{\left(1 + c\right)^n} = 0 \tag{4}$$

where, CI: cash inflow; CO: cash outflow; C: cost opportunity of capital; n: time period.

The annual life cycle savings ALCS is calculated using the following formula: = 1

$$ALCS = \frac{NPV}{1 - \frac{1}{\left(1 + r\right)^N}}\tag{5}$$

From the cash flow diagram depicted in Figure 5 using RETScreen, it can be estimated that it takes 7.2 years for cash flow to become positive and that the simple payback period will be 6.9 years. From the financial viability analysis, we get a Net Present Value (NPV) of BDT 855,428 (USD \$10,183.67) and annual life cycle saving is BDT 73,405 (USD \$873.87) and equity payback period is 7.2 years. In terms of the project's economics, we can say that the proposed hybrid system is the most economical one because after 7 years the project will start to generate profit and reduce the system's overall costs. From the cash flow diagram depicted in Figure 5 using RETScreen, it can be estimated that it takes 7.2 years for cash flow to become positive and that the simple payback period will be 6.9 years. From the financial viability analysis, we get a Net Present Value (NPV) of BDT 855,428 (USD \$10,183.67) and annual life cycle saving is BDT 73,405 (USD \$873.87) and equity payback period is 7.2 years. In terms of the project's economics, we can say that the proposed hybrid system is the most economical one because after 7 years the project will start to generate profit and reduce the system's overall costs.

**Figure 5.** Cumulative cash flows. **Figure 5.** Cumulative cash flows.

#### *4.5. Emissions Analysis*

The emissions' analysis estimates the greenhouse gas emission-reduction (mitigation) potential of the proposed case. RETScreen estimates the annual GHG emission reduction, ∆GHG of electricity by utilising the following Equation:

$$
\Delta\_{\rm GHG} = \left(e\_{\rm base} - e\_{\rm proposed}\right) \mathbf{E}\_{\rm proposed} \left(\mathbf{1} - \boldsymbol{\lambda}\_{\rm proposed}\right) \left(\mathbf{1} - e\_{\rm credit}\right) \tag{6}
$$

where *ebase*: base case GHG emission factor; *eproposed*: proposed case GHG emission factor; *Eproposed*: proposed case annual electricity produced; λ*proposed*: the fraction of electricity lost in transmission and distribution (T&D) for the proposed case. For off grid system consider the value of λ*proposed* is zero; *ecredit*: the GHG emission reduction credit transaction fee.

For a single fuel type or source, the following formula is used to calculate the base case electricity system GHG emission factor, *ebase*:

$$e\_{\rm base} = (e\_{\rm CO2} \, \text{GWP}\_{\rm CO2} + e\_{\rm CH4} \, \text{GWP}\_{\rm CH4} + e\_{\rm N2O} \, \text{GWP}\_{\rm N2O}) \frac{1}{\eta} \frac{1}{1 - \lambda} \tag{7}$$

where *eCO*2, *eCH*4, and *eN*2*<sup>O</sup>* are respectively the CO2, CH<sup>4</sup> and N2O emission factors for the fuel/source considered, *GWPCO*2, *GWPCH*4, and *GWPN*2*<sup>O</sup>* are the global warming potentials for CO2, CH<sup>4</sup> and N2O, η is the fuel conversion efficiency, and λ is the fraction of electricity lost in transmission and distribution. For standard analysis consider *GWPCO*2, *GWPCH*4, and *GWPN*2*<sup>O</sup>* as 1, 21, and 310.

In cases for which there are a number of fuel types or sources, the GHG emission factor *eproposed* for the electricity mix is calculated as the weighted sum of emission factors calculated for each individual fuel source:

$$e\_{\text{proposed}} = \sum\_{i=1}^{n} f\_{\text{'}} e\_{\text{proposed},i} \tag{8}$$

where n is the number of fuels/sources in the mix, *f<sup>i</sup>* is the fraction of end-use electricity coming from fuel/source i, and *eproposed*,*<sup>i</sup>* , is the emission factor for fuel i, calculated through a formula similar to Equation (8):

$$e\_{\text{proposed},i} = (e\_{\text{CO2},i}\text{GWP}\_{\text{CO2}} + e\_{\text{CH4},i}\text{GWP}\_{\text{CH4}} + e\_{\text{N2O},i}\text{GWP}\_{\text{N2O}})\frac{1}{\eta\_i}\frac{1}{1 - \lambda\_i} \tag{9}$$

where, *eCO*2,*<sup>i</sup>* , *eCH*4,*<sup>i</sup>* , and *eN*2*O*,*<sup>i</sup>* are respectively the CO2, CH<sup>4</sup> and N2O emission factors for fuel/source i, η i is the fuel conversion efficiency for fuel i, and λ<sup>i</sup> is fraction of electricity lost in transmission and distribution for fuel i. Consider all λ<sup>i</sup> are zero in case of mix of duel/sources [54].

Table 7 reports the estimated GHG emissions reduction in the proposed system. From the result, it is evident that the proposed system reduces the CO<sup>2</sup> emissions by 0.269 tonne/MWh compared to the base case.

Thus, using Equation (6), we can calculate the yearly CO<sup>2</sup> reduction in the proposed system for 14.161MWh annual electricity production which is 3.81 tonnes. Therefore, we can say that the proposed system is cost effective and also environmentally-friendly.

$$
\Delta\_{\rm GHG} = (0.269 - 0) \text{X } 14.161 = 3.81 \text{ tomes} \tag{10}
$$


**Table 7.** Emissions analysis.

#### **5. Conclusions**

This paper highlighted the benefits of using a hybrid energy system consisting of both solar energy and biomass energy to reduce energy costs and CO<sup>2</sup> emissions. The design was compared with data from the RETScreen data and also with the existing kerosene-based system. Due to the lack of the regional power grid, and available local resources, photovoltaic panel, biogas generator along with battery storage bank are the best solution for providing electricity in future. The paper demonstrated that the hybrid mini-grid system is the most economical and reliable for rural areas. Another useful part of employing a hybrid system is the minimal use of biomass generator which ultimately reduces the greenhouse gas emissions. The only drawback in this system is the battery cost. So, to make this proposed system a reliable one the government should take step to reduce the battery cost. In this hybrid energy model, simulation results showed that 81.2% of total energy is produced by PV and the rest of the energy comes from biomass generator. The simple payback of the project was estimated to be 6.9 years and this project will be able to reduce the CO<sup>2</sup> emissions by approximately 3.81 tonnes per year. The study proved that the proposed system is more reliable and cost-effective and also more environmentally friendly when compared with the kerosene-based system.

**Author Contributions:** N.C. and C.A.H. proposed the core idea, developed the models. N.C. performed the simulations, exported the results and analysed the data. N.C., C.A.H., M.L. and W.Y. contributed to the design of the models and the writing of this manuscript. M.L. and W.Y. revised the paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors declare no conflict of interest.

## **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **From NEDC to WLTP: E**ff**ect on the Energy Consumption, NEV Credits, and Subsidies Policies of PHEV in the Chinese Market**

**Xinglong Liu 1,2, Fuquan Zhao 1,2, Han Hao 1,2,3, Kangda Chen 1,2, Zongwei Liu 1,2,4,\*, Hassan Babiker <sup>5</sup> and Amer Ahmad Amer <sup>5</sup>**


Received: 30 June 2020; Accepted: 15 July 2020; Published: 17 July 2020

**Abstract:** The switching from new European driving cycle (NEDC) to worldwide harmonized light vehicles test procedure (WLTP) will affect the energy consumption of plug-in hybrid electric vehicle (PHEV), and then affect the new energy vehicle (NEV) credit regulation and subsidy policy for PHEVs. This paper reveals the impact on energy consumption, NEV credit regulation, and subsidy policy for PHEV in the Chinese market of the switching from NEDC to WLTP based on qualitative analysis and quantitative calculation. The results show that the WLTP procedure is stricter than NEDC in the determination of road load, test mass, driving resistance forces, and tire selection. Firstly, the electricity consumption (EC) of PHEV in charge-depleting mode (CD) under the WLTP procedure is 26% higher than NEDC on average, which makes the all-electric range (AER) significantly lower under WLTP. The weight EC tested in the WLTP procedure is higher than NEDC. Secondly, the fuel consumption (FC) of PHEV in CD mode is related to the adjustment of the engine management system (EMS) and the size of battery energy under the WLTP procedure. For the FC in the charge-sustaining (CS) mode of PHEV under the WLTP procedure is 20% higher than NEDC on average. However, the weight fuel consumption of PHEVs under WLTP with a long AER may be lower than that of NEDC due to the characteristics of utility factor in the WLTP procedure. Thirdly, most PHEVs fail to meet the requirements of 50 km AER due to the switching of the test procedures. However, the Chinese government reduced the technical specification of PHEV's AER under the WLTP procedure to 43 km to support the development of PHEV technology. It ensures that the switching of test procedures does not change the treatment that they could obtain, the NEV credits, and subsidy as a NEV in China. However, the increasing of the EC in CD mode and the FC in CS mode under the WLTP procedure makes the PHEV obtain lower credit and subsidy multiple compared with NEDC procedure.

**Keywords:** PHEV; NEDC; WLTP; energy consumption; NEV credit regulation; subsidy policy

#### **1. Introduction**

With the development of the automobile industry, China is facing the problems of energy security and environmental pollution [1–3]. The relevant national agencies have made a series of regulatory constraints on the energy consumption and emissions of vehicles to solve these problems [4,5]. According to the technical roadmap of energy-saving and new energy vehicles, the average fuel

consumption (FC) of new passenger cars in 2020 and 2025 will reach 5 L/100 km and 4 L/100 km, respectively [6]. Because of the characteristics of the traditional internal combustion engine (ICE), it has gradually been unable to meet the stringent regulatory requirements [7]. Therefore, electrical upgrading based on the traditional internal combustion engine has become an effective way to solve the problem of energy consumption and emissions [8]. However, considering the battery costs and charging problems, there are still many obstacles in large-scale promotion of fully electrified battery electric vehicles (BEV). Therefore, the plug-in hybrid electric vehicle (PHEV), combining the ideal characteristics of BEVs with the range of traditional internal combustion engine vehicles (ICEV), has become the preferred technology choice for automobile manufacturers to meet increasingly stringent regulations [9–11].

Furthermore, it is an advantage of PHEVs that they can operate in two different modes due to its complicate powertrain configuration [12]. First, in charge-depleting mode (CD), the motor is responsible for propulsion and the ICE is switched off. Secondly, in charge-sustaining (CS) mode, the ICE provides power to drive the wheels and keep the SOC within a certain range [13]. However, due to the complexity of PHEV operation mode, it is difficult to accurately evaluate its energy consumption in tests [14]. In the past, the NEDC has been used to evaluate the fuel economy of Europe and China. However, the European Union has stipulated the gradual implementation of WLTP instead of NEDC to assess the fuel economy of light vehicles since September 2017. Furthermore, WLTP will be adopted as the FC test regulation of light vehicles in China from 2021 to 2025.

However, the switching of test procedures will change the energy consumption and emissions of the whole vehicle, which will have a certain impact on the development strategy of automobile manufacturers [15]. For example, several major European automakers have changed their development strategies for engine turbocharging and discontinued some PHEVs due to the switching from NEDC to WLTP. Thus, the switching of test procedure would affect the energy consumption of PHEV. Moreover, PHEV is one of the new energy vehicles vigorously promoted by the Chinese government. The change of test procedures will further affect the NEV credits and subsidy availability of PHEV on the basis of energy consumption. Therefore, it is necessary to study the effect of NEDC switching to WLTP in advance. On the one hand, it can help the government to accurately evaluate the energy-saving and emission reduction effect brought by the promotion of PHEV. On the other hand, it could help automotive manufacturers to develop new energy vehicle technology development routes.

However, there are few studies on the energy consumption of PHEV under different test procedures. Pavlovic et al. compared the differences from the energy consumption of PHEVs between NEDC and WLTP through experimental tests [16]. They found that the all-electric range (AER) determined by the WLTP procedure was significantly lower than NEDC. However, the FC tested from WLTP were often lower than the corresponding NEDC with the increase of battery energy. Tsiakmakis et al. studied the driving range and FC of PHEV under WLTP and NEDC based on the simulation method [17,18]. They found that the ratio of carbon dioxide emissions of PHEVs at WLTP to NEDC was largely dependent on battery energy. With the increase of battery energy, the proportion decreases rapidly. They also concluded that the ratio of energy consumption of PHEV between WLTP and NEDC is 1 when the battery energy is 25 kWh. Soulouk et al. investigated the main changes that the WLTP test procedure implies to a mid-size sedan electrified vehicle design (series, parallel P2, and power split) and quantifies their impact on the vehicles fuel economy [19]. They found that across different electrified vehicle architectures, the vehicles' fuel economy under the WLTP procedure in CS mode substantially decreases compared to the NEDC. Moreover, the battery needs to deliver more energy in the WLTP cycle compared to the NEDC to meet the AER requirement. The above-mentioned literatures analyze the change of energy consumption due to the switching of test procedures, but it does not systematically compare the differences between the NEDC and WLTP procedures for PHEVs. Moreover, there is a gap in the impact of the switching of test procedures on the policies for PHEVs, such as NEVs credit regulation and subsidy policy.

Therefore, this paper aims to reveal the impact of the switching from NEDC to WLTP on PHEV energy consumption and its external policies. This paper is organized as follows: In the first section, the significance of the research is introduced. Next, the differences between the NEDC and WLTP procedures are analyzed using qualitative analysis method. Then, the impact of test procedures switching on the energy consumption, NEVs credits regulation, and subsidy policy for PHEV is studied in the way of quantitative calculation method. Following that, the paper puts forward relevant policy suggestions for the problems caused by test procedure switching. Finally, the last section summarizes all findings from this paper.

#### **2. Test Procedure Di**ff**erences between the NEDC and WLTP Procedure for PHEV**

#### *2.1. Di*ff*erences in Road Load Determination between NEDC and WLTP*

The first difference is the determination of test mass between NEDC and WLTP. The mass of the test vehicle used to determine the road load is equal to the curb mass plus 100 kg in the NEDC procedure [20]. While for the WLTP procedure, the test mass is equal to the reference mass plus the mass of the fitted equipment of specific vehicles and the representative load mass [21]. It can be seen that test mass determination from WLTP will be significantly higher than NEDC. Therefore, the driving resistance of the whole vehicle will increase during the energy consumption tests, which will improve the energy consumption of PHEV under the WLTP procedure.

The second difference is the tire selection between NEDC and WLTP. The rolling resistance coefficient of a tire is the main contributor to the total rolling resistance. It is well-known that the influencing factors of the rolling resistance coefficient of the tire are mainly the width and circumference of the tire, the inflation pressure, and the tread depth [20]. In tire selection, NEDC requires that the widest tire must be selected for testing, while WLTP selects the tire according to the rolling resistance level of the tire according to the tested vehicles. Although NEDC is stricter than WLTP in terms of requirements, the rolling resistance coefficient of the widest tire specified in NEDC is not the largest. Therefore, the rolling resistance coefficient of the tire selected in WLTP is larger than that in NEDC to a certain extent. Generally speaking, the greater the inflation pressure, the lower the rolling resistance coefficient. There is no regulation on tire pressure in NEDC, so it is usually done to inflate the tire to the maximum allowable pressure. However, the road load is determined when the tire pressure is set to the minimum value as specified in WLTP. In terms of tire pressure, NEDC has more advantages than WLTP. Finally, it is known that the greater the tread depth, the greater the rolling resistance coefficient. The WLTP procedure for the minimum tire tread depth is more stringent (80%) than the NEDC requirement of 50% [21]. In a word, WLTP is stricter than NEDC in terms of tire selection, which leads to the higher energy consumption of PHEV when testing in WLTP.

The third difference is the determination of the coefficient of resistance force between NEDC and WLTP [22]. In the process of the coast down tests, paired runs in alternate directions must be performed due to the practical impossibility to have a perfectly flat test track. In the calculation method, NEDC averages the up and down test time. Unlike this, WLTP averages the resistance force not time in both directions. Due to the difference of test time in two different directions, the average time of the test track is relatively long compared to the real test time. Therefore, this method of average time leads to some errors and results in the final road load coefficient is lower than the coefficient calculated utilizing average force. Therefore, WLTP is stricter than NEDC in the way of calculating driving resistance. It will make PHEV require more energy in tests. Also, NEDC ignores the moment of the inertia effect of components in the process of determining the resistance force. The resistance force determined by the WLTP procedure is about 3% higher than NEDC due to the effect of the moment of inertia [20].

#### *2.2. Di*ff*erences in Test Protocol and Driving Cycles between NEDC and WLTP*

The differences between the driving cycles of NEDC and WLTP are shown in Figure 1. The test cycle corresponding to the WLTP procedure is Worldwide harmonized light vehicles test cycle (WLTC). It can be seen from the figure that the WLTC cycle has a longer test time and distance compared with the NEDC cycle. Furthermore, the WLTC cycle has a significantly higher average maximum

speed and acceleration, which will significantly improve the testing energy consumption of PHEV [23]. Moreover, the large decrease in the idle ratio in the WLTC cycle will weaken the fuel-saving effect of the start/stop and hybrid power technology [24]. compared with the NEDC cycle. Furthermore, the WLTC cycle has a significantly higher average maximum speed and acceleration, which will significantly improve the testing energy consumption of PHEV [23]. Moreover, the large decrease in the idle ratio in the WLTC cycle will weaken the fuel-

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 4 of 18

**Figure 1.** Comparison of new European driving cycle (NEDC) and worldwide harmonized light vehicles test cycle (WLTC) driving cycles. **Figure 1.** Comparison of new European driving cycle (NEDC) and worldwide harmonized light vehicles test cycle (WLTC) driving cycles.

However, the higher engine loads experienced by vehicles under the WLTC cycle can make the engine work in the high-efficiency range, which might compensate for the FC caused by the higher dynamic effect to a certain extent [25]. Furthermore, the proportion of cold start for vehicles under the WLTC cycle is lower than the NEDC cycle, which will reduce the effect of cold start and further reduce FC. Moreover, the WLTP procedure introduced the specific gearshift strategy with a manual transmission by calculated the engine speed and vehicle characteristics, while in the NEDC procedure, the same fixed gear positions are used for all vehicles [21]. Thus, this new WLTP gearshift strategy results in engine speeds ranging in their lower end, which consequently provide better engine efficiency and lower FC. However, the higher engine loads experienced by vehicles under the WLTC cycle can make the engine work in the high-efficiency range, which might compensate for the FC caused by the higher dynamic effect to a certain extent [25]. Furthermore, the proportion of cold start for vehicles under the WLTC cycle is lower than the NEDC cycle, which will reduce the effect of cold start and further reduce FC. Moreover, the WLTP procedure introduced the specific gearshift strategy with a manual transmission by calculated the engine speed and vehicle characteristics, while in the NEDC procedure, the same fixed gear positions are used for all vehicles [21]. Thus, this new WLTP gearshift strategy results in engine speeds ranging in their lower end, which consequently provide better engine efficiency and lower FC.

In a word, the FC for the ICEVs will change limited (far less than expected) by simply analyzing the change from NEDC to the WLTC cycle assuming that other conditions remain unchanged. The amplitude of the FC change is related to the characteristics of the vehicle and the engine. This is also confirmed by other scholars [26–29]. They found that the CO2 emission ratio of WLTC and NEDC cycle is between 0.89 and 1.16 when only considering the difference of the test cycle. However, the EC for BEVs will be significantly improved under the worse WLTC cycle not having the compensating effect from the engine. Furthermore, the NEDC procedure only tests one cycle in CD mode for PHEVs, while WLTP In a word, the FC for the ICEVs will change limited (far less than expected) by simply analyzing the change from NEDC to the WLTC cycle assuming that other conditions remain unchanged. The amplitude of the FC change is related to the characteristics of the vehicle and the engine. This is also confirmed by other scholars [26–29]. They found that the CO<sup>2</sup> emission ratio of WLTC and NEDC cycle is between 0.89 and 1.16 when only considering the difference of the test cycle. However, the EC for BEVs will be significantly improved under the worse WLTC cycle not having the compensating effect from the engine.

extends to the whole CD mode until the battery reaches the SOC level in CS mode [16]. Therefore, the real EC of PHEV in CD mode cannot be accurately measured according to the measurement of NEDC if the EC of PHEV in CD mode is non-linear. Moreover, with the increase of the discharge depth, the energy consumption of the battery gradually increases, considering the discharge characteristics of the power battery [18]. It will result in that the EC in CD mode under WLTP will be significantly higher than NEDC. Meanwhile, there exist some differences when calculating the FC for PHEVs in CS mode between WLTP and NEDC. Under the WLTP procedure, PHEVs should be corrected the FC in CS mode for the difference of the SOC of the battery between the start and end of the CS test. If the change of SOC is greater than 0.5% and SOC is reduced (corresponding to battery discharge) in WLTP, FC correction will be enforced [16]. This is not considered under the NEDC procedure. From this point of view, WLTP will further increase FC in the tests. Furthermore, the NEDC procedure only tests one cycle in CD mode for PHEVs, while WLTP extends to the whole CD mode until the battery reaches the SOC level in CS mode [16]. Therefore, the real EC of PHEV in CD mode cannot be accurately measured according to the measurement of NEDC if the EC of PHEV in CD mode is non-linear. Moreover, with the increase of the discharge depth, the energy consumption of the battery gradually increases, considering the discharge characteristics of the power battery [18]. It will result in that the EC in CD mode under WLTP will be significantly higher than NEDC. Meanwhile, there exist some differences when calculating the FC for PHEVs in CS mode between WLTP and NEDC. Under the WLTP procedure, PHEVs should be corrected the FC in CS mode for the difference of the SOC of the battery between the start and end of the CS test. If the change of SOC is greater than 0.5% and SOC is reduced (corresponding to battery discharge) in WLTP, FC correction will be enforced [16]. This is not considered under the NEDC procedure. From this point of view, WLTP will further increase FC in the tests.

#### *2.3. Di*ff*erences in Post-Processing of the Energy Consumption Data between NEDC and WLTP*

For the calculation of energy consumption under the NEDC procedure, the final FC and EC are calculated by the following formulas [30]:

$$\mathbf{C}\_{\text{weight}}^{\text{NEDC}} = \frac{D\_{\varepsilon} \ast \mathbf{C}\_{\text{CD}}^{\text{NEDC}} + D\_{\text{dW}} \ast \mathbf{C}\_{\text{CS}}^{\text{NEDC}}}{D\_{\varepsilon} + D\_{\text{dW}}} = \text{UF}^{\text{NEDC}} \ast \mathbf{C}\_{\text{CD}}^{\text{NEDC}} + \left(1 - \text{UF}^{\text{NEDC}}\right) \ast \mathbf{C}\_{\text{CS}}^{\text{NEDC}} \tag{1}$$

$$E\_{\text{weight}}^{\text{NEDC}} = \frac{D\_{\varepsilon} \ast E\_{\text{CD}}^{\text{NEDC}}}{D\_{\varepsilon} + D\_{\text{av}}} = \text{UF}^{\text{NEDC}} \ast E\_{\text{CD}}^{\text{NEDC}} \tag{2}$$

where *C NEDC weight E NEDC weight* represent the FC (EC) for PHEV; *C NEDC CD E NEDC CD* indicate the FC (EC) in CD mode; *C NEDC CS* indicates the FC in CS mode. *D<sup>e</sup>* is the electric range of the PHEV and *Dav* is the average distance has driven in CS mode, 25 km. To facilitate comparison with the method to calculate the weight energy consumption of the WLTP procedure, *UFNEDC* is assumed the equivalent utilization factor. It assumes that the EC in CS mode is 0 under the NEDC procedure, which is the same as the WLTP procedure. It makes the comparison simplified between the EC under different test procedures in this paper.

As abovementioned, NEDC only tests one cycle for PHEV in CD mode, so the FC of PHEVs whose electric range is over 11.03 km (one NEDC cycle) in CD mode is 0. Considering that the current AER of PHEV is more than 50 km, the formula adopted to calculate the weight FC of PHEV under NEDC procedure can be simplified as follows:

$$\mathbf{C}\_{\text{weight}}^{\text{NEDC}} = \frac{D\_{\text{av}} \ast \mathbf{C}\_{\text{CS}}^{\text{NEDC}}}{D\_{\text{e}} + D\_{\text{av}}} = \left(1 - \text{UF}^{\text{NEDC}}\right) \ast \mathbf{C}\_{\text{CS}}^{\text{NEDC}} \tag{3}$$

However, the favorable testing assumptions under the NEDC procedure in CD mode will be eliminated with the improvement of WLTP. Thus, the FC in the CD mode test cannot be ignored, which will make the FC of PHEV in CD mode higher than NEDC. In WLTP, the FC in CD mode and final weight FC is calculated according to the different weights of each phase in CD mode. The formula is as follows [21]:

$$\begin{cases} \begin{aligned} \mathcal{C}^{\text{WLTP}}\_{\text{weight}} = \left(\sum\_{j=1}^{k} \text{LF}\_{j}\right) \ast \mathcal{C}^{\text{WLTP}}\_{\text{CD}} + \left(1 - \sum\_{j=1}^{k} \text{LF}\_{j}^{\text{WLTP}}\right) \ast \mathcal{C}^{\text{WLTP}}\_{\text{CS}}\\ \mathcal{C}^{\text{WLTP}}\_{\text{CD}} = \frac{\sum\_{j=1}^{k} \left(\text{LF}\_{j}^{\text{WLTP}} \ast \mathcal{C}\_{\text{CD},j}\right)}{\sum\_{j=1}^{k} \text{LF}\_{j}} \end{aligned} \tag{4}$$

where *C WLTP weight* is the weight FC under the WLTP procedure for PHEVs, L/100 km, *UFWLTP j* is the utility factor of the CD phase j under the WLTP procedure; *C WLTP CD* is the FC under the WLTP procedure in CD mode and *C WLTP CS* is the FC under the WLTP procedure in CD mode, L/100 km; *k* represents the number of velocity segments tested from the CD test to the transition cycle; *CCD*,*<sup>j</sup>* is the FC of phase j in CD mode, L/100 km.

When calculating the weight EC under the WLTP procedure, the EC in CS mode is not considered, so the calculation formula is as follows:

$$\begin{cases} \begin{aligned} ^{\text{WLTTP}} &= \left( \sum\_{j=1}^{k} \text{UF}\_{j} \right) \ast \, ^{\text{WLTP}} \\ ^{\text{WLTTP}} &= \frac{\sum\_{j=1}^{k} \text{UF}\_{j} ^{\text{WLTP}} \ast \, ^{\text{E}\_{\text{CD},j}}}{\sum\_{j=1}^{k} \text{UF}\_{j}} \end{aligned} \tag{5} \\ ^{\text{WLT}} &= \frac{\sum\_{j=1}^{k} \text{UF}\_{j}}{\sum\_{j=1}^{k} \text{UF}\_{j}} \end{aligned} \tag{6}$$

where *E WLTP weight* is the weight EC under the WLTP procedure for PHEVs, kWh/100 km, *E WLTP CD* is the EC under the WLTP procedure in CD mode, kWh/100 km; *ECD*,*<sup>j</sup>* is the EC of phase *j* in CD mode, kWh/100 km.

By comparing the calculation formula of weight energy consumption between NEDC and WLTP, the results of a PHEV are strongly affected by the utility factor related to the electric range [31,32]. Thus, it is not accurate for the NEDC procedure to set 25 km as the average distance of CS mode, which does not conform to the current travel characteristics [32]. It depends on the AER of the specific vehicle. If the vehicle has a high AER, it is likely to drive mainly in CD mode. If the AER of the vehicle is lower, the probability of driving in CS mode is higher. The WLTP procedure introduces the utility factor based on the driver's travel characteristics to more accurately describe the driving probability in CD and CS modes, as shown in Figure 2. It should be noted that since China's current test regulations refer to Europe, its utility factor curve also comes from Europe. However, considering the difference in travel characteristics between China and Europe, China should develop the utility factor that adapts to the travel characteristics of China [33]. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 18 factor based on the driver's travel characteristics to more accurately describe the driving probability in CD and CS modes, as shown in Figure 2. It should be noted that since China's current test regulations refer to Europe, its utility factor curve also comes from Europe. However, considering the difference in travel characteristics between China and Europe, China should develop the utility factor that adapts to the travel characteristics of China [33].

**Figure 2.** Comparison of utility factor between NEDC and worldwide harmonized light vehicles test procedure (WLTP). **Figure 2.** Comparison of utility factor between NEDC and worldwide harmonized light vehicles test procedure (WLTP).

It can be seen from Figure 2 that the utility factor calculated based on NEDC-NEDC and WLTP-WLTP is approximately equal in the range of short driving range. However, the energy consumption of CD mode under the WLTP procedure is calculated by weighting the utility factor of each phase of cycles, so it is more consistent with the travel behaviors. Also, considering that the actual electric range under the two test procedures is not the same, to better evaluate the difference between the two utility factors, the equivalent utility factor of NEDC is calculated by using AER under the WLTP procedure (making the electric range in NEDC transferred to the electric range in WLTP), which is the NEDC-WLTP utility factor curve as shown in Figure 2. It can be found that the equivalent utility factor of the NEDC procedure is higher than that of WLTP when the AER is short. Therefore, this will reduce the proportion of FC of PHEV in CS mode when calculating FC. It results in the FC in CS mode measured by WLTP being relatively high compared to NEDC, derived from formula (4). Meanwhile, the utility factor of WLTP is higher than the equivalent utility factor of NEDC when the AER of PHEV under WLTP is greater than 38km (intersection point in Figure 2). Therefore, when calculating the weight FC, the FC of PHEV with a high driving range under the WLTP procedure will below. On the contrary, with the continuous increase of the AER, the utility factor of WLTP is It can be seen from Figure 2 that the utility factor calculated based on NEDC-NEDC and WLTP-WLTP is approximately equal in the range of short driving range. However, the energy consumption of CD mode under the WLTP procedure is calculated by weighting the utility factor of each phase of cycles, so it is more consistent with the travel behaviors. Also, considering that the actual electric range under the two test procedures is not the same, to better evaluate the difference between the two utility factors, the equivalent utility factor of NEDC is calculated by using AER under the WLTP procedure (making the electric range in NEDC transferred to the electric range in WLTP), which is the NEDC-WLTP utility factor curve as shown in Figure 2. It can be found that the equivalent utility factor of the NEDC procedure is higher than that of WLTP when the AER is short. Therefore, this will reduce the proportion of FC of PHEV in CS mode when calculating FC. It results in the FC in CS mode measured by WLTP being relatively high compared to NEDC, derived from Formula (4). Meanwhile, the utility factor of WLTP is higher than the equivalent utility factor of NEDC when the AER of PHEV under WLTP is greater than 38 km (intersection point in Figure 2). Therefore, when calculating the weight FC, the FC of PHEV with a high driving range under the WLTP procedure will below. On the contrary, with the continuous increase of the AER, the utility factor of WLTP is

In essence, the energy consumption test of PHEVs is to test the energy consumption in CD mode and CS mode, respectively. Then, calculate the weight energy consumption by different weighting methods according to the NEDC and WLTP procedures. To analyze the impact of switching from NEDC to WLTP on the energy consumption of PHEV accurately, this part makes quantitative analysis from three parts: Energy consumption in CD mode, energy consumption in CS mode, and final weight energy consumption. It should be noted that the energy consumption data are mainly

**3. Effect on the Energy Consumption of PHEVs from NEDC to WLTP** 

significantly higher than the equivalent utility factor of NEDC. Hence, the weight value of NEDC may be lower than that of WLTP, when calculating the weight FC.

### **3. E**ff**ect on the Energy Consumption of PHEVs from NEDC to WLTP**

In essence, the energy consumption test of PHEVs is to test the energy consumption in CD mode and CS mode, respectively. Then, calculate the weight energy consumption by different weighting methods according to the NEDC and WLTP procedures. To analyze the impact of switching from NEDC to WLTP on the energy consumption of PHEV accurately, this part makes quantitative analysis from three parts: Energy consumption in CD mode, energy consumption in CS mode, and final weight energy consumption. It should be noted that the energy consumption data are mainly collected from the literature [16,18,19], tested by the China Automotive Technology and Research Center (CATRC) [34], and provided by relevant automobile manufacturers in this paper. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 7 of 18 collected from the literature [16,18,19], tested by the China Automotive Technology and Research Center (CATRC) [34], and provided by relevant automobile manufacturers in this paper.

#### *3.1. Analysis of the Energy Consumption in CD Mode (CCD and ECD) 3.1. Analysis of the Energy Consumption in CD mode*(ܥ *and* ܧ(

As abovementioned, the NEDC procedure only tests one cycle for PHEV in CD mode, the FC will be 0 if the AER of the testing vehicle over 11.03 km. However, the WLTP procedure tests the complete CD phase. The engine of PHEV will start in the condition that the power demand is higher than the power provided by the battery or the SOC of the battery is relatively lower [16]. To understand the test procedure and the determination of energy consumption for PHEVs under WLTP, the testing results of two PHEVs tested in the Joint Research Centre (JRC) of the European Commission Laboratories are cited in this study [16], as shown in Figure 3. The results showed the driving cycles, SOC of battery, and engine revolutions per minute (RPM). As abovementioned, the NEDC procedure only tests one cycle for PHEV in CD mode, the FC will be 0 if the AER of the testing vehicle over 11.03 km. However, the WLTP procedure tests the complete CD phase. The engine of PHEV will start in the condition that the power demand is higher than the power provided by the battery or the SOC of the battery is relatively lower [16]. To understand the test procedure and the determination of energy consumption for PHEVs under WLTP, the testing results of two PHEVs tested in the Joint Research Centre (JRC) of the European Commission Laboratories are cited in this study [16], as shown in Figure 3. The results showed the driving cycles, SOC of battery, and engine revolutions per minute (RPM).

**Figure 3.** WLTP charge depleting test sequence for two plug-in hybrid electric vehicles (PHEVs) from the study of Joint Research Centre (JRC). **Figure 3.** WLTP charge depleting test sequence for two plug-in hybrid electric vehicles (PHEVs) from the study of Joint Research Centre (JRC).

For vehicle 1 in Figure 3, although the SOC was not already at its minimum value, the ICE started during the extra-high speed part of the first WLTC. It may be the consequence that the engine management system (EMS) for the specific vehicle was tuned by the OEMs to behave in a certain way on the NEDC cycle, thus bringing this unexpected behavior in the WLTP testing [16]. As abovementioned, the maximum speed of WLTC is much higher than that of NEDC, and that most For vehicle 1 in Figure 3, although the SOC was not already at its minimum value, the ICE started during the extra-high speed part of the first WLTC. It may be the consequence that the engine management system (EMS) for the specific vehicle was tuned by the OEMs to behave in a certain way on the NEDC cycle, thus bringing this unexpected behavior in the WLTP testing [16]. As abovementioned, the maximum speed of WLTC is much higher than that of NEDC, and that most likely was the cause of the behavior that ICE started early.

likely was the cause of the behavior that ICE started early. For parallel or hybrid PHEVs, if the EMS has not been tuned based on WLTC or the motor power is lower than the required power at the maximum speed, the engine will start even if the SOC is not For parallel or hybrid PHEVs, if the EMS has not been tuned based on WLTC or the motor power is lower than the required power at the maximum speed, the engine will start even if the SOC is

from the wheels or the series PHEV. The engine only started in the last WLTC in CD mode. Therefore, the FC mainly comes from the last cycle of CD mode if the EMS for the engine for PHEV is tuned

At the same time, it can be found that the FC of Vehicle 1 in CD mode is significantly lower than that of Vehicle 2. This is mainly because the Vehicle 2 has a large power battery capacity so that the utility factor in the last cycle is significantly smaller than that of the Vehicle 2 (0.2% < 22.2%). According to formula (4), the utility factor in the last cycle is lower, its FC in CD mode is lower. Therefore, the more WLTC cycles the PHEV experiences, the lower the FC in CD mode. In a word, the contribution of FC in CD mode to the real FC gets lower with the increase of battery capacity.

based on WLTC, as shown in Figure 3.

and NEDC, as shown in Figure 4.

not at its minimum value. Meanwhile, the engine must start in the last WLTC due to the low SOC. Therefore, to reduce the FC of PHEV in CD mode, the vehicle manufacturer needs to adjust the EMS to avoid starting the engine of PHEV frequently in high SOC state. However, the engine decouples from the wheels or the series PHEV. The engine only started in the last WLTC in CD mode. Therefore, the FC mainly comes from the last cycle of CD mode if the EMS for the engine for PHEV is tuned based on WLTC, as shown in Figure 3.

At the same time, it can be found that the FC of Vehicle 1 in CD mode is significantly lower than that of Vehicle 2. This is mainly because the Vehicle 2 has a large power battery capacity so that the utility factor in the last cycle is significantly smaller than that of the Vehicle 2 (0.2% < 22.2%). According to Formula (4), the utility factor in the last cycle is lower, its FC in CD mode is lower. Therefore, the more WLTC cycles the PHEV experiences, the lower the FC in CD mode. In a word, the contribution of FC in CD mode to the real FC gets lower with the increase of battery capacity. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 18

The EC in CD mode under the NEDC procedure is obtained by the meaning of dividing the electricity consumed in a test cycle by the range of an NEDC cycle (11.03 km). The EC in the CD mode under the WLTP procedure is to consider the weight value of each speed phase in the whole CD mode, as shown in Formula (5). Through collecting the EC of PHEVs from the CATRC in China and other research institutions in the world, it is the EC of 10 vehicles in the CD mode under the WLTP and NEDC, as shown in Figure 4. The EC in CD mode under the NEDC procedure is obtained by the meaning of dividing the electricity consumed in a test cycle by the range of an NEDC cycle (11.03 km). The EC in the CD mode under the WLTP procedure is to consider the weight value of each speed phase in the whole CD mode, as shown in formula (5). Through collecting the EC of PHEVs from the CATRC in China and other research institutions in the world, it is the EC of 10 vehicles in the CD mode under the WLTP

**Figure 4.** Electricity consumption (EC) of different PHEVs in charge-depleting (CD) mode under the NEDC and WLTP procedures. **Figure 4.** Electricity consumption (EC) of different PHEVs in charge-depleting (CD) mode under the NEDC and WLTP procedures.

It can be found that the EC in CD mode under the WLTP procedure is significantly higher than that under the NEDC procedure, with an average of about 26%. The results are consistent with those of JRC and ICCT reports [18,32]. It can be explained that the calculation of curb weight, tire selection, and driving resistance calculation of the WLTP procedure are stricter than those under the NEDC procedure. Meanwhile, The EC test of PHEV in CD mode is similar to that of battery electric vehicles, which is more sensitive to driving cycles as abovementioned. So, severe WLTP will cause higher EC. Furthermore, the JRC report points out that the EC ratio of the WLTP and NEDC procedure is related to the PHEV types [18]. It concluded that the heavier the vehicle type is, the larger the ratio is. This is mainly resulting from that the determination of road load is related to the curb weight. The calculation of curb weight under the WLTP procedure is stricter than that under the NEDC procedure. Because there are few vehicles selected in this study, there is no comparison on vehicle types. It could investigate the factor of vehicle type using the simulation method in the future work. According to formula (5), the EC of PHEV in CD mode could represent the AER of the It can be found that the EC in CD mode under the WLTP procedure is significantly higher than that under the NEDC procedure, with an average of about 26%. The results are consistent with those of JRC and ICCT reports [18,32]. It can be explained that the calculation of curb weight, tire selection, and driving resistance calculation of the WLTP procedure are stricter than those under the NEDC procedure. Meanwhile, The EC test of PHEV in CD mode is similar to that of battery electric vehicles, which is more sensitive to driving cycles as abovementioned. So, severe WLTP will cause higher EC. Furthermore, the JRC report points out that the EC ratio of the WLTP and NEDC procedure is related to the PHEV types [18]. It concluded that the heavier the vehicle type is, the larger the ratio is. This is mainly resulting from that the determination of road load is related to the curb weight. The calculation of curb weight under the WLTP procedure is stricter than that under the NEDC procedure. Because there are few vehicles selected in this study, there is no comparison on vehicle types. It could investigate the factor of vehicle type using the simulation method in the future work.

corresponding vehicle. Therefore, it can be concluded that the AER of PHEV decreases by 26% on average with the switching of the testing procedure. Therefore, some PHEVs less than 63 km (NEDC) no longer meet the present regulatory requirements with the 50 km AER. So, the automotive manufacturers need to further increase the energy of the battery to meet the regulatory requirement According to Formula (5), the EC of PHEV in CD mode could represent the AER of the corresponding vehicle. Therefore, it can be concluded that the AER of PHEV decreases by 26% on average with the switching of the testing procedure. Therefore, some PHEVs less than 63 km (NEDC) no longer meet the present regulatory requirements with the 50 km AER. So, the automotive

It can be seen from the calculation formula of weight energy consumption of PHEV that the EC in the CS mode is generally not considered. Therefore, this paper mainly analyzes the FC of PHEV in the CS mode. Considering that PHEV has the same FC characteristics as HEVs with a higher mixing degree when tested in the CS mode, the FC data of the HEVs are taken into account to analyze the FC in the CS mode for PHEV. The FC data are collected from the literature [16,18,19] and provided

for PHEV when designing after the switching of test procedures.

by an automotive manufacturer as shown in Figure 5.

(ܥௌ)

*3.2. Analysis of the FC in CS Mode*

manufacturers need to further increase the energy of the battery to meet the regulatory requirement for PHEV when designing after the switching of test procedures.

## *3.2. Analysis of the FC in CS Mode (CCS)*

It can be seen from the calculation formula of weight energy consumption of PHEV that the EC in the CS mode is generally not considered. Therefore, this paper mainly analyzes the FC of PHEV in the CS mode. Considering that PHEV has the same FC characteristics as HEVs with a higher mixing degree when tested in the CS mode, the FC data of the HEVs are taken into account to analyze the FC in the CS mode for PHEV. The FC data are collected from the literature [16,18,19] and provided by an automotive manufacturer as shown in Figure 5. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 9 of 18

**Figure 5.** Fuel consumption (FC) change in the charge-sustaining (CS) mode of PHEVs from NEDC to WLTP. **Figure 5.** Fuel consumption (FC) change in the charge-sustaining (CS) mode of PHEVs from NEDC to WLTP.

As can be seen from Figure 5, FC in CS mode for PHEV under the WLTP procedure is significantly higher than that under the NEDC procedure, with an average of about 20%. This is explained in three aspects. First, the WLTP procedure is stricter than the NEDC procedure in the determination of test mass, tire selection, and driving resistance. Second, the correction of SOC under the WLTP procedure will also increase the FC of PHEV. Finally, the driving cycle of WLTC is stricter than the NEDC cycle for PHEV. On the one hand, the load can greatly improve FC due to the characteristics of the WLTC cycle. On the other hand, a small proportion of idle speed will reduce the fuel-saving capacity of the hybrid power system, which will also increase the FC of PHEV under the As can be seen from Figure 5, FC in CS mode for PHEV under the WLTP procedure is significantly higher than that under the NEDC procedure, with an average of about 20%. This is explained in three aspects. First, the WLTP procedure is stricter than the NEDC procedure in the determination of test mass, tire selection, and driving resistance. Second, the correction of SOC under the WLTP procedure will also increase the FC of PHEV. Finally, the driving cycle of WLTC is stricter than the NEDC cycle for PHEV. On the one hand, the load can greatly improve FC due to the characteristics of the WLTC cycle. On the other hand, a small proportion of idle speed will reduce the fuel-saving capacity of the hybrid power system, which will also increase the FC of PHEV under the WLTP procedure.

#### WLTP procedure. *3.3. Analysis of the Weight Energy Consumption for PHEV*

*3.3. Analysis of the Weight Energy Consumption for PHEV*  The weight energy consumption of PHEV is calculated by weighting the energy consumption in the CD mode and CS mode. Table 1 shows the weight energy consumption of 10 different PHEVs, among which vehicle 1 and vehicle 2 are measured by JRC through experiments, while other data of eight PHEVs are from CATRC. It can be found that the weight EC of various PHEVs is lower than that in the CD mode, which is mainly due to the advantage of the weighted formula for EC. Furthermore, the weight EC of 10 PHEVs averagely increases by about 38% when switching from NEDC to WLTP, which is significantly higher than that in CD mode (26%). This shows that WLTP is The weight energy consumption of PHEV is calculated by weighting the energy consumption in the CD mode and CS mode. Table 1 shows the weight energy consumption of 10 different PHEVs, among which vehicle 1 and vehicle 2 are measured by JRC through experiments, while other data of eight PHEVs are from CATRC. It can be found that the weight EC of various PHEVs is lower than that in the CD mode, which is mainly due to the advantage of the weighted formula for EC. Furthermore, the weight EC of 10 PHEVs averagely increases by about 38% when switching from NEDC to WLTP, which is significantly higher than that in CD mode (26%). This shows that WLTP is stricter than NEDC in the calculation of weight EC of PHEV.

stricter than NEDC in the calculation of weight EC of PHEV. **Table 1.** Weight energy consumption of 10 PHEVs. **Vehicle ID Curb Mass (kg) Engine Displacement (L) Battery Capacity (kWh) AER (km) Weight FC (L/100km) Weight EC (kWh/100km )**  NED C WLT P NED C WLT P NED C WLT P Vehicle 1 1315 0.65 21.8 161.8 122.8 0.68 0.40 11.67 16.71 The weight FC of PHEV under the WLTP and NEDC procedure is shown in Figure 6. It can be seen that the weight energy consumption of PHEVs has increased significantly except for vehicle 1. The decrease of weight FC of vehicle 1 shows that the weight FC under the WLTP procedure might be lower than that under the NEDC procedure. It gives the credit to the difference of the calculation method of weight FC between NEDC and WLTP. In other words, the calculation method of WLTP is in favor of the PHEVs with a longer AER. Moreover, this phenomenon will be more significant with the increase in battery capacity. Thus, it can be said that the WLTP procedure is more friendly to the weight FC of PHEV with a larger battery.

Vehicle 7 1955 1.50 12.0 81.0 74.0 1.30 2.30 11.32 13.56 Vehicle 8 2037 1.50 13.0 60.0 56.0 1.60 3.30 15.29 19.41 Vehicle 9 2281 2.00 18 70.0 59.0 1.60 3.80 18.95 25.51 Vehicle10 2390 2.00 19 81.0 68.0 1.80 2.90 17.92 23.36 The weight FC of PHEV under the WLTP and NEDC procedure is shown in Figure 6. It can be seen that the weight energy consumption of PHEVs has increased significantly except for vehicle 1. The decrease of weight FC of vehicle 1 shows that the weight FC under the WLTP procedure might

Vehicle 2 1524 1.40 8.7 39.3 27.9 2.23 2.77 13.53 22.83 Vehicle 3 1505 1.80 8.0 55.0 43.0 1.30 2.70 10.00 13.58


**Table 1.** Weight energy consumption of 10 PHEVs. method of weight FC between NEDC and WLTP. In other words, the calculation method of WLTP is

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 10 of 18

be lower than that under the NEDC procedure. It gives the credit to the difference of the calculation

**Figure 6.** Weight FC of PHEVs under the WLTP and NEDC procedures. **Figure 6.** Weight FC of PHEVs under the WLTP and NEDC procedures.

**4. Effect on the Policies of PHEV from NEDC to WLTP**  To promote the rapid development of the new energy automobile industry, the Chinese government proposed corresponding mandatory regulations and policy incentives, such as the NEV credit policy and subsidy policy [35–37]. In particular, the Chinese government delayed the decline of subsidies for two years in order to maintain the development of NEVs due to the COVID-19 outbreak in China. In order to ensure the development of NEVs driven by technology, the threshold value to meet the NEV credits regulations and subsidy policies is directly related to the energy-saving technology of vehicles. The energy consumption of NEVs will change with switching from NEDC to WLTP, which will affect NEV credits and subsidy policies. The newly issued NEV credits regulation [38] and subsidy It can be seen from the figure that the WLTP/NEDC ratio of weight FC is relatively large, with an average of about 1.94. This is quite different from the results (0.6 ~ 1.2) in the JRC report [18]. It needs to be explained that the weight FC data in this paper are all from the vehicle model currently on sale. The EMS of the engine of PHEV developed in China is tuned based on the NEDC cycle now, so the engine will start frequently under the WLTP procedure with the stricter driving cycle. Therefore, the engine will start frequently even if the SOC is high and to generate high FC in the CD mode. With the implementation of the WLTP procedure in China in 2021, automobile manufacturers will tune the EMS according to the WLTC driving cycle. This will make the engine not start frequently in the CD mode under the WLTP procedure. Then, the WLTP/NEDC ratio of weight FC will be further reduced in the future.

#### policy [39] in China, the following analysis is made on the impact of testing procedure switching on the policies for PHEVs. The NEV credit policy and subsidy policy for PHEVs are evaluated by the EC **4. E**ff**ect on the Policies of PHEV from NEDC to WLTP**

index in CD mode and the FC index in CS mode [38,39]. Table 2 presents the EC in CD mode and FC in the CS mode of the above 10 PHEVs under the NEDC and WLTP procedures. Among them, the FC in the CS mode of PHEV (Vehicle 3-Vehicle 8) in the China market is not given specific data. So, this paper assumes that the FC in CS mode under the WLTP procedure is 1.2 times of that in the NEDC procedure based on the above research. **Table 2.** EC in CD mode and FC in CS mode of different PHEVs under the NEDC and WLTP To promote the rapid development of the new energy automobile industry, the Chinese government proposed corresponding mandatory regulations and policy incentives, such as the NEV credit policy and subsidy policy [35–37]. In particular, the Chinese government delayed the decline of subsidies for two years in order to maintain the development of NEVs due to the COVID-19 outbreak in China. In order to ensure the development of NEVs driven by technology, the threshold value to meet the NEV credits regulations and subsidy policies is directly related to the energy-saving technology of vehicles.

procedures. **Energy consumption Vehicle ID 1 2 3 4 5 6 7 8 9 10**  The energy consumption of NEVs will change with switching from NEDC to WLTP, which will affect NEV credits and subsidy policies. The newly issued NEV credits regulation [38] and subsidy policy [39] in China, the following analysis is made on the impact of testing procedure switching on the policies for PHEVs. The NEV credit policy and subsidy policy for PHEVs are evaluated by the EC index in CD mode and the FC index in CS mode [38,39]. Table 2 presents the EC in CD mode and FC in the CS mode of the above 10 PHEVs under the NEDC and WLTP procedures. Among them, the FC

in the CS mode of PHEV (Vehicle 3-Vehicle 8) in the China market is not given specific data. So, this paper assumes that the FC in CS mode under the WLTP procedure is 1.2 times of that in the NEDC procedure based on the above research.


**Table 2.** EC in CD mode and FC in CS mode of different PHEVs under the NEDC and WLTP procedures.

## *4.1. E*ff*ect on the NEV Credit Regulation of PHEV from NEDC to WLTP*

The NEVs credits obtained by PHEV is equal to the basic credit multiplied by the credit multiple. According to the latest NEV credit regulation, the basic credit of PHEV is 1.6, while the credit multiple is determined by the values of the EC in CD mode and the FC in CS mode, which are 1 or 0.5. If the EC in CD mode for PHEV is greater than 135% of the EC targets of the same vehicle type BEV, or the FC in CS mode is greater than 70% of the FC limits for passenger car for the corresponding vehicle type, the credit multiple is 0.5, while the other is 1. The specific calculation is shown in Formula (6):

$$\text{CR} = \begin{cases} \text{0.8, } \left( \text{if } E\_{\text{CD}} \ge 135\% \text{ EC}\_{\text{target}} \text{ or } \text{C}\_{\text{cs}} \ge 70\% \text{ FC}\_{\text{limit}} \right) \\\text{1.6, other} \end{cases} \tag{6}$$

where *CR* is the credits that could be obtained by PHEV; *ECtarget* indicates the EC targets for the same vehicle type BEV, which is the function of curb mass, kwh/100 km; *FClimit* is the FC limits for the corresponding vehicle type, which is the function of curb mass, L/100 km.

The NEVs credits distribution for 10 PHEVs under the NEDC procedure is as shown in Figure 7. It should be noted that only when the EC in CD mode and the FC in CS mode are lower than the target and limit curves at the same time can one credit multiplier be obtained. It can be seen from the figure that 9 PHEVs can obtain NEVs credits except for vehicle 2 whose AER is less than 50 km. Among them, vehicle 3 and vehicle 9 can only get 0.5 times credit multiplier due to high EC, and the other seven PHEVs can get 1 credit multiplier.

The NEVs credits distribution for 10 PHEVs under the WLTP procedure is as shown in Figure 8. It can be seen from the figure that only two PHEVs can obtain one credit multiplier and three PHEVs can obtain a 0.5 credit multiplier after the test procedure is switched from NEDC to WLTP. The remaining five PHEVs could not obtain the NEVs credits. This shows that the switching of the test procedure has a great influence on the NEVs credit acquisition of PHEVs. The reason that five PHEVs cannot obtain NEVs credit is that their AER is less than 50 km under the WLTP procedure. It makes the PHEVs not meet the origin regulatory requirements and not enjoy policy privilege. Therefore, to obtain the NEVs credit of PHEV, the automobile manufacturers need to further increase the total energy of the battery and improve the AER of PHEV in the future. Furthermore, for the PHEV obtaining a 0.5 credit multiplier, the main reason is that the EC is too high, but the FC is still within the technological threshold of the NEV credit policy. It indicates that the influence of switching test procedures on the NEV credit policy for PHEV is higher in CD mode than in CS mode. Thus, the automobile manufacturers need to further consider the energy-saving technology of EC in the future vehicle design, such as the improvement of battery and motor efficiency, the optimization of the battery management system, and the further implementation of vehicle lightweight technology.

other seven PHEVs can get 1 credit multiplier.

NEDC-CD EC

NEDC-CS FC

WLTP-CD EC

WLTP-CS FC

(kWh/100km) 13.47 22.14 14.55 17.74 16.98 20.00 14.81 21.67 25.71 23.46

(L/100km) 6.12 5.21 4.30 4.60 4.30 5.20 5.50 5.50 6.10 7.50

(kWh/100km) 17.75 31.18 18.60 28.21 19.15 26.09 16.22 23.21 30.51 27.91

(L/100km) 6.66 5.78 5.16 5.52 5.16 6.24 6.60 6.60 7.32 9.00

The NEVs credits obtained by PHEV is equal to the basic credit multiplied by the credit multiple. According to the latest NEV credit regulation, the basic credit of PHEV is 1.6, while the credit multiple is determined by the values of the EC in CD mode and the FC in CS mode, which are 1 or 0.5. If the EC in CD mode for PHEV is greater than 135% of the EC targets of the same vehicle type BEV, or the FC in CS mode is greater than 70% of the FC limits for passenger car for the corresponding vehicle type, the credit multiple is 0.5, while the other is 1. The specific calculation is shown in formula (6):

The NEVs credits distribution for 10 PHEVs under the NEDC procedure is as shown in Figure 7. It should be noted that only when the EC in CD mode and the FC in CS mode are lower than the target and limit curves at the same time can one credit multiplier be obtained. It can be seen from the

(6)

( ௧ܥܨ 70% ௦ܥ ݎ ௧௧ܥܧ 135% ܧ ݂݅) 0.8,൜ = ܴܥ

the corresponding vehicle type, which is the function of curb mass, L/100 km.

 ݎ݄݁ݐ 1.6, where ܥܴ is the credits that could be obtained by PHEV; ܥܧ௧௧ indicates the EC targets for the same vehicle type BEV, which is the function of curb mass, kwh/100 km; ܥܨ௧ is the FC limits for

*4.1. Effect on the NEV Credit Regulation of PHEV from NEDC to WLTP* 

**Figure 7.** The new energy vehicles (NEVs) credits distribution for 10 PHEVs under the NEDC **Figure 7.** The new energy vehicles (NEVs) credits distribution for 10 PHEVs under the NEDC procedure. vehicle design, such as the improvement of battery and motor efficiency, the optimization of the battery management system, and the further implementation of vehicle lightweight technology.

**Figure 8.** The NEVs credits distribution for 10 PHEVs under the WLTP procedure. **Figure 8.** The NEVs credits distribution for 10 PHEVs under the WLTP procedure.

#### *4.2. Effect on the Subsidy Policy of PHEV from NEDC to WLTP 4.2. E*ff*ect on the Subsidy Policy of PHEV from NEDC to WLTP*

According to the latest NEVs subsidy policy, the basic subsidy of PHEV is 8500 ¥ (RMB). The subsidy multiplier is related to the AER of PHEV and the FC in CS mode. According to the technological thresholds of subsidy policy, the FC in CS mode should be less than 60% compared with the FC limit of the same vehicle type for PHEVs with an AER of less than 80 km under the NEDC procedure. When the ratio is between 55% and 60%, the subsidy multiplier of the subsidy for PHEVs is 0.5. When the ration is less than 55%, the subsidy multiplier of the subsidy for PHEVs is 1. For PHEVs with AER under the NEDC procedure greater than or equal to 80 km, the EC in CD mode shall meet the requirements of the technological threshold for battery electric passenger vehicles in 2019. The specific calculation is shown in formula (7): According to the latest NEVs subsidy policy, the basic subsidy of PHEV is 8500 ¥ (RMB). The subsidy multiplier is related to the AER of PHEV and the FC in CS mode. According to the technological thresholds of subsidy policy, the FC in CS mode should be less than 60% compared with the FC limit of the same vehicle type for PHEVs with an AER of less than 80 km under the NEDC procedure. When the ratio is between 55% and 60%, the subsidy multiplier of the subsidy for PHEVs is 0.5. When the ration is less than 55%, the subsidy multiplier of the subsidy for PHEVs is 1. For PHEVs with AER under the NEDC procedure greater than or equal to 80 km, the EC in CD mode shall meet the requirements of the technological threshold for battery electric passenger vehicles in 2019. The specific calculation is shown in Formula (7):

$$\text{SU} = \begin{cases} 8500 \ast 0.5, \ (if \text{ 55\%} \text{FC}\_{\text{limit}} \le \text{C}\_{\text{CS}} < 60\% \text{ FC}\_{\text{limit}})\\ 8500 \ast 1.0, (if \text{ C}\_{\text{CS}} < 55\% \text{ FC}\_{\text{limit}} \text{ or } AER \ge 80 \text{ km}) \end{cases} \tag{7}$$

km. The subsidy distribution for 10 PHEVs under the NEDC and WLTP procedure is shown in where *SU* is the subsidy that PHEV could obtain, ¥ (RMB); *AER* is the all-electric range of PHEV, km.

its AER over 80 km (81 km). However, only three PHEVs could obtain 0.5 multiplier subsidies, but no one could obtain one times subsidies when switching from NEDC to WLTP procedure. Five PHEVs could not obtain subsidies because their AER is less than 50 km, which does not meet range requirements for PHEVs in China. Two PHEVs cannot obtain subsidies due to their high FC in CS mode, which is over the FC limits. It can be concluded that the switching of the test procedure has a great impact on the subsidy policy of PHEV. Thus, the automobile manufacturers need to further

The subsidy distribution for 10 PHEVs under the NEDC and WLTP procedure is shown in Figure 9. It can be seen from the figure that all vehicles could obtain the one multiplier of subsidy except Vehicle 1 under the NEDC procedure. It should be noted that although the FC in CS mode of Vehicle 10 is within the range of 0.5 multiplier subsidy, it can also obtain 1 multiplier subsidy due to its AER over 80 km (81 km). However, only three PHEVs could obtain 0.5 multiplier subsidies, but no one could obtain one times subsidies when switching from NEDC to WLTP procedure. Five PHEVs could not obtain subsidies because their AER is less than 50 km, which does not meet range requirements for PHEVs in China. Two PHEVs cannot obtain subsidies due to their high FC in CS mode, which is over the FC limits. It can be concluded that the switching of the test procedure has a great impact on the subsidy policy of PHEV. Thus, the automobile manufacturers need to further increase the total energy of the battery and expand the AER to fit the switching from NEDC to WLTP, which is also the strategy commonly used by European automobile manufacturers. Therefore, to obtain more subsidies for PHEV, it is necessary to further develop energy-saving technology based on increasing battery capacity, to reduce the FC in CS mode to the scope of subsidies. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 13 of 18 increase the total energy of the battery and expand the AER to fit the switching from NEDC to WLTP, which is also the strategy commonly used by European automobile manufacturers. Therefore, to obtain more subsidies for PHEV, it is necessary to further develop energy-saving technology based on increasing battery capacity, to reduce the FC in CS mode to the scope of subsidies.

**Figure 9.** The subsidy distribution for 10 PHEVs under the NEDC and WLTP procedure. **Figure 9.** The subsidy distribution for 10 PHEVs under the NEDC and WLTP procedure.

In a word, the switch from NEDC to WLTP makes PHEV in a disadvantageous situation in the NEVs policy and subsidy policy. The main influence comes from two aspects: First, the switching of working conditions makes the AER of PHEV greatly reduced, which does not meet the requirements for the technical specifications of PHEV of the Chinese government, and cannot continue to enjoy the NEV credit and subsidy policies privilege; on the other hand, the switching of test procedures makes the energy consumption of PHEV in CD mode and CS mode increase, so that the PHEV can obtain In a word, the switch from NEDC to WLTP makes PHEV in a disadvantageous situation in the NEVs policy and subsidy policy. The main influence comes from two aspects: First, the switching of working conditions makes the AER of PHEV greatly reduced, which does not meet the requirements for the technical specifications of PHEV of the Chinese government, and cannot continue to enjoy the NEV credit and subsidy policies privilege; on the other hand, the switching of test procedures makes the energy consumption of PHEV in CD mode and CS mode increase, so that the PHEV can obtain the corresponding multipliers of the NEV credit and subsidy reduced.

the corresponding multipliers of the NEV credit and subsidy reduced. However, the Chinese government has always adhered to the development route of PHEV. Although the test procedures are switched, it is not intended to make it difficult for automotive manufacturers to develop PHEV. In order to ensure that PHEV can still meet the requirements of technical specifications and obtain corresponding NEV credits and subsidies after the test procedure However, the Chinese government has always adhered to the development route of PHEV. Although the test procedures are switched, it is not intended to make it difficult for automotive manufacturers to develop PHEV. In order to ensure that PHEV can still meet the requirements of technical specifications and obtain corresponding NEV credits and subsidies after the test procedure switching, the Chinese government has reduced the AER of PHEV to 43 km in the latest technical requirements.

switching, the Chinese government has reduced the AER of PHEV to 43km in the latest technical requirements. Figure 10 shows the distribution of NEV credits and subsidies obtained under WLTP after reducing the AER technological specification of PHEVs. It can be seen from Figure 10a that three PHEVs can obtain one multiplier NEV credits after the AER of technological requirement is reduced. Figure 10 shows the distribution of NEV credits and subsidies obtained under WLTP after reducing the AER technological specification of PHEVs. It can be seen from Figure 10a that three PHEVs can obtain one multiplier NEV credits after the AER of technological requirement is reduced. Moreover, five PHEVs can obtain 0.5 multiplier NEV credits while only two PHEVs cannot obtain NEV credits due to the AER is less than 43 km. Compared to the Figure 8, the decrease of AER technological requirement could improve

the switching of test procedures could not alter the accessibility of NEV credits when the technological requirement of AER reducing to 43 km under the WLTP procedure. However, the increasing of EC and FC in the CD and CS mode under the WLTP procedure could decrease the

Meanwhile, it can be seen from Figure 10b that six PHEVs can get 0.5 multiplier subsidy under the WLTP procedure after the AER is reduced to 43 km, mainly because the FC in CS mode is too high to meet the technical thresholds of 1 multiplier subsidy. The remaining four PHEVs are not subsidized, of which two are due to the AER less than 43 km, and the other two are due to the FC in CS mode exceeding the threshold value of subsidy, so they are not subsidized. Compared to Figure

multiplier of NEV credits compared to NEDC.

subsidies [40].

**5. Policy Suggestion** 

the accessibility of NEV credits. Compared with Figure 7, the switching of test procedures could not alter the accessibility of NEV credits when the technological requirement of AER reducing to 43 km under the WLTP procedure. However, the increasing of EC and FC in the CD and CS mode under the WLTP procedure could decrease the multiplier of NEV credits compared to NEDC. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 14 of 18 9, the decrease of AER technological requirement could improve the accessibility of subsidy while

the higher FC in CS mode could lead to the lower multiplier subsidy.

**Figure 10.** The NEV credits and subsidy distribution for 10 PHEVs under WLTP procedure after adjusting the all-electric range (AER) to 43 km in Chinese market: (**a**) NEV credits distribution; (**b**) subsidy distribution. **Figure 10.** The NEV credits and subsidy distribution for 10 PHEVs under WLTP procedure after adjusting the all-electric range (AER) to 43 km in Chinese market: (**a**) NEV credits distribution; (**b**) subsidy distribution.

This shows that the switching of test procedures has a limited impact on PHEV, which does not affect the availability of NEV credit and subsidies for PHEV with the support of the Chinese government. The switching of test procedure decreases the multipliers of NEV credit and subsidy due to the higher energy consumption. Thus, in order to further obtain the NEV credits and subsidies of PHEV, automobile manufacturers need to carry out the research and development of vehicle energy-saving technology, so as to reduce the EC in CD mode and the FC of CS mode. Therefore, it can be concluded that the switching of test procedure mainly puts forward higher requirements for Meanwhile, it can be seen from Figure 10b that six PHEVs can get 0.5 multiplier subsidy under the WLTP procedure after the AER is reduced to 43 km, mainly because the FC in CS mode is too high to meet the technical thresholds of 1 multiplier subsidy. The remaining four PHEVs are not subsidized, of which two are due to the AER less than 43 km, and the other two are due to the FC in CS mode exceeding the threshold value of subsidy, so they are not subsidized. Compared to Figure 9, the decrease of AER technological requirement could improve the accessibility of subsidy while the higher FC in CS mode could lead to the lower multiplier subsidy.

the energy-saving technology of PHEV, but does not alter the availability of NEV credits and

This shows that the switching of test procedures has a limited impact on PHEV, which does not affect the availability of NEV credit and subsidies for PHEV with the support of the Chinese government. The switching of test procedure decreases the multipliers of NEV credit and subsidy due to the higher energy consumption. Thus, in order to further obtain the NEV credits and subsidies of PHEV, automobile manufacturers need to carry out the research and development of vehicle energy-saving technology, so as to reduce the EC in CD mode and the FC of CS mode. Therefore, it can be concluded that the switching of test procedure mainly puts forward higher requirements for the energy-saving technology of PHEV, but does not alter the availability of NEV credits and subsidies [40].

#### **5. Policy Suggestion**

The switching from NEDC to WLTP makes great changes in the energy consumption of PHEV in CD and CS modes to the extent that weight energy consumption, which affects the acquisition of NEV credit and subsidy of PHEVs to a extend. Therefore, to promote the reasonable development of PHEV in China after the test procedure change, the following suggestions are proposed:

First, the switching from NEDC to WLTP procedure increases the energy consumption of PHEV and makes most PHEVs unable to meet the requirements of 50 km AER. The Chinese government has changed the AER of PHEV to 43 km, indicating that the government has not tightened the requirements of PHEV, and the automotive manufacturer does not need to increase the battery to improve AER. However, the EC in CD mode and the FC of CS mode under the WLTP procedure are increased due to the condition switching. If the automotive manufacturers want to reach the corresponding fuel regulations, they need to further develop the energy-saving technology of the whole vehicle.

Secondly, the Chinese government's relaxation of PHEV's AER has little impact on the availability of PHEV's NEV credits and subsidies, so automotive manufacturers should not worry too much. However, with the increase of EC in CD mode and FC in CS mode, PHEV will obtain NEV credit and subsidy multiples. Thus, in order to obtain higher NEV credit and subsidy multiples, automotive manufactures need to research and develop energy-saving technologies of PHEV.

Thirdly, China automotive test cycle (CATC) will be used instead of WLTP to test the energy consumption of PHEVs after 2025. The Chinese government needs to evaluate the effect of the switching from WLTP to CATC in advance to make a smooth transition of PHEV's technical route. At the same time, automotive manufactures should prepare in advance, and develop and design PHEV according to the difference between CATC and WLTP.

Fourthly, China only considers the FC in the energy consumption test of PHEVs at present. In order to reasonably evaluate the energy consumption of PHEV, it is necessary to take the power consumption into account for comprehensive evaluation in the future. Then, the Chinese government should do a good job in policy planning, and automotive manufacturers should be prepared in advance.

#### **6. Conclusions and Prospect**

This paper studies the impact of switching from NEDC to WLTP on energy consumption, NEV credits, and subsidy policies of PHEVs through qualitative analysis and quantitative calculation methods. Firstly, the differences between the NEDC and WLTP procedures for testing energy consumption of PHEVs are qualitatively compared. Secondly, the effect of switching of test procedures on the energy consumption of PHEV is quantitative analysis. Finally, the NEV credit and subsidy changes of PHEVs are analyzed based on the energy consumption analysis after test procedure switching. According to the analysis results, the following conclusions are obtained:

(1) For the testing of PHEVs, the WLTP procedure is stricter in the determination of road load and test mass than those in the NEDC procedure. The weight calculation method for weight EC in the WLTP procedure is stricter than that in the NEDC procedure. The weight calculation method for weight FC in the WLTP procedure is stricter than that in the NEDC procedure for the PHEVs with a short AER. However, with the increase of the AER, the weight energy consumption under the WLTP procedure may be lower than NEDC due to the weight calculation for weight FC in the WLTP procedure friendly to PHEVs with longer AER.


This paper only analyzes the differences between NEDC and WLTP procedures for testing energy consumption and to explore the effect of switching of test procedures on the NEV credits and subsidies for PHEV. However, it fails to give how to choose the power assembly of engine, battery, and motor to reduce the energy consumption of PHEV under the WLTP procedure to obtain more NEVs credit and subsidies. Moreover, the design value of the AER of PHEV under the WLTP procedure as well as the specific improvement measures of NEV credits and subsidy policies are not studied. Therefore, the battery and engine sizes of PHEV will be studied based on the WLTP procedure, and then evaluate the reasonability of NEV credit and subsidy policies based on the abovementioned study in the future work.

**Author Contributions:** Conceptualization, X.L., F.Z., and Z.L.; methodology, X.L. and Z.L.; software, X.L.; validation, X.L. and K.C.; formal analysis, K.C.; investigation, X.L.; resources, X.L.; data curation, X.L.; writing—original draft preparation, X.L., Z.L., and K.C.; writing—review and editing, Z.L. and H.H.; visualization, X.L.; supervision, A.A.A. and H.B.; project administration, F.Z. and Z.L.; funding acquisition, Z.L.; A.A.A., and H.B.. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study is supported by the National Natural Science Foundation of China (U1764265, 71774100, 71690241) and the Chinese Academy of Engineering (2019-XZ-55-01-01).

**Acknowledgments:** The authors would like to thank Saudi Aramco for generous financial support, and Xing He for discussions. The authors would like to thank the anonymous reviewers for their reviews and comments.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviations**



#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Hybrid Wind–PV Frequency Control Strategy under Variable Weather Conditions in Isolated Power Systems**

**Ana Fernández-Guillamón 1,\* ,†, Guillermo Martínez-Lucas 2,†, Ángel Molina-García 1,† and Jose-Ignacio Sarasua 2,†**


Received: 7 August 2020; Accepted: 15 September 2020; Published: 19 September 2020

**Abstract:** Over the last two decades, variable renewable energy technologies (i.e., variable-speed wind turbines (VSWTs) and photovoltaic (PV) power plants) have gradually replaced conventional generation units. However, these renewable generators are connected to the grid through power converters decoupled from the grid and do not provide any rotational inertia, subsequently decreasing the overall power system's inertia. Moreover, the variable and stochastic nature of wind speed and solar irradiation may lead to large frequency deviations, especially in isolated power systems. This paper proposes a hybrid wind–PV frequency control strategy for isolated power systems with high renewable energy source integration under variable weather conditions. A new PV controller monitoring the VSWTs' rotational speed deviation is presented in order to modify the PV-generated power accordingly and improve the rotational speed deviations of VSWTs. The power systems modeled include thermal, hydro-power, VSWT, and PV power plants, with generation mixes in line with future European scenarios. The hybrid wind–PV strategy is compared to three other frequency strategies already presented in the specific literature, and gets better results in terms of frequency deviation (reducing the mean squared error between 20% and 95%). Additionally, the rotational speed deviation of VSWTs is also reduced with the proposed approach, providing the same mean squared error as the case in which VSWTs do not participate in frequency control. However, this hybrid strategy requires up to a 30% reduction in the PV-generated energy. Extensive detailing of results and discussion can be also found in the paper.

**Keywords:** frequency control; power system stability; variable renewable energy sources; wind power plants; photovoltaic power plants

#### **1. Introduction**

Over the last two decades, power systems' generation has slowly been changing, replacing conventional generation units (mainly based on the fossil and nuclear fuels) with variable renewable energy sources (vRESs) [1]. This transition has been supported by several aspects, including environmental concern (especially greenhouse gas emissions) and the aim of decreasing the energy dependence of third countries [2–4]. vRES refers to variable-speed wind turbines (VSWTs) and photovoltaic (PV) power plants, which have a stochastic behavior due to their dependence on weather conditions [5]. Together with this point, vRESs are connected to the grid through power inverters decoupled from the grid and, hence, do not inherently provide any inertial or frequency response

under power imbalances [6]. Consequently, the high penetration levels of vRESs will substantially challenge the supply security and reliability of future power systems, thus requiring more flexibility to maintain the balance between generation and demand [7–9]. Moreover, this real-time power balance is required for a stable grid frequency [10].

As an attempt to improve the frequency response of power systems with high vRES integration, several frequency control techniques have been proposed for such generation units in the specific literature [11,12]. In fact, they are usually classified as depicted in Figure 1. Together with these strategies, energy storage systems (i.e., batteries, flywheels, and supercapacitors) have also been considered as suitable options to maintain grid stability [13,14]. With regard to PV power plants, as they are static elements without any rotating parts, they do not have any stored kinetic energy [15]. Therefore, their synchronous inertia constant is nearly zero (*HPV* ≈ 0) [16]. Consequently, they can only provide frequency response based on de-loading strategies. The de-loading technique implies the reduction of the supplied active power under normal operation conditions, going towards the right part of the P–V curve (refer to Figure 2). In this way, when grid frequency is lower than its nominal value, the PV installations can slightly decrease their voltage (from *Vdel* to *VMPP*, see Figure 2) and subsequently increase the corresponding generated power [17]. The ∆*V* decrease is usually estimated based on the grid frequency deviation ∆*f* through a proportional or proportional–integral controller. Similarly to PV power plants, VSWTs can also provide frequency response using the de-loading technique. In this case, it is performed by pitching the blades or over-speeding the rotor [18]. De-loading strategies imply a reduction of the electrical power generated (both for PV and VSWTs) and, subsequently, a considerable decrease in the benefits for their owners [19]. As VSWTs have some stored kinetic energy due to their blades, drive train, and the electrical generator, some authors affirm that VSWTs' inertia constant *HWT* is in line with those of conventional power plants, but it is hidden from the power system point of view due to the power converter [20,21]. Consequently, together with the de-loading approach, VSWTs can provide such rotational inertia to the grid through different inertial response strategies. These strategies require at least one supplementary control loop to be included in the power controller [22]. Under frequency deviation conditions, the frequency control loop(s) will provide an additional input to the VSWTs, aiming to temporarily increase their active power generation. With this type of controller, VSWTs can always generate their maximum active power according to each wind speed value, in contrast to the de-loading technique, in which they slightly curtailed certain amounts of power. As a result, inertial response approaches are preferable to the de-loading technique, since the wind resource is better used [23]. However, inertial response approaches are limited by the VSWTs' speed governor, which prevents the rotational speed of the VSWTs from decreasing below the minimum allowed value [24]. Tables 1 and 2 summarize different vRES frequency control strategies proposed in the specific literature over the last years, including the type of control, PV/VSWT integration level, and the power imbalance (∆*P*) that causes the frequency deviation. As can be seen in Tables 1 and 2, these frequency control strategies are usually analyzed under severe power imbalances (up to 50%). However, in isolated power systems with high wind–PV integration, the variability of wind speed and solar irradiation commonly causes large deviations in the system frequency [25]. Indeed, these fluctuations pose stress on the power system operation, as transmission and distribution system operators (TSOs/DSOs) deal with not only unmanageable demand, but also uncontrollable generation [26]. Consequently, there is an acute need to improve the frequency response of isolated power systems with high vRES integration, especially under typical variable meteorological changes. In fact, over the last decade, some studies have already focused on this topic. However, most of the works proposed are based on including energy storage systems [27–34], demand response [35], and electric vehicles [36], or using independent frequency controllers for VSWT and PV power plants [37–39].

**Figure 1.** Frequency control strategies for variable renewable energy sources (vRESs).

**Figure 2.** De-loading technique of photovoltaic (PV) power plants.


**Table 1.** PV frequency control proposals.

**Table 2.** Variable-speed wind turbine (VSWT) frequency control proposals.


In contrast to the previous studies, this paper proposes a hybrid wind–PV frequency control strategy in which the VSWTs' rotational speed deviation is the proportional–integral (PI) controller input of the PV frequency response strategy. Similar approaches have been previously proposed by the authors, but by linking the rotational speed deviation of VSWTs to hydro-power plants instead of PV power plants [55]. The main contributions of this paper can be summarized as:


The rest of the paper is organized as follows: The models used for the simulations are described in Section 2; the methodology followed to carry out this study is presented in Section 3; in Section 4, the results of the study are shown and analyzed; finally, Section 5 presents the main conclusions of the paper.

## **2. Proposed System Modeling**

A mathematical simulation model is used to analyze the proposed hybrid wind–PV frequency controller. This power system and, consequently, the model implemented in Matlab/Simulink includes conventional generators (hydro-power and thermal power plants), PV power plants, and VSWTs as well as the power demand, as shown in Figure 3. The frequency control is in line with current European requirements (primary and secondary controls), as will be presented in Section 3. In the following subsections, the main components of the proposed model are described in detail.

**Figure 3.** Block diagram of the model.

#### *2.1. Power System and System Inertia*

The power system is modeled considering an aggregated inertial model as proposed in [56]. This formulation has been previously used to model the Irish isolated power system [57] and El Hierro isolated power system [58], among others. Grid frequency variations are the result of the imbalance between the power generation and the corresponding demand (see Equation (1)):

$$f\frac{df}{dt} = \frac{1}{2\ H\_{eq}} \left( p\_w + p\_{PV} + p\_{hyd} + p\_{ther} - p\_{dem} - D\_{net} \Delta f \right),\tag{1}$$

where *pw*, *pPV*, *phyd*, and *pther* represent the power supplied by VSWT, PV, hydro-power, and thermal power plants, respectively; *pdem* is the total power demand; and *Dnet* is the consumer load sensitivity factor to frequency variations. System inertia (*Heq*) is estimated as [59]:

$$H\_{eq} = H\_{hyd} + H\_{ther} \tag{2}$$

where *Hhyd* and *Hther* are the hydro-power and thermal power plants' inertia constants, respectively. The VSWT and PV power plants are connected to the grid through power converters and, therefore, do not inherently provide synchronous inertia to the grid, as already discussed in Section 1.

#### *2.2. Conventional Power Plants*

The conventional power plants (CPPs) considered in this paper are reheat thermal and hydro-power plants. Both of them are modeled following the transfer functions proposed in [60]. Thermal power plants' transfer function, shown in Figure 4a, provides the power variations of these power plants from the frequency deviation (∆*f*) and the automatic generation control (AGC) power reference (*pther*,*re f*). Figure 4b shows the transfer function to model the hydro-power plant, including the conduits' dynamics. The generated power *phyd* also depends on the frequency deviation ∆*f* and the power reference signal provided by the AGC (*phydr*,*re f*). The parameters for both models in Figure 4 are listed in Table 3.

**Figure 4.** Block diagram of conventional power plants: (**a**) thermal power plants' transfer function and (**b**) hydro-power plants' transfer function.

#### *2.3. Wind Power Plants*

Four different wind power plants (WPPs) are considered by modeling each WPP with an aggregated wind turbine. Therefore, one equivalent wind turbine is then simulated by multiplying the corresponding generated power by the number of VSWTs of such a WPP [61]. The VSWT model includes the wind power model, which determines the power extracted from the wind speed; the blade pitch control to regulate the wind input torque; the torque maximum power point (MPP) tracking control, which restores the optimal rotor speed after a rotational speed deviation; and the frequency controller, which is discussed in Section 3.2. Further information with regard to the VSWT model can be found in [62]. The one-mass rotor mechanical model is used for simulations, and is acceptable when the voltage is assumed to be constant [63,64]:

$$
\omega\_{WT} = \frac{p\_{mt} - p\_W}{2 \, H\_{WT} s} \,. \tag{3}
$$

The reference rotational speed *ωre f* is determined from the measured power *pe f* , which is the active generated power *p<sup>W</sup>* after a delay *T<sup>f</sup>* :

$$
\omega\_{ref} = -0.67 \times p\_{ef}^2 + 1.42 \times p\_{ef} + 0.51 \,\text{.}\tag{4}
$$

The VSWT model is represented in Figure 5. The values of the different parameters are listed in Table 3.

**Figure 5.** Block diagram of a VSWT.

#### *2.4. PV Power Plant*

Four different PV power plants are also considered. They are modeled following the diode equation, where the output current *IPV* is given by [65,66]:

$$I\_{PV} = N\_P \times I\_{ph} - N\_P \times I\_{rs} \times \left[ \exp\left(\frac{q}{k\_B \times T\_\odot \times A} \times \frac{V\_{PV}}{N\_s}\right) - 1\right].\tag{5}$$

*Ns* is the number of cells in series, *N<sup>P</sup>* is the number of strings in parallel, *Iph* is the photo-current, *q* is the electron charge, *k<sup>B</sup>* is Boltzmann's constant, *T<sup>C</sup>* is the temperature of the cell, *A* is the diode ideality factor, and *VPV* is the PV voltage. The photo-current *Iph* is calculated by:

$$I\_{ph} = \left[I\_{\rm sc} + k\_I (T\_{\rm C} - T\_{\rm STC})\right] \times \mathcal{G} \tag{6}$$

where *Isc* is the PV cell's short-circuit current, *k<sup>I</sup>* is the short-circuit current temperature coefficient of the PV cell, *TSTC* is the temperature under Standard Test Conditions (STC), and *G* is the solar irradiation. The PV cell's reverse saturation current *Irs* follows Equation (7), where *Ir*,*re f* is the reverse saturation current at *TSTC*, and *E<sup>G</sup>* is the band-gap energy of the PV cell's material:

$$I\_{\rm rs} = I\_{\rm r,ref} \times \left(\frac{T\_{\rm C}}{T\_{\rm STC}}\right)^3 \times \exp\left[\frac{q \times E\_{\rm G}}{k\_B \times A} \times \left(\frac{1}{T\_{\rm STC}} - \frac{1}{T\_{\rm C}}\right)\right]. \tag{7}$$

The voltage at the MPP (*VMPP*) for a given irradiation *G* and cell temperature *T<sup>C</sup>* is estimated depending on its value at STC and a correction factor *αV*:

$$V\_{MPP} = V\_{MPP,STC} \times \left(\frac{\ln(G)}{\ln(1000)}\right) \times (1 + \mathfrak{a}\_V \times (T\_\mathbb{C} - T\_{STC})).\tag{8}$$

By multiplying the PV voltage *VPV* by the output PV current *IPV*, the active power of the PV generator is then estimated. In the case that the PV voltage corresponds to the MPP voltage *VMPP*, the output power of the PV power plants would be the maximum available active power under such *G* and *T<sup>C</sup>* conditions. The values of the different parameters considered are listed in Table 3.


**Table 3.** Power plant parameters [60,62,66].

### **3. Methodology**

Frequency control strategies focus on minimizing grid frequency variations due to generation–demand mismatches [67]. With this aim, generation units must increase/decrease their generation to equal the total power demand (plus the power system losses) [68]. In the following subsections, frequency controls for CPPs, VSWTs, and PV power plants are described.

#### *3.1. Frequency Control in Conventional Power Plants*

CPPs based on synchronous generators inherently release or absorb kinetic energy as a natural inertial response to frequency deviations. In Europe, frequency control with CPPs has a hierarchical structure, usually organized in primary control (frequency containment reserves), secondary control (frequency restoration reserves), and tertiary control (replacement reserves) [69]. Primary frequency control (PFC) is automatically activated by the generator units some seconds after the power imbalance. The CPP power response is proportional to the frequency deviation ∆*f* , following Equation (9), in which *R* is the droop characteristic [70,71]:

$$
\Delta p = -\frac{\Delta f}{R}.\tag{9}
$$

According to the European network of transmission system operators for electricity (ENTSO-E), *R* should range between 2% and 12%, with a dead-band between 10 and 30 mHz in which the PFC is not activated [72]. For this work, *R* is considered as 5% for both thermal and hydro-power plants, and the dead-band is established at 30 mHz. Subsequent to the PFC action, there is still some remaining frequency deviation due to the power imbalance. The secondary frequency control is then required to completely remove this frequency deviation [73,74]. In fact, this secondary frequency control is in charge of modifying the power generation set-point accordingly [75]. The AGC coordinates the effort's dispatch among the different CPPs of the secondary frequency control. In this paper, the equivalent total secondary regulation effort (∆*RR*) is determined by:

$$
\Delta RR = -\Delta f \times \mathcal{K}\_{f'} \tag{10}
$$

where *K<sup>f</sup>* is estimated following the ENTSO-E's recommendations [76]. This ∆*RR* is distributed between the thermal and hydro-power plants depending on their participation factors (*Ku*,*<sup>i</sup>* ), determined according to the droop of each power plant [77]:

$$
\Delta p\_{i, \text{ref}} = \frac{1}{T\_{u, i}} \int \Delta \text{RR} \times \mathbf{K}\_{u, i} \, dt = \frac{-1}{T\_{u, i}} \times \mathbf{K}\_{u, i} \times \mathbf{K}\_f \times \int \Delta f \, dt,\tag{11}
$$

where *i* represents *hyd* or *ther*, and *Ku*,*hyd* + *Ku*,*ther* = 1.

#### *3.2. VSWT Frequency Control Strategy*

Among the three different inertial response strategies for VSWTs (refer to Figure 1), the hidden-inertia emulation technique is selected for this work. The hidden-inertia emulation technique is based on a proportional–derivative (*PD*) control loop, with ∆*f* as input. This *PD* controller provides an additional power proportional to the frequency deviation and its derivative value:

$$
\Delta p\_{\rm FC} = \mathcal{K}\_d \times \frac{d\Delta f}{dt} + \mathcal{K}\_p \times \Delta f \,. \tag{12}
$$

The derivative part represents the inertial control loop (emulating the hidden-inertia of the VSWTs). These inertial (derivative) and proportional control loops allow that some of the kinetic energy stored in the rotating masses of the VSWTs (i.e., rotor, drive train, and electrical generator) is released to provide a fast frequency response from the power converter's capability. This ∆*pFC* signal is added to the power reference output depending on the wind speed.

#### *3.3. PV Frequency Control Strategy*

The de-loading technique is used for PV frequency control. This strategy curtails a certain amount of active power, working the PV power plant on a de-loaded voltage value (*Vdel*) on the right side of the MPP voltage (*VMPP*), as shown in Figure 2. Note that *Vdel* > *VMPP*. In the following subsections, the traditional de-loading strategy and the proposed hybrid wind–PV frequency control approach are analyzed.

#### 3.3.1. Conventional De-Loaded PV Frequency Control Strategy

To provide frequency response, the PV power plant usually works at the de-loaded voltage *Vdel*. In the case of frequency reduction, the PV voltage should be also reduced, moving towards *VMPP* to increase its generated active power. The strategies previously proposed in the specific literature include a proportional (*P*) or proportional–integral (*PI*) controller, with the frequency deviation (∆*f*) as an input of such a controller [40–43]. Frequency deviation (∆*f*) passes through the corresponding control loop (*P* or *PI*), giving an additional voltage ∆*V* (refer to Figure 6). The modified PV voltage is then determined as:

$$V\_{PV} = V\_{del} - \Delta V \,. \tag{13}$$

This *VPV* is used in Equation (5) to determine the PV current, *IPV*, which is the corresponding PV active power:

$$P\_{PV} = I\_{PV} \times V\_{PV} \,. \tag{14}$$

**Figure 6.** Conventional de-loaded PV frequency control.

#### 3.3.2. Hybrid Wind–PV Frequency Control

In this work, and in contrast to previous studies, the authors propose the use of the VSWT rotational speed deviation ∆*ω* as an input for the de-loaded PV frequency controller. Due to the VSWTs' hidden-inertia frequency controller, the rotational speed *ω<sup>j</sup>* of each WPP deviates from the reference value *ωre f* ,*<sup>j</sup>* , following Equation (15):

$$
\Delta\omega\_{\circ} = \omega\_{\circ} - \omega\_{ref,j\circ} \tag{15}
$$

where *ωre f* ,*<sup>j</sup>* is calculated from Equation (4). The global ∆*ω* sent to the PV power plants is the sum of the ∆*ω<sup>j</sup>* of each WPP:

$$
\Delta\omega = \sum\_{j=1}^{4} \Delta\omega\_{j}.\tag{16}
$$

This global ∆*ω* will pass through the *PI* controller, obtaining the value of ∆*V* of Equation (13). An overview of the hybrid wind–PV frequency control approach is shown in Figure 7. By using ∆*ω* as an input, PV power plants modify their generated power according to this rotational speed deviation, improving the VSWTs' rotational speed control and subsequently increasing the inertial control performance of the VSWTs.

**Figure 7.** Hybrid wind–PV frequency control approach.

## **4. Results**

## *4.1. Scenarios under Consideration*

The Ten-Year Network Development Plan (TYNDP) 2020 Scenario Report is used to propose the generation mix under consideration [78]. National trend scenarios are taken into account by the authors, which account for both the supply and demand data collected from the European electricity TSOs. Figure 8 shows the generation mix considered in this paper following [78] for the years 2025 and 2040.

**Figure 8.** Generation mix following the national trends of TYNDP in (**a**) year 2025 and (**b**) year 2040.

Three different electricity demand values (i.e., valley, mean, and peak periods) are also studied based on the Gran Canaria isolated power system (Spain). This power system has been extensively described in [79]. Consequently, by combining the two generation mixes and the three different demand values, six scenarios are under analysis. Power demand is assumed as constant during the simulations, as the timeframe under analysis is 5 min (300 s). As presented in Table 4, this assumption is in line with most TSOs.

**Table 4.** Time interval where demand is considered as constant for different transmission system operators (TSOs).


Wind speed and solar irradiation present some oscillations during the simulation time intervals. These variations cause certain frequency deviations from the nominal value (50 Hz). Real measured values of solar irradiation from a PV power plant in Albacete (Spain) are used for simulation purposes. In addition, synthetic wind speeds estimated from the methodology detailed in [88] are also included in the scenarios. Such synthetic wind speeds have been statistically compared to real measured wind speed series, obtaining similar values.

As mentioned in Section 2, four different WPPs and PV power plants are considered:

$$p\_{\rm WPP} = \sum\_{j=1}^{4} p\_{\rm WPP,j} \,\prime \,\tag{17}$$

$$p\_{PV} = \sum\_{k=1}^{4} p\_{PV,k} \cdot \tag{18}$$

A heterogeneous distribution of wind- and PV-generated power is estimated to give more realistic simulations, considering that WPP<sup>1</sup> and PV<sup>1</sup> account for 50% of the total *pWPP* and *pPV*; WPP<sup>2</sup> and PV<sup>2</sup> are 25% of the total *pWPP* and *pPV*; and WPP<sup>3</sup> =WPP<sup>4</sup> and PV<sup>3</sup> =PV<sup>4</sup> are 12.5% of the total *pWPP* and *pPV*, respectively. The wind speed values *s<sup>w</sup>* and solar irradiation *G* for each WPP and PV power plant are depicted in Figures 9 and 10.

**Figure 9.** Wind speeds considered for the wind power plants (WPPs): (**a**) WPP<sup>1</sup> and WPP<sup>2</sup> as well as (**b**) WPP<sup>3</sup> and WPP<sup>4</sup> .

**Figure 10.** Solar irradiations considered for (**a**) PV<sup>1</sup> and PV<sup>2</sup> as well as (**b**) PV<sup>3</sup> and PV<sup>4</sup> .

#### *4.2. Simulation Results*

The six scenarios presented in Section 4.1 were simulated in a Matlab/Simulink environment. A fixed step with the ode3 (Bogcki-Shampine) solver and a step size of 10−<sup>2</sup> s was used. The Matlab version was r2016a. Four different frequency control strategies were compared:


4. Frequency control is provided by conventional power plants, WPPs with a hidden-inertia emulation technique, and PV power plants with 10% de-loading and a *PI* controller with ∆*ω* of the VSWTs as input (referred to as *PV (ω)*), which is the hybrid wind–PV frequency strategy proposed in this paper.

In each simulation, the authors considered different criteria to compare the frequency control strategies under consideration. Specifically, the parameters taken into account are the following: the minimum and maximum frequency values, the MSE of the grid frequency, the PV-generated energy, the MSE of the thermal and hydro-power plants with respect to their initial assigned power, the minimum and maximum values of the rotational speed of each WPP, and the MSE of the rotational speed of each WPP with respect to their reference value (*ωre f*) for each wind speed. Tables 5–9 show the results of the parameters for the six scenarios simulated under the different frequency control strategies. The PV-generated energy is determined by multiplying the active power *PPV* by the time interval under analysis (5 min).


**Table 5.** Simulation results: frequency, PV electrical energy, and the conventional power plants' mean square error (MSE).

With respect to the frequency deviations, the proposed hybrid wind–PV technique results in the maximum *fmin* and the minimum *fmax* values. This means that smaller frequency deviations are obtained, which is also shown in the MSE. In fact:


Moreover, the authors would like to highlight that similar results are obtained for the three different power demand scenarios in terms of minimum and maximum frequency deviations, together with the MSE of frequency. Consequently, it can be affirmed that this study is scalable to other isolated power systems, and similar results will be obtained independently of the demand as long as the generation mixes are similar to those considered here, which are in line with future European renewable energy integration roadmaps. It is important to remark that, for the 2025 scenario, frequency deviations are nearly within the acceptable range proposed by the ENTSO-E, i.e., ±800 mHz [89], even if the vRESs do not participate in frequency control. However, as vRESs massively replace CPPs, frequency response will be substantially deteriorated. In fact, minimum and maximum frequency values under 45 Hz and over 54 Hz, respectively, are obtained if only CPPs are considered for frequency control response for year 2040. This means that, in the medium term, it is a crucial need that vRESs participate in frequency control to avoid such negative effects due to important (and negative) frequency oscillations.

Focusing on the PV-generated energy, the hybrid wind–PV technique gets the minimum values, mainly due to the initial 10% power de-loading and the action based on the ∆*ω* that comes from VSWTs. Specifically:


Therefore, the hybrid wind–PV strategy implies an additional PV energy reduction of up to 10% in comparison to a conventional PV frequency control strategy. This aspect should be subsequently evaluated by the PV installation operators. Indeed, their benefits should be partially reduced if TSOs/DSOs do not reward them for providing frequency control services.

With regard to the conventional power plants, including any frequency strategy of vRESs reduces the contribution of thermal and hydro-power to the frequency response. The MSE between the power generated by them and their initial assigned value is thus reduced:


The PV(*f*) strategy is then the best technique from the point of view of the MSE of conventional power plants.

Finally, focusing on the rotational speed of WPPs (Tables 6–9):


variations of around 5%–10%. Consequently, the maximum MSE is obtained with the WPP strategy. In some cases, the MSE result is three times higher than the value obtained with the CPP approach. Naturally, the speed deviations with this strategy are the result of both the wind speed changes and the hidden-inertia frequency control approach.


It can be affirmed that the PV(*ω*) technique is the best overall frequency control strategy, as frequency deviations are reduced and the rotational speed variations of the WPPs are closer to their reference values. Nevertheless, this frequency control solution implies certain reductions in PV-generated energy that should be analyzed by the PV installation operators to compensate for possible decreases in benefits. In Figure 11, the frequency evolution, active power of the four generation units, and ∆*ω* input for the *PI* controller of PV power plants for the hybrid wind–PV strategy are presented for the 550 MW load and 2025 year scenario. As can be seen, when CPPs are only considered for frequency control, there are severe frequency oscillations, which are reduced by including vRESs into the frequency control. As was discussed, among the three different vRES frequency control strategies, the proposed hybrid wind–PV technique reduces frequency oscillations, together with lower fluctuations in both thermal and hydro-power generation units. Even though CPPs oscillate less than with the other approaches, they are working further than their initially assigned values, as demonstrated with the MSE of Table 5. Similar output wind power values are determined for the three vRES strategies, as the same *PD* constants are assumed. This response is also less oscillatory than the case in which only CPPs are responding under imbalance conditions. Finally, the PV output power is equal if only the CPP or WPP strategies are considered, as PV active power only changes due to the variations in the solar irradiation *G*. When the de-loading technique is implemented, the PV power plants include a curtailed 10% active power (from 50 to 45 MW, approximately). Moreover, note that considering the ∆*ω* as input significantly reduces the PV-generated power and, consequently, the generated energy (refer to Table 5), but the frequency deviations are also reduced in comparison to the PV(*f*) strategy. In Figures 12 and 13, the active power of the four WPP and PV power plants are shown, respectively.


**Table 6.** Simulation results: rotational speed of WPP<sup>1</sup> .


**Table 7.** Simulation results: rotational speed of WPP2.

**Table 8.** Simulation results: rotational speed of WPP3.


**Table 9.** Simulation results: rotational speed of WPP<sup>4</sup> .


(**c**)

(**a**)

(**d**)

**Figure 11.** Simulation results for 550 MW demand and year 2025: (**a**) frequency evolution, (**b**) rotational speed deviation, (**c**) thermal power, (**d**) hydro-power, (**e**) VSWT power, and (**f**) PV power.

(**a**)

(**b**)

**Figure 12.** WPP power for 550 MW demand and year 2025: (**a**) WPP<sup>1</sup> , (**b**) WPP2, (**c**) WPP3, and (**d**) WPP<sup>4</sup> .

**Figure 13.** PV power for 550 MW demand and year 2025: (**a**) PV<sup>1</sup> , (**b**) PV2, (**c**) PV3, and (**d**) PV<sup>4</sup> .

Figure 14 depicts the frequency evolution, active power of the four generation units, and ∆*ω* input for the *PI* controller of PV power plants for the hybrid wind–PV strategy, corresponding to the 250 MW demand and year 2040 scenario. These results are similar to the previous simulations depicted in Figure 11. In this case, when only CPPs participate in frequency control, the frequency oscillations are higher due to the reduced synchronous inertia of the power system. Grid frequency oscillations are drastically reduced by including vRESs into the frequency response, especially with the proposed hybrid wind–PV solution. Similar active wind power is obtained if they also participate in frequency regulation, providing lower oscillations than the case in which only CPPs are considered. When PV power plants include the de-loading technique, a 10% power reduction is required (from 38 to 34 MW, approximately). When the PV installations receive the ∆*ω* as an input, their active power is reduced accordingly, in line with the generated energy shown in Table 5.

**Figure 14.** Simulation results for 250 MW demand and year 2040: (**a**) frequency evolution, (**b**) rotational speed deviation, (**c**) thermal power, (**d**) hydro-power, **(e)** VSWT power, and (**f**) PV power.

## *4.3. Limitations and Further Work*

This paper tests the benefits of introducing a hybrid wind–PV frequency controller that is able to monitor not only frequency deviations, but also VSWTs' rotational speed deviations from their optimal reference values. This generalization allows conclusions to be drawn in a high casuistry of scenarios, but it makes it necessary to neglect or simplify some aspects without losing accuracy. These simplifications are:


As the results and conclusions of this study can be considered as positive, the next step proposed by the authors is to analyze the impact of establishment in some specific isolated power systems, such as one of the existing ones in the Canary archipelago, in the Aegean islands, or in the Azores archipelago. Since those power systems are well known, it is reasonable to individually introduce each generation unit into the dynamic model. Additionally, current and future renewable energy penetration as well as real wind speed and irradiation data can be used for future simulations. Once the locations of both wind and PV power plants are known, the influence of the communication time on the control actions can be also analyzed. The corresponding time delays with the measurement of errors and the transmission of control signals between wind and PV power plants would then be included in the model. Likewise, this proposal will be able to optimize the power system's operation by associating the control of some specific wind power plants to certain geographically neighboring PV power plants.

### **5. Conclusions**

This paper proposes a hybrid wind–PV frequency control approach for isolated power systems with high vRES integration under variable weather conditions. The proposed controller consists of VSWTs that include a hidden-inertia emulation technique and PV power plants working with a de-load power of 10%. The input of the PV frequency control is the rotational speed deviation of the VSWTs, in contrast to previous studies where the frequency deviation is the input of the de-loaded PV controller. In this way, the PV power plants modify their generated power by following the rotational speed deviation of the VSWTs, reducing the deviation of such rotational speeds, and, consequently, minimizing the frequency deviations under power imbalances. The proposed hybrid frequency approach is compared to three different frequency strategies: (*i*) conventional power plants, (*ii*) conventional power plants and wind power plants, and (*iii*) conventional power plants, wind power plants, and PV power plants with grid frequency as input. The results show that frequency oscillations are drastically reduced by including the proposed hybrid controller. In fact, the mean squared error of frequency variations is reduced by up to 95% in comparison to the case in which only conventional power plants are controlling the frequency variations. Severe reductions are also found when comparing the hybrid wind–PV strategy to the other two strategies in which vRESs participate in frequency control (between 20% and 65%). The rotational speed deviation of the VSWTs

also decreases with the hybrid wind–PV frequency approach, even getting the same mean squared error as when VSWTs do not participate in frequency control. However, the energy generated by the PV power plants decreases by between 20% and 30% when using the proposed approach, and should be subsequently analyzed by the transmission/distribution operators to guarantee some additional benefits for the owners of such generation units. Based on the results, the authors also wish to remark on the importance of vRESs participating in frequency control in the near future in order to avoid significant frequency deviations that will occur if these sources keep replacing conventional power plants without providing any frequency response. Consequently, hybrid frequency control strategies in line with this work should be proposed and analyzed to minimize frequency variations linked to a massive vRES integration.

**Author Contributions:** Conceptualization, A.F.-G. and J.-I.S.; methodology, A.F.-G.; software, A.F.-G. and G.M.-L.; validation, G.M.-L. and J.-I.S.; formal analysis, J.-I.S.; investigation, Á.M.-G. and J.-I.S.; resources, A.F.-G.; data curation, A.F.-G. and J.-I.S.; writing—original draft preparation, A.F.-G. and G.M.-L.; writing—review and editing, J.-I.S. and Á.M.-G.; visualization, Á.M.-G.; supervision, J.-I.S.; project administration, A.F.-G.; funding acquisition, A.F.-G., Á.M.G, J.-I.S., and G.M.-L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was partially supported by 'Ministerio de Educación, Cultura y Deporte' of Spain (ref. FPU16/04282) and by 'Ministerio de Economía y Competitividad' under the project 'Value of pumped-hydro energy storage in isolated power systems with high wind power penetration' of the National Plan for Scientific and Technical Research and Innovation 2013-2016, grant number ENE2016-77951-R.

**Conflicts of Interest:** The authors declare no conflict of interest.

### **Abbreviations**

The following abbreviations and nomenclature are used in this manuscript:



#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **On the Design of an Integrated System for Wave Energy Conversion Purpose with the Reaction Mass on Board**

### **Jinming Wu \* and Zhonghua Ni**

School of Mechanical Engineering, Southeast University, Nanjing 211189, China; nzh2003@seu.edu.cn **\*** Correspondence: jinmingwu@seu.edu.cn;

Received: 12 March 2020; Accepted: 2 April 2020; Published: 3 April 2020

**Abstract:** In this paper, we investigate the design of an integrated system consisting of two non-rigidly connected bodies: A floating buoy and an emerged offshore structure. When waves excite the buoy to oscillate, the relative motion between the two bodies are converted to useful energy through a spring damper system, resulting in wave energy being absorbed. The parameter to design includes the mass and underwater shape of the buoy. The spring stiffness of the power take-off (PTO) system is constrained to be non-negative with the concerns of complexity in implementation and system stability. Results suggest that a larger mass of the buoy is advantageous due to smaller optimal spring stiffness and damping coefficient of the PTO system, more absorbed wave power, and less motion amplitude of the two bodies. A favorable underwater shape of the buoy is characterized by large diameter to draft ratio, with the section profile preferring a circle or square rather than an equilateral triangle. Investigations on the designed buoy in irregular waves show that the integrated system presents its peak power absorption within the common range of energy period, and the motion amplitude of the offshore structure is larger than the wave amplitude in a certain range of sea states.

**Keywords:** integrated system; floating buoy; offshore structure; wave energy converter; non-negative spring stiffness

#### **1. Introduction**

Due to high energy density and excellent predictability, ocean wave energy stands up as a promise substitute of fossil energies, which give rise to environmental issues. Although the first wave energy converter (WEC) had been patented over two centuries ago [1], wave energy conversion technology is still immature and at a pre-commercialized stage. The levelized cost of wave energy is estimated to be \$1.0~1.5/kWh, and not ready for the present power economic request when compared to \$0.07/kWh of onshore wind energy [2]. A possible solution to minimize this discrepancy is to combine WECs with offshore structures, such as navigation aids [3,4], offshore solar [5] and wind [6] electricity generation systems, and even offshore aquaculture devices [2], so that the substructure, mooring system, and electrical grid connections, which account for respectively 20%, 10%, and 6% of the construction cost of an individual WEC [7], can be shared.

In view of wave energy absorption, the aforementioned integrated systems belong to the two-body type, which uses the relative motion between two bodies to harvest energy. A two-body system usually consists of two parts: A floating buoy that moves along with the incoming wave and a reaction mass, which can be submerged in deep water [8,9] or on board [10]. A two-body system with the reaction mass on board has the advantage of lacking submerged moving bodies and actuators, and may be easier for the installation and operation process [10]. Studies on the two-body system with reaction mass on board can be traced back to Parks [11]. French et al. [12] applied phase control to a heaving buoy reacting against an internal mass. Korde [10] revealed that the on board system shows a

power capture width ratio comparable with that of the submerged system over a wide range of wave frequency, demonstrating its potential in wave power absorption. Additionally, Korde [13] proposed a motion compensation system to achieve stationarity of the reaction mass to further increase the absorbed power for the on board system. shows a power capture width ratio comparable with that of the submerged system over a wide range of wave frequency, demonstrating its potential in wave power absorption. Additionally, Korde [13] proposed a motion compensation system to achieve stationarity of the reaction mass to further increase the absorbed power for the on board system.

As concerns how to integrate WECs with offshore structures to form a two-body system, one option is that the offshore structure be on board a floating buoy, which is demonstrated in [14] that a multi-purpose platform is on board floating buoyancy columns with actuators connecting them to absorb wave energy. In order to achieve the best wave power absorption performance, previous studies on two-body systems with reaction mass on board tried to optimize the parameter of the reaction mass [10–13,15]. However, in the integrated system of the offshore structure and the floating buoy, the reaction mass (the offshore structure) is normally provided as fixed parameters ahead of time, hence only the parameter of the floating buoy can be optimized. As concerns how to integrate WECs with offshore structures to form a two‐body system, one option is that the offshore structure be on board a floating buoy, which is demonstrated in [14] that a multi‐purpose platform is on board floating buoyancy columns with actuators connecting them to absorb wave energy. In order to achieve the best wave power absorption performance, previous studies on two‐body systems with reaction mass on board tried to optimize the parameter of the reaction mass [10–13,15]. However, in the integrated system of the offshore structure and the floating buoy, the reaction mass (the offshore structure) is normally provided as fixed parameters ahead of time, hence only the parameter of the floating buoy can be optimized.

The objective of this paper is to design the parameter of the floating buoy with an offshore structure on board to achieve satisfied performance of the integrated system. Here, the parameter of the floating buoy consists of its mass and underwater shape, while the performance of the integrated system includes absorbed wave power and motion amplitude of the two bodies. A 5 MW offshore wind turbine is chosen as the offshore structure for a case study. When WECs are resonant with waves, the absorbed wave power reaches the maximum. However, in order to achieve resonance, the optimal spring stiffness of the power take-off (PTO) system may sometimes be negative, which not only is difficult to implement in reality, but may also shift WECs to unsteady states. Therefore, the spring stiffness of the PTO system is constrained to be non-negative in this paper. The designing process is performed in regular waves, and the designed integrated system is then investigated in irregular waves. The objective of this paper is to design the parameter of the floating buoy with an offshore structure on board to achieve satisfied performance of the integrated system. Here, the parameter of the floating buoy consists of its mass and underwater shape, while the performance of the integrated system includes absorbed wave power and motion amplitude of the two bodies. A 5 MW offshore wind turbine is chosen as the offshore structure for a case study. When WECs are resonant with waves, the absorbed wave power reaches the maximum. However, in order to achieve resonance, the optimal spring stiffness of the power take‐off (PTO) system may sometimes be negative, which not only is difficult to implement in reality, but may also shift WECs to unsteady states. Therefore, the spring stiffness of the PTO system is constrained to be non‐negative in this paper. The designing process is performed in regular waves, and the designed integrated system is then investigated in irregular waves.

#### **2. Methods 2. Methods**

Figure 1 shows the integrated two-body system consisting of the offshore structure and the floating buoy, and with the offshore structure on board the floating buoy. The PTO system makes use of the relative motion between the two bodies to produce useful power, and is simplified as a pair of spring and damper connected in parallel. The stiffness of the spring and the damping coefficient of the damper are denoted as *k<sup>p</sup>* and *cp*, respectively. The mass and displacement of the offshore structure are respectively denoted as *m<sup>s</sup>* and *z<sup>s</sup>* , and that of the buoy as *m<sup>b</sup>* and *z<sup>b</sup>* . For simplicity, the offshore structure and the buoy are subjected to heave motion only. Figure 1 shows the integrated two‐body system consisting of the offshore structure and the floating buoy, and with the offshore structure on board the floating buoy. The PTO system makes use of the relative motion between the two bodies to produce useful power, and is simplified as a pair of spring and damper connected in parallel. The stiffness of the spring and the damping coefficient of the damper are denoted as *kp* and *cp*, respectively. The mass and displacement of the offshore structure are respectively denoted as *ms* and *zs*, and that of the buoy as *mb* and *zb*. For simplicity, the offshore structure and the buoy are subjected to heave motion only.

**Figure 1.** Schematic diagram of the integrated system. **Figure 1.** Schematic diagram of the integrated system.

In regular waves, according to the fundamental investigation on two‐body systems by Falnes [16], the equation of motion of the offshore structure and the buoy can be written as In regular waves, according to the fundamental investigation on two-body systems by Falnes [16], the equation of motion of the offshore structure and the buoy can be written as

$$\begin{aligned} m\_s \ddot{z}\_s + c\_p \left( \dot{z}\_s - \dot{z}\_b \right) + k\_p (z\_s - z\_b) &= 0; \\ \left[ m\_b + m\_{bh}(\omega) \right] \ddot{z}\_b + c\_{bh}(\omega) \dot{z}\_b + k\_{bh} z\_b - c\_p \left( \dot{z}\_s - \dot{z}\_b \right) - k\_p (z\_s - z\_b) &= f\_d; \end{aligned} \tag{1}$$

where *mbh*, *cbh*, and *kbh* denote the added mass, radiation damping coefficient, and hydrostatic stiffness of the buoy, *ω* denotes the wave frequency, and *fd* denotes the excitation force the incoming wave acts on the buoy. The hydrodynamic coefficients and excitation force are calculated numerically by the where *mbh*, *cbh*, and *kbh* denote the added mass, radiation damping coefficient, and hydrostatic stiffness of the buoy, ω denotes the wave frequency, and *f<sup>d</sup>* denotes the excitation force the incoming wave acts

ANSYS AQWA software [17]. The hydrostatic stiffness of the buoy in heave is calculated as

on the buoy. The hydrodynamic coefficients and excitation force are calculated numerically by the ANSYS AQWA software [17]. The hydrostatic stiffness of the buoy in heave is calculated as

$$k\_{bh} = \rho\_{\mathcal{S}} \mathbf{S}\_{\prime} \tag{2}$$

where ρ is the water density, *g* is the gravity acceleration, and *S* is the area of the waterplane. By applying the form of complex amplitude [18,19], which is denoted by a symbol 'ˆ' on top of the variable, Equation (1) can be written as

$$\begin{aligned} -\omega^2 m\_s \hat{\mathbf{z}}\_s + \left( \mathbf{k}\_p + \mathbf{i}\omega c\_p \right) (\hat{\mathbf{z}}\_s - \hat{\mathbf{z}}\_b) &= 0; \\ -\omega^2 [m\_b + m\_{bh}(\omega)] \hat{\mathbf{z}}\_b + [\mathbf{k}\_{bh} + \mathbf{i}\omega c\_{bh}(\omega)] \hat{\mathbf{z}}\_b - \left( \mathbf{k}\_p + \mathbf{i}\omega c\_p \right) (\hat{\mathbf{z}}\_s - \hat{\mathbf{z}}\_b) &= \hat{f}\_d; \end{aligned} \tag{3}$$

Equation (3) can be transferred to the matrix form as

$$
\begin{bmatrix}
\end{bmatrix}
\begin{bmatrix}
\mathfrak{L}\_s \\
\mathfrak{L}\_b
\end{bmatrix} = \begin{bmatrix}
0 \\
f\_d
\end{bmatrix}.
\tag{4}
$$

Here, we denote the impedance of the offshore structure, the buoy, and the PTO system as *Z<sup>s</sup>* , *Z<sup>b</sup>* , and *Zp*, and their real and imaginary parts are denoted by *Zsr*, i*Zsi*, *Zbr*, i*Zbi*, *Zpr*, and i*Zpi*, respectively. The impedance is defined as

$$\begin{aligned} Z\_{\rm sr} &= -\omega^2 m\_{\rm s}; \\ Z\_{\rm br} &= -\omega^2 [m\_{\rm b} + m\_{\rm bh}(\omega)] + k\_{\rm bh}; \\ Z\_{\rm bi} &= \omega c\_{\rm bh}(\omega); \\ Z\_{\rm pr} &= k\_{\rm p}; \\ Z\_{\rm pl} &= \omega c\_{\rm p}. \end{aligned} \tag{5}$$

Since the offshore structure is out of water, *Zsi* = 0. Then, Equation (4) can be written as

$$
\begin{bmatrix} Z\_s + Z\_p & -Z\_p \\ -Z\_p & Z\_b + Z\_p \end{bmatrix} \begin{bmatrix} \hat{z}\_s \\ \hat{z}\_b \end{bmatrix} = \begin{bmatrix} 0 \\ \hat{f}\_d \end{bmatrix}.\tag{6}
$$

By finding the inverse of the matrix in the left term of Equation (6) and matrix multiplication, it gives

$$
\begin{bmatrix}
\hat{\mathbf{z}}\_s \\
\hat{\mathbf{z}}\_b
\end{bmatrix} = \frac{1}{Z\_s \mathbf{Z}\_b + (\mathbf{Z}\_s + \mathbf{Z}\_b) \mathbf{Z}\_p} \begin{bmatrix}
\mathbf{Z}\_p \hat{f}\_d \\
\mathbf{(Z}\_s + \mathbf{Z}\_p) \hat{f}\_d
\end{bmatrix} \tag{7}
$$

For a two-body system, the time-averaged absorbed power can be calculated by [20]

$$P = \frac{\omega^2 c\_p}{2} |\mathfrak{E}\_s - \mathfrak{E}\_b|^2 \tag{8}$$

It can be obtained from Equation (7) that

$$
\hat{\mathbf{z}}\_s - \hat{\mathbf{z}}\_b = \frac{-\hat{f}\_d}{\mathbf{Z}\_b + \left(1 + \frac{\mathbf{Z}\_b}{\mathbf{Z}\_s}\right)\mathbf{Z}\_p}.\tag{9}
$$

By denoting

$$\mathbf{G} = 1 + \frac{\mathbf{Z}\_b}{\mathbf{Z}\_s} \,\tag{10}$$

Equation (9) can be written as

$$
\mathfrak{E}\_s - \mathfrak{E}\_b = \frac{-\hat{f}\_d}{Z\_b + GZ\_p}.\tag{11}
$$

Here, we denote the real and imaginary part of *G* as *G<sup>r</sup>* and i*G<sup>i</sup>* , respectively. Then, we have

$$\begin{array}{c} \mathbf{G}\_{\mathbf{r}} = \mathbf{1} + \frac{Z\_{\mathbf{lr}}}{Z\_{\mathbf{sr}}};\\ \mathbf{G}\_{\mathbf{i}} = \frac{Z\_{\mathbf{il}}}{Z\_{\mathbf{sr}}}; \end{array} \tag{12}$$

Inserting Equation (11) to Equation (8) gives

$$P = \frac{\omega^2 \mathfrak{c}\_p}{2} \frac{f\_d f\_d^\*}{\left(Z\_b + G Z\_p\right)\left(Z\_b^\* + G^\* Z\_p^\*\right)}.\tag{13}$$

Decomposition of the complex parameters to real and imaginary parts makes Equation (13) be written as

$$P = \frac{\omega}{2} \frac{\left| \hat{f}\_d \right|^2}{|\mathbf{G}|^2 \mathbf{Z}\_{p\text{i}} + 2(\mathbf{G}\_{\text{f}} \mathbf{Z}\_{\text{bi}} - \mathbf{G}\_{\text{i}} \mathbf{Z}\_{\text{b}\text{r}}) + \frac{|\mathbf{G}|^2 \mathbf{Z}\_{p\text{r}}^2 + 2(\mathbf{G}\_{\text{f}} \mathbf{Z}\_{\text{b}\text{r}} + \mathbf{G}\_{\text{i}} \mathbf{Z}\_{\text{bi}}) \mathbf{Z}\_{\text{pr}} + \left| \mathbf{Z}\_{\text{b}} \right|^2}{\mathbf{Z}\_{\text{p}\text{i}}} \tag{14}$$

When there is no constraint on the parameters of the PTO system, it can be seen that when

$$Z\_{pr} = Z\_{pro} = -\frac{\mathbf{G}\_r \mathbf{Z}\_{br} + \mathbf{G}\_i \mathbf{Z}\_{bi}}{|\mathbf{G}|^2} \tag{15}$$

And

$$Z\_{\rm pi} = Z\_{\rm pio} = \frac{\mathbf{G}\_r \mathbf{Z}\_{\rm bi} - \mathbf{G}\_i \mathbf{Z}\_{\rm br}}{\left| \mathbf{G} \right|^2} \,, \tag{16}$$

the absorbed power reaches the maximum

$$P\_{\text{max}} = \frac{\omega}{8} \frac{\left|\hat{f}\_d\right|^2}{\text{G}\_r \text{Z}\_{bi} - \text{G}\_i \text{Z}\_{br}}.\tag{17}$$

Equation (5) and Equation (12) gives

$$\mathbf{G}\_{\rm r}\mathbf{Z}\_{\rm bi} - \mathbf{G}\_{\rm i}\mathbf{Z}\_{\rm br} = \left(\mathbf{1} + \frac{\mathbf{Z}\_{\rm br}}{\mathbf{Z}\_{\rm sr}}\right)\mathbf{Z}\_{\rm bi} - \frac{\mathbf{Z}\_{\rm bi}}{\mathbf{Z}\_{\rm sr}}\mathbf{Z}\_{\rm br} = \mathbf{Z}\_{\rm bi}.\tag{18}$$

From Equation (15) to Equation (18) and the relation between *Zpr*, *Zpi*, and *kp*, *c<sup>p</sup>* as shown in Equation (5), we can obtain the optimal values of *kp*, *cp*, and maximum absorbed power as

$$k\_{po} = -\frac{\mathbf{G}\_r \mathbf{Z}\_{br} + \mathbf{G}\_i \mathbf{Z}\_{bi}}{\left| \mathbf{G} \right|^2} \tag{19}$$

$$c\_{po} = \frac{c\_{bh}}{|\mathcal{G}|^2};\tag{20}$$

$$P\_{\text{max}} = \frac{\left| \hat{f}\_d \right|^2}{8c\_{bbl}}.\tag{21}$$

It agrees with the finding of Korde [10] that the maximum absorbed power only depends on the excitation force and radiation damping coefficient of the buoy for an on board system. It is interesting to find that although the integrated system oscillates in two degrees of freedom, it behaves as a single-degree-of-freedom system in terms of power absorption.

However, in some cases, Equation (19) gives the optimal spring stiffness of the PTO system a negative value, which is not only difficult to implement in reality, but also may offset the system to unsteadiness. In this paper, we constrain the spring stiffness of the PTO system to be non-negative. Taking into consideration this constraint, we update the optimal spring stiffness and damping coefficient of the PTO system and the maximum absorbed power of the integrated system as

$$k\_{po} = \max\{-\frac{G\_r Z\_{br} + G\_i Z\_{bi}}{\left|\mathcal{G}\right|^2}, 0\},\tag{22}$$

$$c\_{po} = \begin{cases} \begin{array}{c} \frac{c\_{bh}}{|G|^2} \text{, if } k\_{po} > 0\\ \left| \frac{Z\_b}{\alpha G} \right| \text{, if } k\_{po} = 0 \end{array} \end{cases} \tag{23}$$

And

$$P\_{\text{max}} = \begin{cases} \left| \dot{f}\_d \right|^2 & \text{if } k\_{po} > 0\\ \frac{\omega}{4} \frac{|\dot{f}\_d|^2}{|\text{Cl}| |Z\_b| + Z\_{bi}} & \text{if } k\_{po} = 0 \end{cases} \tag{24}$$

In this paper, a 5 MW offshore wind turbine, which is used in the WindFloat concept [21] proposed by the Principle Power Inc. for offshore wind farm application, is chosen as the offshore structure for a case study. The offshore structure is restricted to oscillate in heave mode. The total mass of the offshore structure is found to be 854 mt [21]. The water depth is considered in offshore scenario to be *h* = 100 m, and the wave amplitude, which is denoted by *A*, is fixed at 1 m for regular waves.

#### **3. E**ff**ect of the Constraint on The Spring Sti**ff**ness of the PTO System**

Since the spring stiffness of the PTO system is constrained to be non-negative, which is not considered in previous literature for the on board two-body system, we investigate the effect of the constraint first. From Equation (19), we find that it is the numerator of the right term that causes the unconstrained optimal spring stiffness of the PTO system to be negative since the denominator is absolutely non-negative. Here we denote

$$H = -(\mathbf{G}\_r \mathbf{Z}\_{br} + \mathbf{G}\_i \mathbf{Z}\_{bi}).\tag{25}$$

Inserting Equation (12) to Equation (25) gives

$$H = -\frac{1}{Z\_{sr}} \left( Z\_{br}^2 + Z\_{sr} Z\_{br} + Z\_{\text{bi}}^2 \right). \tag{26}$$

According to the characteristic of parabolic functions, we find that *H* will be negative only when

$$
\hbar \mathbf{Z}\_{\rm sr}^2 - 4\mathbf{Z}\_{\rm bi}^2 \ge 0 \tag{27}
$$

i.e.,

*Zbi* |*Zsr*| ≤ 1 2 , (28)

and

$$\frac{Z\_{br}}{|Z\_{sr}|} \rightarrow \frac{1}{2}, \frac{Z\_{br}}{|Z\_{sr}|} \in \left[\frac{1}{2} - \sqrt{\frac{1}{4} - \left(\frac{Z\_{bi}}{|Z\_{sr}|}\right)^2}, \frac{1}{2} + \sqrt{\frac{1}{4} - \left(\frac{Z\_{bi}}{|Z\_{sr}|}\right)^2}\right] \tag{29}$$

Here, *Zbr*/|*Zsr*| and *Zbi*/|*Zsr*| can be recognized as the real and imaginary part of the non-dimensional buoy impedance, respectively. It can be seen that a negative unconstrained optimal spring stiffness of the PTO system results from the imaginary part of the non-dimensional buoy impedance being smaller than 0.5 and the real part approaching 0.5, and the extent of approaching is determined from the imaginary part. To avoid negative values, the spring stiffness of the PTO system should be constrained.

We selected a truncated circular cylinder as the reference underwater shape of the floating buoy as shown in Figure 2, where d denotes the draft and D for the diameter. The mass of the buoy is set as

*m<sup>b</sup>* = µ*m<sup>s</sup>* , in which µ is defined as the mass ratio. As for a simple case study here, we set the diameter to draft ratio γ = *D*/*d* = 2, and µ = 0.5. Since the displacement of the buoy should be equal to the mass of both the offshore structure and the buoy, the dimension of the underwater part of the buoy can be obtained based on the mass of the offshore structure as *Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 16 of both the offshore structure and the buoy, the dimension of the underwater part of the buoy can be *Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 16 of both the offshore structure and the buoy, the dimension of the underwater part of the buoy can be

$$d = \sqrt[3]{\frac{4(1+\mu)m\_s}{\pi\rho\gamma^2}}\tag{30}$$

$$D = d\gamma \tag{31}$$

**Figure 2.** Diagram of the truncated circular cylinder. **Figure 2.** Diagram of the truncated circular cylinder. **Figure 2.** Diagram of the truncated circular cylinder.

Figure 3 shows the hydrodynamic coefficients and non‐dimensional impedance of the buoy as a function of wave frequency. The hydrodynamic coefficients are calculated by the boundary element method (BEM) using the ANSYS AQWA software. Based on the hydrodynamic coefficients as shown in Figure 3(a), the non‐dimensional impedance of the buoy can be easily obtained as shown in Figure 3(b). We find that, within the investigated range of wave frequency, the imaginary part of the non‐ dimensional buoy impedance is smaller than 0.5, which implies that the unconstrained optimal spring stiffness of the PTO system is possible to be negative. Further, the real part of the non‐ dimensional buoy impedance decreases monotonously with wave frequency, and crosses 0.5 at *ω* = 1.06 rad/s. Therefore, within the wave frequency range determined from Equation (29), the unconstrained optimal spring stiffness of the PTO system will be negative. Figure 3 shows the hydrodynamic coefficients and non-dimensional impedance of the buoy as a function of wave frequency. The hydrodynamic coefficients are calculated by the boundary element method (BEM) using the ANSYS AQWA software. Based on the hydrodynamic coefficients as shown in Figure 3a, the non-dimensional impedance of the buoy can be easily obtained as shown in Figure 3b. We find that, within the investigated range of wave frequency, the imaginary part of the non-dimensional buoy impedance is smaller than 0.5, which implies that the unconstrained optimal spring stiffness of the PTO system is possible to be negative. Further, the real part of the non-dimensional buoy impedance decreases monotonously with wave frequency, and crosses 0.5 at ω = 1.06 rad/s. Therefore, within the wave frequency range determined from Equation (29), the unconstrained optimal spring stiffness of the PTO system will be negative. Figure 3 shows the hydrodynamic coefficients and non‐dimensional impedance of the buoy as a function of wave frequency. The hydrodynamic coefficients are calculated by the boundary element method (BEM) using the ANSYS AQWA software. Based on the hydrodynamic coefficients as shown in Figure 3(a), the non‐dimensional impedance of the buoy can be easily obtained as shown in Figure 3(b). We find that, within the investigated range of wave frequency, the imaginary part of the non‐ dimensional buoy impedance is smaller than 0.5, which implies that the unconstrained optimal spring stiffness of the PTO system is possible to be negative. Further, the real part of the non‐ dimensional buoy impedance decreases monotonously with wave frequency, and crosses 0.5 at *ω* = 1.06 rad/s. Therefore, within the wave frequency range determined from Equation (29), the

unconstrained optimal spring stiffness of the PTO system will be negative.

**Figure 3.** Hydrodynamic coefficients and non‐dimensional impedance of the buoy as a function of wave frequency: (a) hydrodynamic coefficients; (b) non‐dimensional impedance. **Figure 3.** Hydrodynamic coefficients and non‐dimensional impedance of the buoy as a function of wave frequency: (a) hydrodynamic coefficients; (b) non‐dimensional impedance. **Figure 3.** Hydrodynamic coefficients and non-dimensional impedance of the buoy as a function of wave frequency: (**a**) hydrodynamic coefficients; (**b**) non-dimensional impedance.

Figure 4 shows the optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of wave frequency. It can be clearly seen that, within the wave frequency range from 0.94 rad/s to 1.22 rad/s, the constrained optimal spring stiffness of the PTO system is compulsorily tuned to 0 since the unconstrained optimal value is negative, as inferred from Figure 3. Another obvious finding in Figure 4(a) is that, around *ω* = 0.93 rad/s, both the optimal spring stiffness and damping coefficient of the PTO system are impractically large, which can be attributed to the norm of *G*. The definition of *G* can be found in Equation (10) and Equation (12). The imaginary Figure 4 shows the optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of wave frequency. It can be clearly seen that, within the wave frequency range from 0.94 rad/s to 1.22 rad/s, the constrained optimal spring stiffness of the PTO system is compulsorily tuned to 0 since the unconstrained optimal value is negative, as inferred from Figure 3. Another obvious finding in Figure 4(a) is that, around *ω* = 0.93 rad/s, both the optimal spring stiffness and damping coefficient of the PTO system are impractically large, which can be attributed to the norm of *G*. The definition of *G* can be found in Equation (10) and Equation (12). The imaginary Figure 4 shows the optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of wave frequency. It can be clearly seen that, within the wave frequency range from 0.94 rad/s to 1.22 rad/s, the constrained optimal spring stiffness of the PTO system is compulsorily tuned to 0 since the unconstrained optimal value is negative, as inferred from Figure 3. Another obvious finding in Figure 4a is that, around ω = 0.93 rad/s, both the optimal spring stiffness and damping coefficient of the PTO system are impractically large, which can be attributed to the norm of *G*. The definition of *G* can be found in Equation (10) and Equation (12). The imaginary part of *G* is

part of *G* is the negative of the imaginary part of the non‐dimensional buoy impedance, while the

part of *G* is the negative of the imaginary part of the non‐dimensional buoy impedance, while the

magnitude than the real part, indicating that the norm of *G* is dominated by the real part. Around the

the negative of the imaginary part of the non-dimensional buoy impedance, while the real part of *G* is the negative of the real part of the non-dimensional buoy impedance plus one. From Figure 3, we find that the imaginary part of the non-dimensional buoy impedance is much smaller in magnitude than the real part, indicating that the norm of *G* is dominated by the real part. Around the wave frequency of 0.93 rad/s, the real part of *G* is crossing the zero point, which will result in impractically large values of *kpo* and *cpo* since the norm of *G* is in the denominator of both. It is observed in Figure 4b that the maximum absorbed power is less than the unconstrained case, which is denoted as *Pmaxunconstrained*, within the frequency range when the constrained optimal spring stiffness of the PTO system is tuned to zero. This is as expected since the global extreme point of the absorbed power cannot be reached due to the constraint. This suggests that, to absorb the most wave power, the integrated two-body system should be carefully designed so that the wave frequency range where the constrained optimal spring stiffness of the PTO system is tuned to zero stays far away from the peak frequency of a spectrum, around which wave energy is concentrated. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 7 of 16 wave frequency of 0.93 rad/s, the real part of *G* is crossing the zero point, which will result in impractically large values of *kpo* and *cpo* since the norm of *G* is in the denominator of both. It is observed in Figure 4(b) that the maximum absorbed power is less than the unconstrained case, which is denoted as *Pmaxunconstrained*, within the frequency range when the constrained optimal spring stiffness of the PTO system is tuned to zero. This is as expected since the global extreme point of the absorbed power cannot be reached due to the constraint. This suggests that, to absorb the most wave power, the integrated two‐body system should be carefully designed so that the wave frequency range where the constrained optimal spring stiffness of the PTO system is tuned to zero stays far away from the peak frequency of a spectrum, around which wave energy is concentrated.

**Figure 4.** Optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of wave frequency: (a) Optimal parameters of the PTO system; (b) maximum absorbed power. **Figure 4.** Optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of wave frequency: (**a**) Optimal parameters of the PTO system; (**b**) maximum absorbed power.

#### **4. Design of the Integrated System 4. Design of the Integrated System**

For a given offshore structure to achieve satisfied performance of the integrated system, parameters of the floating buoy should be properly designed. In this paper, parameters of the buoy designed include the mass and the underwater shape, while performance of the integrated system is focused on the absorbed wave power and motion amplitude of the two bodies. For a given offshore structure to achieve satisfied performance of the integrated system, parameters of the floating buoy should be properly designed. In this paper, parameters of the buoy designed include the mass and the underwater shape, while performance of the integrated system is focused on the absorbed wave power and motion amplitude of the two bodies.

#### *4.1. Design of the Mass of the Buoy 4.1. Design of the Mass of the Buoy*

In order to investigate the influence of the mass of the buoy on the performance of the integrated system, we fixed the underwater shape of the buoy while varying the mass. However, when varying the mass of the buoy, to support the varying total mass of the integrated system, the underwater shape will be changed accordingly to provide adequate displacement. Therefore, we take an alternative way to set the key feature of the underwater shape of the buoy to be fixed, and the diameter to draft ratio is chosen. Here, the underwater shape of the buoy employs the truncated circular cylinder, as shown in Figure 2, and the diameter to draft ratio is set to 2. It is assumed that the density of the buoy can be arbitrarily adjusted to fulfill the relation between the volume and mass In order to investigate the influence of the mass of the buoy on the performance of the integrated system, we fixed the underwater shape of the buoy while varying the mass. However, when varying the mass of the buoy, to support the varying total mass of the integrated system, the underwater shape will be changed accordingly to provide adequate displacement. Therefore, we take an alternative way to set the key feature of the underwater shape of the buoy to be fixed, and the diameter to draft ratio is chosen. Here, the underwater shape of the buoy employs the truncated circular cylinder, as shown in Figure 2, and the diameter to draft ratio is set to 2. It is assumed that the density of the buoy can be arbitrarily adjusted to fulfill the relation between the volume and mass of the buoy.

of the buoy. Figure 5 shows the optimal parameters of the PTO system as a function of wave frequency at different mass ratios. A general finding is that the maximum values of *kpo* and *cpo* decrease with the Figure 5 shows the optimal parameters of the PTO system as a function of wave frequency at different mass ratios. A general finding is that the maximum values of *kpo* and *cpo* decrease with the mass ratio, and the range of wave frequency when *kpo* is tuned to zero becomes narrower as well. Both

mass ratio, and the range of wave frequency when *kpo* is tuned to zero becomes narrower as well. Both suggest that the mass of the buoy should be larger to achieve better design of the PTO system, since suggest that the mass of the buoy should be larger to achieve better design of the PTO system, since smaller spring stiffness and damping coefficient seem easier to implement in engineering. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 16 *Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 16

**Figure 5.** Optimal parameters of the PTO system as a function of wave frequency at different mass ratios: (a) Optimal spring stiffness; (b) Optimal damping coefficient. **Figure 5.** Optimal parameters of the PTO system as a function of wave frequency at different mass ratios: (**a**) Optimal spring stiffness; (**b**) Optimal damping coefficient. **Figure 5.** Optimal parameters of the PTO system as a function of wave frequency at different mass ratios: (a) Optimal spring stiffness; (b) Optimal damping coefficient.

Figure 6 shows the maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios. A narrower range of wave frequency, where power absorption ability degrades due to the constraint on the spring stiffness of the PTO system, is observed at a larger mass ratio. Additionally, the absorbed power increases slightly with the mass ratio when the wave frequency is large. The above findings indicate that an integrated system with a larger mass ratio shows better power absorption performance. Figure 6 shows the maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios. A narrower range of wave frequency, where power absorption ability degrades due to the constraint on the spring stiffness of the PTO system, is observed at a larger mass ratio. Additionally, the absorbed power increases slightly with the mass ratio when the wave frequency is large. The above findings indicate that an integrated system with a larger mass ratio shows better power absorption performance. Figure 6 shows the maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios. A narrower range of wave frequency, where power absorption ability degrades due to the constraint on the spring stiffness of the PTO system, is observed at a larger mass ratio. Additionally, the absorbed power increases slightly with the mass ratio when the wave frequency is large. The above findings indicate that an integrated system with a larger mass ratio shows better power absorption performance.

**Figure 6.** Maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios. **Figure 6.** Maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios. **Figure 6.** Maximum absorbed power of the integrated system as a function of wave frequency at different mass ratios.

Equation (7) gives the response of the offshore structure and the floating buoy respectively as Equation (7) gives the response of the offshore structure and the floating buoy respectively as Equation (7) gives the response of the offshore structure and the floating buoy respectively as

$$\mathfrak{E}\_{\rm s} = \frac{Z\_p \hat{f}\_d}{Z\_{\rm s} Z\_b + (Z\_{\rm s} + Z\_b) Z\_p} \tag{32}$$

$$\hat{\mathbf{z}}\_b = \frac{\left(\mathbf{Z}\_\mathbf{s} + \mathbf{Z}\_p\right)\hat{\mathbf{f}}\_d}{\mathbf{Z}\_\mathbf{s}\mathbf{Z}\_b + (\mathbf{Z}\_\mathbf{s} + \mathbf{Z}\_b)\mathbf{Z}\_p}.\tag{33}$$

Figure 7 shows the non‐dimensional motion amplitude of the integrated system, which is defined by the motion amplitude divided by the wave amplitude imitating the definition of RAO as in [22], i.e., *zsn*=|*z*ො*s*|/*A* and *zbn*=|*z*ො*b*|/*A*, as a function of wave frequency at different mass ratios. The Figure 7 shows the non‐dimensional motion amplitude of the integrated system, which is defined by the motion amplitude divided by the wave amplitude imitating the definition of RAO as in [22], i.e., *zsn*=|*z*ො*s*|/*A* and *zbn*=|*z*ො*b*|/*A*, as a function of wave frequency at different mass ratios. The Figure 7 shows the non-dimensional motion amplitude of the integrated system, which is defined by the motion amplitude divided by the wave amplitude imitating the definition of RAO as in [22],i.e., *<sup>z</sup>sn* <sup>=</sup> <sup>|</sup>*z*ˆ*s*|/*<sup>A</sup>* and *<sup>z</sup>bn* <sup>=</sup> <sup>|</sup>*z*ˆ*<sup>b</sup>* |/*A*, as a function of wave frequency at different mass ratios. The

non‐dimensional motion amplitude of both the offshore structure and the floating buoy generally

non‐dimensional motion amplitude of both the offshore structure and the floating buoy generally

shown as the obvious reduction of the motion amplitude at corresponding wave frequencies, while

shown as the obvious reduction of the motion amplitude at corresponding wave frequencies, while

non-dimensional motion amplitude of both the offshore structure and the floating buoy generally shows a tendency of fall first and then rise with the wave frequency. However, the motion amplitude of the buoy seems to be more sensitive to the constraint on the spring stiffness of the PTO system shown as the obvious reduction of the motion amplitude at corresponding wave frequencies, while the motion amplitude of the offshore structure seems to be only slightly influenced. A larger mass ratio generally shifts both curves of zsn and zbn to the left. Within the common range of wave period in the ocean, i.e., from 6 s to 10 s (wave frequency from 1.05 rad/s to 0.63 rad/s), a larger mass ratio results in smaller motion amplitude of both the offshore structure and the floating buoy, which is preferred since smaller motion amplitude will not only reduce the influence on the function of the offshore structure due to the motion of itself, but also reduce viscous energy loss due to the motion of the buoy. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 9 of 16 the motion amplitude of the offshore structure seems to be only slightly influenced. A larger mass ratio generally shifts both curves of zsn and zbn to the left. Within the common range of wave period in the ocean, i.e., from 6 s to 10 s (wave frequency from 1.05 rad/s to 0.63 rad/s), a larger mass ratio results in smaller motion amplitude of both the offshore structure and the floating buoy, which is preferred since smaller motion amplitude will not only reduce the influence on the function of the offshore structure due to the motion of itself, but also reduce viscous energy loss due to the motion of the buoy.

**Figure 7.** Non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different mass ratios: (a) For the offshore structure; (b) for the floating buoy. **Figure 7.** Non-dimensional motion amplitude of the integrated system as a function of wave frequency at different mass ratios: (**a**) For the offshore structure; (**b**) for the floating buoy.

Combined with the findings in above paragraphs, we conclude that a larger mass ratio is preferred for the integrated system due to smaller optimal parameters of the PTO system, more absorbed wave power, and less motion amplitude of the two bodies. However, in practical applications, the mass ratio should not be too large since it will introduce additional structural costs. Therefore, a balance between the performance of the integrated system and the structural cost should be made so that the overall profit reaches the maximum. In the following section, we set the mass ratio to be one for the designing process of the underwater shape of the buoy. Combined with the findings in above paragraphs, we conclude that a larger mass ratio is preferred for the integrated system due to smaller optimal parameters of the PTO system, more absorbed wave power, and less motion amplitude of the two bodies. However, in practical applications, the mass ratioshould not be too large since it will introduce additional structural costs. Therefore, a balance between the performance of the integrated system and the structural cost should be made so that the overall profit reaches the maximum. In the following section, we set the mass ratio to be one for the designing process of the underwater shape of the buoy.

#### *4.2. Design of the Underwater Shape of the Buoy 4.2. Design of the Underwater Shape of the Buoy*

in the following.

First of all, we focus on the influence of the diameter to draft ratio γ of the underwater shape of the buoy. Figure 8 shows the maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different diameter to draft ratios. At *γ* = 1, the absorbed power is strongly influenced by the constraint on the spring stiffness of the PTO system, while at *γ* = 4, the influence almost vanishes. Although smaller motion amplitude of the offshore structure from 0.76 rad/s to 1 rad/s is observed at *γ* = 1, it is at the expense of the maximum absorbed power being significantly decreased due to the constraint. An overall comparison suggests that a larger diameter to draft ratio results in both more absorbed power and reasonable small motion amplitude of the two bodies. Therefore, we set the diameter to draft ratio to be 4 for further studies First of all, we focus on the influence of the diameter to draft ratio γ of the underwater shape of the buoy. Figure 8 shows the maximum absorbed power and non-dimensional motion amplitude of the integrated system as a function of wave frequency at different diameter to draft ratios. At γ = 1, the absorbed power is strongly influenced by the constraint on the spring stiffness of the PTO system, while at γ = 4, the influence almost vanishes. Although smaller motion amplitude of the offshore structure from 0.76 rad/s to 1 rad/s is observed at γ = 1, it is at the expense of the maximum absorbed power being significantly decreased due to the constraint. An overall comparison suggests that a larger diameter to draft ratio results in both more absorbed power and reasonable small motion amplitude of the two bodies. Therefore, we set the diameter to draft ratio to be 4 for further studies in the following.

= 1

1

integrated system.

integrated system.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 10 of 16

10

10

**Figure 8.** Maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different diameter to draft ratios: (a) Maximum absorbed power; (b) non‐dimensional motion amplitude of the offshore structure; (c) non‐dimensional motion amplitude of the floating buoy. **Figure 8.** Maximum absorbed power and non-dimensional motion amplitude of the integrated system as a function of wave frequency at different diameter to draft ratios: (**a**) Maximum absorbed power; (**b**) non-dimensional motion amplitude of the offshore structure; (**c**) non-dimensional motion amplitude of the floating buoy. (**a**) (**b**) (**c**) **Figure 8.** Maximum absorbed power and non‐dimensional motion amplitude of the integrated system 0.5 <sup>1</sup> 1.5 <sup>0</sup> , rad/s 0.5 <sup>1</sup> 1.5 <sup>0</sup> , rad/s 0.5 <sup>1</sup> 1.5 <sup>0</sup> , rad/s (b) non‐dimensional motion amplitude of the offshore structure; (c) non‐dimensional motion amplitude of the floating buoy. Secondly, we focus on the influence of non‐axisymmetry of the section of the underwater shape.

as a function of wave frequency at different diameter to draft ratios: (a) Maximum absorbed power;

as a function of wave frequency at different diameter to draft ratios: (a) Maximum absorbed power;

Secondly, we focus on the influence of non‐axisymmetry of the section of the underwater shape. Two representative non‐axisymmetric section profiles, i.e., the equilateral triangle and the square as shown in Figure 9, are analyzed as opposed to the circle with axisymmetric section. In Figure 9, D represents the side length of the triangle and the square. Due to non‐axisymmetry, the performance of the integrated system varies at different wave directions, which are defined by the wave angle *β* in Figure 9. Here, we take two typical wave directions for both non‐axisymmetric buoys into consideration, i.e., 0° and 60° for the triangle, and 0° and 45° for the square. Figure 10 shows the maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles. An interesting finding is that non‐ axisymmetry of the section does not bring about significant influence on the performance of the integrated system, and only a slight increase of the motion amplitude of the buoy is observed for the triangular section. Here, the circular or square sections are recognized as the better choice for the Secondly, we focus on the influence of non-axisymmetry of the section of the underwater shape. Two representative non-axisymmetric section profiles, i.e., the equilateral triangle and the square as shown in Figure 9, are analyzed as opposed to the circle with axisymmetric section. In Figure9, D represents the side length of the triangle and the square. Due to non-axisymmetry, the performanceof the integrated system varies at different wave directions, which are defined by the wave angle<sup>β</sup> in Figure 9. Here, we take two typical wave directions for both non-axisymmetric buoys intoconsideration, i.e., 0◦ and 60◦ for the triangle, and 0◦ and 45◦ for the square. Figure 10shows themaximum absorbed power and non-dimensional motion amplitude of the integrated system as afunction of wave frequency at different section profiles. An interesting finding is that non-axisymmetry of the section does not bring about significant influence on the performance of the integrated system, and only a slight increase of the motion amplitude of the buoy is observed for the triangular section. Here, the circular or square sections are recognized as the better choice for the integrated system. (b) non‐dimensional motion amplitude of the offshore structure; (c) non‐dimensional motion amplitude of the floating buoy. Secondly, we focus on the influence of non‐axisymmetry of the section of the underwater shape. Two representative non‐axisymmetric section profiles, i.e., the equilateral triangle and the square as shown in Figure 9, are analyzed as opposed to the circle with axisymmetric section. In Figure 9, D represents the side length of the triangle and the square. Due to non‐axisymmetry, the performance of the integrated system varies at different wave directions, which are defined by the wave angle *β* in Figure 9. Here, we take two typical wave directions for both non‐axisymmetric buoys into consideration, i.e., 0° and 60° for the triangle, and 0° and 45° for the square. Figure 10 shows the maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles. An interesting finding is that non‐ axisymmetry of the section does not bring about significant influence on the performance of the integrated system, and only a slight increase of the motion amplitude of the buoy is observed for the triangular section. Here, the circular or square sections are recognized as the better choice for the Two representative non‐axisymmetric section profiles, i.e., the equilateral triangle and the square as shown in Figure 9, are analyzed as opposed to the circle with axisymmetric section. In Figure 9, D represents the side length of the triangle and the square. Due to non‐axisymmetry, the performance of the integrated system varies at different wave directions, which are defined by the wave angle *β* in Figure 9. Here, we take two typical wave directions for both non‐axisymmetric buoys into consideration, i.e., 0° and 60° for the triangle, and 0° and 45° for the square. Figure 10 shows the maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles. An interesting finding is that non‐ axisymmetry of the section does not bring about significant influence on the performance of the integrated system, and only a slight increase of the motion amplitude of the buoy is observed for the triangular section. Here, the circular or square sections are recognized as the better choice for the integrated system.

(**a**) (**b**) **Figure 9.** Two representative non‐axisymmetric section profiles: (a) Equilateral triangular section; (b) **Figure 9.** Two representative non‐axisymmetric section profiles: (a) Equilateral triangular section; (b) square section. **Figure 9.** Two representative non-axisymmetric section profiles: (**a**) Equilateral triangular section; (**b**) square section. **Figure 9.** Two representative non‐axisymmetric section profiles: (a) Equilateral triangular section; (b) square section.

(**a**) (**b**) (**c**) **Figure 10.** Maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles: **(a)** Maximum absorbed power; **Figure 10.** Maximum absorbed power and non‐dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles: **(a)** Maximum absorbed power; **Figure 10.** Maximum absorbed power and non-dimensional motion amplitude of the integrated system as a function of wave frequency at different section profiles: (**a**) Maximum absorbed power; (**b**) non-dimensional motion amplitude of the offshore structure; (**c**) non-dimensional motion amplitude of the floating buoy.

#### **5. Performance of the Designed Integrated System in Irregular Waves**

The designing process in the previous section suggests that a larger mass of the buoy is better, and a favorable underwater shape of the buoy is characterized by large diameter to draft ratio and constant section size, with the section profile preferring a circle or square rather than an equilateral triangle. In this section, we progress the investigation into irregular waves to study the performance of the designed integrated system in real ocean sea. The mass ratio is set to 1, and the truncated circular cylinder with a diameter to draft ratio of 4 is chosen as the underwater shape of the buoy.

An irregular wave can be recognized as the superposition of regular wave components [23] with the wave amplitude weighted by the energy distribution function, which in this paper employs the Pierson-Moskowitz spectrum [24], defined as

$$S(\omega) = 262.9 \frac{H\_s^2}{\omega^5 T\_\varepsilon^4} \exp\left(-\frac{1054}{\omega^4 T\_\varepsilon^4}\right) \tag{34}$$

where *H<sup>s</sup>* is the significant wave height and *T<sup>e</sup>* is the energy period. For the *n*-th regular wave component with frequency ω*n*, the wave amplitude is

$$A\_{\mathfrak{n}} = \sqrt{2\mathcal{S}(\omega\_{\mathfrak{n}})\Delta\omega}.\tag{35}$$

The response of the offshore structure and the floating buoy is given by

$$\mathbf{z}\_{s} = \sum\_{n} \text{Re}\left\{ \frac{Z\_{p}\hat{f}\_{\text{d}}(\omega\_{n})}{Z\_{\text{s}}(\omega\_{n})Z\_{\text{b}}(\omega\_{n}) + [Z\_{\text{s}}(\omega\_{n}) + Z\_{\text{b}}(\omega\_{n})]Z\_{p}} \exp[\mathbf{i}(\omega\_{n}\mathbf{t} + \varphi\_{n})]} \right\};\tag{36}$$

$$z\_b = \sum\_n \text{Re}\left\{ \frac{\left[Z\_s(\omega\_n) + Z\_p\right] \xi\_d(\omega\_n)}{Z\_s(\omega\_n) Z\_b(\omega\_n) + \left[Z\_s(\omega\_n) + Z\_b(\omega\_n)\right] Z\_p} \exp[\mathbf{i}(\omega\_n t + \varphi\_n)]\right\} \tag{37}$$

where *t* is the time, and ϕ*<sup>n</sup>* is the phase of the *n*-th regular wave designated randomly. The absorbed power of the integrated system is given as

$$P = \frac{c\_p}{t\_2 - t\_1} \int\_{t\_1}^{t\_2} (z\_s - z\_b)^2 \,\mathrm{d}t. \tag{38}$$

Due to orthogonality, influence of the coupled terms between different wave frequencies when inserting Equation (36) and Equation (37) into Equation (38) will vanish if the time interval between *t<sup>1</sup>* and *t<sup>2</sup>* is long enough [25]. Then, the absorbed power of the integrated system reduces to

$$P = \sum\_{\mathbf{n}} \frac{\omega\_{\mathbf{n}}}{2} \frac{\left| \hat{f}\_{\mathbf{d}}(\omega\_{\mathbf{n}}) \right|^{2}}{\left| \mathbf{G}(\omega\_{\mathbf{n}}) \right|^{2} \mathbf{Z}\_{\mathbf{p}i} + 2 \mathbf{Z}\_{\mathbf{d}i}(\omega\_{\mathbf{n}}) + \frac{\left| \mathbf{G}(\omega\_{\mathbf{n}}) \right|^{2} \mathbf{Z}\_{\mathbf{p}r}^{2} + 2 \left[ \mathbf{G}\_{\mathbf{r}}(\omega\_{\mathbf{h}}) \mathbf{Z}\_{\mathbf{h}}(\omega\_{\mathbf{n}}) + \mathbf{G}\_{\mathbf{r}}(\omega\_{\mathbf{h}}) \mathbf{Z}\_{\mathbf{h}}(\omega\_{\mathbf{n}}) \right] \mathbf{Z}\_{\mathbf{p}r} + \left| \mathbf{Z}\_{\mathbf{p}}(\omega\_{\mathbf{n}}) \right|^{2}} \tag{39}$$

Figure 11 shows the contour of the absorbed power as a function of the spring stiffness and damping coefficient of the PTO system at *H<sup>s</sup>* = 2 m and *T<sup>e</sup>* = 8 s. It is found that the absorbed power is a unimodal function of the spring stiffness and damping coefficient of the PTO system. As with previous sections, the spring stiffness of the PTO system should be constrained to be non-negative. Therefore, finding the maximum absorbed power of the integrated system is a constrained optimization problem, which is solved here by employing the *fmincon* function using the active-set method within the MATLAB environment.

**Figure 11.** Contour of the absorbed power as a function of the spring stiffness and damping coefficient of the PTO system at *Hs* = 2 m and *Te* = 8 s. **Figure 11.** Contour of the absorbed power as a function of the spring stiffness and damping coefficient of the PTO system at *Hs* = 2 m and *Te* = 8 s. **Figure 11.** Contour of the absorbed power as a function of the spring stiffness and damping coefficient the at = 2 m and *Te* = 8

Figure 12 shows the optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of energy period at *Hs* = 2 m. It is found that the optimal spring stiffness of the PTO system automatically fulfills the constraint within the investigated range of energy period without being compulsorily tuned to zero. Both the optimal spring stiffness and damping coefficient experience rather milder variation with the energy period than in regular wave conditions, as shown in Figure 4; Figure 5. This can be explained by the fact that irregular waves are composed of wave components of different wave periods, and the behavior of the integrated system in irregular waves is the average of that in regular waves weighted by the energy distribution function. As a result, the violent variation tendency is filtered out by the averaging process. The maximum absorbed power peaks at the energy period of 7.3 s, which is within the common range of wave period in the ocean, i.e., from 6 s to 10 s, indicating that the power absorption ability of the integrated system is at the favorable state. When wave periods are larger than 10 s, it is found that the curves in both subfigures of Figure 12 keep the current tendency and monotonically decrease with the wave period. Therefore, to focus on the WEC behavior in the common range of wave period, the data out of the range is not shown in the figure. Figure 12 shows the optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of energy period at *H<sup>s</sup>* = 2 m. It is found that the optimal spring stiffness of the PTO system automatically fulfills the constraint within the investigated range of energy period without being compulsorily tuned to zero. Both the optimal spring stiffness and damping coefficient experience rather milder variation with the energy period than in regular wave conditions, as shown in Figure 4; Figure 5. This can be explained by the fact that irregular waves are composed of wave components of different wave periods, and the behavior of the integrated system in irregular waves is the average of that in regular waves weighted by the energy distribution function. As a result, the violent variation tendency is filtered out by the averaging process. The maximum absorbed power peaks at the energy period of 7.3 s, which is within the common range of wave period in the ocean, i.e., from 6 s to 10 s, indicating that the power absorption ability of the integrated system is at the favorable state. When wave periods are larger than 10 s, it is found that the curves in both subfigures of Figure 12 keep the current tendency and monotonically decrease with the wave period. Therefore, to focus on the WEC behavior in the common range of wave period, the data out of the range is not shown in the figure. Figure <sup>12</sup> showsthe optimal parametersof the PTO system andmaximumabsorbed power ofthe integrated systemasa function of energy period It isfoundthat theoptimalspringstiffnessofthe PTOsystem automaticallyfulfillsthe constraintwithintheinvestigatedrange ofenergy period without being compulsorily tozero.Boththeoptimal springstiffness damping coefficient experience rather milder variation with the energy period than in regular wave conditions, asshownin Figure4; Figure5. This can be explained by thefact that irregular are composed of wave components of different wave periods, and the behavior of the integrated system in is the average of thatin regular waves weighted by the energy distribution function. As a result, the violent variation tendency is filtered out by the averaging process. The maximum absorbedpowerpeaks at the energy of 7.3s, whichiswithin the common range ofwave period in the ocean, i.e., from <sup>6</sup> <sup>s</sup> to <sup>10</sup> s, indicating that the power absorption ability of the integrated system is at the favorable state.When waveperiodsare larger than <sup>10</sup> s, it isfoundthatthe curves in both subfigures of Figure <sup>12</sup> keep the current tendency and monotonically decrease withthe wave period. Therefore, to focus on the WEC in rangeof wave period, the data out of the range is not shown in the figure.

**Figure 12.** Optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of energy period at *Hs* = 2 m: **(a)** Optimal parameters of the PTO system; **(b)** maximum absorbed power of the integrated system. **Figure 12.** Optimalparameters ofthePTOsystemandmaximum absorbedpowerof the integrated system asafunction ofenergy period at *Hs* =2 m:**(a)**Optimal parameters of thePTO system;**(b)Figure 12.** Optimal parameters of the PTO system and maximum absorbed power of the integrated system as a function of energy period at *H<sup>s</sup>* = 2 m: (**a**) Optimal parameters of the PTO system; (**b**) maximum absorbed power of the integrated system.

Inserting the optimal parameters of the PTO system into Equation (36) and Equation (37) gives the motion response of the offshore structure and the floating buoy when the absorbed wave power maximumabsorbedpower of theintegratedsystem.Inserting theoptimal parameters of thePTO system into Equation (36)and Equation (37)givesthe motion response of the offshore structure and the floating buoy when the absorbed wave power Inserting the optimal parameters of the PTO system into Equation (36) and Equation (37) givesthe motion response of the offshore structure and the floating buoy when the absorbed wave power is

as

maximized. Figure 13 shows an example of the time history of the motion response of the integrated system at *T<sup>e</sup>* = 6 s and *H<sup>s</sup>* = 2 m when the absorbed wave power is maximized. It can be observed that the offshore structure always follows the motion of the buoy, but with a different magnitude and lag phase. system at *Te* = 6 s and *Hs* = 2 m when the absorbed wave power is maximized. It can be observed that the offshore structure always follows the motion of the buoy, but with a different magnitude and lag phase.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 13 of 16

**Figure 13.** An example of the time history of the motion response of the integrated system at *Te* = 6 s and *Hs* = 2 m when the absorbed wave power is maximized. **Figure 13.** An example of the time history of the motion response of the integrated system at *Te*= 6 s and *Hs* = 2 m when the absorbed wave power is maximized.

To quantitatively estimate the motion amplitude of the integrated system in irregular waves, we imitate the definition of the significant wave height, which is the mean of the highest one‐third wave height in a wave record [26], to define the significant motion amplitude as the mean of largest one‐ third motion amplitude in a time history record. The significant motion amplitude of the offshore structure and the buoy are denoted by *zss* and *zbs*, respectively, and they probably represent the visual estimate of the motion amplitude of integrated system. As with previous representation of the motion amplitude of the integrated system, the non‐dimensional form of *zss* and *zbs* are respectively defined To quantitatively estimate the motion amplitude of the integrated system in irregular waves, we imitate the definition of the significant wave height, which is the mean of the highest one-third wave height in a wave record [26], to define the significant motion amplitude as the mean of largest one-third motion amplitude in a time history record. The significant motion amplitude of the offshore structure and the buoy are denoted by *zss* and *zbs*, respectively, and they probably represent the visual estimate of the motion amplitude of integrated system. As with previous representation of the motion amplitude of the integrated system, the non-dimensional form of *zss* and *zbs* are respectively defined as

$$z\_{\rm sss} = \frac{z\_{\rm ss}}{H\_{\rm s}/2};\tag{40}$$

$$z\_{\rm ss}$$

$$z\_{\rm bsn} = \frac{z\_{\rm bs}}{H\_{\rm s}/2}.\tag{41}$$

௦⁄2 Figure 14 shows the non‐dimensional significant motion amplitude of the integrated system as a function of energy period when the absorbed wave power is maximized. The calculated result is the mean of 10 different cases of random phases for the regular wave components. Within the investigated range of energy period, the motion amplitude of the offshore structure is larger than the buoy, and both increase with energy period. One may note that, in the designing process, we have tried our best to choose the design with smaller motion amplitude of the offshore structure, which however exceeds the wave amplitude for a certain range of energy period. This suggests that the offshore structure should be sophisticatedly designed to function insensitively to its heave motion. When offshore structures are not integrated with WECs, they are normally designed to suppress oscillation with waves. However, when they are integrated with WECs, their motion amplitude is even larger than waves. Therefore, to employ offshore structures to absorb wave power, functional devices in the offshore structure should work normally even when the structural support is under large oscillation. In this condition, measures may be taken to suppress oscillation of functional devices, such as motion compensation systems, vibration attenuation systems, etc. On the other hand, Figure 14 shows the non-dimensional significant motion amplitude of the integrated system as a function of energy period when the absorbed wave power is maximized. The calculated result is the mean of 10 different cases of random phases for the regular wave components. Within the investigated range of energy period, the motion amplitude of the offshore structure is larger than the buoy, and both increase with energy period. One may note that, in the designing process, we have tried our best to choose the design with smaller motion amplitude of the offshore structure, which however exceeds the wave amplitude for a certain range of energy period. This suggests that the offshore structure should be sophisticatedly designed to function insensitively to its heave motion. When offshore structures are not integrated with WECs, they are normally designed to suppress oscillation with waves. However, when they are integrated with WECs, their motion amplitude is even larger than waves. Therefore, to employ offshore structures to absorb wave power, functional devices in the offshore structure should work normally even when the structural support is under large oscillation. In this condition, measures may be taken to suppress oscillation of functional devices, such as motion compensation systems, vibration attenuation systems, etc. On the other hand, this is the accompanying price when the offshore structure is employed to absorb wave power.

this is the accompanying price when the offshore structure is employed to absorb wave power. Levelized costs of energy (LCOE) are defined as the ratio of total lifetime costs versus total expected outputs, and are used to measure economical competitiveness of a renewable energy technology. Total lifetime costs usually include investment costs (i.e., construction costs), operations, and maintenance costs. Since wave energy is still at an infant stage, lifetime costs of WECs vary significantly depending on many factors, such as the initial capital cost, the capture ratio, the discount rate, and the installation size. Some papers have done preliminary work on estimating levelized costs of wave energy. However, due to immaturity of WEC technology, the estimation process is usually based on many assumptions, and the estimated results show high level of uncertainty. For the integrated system in this paper, levelized costs of wave energy are also quite hard to estimate precisely. However, we can employ some basic data in the literature to provide a rough estimation to find the improvement of levelized costs of wave energy when WECs are integrated to offshore structures. It can be seen from [7] that levelized costs of wave energy are about 190 ₤/MWh, in which the construction component occupies 155 ₤/MWh. When integrating WECs to offshore structures, the substructure, mooring system, and electrical grid connections can be shared. Therefore, when calculating levelized costs of wave energy of the integrated system, the above three parts that account for 36% of the construction cost [7] can be subtracted. As a result, the total construction costs reduce by 55.8 ₤/MWh. If we assume other costs beside construction to be consistent, levelized costs reduce by 29.4%. Therefore, when integrating WECs with offshore structures, levelized costs of wave energy will be significantly reduced. However, we should note that the above calculation is quite simple and under a high level of uncertainty. Precise improvement of levelized costs of wave energy should take into consideration more factors and will be the work of next phase. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 14 of 16

**Figure 14.** Non‐dimensional significant motion amplitude of the integrated system as a function of energy period when the absorbed wave power is maximized. **Figure 14.** Non-dimensional significant motion amplitude of the integrated system as a function of energy period when the absorbed wave power is maximized.

#### Levelized costs of energy (LCOE) are defined as the ratio of total lifetime costs versus total **6. Conclusions**

**6. Conclusions** 

expected outputs, and are used to measure economical competitiveness of a renewable energy technology. Total lifetime costs usually include investment costs (i.e., construction costs), operations, and maintenance costs. Since wave energy is still at an infant stage, lifetime costs of WECs vary significantly depending on many factors, such as the initial capital cost, the capture ratio, the discount rate, and the installation size. Some papers have done preliminary work on estimating levelized costs of wave energy. However, due to immaturity of WEC technology, the estimation process is usually based on many assumptions, and the estimated results show high level of uncertainty. For the In this work, our main interest is focused on the design of an integrated system, which consists of a floating buoy and an offshore structure on board and employs the relative motion between the two bodies to absorb wave energy, so that satisfied performance in the power absorption process can be achieved. The parameter to design is the mass and underwater shape of the buoy, and the performance concerned includes the absorbed power and motion amplitude of the two bodies. The spring stiffness of the power take-off (PTO) system is constrained to be non-negative with the concerns of complexity in implementation and system stability.

integrated system in this paper, levelized costs of wave energy are also quite hard to estimate precisely. However, we can employ some basic data in the literature to provide a rough estimation to find the improvement of levelized costs of wave energy when WECs are integrated to offshore structures. It can be seen from [7] that levelized costs of wave energy are about 190 ₤/MWh, in which When PTO parameters are not constrained, the optimal spring stiffness will be negative when imaginary and real parts of the non-dimensional buoy impedance are smaller than and close to 0.5, respectively. The constraint on the spring stiffness of the PTO system will reduce the maximum absorbed power when the unconstrained optimal value constrained is negative.

the construction component occupies 155 ₤/MWh. When integrating WECs to offshore structures, the substructure, mooring system, and electrical grid connections can be shared. Therefore, when calculating levelized costs of wave energy of the integrated system, the above three parts that account for 36% of the construction cost [7] can be subtracted. As a result, the total construction costs reduce by 55.8 ₤/MWh. If we assume other costs beside construction to be consistent, levelized costs reduce by 29.4%. Therefore, when integrating WECs with offshore structures, levelized costs of wave energy will be significantly reduced. However, we should note that the above calculation is quite simple and A larger mass of the buoy is advantageous due to smaller optimal spring stiffness and damping coefficient of the PTO system, more absorbed wave power, and less motion amplitude of the two bodies. The best underwater shape of the buoy is characterized by large diameter to draft ratio with the section profile prefers a circle or square rather than an equilateral triangle. Investigations on the designed buoy in irregular waves show that the integrated system presents its peak power absorption ability within the common range of energy period, and the motion amplitude of the offshore structure is larger than the wave amplitude in a certain range of sea states.

In this work, our main interest is focused on the design of an integrated system, which consists of a floating buoy and an offshore structure on board and employs the relative motion between the two bodies to absorb wave energy, so that satisfied performance in the power absorption process can be achieved. The parameter to design is the mass and underwater shape of the buoy, and the performance concerned includes the absorbed power and motion amplitude of the two bodies. The spring stiffness of the power take‐off (PTO) system is constrained to be non‐negative with the

When PTO parameters are not constrained, the optimal spring stiffness will be negative when imaginary and real parts of the non‐dimensional buoy impedance are smaller than and close to 0.5, respectively. The constraint on the spring stiffness of the PTO system will reduce the maximum

A larger mass of the buoy is advantageous due to smaller optimal spring stiffness and damping coefficient of the PTO system, more absorbed wave power, and less motion amplitude of the two

into consideration more factors and will be the work of next phase.

concerns of complexity in implementation and system stability.

absorbed power when the unconstrained optimal value constrained is negative.

under a high level of uncertainty. Precise improvement of levelized costs of wave energy should take

**Author Contributions:** Conceptualization, Z.N.; formal analysis, Z.N.; methodology, J.W.; writing—original draft, J.W.; writing—review & editing, J.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the NSFC of China, grant number 51905096, and the Basic Research Plan of Jiangsu Province, grant number BK20190373, and was also supported by "the Fundamental Research Funds for the Central Universities".

**Conflicts of Interest:** The authors declare no conflict of interest.

## **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Optimizing Wave Overtopping Energy Converters by ANN Modelling: Evaluating the Overtopping Rate Forecasting as the First Step**

**José Manuel Oliver 1,2,\*, Maria Dolores Esteban <sup>2</sup> , José-Santos López-Gutiérrez <sup>2</sup> , Vicente Negro <sup>2</sup> and Maria Graça Neves <sup>3</sup>**


**Abstract:** Artificial neural networks (ANN) are extremely powerful analytical, parallel processing elements that can successfully approximate any complex non-linear process, and which form a key piece in Artificial Intelligence models. Its field of application, being very wide, is especially suitable for the field of prediction. In this article, its application for the prediction of the overtopping rate is presented, as part of a strategy for the sustainable optimization of coastal or harbor defense structures and their conversion into Waves Energy Converters (WEC). This would allow, among others benefits, reducing their initial high capital expenditure. For the construction of the predictive model, classical multivariate statistical techniques such as Principal Component Analysis (PCA), or unsupervised clustering methods like Self Organized Maps (SOM), are used, demonstrating that this close alliance is always methodologically beneficial. The specific application carried out, based on the data provided by the CLASH and EurOtop 2018 databases, involves the creation of a useful application to predict overtopping rates in both sloping breakwaters and seawalls, with good results both in terms of prediction error, such as correlation of the estimated variable.

**Keywords:** artificial neural network; principal component analysis; wave energy converters; wave overtopping rate

#### **1. Introduction**

During the last decade, the power generation sector has experienced a huge rise in what are known as renewable energy. The sea as one of the most powerful energy sources on earth, over 120,000 TWh/year capacity [1], is one of the key pieces in this sustainability strategy, such that it has even been included in the Sustainable Development Goals (SDGs) [2]. Among its potentials as a generator of clean energy [3] can be cited those derived from the use of marine currents [4], tides [5], thermal gradients [6], salt flats [7], and finally the use of waves to generate energy [8]. These uses have experienced one of the highest growth rates among renewable energy technology in recent years [9], and this may mean a change in trend in the production of sustainable energy, although it is not without major drawbacks at this early stage of development, one of which is precisely associated with the disparity in technologies [10].

This change in trend has undoubtedly been favored by technological advances in all the sectors involved. This is especially so in relation to hydrodynamic systems, with the improvement of equipment and control systems, as well as the utilization of new, more durable materials in such a strongly aggressive environment as the marine one [11]. But these are not the only motivations that direct the focus of interest towards this sector. There

**Citation:** Oliver, J.M.; Esteban, M.D.; López-Gutiérrez, J.-S.; Negro, V.; Neves, M.G. Optimizing Wave Overtopping Energy Converters by ANN Modelling: Evaluating the Overtopping Rate Forecasting as the First Step. *Sustainability* **2021**, *13*, 1483. https://doi.org/10.3390/ su13031483

Academic Editor: Gregorio Iglesias Rodriguez Received: 7 November 2020 Accepted: 21 January 2021 Published: 1 February 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

is no denying the change in trend in energy production strategies, linked it with a change in social sensitivity. A sensitivity which advocates the search for new sources of energy not linked to the consumptive use of finite natural resources, and where the search for more sustainable harvesting strategies is necessary. Especially if these coexist with a catalytic impulse advocated by different administrations [12].

Among the quoted wide range of uses, the one attributed to the inexhaustible energy of the waves stands out for its potential, especially when it is estimated that its associated energy power ranges between 8000–80,000 TWh/year [1]. Although this is a figure currently under intense debate, since different and detailed approximations will cause variation when specific dependent factors are introduced in the analysis [13]. Although faced with the common drawbacks that characterize developing marine technologies, advantages are presented that make structures for harnessing wave energy strategically advantageous alternatives compared to other marine energy converting technologies. For example, the clear correlation between areas of high energy demand such as densely populated coastal regions and production areas, or the greater stationarity of the generating capacity of the waves compared to more established renewable energy sources, such as wind energy [14].

As a crucial part of this search we focused on those that profit from energy generated when a singular structure is overtopped by incoming waves. Due to the fact that wave height decreases as waves travel from offshore to onshore, more powerful energy generation will commonly be associated with floating structures placed offshore [15], like popular Wave Dragon [16]. There also exists the possibility of combining power generation with other infrastructural needs. It can be combined with building defense structures, or take advantage of existing ones, which will result in a significant reduction of installation costs [17,18], and maintenance costs will be lower because of the accessibility of sites compared to those associated with offshore structures [15]. The sharing of construction techniques commonly used in coastal engineering structures is well documented [18,19], and may be extended to taking advantage of those areas associated with wave propagation, such as refraction, making its implementation even more interesting [20]. These reasons justify that one part of the research effort should be directed to the use of this type of energy converter in existing breakwater structures.

One of the most attractive challenges related to the wave energy generation field of application, is the application of Artificial Intelligence (AI) techniques. This can be applied to both the characterization of the problems associated with the technologies of the converters, and for the evaluation of the energy resources themselves [21]. As proof of its high potential and versatility, AI techniques are commonly applied in the majority of scientific fields, especially in those applied sciences such as engineering, sometimes with significant success [22].

The present work particularly explores the capacities of one of the most popular AI techniques: Artificial Neural Networks, applied in the field of coastal engineering, and specifically for the forecasting of the overtopping rate as an essential part of the design of wave energy converter structures of the overtopping type [11,19].

The potential power of the wave energy converter during its life time *T* can be defined by the following expression:

$$Energy = \int\_{0}^{T} fP\_{h} \tag{1}$$

where *P<sup>h</sup>* is the hydrodynamic power, and *f* is a factor that comprises several efficiency related factors, as electrical, mechanical, or relative to the electrical energy transmission.

There is a direct relationship between the hydrodynamic power energy in the incoming waves with the power stored in the reservoirs of the overtopping breakwater for energy conversion from potential energy to useful energy, but it is necessary to adequately determine certain parameters as the height of the crest or the slope characteristics, to determine the overtopping rate associated with it. The accuracy of such information together with the stochastic nature of the variation both in height and in period that characterize the wave field make it a complex problem and difficult to solve, which is not always adequately

solved in the related literature [11]. It is in this multivariate, nonlinear, uncertain situation that ANN can outperform classical approaches [23] and is what directs this research.

The research focuses on building an ANN model where its predictive properties can be applied to a wide range of structures. Data from the CLASH project and EurOtop 2018 are used to develop a new methodological approach that can be differentiated from those proposed in contemporary works [24–26]. The clear objective being the incorporation of this knowledge into the design of wave energy converters for breakwaters and coastal structures, maintaining their defense purpose.

As previously mentioned, other more specific works have developed ANN applications [27], some introduce the prediction of new parameters related to the transmission coefficient [28,29], or the reflection coefficient [26,30,31], or, in one exceptional case, used an ANN as an indirect means for the proposal of a new empirical expression for the calculation of the overtopping [32]. All of them validate the application of artificial intelligence techniques in this specific field of maritime engineering. The application of which has already been related to the more classic parameters of the waves or coastal engineering [33], and even in fields as specific as the scour depth around of marine structures [34–37], or newer ones, such as interaction with coastal biocenosis [38].

#### **2. Materials and Methods**

The proposed methodological approach includes a first phase of descriptive analysis of the available variables in the database, where the data are scaled to a common scale, and then proceeds to the detection and elimination of outliers, using univariate and multivariate techniques. As the next step, the identification of the most significant predictor variables is carried out. This is followed by moving to a process of dimensionality reduction of the input data space, using various techniques: Principal Component Analysis, and artificial neural networks with clustering ability (Kohonen networks). Once these pre-processing phases have been completed, the predictive Artificial Neural Network modelling is carried out, and the results of the prediction are analyzed.

#### *2.1. The Data Base and Its Component Parameters*

The overtopping is defined as the physical phenomenon that causes a certain flow over the top of a structure when the height of top is less than the rise of the waves (run-up) of the successive wave trains that affect that structure. Its quantification is carried out mainly by means of the variable called overtopping rate (*q*), which is the flow that exceeds a length of structure per unit of time, when a certain number of waves impact on it.

In the case during time interval *t*<sup>0</sup> there were *N*<sup>0</sup> waves falling upon the structure, of heights and periods (*H<sup>i</sup>* , *T<sup>i</sup>* ), where each wave produces a certain volume of overflow *V<sup>i</sup>* (*H<sup>i</sup>* , *T<sup>i</sup>* ), the overtopping rate can be defined as:

$$q = \frac{1}{t\_0} \sum\_{i=1}^{N\_0} V\_i(H\_{i,\prime}, T\_i) \tag{2}$$

where is the overtopping rate (m3/s/m or m2/s); *N*<sup>0</sup> is the number of total waves; *Hi*, , *T<sup>i</sup>* are the height and period of every *i* wave that fall upon the structure (*m*; *s*); *Vi*(*Hi*, , *Ti*) is the overflow volume produced by each wave of the wave series, per unit length; and *t*<sup>0</sup> = ∑ *N*<sup>0</sup> *i*=1 *Ti* is the duration of the record of waves in a storm (*s*).

Quantification is usually carried out by both prototype test and reduced scale tests, but the latter is mainly used due to its greater economy and simplicity [37]. Both methodological approaches have been used to compile the database that concerns this investigation: CLASH-EurOtop.

The improvement of the databases that have led to their application, has recently borne fruit in a new edition of the EurOtop manual [25]. There are several parameters that must be taken into account when building any overtopping rate prediction model, most of which are collected in this database, as are those used in this study. Among them, the main ∑ ேబ

tion: CLASH-EurOtop.

ones will be geometric and hydraulic parameters [39]. Hydraulic parameters such as the significant wave height (*Hs*), the mean and peak wave period (*T*), the direction of incidence of the waves on the structure (*β*). Geometric, or related to the typology of the structure itself, such as the freeboard of the structure (*Rc*). Other factors that also influence, although to a lesser extent, would be the bottom slope, the depth at the foot of the structure, the wind (direction and intensity), the wave grouping, the run-up interference, and so on. So, any database related to the quantification of that phenomenon will necessarily be enriched by data of this nature. ture, the wind (direction and intensity), the wave grouping, the run-up interference, and so on. So, any database related to the quantification of that phenomenon will necessarily be enriched by data of this nature. There are many empirical formulations in the specialized literature [40], most of them limited to a certain field or range of application, set either by the nature or design of the structure, or by other conditions specific to the environment (wind, wave conditions, angle of incidence). The desire and need to unify all these existing formulations, at least as

There are many empirical formulations in the specialized literature [40], most of them limited to a certain field or range of application, set either by the nature or design of the structure, or by other conditions specific to the environment (wind, wave conditions, angle of incidence). The desire and need to unify all these existing formulations, at least as far as the data from trials in which they had their genesis, converged in the CLASH Project and its subsequent modifications at EurOtop 2018 [25,41,42]. Therefore, they are the sources of data input used in the modelling of this work. far as the data from trials in which they had their genesis, converged in the CLASH Project and its subsequent modifications at EurOtop 2018 [25,41,42]. Therefore, they are the sources of data input used in the modelling of this work. The parameters that describe each test, with an initial total number of 34, including the measurement of the overtopping rate, were grouped into three categories: general,

The parameters that describe each test, with an initial total number of 34, including the measurement of the overtopping rate, were grouped into three categories: general, hydraulic and structural parameters. The general parameters refer to the reliability that is given to each test carried out (*RF*), and his degree of complexity (*CF)*. The hydraulic parameters refer to those characteristics related to the waves. While the structural parameters, a total of 17, are proposed to geometrically define the structures under analysis, as well as their boundary conditions relative to depths of water (See Figure 1). hydraulic and structural parameters. The general parameters refer to the reliability that is given to each test carried out (*RF*), and his degree of complexity (*CF)*. The hydraulic parameters refer to those characteristics related to the waves. While the structural parameters, a total of 17, are proposed to geometrically define the structures under analysis, as well as their boundary conditions relative to depths of water (See Figure 1).

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 4 of 25

ୀଵ is the duration of the record of waves in a storm (*s*).

the overflow volume produced by each wave of the wave series, per unit length; and =

but the latter is mainly used due to its greater economy and simplicity [37]. Both methodological approaches have been used to compile the database that concerns this investiga-

borne fruit in a new edition of the EurOtop manual [25]. There are several parameters that must be taken into account when building any overtopping rate prediction model, most of which are collected in this database, as are those used in this study. Among them, the main ones will be geometric and hydraulic parameters [39]. Hydraulic parameters such as the significant wave height (*Hs*), the mean and peak wave period (*T*), the direction of incidence of the waves on the structure (*β*). Geometric, or related to the typology of the

although to a lesser extent, would be the bottom slope, the depth at the foot of the struc-

Quantification is usually carried out by both prototype test and reduced scale tests,

The improvement of the databases that have led to their application, has recently

**Figure 1.** Schematic of the initial descriptive parameters of the model, representing a generic **Figure 1.** Schematic of the initial descriptive parameters of the model, representing a generic model of a coastal structure.

model of a coastal structure. The hydraulic parameters related to the wave field proposed are: the spectral significant wave height offshore and at the toe of the structure (*Hm*0 *deep*; *Hm*0), the mean spectral wave period offshore and at the toe (*Tm*−1,0 *deep*; *Tm−*1,0), the average and peak period offshore (*Tm*,*deep*; *Tp*,*deep*), the peak period at toe of the structure (*Tp*,0), and the wave incidence angle (*β*). The structural parameters, considered to geometrically define the structure: the water depth offshore (*hdeep*), the water depth in front of the structure (*h*), the water depth at the toe of the structure (*ht*), the width of the toe berm (*Bt*), the width of the berm (*B*), the berm The hydraulic parameters related to the wave field proposed are: the spectral significant wave height offshore and at the toe of the structure (*Hm*<sup>0</sup> *deep*; *Hm*0), the mean spectral wave period offshore and at the toe (*Tm*−1,0 *deep*; *Tm*−1,0), the average and peak period offshore (*Tm*,*deep*; *Tp*,*deep*), the peak period at toe of the structure (*Tp*,0), and the wave incidence angle (*β*). The structural parameters, considered to geometrically define the structure: the water depth offshore (*hdeep*), the water depth in front of the structure (*h*), the water depth at the toe of the structure (*ht*), the width of the toe berm (*Bt*), the width of the berm (*B*), the berm submergence (*h<sup>b</sup>* ), the slope of the structure downward of the berm (*cotα<sup>d</sup>* ), the slope of the structure upward of the berm (*cotαu*), the average co-tangent, considering the contribution or not of the berm and the slope (*cotαincl*; *cotαexcl*), the slope of the berm (*tanα<sup>b</sup>* ), the crest freeboard of the structure (*Rc*), the armour crest freeboard of the structure (*Ac*) and the crest width of the structure (*Gc*), and finally among the structural parameters, those related with the armour elements characterization, like the permeability/roughness factor of the armour layer (*γ<sup>f</sup>* ) or the size of the structure elements along the slope (*D*).

#### *2.2. Artificial Neural Network Models*

ANNs are data-driven, parallel processing structures that offer highly nonlinear problem solving applications that are very difficult to solve using traditional techniques. Among the great variety of existing processing paradigms, the use of two of the most common is proposed in this study: Multilayer Perceptrons (MLP) [43], and the Kohonen Neural Networks (KNN) which is also named as Self Organizing Maps (SOM) [44].

#### 2.2.1. Multilayer Perceptron

MLP networks are a supervised training type of ANN, where neurons are strongly interconnected with the previous layer, from where they receive information, and with the posterior layer, towards where they transmit it. In the present case, the output layer is composed of a single neuron, corresponding to the overtopping rate, whose output response can be represented mathematically such that:

$$y = \sum\_{k=1}^{n} v\_k \rho \left(\sum\_{j=1}^{m} w\_{kj} \cdot x\_j + b\_k\right) \tag{3}$$

with input data represented by the matrix *X* ∈ *R <sup>m</sup>*, output data *<sup>y</sup>* <sup>∈</sup> *<sup>R</sup>*, the weight matrix being *V* ∈ *R* 1*xn* , *W* ∈ *R nxm*, and where the vector of bias terms *<sup>b</sup>* <sup>∈</sup> *<sup>R</sup> n* , where *m* is the dimension of the input space, *n* is the number of neurons in the hidden layer, and *R* is the set of real numbers. More detailed information for the MLP networks can be found in Haykin (1991) [43].

#### 2.2.2. Kohonen Neural Network

The KNN is an ANN model, with the ability to classify data according to their similarity, in a preserving topological way.

Kohonen networks are basically made up of two layers, the input layer and the competitive or exit layer (or self-organized map), both layers are fully interconnected, and unlike MLPs they respond to unsupervised training.

In a sequence of phases of competition, comparison, cooperation and adaptation, a process is structured in which a neuron of the output layer will be activated by a process of comparison or measurement of similarity between the input pattern and that neuron, the candidate for winning. This measure of similarity is usually the Euclidean distance *d<sup>j</sup>* between the input vector *X* and the vector of synaptic weights *W<sup>j</sup>* :

$$d\_{\hat{j}} = \left\| \left| \left. X - \mathcal{W}\_{\hat{j}} \right| \right\| = \sqrt{\sum\_{i=1}^{M} \left( x\_{i} - w\_{i\hat{j}} \right)^{2}} \tag{4}$$

Being the winning neuron, the one with the smallest distance between them. The weights of this winning neuron will therefore be adjusted, in the direction of the input vector, according to the expression:

$$\mathcal{W}\_{\text{I}}(t+1) = \mathcal{W}\_{\text{\textquotedblleft}}(t) + \eta(t) \cdot h\_{\text{\textquotedblright}}(t) \cdot \left(\mathbf{X} - \mathcal{W}\_{\text{\textquotedblleft}}(t)\right) \tag{5}$$

where *η* is the learning rate, and therefore the update of the weight vector *W<sup>J</sup>* (*t +* 1) for time *t +* 1, and *h<sup>j</sup>* (*t*) is the neighborhood function for time *t*.

With this adjustment, each node of the output layer develops the ability to recognize future input vectors that are presented to the network and are similar to it, grouping them in this environment according to a self-organizing process, which gives them propensity for grouping (clustering) [44]. More detailed information for the Kohonen networks and theirs mathematical foundations can be found in Kohonen (2001) [45].

#### *2.3. Pre-Processing of Data*

On the initial data set consisting of 17,942 trials, an exploratory and descriptive analysis phase of the data was carried out, conditioned especially by the general parameter *RF* and *CF*, and which allowed abundant discarding of trials (including all those vectors with missing data). Similarly, some parameters were discarded, based on bibliographic recommendations [24,41], due to their low significance. Among them: all the deep water parameters and some redundant structural parameters. The discards resulted in a dimensional reduction to a total of 23 variables.

#### 2.3.1. Data Escalation

The original data, once those considered anomalous had been discarded, underwent a scaling process based on the fact that they came from two very different sources such as: those from laboratory tests at different scales, and those from prototypes. Maintaining this lack of dimensional coherence would introduce an additional problem when modelling, and would also greatly hinder an elementary descriptive analysis of the data. Therefore, the adoption of a single scale is proposed [46].

The scaling process was carried out by applying a scaling based on the theory of dimensional analysis to represent the general equations of hydraulics and obtain the dimensionless Buckingham-Pi monomials, and specifically those based on Froude similarity [47].

For practical purposes, it is established that *Hm*<sup>0</sup> *<sup>t</sup>* acquires a value common to all trials of 1 m [24,25,48].

#### 2.3.2. Debugging the Data

Once the data has been scaled, the sample is debugged in order to use it as an input space in the training of neural networks. To do this it must be ensured that the integrating patterns do not contain outliers, since their presence can be highly detrimental the network training [49]. Another significant aspect that makes the elimination of outliers advisable is the improvement in the performance of the error functions that will determine the goodness of the model, and is especially desirable when we work with the MSE function [50].

For dealing with the problem caused by the presence of outliers in the input data, this work chose from different strategies [51], the one that proposes the early detection of them, later proceeding with their removal from the sample and finally facing the modelling process with a sample clean of outliers.

To check if an input pattern can be considered as a multidimensional outlier, a process based on the Mahalanobis distance is proposed [52].

To apply this test, a detection of those data vectors that present one-dimensional outliers is previously performed, establishing a class that will be compared by means of the Mahalanobis distance test with the other sample that does not contain univariate outliers, established through statistics such as Box Fisher's *F*, or Wilks' lambda statistic and supported by graphics (box-plots).

For practical purposes, the strategy referred to above means that from the univariate and multivariate analysis of the data, it turns out that the data set is reduced to another set made up of 23 variables and a total of 10,097 patterns, which constitute the definitive base, or sample space, available for a subsequent dimensionality reduction by applying diverse techniques. In the following table (Table 1) the main statistical parameters of this debugged data base are summarized.


**Table 1.** Statistical parameters of the scaled input–output patterns of the database used for training of the ANN, after the debugging process (scaled to *Hm*<sup>0</sup> *<sup>t</sup>* = 1 m).

#### 2.3.3. Dimensionality Reduction

The good generalizability of an ANN model is linked to its complexity. The presence of a very high number of features (>30) results in the well-known "curse of dimensionality" [23] to which the ANN are not alien [43]. Avoiding overfitting is one of the most important goals to achieve with dimensionality reduction. Therefore, an attempt is made to reduce the input dimension as much as possible, without loss of information associated with the sample variance.

In the present study, this objective has been achieved by applying, on the one hand, the discriminant analysis technique of Principal Component Analysis (PCA), and on the other hand, the application of Kohonen networks (SOM).

Assuming the matrix *X* of the sample data with *p* features and *n* integrating patterns (vectors) of the sample:

$$X = \begin{pmatrix} \begin{array}{cccc} \times & \mathbf{x}\_{11} & \mathbf{x}\_{12} & \cdots & \mathbf{x}\_{1p} \\ \times & \mathbf{x}\_{21} & \mathbf{x}\_{22} & \cdots & \mathbf{x}\_{2p} \\ \vdots & \vdots & \ddots & \vdots \\ \times & \mathbf{x}\_{n1} & \mathbf{x}\_{n2} & \cdots & \mathbf{x}\_{np} \end{array} \end{pmatrix} \tag{6}$$

PCA consists of finding orthogonal transformations of the original features to obtain a new set of uncorrelated ones, called Principal Components. These uncorrelated features are the eigenvectors, and are obtained in decreasing order of importance, this importance being associated with the amount of variance explained by them. As a result, the components are linear combinations of the original features and it is expected that only a few (the first of them) will capture most of the variability of the data, thus obtaining a reduction in size after the transformation that this technique involves [53]. Geometrically, transformation can in fact be explained as a rotation in the *p*-dimensional space (see Figure 2), looking for the projection that maximizes the information provided by the multivariate pattern in terms of variance.

tern in terms of variance.

**Figure 2.** Spatial interpretation of the PCA technique. The projection on the new axes PCA1 and PCA2 maximizes the information provided by the multivariate variable in terms of variance. **Figure 2.** Spatial interpretation of the PCA technique. The projection on the new axes PCA1 and PCA2 maximizes the information provided by the multivariate variable in terms of variance.

The space generated by the first *q* components is then a *q*-dimensional vector subspace of the original *p*-dimensional space. Thus, the principal components of *X* will be the The space generated by the first *q* components is then a *q*-dimensional vector subspace of the original *p*-dimensional space. Thus, the principal components of *X* will be the new variables:

$$Y\_j = X \cdot t\_{j\nu} \text{ j} = 1, \ldots, p \tag{7}$$

being associated with the amount of variance explained by them. As a result, the components are linear combinations of the original features and it is expected that only a few (the first of them) will capture most of the variability of the data, thus obtaining a reduction in size after the transformation that this technique involves [53]. Geometrically, transformation can in fact be explained as a rotation in the *p*-dimensional space (see Figure 2), looking for the projection that maximizes the information provided by the multivariate pat-

 =∙, = 1, … , (7) for each, the new variable *j* is constructed from the *j*-th eigenvector of *S* = var(*X*). Expressing it in a less compact way:

$$
\begin{pmatrix} Y\_1 \\ Y\_2 \\ \vdots \\ Y\_P \end{pmatrix} = \begin{pmatrix} t\_{11} & t\_{12} & \cdots & t\_{1p} \\ t\_{21} & t\_{22} & \cdots & t\_{2p} \\ \vdots & \vdots & \ddots & \vdots \\ t\_{n1} & t\_{n2} & \cdots & t\_{np} \end{pmatrix} \cdot \begin{pmatrix} x\_1 \\ x\_2 \\ \vdots \\ x\_p \end{pmatrix} \tag{8}
$$

$$
\text{The transformed data matrix is } Y = X \cdot T \text{, and represents the \textquotesingle observations\textquotesingle} \text{ of the new}
$$

ቌ ⋮ ⋮ ⋮ ⋱ ⋮ ଵ ଶ ⋯ ⋮ ቍ (8) The transformed data matrix is *Y = X*·*T*, and represents the "observations" of the new variables (principal components) on the *n* sample patterns.

The transformed data matrix is *Y = X∙T*, and represents the "observations" of the new variables (principal components) on the *n* sample patterns. In another way, the application in this case of a Kohonen network model, also indirectly allows the interpretation of existing relationships at the level of self-similarity. Information that will ultimately allow decisions to be made on reducing the dimensionality of the input space with topology preserving [54], either by interpretation of these detected relationships between the variables, or by confirmation of those already detected using In another way, the application in this case of a Kohonen network model, also indirectly allows the interpretation of existing relationships at the level of self-similarity. Information that will ultimately allow decisions to be made on reducing the dimensionality of the input space with topology preserving [54], either by interpretation of these detected relationships between the variables, or by confirmation of those already detected using the PCA technique. This is achieved through the interpretation of the distance matrix (U-matrix), and significantly, of the plane components (P-Matrix) [55]. The U-matrix represents in a 2D lattice the Euclidean distance between neighboring nodes. The component planes allow the graphic display, also in 2D, of each of the variables in the data set, by expressing the values of the weight vectors.

the PCA technique. This is achieved through the interpretation of the distance matrix (U-Although it should be noted that this procedure has an obvious drawback, which lies in the subjectivity of the criteria for establishing these relationships between variables.

#### matrix), and significantly, of the plane components (P-Matrix) [55]. The U-matrix repre-*2.4. Proposed Models*

new variables:

sents in a 2D lattice the Euclidean distance between neighboring nodes. The component planes allow the graphic display, also in 2D, of each of the variables in the data set, by expressing the values of the weight vectors. Although it should be noted that this procedure has an obvious drawback, which lies in the subjectivity of the criteria for establishing these relationships between variables. Prior to the construction of the first models, a parametric contrast of homogeneity of the joint sample is carried out, demonstrating that there are significant differences between the samples, once they have been scaled, from both tests on reduced models and from prototypes [30], thus breaching one of the premises that motivated the creation of an international and homogeneous base on wave overtopping [56].

It is well known that there are founded differences between overtopping rates determined from conventional scale models of breakwaters (generally based on Froude's law of similarity) [57], especially for rubble mound breakwaters in laboratories, and those measured on similar prototypes [42,58–61]. To confirm this lack of homogeneity, that could significantly influence any approach of ANN models, it was decided to carry out certain hypothesis tests, understanding these as a test of significance to demonstrate whether certain hypotheses that are assumed to be true, are with a certain degree of security. The hypothesis test proposed is a parametric test, checking for its certain population parameters, and also assuming that the distribution of the data is of a certain known type (distributed according to a normality hypothesis).

The parametric tests carried out consist of the Fisher test of Snedecor's *F*-contrast statistic over the sample variance and the bilateral hypothesis test on the equality of the sample means carried out using the Student's *t*-test statistic. Table 2 shows the results of the bilateral contrast carried out over the means and sample variances.

**Table 2.** *p*-valor results on the bilateral *F*-Snedecor and *t*-Student contrast (confidence level α = 0.05). In bold, those values that allow accepting the null hypothesis of equality of means or variances have been highlighted.


The results highlighted in bold in Table 2, finally allow concluding that the two samples are not homogeneous, due to the fact that only five of the variables (*cotαu*, *cotαincl*, *h*, *h<sup>t</sup>* , *h<sup>b</sup>* ) meet the hypothesis of equality of means, but additionally, none of them meet the hypothesis of equality in variances.

The proven existence of such a lack of homogeneity leads to proposing a differentiating strategy, which implies proposing suitably differentiated models for each sample with its own sampling characteristics. Therefore, two initial models are proposed, that will be trained with a total number of 9997 patterns, preserving an additional number of 100 patterns for extra validation purposes:


Model I will result in obtaining a single ANN, while Model II will involve the construction of 2 different ANNs: Sub model II.1 and sub model II.2. Sub model II.2 implies the application of a Kohonen network as a previous step for the optimized obtaining of the training, verification and test subsets on a very small sample of 171 patterns from prototype tests, according to the methodology described by Bowden and Maier in 2002 [62]. This is not necessary for the rest of models for which this division is randomly conducted. The quoted methodology, which includes a Kohonen network with 10 × 10 nodes that allows selecting three cases for each of these nodes, where the first of the data will be used for

1

the training subset, the second for the verification subset and finally the third goes to the validation subset. In the case that in one of the nodes of the self-organized map only one pattern existed, it would be mandatorily destined to the training subset, while if there were only two existing patterns, it would be acted so that the first one was destined to the subset of training, while the second would be destined to the verification one. In this way it is possible to reduce the patterns necessary to train the model to a minimum.

Univariate MLP networks are proposed for the construction of the predictive model. For the determination of its structure, it is assumed that more than one hidden layer does not represent an appreciable improvement [63], and on the contrary, would suppose a substantial increase in training time in addition to an increase in the possibility of network overfitting, with a lower capacity for generalization associated [23,43,64]. Therefore, the incremental calculation of the number of neurons is proposed as a valid criterion, by a trial and error procedure, and not the number of layers [65].

Before the training of the networks, the variables of the input patterns will be standardized to the range [−1, 1] by means of the following function:

$$\mathbf{x}' = \frac{(\mathcal{U}\_{\mathbf{x}} - L\_{\mathbf{x}}) \cdot (\mathbf{x} - \mathbf{x}\_{\min})}{(\mathbf{x}\_{\max} - \mathbf{x}\_{\min})} + L\_{\mathbf{x}} \tag{9}$$

where x 0 is the escalated variable; *x* is the original variable; *xmax* is the maximum of variable *x* in the original sample; *xmin* is the minimum of the variable *x* in the original sample; *U<sup>x</sup>* the transformed value of the maximum of the variable *x*; *L<sup>x</sup>* is the transformed value of the minimum of the variable x; and where the maximum and minimum escalation ranges are *U<sup>x</sup>* = 1, and *L<sup>x</sup>* = −1.

This scaling proposal implies that the activation function will be the hyperbolic tangent, which entails the mandatory use of a linear activation function in the output layer [66].

Candidate networks will be trained with MATLAB software (Mathworks ©) according to a cross-verification procedure [43], and using the Levenberg-Marquardt algorithm, as it is better adapted to the characteristics of the available sample space, providing better results, both in terms of error, as well as computational time and stability [67,68].

For practical purposes, from the complete sample of 9997 patterns, 85% will be used in the network training process, which means randomly dividing up that sample into a training subset consisting of 6997 patterns, and 1500 patterns intended for cross-verification process, while the remaining subset consisting of 1500 additional patterns will serve for the model test.

#### **3. Discussion**

This study investigates the capabilities of two differentiated ANN models to predict the overtopping rate based on different boundary conditions and in the capacity of the ANN to work optimally in homogeneous sample spaces [69]. The performance of these two models will be carried out based on the statistics of the mean squared error (MSE), and the correlation coefficient (*r*).

The mean squared error measures the average of the squared errors, that is, the difference between the simulated value (*q*ˆ*i*) and the observed value (*qi*) across the range of data (*n*):

$$MSE = \frac{\sum\_{i=1}^{n} (\mathfrak{q}\_i - q\_i)}{n} \tag{10}$$

The mean squared error is one of the most used functions, with interesting properties that make its use generalized, such as it is easily calculated and penalizes large errors. As a disadvantage, it requires errors to be distributed independently and normally [50].

The correlation coefficient (*r*) is an indicator of the degree of linear statistical dependence, and is calculated according to:

$$r = \frac{\sum\_{i=1}^{n} \left(\boldsymbol{\eta}\_{i} - \hat{\boldsymbol{\eta}}\right) \cdot \left(\boldsymbol{q}\_{i} - \overline{\boldsymbol{\eta}}\right)}{\sqrt{\sum\_{i=1}^{n} \left(\hat{\boldsymbol{q}}\_{i} - \hat{\boldsymbol{\eta}}\right)^{2} \cdot \sum\_{i=1}^{n} \left(\boldsymbol{q}\_{i} - \overline{\boldsymbol{q}}\right)^{2}}} \tag{11}$$

where *q*ˆ is the mean value of the simulated variable and *q* is the average value of the observed variable.

#### *3.1. Obtaining the Reduced Dimension of the Input Vector*

A first step will be to carry out a correlational analysis, by obtaining the correlation matrix. The information it provides is sufficient to make decisions about discarding some variables, but not enough to others, so it is necessary to resort to more sophisticated techniques such as those that will be used later. The existing correlations of the different explanatory variables with the variable to be predicted are very low (maximum *r* value of 0.322), which shows the high non-linearity of the process to be analyzed (Table 3).

**Table 3.** Correlation with explained variable (*q*).


In general, the correlation coefficients between the different variables that make up the set of input variables are low, or very low, reaching only significant correlation values in the following pairs of values:


Regarding PCA analysis, Table 4 represents the contribution of each variable to the first eight principal components, together with their correspondent eigenvalues and the cumulative variance explained by them. The total variance cumulated by them is higher than 75%, that is one of the criteria accepted in practice for stablishing the contributing limit to an effective model. Another adopted criterion will be that the variance explained by them be major than the mean, so major than one, rule also proportionated by the first eight components.


**Table 4.** Loadings of the first eight variables.

The first component (F1) explains 15.09% of the variance and is dominated by variables related with the period, slope of the structure and by the wave steepness. The second component (F2) with a similar percentage of the variance explained (13.55%) is dominated by variables related with the submergence and *cotαexcl*. And the third component (F3) explains 12.64% of the total variance and is dominated by the variables *cotαincl* and *Ac*.

The analysis of the correlation circle, that corresponds to a projection of the initial variables of the first two factors of the PCA onto a two-dimensional plane, provides relevant information that allows observing correlations between the variables and interpreting the axes, or main factors, and thus being able to eliminate correlations that could be redundant and therefore detrimental to the predictability of the model. For the present case, it is shown, in the correlation circle (in the projection of both F1 and F2 axes) (See Figure 3a), that the percentage of variability represented by the first two factors is not particularly high (28.64%). Therefore, to avoid a misinterpretation of the graphics, it also requires a visualization in axes 1 and 3, and interpretation of the influence of presence or absence of certain parameters (See Figure 3b).

**Figure 3.** Dimensionality reduction using the PCA technique: (**a**) Correlation circle on the F1–F2 plane; (**b**) Correlation circle on the F1–F3 plane. **Figure 3.** Dimensionality reduction using the PCA technique: (**a**) Correlation circle on the F1–F2 plane; (**b**) Correlation circle on the F1–F3 plane.

The projection on the F1 and F3 axes explain a total variability of 27.7%, which is a percentage very similar to that explained by the previous projection (F1 and F2). Additionally, in this projection the correlations established for the first circle of projections are Both plots confirm the results shown in Table 4, without a significant relevance of the variables, accompanied by a lack of clear interpretation of the axes, but on the other hand, with the evidence of certain interesting relationships between the variables.

Figure 3a), that the percentage of variability represented by the first two factors is not particularly high (28.64%). Therefore, to avoid a misinterpretation of the graphics, it also requires a visualization in axes 1 and 3, and interpretation of the influence of presence or

with the evidence of certain interesting relationships between the variables.

in the breakwater is clearly reflected along with its strong link with the F2 axis.

Both plots confirm the results shown in Table 4, without a significant relevance of the variables, accompanied by a lack of clear interpretation of the axes, but on the other hand,

It can be seen that there is a strong grouping between the variables related to the period (*Tm*1.0*t*; *Tp t*; *Tm t*), with a high positive correlation between them, a trivial matter already detected in the correlation matrix, which at least allows reducing their number, in any case keeping only one of them. Another group with a strong positive correlation is composed of those variables related to the geometric characterization of the slope (*cotαincl*; *cotαexcl*; *cotαd*) on which it will act in a similar way. The same procedure can be carried out with the variables relative to the width of the berm and its horizontal projection (*B*; *Bh*), and from which it is inferred that only variable *B* will be preserved. The grouping of variables in the correlation circle, in the projection of both axes F1 and F2, seems to determine the lack of correlation between the variables that make up the most obvious groupings, with similar direct cosines, such as those determined by *cotαincl*, *cotαexcl*, *cotα*d, and those like: *cotαu*, *tanαB*, *Hm0 t*. The foregoing leads to considering that both groupings of variables must be present in the input space, although with the particular restrictions indicated previously for some of them. The spectral wave steepness variable (*Hm*0 *t/Lm*1 *t*), negatively correlated with the freeboard variables (*Rc*, *Ac*), should be kept as above. Finally, the strong link between the width of the crest and the characteristic size of the protection elements

absence of certain parameters (See Figure 3b).

maintained, even the observed groupings are very similar. This robustness in the projection reaffirms the initial idea of finally discarding several of these correlated variables. It can be seen that there is a strong grouping between the variables related to the period (*Tm*1.0*<sup>t</sup>* ; *Tp t*; *Tm t*), with a high positive correlation between them, a trivial matter already detected in the correlation matrix, which at least allows reducing their number, in any case keeping only one of them. Another group with a strong positive correlation is composed of those variables related to the geometric characterization of the slope (*cotαincl*; *cotαexcl*; *cotα<sup>d</sup>* ) on which it will act in a similar way. The same procedure can be carried out with the variables relative to the width of the berm and its horizontal projection (*B*; *B<sup>h</sup>* ), and from which it is inferred that only variable *B* will be preserved. The grouping of variables in the correlation circle, in the projection of both axes F1 and F2, seems to determine the lack of correlation between the variables that make up the most obvious groupings, with similar direct cosines, such as those determined by *cotαincl*, *cotαexcl*, *cotα*d, and those like: *cotαu*, *tanαB*, *Hm0 t*. The foregoing leads to considering that both groupings of variables must be present in the input space, although with the particular restrictions indicated previously for some of them. The spectral wave steepness variable (*Hm*<sup>0</sup> *<sup>t</sup> /Lm*<sup>1</sup> *<sup>t</sup>* ), negatively correlated with the freeboard variables (*Rc*, *Ac*), should be kept as above. Finally, the strong link between the width of the crest and the characteristic size of the protection elements in the breakwater is clearly reflected along with its strong link with the F2 axis.

The projection on the F1 and F3 axes explain a total variability of 27.7%, which is a percentage very similar to that explained by the previous projection (F1 and F2). Additionally, in this projection the correlations established for the first circle of projections are maintained, even the observed groupings are very similar. This robustness in the projection reaffirms the initial idea of finally discarding several of these correlated variables.

A technique that is usually used when the information provided by the PCA analysis carried out on the total of the variables is not very informative, is to consider the contribution of some variables whose contribution is doubted, such as that of supplementary variables studying the effect of their elimination on the projection space. In this case, the input space is censored by considering as supplementary variables all those that have shown an evident correlation in the previous projections. Figure 4 shows the new correlation circle with this elimination step applied. initial projections, indicates that a reduction in dimensionality may be beneficial for the explanation of the problem without a significant loss of information [53], and therefore this reasoning can be valid for the composition of a model with a smaller input dimension.

A technique that is usually used when the information provided by the PCA analysis carried out on the total of the variables is not very informative, is to consider the contribution of some variables whose contribution is doubted, such as that of supplementary variables studying the effect of their elimination on the projection space. In this case, the input space is censored by considering as supplementary variables all those that have shown an evident correlation in the previous projections. Figure 4 shows the new corre-

Figure 4 demonstrates that the variability explained by the first two factors increases slightly with a total of the variance explained equal to 35.10%. Additionally, that there is strong linking between variables relates with freeboard parameters and F1 axis, which could explain the major variance in the sample. The above, and its comparison with the

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 14 of 25

lation circle with this elimination step applied.

**Figure 4.** Correlation circle in the first two principal components with variable elimination. In blue colored appear those variables that are candidates to be eliminated. **Figure 4.** Correlation circle in the first two principal components with variable elimination. In blue colored appear those variables that are candidates to be eliminated.

Alternatively, the application of a Kohonen network model on the same input pattern space, with a dimension of 23 factors (all of which come from the previous pre-processing processes), will indirectly allow the interpretation of the relationships existing at the level of self-similarity between input patterns, in a two-dimensional projection, where every pixel in that 2-dimensional map is characterized by a multidimensional vector. Information that will ultimately allow the reduction of the dimensionality of the entrance space, either by interpretation of these detected relationships, or by confirmation of those Figure 4 demonstrates that the variability explained by the first two factors increases slightly with a total of the variance explained equal to 35.10%. Additionally, that there is strong linking between variables relates with freeboard parameters and F1 axis, which could explain the major variance in the sample. The above, and its comparison with the initial projections, indicates that a reduction in dimensionality may be beneficial for the explanation of the problem without a significant loss of information [53], and therefore this reasoning can be valid for the composition of a model with a smaller input dimension.

already detected using the PCA technique. For the specific purposes of the present study, the constructed model, built with a Gaussian neighborhood function in every node, has been carried out using the following training scheme, characterized by two different phases [66], where each set is shown 500 times to the SOM. During the training phase, the complete preprocessed data set is shown 500 times to the SOM, built with a Gaussian neighborhood function in every node. The first of these, or rough adjustment phase, is performed with a learning rate with values in the range between 0.9 to 0.1, and with a neighborhood ratio that varies from 2 to 1, and with Alternatively, the application of a Kohonen network model on the same input pattern space, with a dimension of 23 factors (all of which come from the previous pre-processing processes), will indirectly allow the interpretation of the relationships existing at the level of self-similarity between input patterns, in a two-dimensional projection, where every pixel in that 2-dimensional map is characterized by a multidimensional vector. Information that will ultimately allow the reduction of the dimensionality of the entrance space, either by interpretation of these detected relationships, or by confirmation of those already detected using the PCA technique.

For the specific purposes of the present study, the constructed model, built with a Gaussian neighborhood function in every node, has been carried out using the following training scheme, characterized by two different phases [66], where each set is shown 500 times to the SOM. During the training phase, the complete preprocessed data set is shown 500 times to the SOM, built with a Gaussian neighborhood function in every node. The first of these, or rough adjustment phase, is performed with a learning rate with values in the range between 0.9 to 0.1, and with a neighborhood ratio that varies from 2 to 1, and with an extension of training up to 100 times. The second or fine-tuning phase is completed with a unique learning rate of *η* = 0.01, with a neighborhood ratio of 0, and with training extension up to 100 epochs. With this, the total length of the training will be:

280

units).

bles.

and output space).

100 + 100 = 200 times. The dimension of the input tile for this model will be 25 × 25 units (625 total units). can be redundant, and informs which two of them should be discarded. However, it is noteworthy that another of the variables related to that group of parameters, which refers

an extension of training up to 100 times. The second or fine-tuning phase is completed with a unique learning rate of *η* = 0.01, with a neighborhood ratio of 0, and with training extension up to 100 epochs. With this, the total length of the training will be: 100 + 100 = 200 times. The dimension of the input tile for this model will be 25 × 25 units (625 total

The following figure (Figure 5) shows the different component planes obtained after network training, with a total of 23 units (one for each component variable of the input

From the analysis of these component planes the existence of several evident relationships between the variables is deduced. The first of these concerns the parameters related to the definition of the slopes of the dykes and which comparatively shows the existence of a direct and significant correlation across the entire range of the data between the variables (*cotαexcl*, *cotαincl*, *cotαd*). A reason why the information contributed by them

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 15 of 25

The following figure (Figure 5) shows the different component planes obtained after network training, with a total of 23 units (one for each component variable of the input and output space). to the cotangent of the slope of the structure in the part of the slope above the berm (*cotαu*), presents a projection pattern that is notoriously different from the previous ones, but without a distinctive response in the component plane, so it should not be taken into account.

**Figure 5.** Dimensionality reduction. Self-organizing map (SOM): component planes of 23 varia-**Figure 5.** Dimensionality reduction. Self-organizing map (SOM): component planes of 23 variables.

Another very significant relationship detected by the SOM is the one that shows the component planes of the roughness factor variables (*γf*) and the mean diameter (*D*). The comparison of both planes shows the existence of a negative correlation between them, and thus the greater sizes, the lower the roughness factor. This relationship is evidence in the existing empirical knowledge and taken into account [32,41], but it is comforting to From the analysis of these component planes the existence of several evident relationships between the variables is deduced. The first of these concerns the parameters related to the definition of the slopes of the dykes and which comparatively shows the existence of a direct and significant correlation across the entire range of the data between the variables (*cotαexcl*, *cotαincl*, *cotα<sup>d</sup>* ). A reason why the information contributed by them can be redundant, and informs which two of them should be discarded. However, it is noteworthy that another of the variables related to that group of parameters, which refers to the cotangent of the slope of the structure in the part of the slope above the berm (*cotαu*), presents a projection pattern that is notoriously different from the previous ones, but without a distinctive response in the component plane, so it should not be taken into account.

Another very significant relationship detected by the SOM is the one that shows the component planes of the roughness factor variables (*γ<sup>f</sup>* ) and the mean diameter (*D*). The comparison of both planes shows the existence of a negative correlation between them, and thus the greater sizes, the lower the roughness factor. This relationship is evidence in the existing empirical knowledge and taken into account [32,41], but it is comforting to confirm that the ANN is capable of detecting it as well. Both parameters should be preserved a priori.

The next detected relationship is the one between the width of the berm (*B*) and its projected width (*B<sup>h</sup>* ), and that is also preserved across the data range. Therefore, only one of them should be selected, discarding the other.

It could be thought that, based on design criteria, there was a direct relationship between the width of the berm and the width of the toe (*Bt*), or with the width of the crest, however, this has not been detected at the data base analyzed for the crest width of the structure, so this supposed relationship will be discarded. While the existence of a partial correlation, at least in a region of the projection plane, between the variables of the berm width and the width of the bench (see Figure 5) is detected, which indicates that some of the breakwaters that have been tested have been designed with a theoretical pattern that relates both variables. The foregoing forces not discarding these variables, but to keep them in the input space, since this relationship is partial in the sample space it is necessary to preserve that differentiation.

A last relationship highlighted (Figure 5), and also expected by the existing empirical knowledge, is the one presented by the depth variables at the toe of the structure (*h*), and the one that define the submergence of it (*ht*). In this case, as expected, its correlation is direct or positive. However, the lack of correlation between both variables and the berm (*h<sup>b</sup>* ) is also striking, therefore, following the above reasoning, at least two of them should be maintained, discarding the third of them.

In view of the results obtained and interpreted after applying both the PCA technique and the SOM maps, a reduction in the size of the input patterns can be achieved. It results in a final dimension of the input vector of 15 parameters (see Table 5).


**Table 5.** Selected variables that finally composes the input vector for the ANN modelling.

<sup>1</sup> Windward side.

#### *3.2. Model Selection*

The results obtained after the training process of the different architectures tested for each model, show better performance of the aggregate model (Model I) over the disaggregated model (Model II), both in terms of error and correlation, as shown in Tables 6–8, and Figure 6, in which is possible to distinguish the results for each of the subsets used in the cross-verification process: Training (TR), Verification (V), Test (T) of the better Model I.


**Table 6.** Model I: Results in the test subset based on the number of neurons in the hidden layer. MSE

**Table 6.** Model I: Results in the test subset based on the number of neurons in the hidden layer.

**N° ud. 5 10 15 20 25 30 35**  MSE 7.02 × 10−5 6.75 × 10−5 5.69 × 10−5 5.72 × 10−5 3.82 × 10−5 5.06 × 10−5 7.08 × 10−<sup>5</sup> *r* 96.51 96.28 97.33 97.17 98.17 97.68 97.28

The finally selected architecture for Model I, based on the results obtained, is an MLP network with 15 input variables, 25 neurons in the hidden layer, and a single neuron in the output layer (see Tables 6–8, with the trial results to determine the best architecture in

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 17 of 25

the results of the different models proposed).

MSE and *r* statistic.


*r* 96.98 96.24 96.31 97.37 96.13 97.28 96.81

**Table 7.** Model II.I: Results in the test subset based on the number of neurons in the hidden layer. MSE and *r* statistic. **Table 8.** Model II.II: Results in the test subset based on the number of neurons in the hidden layer. MSE and *r* statistic.


**Table 8.** Model II.II: Results in the test subset based on the number of neurons in the hidden layer. MSE and *r* statistic. The results are shown in the form of correlation plots for test subset in Figure 6. Noting that for the test subset the correlation values are greater than 0.98. Although they are


**Figure 6.** Results on the proposed models. Correlation graphs are shown for each one and correspond to the test subset: (**a**) Model I; (**b**) Model II.1; (**b**) Model II.2. In each figure the dashed line corresponds to a perfect fit. Additionally, a linear adjustment of the results is shown for each model (only for test subset). **Figure 6.** Results on the proposed models. Correlation graphs are shown for each one and correspond to the test subset: (**a**) Model I; (**b**) Model II.1; (**b**) Model II.2. In each figure the dashed line corresponds to a perfect fit. Additionally, a linear adjustment of the results is shown for each model (only for test subset).

The analysis of the residuals establishes, as a desirable objective for an ideal model, that its distribution be carried out according to a pattern as close as possible to a normal distribution as a clear indicator of the absence of any hidden trend or bias in the modelling performed. In the present case, a careful analysis of this distribution shows that although The finally selected architecture for Model I, based on the results obtained, is an MLP network with 15 input variables, 25 neurons in the hidden layer, and a single neuron in the output layer (see Tables 6–8, with the trial results to determine the best architecture in the results of the different models proposed).

it is close to normal, it does not fit significantly to it. This is demonstrated by the chi-square The results are shown in the form of correlation plots for test subset in Figure 6. Noting that for the test subset the correlation values are greater than 0.98. Although they are similar to those obtained for sub model II.1 (0.96), they are much higher than those obtained with sub model II.2 (0.84). The results in terms of error (MSE) are similar for both model I (3.85 <sup>×</sup> <sup>10</sup>−<sup>5</sup> ) and model II (3.82 <sup>×</sup> <sup>10</sup>−<sup>5</sup> ), with the known exception that the MSE is not an absolute statistic, but a relative one [64].

The analysis of the residuals establishes, as a desirable objective for an ideal model, that its distribution be carried out according to a pattern as close as possible to a normal distribution as a clear indicator of the absence of any hidden trend or bias in the modelling performed. In the present case, a careful analysis of this distribution shows that although it is close to normal, it does not fit significantly to it. This is demonstrated by the chi-square

and Kolmogorov-Smirnov fit tests carried out, and that they are presented below (see Table 9), together with their correspondent graphical adjustment (Figure 7a). **Table 9.** Results of the different contrast test on the residuals for the selected model.

**Statistics Chi-Squared Kolmogorov-Smirnov** 

and Kolmogorov-Smirnov fit tests carried out, and that they are presented below (see Ta-

**Table 9.** Results of the different contrast test on the residuals for the selected model. Empirical Parameter 3.922 × 1015 0.161

ble 9), together with their correspondent graphical adjustment (Figure 7a).

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 18 of 25


**Figure 7.** Results for the selected model: (**a**) Graphical fit of the residual histogram to a normal distribution; (**b**) Scatter plot of the model residual versus outputs. **Figure 7.** Results for the selected model: (**a**) Graphical fit of the residual histogram to a normal distribution; (**b**) Scatter plot of the model residual versus outputs.

The graphical analysis of the scatter plot of the residuals (Figure 7b) shows adequate behavior across the entire response range, except in the range of low values of overtopping rate, for which it does show a certain tendency towards non-compliance with the hypothesis of constant variance of the residuals. This heteroscedasticity may be linked to scale problems in the tests or introduced by iso-energetic sequences of waves [60,61], since the behavior of the prediction for very low overtopping rates has been associated with high levels of uncertainty [69], or may be due to the need to perform further specific trans-The graphical analysis of the scatter plot of the residuals (Figure 7b) shows adequate behavior across the entire response range, except in the range of low values of overtopping rate, for which it does show a certain tendency towards non-compliance with the hypothesis of constant variance of the residuals. This heteroscedasticity may be linked to scale problems in the tests or introduced by iso-energetic sequences of waves [60,61], since the behavior of the prediction for very low overtopping rates has been associated with high levels of uncertainty [69], or may be due to the need to perform further specific transformations on the input variables beyond those already applied in the present study [25].

formations on the input variables beyond those already applied in the present study [25]. An extra validation test performed on an additional sample of 100 patterns, and with the selected ANN, provides good performance, with correlations of 0.98, which could validate its generalizability. However, what is more interesting, after classifying the component patterns into two different classes, the first one corresponding to tests on seawalls, and the second corresponding to sloped breakwaters, they show a similar aspect and is very suitable for the prediction of the overtopping rate on both (see Figure 8). Thus, the An extra validation test performed on an additional sample of 100 patterns, and with the selected ANN, provides good performance, with correlations of 0.98, which could validate its generalizability. However, what is more interesting, after classifying the component patterns into two different classes, the first one corresponding to tests on seawalls, and the second corresponding to sloped breakwaters, they show a similar aspect and is very suitable for the prediction of the overtopping rate on both (see Figure 8). Thus, the results for the typology of seawalls provides a correlation coefficient of 0.996, while for slope breakwaters it provides a similar result of 0.998.

results for the typology of seawalls provides a correlation coefficient of 0.996, while for

slope breakwaters it provides a similar result of 0.998.

**Figure 8.** Extra-validation results on the selected model: (**a**) Correlation graph on the seawall models; (**b**) Correlation graph on the slope breakwaters models. **Figure 8.** Extra-validation results on the selected model: (**a**) Correlation graph on the seawall models; (**b**) Correlation graph on the slope breakwaters models.

#### *3.3. Sensitivity Analysis 3.3. Sensitivity Analysis*

Finally, a sensitivity analysis is carried out on the selected ANN, specifically on the component parameters of the input vector. This analysis is performed using a pruning technique, and the ratio that was proposed for this purpose: Finally, a sensitivity analysis is carried out on the selected ANN, specifically on the component parameters of the input vector. This analysis is performed using a pruning technique, and the ratio that was proposed for this purpose:

$$r\_s = \frac{er\_j}{er\_T} \tag{12}$$

(12)

where x′ is the sensibility ratio, and the error function value for the trained network. where x 0 is the sensibility ratio, and *er<sup>j</sup>* the error function value for the trained network. In this case, the MSE is chosen as the error criterion to define the sensitivity ratio.

௦ <sup>=</sup> ்

In this case, the MSE is chosen as the error criterion to define the sensitivity ratio. This procedure is especially useful when the input variables are essentially independent of each other [64], and conversely, the more interdependencies there are between the variables, the less reliable they will be. Hence, among other reasons, the importance of the previously performed dimensionality reduction procedure, which now supports This procedure is especially useful when the input variables are essentially independent of each other [64], and conversely, the more interdependencies there are between the variables, the less reliable they will be. Hence, among other reasons, the importance of the previously performed dimensionality reduction procedure, which now supports the application of that sensitivity analysis.

the application of that sensitivity analysis. It is observed in the Figure 9 that all the variables have a significance ratio greater than 1.05, which according to that criterion implies that all the variables are significant a priori, and that therefore it would be desirable to maintain them for adequate network performance. The above has a clear derivative, which is to suppose that the process of dimensionality reduction has been successful, since any of the existing variables will pro-It is observed in the Figure 9 that all the variables have a significance ratio greater than 1.05, which according to that criterion implies that all the variables are significant a priori, and that therefore it would be desirable to maintain them for adequate network performance. The above has a clear derivative, which is to suppose that the process of dimensionality reduction has been successful, since any of the existing variables will provide enough relevant information and their elimination may imply worse predictive capacity of the network.

vide enough relevant information and their elimination may imply worse predictive capacity of the network. Beyond the previous observations, it is noted that the most influential variable is the freeboard of the wall with respect to swl (*Rc*), an issue that is confirmed by PCA analysis. It is noteworthy that a variable closely related to it, the other freeboard parameter, the crest height with respect to swl (*Ac*), is quite far, in terms of significance, from the parameter *Rc*. This fact is relevant since in some works [59] it has been determined that the scale effect seems to depend a lot on the superior geometry of the breakwater. This results in many more significant associated effects on small overtopping rates, which are incidentally, also the most numerous in the database. Given this, and to try to mitigate these effects as much as possible, some authors propose the dimensionless of these variables [25,30,59].

**Figure 9.** Sensibility analysis. The histogram represents the contribution of each variables of the

els; (**b**) Correlation graph on the slope breakwaters models.

technique, and the ratio that was proposed for this purpose:

௦ <sup>=</sup> ்

the application of that sensitivity analysis.

model input space in terms of significance of the *rs* ratio.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 20 of 25

*3.3. Sensitivity Analysis* 

pacity of the network.

(**a**) (**b**)

**Figure 8.** Extra-validation results on the selected model: (**a**) Correlation graph on the seawall mod-

Finally, a sensitivity analysis is carried out on the selected ANN, specifically on the component parameters of the input vector. This analysis is performed using a pruning

where x′ is the sensibility ratio, and the error function value for the trained network.

This procedure is especially useful when the input variables are essentially independent of each other [64], and conversely, the more interdependencies there are between the variables, the less reliable they will be. Hence, among other reasons, the importance of the previously performed dimensionality reduction procedure, which now supports

It is observed in the Figure 9 that all the variables have a significance ratio greater than 1.05, which according to that criterion implies that all the variables are significant a priori, and that therefore it would be desirable to maintain them for adequate network performance. The above has a clear derivative, which is to suppose that the process of dimensionality reduction has been successful, since any of the existing variables will provide enough relevant information and their elimination may imply worse predictive ca-

In this case, the MSE is chosen as the error criterion to define the sensitivity ratio.

(12)

**Figure 9.** Sensibility analysis. The histogram represents the contribution of each variables of the model input space in terms of significance of the *rs* ratio. Overall, the results are consistent in terms of the significance of these parameters with similar studies carried out with different preprocessing techniques [30]. And it should be mentioned that some parameters in this study are may be penalized, due to the

Another significantly interesting variable is the average cotangent where the contribution of the berm (*cotαincl*) is considered [42,59]. Similarly of interest, the wave steepness (*Hm*0*/Lm*−1) is highlighted. In addition to these, are both parameters related to roughness (and in essence, to the porosity of the mantle) where their close relationship with overtopping is already known empirically [32], and which in turn have a substantial dependence on the dimensionless freeboard (*Rc/Hm*0). fact that they have been poorly represented in the database. For example, this happens with the wave incidence angle parameter (*β*) that shows a significant lack of data in some ranges of that continuous variable. In the following figure (see Figure 10a) the distribution of the aforementioned parameter with respect to the significant wave height at the toe structure (*Hm*o *t*) is presented since, as Van der Meer cites [48], this relationship is strongly related to the overtopping phenomenon, and shows the existence of poor representability

Overall, the results are consistent in terms of the significance of these parameters with similar studies carried out with different preprocessing techniques [30]. And it should be mentioned that some parameters in this study are may be penalized, due to the fact that they have been poorly represented in the database. For example, this happens with the wave incidence angle parameter (*β*) that shows a significant lack of data in some ranges of that continuous variable. In the following figure (see Figure 10a) the distribution of the aforementioned parameter with respect to the significant wave height at the toe structure (*Hm*<sup>o</sup> *<sup>t</sup>*) is presented since, as Van der Meer cites [48], this relationship is strongly related to the overtopping phenomenon, and shows the existence of poor representability in the ranges greater than 50◦ . in the ranges greater than 50°. The importance given to the dimensionless parameter of the wave steepness, particularly for wave overtopping energy conversion [70], which has good representation in the database, both in its distribution and in the quality of that distribution (normalized distribution), should also be highlighted (see Figure 10b). Faced with possible uncertainties associated with scale phenomena [59], although the parameter's existence in the field of validity is remarkable with values over 0.07 that are physically not possible as the wave breaks on steepness [56], the use of wave steepness as a variable is recommended. This also represents the effects induced by local breakage and waves [25], and is therefore strongly related with the overtopping.

**Figure 10.** Analysis of the model sample: (**a**) Scatterplot of the parameter *β* vs. *Hm*<sup>o</sup> *<sup>t</sup>* ; (**b**) Histogram of the *<sup>S</sup>m*−1,0 parameter (spectral wave steepness), with adjustment of a normal probability density function.

The importance given to the dimensionless parameter of the wave steepness, particularly for wave overtopping energy conversion [70], which has good representation in the database, both in its distribution and in the quality of that distribution (normalized distribution), should also be highlighted (see Figure 10b). Faced with possible uncertainties associated with scale phenomena [59], although the parameter's existence in the field of validity is remarkable with values over 0.07 that are physically not possible as the wave breaks on steepness [56], the use of wave steepness as a variable is recommended. This also represents the effects induced by local breakage and waves [25], and is therefore strongly related with the overtopping. happen, and is shown in the following figure (Figure 11a). Checking the model for those exceedance rates corresponding to the range in the previously mentioned sample of 100 extra cases (corresponding to a total of 57 cases), the result is encouraging, with values of the correlation coefficient greater than 0.98, as shown in the Figure 11b. Thus, and in accordance with the above mentioned, any future improvement in the model should necessarily focus on that data range. This desired approach is in practice the opposite of what is usually done for defense structures. Another crucial issue related to the generation of the data is the need to make the

Due to the fact that for wave overtopping conversion the maximum overtopping rates correlated with the lower *Rc/Hs* relationships [71] are highly desirable. These will generally be associated with low crested structures, specifically with *Rc/Hs* lower than 1. It would be desirable that the training sample be well represented in this range, as does

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 21 of 25

**Figure 10.** Analysis of the model sample: (**a**) Scatterplot of the parameter *β* vs. *Hm*o *t*; (**b**) Histogram of the *Sm*−1*,*0 parameter

(spectral wave steepness), with adjustment of a normal probability density function.

Due to the fact that for wave overtopping conversion the maximum overtopping rates correlated with the lower *Rc/H<sup>s</sup>* relationships [71] are highly desirable. These will generally be associated with low crested structures, specifically with *Rc/H<sup>s</sup>* lower than 1. It would be desirable that the training sample be well represented in this range, as does happen, and is shown in the following figure (Figure 11a). Checking the model for those exceedance rates corresponding to the range in the previously mentioned sample of 100 extra cases (corresponding to a total of 57 cases), the result is encouraging, with values of the correlation coefficient greater than 0.98, as shown in the Figure 11b. range of data tested wide enough to include extraordinary events, given that ANNs are usually unable to extrapolate beyond the range of the data used for training [65,72]. This ensures that they always work in the expected range, avoiding poor predictions when the validation data contain values outside of the range of those used for training. In this sense, there is a preponderance of low flow rates that reinforces the idea of a disaggregated approach in future models.

**Figure 11.** Results on the selected model for the optimum specific range for wave overtopping conversion: (**a**) Vertical freeboard (*Rc*) vs. overtopping rate relationship in the model sample; (**b**) Correlation graph for the *Rc/Hs* < 1 specific range. **Figure 11.** Results on the selected model for the optimum specific range for wave overtopping conversion: (**a**) Vertical freeboard (*Rc*) vs. overtopping rate relationship in the model sample; (**b**) Correlation graph for the *Rc/Hs* < 1 specific range.

**4. Conclusions**  As part of a sustainable strategy to take advantage of some existing breakwater in-Thus, and in accordance with the above mentioned, any future improvement in the model should necessarily focus on that data range. This desired approach is in practice the opposite of what is usually done for defense structures.

frastructure, and its partial reconversion as a Wave Energy Converter while maintaining its defense purpose, this study is framed in which a model based on artificial neural networks for the overtopping rate forecasting is proposed for a wide range of breakwaters. The adjusted prediction of the overtopping rate constitutes the first step in the study of subsequent modifications to be made to these structures. To achieve this purpose, existing data from CLASH-EurOtop have been subjected to Another crucial issue related to the generation of the data is the need to make the range of data tested wide enough to include extraordinary events, given that ANNs are usually unable to extrapolate beyond the range of the data used for training [65,72]. This ensures that they always work in the expected range, avoiding poor predictions when the validation data contain values outside of the range of those used for training. In this sense, there is a preponderance of low flow rates that reinforces the idea of a disaggregated approach in future models.

a preprocessing step, where only the parameters from laboratory tests have been previously scaled according to the Froude model law (to *Hm*0 *toe* = 1 m). Subsequently, the entire

#### **4. Conclusions**

As part of a sustainable strategy to take advantage of some existing breakwater infrastructure, and its partial reconversion as a Wave Energy Converter while maintaining its defense purpose, this study is framed in which a model based on artificial neural networks for the overtopping rate forecasting is proposed for a wide range of breakwaters. The adjusted prediction of the overtopping rate constitutes the first step in the study of subsequent modifications to be made to these structures.

To achieve this purpose, existing data from CLASH-EurOtop have been subjected to a preprocessing step, where only the parameters from laboratory tests have been previously scaled according to the Froude model law (to *Hm*<sup>0</sup> *toe* = 1 m). Subsequently, the entire data base was subjected to an extensive process of exploration, debugging, and dimensionality reduction, until an optimized input pattern in the ANN model was obtained. Using only 15 of the 34 initial features, sufficient relevant information was used to train a model with generalization skills and high predictive efficiency. This preliminary phase derives substantial conclusions such as:


Several ANN models have been proposed and the architecture finally selected has been an MLP 15-25-1. This was obtained after a cross-verification training process with the Levenberg-Marquardt algorithm, and which corresponds to a model that takes into account both the data from prototypes and small-scale tests.

The results are very encouraging since they allow obtaining predictions with very high correlation coefficients (>0.98) and where the validation process carried out shows that the model is equally suitable for both seawalls and slope breakwaters. This has been justified by the prevalence of those that refer to crest freeboard of the structure with respect to swl over the rest of the parameters, and the average cotangent of the slope of the structure considering the contribution of the berm.

This final conclusion reinforces the belief that subsequent studies, in which an adequate classification criterion of the input parameters will be obtained, will undoubtedly reinforce the good performance of the ANN model. For wave energy conversion, the lower *Rc/H<sup>s</sup>* relationship the higher overtopping rate, therefore this criterion will allow future models to be developed and trained in that specific range of patterns.

**Author Contributions:** J.M.O. contributed to the work described in this paper by carrying out the analysis and treatment of the data, proposing and analyzing the models and results, and writing— Original Draft preparation. M.D.E. contributed to the work by supervising the methodology and the obtained results and gave the final approval. J.-S.L.-G. contributed with knowledgeable discussion and suggestion. V.N. and M.G.N. contributed with knowledgeable discussion and suggestion. All of the co-authors participated in editing the final paper. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors acknowledge the assignment of the EurOtop 2018 Database, specially to Formentin, Z.M., Zanuttigh, B., and Van der Meer, J.W.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Probabilistic Assessment of Hybrid Wind-PV Hosting Capacity in Distribution Systems**

**Dichen Liu <sup>1</sup> , Chenxu Wang 1,\*, Fei Tang 1,\* and Yixi Zhou <sup>2</sup>**


Received: 14 February 2020; Accepted: 10 March 2020; Published: 11 March 2020

**Abstract:** In recent years, hybrid wind-photovoltaic (PV) systems are flourishing due to their advantages in the utilization of renewable energy. However, the accurate assessment of the maximum integration of hybrid renewable generation is problematic because of the complex uncertainties of source and demand. To address this issue, we develop a stochastic framework for the quantification of hybrid energy hosting capacity. In the proposed framework, historical data sets are adopted to represent the stochastic nature of production and demand. Moreover, extreme combinations of production and demand are introduced to avoid multiple load flow calculations. The proposed framework is conducted in the IEEE 33-bus system to evaluate both single and hybrid energy hosting capacity. The results demonstrate that the stochastic framework can provide accurate evaluations of hosting capacity while significantly reducing the computational burden. This study provides a comprehensive understanding of hybrid wind-PV hosting capacity and verifies the excellent performance of the hybrid energy system in facilitating integration and energy utilization.

**Keywords:** Hybrid energy system; wind power; photovoltaic; hosting capacity; distribution system

#### **1. Introduction**

Over the past decades, the utilization of renewable energy sources (RES) has grown dramatically due to concerns over the environment and carbon emissions [1]. Wind power and photovoltaic (PV) comprise a significant proportion of RES due to mature technologies, economic advantages, and abundant primary energy. The strategical deployment of RES in distribution systems as distributed generation (DG) can bring serval benefits [2,3], such as voltage profile improvement, power loss reduction, and reliability enhancement. However, the excessive penetration of DG can alter the normal operational behavior of distribution systems and cause various negative impacts, such as overvoltage [4,5], voltage unbalance [6], thermal overloading [7], and harmonic distortions [8,9]. The performances of systems becomes unacceptable when the amount of DG exceeds the hosting capacity [10]. For this reason, it is of paramount importance to system planners to determine the maximum amount of DG that can be connected.

Hosting capacity is defined as the amount of DG that can be connected without endangering the reliability or voltage quality of systems [11]. The idea of hosting capacity was introduced in 2004, and Math Bollen et al. [12] further developed this concept. Throughout the next few years, hosting capacity analysis was regarded as a transparent tool for the planning and design of distribution networks. Furthermore, based on the time-varying nature of generation and demand, Castelo de Oliveira et al. [13] emphasized that the hosting capacity needs to be analyzed throughout different periods, and developed the concept of dynamic hosting capacity. The analysis of dynamic hosting capacity requires time series data, e.g., forecasting the values of generation and demand [14,15]. In this

context, the combined dynamic hosting capacity with forecasting data has the potential to predict and facilitate the operation of systems.

In addition to the development of the concept, abundant studies have been conducted using various techniques for the assessment and enhancement of hosting capacity. To provide guidelines for system planners and researchers on various assessment techniques, the authors in [16] presented a comprehensive overview of three different methods for quantifying PV hosting capacity, i.e., deterministic [17], stochastic [18,19], and time series methods [20]. The main differences between the three categories of methods are whether and how uncertainties are considered. Deterministic methods usually adopt the worst-case scenario (e.g., maximum production and minimum load) to evaluate the extreme impact of DG. Therefore, these methods may lead to conservative results. To realistically assess the states of systems, the stochastic and time series methods adopt probability distributions and time-series data, respectively, to represent uncertainties of source and demand.

The accurate analysis of hosting capacity involves, not only the variations of DG production and load demands, but also the uncertainties of DG location and rated power [16]. In this context, Smith et al. [21] developed a Monte Carlo simulation-based (MCS) stochastic method to provide a comprehensive assessment with various uncertainties. This stochastic method takes into account the uncertainties in the size and location of DG by generating thousands of potential DG deployment schemes. Then, load flow calculations are performed to address the stochastic nature associated with DG production and load consumption. For a specific deployment scheme, the hosting capacity result can be obtained if any technological limitations are violated. With the multiple potential DG deployment schemes, the final result is no longer one exact value but a probability distribution.

Based on the stochastic method in [21], some applications and variations have been developed. This method was conducted in [22] to estimate the PV hosting capacity of 16 utility distribution feeders. Compared with the 15% rule adopted by system planners [10] (i.e., the total DG rating should be lower than 15% of the feeder annual peak load), the stochastic analysis provides more realistic estimations of hosting capacity. Moreover, some studies [23,24] adopted the stochastic method to perform the sensitivity analysis of hosting capacity to the characteristics of the feeder (e.g., voltage class, peak load demand, and voltage regulation equipment) and PV generation (e.g., location and power factor). Dubey et al. [25] presented a comprehensive understanding of the PV hosting capacity problem by developing a mathematical formulation and an hourly stochastic analysis framework of hosting capacity assessment. Given that the analysis framework is stochastic, an approach was also established to quantify the accuracy of the obtained results. A simplified MCS-based method was developed in [26] to estimate the PV hosting capacity of 50000 real low-voltage systems from a distribution utility in Brazil. The obtained statistical results provide a risk-based guide to determine strategies to deal with increasing PV penetration. Instead of using load flow calculations, Abad et al. [27] proposed an optimization model of hosting capacity and embedded this model into the stochastic framework to address the variability of production and consumption.

The literature review in the realm of hosting capacity assessment illustrates that abundant methods have been developed for PV hosting capacity problems. These studies have contributed significantly to the understanding of the impact of different variables on PV hosting capacity. However, less research work has been done on the hybrid wind-PV hosting capacity assessment. Wind power and PV, to some extent, are complementary [28,29]. Therefore, the hybrid wind-PV system may have the potential to generate smoother production than single renewable generation. Some studies have reported the advantages of the hybrid wind-PV system in facilitating energy production [30] and enhancing power supply reliability [31]. In this context, if the complementarity can be captured through the hybrid deployment of wind power and PV, distribution systems could have the potential to accommodate more DG capacity. More importantly, hybrid renewable energy can provide more power for demand, which helps achieve the goal of reducing carbon emissions.

Since the output of wind power and PV have different characteristics [20], the methods for hybrid wind-PV hosting capacity assessment can be different compared to those methods for only PV. The most significant difference is how to consider the uncertainties of DG production. For PV generation, the maximum output usually happens at midday due to solar irradiation. Therefore, only critical periods, in terms of potential limit violations, are evaluated in some studies. For instance, the authors in [23,26] only considered periods 10 am-2 pm and 11 am-1 pm, respectively. The adoption of critical periods can significantly reduce the computational burden while ensuring the accuracy of assessments. However, it is difficult to determine such periods for wind power. Under such circumstances, the evaluation of hybrid wind-PV hosting capacity can be very time-consuming due to the need for a large number of load flow calculations to capture the potential technological violations.

Motivated by the research gap and challenges, this paper aims to present a comprehensive assessment of hybrid wind-PV hosting capacity in distribution systems. To achieve this goal, we develop a stochastic framework that takes into account the uncertainties of load demands, DG production, location, size, and type. Moreover, to avoid multiple load flow calculations, the extreme combinations of DG production and load demands are defined. The proposed framework is implemented to evaluate both single and hybrid DG hosting capacity. Compared with the comprehensive load flow method, both the efficiency and accuracy of the proposed framework are verified. The simulation results provide system planners with an understanding of hosting capacity and energy utilization of the hybrid wind-PV system from the probabilistic perspective.

The paper organization is as follows. Section 2 presents the modeling of time-varying DG production and load demands. Section 3 introduces the stochastic framework for DG hosting capacity assessment. Section 4 performs the proposed framework on the IEEE 33-bus system to evaluate single and hybrid DG hosting capacity. Section 5 presents the discussion on the main features and applications of the proposed framework. Finally, Section 6 summarizes the main conclusions of this study.

#### **2. Time-Varying Renewables and Demand**

In hosting capacity assessment, the time-varying characteristics of both generation and demand need to be considered in evaluating the operating state of the systems. In this context, some studies [32,33] adopted the historical data of generation and demand as inputs to perform time series analysis to identify potential constraint violations. In general, time series analysis requires historical data sets with a long time scale and high-resolution. On the one hand, long time scale historical data, such as one year or a few years [16], is needed to capture various combinations of production and demand. On the other hand, a relatively short time interval [34] (e.g., 1-min, 10-min, and 15-min) is critical for capturing the variabilities of these uncertainties. However, the adoption of time series analysis introduces a significant number of power flow calculations into assessments. For instance, a time series analysis on the data set of one year at the interval of 15-min requires 35,040 power flow simulations to cover the complete assessment. Therefore, the significant computational burden makes time series analysis laborious and intractable.

One of the solutions to improve the efficiency of host capacity assessment is to extract the historical data that has the most significant impact on system constraints. In this context, Ochoa et al. [35] proposed a two-step (i.e., discretization and aggregation steps) technique to allocate the historical data of wind power and load demand into a finite number of bins based on their joint probability of occurrence. Subsequently, Sun et al. [30] further developed this technique by addressing the "coincidence" of wind power, PV, and load.

To outline the procedure of the two-step technique, we adopt wind power and load demand as an example. In Figure 1a, the curves present a 10-day snapshot of 15-min wind power and demand data with the values normalized against respective peak values. The discretization process allocates the time-dependent wind power and demand data into a series of bins. For instance, when the width of the bins is set to 0.1 p.u, the data points 0.43 p.u (wind power) and 0.64 p.u (load demand) are allocated into bins (0.40 p.u, 0.50 p.u] and (0.60 p.u, 0.70 p.u], respectively. Then, the aggregation process groups the bins into multiple combinations, e.g., {(0.40 p.u, 0.50 p.u] and (0.60 p.u, 0.70 p.u]}, as shown as the two-dimensional mesh in Figure 1b, using the above processes to sweep all data points and allocate

them into corresponding combinations. Since load demand is never below 0.40 p.u, only 52 out of 100 potential combinations have non-zero probabilities of occurrence. These combinations are labeled with " √ " in Figure 1c. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 4 of 19 p.u, only 52 out of 100 potential combinations have non-zero probabilities of occurrence. These combinations are labeled with "**√**" in Figure 1c.

**Figure 1.** The procedure of discretization and aggregation of historical data: (**a**) historical data of wind power and load; (**b**) scatter plot of wind power and load; (**c**) visualization of wind power-load combinations. **Figure 1.** The procedure of discretization and aggregation of historical data: (**a**) historical data of wind power and load; (**b**) scatter plot of wind power and load; (**c**) visualization of wind power-load combinations.

The discretization-aggregation technique allocates the time series data into a finite number of combinations, thus reducing the number of load flow calculations. More importantly, this technique can preserve the interrelationships between production and demand. However, two challenges still need to be addressed when applying this technique in the hosting capacity assessment. The first is to determine which combinations are most critical in driving the system constraints. Second, the selection of bin width is empirical. Therefore, it is imperative to evaluate the impact of different bin The discretization-aggregation technique allocates the time series data into a finite number of combinations, thus reducing the number of load flow calculations. More importantly, this technique can preserve the interrelationships between production and demand. However, two challenges still need to be addressed when applying this technique in the hosting capacity assessment. The first is to determine which combinations are most critical in driving the system constraints. Second, the selection of bin width is empirical. Therefore, it is imperative to evaluate the impact of different bin widths on hosting capacity results.

widths on hosting capacity results. For the first challenge, consider the combinations labeled with the red "**√**" in Figure 1c, which are characterized by relatively high generation and low demand. If technical constraints are not violated in these combinations, they are unlikely to be violated in other combinations. Therefore, these combinations can represent the worst-case scenarios in determining the hosting capacity, and we define them as extreme combinations. For the second challenge, different bin widths need to be considered. Figure 2 presents the combinations of wind power and load demand when the bin width is set to 0.05 p.u. It is evident that the total number of potential combinations increases significantly compared with the results in Figure 1c, but the number of extreme combinations is almost the same. The adoption of a smaller bin width can generate more compact intervals to be evaluated. Therefore, the selection of bin width may affect the accuracy of hosting capacity assessment, and the quantitative analysis of this impact is presented in Section 4. For the first challenge, consider the combinations labeled with the red "√ " in Figure 1c, which are characterized by relatively high generation and low demand. If technical constraints are not violated in these combinations, they are unlikely to be violated in other combinations. Therefore, these combinations can represent the worst-case scenarios in determining the hosting capacity, and we define them as extreme combinations. For the second challenge, different bin widths need to be considered. Figure 2 presents the combinations of wind power and load demand when the bin width is set to 0.05 p.u. It is evident that the total number of potential combinations increases significantly compared with the results in Figure 1c, but the number of extreme combinations is almost the same. The adoption of a smaller bin width can generate more compact intervals to be evaluated. Therefore, the selection of bin width may affect the accuracy of hosting capacity assessment, and the quantitative analysis of this impact is presented in Section 4.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 5 of 19

**Figure 2.** Visualization of wind power-load combinations. **Figure 2.** Visualization of wind power-load combinations.

#### **3. Framework for Hosting Capacity Assessment 3. Framework for Hosting Capacity Assessment**

In this section, we develop a stochastic framework for hosting capacity assessment. The determination of hosting capacity requires clear performance limits as criteria. Note that the most restrictive impact of DG integration is bus overvoltage [23,26]. This paper primarily considers the voltage limits to determine the hosting capacity. The proposed approach can also be extended to consider other criteria, such as conductor thermal capacity and transformer overload [26]. In this section, we develop a stochastic framework for hosting capacity assessment. The determination of hosting capacity requires clear performance limits as criteria. Note that the most restrictive impact of DG integration is bus overvoltage [23,26]. This paper primarily considers the voltage limits to determine the hosting capacity. The proposed approach can also be extended to consider other criteria, such as conductor thermal capacity and transformer overload [26].

#### Two types of variables need to be considered in hosting capacity assessment. The first category *3.1. Definition of Variables*

*3.1. Definition of Variables* 

of variables relates to the deployment of DG, including location, type, and the rated power of DG. The second category of variables includes DG production and load consumption. To consider the above variables, we define some terms as follows: 1) Location penetration (*Loci*): The definition of location penetration is the ratio of the number of selected DG locations *NS* to the number of all potential locations *NP*. This variable denotes the number Two types of variables need to be considered in hosting capacity assessment. The first category of variables relates to the deployment of DG, including location, type, and the rated power of DG. The second category of variables includes DG production and load consumption. To consider the above variables, we define some terms as follows:


values are considered to provide comparative results. Let the 4*d* be the set of bin widths, i.e., 4*d* = {0.10 p.u, 0.05 p.u, 0.025 p.u, 0.01 p.u, 0.005 p.u, 0.0025 p.u, 0.001 p.u}. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 6 of 19 determine the extreme combinations. Since the selection of bin width is empirical, seven different

#### *3.2. Stochastic Analysis Framework* values are considered to provide comparative results. Let the △*d* be the set of bin widths, i.e., △*d* = {0.10 p.u, 0.05 p.u, 0.025 p.u, 0.01 p.u, 0.005 p.u, 0.0025 p.u, 0.001 p.u}.

hosting capacity results.

In this section, the computational procedure of the proposed stochastic framework is developed. As shown in Figure 3, the proposed framework has three modules. Module 1 aims to generate various DG deployment schemes. Then, Module 2 conducts the impact analysis to obtain the hosting capacity results of different deployment schemes. Module 3 performs the statistical analysis of the obtained hosting capacity results. *3.2. Stochastic Analysis Framework*  In this section, the computational procedure of the proposed stochastic framework is developed. As shown in Figure 3, the proposed framework has three modules. Module 1 aims to generate various DG deployment schemes. Then, Module 2 conducts the impact analysis to obtain the hosting capacity results of different deployment schemes. Module 3 performs the statistical analysis of the obtained

**Figure 3.** The proposed stochastic framework for hosting capacity analysis.(**a**) Histogram of hosting capacity; (**b**) Cumulative probability curve of hosting capacity; (**c**) Histogram of total generation **Figure 3.** The proposed stochastic framework for hosting capacity analysis.(**a**) Histogram of hosting capacity; (**b**) Cumulative probability curve of hosting capacity; (**c**) Histogram of total generation

1) Deployment schemes of DG (1) Deployment schemes of DG

This module addresses the generation of multiple DG deployment schemes. The variables involved in this module include DG location penetration, rated power, and type. The steps of Module 1 are as follows: This module addresses the generation of multiple DG deployment schemes. The variables involved in this module include DG location penetration, rated power, and type. The steps of Module 1 are as follows:

Step 1: Determine the location penetration *Loci* ∈ *Locpen*, (*i* = 1, 2, …, 10); Step 1: Determine the location penetration *Loc<sup>i</sup>* ∈ *Locpen*, (*i* = 1, 2, . . . , 10);

Step 2: Perform MCS to generate *k* DG deployment schemes with location penetration level *Loci*. The set of deployment schemes is represented as *Si* = {*Si*1, …, *Sij*, …,*Sik*}; Step 3: Determine the initial total hosting capacity (e.g., 1 MW) of the test system. For each Step 2: Perform MCS to generate *k* DG deployment schemes with location penetration level *Loc<sup>i</sup>* . The set of deployment schemes is represented as *S<sup>i</sup>* = {*Si*1, . . . , *Sij*, . . . ,*Sik*};

deployment scheme, *Sij*, the initial DG rated power at each bus is allocated based on the corresponding peak load demand; Step 4: Determine the DG type at each selected bus (e.g., 50% wind, 50% PV). Step 3: Determine the initial total hosting capacity (e.g., 1 MW) of the test system. For each deployment scheme, *Sij*, the initial DG rated power at each bus is allocated based on the corresponding peak load demand;

Step 4: Determine the DG type at each selected bus (e.g., 50% wind, 50% PV).

(2) Assessment of hosting capacity

This module aims to analyze the impact of DG production on the operating states of systems. The hosting capacity of each deployment scheme is determined by performing deterministic power

flow calculations on extreme combinations. Therefore, the variable considered in this module is the bin width 4*d*. The steps of Module 2 are as follows:

Step 5: Determine the bin width 4*d*, and obtain the extreme combinations to be evaluated;

Step 6: For a specific deployment scheme, *Sij*, perform deterministic power flow calculations on the extreme combinations to estimate the upper bounds of voltages. For instance, if the extreme combination is {(0.65 p.u, 0.70 p.u], (0.45 p.u, 0.50 p.u]}, the upper bounds of voltages can be obtained with DG generation and load demand set to 0.70 p.u and 0.45 p.u, respectively;

Step 7: If none of the bus voltages exceed the technical constraints, i.e., 1.05 p.u, the total hosting capacity is increased by a fixed step size (e.g., 0.01 MW) and the procedure is repeated from Step 6. Otherwise, the procedure is terminated and the hosting capacity result, *HC<sup>j</sup>* , is obtained for the test system with deployment scheme *Sij*;

Step 8: Repeat steps 6–7 to obtain the hosting capacity results for each deployment scheme in *S<sup>i</sup>* . The hosting capacity results are represented as *HC* = {*HC*1, . . . *HC<sup>j</sup>* , . . . ,*HC<sup>k</sup>* }.

(3) Analysis of hosting capacity results

In Module 3, statistical analysis of the obtained hosting capacity results is performed. The results of this module are as follows:

Histogram of hosting capacity: The histogram of hosting capacity can be obtained based on the results *HC* = {*HC*1, . . . *HC<sup>j</sup>* , . . . ,*HC<sup>k</sup>* }. As shown in Figure 3a, the results provide an understanding of the probabilistic hosting capacity of the test system with a specific location penetration.

Cumulative probability curve of hosting capacity: This curve can be obtained based on the histogram of hosting capacity. As shown in Figure 3b, the curve is presented in a descending trend. This curve helps system planners to estimate the probability of having a hosting capacity higher than a specific value.

Histogram of total generation: The total energy generation of DG is a valuable indicator for the utilization of renewable energy. Therefore, the histogram of energy generation in one year, as shown in Figure 3c, is calculated to provide a comparison of the energy utilization of different types of DG.

#### **4. Numerical Results**

In this section, numerical studies are carried out on the IEEE 33-bus distribution system to demonstrate the efficacy of the proposed framework. Firstly, brief introductions to the test system and historical data sets of load demand, wind, and PV production are presented. Then, two critical parameters (bin width and trail number) in the stochastic framework are determined through comparisons with the comprehensive assessment. Following that, detailed assessments of both single and hybrid DG hosting capacity are performed.

#### *4.1. Data*

The proposed framework is evaluated on the IEEE 33-bus distribution network, whose single line diagram is presented in Figure 4. We set the voltage of the grid supply point to 1.0 p.u, and the lower and upper bounds of voltage for each bus are 0.95 p.u and 1.05 p.u, respectively. The total peak demand for this test system is 3.715 MW. The detailed parameters and configurations of the test system are available in [36].

The historical data of load demand, wind speed, and solar irradiation are derived from a typical distribution system in Hubei province, China. The data set of one year at intervals of 15 min has a total of 35,040 data points. Load demand is normalized against its peak value. Wind speed and solar irradiation are processed and applied to their characteristic curves [37], respectively. Figure 5 presents the variations of load demand, wind power, and PV production in the whole year. It can be seen from the curves that both load demand and PV output have visible seasonal patterns. The load in summer is relatively lower than in winter, while PV output shows the opposite trend. However, the pattern of wind power is less clear than the load and PV output. In the test system, different buses

are geographically close. Therefore, it is reasonable to assume that the production of the same type of DG follow the same time series curve. the pattern of wind power is less clear than the load and PV output. In the test system, different buses are geographically close. Therefore, it is reasonable to assume that the production of the same type of DG follow the same time series curve. are geographically close. Therefore, it is reasonable to assume that the production of the same type of DG follow the same time series curve. 10 11 12 13 15 14 16 17 18

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 19

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 19

in summer is relatively lower than in winter, while PV output shows the opposite trend. However,

the pattern of wind power is less clear than the load and PV output. In the test system, different buses

**Figure 4.** Single line diagram of the IEEE 33-bus distribution system. **Figure 4.** Single line diagram of the IEEE 33-bus distribution system. **Figure 4.** Single line diagram of the IEEE 33-bus distribution system.

(**c**) **Figure 5.** One-year time series data: (**a**) load demand; (**b**) wind power; (**c**) PV power.

The discretization-aggregation technique in Section 2 is adopted to establish the combinations of renewables and load demands. Figure 6 shows the coincident periods for wind power-load, PV-load, and wind power-PV when the bin width is selected as 0.05 p.u. For the 400 potential combinations of wind power-load, only 175 have non-zero probabilities of occurrence. Similarly, only 133 combinations are non-zero in the PV-load case. For the coincidence of renewable energies, there is a negative correlation between wind power and PV, and the correlation coefficient between two resources is −0.162. The PV output mainly depends on solar irradiation. As a result, solar power is negligible during nighttime. In contrast, wind power during daytime is typically less than at nighttime. The complementarity between wind and solar power has the potential to facilitate energy integration. combinations of wind power-load, only 175 have non-zero probabilities of occurrence. Similarly, only 133 combinations are non-zero in the PV-load case. For the coincidence of renewable energies, there is a negative correlation between wind power and PV, and the correlation coefficient between two resources is −0.162. The PV output mainly depends on solar irradiation. As a result, solar power is negligible during nighttime. In contrast, wind power during daytime is typically less than at nighttime. The complementarity between wind and solar power has the potential to facilitate energy integration. In hosting capacity assessment, the occurrences of relatively high production and low demand are critical in driving in system constraints. Table 1 presents the extreme combinations to be evaluated for wind power and PV hosting capacity. For wind power-load and PV-load cases, only three and four extreme combinations need to be considered, respectively. Compared with the historical dataset, the introduction of extreme combinations can significantly reduce the number of load flow

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 9 of 19 **Figure 5.** One-year time series data: (**a**) load demand; (**b**) wind power; (**c**) PV power.

load, and wind power-PV when the bin width is selected as 0.05 p.u. For the 400 potential

**Figure 6.** Coincident periods for (**a**) wind power-load; (**b**) PV-load; (**c**) PV-wind power. **Figure 6.** Coincident periods for (**a**) wind power-load; (**b**) PV-load; (**c**) PV-wind power.

**Table 1.** Extreme combinations for wind power-load and PV-load cases. **Renewables and Load Extreme Combinations {(renewables], (load]} (p.u)**  Wind Power-Load {(0.65 0.70], (0.45 0.50]}, {(0.75 0.80], (0.50 0.55]} In hosting capacity assessment, the occurrences of relatively high production and low demand are critical in driving in system constraints. Table 1 presents the extreme combinations to be evaluated for wind power and PV hosting capacity. For wind power-load and PV-load cases, only three and four extreme combinations need to be considered, respectively. Compared with the historical dataset, the introduction of extreme combinations can significantly reduce the number of load flow calculations.


**Table 1.** Extreme combinations for wind power-load and PV-load cases.

#### *4.2. Impacts of Simulation Parameters on Hosting Capacity*

Before the evaluation of hosting capacity, two critical parameters should be determined. One of them is the bin width, which is used to determine the extreme combinations of generation and demand. This parameter may have an impact on the accuracy of hosting capacity. Another parameter is the trial number of MCS, which represents the number of DG deployment schemes. Since the analysis framework is stochastic, this parameter may affect the accuracy of the probability distribution of the obtained results.

Firstly, the selection of different bin widths is conducted, and only wind power is considered to simplify the analysis. The proposed stochastic framework is implemented with seven different bin widths, 4*d*, i.e., 0.1 p.u, 0.05 p.u, 0.025 p.u, 0.01 p.u, 0.005 p.u, 0.0025 p.u, and 0.001 p.u. Moreover, the comprehensive load flow method is adopted to render comparative results. The computational procedure of comprehensive assessment is similar to Module 2 of the proposed framework, i.e., increase the total DG capacity until the limitation is violated. Still, the difference is that the former method performs load flow calculations with all historical data of wind power and load to capture the violation of system constraints. Therefore, the results of the comprehensive load flow method can be treated as the exact values.

Taking a specific DG deployment as an example, the locations of wind power integration are Bus 2, 7, 24, and 33. Table 2 shows the number of extreme combinations and hosting capacity results under different bin widths. As a comparison, the result obtained by the comprehensive assessment is 10.87 MW, and the computational time is 894 s. The result obtained by the proposed framework with 4*d* = 0.1 p.u is 12.97% less than the exact value. That means the adoption of a relatively large bin width underestimates the hosting capacity and leads to a conservative result. In contrast, when 4*d* is less than 0.01 p.u, the relative errors are less than 1%. The accurate estimations demonstrate that the proposed extreme combination can provide a suitable alternative to the time series data in hosting capacity assessment when an appropriate bin width is selected.


**Table 2.** Hosting capacity results with different bin widths.

The reason for the conservative estimation of hosting capacity can be further explained through the time series results of bus voltage. Figure 7 presents the 1-day snapshot of the time series voltage curves at Bus 18. The blue curve denotes the accurate results obtained by the comprehensive load flow method. Three different 4*d*, i.e., 0.1 p.u, 0.01 p.u, and 0.001 p.u, are conducted as the comparative cases. For these cases, the corresponding results in Table 2, i.e., 9.46 MW, 10.77 MW, and 10.86 MW, are used to calculate the upper and lower bounds of bus voltage. As described in Section 3.2, the upper bounds of the voltages are calculated using the lower and upper bounds of load and wind power bins, respectively. The lower bounds can be calculated using the opposite rule. It is evident from the results that the accurate voltage curve is always contained by the range of upper and lower bounds in each case. Therefore, when the corresponding hosting capacity in each case is applied to the actual time series data, it can be guaranteed that no voltage violation occurs. The comparison of the upper bounds of the voltages in different cases can explain the conservatism of the obtained hosting capacity. When 4*d* is relatively large, the upper bound of voltages is overestimated, and the voltage violation can be observed with less DG capacity. In contrast, when 4*d* is appropriately selected, the interval between

the upper and lower bounds of voltage is much narrower than those with larger bin widths. Under such circumstances, the hosting capacity can be accurately estimated. framework can provide accurate estimations of hosting capacity with at least 400 times (894/2.23) speed up in efficiency.

increases slightly, as shown in Table 2. Compared with the comprehensive assessment, the proposed

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 11 of 19

voltage violation can be observed with less DG capacity. In contrast, when △*d* is appropriately selected, the interval between the upper and lower bounds of voltage is much narrower than those with larger bin widths. Under such circumstances, the hosting capacity can be accurately estimated.

**Figure 7.** Time-series curves of voltage at Bus 18 with different bin widths: (**a**) △*d* = 0.1 p.u; (**b**) △*d* = 0.01 p.u; (**c**) △*d* = 0.001 p.u. **Figure 7.** Time-series curves of voltage at Bus 18 with different bin widths: (**a**) 4*d* = 0.1 p.u; (**b**) 4*d* = 0.01 p.u; (**c**) 4*d* = 0.001 p.u.

In the test system, the number of candidate buses is 32, and each of them has an equal probability of DG integration. For location penetration of 20%, the total possible number of DG deployment schemes is more than 4 × 105. The hosting capacity assessment for such a large number of potential DG deployments results in a significant computational burden. Therefore, MCS is adopted to simulate a relatively small number of scenarios to obtain approximated results. Moreover, the trial On the comparison of computational burden, due to the increase of the number of extreme combinations, the computational time required by the proposed framework with smaller 4*d* increases slightly, as shown in Table 2. Compared with the comprehensive assessment, the proposed framework can provide accurate estimations of hosting capacity with at least 400 times (894/2.23) speed up in efficiency.

number of MCS needs to be determined to achieve a compromise between accuracy and computational burden. In this context, we adopt variance coefficient *β* [38] of the obtained hosting In the test system, the number of candidate buses is 32, and each of them has an equal probability of DG integration. For location penetration of 20%, the total possible number of DG deployment schemes is more than 4 <sup>×</sup> <sup>10</sup><sup>5</sup> . The hosting capacity assessment for such a large number of potential DG deployments results in a significant computational burden. Therefore, MCS is adopted to simulate a relatively small number of scenarios to obtain approximated results. Moreover, the trial number of MCS needs to be determined to achieve a compromise between accuracy and computational burden. In this context, we adopt variance coefficient β [38] of the obtained hosting capacity results as the

stopping criteria of MCS. That means the results of MCS are assumed to be accurate if β is less than a specific value. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 12 of 19

The variance coefficient, β, should be appropriately selected to ensure the accuracy of the obtained results. Therefore, we consider two different values of β, i.e., 1% and 0.5%, to provide comparative results. Assume that there are four buses in each deployment scheme that are the candidate locations for DG integration. Under such circumstances, the total number of potential deployments can be obtained by the enumeration method (EM). The hosting capacity results of a total of 35,960 deployment scenarios are treated as accurate results. The mean and standard deviation (STD) values of hosting capacity obtained by EM are 6.66 MW and 2.45 MW. As a comparison, the results obtained by MCS are presented in Table 3. The trial number required by MCS with two different β values are 1300 and 5400, respectively. It can be found that the approximated results are more accurate when stopping criteria is relatively small. Figure 8 presents the comparisons between the probability distribution curves. The results indicate that MCS can well approximate the accurate curves obtained by the EM with β set to 0.5%. Therefore, in the following case study, the stopping criterion of MCS is set as β ≤ 0.5%. The variance coefficient, *β*, should be appropriately selected to ensure the accuracy of the obtained results. Therefore, we consider two different values of *β*, i.e., 1% and 0.5%, to provide comparative results. Assume that there are four buses in each deployment scheme that are the candidate locations for DG integration. Under such circumstances, the total number of potential deployments can be obtained by the enumeration method (EM). The hosting capacity results of a total of 35,960 deployment scenarios are treated as accurate results. The mean and standard deviation (STD) values of hosting capacity obtained by EM are 6.66 MW and 2.45 MW. As a comparison, the results obtained by MCS are presented in Table 3. The trial number required by MCS with two different *β* values are 1300 and 5400, respectively. It can be found that the approximated results are more accurate when stopping criteria is relatively small. Figure 8 presents the comparisons between the probability distribution curves. The results indicate that MCS can well approximate the accurate curves obtained by the EM with *β* set to 0.5%. Therefore, in the following case study, the stopping

**Table 3.** Statistical results of hosting capacity with different beta values. criterion of MCS is set as *β ≤* 0.5%.

**Figure 8.** Comparison of approximate results and exact results. MCS: Monte Carlo simulation-based; EM: enumeration method. **Figure 8.** Comparison of approximate results and exact results. MCS: Monte Carlo simulation-based; EM: enumeration method.

#### *4.3. Hosting Capacity Assessment of Single DG Technology 4.3. Hosting Capacity Assessment of Single DG Technology*

In this section, we perform the stochastic framework to assess the hosting capacity of single DG technology. Based on the analysis in Section 4.2, the parameters bin width, △*d*, and variance coefficient, *β*, are set to 0.001 p.u and 0.5%, respectively, to ensure the accuracy of the results. For each In this section, we perform the stochastic framework to assess the hosting capacity of single DG technology. Based on the analysis in Section 4.2, the parameters bin width, 4*d*, and variance coefficient, β, are set to 0.001 p.u and 0.5%, respectively, to ensure the accuracy of the results. For each DG technology, location penetration is increased from 10% to 100%, with a step size of 10%.

DG technology, location penetration is increased from 10% to 100%, with a step size of 10%. Table 4 presents the statistical characteristics of wind power and PV hosting capacity results (*HCWP* and *HCPV*) under location penetration 20%, 50%, and 80%. It is evident that the mean values of hosting capacity increase with higher location penetration, while the standard deviations present an opposite trend. The reasons for such results are as follows. Firstly, with increased location penetration, the installed DG capacity at each location reduces (the initial total capacity is assumed as 1 MW and increased with a fixed step). In this case, the impact of DG on the system is less severe, and the test system can accommodate more DG capacity. Secondly, higher location penetration alleviates the uncertainties of DG location. Therefore, the obtained hosting capacity results with different deployment schemes are less dispersed. Table 4 presents the statistical characteristics of wind power and PV hosting capacity results (*HCWP* and *HCPV*) under location penetration 20%, 50%, and 80%. It is evident that the mean values of hosting capacity increase with higher location penetration, while the standard deviations present an opposite trend. The reasons for such results are as follows. Firstly, with increased location penetration, the installed DG capacity at each location reduces (the initial total capacity is assumed as 1 MW and increased with a fixed step). In this case, the impact of DG on the system is less severe, and the test system can accommodate more DG capacity. Secondly, higher location penetration alleviates the uncertainties of DG location. Therefore, the obtained hosting capacity results with different deployment schemes are less dispersed.


**Table 4.** Statistical results of hosting capacity with different location penetration.

When comparing the hosting capacity of different DG types, it is worth mentioning that the test system has a higher PV hosting capacity than wind power at the same location penetration. For instance, the results for PV is 3.3% higher than that of wind power when location penetration is 50%. In general, the occurrences of high DG production and low demand are of great concern as they promote voltage rise. Therefore, the difference in wind power and PV hosting capacity can be explained using extreme combinations. For the case of wind power, most of the deployment schemes are constrained by the combination {(0.729 p.u, 0.730 p.u], (0.507 p.u, 0.508 p.u]}. While the corresponding combination for the PV case is {(0.749 p.u, 0.750 p.u], (0.590 p.u, 0.591 p.u]}. The former combination can cause a slightly higher voltage rise at the same DG capacity.

In addition to the hosting capacity results, Table 5 presents the comparisons of mean values of total energy production from wind power and PV (*EWP* and *EPV*). Although the hosting capacity for PV is slightly higher than wind power, total energy production from PV is less than 50% of wind. The reason for the difference between total energy production is the higher wind power capacity factor in this area. Total generation production can realistically reflect energy utilization. Therefore, in the planning of DG, not only the maximum DG capacity, but also the total generation production should be considered.

**Table 5.** Mean values of energy production with different location penetration.


Figure 9 presents the probability distributions of wind power and PV hosting capacity under 50% location penetration. The results are shown using histograms and cumulative probability curves. It can be found that Gamma distribution can approximate both wind power and PV hosting capacity. Moreover, the obtained results with other location penetrations also follow the same rule. The histograms provide the estimations of extreme values of hosting capacity. For wind power hosting capacity, the obtained minimum and maximum values are 4.10 MW and 12.65 MW, respectively. As a comparison, the corresponding results for PV hosting capacity are 4.26 MW and 13.21 MW, respectively. The difference between the extreme values is partly due to the locations of DG. In general, the relatively high hosting capacity can be obtained with DG located near the first bus. Figure 9b,d presents the cumulative probability curves of hosting capacity. These curves provide system planners with a straightforward tool for estimating the probability of having a hosting capacity higher than a specific value. For example, the probability of having wind power hosting capacity higher than 6 MW is 0.9072, while in the PV case, the result is 0.9353.

Since the obtained hosting capacity results with different location penetration can be approximated by Gamma distributions, it is imperative to establish the relationship between location penetration and the parameters of Gamma distributions. The probability density function of Gamma distribution is given as follows:

$$f(\mathbf{x}|\alpha, \beta) = \frac{1}{\beta^{\alpha} \Gamma(\alpha)} \mathbf{x}^{\alpha - 1} e^{\frac{\overline{\mathbf{x}}^{\alpha}}{\beta}} \tag{1}$$

where Γ(·) is the Gamma function; parameters α and β are shape and scale parameters, respectively.

Probability density

Probability density

The shape and scale parameters can be calculated as follows:

$$\begin{cases} \alpha = \mu\_x^2 / \sigma\_x \\ \beta = \sigma\_x / \mu\_x \end{cases} \tag{2}$$

Gamma fit

where µ*<sup>x</sup>* and σ*<sup>x</sup>* are the expected value and standard deviation of variable *x*, respectively. where *μx* and *σx* are the expected value and standard deviation of variable *x*, respectively.

Based on the hosting capacity results with different location penetrations, the curves of the parameter of Gamma distribution to location penetration can be obtained and fitted by the fifth-order polynomials, as shown in Figure 10. These curves can provide planners with useful information. For instance, when the location penetration is 35%, system planners can utilize the curve to obtain the shape and scale parameters and estimate the probability distribution of hosting capacity under the target location penetration. Moreover, once such curves are established for a specific distribution system, the probabilistic hosting capacity of the system with any location penetration can be obtained without simulating thousands of DG deployment schemes. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 14 of 19 where *μx* and *σx* are the expected value and standard deviation of variable *x*, respectively. Based on the hosting capacity results with different location penetrations, the curves of the parameter of Gamma distribution to location penetration can be obtained and fitted by the fifth-order polynomials, as shown in Figure 10. These curves can provide planners with useful information. For instance, when the location penetration is 35%, system planners can utilize the curve to obtain the shape and scale parameters and estimate the probability distribution of hosting capacity under the target location penetration. Moreover, once such curves are established for a specific distribution system, the probabilistic hosting capacity of the system with any location penetration can be obtained Based on the hosting capacity results with different location penetrations, the curves of the parameter of Gamma distribution to location penetration can be obtained and fitted by the fifth-order polynomials, as shown in Figure 10. These curves can provide planners with useful information. For instance, when the location penetration is 35%, system planners can utilize the curve to obtain the shape and scale parameters and estimate the probability distribution of hosting capacity under the target location penetration. Moreover, once such curves are established for a specific distribution system, the probabilistic hosting capacity of the system with any location penetration can be obtained without simulating thousands of DG deployment schemes. 0.3 Results 0.9072 Results

**Figure 9.** Hosting capacity results of single DG technology: (**a**) and (**b**) probability distribution of wind power hosting capacity; (**c**) and (**d**) probability distribution of PV hosting capacity. **Figure 9.** Hosting capacity results of single DG technology: (**a**) and (**b**) probability distribution of wind power hosting capacity; (**c**) and (**d**) probability distribution of PV hosting capacity. **Figure 9.** Hosting capacity results of single DG technology: (**a**) and (**b**) probability distribution of wind power hosting capacity; (**c**) and (**d**) probability distribution of PV hosting capacity.

**Figure 10.** The curves of shape and scale parameters to location penetration: (**a**) shape parameter; (**b**) scale parameter. **Figure 10.** The curves of shape and scale parameters to location penetration: (**a**) shape parameter; (**b**) **Figure 10.** The curves of shape and scale parameters to location penetration: (**a**) shape parameter; (**b**) scale parameter.

*4.4. Hosting Capacity Assessment of Hybrid Wind-PV* 

*4.4. Hosting Capacity Assessment of Hybrid Wind-PV* 

scale parameter.

#### *4.4. Hosting Capacity Assessment of Hybrid Wind-PV Sustainability* **2020**, *12*, x FOR PEER REVIEW 15 of 19

In this section, the proposed framework is conducted to estimate the hybrid wind power-PV hosting capacity. This section only considers the different ratios of wind power and PV under 50% location penetration to simplify the presentation. In this section, the proposed framework is conducted to estimate the hybrid wind power-PV hosting capacity. This section only considers the different ratios of wind power and PV under 50% location penetration to simplify the presentation.

Monte Carlo simulation is performed to generate various DG deployment schemes. For a specific DG deployment scenario, the percentage of wind power generation is increased from 0% to 100% with 5% increment steps. The proposed framework is performed to calculate the corresponding hosting capacity. Figure 11 presents the probability distributions of the obtained hosting capacity results with different wind power percentages. As the ratio of wind power increases, the transition of hybrid DG hosting capacity follows a two-stage process. In the first stage, the probability distributions of hosting capacity shift toward the right until the percentage of wind power reaches 75%. In the second stage, as the proportion of wind power continues to increase, the probability distribution curves of hosting capacity move to the left. Monte Carlo simulation is performed to generate various DG deployment schemes. For a specific DG deployment scenario, the percentage of wind power generation is increased from 0% to 100% with 5% increment steps. The proposed framework is performed to calculate the corresponding hosting capacity. Figure 11 presents the probability distributions of the obtained hosting capacity results with different wind power percentages. As the ratio of wind power increases, the transition of hybrid DG hosting capacity follows a two-stage process. In the first stage, the probability distributions of hosting capacity shift toward the right until the percentage of wind power reaches 75%. In the second stage, as the proportion of wind power continues to increase, the probability distribution curves of hosting capacity move to the left.

**Figure 11.** Probability distributions of hosting capacity with different DG technology combinations. **Figure 11.** Probability distributions of hosting capacity with different DG technology combinations.

Table 6 presents the mean values of hosting capacity and total energy production under two maximum situations. The hybrid wind power-PV hosting capacity reaches its maximum value when the percentage of wind power is 75%. In this case, the mean value of total hosting capacity is 9.39 MW, of which wind power and PV capacity are 7.04 MW and 2.35 MW, respectively. Compared with the results of single wind power and PV technologies, the total hosting capacity increases by about 1.21 (9.39/7.78) and 1.17 (9.39/8.05) times, respectively. The mean value of total energy production for one year is 17,999.26 MW·h, of which the energy production of wind power accounts for about 87%. The total energy production increases by about 1.04 and 2.20 times when compared with single wind power and PV, respectively. Another important result is the maximum energy production, and this value is reached when the wind power percentage is 85%. In this case, the mean values of hosting capacity and total energy production are 9.12 MW and 18,596.31 MW·h, respectively. Table 6 presents the mean values of hosting capacity and total energy production under two maximum situations. The hybrid wind power-PV hosting capacity reaches its maximum value when the percentage of wind power is 75%. In this case, the mean value of total hosting capacity is 9.39 MW, of which wind power and PV capacity are 7.04 MW and 2.35 MW, respectively. Compared with the results of single wind power and PV technologies, the total hosting capacity increases by about 1.21 (9.39/7.78) and 1.17 (9.39/8.05) times, respectively. The mean value of total energy production for one year is 17,999.26 MW·h, of which the energy production of wind power accounts for about 87%. The total energy production increases by about 1.04 and 2.20 times when compared with single wind power and PV, respectively. Another important result is the maximum energy production, and this value is reached when the wind power percentage is 85%. In this case, the mean values of hosting capacity and total energy production are 9.12 MW and 18,596.31 MW·h, respectively.

**Table 6.** Mean values of hosting capacity and energy production. **Table 6.** Mean values of hosting capacity and energy production.


As can be seen from the above results, wind power has a dominant role in both cases of maximum hosting capacity and energy production. However, PV is also essential as the complementarity generation to increase total hosting capacity and energy production. The results As can be seen from the above results, wind power has a dominant role in both cases of maximum hosting capacity and energy production. However, PV is also essential as the complementarity

production do not reach the optimal values simultaneously. Therefore, considering the investment

generation to increase total hosting capacity and energy production. The results demonstrate that the complementarity between wind power and PV can promote the system to accommodate more DG capacity, thus increasing the utilization of renewable energy. Besides, the difference between the two maximum results demonstrates that hosting capacity and energy production do not reach the optimal values simultaneously. Therefore, considering the investment of different DG types, system planners and DG investors may need to resort to multi-objective optimization to achieve compromise solutions. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 16 of 19 of different DG types, system planners and DG investors may need to resort to multi-objective optimization to achieve compromise solutions.

Similar to the hosting capacity of single DG technology, the probability distribution of hybrid wind power-PV hosting capacity with different wind power ratios can also be approximated by Gamma distribution. Moreover, the probability distribution curves of hosting capacity in Figure 11 have a similar shape. Using Equation (2) to calculate the parameters of Gamma distribution corresponding to each curve, the obtained mean and STD values of the shape parameters are 32.13 and 0.4, which indicates that the shape parameters of different curves are almost the same. In this context, only the relationship between wind power percentage and scale parameter needs to be established. Moreover, the change of scale parameters can represent the trend of hosting capacity. As can be seen from the curve in Figure 12, the system can accommodate relatively higher DG capacity when the wind power percentage is 65–75%. This result is consistent with the previous analysis in Table 6. This curve has two applications. Firstly, system planners can obtain the probability distributions of hosting capacity with different wind power and photovoltaic combinations based on the curve. Secondly, DG investors can adjust their investment plans for wind power and PV generation based on the curve so that they can maximize the DG capacity and obtain more revenue from energy production. Similar to the hosting capacity of single DG technology, the probability distribution of hybrid wind power-PV hosting capacity with different wind power ratios can also be approximated by Gamma distribution. Moreover, the probability distribution curves of hosting capacity in Figure 11 have a similar shape. Using Equation (2) to calculate the parameters of Gamma distribution corresponding to each curve, the obtained mean and STD values of the shape parameters are 32.13 and 0.4, which indicates that the shape parameters of different curves are almost the same. In this context, only the relationship between wind power percentage and scale parameter needs to be established. Moreover, the change of scale parameters can represent the trend of hosting capacity. As can be seen from the curve in Figure 12, the system can accommodate relatively higher DG capacity when the wind power percentage is 65–75%. This result is consistent with the previous analysis in Table 6. This curve has two applications. Firstly, system planners can obtain the probability distributions of hosting capacity with different wind power and photovoltaic combinations based on the curve. Secondly, DG investors can adjust their investment plans for wind power and PV generation based on the curve so that they can maximize the DG capacity and obtain more revenue from energy production.

**Figure 12.** The curves of scale parameters to wind power percentage. **Figure 12.** The curves of scale parameters to wind power percentage.

#### **5. Discussion**

**5. Discussion**  In this section, we discuss the main features, applications, and future research of this study.

In this section, we discuss the main features, applications, and future research of this study. Firstly, the proposed stochastic framework provides system planners with an accurate and efficient tool for hosting capacity assessment. The framework adopts the discretization-aggregation technique to deal with the time series data. Compared with the application of this technique in previous studies [30,35], this study concentrates more on the extreme combinations and the appropriate selection of bin width. Thanks to the introduction of extreme combinations, the number of power flow calculations required by the hosting capacity evaluation is significantly reduced. More Firstly, the proposed stochastic framework provides system planners with an accurate and efficient tool for hosting capacity assessment. The framework adopts the discretization-aggregation technique to deal with the time series data. Compared with the application of this technique in previous studies [30,35], this study concentrates more on the extreme combinations and the appropriate selection of bin width. Thanks to the introduction of extreme combinations, the number of power flow calculations required by the hosting capacity evaluation is significantly reduced. More importantly, the simulation results demonstrate the necessity of appropriate selection of bin width.

importantly, the simulation results demonstrate the necessity of appropriate selection of bin width. Secondly, this study presents the assessments of wind power, PV, and hybrid wind-PV hosting capacity. The simulation results indicate that the hybrid energy system has the potential to increase hosting capacity and energy production due to the complementarity of wind speed and solar irradiation. The performance of the hybrid wind-PV system in facilitating energy integration and Secondly, this study presents the assessments of wind power, PV, and hybrid wind-PV hosting capacity. The simulation results indicate that the hybrid energy system has the potential to increase hosting capacity and energy production due to the complementarity of wind speed and solar irradiation. The performance of the hybrid wind-PV system in facilitating energy integration and utilization varies with different wind power and PV ratios. Compared with the studies that focused on PV hosting

system planners and DG investors with serval useful tools, such as the curves in Figures 10 and 12. Such results can be generally established for other networks. These curves can, not only help system

utilization varies with different wind power and PV ratios. Compared with the studies that focused on PV hosting capacity [23–26], this study can contribute a better understanding of hybrid wind-PV capacity [23–26], this study can contribute a better understanding of hybrid wind-PV hosting capacity from a probabilistic point of view.

Thirdly, the simulation results identify that Gamma distribution can approximate the probability distributions of both single and hybrid DG hosting capacity. These findings can provide system planners and DG investors with serval useful tools, such as the curves in Figures 10 and 12. Such results can be generally established for other networks. These curves can, not only help system planners assess the hosting capacity under different location penetrations, but also assists DG investors in making informed decisions on the investment plan of different DG types.

This study performs the hosting capacity assessment without considering active network managements (ANMs). ANMs, such as on-load tap changers [39] and energy storage [40,41], are verified to be effective ways for the enhancement of hosting capacity. However, due to the need for the knowledge of the succession of operating states (e.g., state of charge of energy storage), the assessment of enhanced hosting capacity can be more challenging when considering ANMs. Therefore, the establishment of a critical time series of DG production and load demand can be one of the future research works.

#### **6. Conclusions**

In this paper, a stochastic framework is developed to study the hybrid wind-PV system hosting capacity from a probabilistic view. To reduce the computational burden of assessments, we define the extreme combinations of DG productions and load demand based on historical data sets. The proposed framework is implemented in the IEEE 33-bus system to evaluate wind power, PV, and hybrid energy hosting capacity. The main findings of this study are as follows:


The main application of this study is to provide a useful tool for system planners to assess DG hosting capacity. This study can be extended to consider other low carbon technologies, such as electric vehicles and biomass-fueled gas engines. Also, the proposed combination of production and demand can be used in other energy management studies that require extensive power flow calculations, such as analysis of system losses and energy reductions. Regarding future research, more binding constraints, such as voltage unbalance, conductor thermal capacity, and transformer overload, can be incorporated to establish a more comprehensive hosting capacity assessment. Furthermore, a critical time series for DG production and load demand needs to be developed to evaluate the hosting capacity enhanced with ANMs.

**Author Contributions:** The authors confirm their contributions to the paper as follows: Supervision, D.L.; methodology, software, and writing—original draft preparation, C.W.; project administration, F.T.; visualization and writing—review and editing, Y.Z. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the National Natural Science Foundation of China under grant No. 51977157.

**Acknowledgments:** This work was supported by the National Natural Science Foundation of China under grant No. 51977157.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Renewable Energy for Sustainable Growth and Development: An Evaluation of Law and Policy of Bangladesh**

#### **Mohammad Ershadul Karim 1,\*, Ridoan Karim 1,2, Md. Toriqul Islam 1,3 , Firdaus Muhammad-Sukki 4,\*, Nurul Aini Bani <sup>5</sup> and Mohd Nabil Muhtazaruddin <sup>5</sup>**


Received: 23 September 2019; Accepted: 15 October 2019; Published: 17 October 2019

**Abstract:** Bangladesh's constant growth with an annual 6% plus Gross Domestic Product (GDP) for more than the last two decades and achievements in other socio-economic metrics in recent times is impressive and recognized by various global authoritative bodies. The extent of overwhelming economic ventures in the private sector coupled with the commitments of the government clearly demonstrates the transformation of the country from a primarily agro-based economy to one influenced by the manufacturing and service sectors. Bangladesh is fortunate to have fossil fuel reserves on a limited scale, though these are not enough to run the ongoing massive scale development activities, both in private and public sectors. Thus, the constant and uninterrupted supply of energy at an affordable price remains a serious concern for the successive governments. Therefore, this issue of supply of constant energy has turned to be an important part in the national development agenda. Besides, the country is one of the worst victim nations of the devastating effects of global warming and climate change. As Bangladesh is geographically located in a favorable place in the world map with the availability of plenty of renewable energy sources (RES), the policymakers started to take initiatives leading to exploiting these sources to meet the energy demand of the country. There are both prospects and administrative, legal, technological, socio-cultural and environmental challenges. To address these challenges, it requires comprehensive policy initiatives. A good number of technical and scientific research containing findings and recommendations are available. This paper, which is based on adopting a qualitative research methodology where the contents of secondary sources were analyzed, is an initial attempt to highlight the renewable energy developments in Bangladesh, and subsequently, to evaluate the relevant legal and policy initiatives in the light of international best practices. We advance several recommendations that the stakeholders can consider exploiting RES effectively to attain inclusive, equitable and sustainable development in Bangladesh. These include, *inter alia*: (1) Enhancing government participation to lead the development of renewable energy (RE); (2) ensuring localization of RE technology; (3) reducing the expenses of energy generation through RES and providing assistance in initial investments; (4) introducing comprehensive legal and regulatory policy for the development of RE industry in Bangladesh; and (5) conducting effective public awareness.

**Keywords:** RE prospects and challenges; RE regulations and policy; RE in Bangladesh

#### **1. Introduction**

Energy generated from fossil or traditional fuel sources, such as natural oil, gas and coal, etc., causes negative impacts on our surrounding environment as they produce greenhouse gases (GHGs) which are responsible for climate change and global warming. Climate change has devastating effects and consequences on the existence and survival of human beings, biodiversity and ecosystems on this earth. Therefore, the reduction of emission of GHGs in order to mitigate climate change has turned out to be an important concern for almost everyone. The global community has proposed and adopted various policy initiatives which advocated to shift the energy production from fossil fuels to energy generated from renewable energy sources (RES), such as solar, geothermal, biomass, wind, biogas, hydro-power, etc., since these are reported to be cleaner, more sustainable and relatively less pollutant [1].

Bangladesh's constant growth with an annual 6% plus Gross Domestic Product (GDP) for more than the last two decades and achievements in other socio-economic metrics in recent times is impressive and recognized by various global authoritative bodies. Bangladesh is frequently projected as an agricultural country that encounters various catastrophic natural disasters every year. However, in recent years, it aims to transform from an agro-based nation to an industrial one, and this is reflected through a slow, but massive industrial growth. The extent of overwhelming economic ventures and the commitments of the government clearly demonstrates Bangladesh's economic shift. The country is fortunate to have fossil fuel reserves on a limited scale, and these are not enough to run the ongoing massive scale development activities, both in private and public sectors. Thus, the constant and uninterrupted supply of energy at an affordable price remains a serious concern for the successive governments. Therefore, this issue has been getting importance in the development agenda constantly.

The Government of Bangladesh has promised and has taken initiatives to ensure quality energy production and supply to every door by 2020. In implementing this high ambition, the government set plans to generate 10% of its power's demand through energy generated from RES by 2020 [2]. Fuel diversification seems to be an exclusive approach for ensuring sufficient electricity supply and stimulating economic advancement of the country. Consequently, Bangladesh has adopted this fuel diversification program for extending its renewable energy (RE) industry and ensuring the ever-increasing electricity demand. In 2008, the government formulated the Renewable Energy Policy. Public-private investment is welcomed in promoting the RE ventures to replace the fossil fuel-based energy generation to the RE contributing to the national grid [3]. Government's Renewable Energy Policy presumed to generate 5% of energy from the RES by 2015 and to increase a total of 10% of the total electricity generation by 2020 [4]. Meanwhile, the government has launched a new project "500 MW Solar Power Mission" to satisfy the escalating need for energy consumption [5].

Bangladesh has signed the Paris Agreement to the United Nations Framework Convention on Climate Change (UNFCC) on the 21st September 2016 and ratified the same on the 4th November, 2016. The Paris Agreement is not like the Kyoto Protocol that puts an obligation toward the advanced nations of mitigating emissions of GHG. It is a bottom-up treaty that obliges every country to undertake essential efforts for the protection of the environment and keeping climate stable [6]. Having ratified the Paris Agreement, the country has pledged to reduce its GHG emissions by 5% below 'business-as-usual' level by 2030 using domestic resources regardless of the country's GDP. In such a context, Bangladesh is also responsible to the establishment, development and implementation of the legal framework for addressing the environmental concerns, formulate rules and policies for reducing the GHG emissions, and realize a sustainable economic strategy to generate electricity [6].

Although the government wishes seriously to expand a significant portion of RE contribution in the national grid, such expansion can only be achieved if fundamental challenges encompassing the RE projects are properly addressed. Despite the execution of comprehensive policies and development strategies, which eventually worked as the catalyst for RE penetration in power generation, RE projects and green technologies still face the financial, technical and even political barriers. Even though many scholars, researchers, bodies, and agencies from both public and private sectors have shown interests and great concerns about the scientific and technical challenges encircling the expansion of RE in Bangladesh, there is a dearth of literature on this issue, particularly, from the legal point of view. This paper is primarily legal in nature which purports to share the prospects of RES and problems in exploiting the RES in Bangladesh context. In doing so, it has attempted to identify the key obstacles within the existing legal and regulatory mechanisms and suggest some solutions that are based on international best practices. To this end, this paper is divided into seven parts, including the introduction and conclusion. Part 2 of the paper will discuss the basics of RE and its growing importance, while Part 3 will highlight the prospects of RE in Bangladesh context and Part 4 will share some challenges for RE in Bangladesh context. Part 5 of this paper covers discussion on the legal, regulatory and policy aspects of RE in Bangladesh and Part 6, finally, proposes some suggestions and recommendations.

#### **2. Renewable Energy Basics and its Growing Importance**

Energy generated from RES, such as solar, wind, biomass, thermal or hydro, etc., is theoretically considered as renewable as each one could yield an infinite amount of energy [7]. Uses of energy generated via these sources are not new and ancient people successfully utilized these also. For example, the Egyptian Nile civilization used to utilize the wind to move ships, grain-grinding facilities, and boat propellers. Even the Chinese and Japanese started using wind-run water pumping systems in the historical past, leading to cost minimization dramatically [8]. The succeeding civilizations opted to use windmills as alternative sources, since it is accessible and comparatively less expensive [9]. In fact, in recent times, the increasing energy crisis and stringent enforcement of carbon emission laws for reducing the GHG emissions has forced many nations to think about alternative supplies of energy [10]. This segment will highlight some basics about RES, RE, their importance and relevant concerns in exploiting them.

#### *2.1. Renewable Energy: Definitional Challenges*

Generally, a universally accepted definition of RE and RES are difficult to pinpoint because of divergent understanding of these terms by various stakeholders. As a result, various definitions of RE can be found in both scholarly works and in jurisdiction specific literature. For instance, the definition of RE can be found in Article III of the Statute of the International Renewable Energy Agency (IRENA) [11], the EU Directive 2009/28/EC [12] on the development and progression of the utilization concerning electricity from sustainable resources, the Glossary of Statistical Terms of Organization for Economic Co-operation and Development (OECD) [13], and the International Energy Agency (IEA) in its 2010 Renewable Information Report [14], etc. A cautious look at these definitions will reveal that these have incorporated specific attributes of either RE or the RES, i.e., RE means the electricity produced from solar sources, wind, geothermal, biomass, and hydropower assets, etc.

The notion of the definition is essential not only to get an agreement on a term's meaning and scope, but also to delimit its parameters. Moreover, it is critically significant in matters of jurisprudence and the regulatory frameworks too. The absence of consensus on specific lawful definition of 'RES' creates much confusion and invites more questions. Various definitions of the same term in authoritative literature invites heated debates as the stakeholders may reveal unexpected contradictions, while undertaking RE-related activities [15]. Since the significance of RE has been emerging increasingly as a result of its inherent benefits and prospects, the global community should reach a consensus regarding the relevant definitions for legal and administrative purposes, which is imperative for its ultimate success.

In the domestic level of Bangladesh, according to section 2 of the Sustainable and Renewable Energy Development Authority Act, 2012 (Act No. 48 of 2012) [16], "renewable energy" is defined as the energy and power originated from biomass, biofuel, biogas, hydropower, solar energy, wind power, hydrogen cell, geothermal, tide and wave or energy and power generated from any other sources declared by the government as renewable energy in the official gazette notification. The same section

further defines 'sustainable energy' and 'non-renewable energy' too. It is provided that "non-renewable energy" is the energy and power produced from mineral gas, coal, peat coal, natural oil, any other fossil fuel, nuclear power, and any other non-renewable energy resources declared by the government as non-renewable energy in the official notification gazette notification [16].

While defining any term(s) and/or phrase(s), the words either 'means' or 'includes' are used in legal texts. The use of the word 'means' denotes exclusivity, i.e., the definitions are exclusive and cannot be expanded. On the other hand, the word 'includes' is used to mean flexibility to allow to consider similar kinds of things. It is evident that in both the definitions of the words 'renewable energy' and 'non-renewable energy', the word 'means' is used, which indicates that it is not possible to include any things else other than those included in the definition. This legal definition of RE will eventually help to understand the regulatory measures in describing the prospects and challenges of RE of Bangladesh in this paper.

#### *2.2. Renewable Energy Sources: Importance and Concerns*

RES, due to the utilization of indigenous assets, can possibly give energy with near-zero discharges of both air toxins and GHG emissions [17]. That is why, clean energy, environmentally friendly power energy, sustainable energy, alternative energy, green energy, etc., are used as the synonyms of RE. The importance of this type of energy is documented in many scientific and policy research. Aside from the researchers and specialists, even the legal experts have embraced the significance of RE for the preservation of the global environment and combating climate change [6]. In the case of *Preussen Elektra AG v Schhleswag AG* [2001], *Case C-379*/*98*, the Court of Justice of the European Union (CJEU) reiterated that the use of sustainable energy resources for generating power is paramount in connection with saving the planet because of its undeniable role in reducing GHGs, which scientists ascribe to be the chief driver of climatic doom [18].

Environmental experts and legal researchers have been campaigning for a shift toward the concentration from finite non-RES to renewables to maintain the development activities, despite the fact that there are some inherent initial concerns [19]. Nevertheless, these concerns are not exceptional in this sector alone; rather, they are generally present whenever any new technological developments are introduced [6]. With the entry of, and advance in every sector, triggered by the scientific advancements, the global community has understood the significance of improvement of the RES in a practical, scalable, and capable ways [20].

In the global context, the first and formal, even though indirect, talk on RE started in the 1970s, in particular, in the United Nations (UN) Stockholm Declaration on the Human Environment 1972 [6]. While sharing on the dangers of depletion of non-renewable materials, the world leaders emphasized on the utilization of RES [21]. Subsequently, the UN and many other global and regional agencies stressed using RE resources, especially, in power generation in an environment-friendly and sustainable manner. Numerous international bodies, forums, non-governmental organizations (NGOs) and instruments, are established and developed to concentrate on promoting RE resources for securing sustainable energy solutions in the context of global warming and climate change.

Above all, the international endeavor proceeded steadily for moving forward through many initiatives. A list of such initiatives include, but not limited to, World Commission on Environment and Development Report, 1987 [22]; Nairobi Program of Action for the Development and Utilization of New and Renewable Sources of Energy, 1981 [23]; Intergovernmental Panel on Climate Change (IPCC), 1988 by United Nations Environment Program (UNEP); World Meteorological Organization (WMO), 1988; UNFCCC (Rio Earth Summit) 1992; the 3rd Conference of the Parties (COP3), 1997; Kyoto Protocol to the UNFCCC; UN Secretary General's declaration, 'Sustainable Energy for All (SE4ALL)' to attain 30% of the global RE target by 2030 [6]; Resolution 65/151, UN General Assembly's declaration on 'International Year of Sustainable Energy for All, (2012)' by 2014–2024 [24]; UN Sustainable Development Goals (SDGs), 2015, especially, the goals No. 7 and 13, and finally, the Paris Agreement to the UNFCCC, 2016. 206

Since the late 1970s, the worldwide network's dependence on RE has been emerging over 10% every year [25], and starting from 2014, more than 164 nations have embraced the renewables targets [26]. In 2012, the utilization of RES helped to provide up to 13.2% of the worldwide essential energy supply. The same figure rose to 22% of worldwide power usage in 2013. It was estimated that this number is expected to rise to 26% in 2020. To share this in a practical setting, this number is greater than the current overall power demands of Brazil, Russia, India, China and South Africa (BRICS) nations put together [27]. Figure 1 shows the world's total primary consumption from 2007 until 2016, illustrating the rising trend of RE share. 198 Since the late 1970s, the worldwide network's dependence on RE has been emerging over 10% 199 every year [25], and starting from 2014, more than 164 nations have embraced the renewables targets 200 [26]. In 2012, the utilization of RES helped to provide up to 13.2% of the worldwide essential energy 201 supply. The same figure rose to 22% of worldwide power usage in 2013. It was estimated that this 202 number is expected to rise to 26% in 2020. To share this in a practical setting, this number is greater 203 than the current overall power demands of Brazil, Russia, India, China and South Africa (BRICS) 204 nations put together [27]. Figure 1 shows the world's total primary consumption from 2007 until 2016, 205 illustrating the rising trend of RE share.

*Sustainability* **2019**, *11*, x FOR PEER REVIEW 5 of 30

207 **Figure 1.** Total primary energy consumption (2007–2016). Adapted from British Petroleum (BP) 208 Statistical Review of World Energy [28–32]. **Figure 1.** Total primary energy consumption (2007–2016). Adapted from British Petroleum (BP) Statistical Review of World Energy [28–32].

209 Despite the positive outlook and promises, worldwide demand for RE has been expanding at an 210 underwhelming rate, especially, compared to fossil fuel-based energy demands. About USD 16 211 billion and USD 17 billion were put for RE resources in 2015 and 2016, respectively [33]. This figure 212 is marginally lower than the normal interest in 2014, which was USD 19 billion. Such a circumstance 213 has urged the private sector entrepreneurs to approach to dedicate resources to creating power 214 utilizing renewables. The dedication of the private sector commitments—including speculation—is 215 additionally obvious as the interest in the RE limit exceeding that of hydrocarbon incumbents Despite the positive outlook and promises, worldwide demand for RE has been expanding at an underwhelming rate, especially, compared to fossil fuel-based energy demands. About USD 16 billion and USD 17 billion were put for RE resources in 2015 and 2016, respectively [33]. This figure is marginally lower than the normal interest in 2014, which was USD 19 billion. Such a circumstance has urged the private sector entrepreneurs to approach to dedicate resources to creating power utilizing renewables. The dedication of the private sector commitments—including speculation—is additionally obvious as the interest in the RE limit exceeding that of hydrocarbon incumbents demonstrating a steady trend of difference for the fifth year in a row [34].

216 demonstrating a steady trend of difference for the fifth year in a row [34]. 217 Experts believe that RE frameworks perform best at little to medium scale and are perfect for 218 rural and geographically disadvantaged regions where it is difficult to cover these territories through 219 traditional fossil-based energy sources [35]. RE frameworks offer an attractive alternative prospect in 220 such regions. Besides, the energy generated utilizing renewables is less vulnerable to the volatility of 221 price and value, which is generally suffered by the oil and gas markets [36]. Thus, buyers can remain 222 certain about the supply of energy and can additionally be profited through—among other things— 223 delivering and exchanging additional energy. On the other hand, a few administrative and policy-224 oriented difficulties, consumption payments, tax benefits, and other surfeit expenses tend to depress Experts believe that RE frameworks perform best at little to medium scale and are perfect for rural and geographically disadvantaged regions where it is difficult to cover these territories through traditional fossil-based energy sources [35]. RE frameworks offer an attractive alternative prospect in such regions. Besides, the energy generated utilizing renewables is less vulnerable to the volatility of price and value, which is generally suffered by the oil and gas markets [36]. Thus, buyers can remain certain about the supply of energy and can additionally be profited through—among other things—delivering and exchanging additional energy. On the other hand, a few administrative and policy-oriented difficulties, consumption payments, tax benefits, and other surfeit expenses tend to depress the investors of RE projects compared to investment in fossil fuel-based projects [37].

225 the investors of RE projects compared to investment in fossil fuel-based projects [37]. 226 Asides from the concerns raised above, many years of systemic abuse of the RES have led to a 227 plethora of worries stemming from the lack of adequacy of productivity of the existing power 228 management framework as the energy bills are still quite high. Moreover, private contractors are 229 raking in enormous financial benefits, and the cash spent to help the exercises of various 230 foundations—such as controllers, transport framework administrators, open utilities and universal 231 organizations—further magnify the battle of competing with fossil-based power sources [38]. 232 Disheartening as these worries may be, there are still signs of hope. In recent times, there are signs 233 that the RE projects that have been utilizing the RES judiciously are overcoming these challenges Asides from the concerns raised above, many years of systemic abuse of the RES have led to a plethora of worries stemming from the lack of adequacy of productivity of the existing power management framework as the energy bills are still quite high. Moreover, private contractors are raking in enormous financial benefits, and the cash spent to help the exercises of various foundations—such as controllers, transport framework administrators, open utilities and universal organizations—further magnify the battle of competing with fossil-based power sources [38]. Disheartening as these worries may be, there are still signs of hope. In recent times, there are signs that the RE projects that have been utilizing the RES judiciously are overcoming these challenges leading to financial improvement [39]. Furthermore, it should additionally be appreciated that the issue of initial outflow relating to the

234 leading to financial improvement [39]. Furthermore, it should additionally be appreciated that the 235 issue of initial outflow relating to the capital expenses is not limited only to the RE, rather any power capital expenses is not limited only to the RE, rather any power development project has to incur similar initial high up-front costs.

Renewables are taking a consistently growing pace in the overall energy industry with an evolving number of driving organizations focusing on aggressive inexhaustible power targets [40]. Unfortunately, even such growth cannot decrease the level of GHG emissions that the world desires. Hence, effective, consolidated and synergic strategies and efforts from the stakeholders are required to utilize RES across the board at a quicker pace. Moreover, the significance of an all-inclusive concession to the environment, as advocated by the Paris Agreement, holds guarantees for a feasible eco-accommodating development and progression of the world for the future.

#### **3. Prospects of Renewable Energy in Bangladesh**

Fossil-fuel sources, more specifically natural gas, oil and coal, play the lead role in electricity generation in Bangladesh [3]. The electricity demand in the country increased considerably from 0.4 to 1.38 quadrillion between 1997 and 2016 [41] and the country so far could not provide 100% access to electricity, due to *inter alia*, the inadequate volume of energy sources [42]. The total coal assets in Bangladesh were only 1063 million tons, whereas, the natural gas reserve demonstrated to be available with a volume of 9.7 trillion cubic feet (TCF) according to the statistics of 2013 [43]. Besides, to supply fuel in different industries, including energy, Bangladesh imports nearly 1.2 million tons of crude oil and 2.6 million tons of refined petroleum commodities every year [44]. Bangladesh absorbed approximately 2.132 million metric tons of coal, 175.69 kilo barrels of oil per day and 28.37 billion cubic meters of natural gas in 2018 [45].

Energy application in the country has expanded significantly, and the generation altogether relies mainly upon non-RES. Energy production through these ways adds practically 40% of the absolute carbon dioxide (CO2) discharges by the nation [46]. In 2018, Bangladesh generated energy amounting 7,418 MW, whereas the demand was 11,534 MW. Such a gap between supply and demand results in the impediment of the financial and innovative advancement of the nation. Such a situation has influenced the policymakers to officially undertake activities to utilize the RES to alleviate the energy needs while maintaining the ecological effects.

Bangladesh is blessed to have a generous amount of RES and the effective exploitation of these promises to satisfy the energy need of the country. Among the accessible energy assets, biomass is considered as the significant RES available in the country, which can reduce the utilization of and reliance on non-RES. Solar energy is also very promising as the country gets an immense scale of sunlight, due to its geographical location. It may be pertinent to share that the rural and coastal areas of Bangladesh have been enjoying the benefits of the installed solar photovoltaic (PV) panels. Besides, several government agencies and NGOs undertook some activities for delivering electricity from small scale hydro plants and wind turbines despite the fact that the country is not very suitable to consider hydropower and wind for power generation. Hence, the RE resource, such as biomass and solar can be utilized to ensure energy security.

RES are the assets that are renewed persistently through natural transformations and can be reused because of their inherent properties. The RES incorporate biomass, solar, wind, geothermal and hydropower energy, but exclude conventional nuclear fuels. Figure 2 exhibits a preview of various RES available for electricity or power generation [47].

From the discussion above, it can be revealed that an immense volume of RES is available in Bangladesh. Even though the use of RE technology has become a worldwide trend, the country yet greatly struggling to utilize these sources. From among the list, biogas, biomass, and solar are regarded as the probable, favorable and productive sources for the sustainable energy generation in the country. Table 1 provides a review of RE resources and potentials in Bangladesh [46].

Even with the availability of all these RES, it is a matter of great concern that only 1.0% of the nation's cumulative electricity generation originates from RE resources; however, RE estimates to provide approximately 19% of cumulative global electricity supply [48]. In this given context, 271 solar can be utilized to ensure energy security.

the Government of Bangladesh has formulated and adopted some strategies and policies to utilize RES for electricity production up to 10% by 2020 [4]. The following segment will provide a brief survey of the available RES in Bangladesh. 272 RES are the assets that are renewed persistently through natural transformations and can be 273 reused because of their inherent properties. The RES incorporate biomass, solar, wind, geothermal 274 and hydropower energy, but exclude conventional nuclear fuels. Figure 2 exhibits a preview of 275 various RES available for electricity or power generation [47].

*Sustainability* **2019**, *11*, x FOR PEER REVIEW 6 of 30

 Renewables are taking a consistently growing pace in the overall energy industry with an evolving number of driving organizations focusing on aggressive inexhaustible power targets [40]. Unfortunately, even such growth cannot decrease the level of GHG emissions that the world desires. Hence, effective, consolidated and synergic strategies and efforts from the stakeholders are required to utilize RES across the board at a quicker pace. Moreover, the significance of an all-inclusive concession to the environment, as advocated by the Paris Agreement, holds guarantees for a feasible

 Fossil-fuel sources, more specifically natural gas, oil and coal, play the lead role in electricity generation in Bangladesh [3]. The electricity demand in the country increased considerably from 0.4 to 1.38 quadrillion between 1997 and 2016 [41] and the country so far could not provide 100% access to electricity, due to *inter alia*, the inadequate volume of energy sources [42]. The total coal assets in Bangladesh were only 1063 million tons, whereas, the natural gas reserve demonstrated to be available with a volume of 9.7 trillion cubic feet (TCF) according to the statistics of 2013 [43]. Besides, to supply fuel in different industries, including energy, Bangladesh imports nearly 1.2 million tons of crude oil and 2.6 million tons of refined petroleum commodities every year [44]. Bangladesh absorbed approximately 2.132 million metric tons of coal, 175.69 kilo barrels of oil per day and 28.37

 Energy application in the country has expanded significantly, and the generation altogether relies mainly upon non-RES. Energy production through these ways adds practically 40% of the absolute carbon dioxide (CO2) discharges by the nation [46]. In 2018, Bangladesh generated energy amounting 7,418 MW, whereas the demand was 11,534 MW. Such a gap between supply and demand results in the impediment of the financial and innovative advancement of the nation. Such a situation has influenced the policymakers to officially undertake activities to utilize the RES to alleviate the

 Bangladesh is blessed to have a generous amount of RES and the effective exploitation of these promises to satisfy the energy need of the country. Among the accessible energy assets, biomass is considered as the significant RES available in the country, which can reduce the utilization of and reliance on non-RES. Solar energy is also very promising as the country gets an immense scale of sunlight, due to its geographical location. It may be pertinent to share that the rural and coastal areas of Bangladesh have been enjoying the benefits of the installed solar photovoltaic (PV) panels. Besides, several government agencies and NGOs undertook some activities for delivering electricity from

243 eco-accommodating development and progression of the world for the future.

244 **3. Prospects of Renewable Energy in Bangladesh**

254 billion cubic meters of natural gas in 2018 [45].

261 energy needs while maintaining the ecological effects.

277 **Figure 2.** An overview of renewable energy sources (RES) and technologies. **Figure 2.** An overview of renewable energy sources (RES) and technologies.


**Table 1. Renewable energy** (RE) potential in Bangladesh.

#### *3.1. Biomass*

276

The natural substance originating from living or dead life forms like the plant, yields, tree and its deposits are considered as biomass. Biomass is contemplated as an outstanding, reliable and sustainable energy source. Biomass absorbs CO<sup>2</sup> for photosynthesis within the sight of sun-based energy to deliver natural mixes required for its development. However, it is well documented that biomass discharges CO2. Hence, even though biomass is a sustainable energy source, it is not relatively so clean for the environment. Notwithstanding, on the planet, approximately 2.6 billion individuals rely upon biomass for purposes, such as heating or cooking [49].

Being a primarily agricultural country, Bangladesh has an enormous volume of biomass assets that incorporates rice shell, animal waste, crop sediment, timber, municipal waste, jute stock, sugarcane residue and additional complementary springs because of the country's rainforest, the ecological and biological system. In Bangladesh, practically 64% of all-out lands are utilized for agricultural reasons [50]. Along these lines, the nation has a huge extent of agricultural deposits from different harvests, including sugarcane, rice, vegetables, jute, wheat, beets, maize, coconut, cotton groundnut and millet developed all the year round. As a result, practically 70% of individuals, immediately or discursively, depend on biomass energy in Bangladesh [51].

Other than biomass, biogas is created by anaerobic processing that can be utilized for cooking, lighting, and power production and the residue can be utilized for fertilizer, fish feed and compost. Thus, Bangladesh has a tremendous opportunity concerning biogas production from current build-ups

and waste assets. It was affirmed that it was possible to utilize 2.91 billion m<sup>3</sup> biogas in Bangladesh in 2012–2013, which was equal to 1.455 billion liters of diesel [52].

Bio-fuel creation is still on the embryonic step in Bangladesh in which ten pyrolysis plants were set up to deliver bio-oil from different biomasses [53]. Be that as it may, the plants are not operating legitimately because of the absence of suitable support and specialized technology [54].

The Renewable Energy Policy of Bangladesh 2008 intends to outfit the possibilities, expansion and utilization of RES. For instance, it is provided for biomass gasification and clean energy advancement, discouraging the use of energy generated from fossil fuel sources. Clean energy from biomass can result in relatively less carbon emission contrasted with the reliance on fossil fuel substances. Therefore, biomass has great potential in Bangladesh for achieving the clean environment goals set by the government.

#### *3.2. Solar*

Solar-based electricity is the most popular, infinite and effective energy source which is well-accepted everywhere throughout the world. Concentrating solar power (CSP) and solar PV are very promising advancements, and the solar-based home system can deliver power using solar-oriented radiation. In the case of Bangladesh, there is a tremendous extent of possibilities to use solar-powered radiation because the country is situated in the topographical area [55]. The country gets a normal daily solar irradiation of 4.2–5.5 kWh/m<sup>2</sup> that can create roughly 1,862.5 kWh/m<sup>2</sup> per year (see Figure 3) [56].

Solar oriented PV panels assume a significant job in the worldwide power sector and offer about 0.7% of complete energy production. Solar PV provides approximately 7.8% of annual electricity creation in Italy, 5% in Germany and 6% in Greece [57]. Japan and China are the Asian driving nations creating around 13.6 GW and 20 GW sun-oriented power individually [58]. Similarly, modern technological advancement in solar home-system and solar-based innovations are progressively appealing and compelling in Bangladesh.

Bangladesh has already experienced a few success and fruitful execution of solar-based energy usage [59]. The nation has very nearly 234 MW energy production potential from sunlight-based home frameworks [60]. Bangladesh possesses a capability of 50,174 MW electricity generation from solar PV, as displayed in Table 2 [61]. With about 5 million Solar Home System (SHS), Bangladesh has the world's largest SHS. Be that as it may, the nation is creating and delivering just 3 MW from rooftop top panels to the national grid. Nevertheless, Bangladesh Power Development Board (BPDB) is attempting to additionally introduce joined solar cycle plants and LED road light, in order to reduce the necessary electricity demand. Very recently, the country has successfully inaugurated the largest solar power plant in Teknaf area with a capacity to produce 28 MW, which can feed 20 MW to the local substation. With regards to concentrated solar power (CSP), this is a promising technology for power generation in which the solar radiation is concentrated to generate high temperature for producing steam in a solar thermal power plant [62]. A number of researchers have identified the potential of using CSP in Bangladesh [63–67]. In 2011, the State Minister for Power, Energy and Mineral Resources mentioned that the Asian Development Bank (ADB) was willing to finance a 10 MW to 20 MW capacity CSP plant inside Kaptai Hydro-electric Plant which was expected to be ready by 2016 [68]. In 2014, Reliance Power commissioned a 100 MW grid-connected CSP plant in Rajashtan, the largest CSP that utilized Fresnel technology [69].



355 **Figure 3.** Global horizontal solar irradiation in Bangladesh [56]. **Figure 3.** Global horizontal solar irradiation in Bangladesh [56].

## *3.3. Hydro*

354

356 *3.3. Hydro* 357 Hydro energy implies a type of sustainable power source which utilizes the force of water stream 358 to generate electricity. The massive flow of water transforms its energy into electricity. Bangladesh is 359 blessed with various rivers and waterways; henceforth, the country holds a decent extent of 360 possibilities for hydropower extraction. BPDB and Bangladesh Water Development Board (BWDB) 361 identified some proper micro-hydropower production places where the hydropower installations Hydro energy implies a type of sustainable power source which utilizes the force of water stream to generate electricity. The massive flow of water transforms its energy into electricity. Bangladesh is blessed with various rivers and waterways; henceforth, the country holds a decent extent of possibilities for hydropower extraction. BPDB and Bangladesh Water Development Board (BWDB) identified some proper micro-hydropower production places where the hydropower installations may provide

a satisfying amount of energy in the future. Table 3 provides a summary of such identification of suitable sites.


**Table 3.** Potential of micro-hydropower sites in Bangladesh identified by Bangladesh Power Development Board (BPDB) and Bangladesh Water Development Board (BWDB) [3].

Sustainable Rural Energy additionally ventures and investigates more conceivable destinations for miniaturized scale hydropower plants in Bangladesh. According to their assessment, Chittagong possesses huge potentials with an expected electricity generation of 135 kW from its rivers and lakes [70].

In 1962, BPDP introduced the first hydro plant in Bangladesh which consists of two units of limit 40 MW, and each one was installed in the Karnafuli stream. Three additional units of limit 50 MW were introduced constantly in 1988. The Kaptai power plant is known as the hydroelectric power plant that utilizes a vertical hub Kaplan turbine [71]. Bangladesh owns huge hydropower potential in Karnafuli waterway, Matamuhuri basin, Sangu stream and in Brahmaputra river [72].

#### *3.4. Wind*

Wind energy is known for its eco-accommodating nature and considered as the best sustainable power source that outfits for future energy solutions. The dynamic electricity is created from the moving air as the kinetic energy of wind provides the turbine shaft. The geographical situation of Bangladesh makes the country suitable for little scale wind turbines [73]. BPDB introduced four units of the primary wind power plant at the Muhuri dam zone of Sonagazi in Feni district with a generation capacity of 0.90 MW. Likewise, BPDB introduced another 1 MW wind power plant at Kutubdia Island in the year 2008 which comprises 50 wind turbines of 20 kW limit each [74].

In addition, various government agencies and non-government organizations have introduced some initiatives to establish more wind plants in the country [75]. The country has also consented to an arrangement for a joint project with US-DK Green Energy (BD) Ltd. Which proposes the assistance of the USA and Denmark to introduce the nation's biggest ever wind control plant of limit 60 MW at Cox's Bazar [3].

#### *3.5. Other Renewable Resources*

Bangladesh can consider utilizing the prospects of other RES, e.g., tidal electricity, oceanic wave and geothermal energy for power generation in the future in addition to biomass, wind, solar and hydro energy [3]. Ocean waves and tides can be considered as a few tornados hit in the Bay of Bengal consistently. Nevertheless, the prospects of these resources are still under scrutiny, and pragmatic endeavors need to be taken in this regard. All these options are yet to be utilized, due to the absence of appropriate information, innovation, and instruments [76].

#### **4. Renewable Energy in Bangladesh: Issues and Challenges**

Bangladesh has a high potential to move towards a better and more sustainable country with greener energy without compromising the country's economic efficiency, human dignity, standards of living, and financial development. Nevertheless, considerable development activities, including a sustainable power source and green arrangements in Bangladesh have had limited accomplishments and are confronting various challenges which are mostly financial, technical, and regulatory in nature [77]. It may be pertinent to share here that these challenges are not unique in Bangladesh's context, rather these present in most jurisdictions.

Since RE projects are more complex, unpredictable and full of unforeseen risks and dangers, the investors in this field may face serious financial challenges having impacts on future development and commercialization of the projects and technologies [78]. For the investors, it is difficult to convince the financial institutions and other financers to get the necessary support. Since they use new technologies that are uncertain, this may cause the payback period to exceed their expectations [79]. Hence, it acts as an impediment to innovative projects. This vulnerability results in high financing expenses for research, improvement and preparation in this sector. Thus, this deceptively raises the cost of clean energy generation, postponing their full assimilation into the energy market.

Sometimes, different RE related projects require huge initial investments. This is again very challenging for the entrepreneurs, since there are already available standard, but cheaper alternatives in the market [80,81]. Besides, there are geographical factors that have effects on the performance of RE projects. Hence, it can be concluded that an already successful RE project in one country probably will not be realistic to another nation for the presence of sunlight-based variables, biomass, wind and sea between nations in the tropical and other areas.

Usually, RE project entrepreneurs are small companies with limited resources [82]. The eagerness for their undertakings depends on the capacity to support the improvement at an insignificant expense contrasted with the gigantic Independent Power Producers (IPP) [83]. In any case, the output of RE projects run by the small companies is not necessarily the same when compared to financially feasible and promising projects, run by the IPPs. Thus, the financial strength of the company is important. All these above-mentioned challenges are crucial for Bangladesh to develop and promote RE in the future. Without addressing these financial barriers, Bangladesh cannot achieve its national goals relating to RE. Some other relevant challenges are discussed as follows.

#### *4.1. Employment of Advanced RE Technologies*

Technological constraints plague the RE sector worldwide, and this phenomenon is not limited to Bangladesh alone [84]. Hence, these constraints are hindering the growth of innovation within the sustainable energy sector.

To begin with, the unreliable power supply with respect to Bangladesh's geographical variables makes it difficult and vulnerability for RE's sustainable advancement [85]. Likewise, uncertain technological innovation ensures the development of the RE industry with negative intensity, whereas, the orthodox methods of energy generation still offers a financially feasible choice to the energy

stakeholders [86]. Furthermore, there is restricted expertise when it comes to productive practices and hardware aspects of RE [87]. The deficiency of qualified labour and skilled workers with adequate engineering or technical training additionally blocks the pace of the innovative improvement in the sustainable energy industry in Bangladesh. In particular, the inefficiency of human resources with engineering and technical skills in handling the equipment hinders security issues on the RE supply. To prepare the industry for these challenges, it is necessary to train and equip the workforce with necessary facilities though such facilities add additional costs to the RE projects. Moreover, due to the technical issues, there has been ample evidence of time-delays in generating sufficient energy through renewable means, and in such cases, the financers lose the interest to invest in RE projects.

The insecurity of available renewable source supplies for long-term and the price instability is making the situations more vulnerable for the biomass energy projects in Bangladesh [88]. Besides, although the government policy promised to introduce tax incentives in the development of RE projects, however, other than the feed-in-tariff (FiT) system provided by the Bangladesh Energy Regulatory Commission Act, 2003 (Act No. 13 of 2003), no other financial relieve system or tax incentive measures have been formulated so far [83]. Such a position also indirectly discourages entrepreneurs in investing in this sector.

#### *4.2. Policy Related Institutional Barriers*

Although the government's intention to promote RE through institutional, legal and regulatory frameworks is appreciable, the policies relating to RE are not firm and comprehensive [89]. The financing bodies in RE projects are more concerned with the heavy return of profits, while the Government of Bangladesh is worried about the allocation of subsidies to bear to achieve the ultimate objective of fuel diversification policy [90]. Hence, the government has to deal with different and conflicting interests with the potential industry players [3]. Such disparities in the RE industry create negative impacts on the investment of RE projects.

Additionally, financial investors find themselves in a continuous dilemma because of the unreliable fuel supply in the industry. Moreover, the stakeholders face several investment barriers, due to lack of practice and governance-related discrepancies, because the regulation of the energy sector in Bangladesh, like in most jurisdictions, is both fragmented and inconsistent [91]. Regulatory expertise is lacking too at times, due to the dearth of technical know-how. In practice, the arrangements relating to RE development and strategy execution in Bangladesh are viewed as isolated exercises as there are more than one authority that deal with this issue and no single authority is made responsible for monitoring the implementation of the Policy.

In general, the sustainable and RE policies are formulated at Ministerial and Parliamentary political process, and, then, such policy is conveyed and executed by the local governments and other agencies which are equipped with specialized technical tools, administrative arrangements, and regulatory structures for proper implementation. However, this top-down methodology seems non-functional in Bangladesh.

The participants and stakeholders in the energy business in Bangladesh additionally seem, by all accounts, to be less organized. There is no proper forum that listens to their perceptions. Even though they are heard occasionally by way of a public hearing before taking any policy initiative by the regulators, their inputs are less counted in the policy-making process, and the whole process is turned as a routine exercise without any significant impact. There is no other better alternative to develop without addressing the issues of active engagements of the potential stakeholders of the RE industry. Additionally, inputs from experts, producers, consumers, and engineers should also be counted to upgrade the ability of RES.

#### *4.3. Legal and Regulatory Concerns*

With respect to the legal and regulatory barriers, genuine and significant initiatives are missing toward the advancement and extension of RE in Bangladesh. Without addressing the regulatory concerns, the policy related to fuel diversification can never be achieved. The Bangladesh Energy Regulatory Commission Act, 2003 (Act No. 13 of 2003) and subsequent amendments directly or indirectly support to implement a FiT scheme to develop the RE industry in the country. However, such laws have limited enforcement and impact in the practical field as a FiT scheme functions in a project-to-project basis. Hence, there is no positive output on such regulations. Additionally, the government needs to consider several other incentives in the RE industry, such as higher selling tariffs and tax reductions or tax relief, etc. Associated administrative and regulatory bodies should reallocate the subsidies from conventional energy generation to RES to sponsor the existing efforts of the RE industry. Such initiatives may progressively reduce the burden of different stakeholders and play a significant role in the growth of the sustainable energy industry.

In the case of regulatory and administrative difficulties to cope with the complex issues of RE, two remedies may be considered. It is important to note that although the Government of Bangladesh initiated several policies to support RE, such policies are not found successful and effective, due to the fact, *inter alia*, that Bangladeshi enterprises still do not consider the RE industry as a suitable place for investment [92]. Hence, the first task of the regulatory bodies is to review and evaluate the policies comprehensively and make necessary practical amendments to refocus the country's target of achieving energy fuel mix to offer significance to the sustainable energy source.

The regulators need to consider distributing subsidies for RE utilization. The subsidies for traditional fossil fuel source ought to be periodically eradicated, as well as converted and reallocated to RE assets to develop the sustainable energy industry establishments [93].

Furthermore, the second task of the regulatory bodies is to address the issues relating to the institutional framework relating to the industry. The absence of an efficient working institutional framework on RE must be overwhelmed by empowering joint exertion between government organizations and private establishments with the ultimate objective to explore the financial and technical viability of RE generation. Enhancing the institutional network between government leaders, enterprises and utilities promise to implement carefully designed RE policies. In addition, a portion of the policies and strategic activities need reviews or explicit clarifications on how the existing or newly incorporated legal and regulatory system and standards would approach and direct the execution of such adopted policies.

#### *4.4. Lack of Awareness*

There are likewise various social difficulties relating to RES and green innovations. There is a lamentable, but noticeable absence of public awareness and participation with respect to feasible advancements in the energy industry in Bangladesh. The public participation programs, particularly in the rural areas relating to RE development within the country, is not satisfactory [50]. Public participation and awareness about the energy sector promise to deliver the required advancement and development of the commercialization process of the energy industry. Hence, the viability of public awareness programs needs to be increased and boosted with the goal to raise public support for the advancement of RE, which promises to prompt effective usage of sustainable power policies and green approaches ultimately [94]. Relevant government organizations can also come forward to help and guide the potential beneficiaries of RE to motivate them to engage and disseminate positive information relating to the industry.

#### **5. Renewable Energy in Bangladesh: Legal, Regulatory and Policy Aspects**

It has already been shared that energy production, management, and supply to the people have become a serious concern for the Government of Bangladesh. Therefore, the government has formulated various policies and adopted regulatory measures [42]. In this context, the contribution of RES in the energy mix can be an exceptional accomplishment that would influence the economic development of the nation significantly.

In 2008, the National Renewable Energy Policy was introduced to promote RE by encouraging distinctive private and public platforms concerning the investment in the RE industry in place of the fossil fuel-based power sources. Since then, the uses, development, preparation, and research on the domestic RE industry have been accelerating; though for exploiting the commercial benefits of these, it will still require a comprehensive strategy.

In recent times, the domestic RE generation and the government's participation in expanding electricity generation through RE increased approximately 560 MW [95]. Bangladesh attempted to increase the electricity supply up to 16,000 MW by 2016 and subsequently, set the target to obtain up to 39,000 MW by 2030 according to the adopted the Power System Master Plan 2010 (PSMP) [96]. The electricity generation target as encapsulated in the PSMP is presented in the following Table 4.


**Table 4.** Power System Master Plan 2010 [97].

In order to meet these targets, the Government of Bangladesh initiated multiple projects, including the utilization of coal as a commanding energy source for electricity generation and supply in the future [98]. Accordingly, two large coal energy plants with a target to produce 1320 MW are under construction at Khulna and Chittagong. Moreover, the government has undertaken major policy decisions to set-up nuclear energy plants to ensure reliable and environment-friendly electricity for the future [99]. The atomic power plant is anticipated to add another 2000 MW by 2020, and the government wishes to increase to 5000 MW by 2030. In implementing those wishes, the government initiated the process of installation of nuclear power plants in the country. Accordingly, the long-cherished nuclear power plant, with the technological support of Russia, is under construction at Rooppur in the Pabna district [8].

Despite the adoption of so many energy policies, it seems that the government may not be successful in achieving the desired objectives unless it exploits the RES to produce energy. The Government of Bangladesh, hence, concentrated on electricity production up to 500 MW by RE within 2015; but failed to succeed.

Fortunately, the government recognized the reality, revised the existing policies and strategies and attempted to formulate pragmatic strategies for the sector. Consequently, the policymakers of Bangladesh aim to achieve its objectives by producing 10% of its total electricity demand by 2020 from RES [46]. Now Bangladesh is producing around 560 MW of power from renewables which contributes 2.95% of the cumulative energy generation of the country. Specialists figure that achieving 10% of the aggregate supply with around 2000 MW of electricity by the next two years will be a challenge for the country. In accomplishing the goals, the government has initiated numerous endeavors particularly, to achieve the energy-specific objectives.

The government attempts to fostering the cooperative approaches within private and public sectors in reaching that milestone. Accordingly, different government organizations, such as BPDB, Bangladesh Rural Electrification Board, Bangladesh Council of Scientific and Industrial Research and Local Government Engineering Department, started executing numerous RE related activities. Infrastructure Development Company Limited, on the other hand, is a private NGO, is trying to develop the market competitiveness by commercializing RE in rural areas of Bangladesh.

Securing access to electricity for all is one of the prime visions of Bangladesh since its independence, which has been evidenced by incorporating the State's promise in the Fundamental Principles of the State Policy (FPSP). Rural development and agricultural revolution are important FPSP as enshrined in article 16 of the Constitution of the People's Republic of Bangladesh, 1972. In this article, Bangladesh

has given words of undertaking adequate measures to bring radical change in rural areas by introducing a revolution in agriculture, ensuring rural electrification, developing cottage and other industries, and improving the education system. In addition to these, it is further provided that the State shall endeavor to develop the communication and public health systems with a view to removing the disparity of living standards between rural and urban settings. The provision of affordable and continuous energy is instrumental to achieve this.

Nonetheless, since independence to date, Bangladesh is suffering in ensuring 100% access to electricity for all. In addition, the findings of various scientific studies have revealed that domestic fossil fuel sources are either depleting or will naturally be depleting in the course of time. Therefore, the promotion of RE has been taken as an alternative source in the Renewable Energy Policy, 2008 for the long-term energy sustainability, pollution control and electricity generation. Because fuel diversification contributes to sustainable electricity production [100], and consumption within the socio-cultural and economic parameters, Bangladesh is in great need to promote energy production through renewable means.

By assessing the international environmental obligations, the country made in the international forum, Bangladesh should design the national policies and regulatory frameworks for RE. From various initiatives taken by the Government of Bangladesh, it is evident that the government truly values sustainable development [101]. For example, the Government of Bangladesh has signed and ratified most of the international environmental law instruments and have been submitting the required state reports regularly. Additionally, the government has enacted national laws and taken policy initiatives to give effect to the provisions of these international instruments.

In order to understand the energy policies of the country, it will be relevant to share here energy-related legal and policy instruments. In the undivided India when present Bangladesh was part of it, the Electricity Act, 1910 (Act No. IX of 1910) was enacted containing provisions, *inter alia*, on grant of licenses for the supply, transmission and use of electricity. That law provides for punishments for a number of offences such as dishonest abstraction of energy, installation of artificial means, maliciously wasting energy or injuring works, theft of line materials, tower members, equipment, etc., from any electric supply system, dishonestly receiving stolen property, unauthorized supply of energy by non-licensees, illegal or defective supply or for non-compliance with order, illegal transmission or use of energy, etc. After more than a century, the government repealed the law and enacted the Electricity Act, 2018 (Act No. VII of 2018) containing similar but updated provisions on these matters to meet the continuously increasing demands of electricity. It will be pertinent to mention here that the law does not contain provisions on the sources of energy, i.e., energy generated from fossil-fuel sources or RES, rather the main focus of the law is to ensure the supply, transmission and use of electricity.

After the independence of the country, the government enacted the Rural Electrification Board Ordinance, 1977 (Ordinance No. LI of 1977) to establish the Rural Electrification Board that will primarily be responsible to take measures for effective use of electrical power for development of the rural economy of the country. This Ordinance of 1977 was also repealed, and the government enacted the Rural Electrification Board Act, 2013 (Act No. 57 of 2013). Same as with the Electricity Act, 2018 (Act No. VII of 2018), the provisions of the law are mainly concerned about the supply, use and transmission of electricity in the rural area and the sources used to produce electricity is not the primary concern. The law provides to establish a body known as Bangladesh Rural Electrification Board (BREB) to this end. Most importantly, this Board has been playing a pivotal role in the promotion and utilization of RES in the rural area of the country as it has taken various initiatives [42], including the introduction of SHS for the first time in the country in 1993 through the project 'Diffusion of Renewable Energy technologies' with the financial assistance of France. In the agro-based economy, BREB is now working to install 2000 solar irrigation pumps.

Moreover, a list of such instruments include: Policy Guidelines for Small Power Plants in the Private Sector, 1996, Private Sector Power Generation Policy, 1996, the Bangladesh Energy Regulatory Commission Act, 2003, Import Duty Exemptions for Solar and Wind of Bangladesh (Statutory Regulatory Order), 2004, Renewable Energy Policy, 2008, the Sustainable and Renewable Energy Development Authority Act, 2012, Scaling Up Renewable Energy Program for Bangladesh, 2015(SREP Bangladesh), Bangladesh Energy Regulatory Commission (Tariff for Roof Top Solar PV Electricity) Regulations, 2016 (Draft), etc. Some of these are evaluated and discussed below-

#### *5.1. Policy Guidelines for Small Power Plants in the Private Sector (1996)*

In fulfilling the promises encompassing energy supply as enshrined in the Constitution, the Government of Bangladesh has taken numerous policies on power production, supply and consumption. Even though Bangladesh has to depend on large-scale nationalized efforts to produce sufficient electricity for the national grid, it has also been encouraging privatized endeavors. In line with the aims, the Government has formulated a Policy Guidelines for Small Power Plants in the Private Sector, 1996 allowing private investors to install Small Power Plants (SPP) at the earliest possible time for generating electricity on a commercial basis. Accordingly, private parties could generate power for their own and sell the residue, if any, to anybody else. Initially, the plant size was fixed for producing up to 10 MW and allowing, even more, based on necessity, demand and loading capacity [102].

To encourage private entrepreneurs in power production, the Policy suggests that Petrobangla, a government-owned national oil company of Bangladesh, may supply natural gas to those power plants on a commercial basis with a usual rate if the SPP is not located too far from the gas supply reticulation (section 3). To ease the business, the Policy also provides that the sponsors can use the extant transmission and distribution systems if there is no problem in terms of capacity though the owner of SPP or the sponsors will have to pay a mutually set wheeling charge for using transmission/distribution facilities (section 4). Moreover, it is also provided that the government would not interfere in pricing; rather, it would be fixed on negotiation between the sponsor and the consumers (section 5).

The provisions of 'fiscal and other incentives' for national and foreign investors are included under the heading of "Captive Independent Power Producer' and 'Captive Power Generation' policies in sections 5 and 6, respectively. Moreover, the Policy provisions clarified that the government has no obligation of purchasing the power produced by SPP. Nevertheless, the government may, if it considers essential, purchase the power from the SPP (section 9). Globally, community-based energy production and supply systems are gaining momentum. Many scholars suggest that Bangladesh can consider even the household level system of energy production and distribution in resolving its energy crisis though which it may add sufficient power into the national grid [103].

#### *5.2. Private Sector Power Generation Policy (1996)*

The Government of Bangladesh formulated the Private Sector Power Generation Policy (PSPGP) in 1996 with a view to ensuring the participation of the private bodies in the power generation leading to promoting the economic growth of the country. The PSPGP set several specific objectives, such as ensuring the access to electricity for all; increasing the annual per capita generation of power; reaching and sustaining minimum 6-7% annual GDP growth; achieving desired socio-economic progress by alleviating poverty; securing adequate electricity supply at an affordable cost by expanding the production of electricity as a whole.

Starting from 11666 GWh, the Policy set a target of generating 16500 GWh power within 2000 and 24160 GWh by 2005, which figured out that an average of 300 MW of power has to be generated more annually. It was estimated that a total of US\$ 6.6 billion would be required for expanding, reinforcing, transmitting and distributing electricity (section 1.3).

Other provisions of the said Policy include: Formation of a power cell for facilitating the promotion, improvement, execution, commissioning and operations of private energy production schemes (section 2.0); modality for implementation of private power projects (section 3.0); financing arrangements (section 3.3); security package (section 3.4); tariff for bulk purchase of power at busbar (section 4.0); fiscal incentives (section 5.0); other facilities and incentives for foreigners (section 6.0), and right of interpretation (section 8.0) [104]. Taking the advantages of the Policy, the first private power plant

was installed in October 1998, having a capacity of producing 110 MW power to add in the national grid [105].

In Bangladesh, there was no engagement of the private bodies in the energy generation and distribution sectors in the early 1990s. Since 1996, the Government of Bangladesh undertook several reforms plans under the National Energy Policy to offer competition, bring foreign investment with a view to increasing the energy generation and distribution. One of the key Policies was the Policy Guidelines for SPP in Private Sector, 1996 discussed above [106]. During the period, some private energy generation schemes were established as the IPPs selling power to the BPDB. However, due to the friendly investment policy of the government, now private enterprises dominate the energy sector contributing a total of 54.35% in the national power generation [107].

#### *5.3. The Bangladesh Energy Regulatory Commission Act, 2003*

With a view to achieving some broad goals, especially, making provisions for the establishment of a self-governing and fair energy regulatory commission, the Parliament of Bangladesh enacted the Bangladesh Energy Regulatory Commission Act, 2003 (Act No.13 of 2003). One of the prime objectives of the Act was to establish the national regulator Bangladesh Energy Regulatory Commission (BERC) for regulating the electricity, gas, and petroleum commodities in Bangladesh. The vision of the BERC was to foster an amicable environment to establish an efficient, properly organized and sustainable energy sector in Bangladesh for ensuring energy at a fair and sensible price, and to secure the interest and satisfaction of the customers by fair practice [108].

Major provisions of the Bangladesh Energy Regulatory Commission Act, 2003 are as follows: The establishment of the BERC as a legal person (section 4); functions of the BERC (section 22); emergency power of the government to control energy use (section 25); settlement of disputes (section 26); issuance of license by the BERC (section 28); renewal, revision and cancellation of license (section 30); tariff (section 34); emergency provision (in favor of the BERC) (section 36); restrictions on publishing information (section 39;) arbitration-settlement by BERC (section 40); appeal against the decision of inspector (section 41); penalty (section 42); penalty for stealing energy (section 44); penalty for obstruction of the construction during the installation or repair of electric line or gas pipeline, etc. (section 45); jurisdiction of trial court (section 50); collection of fee, fine and charges (section 56); power to make rules (section 58) and regulations (section 59), and provisions for issuance of license during transitional period (section 66).

The duties and responsibilities of the Commission include, but not limited to: Creating a conducive atmosphere for the private investors in power generation activities; ensuring transport, transmission, and promoting the petroleum products; maintaining transparency in every sphere of the sector, and finally, to protect the rights of the consumers by creating the competitive market [109]. However, the BERC is conducting widespread activities in line with its missions and visions as a major regulatory body since its inception in 2004.

#### *5.4. Bangladesh (Statutory Regulatory Order), 2004*

The government framed the Import Duty Exemptions for Solar and Wind of Bangladesh (Statutory Regulatory Order), 2004 [110] targeting certain resources, such as solar, SPV, solar thermal, solar heat, wind in the electricity, heating and cooling sector. It is applicable for both small- and large-scale ventures of RE project. Through the provisions of this Order, the National Board of Revenue and the regulatory code 155, an exemption in import duties for certain RE products, in particular, the solar cells, modules and lanterns are offered. Additionally, according to this instrument, the rate of duties imposed for other RE resources is comparatively inexpensive too. Moreover, these duty exemptions are not limited to the national companies only, rather extended to foreign companies also. As a result, the duly registered foreign companies also enjoy the same facilities as are enjoyed by the locally owned companies [110].

#### *5.5. The Renewable Energy Policy of Bangladesh, 2008*

The Power Division, Ministry of Power, Energy and Mineral Resources, Bangladesh formulated the Renewable Energy Policy on 18 December 2008 [4]. The Policy identified the following three reasons of global crisis encompassing the energy, e.g., the gradual decline of fossil fuel, and the consequent price fluctuation because of the gap between the demand and supply; the necessity of reducing global emissions (up to 80% by 2050) in responding to the climate change, and the continuous demand for energy security.

The objectives of the Policy are to put to use the potentials of RE resources and technologies everywhere; encourage and facilitate both public and private sector investors in RE sectors; scaling up the power generation; conducting training facilitating the use of RE in each unit of energy usage; providing encouragement, legal and environment assistance in the use of RE, etc. In particular, the ultimate goal was to produce 5% of the total electricity demand utilizing RES by 2015 and to increase the portion up to 10% by 2020 [4], which means that 2000 MW has to be generated from RES. Some of the main provisions of the Policy include: Institutional arrangements (section 3); provisions regarding the resource, technology and program development (section 4); investment and fiscal incentives (section 5); regulatory policy (section 6), and the right to interpretation (section 7), etc.

It may appear that the production costs of RE are comparatively higher than the fossil fuels; however, they can be economically viable if all other ancillaries (eco-friendliness, beneficial to health, low operating costs) are taken into consideration. In recent years, Bangladesh is witnessing commendable progress in the RE sector because of the implementation of several policies, particularly, the Renewable Energy Policy of Bangladesh, 2008. Recent statistics show that the country is generating a total of 404 MW of electricity from the RES (see Table 5). Meanwhile, the solar home system has appeared as a success story and getting popularity gradually, principally in the off-grid areas [111].



#### *5.6. The Sustainable and Renewable Energy Development Authority Act, 2012*

To secure the energy security, the government established the Sustainable and Renewable Energy Development Authority (SREDA) though the enactment of the law—the Sustainable and Renewable Energy Development Authority Act, 2012 (Act No. 48 of 2012). While enacting the law, the policymakers of Bangladesh realized the need to control global warming, prevent the misuse of energy, reduce the hazard of natural disasters and gradually reduce the dependence of fossil fuel in power generation by promoting the use of RE.

According to this Act, energy refers to power generated using both the renewable and non-RE sources [section 2(5)]. The law bestows legal personality to the authority, i.e., SREDA [section 2(2)]. Some other relevant provisions of the law include: Responsibilities and functions of the authority

(section 6); the power of the authority to impose fees (section 7); the constitution of the board of directors (section 9); fund, budget, and accounts and audit (sections 19–21); power to make rules and regulations (sections 26–27), and abolition of Energy Audit Cell, etc. (section 28).

In spite of having diverse plans, so far, Bangladesh achieved a limited success encircling the RE sectors because of numerous reasons, and principally, the lack of project execution experience; lack of strong institutional capability; financial challenges; land constraint; project development challenges, etc. Keeping all these in context, the Government of Bangladesh established the SREDA in 2014 to promote the RE sector and achieve energy efficiency [112]. Since its inception, the SREDA is actively operating multiple action plans with a view to ensuring electricity for all within 2020 by increasing the RE share 1.5% to an increase of 50% in grid supply [113].

### *5.7. Bangladesh Energy Regulatory Commission (Tari*ff *for Roof Top Solar PV Electricity) Regulations, 2016 (Draft)*

The government has recently drafted the Bangladesh Energy Regulatory Commission (Tariff for Roof Top Solar PV Electricity) Regulations, 2016 [114]. Once approved, the provisions of the Regulations shall apply to all new RE power plants operated for generating and promoting the sale of electricity produced in such RE power plants. It is also provided that in the Draft that in the current RE power plants, all terms and conditions, tariffs, etc. shall be administered by the prevailing notifications, of course, with the prior approval of the BERC. All terms and conditions required for getting a license are governed as per the provisions of the Bangladesh Energy Regulatory Commission Act, 2003, Renewable Energy Policy, 2008, and the subsequent amendments thereof. In addition to these, no license is required for a plant installed for producing up to 5 MW, and in such a case, a waiver certificate shall have to be taken subject to the fulfillment of the conditions determined by the BERC [114].

BERC remains as authority to govern specific tariffs concerning RE production. The BERC has been working to conceive similar laws and regulations pertaining to the specific FiT system, especially for solar and wind energy ventures. The government in their several policies and laws has mentioned that a broad range of RE projects may serve the country with prosperity that can additionally address four key intentions, i.e., Efficient Energy Access, Energy Security, Industrial Advancement and Environmental Protection.

The draft Regulations imply to undertake FiTs as the primary common essential measure to promote RE in Bangladesh. Such a system is assumed to provide an outstanding result relating to the innovative investments in the RE industry. To improve the RE industry, and enhance the grid-connected electricity production, it is essential to have an efficient administration and the adoption of recent innovative technologies. Such a contribution of RE can deliver advantages concerning every stakeholder in the energy industry of Bangladesh.

The Draft Regulations of 2016 is a substantial foundation to initiate supplementary innovations for reliable and affordable energy solutions. The draft Regulations proposed for a universal FiT concerning the special model of RE technology managed by BERC on a specific basis that is presumed to initiate assistance for new investors. Such a universal approach is much effective than the existing project-by-project strategy. This approach concerning universal FiT for diverse RE exercises has produced concrete outcomes in several nations. Such a method is also capable of reducing excessive profits of private energy enterprises. Nevertheless, the regulations have been in the draft stage since 2016, and have not been enforced yet.

In achieving the desired goals, the Government of Bangladesh has planned to generate electricity by the following means and approaches:


Net energy metering (NEM), also known as net metering, which allows the prosumers to use the produced electricity anytime of personal need, is gaining popularity all over the world. Such an initiative encourages the prosumers to produce more energy using RES. It is a matter of great hope that the Government of Bangladesh has actively been considering to introduce the NEM system in the country and has released the draft of the Guidelines with the aim to support the government to develop a net metering policy for individual energy generation. It is anticipated that such an initiative, when successfully be implemented, will encourage the consumers to use their self-produced electricity reducing the dependency on the grid power. Thus, such an initiative offers various benefits - the consumers will need to spend less on their electricity bill, emission of GHGs will be reduced, government needs to spend less on the import of fossil-fuel sources for energy production and the so save national money can be used for other socio-economic development activities.

Recently, the BERC has formulated the Electricity Grid Code, 2018 to govern the boundary between the licensee and users, and to establish the procedures for operations of facilities that will use the transmission system. This Grid Code specifies criteria, guidelines, basic rules, procedures, responsibilities, standards and obligations for the operation, maintenance and development of the electricity transmission system. It is anticipated that this Code will help to ensure a transparent, non-discriminatory and economic access and use of the grid, whilst maintaining a safe, reliable and efficient operation which will facilitate to provide a quality and secure electricity supply as reasonably as practicable. Most importantly, this Code contains provisions on RE.

Finally, it can be submitted that the policies, such as policies on fuel diversification, energy efficiency improvement and low carbon emission, can never be achieved without promoting RE in the country. Hence, the government is firmly committed to bringing every possible solution to increase the contribution of the RE industry in the national energy mix and thus, has taken various initiatives. While the initiatives taken by the government can be applauded, there are still avenues to improve the situation and therefore, we have advanced some suggestions and recommendations in the following segment.

#### **6. Suggestions and Recommendations**

### *6.1. Government Leadership*

Achieving green and RE goals require the presence and coordination of synergic move by the stakeholders under the auspices of the appropriate regulatory watch. The Government authority is fundamental in planning proper RE approaches. Bangladesh has organized different activities to advance a sustainable power source [115]. However, there is correspondingly a need to give arrangements and support to outstanding business people, enterprises and the business networks in supporting the practicable power drive. This requires a wide understanding and contribution at different dimensions.

Bangladesh still battles to keep its international promises to enhance the usage of renewable resources to promote a green environment. The heavily anticipated Renewable Energy Policy leads Bangladesh's change into a green country and latches onto the prospects of practical improvement. Nevertheless, although the policies attempted to develop the RE industry in Bangladesh and achieve its sustainable goals, there remain many failures on the part of the government agencies in exploiting RES to generate energy.

The awareness regarding environmental sustainability and the need for creative alternative sources for the energy sector are a global phenomenon. Whatever the case, it is imperative for the policymakers of Bangladesh to categorically execute the propelled tasks and plans successfully. It has been seen from historical encounters that there remains an immense mismatch in accomplishing the goals of power designs. Along these lines, the Government of Bangladesh should take the lead position with the end goal to advance RE and execute the undertakings that create RE inside its regional area for the benefit of its citizens.

As the economy of Bangladesh is shifting toward industrialization and the shifting into industry-based economy produces increased earnings to the citizens and stimulated urbanization. Fast accelerating commercial ventures, including increasing income levels, produced an immense need for transportation settings, particularly in immediately developing metropolitan cities. Thus, the country is also increasing its contribution to polluting the environment within the region.

While the economy is fast developing, a difficult assignment remains for the Government of Bangladesh, i.e., to fulfill its expanding need for electricity. It is a fact that the costs of fossil fuel sources far and wide are unpredictable. The lack of adequate natural gas reserves in the country intensified the situation further. In total, these two forces increase Bangladesh's electricity demands. At present, Bangladesh supplies her electricity prerequisites overwhelmingly through gaseous petrol, additionally supported by a bunch of coal and oil plants. Consequently, because of the shortage of gas, coal and oil assets, sustainable power source surfaces as a precise vital alternative for Bangladesh's future advancement plan. Nonetheless, a productive, sustainable power source program involves a broad framework. Therefore, the Government of Bangladesh should take a leadership position in order to promote RE and implement the projects that help develop RE within its territory.

Other than the traditional approaches, the Government should also focus on the modern technical and regulatory approaches to overcome the challenges that they have been facing. Building codes, energy rating schemes, supportive grid connection arrangements, etc. can also be very useful tools in promoting the RE industry. Building codes are the significant mechanism which provides indirect incentive to promote RES. For instance, Australia has been using energy rating schemes that provide criteria for calculating the credit system for PV and solar water heaters to empower new buildings or substantial renovations with the capacity to generate electricity [116]. Such set up in commercial buildings generally implement solar PV to achieve higher energy ratings, which then helps them to achieve higher rentals. It cannot be ignored that for such construction with RE tools, the owners had to make the huge initial investment; however, such appearance helps the construction firms with tax incentives from the government and add to high resale value, especially in the countries where the requirement of energy rating on building sales is mandatory. Hence, building codes and rating schemes can help to promote a green environment and reduce electricity needs with limited public investment. Additionally, regulating little categorized generators like residential PV linking with grid connection can enhance the energy efficiency of the country. Such specific regulation can be a significant step to reduce the skepticism and danger encountered by private sector investors and also reduces management expenses of the energy industry.

#### *6.2. Ensuring Renewable Energy Development through Technological Means*

To achieve public procurement on RE and to enhance the output of technology, it is important to encourage research, especially when the government intended for the localization of foreign imported technologies. Advancement of confined clarifications and explanations on innovative technologies or methods can frame civic perspectives, including public participation. Additionally, RE establishments in Bangladesh are designed by regional corporations, including external specialists. Information and technology directions are prerequisites to encourage the potential raising in the RE industry and to support the progressive growth for long-term goals even though such technology is imported by foreign ventures and experts.

The idea of RE and energy efficiency could be brought into the curricular exercises in conventional educational and vocational institutions, such as schools and colleges. One of the major challenges relating to RE is the scarcity of sufficient human resource with relevant skill, expertise and technical knowledge in regards to, *inter alia*, handling of equipment [115]. This is further exacerbated by inefficient energy practices, the reluctance of administrative and regulatory bodies, and lack of culture embracing innovations.

Hence, instruction developed and programs preparation on sustainable energy at the tertiary training level promise to facilitate the advancement of RE in the country. Such advancements and developments seem pre-requisites for achieving the sustainable goals of Bangladesh.

#### *6.3. Minimization of Renewable Power Generation Cost and Promoting Market Competitiveness*

The costs of generating and transmitting power in the modern age despite all the technological advancements are staggering [117]. The matter is manifold complicated for the RE sector as the industry yet does not benefit from economies of scale. When RE is compared with existing fuel-based energy sources, the economic benefit is prospectively and considerably diminutive [118]. Hence, the major challenge relating to the RE is the difficulties in anchoring the investment in the industry.

In any case, the bankers and financial investors seem to have a lack of confidence to make required speculation about the RE industry. Additionally, the long process of granting bank loans for RE demotivates the participants to engage in business [119]. In this manner, the Government of Bangladesh needs to build up a worthwhile, practical, and controllable subsidized legal, regulatory or political instrument for RE projects.

Because of the absence of such instruments, both the financial specialists and industry players consider it monetarily not lucrative to invest in the RES activities and advancements. Consequently, lack of coordination and non-responsive attitude on the part of the administration decreases the chances of prospective foreign or local investments on RE projects.

Other project motivators, for example, tax incentives, research funding or any other technological support from the part of government can bring down the expense of RE innovation [120]. European nations have embraced such sponsorships and motivating forces as a major aspect of their advanced techniques to build such a complex industry like RE [121]. Financial help, dedicated lessons and embracing the RE experts to create a comprehensive RE study syllabus for universities, colleges and even in schools might catalyze the discourse about advancement, exhibition and systematic awareness-raising campaigns related to the sustainable energy source and green innovations.

In Bangladesh, until now, RE significant ventures have been actualized with the project-to-project base assessment. Specific separate investors invested, and the government fixed the tariff-based on power purchase agreements (PPAs). The countries which have long experiences of RE production, such as Germany, India, China or Spain, have never followed project-by-project contracts. Instead, a market approach was established to drive the considerations and regulation of energy generation stipulations. With the establishments of FiT regime, private sectors of the said countries plan for their investments, based on the offerings of RE premiums (the extra amount in addition to the market price for electricity) or RE PPAs. Furthermore, the outcome of certain applications has been enormous, covering the preceding ten years concerning the penetration of the RE industry as reducing the expenses associated with it.

#### *6.4. Creating a Comprehensive Policy, Legal and Regulatory System*

The regulatory framework plays a significant role in the advancement and development of RE technologies within a country's existing energy formation. The regulatory direction is necessary to enhance the competitiveness of the energy market. Such a regulatory framework is also vital to ensure that the stakeholders comply with the established rules and regulations for energy trading. Any deficiency in the regulatory structure may cause a threat to the market liberalization process. Likewise, the regulatory framework ought to increase public awareness about RE technologies and incentive mechanisms to promote investment in the RE industry.

World Bank's 'Turn Down the Heat', and The Intergovernmental Panel on Climate Change's 'Fifth Assessment Report' were two reports which uncover large haul suggestions for Bangladesh and its kin from likely cataclysmic effects of environmental change. The two reports draw a terrible situation to face in the future if the environmental changes are not addressed.

The Government of Bangladesh has been allocating funds in the research targeting the control of environmental degradation. Regardless of these activities, as all examinations call attention to, Bangladesh will remain helpless, and its kin will confront extreme monetary difficulties from environmental consequences in the future. Accordingly, Bangladesh should approach all countries to target a low-carbon economy dependent on sustainable power source supply. Such a request can only be considered positive, given that Bangladesh itself change approaching towards environment-friendly ventures.

The Government of Bangladesh has schemed different innovative approaches, for example, the Renewable Energy Policy 2008. Be that as it may, there must have a subsequent execution of the components set up by the Policy. Other than that, there must have severe courses of events for each strategy for achieving administrative success.

Financial experts urged that a fruitful sustainable power source framework needs long haul security to recuperate the costs of investors and to support their revenues. The draft Regulation released by BERC in 2016, includes remarks and commitments from various partners, which appears coordinating and immensely conducive in preparing an appropriate structure. Nevertheless, the most critical part of any policy for achieving success is its unwavering quality and dependability. Ventures need a considerate long-haul connection between proprietors of these offices in one hand, and the circulation or transmission matrix between administrators is necessary to deal with the price of electricity produced, on the other. This dependability must be demonstrated in the policies also and ought not to be influenced by political insecurity or regime change. Policies, in general, should incorporate every statutory and legal term relating to the grid technology, including energy reliability, security and essential expansions.

Grid access, including a flexible licensing system, for RE is vital concerning advancement in the treatment of solar PV and wind power. Additionally, complementary measures for the RE generation, such as land acquisition, preparation of human resources, tax exemptions or tariff system, etc. should be comfortable to procure. The draft Regulation of 2016 containing provisions on FiT is promising as the provision contains no fees for licensing in the installations of RE plants with the capacity up to 1 MW. Notwithstanding, a special provision ought to additionally render the guidance on the usage of lands, tenancy contracts, operational standards, availability of equipment and the technological expectancy.

Furthermore, to improve the monetary appeal of RE, there should be no retroactive reduction of tariffs or modification of practices for existing plants under the agreement. To identify every single installation correctly, a registry should be created, preferably by SREDA containing the publicly available information on the type of equipment, year of installation/grid connection, the expiry date of FiT contract, etc. Such registry should be declared "protective" against retroactive changes by law or regulation. Concerning the vulnerable situation in Bangladesh for the energy industry, advanced policy and comprehensive regulatory framework are essential to obtain all possible solutions.

#### *6.5. Increasing Judicious Public Relation Work to Spur Public Awareness*

The government's decision to tackle climate change crises and developing energy in a sustainable fashion is receiving more traction. From nearly two decades onwards, the Government of Bangladesh has adopted various policies and action plans to promote implementing greener and RE as an alternative source. In spite of continual strides made within this field, developing true renewable sources is not yet able to fulfill its utmost potential. Such a sluggish response is holding back the achievement of broader energy goals [122].

The tepid acceptance of renewable alternatives by the general public is another important factor for diffusion and solidification of RE development [8]. This is why a certain necessity exists for raising the general public's awareness regarding the pros and cons of using fossil fuel generated energies and the benefits of using sustainable and RE. For this to happen, appropriate campaigns and programs should be designed with the help of modern information and technological communication tools. Furthermore, education and material propagation tied to RE should be made as part of the curriculum and education course design at all levels of schooling. Without a proper introduction to the costs and benefits of RE resources and technologies, the public cannot be expected to be conscious about achieving the nation's broader goals.

This process can be helped through government-subsidized programs and initiatives. Besides, once the public comes to know the pivotal role played by RE, research and development programs will most likely be receiving greater funding leading to the development of more efficient and cheaper technology. In this regard, the Government of Bangladesh may want to prioritize budget allocations for the research and development of technologies in engineering and relevant environmental science disciplines [123].

#### **7. Conclusions**

The planet-wide impacts of environmental change and the nature of global warming urge to adopt 'hard' International law and relevant arrangements so as to advance electricity generation through a sustainable manner. Notwithstanding, there is no urgent or explicit authoritative decision that binds the countries legally to advance the local utilization of RES to generate electricity [6], though the provisions of the recently adopted Paris Agreement to UNFCCC encourages the use of RE. The scarcity of universal legal arrangements on RE does not, in any case, undermine the significance of its utilization to limit environmental degradation. Henceforth, a binding international instrument, alongside the positive activities and mutual cooperation of the stakeholders worldwide, regional and municipal players in terms of financial and technological information exchange, may assume an instrumental job in the advancement of RE [6].

Since independence, Bangladesh remained extremely reliant on fossil fuel sources to generate electricity, and these sources are depleting naturally. Continuous reduction of these fossil fuels sources, increasing expenses of such sources, energy security and international attention on GHG emissions drive Bangladesh, like many other countries, to consider sustainable policies for the electricity industry. The Government of Bangladesh is constitutionally obliged to eliminate specific inequality in living standards within the metropolitan and rural areas. Moreover, the government is constitutionally bound to achieve a specific end of development by ensuring rural electrification [124]. Therefore, the government has taken various initiatives, including the formulation of the Renewable Energy Policy 2008, aiming at promoting RE. Besides, the government has adopted policies to establish the SREDA with the prime purpose of obtaining sustainable advancement and improvement of the RE industry. Besides, Bangladesh adopted other policies, *inter alia*, to advance the industry by providing subsidies and tax incentives to draw investors into the industry. The investment relating to RE are commonly expensive in comparison with fossil fuel-based power plants, though RE plants can be profitable if one considers every external opportunity that RE offers, such as environmental protection, sustainability and inclusive social development, etc.

Achieving the goals of sustainable and greener development of energy infrastructure in Bangladesh is constrained by several issues in terms of the development and frameworks. A practical breakthrough is necessary to surmount the high upfront costs borne by developing a fresh market to let the RE market growth. This will have the effect of scaling up the renewable sector and drives down the cost of technology. In order for Bangladesh to make headway in building a low carbon society, legal instruments on RE are in dire need of institutional patronage to ensure level playing fields for all key players, the stakeholders, and the general public. Since it is undeniable that energy is the key antecedent for wide-scale development in the global arena, ensuring this development is

sustainable and counters inequality and remains inclusive for all segments of the population is equally relevant [125]. The government agencies are not the only ones who should make efforts. The private sector should also be more corporate, socially responsible and make compromises by accepting longer payback periods in RE projects.

Globally, energy is imperative to pursue any type of development, and for attaining sustainable and all-inclusive development [6]. In addition, creating a friendly green climate for the future remains a challenge for every country, not particularly, for Bangladesh [126]. To mitigate climate degradation and to achieve the SDG goals, Bangladesh should lead its energy sector into renewable sources. Providing an equitable opportunity to its citizens and to promote inclusive development for the society, choosing RE is one of the most strategic options that Bangladesh has in its hand now.

**Author Contributions:** Conceptualization, M.E.K. and R.K.; Data curation, M.E.K. and R.K.; Formal analysis, M.E.K., R.K., M.T.I. and F.M.-S.; Funding acquisition, F.M.-S., N.A.B. and M.N.M.; Investigation, M.E.K., R.K., M.T.I. and F.M.-S.; Methodology, M.E.K. and R.K.; Project administration, M.E.K. and F.M.-S.; Resources, M.E.K.; Software, F.M.-S.; Supervision, M.E.K.; Validation, M.E.K.; Visualization, F.M.-S.; Writing—original draft, M.E.K., R.K., M.T.I. and F.M.-S.; Writing—review and editing, M.E.K., R.K., M.T.I., F.M.-S., N.A.B. and M.N.M.

**Funding:** This research was partly funded by Universiti Teknologi Malaysia (UTM) and Ministry of Education, grant number Q.K 130000.3556.06G43 and R.K 130000.7756.4J380.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Residential Energy-Related CO<sup>2</sup> Emissions in China's Less Developed Regions: A Case Study of Jiangxi**

**Yong Yang 1,2, Junsong Jia 1,2,\* and Chundi Chen <sup>3</sup>**


Received: 26 January 2020; Accepted: 28 February 2020; Published: 5 March 2020

**Abstract:** The residential sector is the second-largest consumer of energy in China. However, little attention has been paid to reducing the residential CO<sup>2</sup> emissions of China's less developed or undeveloped regions. Taking Jiangxi as a case study, this paper thus aims at fully analyzing the difference of the residential energy-related CO<sup>2</sup> emissions between urban and rural regions based on the Log-Mean Divisia Index (LMDI) and Tapio decoupling model. The main results are showed as follows: (1) Since 2008, residential energy-related CO<sup>2</sup> emissions have increased rapidly in both urban and rural Jiangxi. From 2000 to 2017, the residential energy-related CO<sup>2</sup> emissions per capita in rural regions rapidly increased and exceeded that in urban regions after 2015. Furthermore, the residential energy structures had become multiple in both urban and rural regions, but rural regions still had room to optimize its energy structure. (2) Over the study period, consumption expenditure per capita played the dominant role in increasing the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions, followed by energy demand and energy structure. Energy price had the most important effect on decreasing the urban and rural residential energy-related CO<sup>2</sup> emissions, followed by the carbon emission coefficient. However, urbanization increased the urban residential energy-related CO<sup>2</sup> emissions but decreased the CO<sup>2</sup> emissions in rural regions. Population made marginal and the most stable contribution to increase the residential energy-related CO<sup>2</sup> emissions both in urban and rural regions. (3) Overall, the decoupling status showed the weak decoupling (0.1) and expansive negative decoupling (1.21) in urban and rural regions, respectively.

**Keywords:** residential energy-related CO<sup>2</sup> emissions; less developed regions; urban and rural regions; LMDI; Tapio decoupling; Jiangxi province

### **1. Introduction**

The Intergovernmental Panel on Climate Change (IPCC) suggests that climate change is a great threat for human survival and development owing to the anthropogenic greenhouse gas (GHG) emissions [1,2]. Since 2006, China has become the largest CO<sup>2</sup> emitter globally [3]. Currently, its emissions account for approximate one third of the global total emissions [4,5]. In this context, China has made commitments and formulated a series of policies to reduce emissions. Meanwhile, there is a consensus that reducing the fast increase of residential CO<sup>2</sup> emissions is a significant pathway to achieve energy-saving and emission-reduction targets in China [6,7]. For China, the residential sector is the second-largest energy consumer category as well as primary CO<sup>2</sup> emissions source [8,9]. Specifically, the residential sector is responsible for 11.7% of energy consumption and 12.6% of CO<sup>2</sup> emissions in 2015 [10]. On one hand, after 40 years of "reform and opening up", China steps into the new development phase. In addition, with its further development of urbanization and industrialization, the residential CO<sup>2</sup> emissions are expected to increase continuously. Thus, it is necessary and significant to study the key driving forces to help raise reasonable and effective mitigation policies to combat climate change [9]. On the other hand, it is widely appreciated that investment, consumption and export are the troika for economic growth, and it is not wise to simply restrain consumption to mitigate the CO<sup>2</sup> emissions [11]. In addition, as reported by a recent study that China's economic prosperity had been coupled with environmental degradation, and suggested that decoupling economic growth from ecological impact had been central to achieve the national sustainable development [12]. Therefore, it is also necessary and imperative for China to discover a green, low-carbon and sustainable path to coordinate the link between developmental and environmental protection [13].

In reality, plenty of studies have been conducted to respond to the above concerns for decades in China. They often focused on exploring influencing factors of the CO<sup>2</sup> emissions or energy consumption in the residential sector, based on national or regional perspectives, or analyzing the relationship between economic growth and CO<sup>2</sup> emissions from residential sector. When exploring the drivers affecting residential CO<sup>2</sup> emissions or energy consumption, many factors such as income [14–16], population [9,11,17], urbanization [18,19], energy intensity (or energy price) [10,20,21], energy structure [6,22,23], floor space (per capita) [24,25], education level [26], etc., were discussed. For example, based on structural decomposition analysis (SDA), Zhu, et al. [11] studied the indirect residential CO<sup>2</sup> emissions during the period 1992–2005. They argued that population size was not the main reason for the emissions growth anymore, although it promoted the indirect emissions to a certain extent. Using Log-Mean Divisia Index (LMDI), Zhao, et al. [27] decomposed the residential energy consumption in urban China from 1998–2007 at a disaggregated product level. The results demonstrated that energy price had the positive effect on the emission reduction, while population and income played the key role in the growth of the CO<sup>2</sup> emissions. Simultaneously, more scholars were also awareness of the urban–rural gap or dual society in China, thereby they attempted to research Chinese residential CO<sup>2</sup> emissions or energy consumption based on the perspective. Based on Sato–Vartia index, Liu, et al. [17] analyzed the impact of China's increased urban and rural residential consumption on CO<sup>2</sup> emissions from 1992–2007. Furthermore, Fan, et al. [6] studied the residential CO<sup>2</sup> emissions evolutions in urban–rural divided China and explored the underlying driving forces from the perspective of end-use and behavior by applying Adaptive Weighting Divisia (AWD) decomposition. With the further study on the issue, some researchers extended their scopes and took into account Chinese regional disparity in their related studies. Taking Guangdong, the most developed province in China, as a case study, Wang, et al. [21] decomposed the influencing factors of direct residential energy-related CO<sup>2</sup> emissions into eight factors by using LMDI. The results showed that residential living standard had the largest contribution to the increase of CO<sup>2</sup> emissions, while energy price was the first inhibiting factor. Similarly, taking Liaoning, a coastal province in northeastern China, as an example, Tian, et al. [9] explored the driving forces of the residential CO<sup>2</sup> emissions from 2002–2007 based on LMDI. The results suggested that population and per-capita consumption were the main factors to the increase of residential CO<sup>2</sup> emissions, whereas carbon intensity had the negative effect on the residential CO<sup>2</sup> emissions growth. Moreover, based on spatial–regional level, Yuan, et al. [22] applied LMDI to analyze the drivers of urban and rural residential CO<sup>2</sup> emissions in China's 30 provinces between 2007 and 2012. They found that population and income-per-capita effects were the main drivers of high urban residential CO<sup>2</sup> emissions in most of the coastal provinces. There were so many studies conducted in many other regions of China. The explanations for these studies can be similarly illustrated, here; we condense some explanations and list some presentative references in Table 1 to save space.

In the analysis of the pathway to coordinate CO<sup>2</sup> emissions and economic growth, numerous studies were performed using the decoupling analysis. Generally, there are two main decoupling methods, i.e., OECD indicator and Tapio decoupling indicator [28,29]. Compared with OECD indicator, the latter was more accurate and flexible, which greatly improved OECD indicator. Therefore, Tapio decoupling indicator was more popular and widely used to analyze the relationship between CO<sup>2</sup>

emissions and economic growth. In addition, in China, most of these previous studies focused on the main emission sectors: industry [13,30]; transportation [31,32]; construction [33,34]; agriculture [35,36]. To the best of our knowledge, a few scholars applied the Tapio decoupling indicator to analyze the nexus between CO<sup>2</sup> emissions and economic growth from the residential sector. Based on the Tapio decoupling indicator, Ye, et al. [20] conducted the study to analyze the relationship between residential CO<sup>2</sup> emissions and economic growth in China from 1994–2012, which ignored the regional disparity and urban–rural gap. Using the OECD indicator, Yuan, et al. [22] studied the decoupling status in China's 30 provinces from the residential sector with little content.

Therefore, from Table 1, we can conclude that to date, researchers have mainly focused on developed regions especially coastal provinces and cities such as Guangdong, Liaoning, Jiangsu, Shandong and urban agglomerations, or on national and its 30 provinces when they conducted the studies on exploring the driving forces of CO<sup>2</sup> emissions in the residential sector. Moreover, few studies were conducted on decoupling analysis of the relationship between CO<sup>2</sup> emissions and economic growth from the residential sector compared with other main sectors. The existing related decoupling analysis studies on the residential sector were far from straightforward. As we all know, there is a consensus that China is a multiple-regional and heterogeneous country with the regional disparity and urban-gap [10,23], such as socio-economic, consumption pattern, urbanization, natural resource endowment, energy use, etc. Indeed, China's mitigation targets need all regions' efforts. However, this country's less developed and underdeveloped regions have not yet received academic attention, and few findings can be found in the existing studies worldwide. It is thus crucial to study these less developed and undeveloped regions. This can complete CO<sup>2</sup> emissions profile and help those regions' decision-makers make reasonable mitigation policies on residential sector by fully considering the local situations.


**Table 1.** Summary of representative studies on investigating the driving forces of Chinese residential sector.

Under such a circumstance, this paper aims to make up for the above-mentioned deficiencies by fully considering urban–rural gap and regional disparity; we take Jiangxi, a typical less developed province in central China, as a case study. Figure 1 shows its location. Jiangxi is in central China with an administrative area of 166,900 km<sup>2</sup> . Statistically, its total gross domestic product (GDP) reached 2001 billion RenMinBi (RMB) in 2017 [37], which ranked 16 among all 31 regions of China. To the best of our knowledge, its capital city, Nanchang, was assigned the task of pursuing a low-carbon economic transformation in China's first "Low-Carbon Pilot Cities" national project in 2010 [9,38]. More recently, Jiangxi, Fujian, and Guizhou are selected as the first national ecological civilization test beds in 2016, aiming to set up the relatively sound ecological mechanism for China. According to these facts, we can know that Jiangxi had the huge potential of increasing its CO<sup>2</sup> emissions on residential sector with its further development. However, little academic attention was paid to Jiangxi, especially

its residential sector. Thus, based on data availability, we first attempt to overview the situation of residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions. Then, the driving forces affecting the residential energy-related CO<sup>2</sup> emissions in urban and rural regions are explored by using LMDI. In addition, the Tapio decoupling model was applied to analyze the relationship between the residential energy-related CO<sup>2</sup> emissions and economic growth (consumption expenditure). We believe this study has some innovative significance to help this province make mitigation policies and provide important insight for China's other less developed or undeveloped regions to reduce emissions. relationship between the residential energy-related CO2 emissions and economic growth (consumption expenditure). We believe this study has some innovative significance to help this province make mitigation policies and provide important insight for China's other less developed or undeveloped regions to reduce emissions. The remainder of this paper is organized as follows: Section 2 depicts the methodology and data description. Results and discussions are given in Section 3. The main conclusions and corresponding policy implications are provided in Section 4.

explored by using LMDI. In addition, the Tapio decoupling model was applied to analyze the

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 4 of 27

economic transformation in China's first "Low-Carbon Pilot Cities" national project in 2010 [9,38]. More recently, Jiangxi, Fujian, and Guizhou are selected as the first national ecological civilization test beds in 2016, aiming to set up the relatively sound ecological mechanism for China. According to these facts, we can know that Jiangxi had the huge potential of increasing its CO2 emissions on residential sector with its further development. However, little academic attention was paid to Jiangxi, especially its residential sector. Thus, based on data availability, we first attempt to overview

**Figure 1.** Location of Jiangxi. **Figure 1.** Location of Jiangxi.

**2. Methods and Data Description**  *2.1. Estimation of the Residential Energy-Related CO2 Emissions*  The remainder of this paper is organized as follows: Section 2 depicts the methodology and data description. Results and discussions are given in Section 3. The main conclusions and corresponding policy implications are provided in Section 4.

consumption, e.g., lighting, cooking, heating, using household appliances and private transport, in

#### The residential energy-related CO2 emissions was caused by the residential energy **2. Methods and Data Description**

#### addition, including electricity and heat [9]. Along with these activities, quantities of fossil fuels (e.g., *2.1. Estimation of the Residential Energy-Related CO<sup>2</sup> Emissions*

coal, oil, natural gas) and secondary energy sources (e.g., electricity, heat) were consumed, which resulted in CO2 emissions directly. According to IPCC [39], the residential energy-related CO2 emissions can be calculated by the following equation: = =⋅ *ij ij j ij ij C C Ef* (1) where *i* represents the resident type, i.e., urban residents and rural residents. *j* represents the fuel The residential energy-related CO<sup>2</sup> emissions was caused by the residential energy consumption, e.g., lighting, cooking, heating, using household appliances and private transport, in addition, including electricity and heat [9]. Along with these activities, quantities of fossil fuels (e.g., coal, oil, natural gas) and secondary energy sources (e.g., electricity, heat) were consumed, which resulted in CO<sup>2</sup> emissions directly. According to IPCC [39], the residential energy-related CO<sup>2</sup> emissions can be calculated by the following equation:

$$\mathcal{C} = \sum\_{i} \sum\_{j} \mathcal{C}\_{ij} = \sum\_{i} \sum\_{j} E\_{ij} \cdot f\_{j} \tag{1}$$

electricity, and heat. Specifically, coal includes raw coal, briquettes, coke oven gas, and other gases. where *i* represents the resident type, i.e., urban residents and rural residents. *j* represents the fuel type. It should be noted that all kinds of fuel types consumed by residents are subdivided into five categories in this study based on China Energy Statistical Yearbook [40], i.e., coal, oil, natural gas, electricity, and heat. Specifically, coal includes raw coal, briquettes, coke oven gas, and other gases. Oil includes gasoline, kerosene, diesel oil, lubricants, liquefied petroleum gas (LPG) and other petroleum products. Natural gas includes natural gas and liquefied natural gas (LNG). *Eij* denotes energy consumption of the *i* type resident, *f<sup>j</sup>* denotes the carbon emission coefficient of *j* fuel type. With reference to Kennedy, et al. [41], carbon emission coefficients of different energy types are listed in Table A2. In addition, it is

difficult to acquire the carbon emission coefficients of electricity and heat in Jiangxi. In the light of the data of heat can be only found from 2000–2002, which has small influence on the whole results. For this reason, we use China's corresponding carbon emission coefficients to substitute for Jiangxi's carbon emission coefficients of heat. With references to Jia, et al. [38], the carbon emission coefficient of heat can be measured as 0.11 t-CO2/GJ by the equivalent calorific value. As for the carbon emission coefficient of electricity, here, we calculate and list the coefficient in Table A1.

#### *2.2. Decomposition Method*

Decomposition analysis has been widely applied to explore the impact factors affecting the changes of CO<sup>2</sup> emissions in the economic and environmental field. To date, there are two mainstream factor decomposition methods [25,36], viz., SDA and index decomposition analysis (IDA). Generally, SDA depends on input–output (IO) table and produces more accurate decomposition results. However, the IO tables are not available every year so that SDA has the limitation to the annual analysis. Inversely, IDA is easier to acquire data and more alterative to use aggregated data to analyze any years' changes [10]. Hence, IDA is widely applied to decompose the driving factors of energy-related CO<sup>2</sup> emissions. In addition, IDA provides many different indexes to choose [38]. In particular, among these indexes, LMDI has the incomparable advantages for its ease of use and no unexplainable residuals [42]. Moreover, it also provides eight effective strategies to handle zero values problem [43]. In view of these advantages, LMDI has been regarded as the most perfect decomposition method and widely used by many researchers [44,45]. Thus, in this paper, we apply the LMDI to explore the influencing factors affecting the residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions.

The Kaya identity was widely used to reveal the influencing factors of CO<sup>2</sup> emissions. With the further development of the studies, more influencing factors are discussed. According to Kaya [46], the extended determinants of the residential energy-related CO<sup>2</sup> emissions can be expressed as:

$$\mathbf{C} = \sum\_{i} \sum\_{j} \mathbf{C}\_{ij} = \sum\_{i} \sum\_{j} \frac{\mathbf{C}\_{ij}}{\mathbf{E}\_{ij}} \cdot \frac{\mathbf{E}\_{ij}}{\mathbf{E}\_{i}} \cdot \frac{\mathbf{E}\_{i}}{\mathbf{Y}\_{i\varepsilon}} \cdot \frac{\mathbf{Y}\_{i\varepsilon}}{\mathbf{Y}\_{i}} \cdot \frac{\mathbf{Y}\_{i\varepsilon}}{\mathbf{P}\_{i}} \cdot \frac{\mathbf{Y}\_{i}}{\mathbf{P}\_{i}} \cdot \frac{\mathbf{P}\_{i}}{\mathbf{P}} \cdot \mathbf{P} \tag{2}$$

Let *Kij* = *Cij Eij* , *ESij* = *Eij Ei* , *EP<sup>i</sup>* = *Ei Yie* , *ED<sup>i</sup>* = *Yie Yi* , *CP<sup>i</sup>* = *Yi Pi* , *U<sup>i</sup>* = *Pi P* , *P* = *P* in Equation (2) can be written as:

$$\mathcal{C} = \sum\_{i} \sum\_{j} \mathbf{K}\_{ij} \cdot \mathbf{ES}\_{ij} \cdot \mathbf{EP}\_i \cdot \mathbf{ED}\_i \cdot \mathbf{CP}\_i \cdot \mathbf{U}\_i \cdot \mathbf{P} \tag{3}$$

where the variables are defined in Table 2.

**Table 2.** Definition of variables.


Thereafter, to explore the contribution of the influencing factors to the total CO<sup>2</sup> emissions, next, we use LMDI, and it has two formations, i.e., additive LMDI (LMDI-I) and multiplicative LMDI (LMDI-II). Based on the additive LMDI, the changes of the total CO<sup>2</sup> emissions (*CTOT*) from the baseline period (*C*0) to the target period (*CT*) can be decomposed as the following seven effects:

$$
\Delta \mathcal{C}\_{TOT} = \mathcal{C}\_{T} - \mathcal{C}\_{0} = \Delta \mathcal{C}\_{K\_{\overline{i}}} + \Delta \mathcal{C}\_{ES\_{\overline{i}\overline{j}}} + \Delta \mathcal{C}\_{EP\_{\overline{i}}} + \Delta \mathcal{C}\_{ED\_{\overline{i}}} + \Delta \mathcal{C}\_{CP\_{\overline{i}}} + \Delta \mathcal{C}\_{L\overline{l}\_{\overline{i}}} + \Delta \mathcal{C}\_{P} \tag{4}$$

Each effect can be calculated as follows:

∆*CKij* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln*Kij*,*<sup>T</sup>* − ln*Kij*,0) ∆*CESij* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln *ESij*,*<sup>T</sup>* − ln *ESij*,0) ∆*CEP<sup>i</sup>* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln *EPi*,*<sup>T</sup>* − ln *EPi*,0) ∆*CED<sup>i</sup>* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln *EDi*,*<sup>T</sup>* − ln *EDi*,0) ∆*CCP<sup>i</sup>* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln*CPi*,*<sup>T</sup>* − ln*CPi*,0) ∆*CU<sup>i</sup>* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln *Ui*,*<sup>T</sup>* − ln *Ui*,0) ∆*C<sup>P</sup>* = P *i* P *j* (*Cij*,*<sup>T</sup>* − *Cij*,0)/(ln*Cij*,*<sup>T</sup>* − ln*Cij*,0) · (ln *P<sup>T</sup>* − ln *P*0) (5) *<sup>L</sup>*(*a*, *<sup>b</sup>*) = ( (*a* − *b*)/(ln *a* − ln *b*), *a* , *b*(*a* > 0, *b* > 0) 0, *a* = *b*(*a* > 0, *b* > 0) (6)

In addition, to present the decomposition results clearer, here, we also use the multiplicative LMDI. According to Shao, et al. [47], the corresponding multiplicative LMDI can be written as:

$$\mathbf{\dot{Y}C\_{TOT}} = \frac{\mathbf{C\_T}}{\mathbf{C\_0}} = \mathbf{\dot{Y}C\_{K\_{ij}}} \cdot \mathbf{\dot{Y}C\_{ES\_{ij}}} \cdot \mathbf{\dot{Y}C\_{EP\_i}} \cdot \mathbf{\dot{Y}C\_{ED\_i}} \cdot \mathbf{\dot{Y}C\_{CP\_i}} \cdot \mathbf{\dot{y}C\_{U\_i}} \cdot \mathbf{\dot{y}C\_P} \tag{7}$$

where <sup>Ψ</sup>*C<sup>w</sup>* <sup>=</sup> exp P *i* ln ω*<sup>T</sup>* ln ω<sup>0</sup> · (*Cij*,*T*−*Cij*,0)/(ln*Cij*,*T*−ln*Cij*,0) (*CT*−*C*0)/(ln*CT*−ln*C*0) ! , and ω means *Kij*, *ESij*, *EP<sup>i</sup>* , *ED<sup>i</sup>* , *CP<sup>i</sup>* , *U<sup>i</sup>* and *P*. Ψ*TOT* means the changes of the total CO<sup>2</sup> emissions between the baseline period and the target period with the corresponding multiplicative formation of the LMDI.

To the best of our knowledge, the CO<sup>2</sup> emission coefficient is usually assumed as a constant value, which has no contributions to the changes of CO<sup>2</sup> emissions. However, in this paper, the CO<sup>2</sup> emission coefficient of electricity changes every year, and it will lead to the changes of CO<sup>2</sup> emissions. Therefore, we finally analyze the following seven effects: the carbon emission coefficient effect (∆*CKij* and Ψ*CKij*), energy structure effect (∆*CESij* and Ψ*CESij*), energy price effect (∆*CEP<sup>i</sup>* and Ψ*CEP<sup>i</sup>* ), energy demand effect (∆C*ED<sup>i</sup>* and Ψ*CED<sup>i</sup>* ), consumption expenditure per-capita effect (∆*CCP<sup>i</sup>* and Ψ*CCP<sup>i</sup>* ), urbanization effect (∆*CU<sup>i</sup>* and Ψ*CU<sup>i</sup>* ) and population effect (∆*C<sup>P</sup>* and Ψ*CP*).

#### *2.3. Decoupling Model*

The decoupling theory initially originated from the field of physics, which denotes the de-linkage relationship between two or more variables. Organization for Co-operation and Development (OECD) first categorized the decoupling indicators into relative decoupling and absolute decoupling [28]. However, there are some limitations and shortcomings of the OECD decoupling theory. Specifically, it is liable to be affected by the decoupling elasticity and lacked obvious criteria for choosing appropriate factors [36]. In view of this, Tapio [29] subdivided decoupling indicators into eight sub-categories to analyze the relationship between the road traffic and CO<sup>2</sup> emissions in the EU 15 countries from 1970–2001, which greatly improved the framework of OECD decoupling indicators. Since then, the Tapio decoupling model (see Table 3) was widely used to explore the dynamics nexus between economic

growth and environmental protection. However, based on the context above, the decoupling analysis of the CO<sup>2</sup> emissions in the residential sector was far from straightforward in China, especially in less developed regions. Therefore, in this paper, the Tapio decoupling model was selected to analyze the decoupling status and the underlying drivers of the residential consumption expenditure from the residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions.

According to Tapio [29], the decoupling indicator of residential energy-related CO<sup>2</sup> emissions of the *i* type resident from residential consumption expenditure can be expressed as:

$$D\_i = \frac{\mathbb{C}\_i \%}{Y\_i \%} = \frac{\Delta \mathbb{C}\_i / \mathbb{C}\_i^0}{\Delta Y\_i / Y\_i^0} \tag{8}$$


**Table 3.** Framework of the Tapio decoupling indicators.

#### *2.4. Data Description*

Based on data availability, in this paper, the study period ranges from 2000 to 2017. To the best of our knowledge, since 1953, with a gap from 1963 to 1965, Chinese government has formulated plans for national economic and social development every five years, namely "Five-Year Plan (FYP)". To make the related results and analysis clearer, here, the study period is consistent with the FYP and subdivided into four stages, i.e., 2000–2005 (10th FYP), 2005–2010 (11th FYP), 2010–2015 (12th FYP) and 2015–2017 (13th FYP). In addition, it is noteworthy that Chinese government proposed a new stage, called "the new normal", of China's economic development during the 12th FYP, which was coincided with the last two stages.

All residential energy data are collected from Energy Balance Sheet of Jiangxi Province in the China Energy Statistical Yearbook (2001–2018) [40]. The related consumption expenditure per capita and population data directly derive from the Jiangxi Statistical Yearbook (JSY) (2001–2018) [37]. It should be noticed that consumption expenditure is subdivided into eight categories (e.g., food, cigarettes and wine, clothing, residence, household appliances, and services) in the JSY [37]. Additionally, among these categories, the data of residence expenditure (including the expenditure of housing, electricity, water, fuel and others) is the main source of the residential energy-related CO<sup>2</sup> emissions [21]. Thus, we split the residence expenditure from the consumption expenditure to make the analyses clearer. The share of residence expenditure to consumption expenditure is defined the residential energy demand in this study.

#### **3. Results and Discussion**

### *3.1. Overview the Situation of Residential Energy-Related CO<sup>2</sup> Emissions in Urban and Rural Jiangxi*

3.1.1. The Trends of Residential Energy-Related CO<sup>2</sup> Emissions

As shown in Figure 2, the urban residential energy-related CO<sup>2</sup> emissions grew from 540.38 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2000 to 1112.84 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2017, with an annual average increase amount of 33.67 <sup>×</sup> <sup>10</sup><sup>4</sup> t and a growth rate of 4.34%. The trend of urban residential energy-related CO<sup>2</sup> emissions could be subdivided into three phases: a fluctuating phase (2000–2005), a slow growth phase (2005–2008) and a rapid increase phase (2008–2017). From 2000–2005, it could be easily seen that the urban residential energy-related CO<sup>2</sup> emissions was from 540.38 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 548.28 <sup>×</sup> <sup>10</sup><sup>4</sup> t, with an annual growth rate of only 0.29%. This phenomenon might be mainly attributed to the large-scale adjustment of the energy structure and cut down the share of coal use. From 2005–2008, because the global economic crisis took place around 2008, people restricted and reduced all kinds of expenditure in daily life, but the consumption of electricity could not fall for people's habits and customs [38]. Specifically, the urban residential energy-related CO<sup>2</sup> emissions were not more than 600 <sup>×</sup> <sup>10</sup><sup>4</sup> t. After 2008, the urban residential energy-related CO<sup>2</sup> emissions grew to 615.17 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2009, then it steadily and quickly increased to 1112.84 <sup>×</sup> <sup>10</sup><sup>4</sup> t, with an annual growth rate of 7.47%. The rapid growth of the residential energy-related CO<sup>2</sup> emissions might be explained by the recovery from the crisis and the improvement of economy and living standard for urban residents. This situation was mainly caused by the growing demand of high-carbon appliances and private transport (Table 4), consuming large quantities of oil and electricity [23]. For example, the number of family cars, air conditioners and computers in 2017 reached 40-fold, 1.9-fold and 2.3-fold than that in 2005, respectively. growth rate of 4.34%. The trend of urban residential energy-related CO2 emissions could be subdivided into three phases: a fluctuating phase (2000–2005), a slow growth phase (2005–2008) and a rapid increase phase (2008–2017). From 2000–2005, it could be easily seen that the urban residential energy-related CO2 emissions was from 540.38 × 104 t to 548.28×104 t, with an annual growth rate of only 0.29%. This phenomenon might be mainly attributed to the large-scale adjustment of the energy structure and cut down the share of coal use. From 2005–2008, because the global economic crisis took place around 2008, people restricted and reduced all kinds of expenditure in daily life, but the consumption of electricity could not fall for people's habits and customs [38]. Specifically, the urban residential energy-related CO2 emissions were not more than 600 × 104 t. After 2008, the urban residential energy-related CO2 emissions grew to 615.17 × 104 t in 2009, then it steadily and quickly increased to 1112.84 × 104 t, with an annual growth rate of 7.47%. The rapid growth of the residential energy-related CO2 emissions might be explained by the recovery from the crisis and the improvement of economy and living standard for urban residents. This situation was mainly caused by the growing demand of high-carbon appliances and private transport (Table 4), consuming large quantities of oil and electricity [23]. For example, the number of family cars, air conditioners and computers in 2017 reached 40-fold, 1.9-fold and 2.3-fold than that in 2005, respectively.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 8 of 27

104 t in 2000 to 1112.84 × 104 t in 2017, with an annual average increase amount of 33.67×104 t and a

*3.1. Overview the Situation of Residential Energy-Related CO2 Emissions in Urban and Rural Jiangxi* 

3.1.1. The Trends of Residential Energy-Related CO2 Emissions

**Figure 2.** Trends of residential energy-related CO2 emissions in Jiangxi urban and rural regions. **Figure 2.** Trends of residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions.

By contrast, the rural residential energy-related CO2 emissions grew from 210.13 × 104 t to 954.82×104 t between 2000 and 2017 (Figure 2), with an annual growth amount of 43.81 × 104 t and a growth rate of 9.31%. However, the growth rate of rural residential energy-related CO2 emissions was quicker than that of urban Jiangxi. Here, we also subdivided the rural residential energy-related CO2 emissions into three phases: a rapid growth phase (2000–2005), a slight decrease phase (2005– 2008) and a steadily growth phase (2008–2017). From 2000–2005, the rural residential energy-related CO2 emissions grew from 210.13 × 104 t to 570.94×104 t, with an annual growth rate of 22.13%. This rapid growth may be explained by the increase of coal and electricity use, it was worth noting that coal still played the dominant role in rural residents' energy use (Figure 3b), leading to 154.64 × 104 t in 2000 and 331.08×104 t CO2 in 2005, respectively. Analogous to the urban residents, the rural residents were also affected by the economic crisis from 2005–2008, making the corresponding CO2 emissions declined by −1.27%. From 2008–2017, the rural residential energy-related CO2 emissions increased from 563.69 × 104 t to 954.82 × 104 t, with an annual growth rate of 6.03%. By contrast, the rural residential energy-related CO<sup>2</sup> emissions grew from 210.13 <sup>×</sup> <sup>10</sup><sup>4</sup> t to954.82 <sup>×</sup> <sup>10</sup><sup>4</sup> <sup>t</sup> between 2000 and 2017 (Figure 2), with an annual growth amount of 43.81 <sup>×</sup> <sup>10</sup><sup>4</sup> t and agrowth rate of 9.31%. However, the growth rate of rural residential energy-related CO<sup>2</sup> emissions wasquicker than that of urban Jiangxi. Here, we also subdivided the rural residential energy-related CO<sup>2</sup> emissions into three phases: a rapid growth phase (2000–2005), a slight decrease phase (2005–2008) and a steadily growth phase (2008–2017). From 2000–2005, the rural residential energy-related CO<sup>2</sup> emissions grew from 210.13 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 570.94 <sup>×</sup> <sup>10</sup><sup>4</sup> t, with an annual growth rate of 22.13%. This rapidgrowth may be explained by the increase of coal and electricity use, it was worth noting that coal stillplayed the dominant role in rural residents' energy use (Figure 3b), leading to 154.64 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2000 and 331.08 <sup>×</sup> <sup>10</sup><sup>4</sup> t CO<sup>2</sup> in 2005, respectively. Analogous to the urban residents, the rural residentswere also affected by the economic crisis from 2005–2008, making the corresponding CO<sup>2</sup> emissions declined by <sup>−</sup>1.27%. From 2008–2017, the rural residential energy-related CO<sup>2</sup> emissions increasedfrom 563.69 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 954.82 <sup>×</sup> <sup>10</sup><sup>4</sup> t, with an annual growth rate of 6.03%.

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 9 of 27

**Figure 3.** Changes of residential energy-related CO2 emissions structure in Jiangxi urban and rural regions from 2000–2017. (**a**) Urban. (**b**) Rural. **Figure 3.** Changes of residential energy-related CO<sup>2</sup> emissions structure in Jiangxi urban and rural regions from 2000–2017. (**a**) Urban. (**b**) Rural.


**Table 4.** Urban: Ownership of major durable consumer goods per 100 urban households at year-**Table 4.** Urban: Ownership of major durable consumer goods per 100 urban households at year-end.

Shower heater 58.02 81.77 92.28 91.99 95.9 Figure 2 also showed the ratio of the residential energy-related CO2 emissions for urban and rural regions widened and narrowed, felling from 2.57 in 2000 to 1.17 in 2017. Over the study period, the urban–rural ratio of residential energy-related CO2 emission most remained the interval 1.0–1.2 except 2000–2005. Specifically, the ratio was more than 1.3 between 2000–2003. To the best of our knowledge, urban residents owned more energy equipment and consumed more commercial energy than that of rural residents, leading to more residential energy-related CO2 emissions [17,48]. The ratio was 0.84 and 0.96 between 2003 and 2004, respectively. The reason might be that the large-scale reduction of coal use in urban regions as well as the growing popularity of appliances in rural regions Figure 2 also showed the ratio of the residential energy-related CO<sup>2</sup> emissions for urban and rural regions widened and narrowed, felling from 2.57 in 2000 to 1.17 in 2017. Over the study period, the urban–rural ratio of residential energy-related CO<sup>2</sup> emission most remained the interval 1.0–1.2 except 2000–2005. Specifically, the ratio was more than 1.3 between 2000–2003. To the best of our knowledge, urban residents owned more energy equipment and consumed more commercial energy than that of rural residents, leading to more residential energy-related CO<sup>2</sup> emissions [17,48]. The ratio was 0.84and 0.96 between 2003 and 2004, respectively. The reason might be that the large-scale reduction of coal use in urban regions as well as the growing popularity of appliances in rural regions (Table 5). From2006–2017, the ratio of urban–rural always over 1.0, which mainly resulted from the development of urbanization.

Mobile telephone 14.37 136.26 181.18 226.16 235.73

(Table 5). From 2006–2017, the ratio of urban–rural always over 1.0, which mainly resulted from the development of urbanization. **Table 5.** Rural: Ownership of major durable consumer goods per 100 rural households at year-end.


Air conditioner 0.04 2 10.24 42.41 56.48

#### 3.1.2. The Changes of Residential Energy-Related CO<sup>2</sup> Emissions Per Capita 3.1.2. The Changes of Residential Energy-Related CO2 Emissions Per Capita

The changes of residential energy-related CO<sup>2</sup> emissions per capita in Jiangxi urban and rural regions were presented in Figure 4. The urban residential energy-related CO<sup>2</sup> emissions per capita decreased from 470.41 kg-CO<sup>2</sup> in 2000 to 440.96 kg-CO<sup>2</sup> in 2017, with an annual average decrease amount of 1.73 kg-CO<sup>2</sup> and a decrease rate of 0.38%. The trends of urban residential energy-related CO<sup>2</sup> emissions per capita could be also subdivided into three stages: a fluctuating decrease stage (2000–2005), a slow growth stage (2005–2010) and a slightly more rapid growth stage (2010–2017). The urban residential energy-related CO<sup>2</sup> emissions per capita experienced a waved decrease from 470.41 kg-CO<sup>2</sup> in 2000 to 342.79 kg-CO<sup>2</sup> in 2005, then it slowly increased to 361.35 kg-CO<sup>2</sup> in 2010. The decrease of the urban residential energy-related CO<sup>2</sup> emissions per capita could be explained by the following reasons. The proportion of high-quality energy types, such as oil, natural gas, and electricity greatly improved due to the shift of energy structure and the improvement of appliances' energy efficiency [24]. Additionally, urban residents reduced some unnecessary energy consumption when the financial crisis occurred. The urban residential energy-related CO<sup>2</sup> emissions per capita grew from 361.35 kg-CO<sup>2</sup> to 440.96 kg-CO<sup>2</sup> from 2010–2017, this might be explained by the increase of appliances and private transport (Table 4). The changes of residential energy-related CO2 emissions per capita in Jiangxi urban and rural regions were presented in Figure 4. The urban residential energy-related CO2 emissions per capita decreased from 470.41 kg-CO2 in 2000 to 440.96 kg-CO2 in 2017, with an annual average decrease amount of 1.73 kg-CO2 and a decrease rate of 0.38%. The trends of urban residential energy-related CO2 emissions per capita could be also subdivided into three stages: a fluctuating decrease stage (2000–2005), a slow growth stage (2005–2010) and a slightly more rapid growth stage (2010–2017). The urban residential energy-related CO2 emissions per capita experienced a waved decrease from 470.41 kg-CO2 in 2000 to 342.79 kg-CO2 in 2005, then it slowly increased to 361.35 kg-CO2 in 2010. The decrease of the urban residential energy-related CO2 emissions per capita could be explained by the following reasons. The proportion of high-quality energy types, such as oil, natural gas, and electricity greatly improved due to the shift of energy structure and the improvement of appliances' energy efficiency [24]. Additionally, urban residents reduced some unnecessary energy consumption when the financial crisis occurred. The urban residential energy-related CO2 emissions per capita grew from 361.35 kg-CO2 to 440.96 kg-CO2 from 2010–2017, this might be explained by the increase of appliances and private transport (Table 4).

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 10 of 27

**Figure 4.** Changes of residential energy-related CO2 emissions per capita in Jiangxi urban and rural regions from 2000–2017. **Figure 4.** Changes of residential energy-related CO<sup>2</sup> emissions per capita in Jiangxi urban and rural regions from 2000–2017.

By contrast, the rural residential energy-related CO2 emissions per capita rapidly increased from 70.05 kg-CO2 to 455.02 kg-CO2 from 2000–2017 (Figure 4), with an annual average increase amount of 22.65 kg-CO2 and a growth rate of 11.64%. In particular, the rural energy-related CO2 emissions per capita were 401.41 kg-CO2 and 455.02 kg-CO2 from 2016–2017 which exceeded that of urban regions. Compare to urban regions, the rural residential energy-related CO2 emissions presented a tedious upward trend. This situation could be explained by the traditional biomass was gradually outpaced by the commercial energy. Moreover, with the improvement of living standards, rural residents had the capacity to pursuit high-quality life, such as buying more energy-intensive appliances and private transport tools (Table 5). However, many of them lacked the consciousness of environmental protection, resulting in more residential CO2 emissions [49]. By contrast, the rural residential energy-related CO<sup>2</sup> emissions per capita rapidly increased from 70.05 kg-CO<sup>2</sup> to 455.02 kg-CO<sup>2</sup> from 2000–2017 (Figure 4), with an annual average increase amount of 22.65 kg-CO<sup>2</sup> and a growth rate of 11.64%. In particular, the rural energy-related CO<sup>2</sup> emissions per capita were 401.41 kg-CO<sup>2</sup> and 455.02 kg-CO<sup>2</sup> from 2016–2017 which exceeded that of urban regions. Compare to urban regions, the rural residential energy-related CO<sup>2</sup> emissions presented a tedious upward trend. This situation could be explained by the traditional biomass was gradually outpaced by the commercial energy. Moreover, with the improvement of living standards, rural residents had the capacity to pursuit high-quality life, such as buying more energy-intensive appliances and private transport tools (Table 5). However, many of them lacked the consciousness of environmental protection, resulting in more residential CO<sup>2</sup> emissions [49].

As presented in Figure 4, the gap in residential energy-related CO2 emissions per capita between Jiangxi urban and rural regions decreased from 6.72 to 0.97 from 2000–2017. In addition, it could be divided into two phases: a rapid decrease phase in 2000–2005 and a slow decrease phase between 2005–2017. The approximate reason might be owing to the development of urbanization and the rural economy [23]. As presented in Figure 4, the gap in residential energy-related CO<sup>2</sup> emissions per capita between Jiangxi urban and rural regions decreased from 6.72 to 0.97 from 2000–2017. In addition, it could be divided into two phases: a rapid decrease phase in 2000–2005 and a slow decrease phase between 2005–2017. The approximate reason might be owing to the development of urbanization and the rural economy [23].

## 3.1.3. Analysis on the Residential Energy-Related CO<sup>2</sup> Emissions Structure

As shown in Figure 3a, the urban residential energy-related CO<sup>2</sup> emissions caused by coal waned from 324.57 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2000 to 65.64 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2007 and its corresponding share decreased from 60.06% to 5.90%, thereby indicating the reduction of coal use in urban regions. In addition, the residential energy-related CO<sup>2</sup> emissions caused by heat also presented a decrease tendency (46.86 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2000 and 2.01 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2017). The reason was that Jiangxi reduced the supply of heat, eventually, cut it off after 2002. See other energy types, electricity, oil, and natural gas overall presented an upward trend. The residential energy-related CO<sup>2</sup> emissions caused by electricity increased from 101.81 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 605.55 <sup>×</sup> <sup>10</sup><sup>4</sup> t from 2000–2017, the corresponding share increased from 18.84% to 54.42%. It should be noteworthy that the residential energy-related CO<sup>2</sup> emissions induced by electricity exceeded coal and ranked first, which became the most important energy for urban residents after 2004. Since 2007, the residential energy-related CO<sup>2</sup> emissions induced by oil also exceeded the coal and became the second-largest energy type, and its corresponding share grew from 12.42% in 2001 to 29.24% in 2017. It is likely that the private transport increased rapidly (Table 4). However, starting from 2005, urban residents began to use natural gas. The share of residential energy-related CO<sup>2</sup> emissions induced by natural gas increased quickly from 1.09 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2005 to 116.29 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2007, with an annual growth rate of 47.57%, the corresponding share was from 0.2% to 10.45%. Since 2012, the share of residential energy-related CO<sup>2</sup> emissions induced by natural gas exceeded coal. At present, the residential energy-related CO<sup>2</sup> emissions structure was dominated by electricity, followed by oil, natural gas, and coal in urban regions.

The residential energy-related CO<sup>2</sup> emissions caused by heat was not found in rural regions according to JSY [37]. Here, we only considered coal, oil, natural gas, and electricity. As shown in Figure 3b, the residential energy-related CO<sup>2</sup> emissions for rural residents mainly induced by coal, oil, and electricity, which accounted for over 99% over the study period. In 2000, the residential energy-related CO<sup>2</sup> emissions induced by coal ranked the first, followed by electricity and oil. The share of residential energy-related CO<sup>2</sup> emissions induced by coal decreased from 73.59% to 27.52% from 2000–2007. The residential energy-related CO<sup>2</sup> emissions induced by electricity rapidly increased from 49.02 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 513.47 <sup>×</sup> <sup>10</sup><sup>4</sup> t over the study period, with an annual growth rate of 14.82%. In 2009, the residential energy-related CO<sup>2</sup> emissions induced by electricity exceeded the coal, which might be explained by the growing increase of appliances. From 2009–2017, electricity became the most important energy type of the residential energy-related CO<sup>2</sup> emission for rural residents. For oil, the residential energy-related CO<sup>2</sup> emissions induced by oil occupied from 3.08% in 2000 to 18.25% in 2017. In addition, its share was still lower than the coal, although the share of coal decreased. Therefore, it could be deduced that the energy structure had further space to optimize for rural regions. The residential energy-related CO<sup>2</sup> emissions induced by natural gas was minimal among other energy types, which increased from 0.66 <sup>×</sup> <sup>10</sup><sup>4</sup> t to 4.37 <sup>×</sup> <sup>10</sup><sup>4</sup> t from 2008–2017. It should be noted that Jiangxi belonged to the "second pipeline of West–East Natural Gas transmission Project (To provide clean energy and shift energy structure in central and south China, the second pipeline of "West–East Natural Gas transmission Project" was launched in February 2008. Specifically, the pipeline went through Xinjiang, Gansu, Ningxia, Shaanxi, Henan, Hubei, Hunan, Jiangxi, Anhui, Jiangsu, Shanghai, Zhejiang, Guangdong, Guangxi, and H.K.)", thereby leading to the use of natural gas after 2008. With the implementation of the "coal to gas" project, the demand of natural gas of rural regions in Jiangxi might increase rapidly and expand the of share of residential energy-related CO<sup>2</sup> emissions induced by natural gas in the foreseeable future. Correspondingly, the share of the residential energy-related CO<sup>2</sup> emissions induced by coal would fall further. To the best of our knowledge, coal was a type of low-quality energy; the direct combustion of coal indoors was harmful to human health for rural residents, especially for women and children [17], which in turn led to air pollution even aggregated climate change. From the perspective of socio-development, if the energy structure were outdated, it would be bad for the development of rural economy and education due to the vulnerability of Chinese

rural regions. To date, the share of residential energy-related CO<sup>2</sup> emissions structure was electricity, coal, oil, and natural gas in sequence for rural regions. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 12 of 27

#### *3.2. Decomposition Analysis of Residential Energy-Related CO<sup>2</sup> Emissions at Four Stages* As shown that the urban residential energy-related CO2 emissions increased by only 7.91×104 t

As shown that the urban residential energy-related CO<sup>2</sup> emissions increased by only 7.91 <sup>×</sup> <sup>10</sup><sup>4</sup> t with a growth rate of 1.46% during the first stage (Figures 5a and 6a), and then rapidly grew by 162.15 <sup>×</sup> <sup>10</sup><sup>4</sup> t with a growth rate of 29.58% in the second stage (2005-2010) (Figures 5a and 6b). During the third stage, the CO<sup>2</sup> emissions increased by 215.42 <sup>×</sup> <sup>10</sup><sup>4</sup> t at a rate of 30.32% (Figures 5a and 6c) and grew by 186.98 <sup>×</sup> <sup>10</sup><sup>4</sup> t at a rate of 20.19% during the fourth stage (Figures 5a and 6d). Therefore, the conclusion could be drawn that the urban residential energy-related CO<sup>2</sup> emissions presented a sequentially upward trend with a total growth rate of 105.94% over the study period (Figure 5a). with a growth rate of 1.46% during the first stage (Figure 5a and Figure 6a), and then rapidly grew by 162.15×104 t with a growth rate of 29.58% in the second stage (2005-2010) (Figure 5a and Figure 6b). During the third stage, the CO2 emissions increased by 215.42×104 t at a rate of 30.32% (Figure 5a and Figure 6c) and grew by 186.98×104 t at a rate of 20.19% during the fourth stage (Figure 5a and Figure 6d). Therefore, the conclusion could be drawn that the urban residential energy-related CO2 emissions presented a sequentially upward trend with a total growth rate of 105.94% over the study period (Figure 5a).

**Figure 5.** Factors affecting the residential energy-related CO2 emissions in Jiangxi urban and rural regions from 2000–2017. (**a**) Urban. (**b**) Rural. **Figure 5.** Factors affecting the residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions from 2000–2017. (**a**) Urban. (**b**) Rural.

With respect to rural regions, it could be easily observed that the rural residential energy-related CO2 emissions grew by 360.81×104 t at a rate of 171.71% in the first stage (Figure 5b and Figure 6a), and then increased by 69.11×104 t at a rate of 12.11% in the second stage (Figure 5b and Figure 6b). During the third and fourth stage, the rural residential CO2 emissions increased by 181.59×104 t with a growth rate of 28.37% (Figure 5b and Figure 6c) and grew by 133.18×104 t with a growth rate of 16.21% (Figure 5b and Figure 6d), respectively. Analogous to the urban regions, it could be easily noticed that the rural residential energy-related CO2 emissions also exhibited a sequentially increasing trend at a rate of 354.39% (Figure 5b). With respect to rural regions, it could be easily observed that the rural residential energy-related CO<sup>2</sup> emissions grew by 360.81 <sup>×</sup> <sup>10</sup><sup>4</sup> t at a rate of 171.71% in the first stage (Figures 5b and 6a), and then increased by 69.11 <sup>×</sup> <sup>10</sup><sup>4</sup> t at a rate of 12.11% in the second stage (Figures 5b and 6b). During the third and fourth stage, the rural residential CO<sup>2</sup> emissions increased by 181.59 <sup>×</sup> <sup>10</sup><sup>4</sup> t with a growth rate of 28.37% (Figures 5b and 6c) and grew by 133.18 <sup>×</sup> <sup>10</sup><sup>4</sup> t with a growth rate of 16.21% (Figures 5b and 6d), respectively. Analogous to the urban regions, it could be easily noticed that the rural residential energy-related CO<sup>2</sup> emissions also exhibited a sequentially increasing trend at a rate of 354.39% (Figure 5b).

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 13 of 27

**Figure 6.** Multiplicative decomposition results of the residential energy-related CO2 emissions in Jiangxi urban and rural regions in the four stages (**a**) 2000–2005. (**b**) 2005–2010. (**c**) 2010–2015. (**d**) 2015– **Figure 6.** Multiplicative decomposition results of the residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions in the four stages (**a**) 2000–2005. (**b**) 2005–2010. (**c**) 2010–2015. (**d**) 2015–2017.

#### 2017. 3.2.1. Consumption Expenditure Per Capita (CE)

3.2.1. Consumption Expenditure Per Capita (CE) To the best of our knowledge, consumption expenditure per capita could reflect folks' living standards of a region to some extent. As shown in Table A5, it could be intuitively seen that consumption expenditure per capita always exerted a positive influence on the urban residential energy-related CO2 emissions, which was in consonance with the results in Figure 5a and Table 6. Specifically, the additive decomposition effect of consumption expenditure per capita was 257.97 × 104 t in 2000–2005 (Figure 5a and Table A3). Then, from 2005–2010, 2010–2015, and 2015–2017, the decomposition effects were 162.15 × 104 t, 215.42 × 104 t and 186.98×104 t (Figure 5a and Table A3). In addition, the consumption expenditure per-capita effect contributed to the largest annual growth rate of the CO2 emissions (11.77%), and the corresponding average annual contribution rates at four stages were 9.55%, 11.68%, 10.24% and 7.51% (Table 6), respectively. Thus, the conclusion could be drawn that the consumption expenditure per-capita effect played the most important role in promoting the To the best of our knowledge, consumption expenditure per capita could reflect folks' living standards of a region to some extent. As shown in Table A5, it could be intuitively seen that consumption expenditure per capita always exerted a positive influence on the urban residential energy-related CO<sup>2</sup> emissions, which was in consonance with the results in Figure 5a and Table 6. Specifically, the additive decomposition effect of consumption expenditure per capita was 257.97 <sup>×</sup> <sup>10</sup><sup>4</sup> t in 2000–2005 (Figure 5a and Table A3). Then, from 2005–2010, 2010–2015, and 2015–2017, the decomposition effects were 162.15 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 215.42 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 186.98 <sup>×</sup> <sup>10</sup><sup>4</sup> t (Figure 5a and Table A3). In addition, the consumption expenditure per-capita effect contributed to the largest annual growth rate of the CO<sup>2</sup> emissions (11.77%), and the corresponding average annual contribution rates at four stages were 9.55%, 11.68%, 10.24% and 7.51% (Table 6), respectively. Thus, the conclusion could be drawn that the consumption expenditure per-capita effect played the most important role in promoting the urban residential energy-related CO<sup>2</sup> emissions.

urban residential energy-related CO2 emissions. For rural regions, the consumption expenditure per capita also always had a positive effect on the rural residential energy-related CO2 emissions (Figure 5b and Table A6). At different four stages, the effect contributed to 166×104 t, 258.04×104 t, 472.22 × 104 t and 160.16 × 104 t, and the corresponding average annual contribution rates were 15.88%, 9.04%, 14.76% and 9.75% (Figure 5b and Table 7), respectively. Overall, the average annual led to the largest annual contribution rate (29.60%) over the studied decade, suggesting the effect of consumption expenditure per capita was also the most critical For rural regions, the consumption expenditure per capita also always had a positive effect on the rural residential energy-related CO<sup>2</sup> emissions (Figure 5b and Table A6). At different four stages, the <sup>e</sup>ffect contributed to 166 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 258.04 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 472.22 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 160.16 <sup>×</sup> <sup>10</sup><sup>4</sup> t, and the corresponding average annual contribution rates were 15.88%, 9.04%, 14.76% and 9.75% (Figure 5b and Table 7), respectively. Overall, the average annual led to the largest annual contribution rate (29.60%) over the studied decade, suggesting the effect of consumption expenditure per capita was also the most critical promoting factor to the rural residential energy-related CO<sup>2</sup> emissions.

promoting factor to the rural residential energy-related CO2 emissions. Therefore, simply stated, these results implicated that the effect of consumption expenditure was Therefore, simply stated, these results implicated that the effect of consumption expenditure was the first promoting factor to residential energy-related CO<sup>2</sup> emissions for both urban and rural

residents, which was in line with the previous studies [9,17]. In fact, with the development of Jiangxi's economy, both urban and rural residents' living standards had greatly improved. Specifically, the consumption expenditure per capita of urban and rural residents increased from 3623.52 RMB and 1642.66 RMB in 2000 to 19244.46 RMB and 9230.21 RMB in 2017 (Figure 7), with a growth rate of 10.32% and 10.69%, respectively. As a result, people had the capacity to shift their consumption patterns to buy more appliances and transport tools, resulting in more energy consumption and CO<sup>2</sup> emissions. According to Pachauri [50], future increases in expenditure levels will lead to further increases in household energy requirement, in turn, a further increase in the urban and rural residential energy-related CO<sup>2</sup> emissions caused by the steady increase of the consumption expenditure per capita in Jiangxi in the future. residential energy-related CO2 emissions caused by the steady increase of the consumption expenditure per capita in Jiangxi in the future. 3.2.2. Energy Demand (ED) It could be intuitively observed that the additive decomposition effect of energy demand was – 73.42 × 104 t and –0.11 × 104 t in urban regions (Figure 5a and Table A3), exerting a negative effect on decreasing the urban residential energy-related CO2 emissions during the first and second stages. However, the negative effect weakened year by year. This situation was mainly because the energy consumption growth lagged behind the growth of the consumption expenditure. However, during the third and fourth stages, the decomposition effect increased to 597.57 × 104 t and 98.05 × 104 t (Figure

increases in household energy requirement, in turn, a further increase in the urban and rural

*Sustainability* **2020**, *12*, x FOR PEER REVIEW 14 of 27

residents, which was in line with the previous studies [9,17]. In fact, with the development of Jiangxi's economy, both urban and rural residents' living standards had greatly improved. Specifically, the consumption expenditure per capita of urban and rural residents increased from 3623.52 RMB and 1642.66 RMB in 2000 to 19244.46 RMB and 9230.21 RMB in 2017 (Figure 7), with a growth rate of 10.32% and 10.69%, respectively. As a result, people had the capacity to shift their consumption

#### 3.2.2. Energy Demand (ED) 5a and Table A3), respectively. This could be attributable to the rapid growth of the energy demand, which increased from 10.45% in 2010 to 21.64% in 2015, and then to 23.85% in 2017 (Figure 7). In total,

It could be intuitively observed that the additive decomposition effect of energy demand was –73.42 <sup>×</sup> <sup>10</sup><sup>4</sup> t and –0.11 <sup>×</sup> <sup>10</sup><sup>4</sup> t in urban regions (Figure 5a and Table A3), exerting a negative effect on decreasing the urban residential energy-related CO<sup>2</sup> emissions during the first and second stages. However, the negative effect weakened year by year. This situation was mainly because the energy consumption growth lagged behind the growth of the consumption expenditure. However, during the third and fourth stages, the decomposition effect increased to 597.57 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 98.05 <sup>×</sup> <sup>10</sup><sup>4</sup> t (Figure 5a and Table A3), respectively. This could be attributable to the rapid growth of the energy demand, which increased from 10.45% in 2010 to 21.64% in 2015, and then to 23.85% in 2017 (Figure 7). In total, the effect of energy demand made the second-largest contribution to the total annual average rate of urban residential CO<sup>2</sup> emissions (6.77%), and for each of the four stages, the corresponding average annual contribution rates were –2.72%, –0.01%, 16.82% and 5.29%, respectively (Table 6). Overall, the energy demand effect played a key role in increasing the urban residential energy-related CO<sup>2</sup> emissions although it slightly mitigated the CO<sup>2</sup> emissions in Stage 1 and 2 (Figure 6 and Table A5). the effect of energy demand made the second-largest contribution to the total annual average rate of urban residential CO2 emissions (6.77%), and for each of the four stages, the corresponding average annual contribution rates were –2.72%, –0.01%, 16.82% and 5.29%, respectively (Table 6). Overall, the energy demand effect played a key role in increasing the urban residential energy-related CO2 emissions although it slightly mitigated the CO2 emissions in Stage 1 and 2 (Figure 6 and Table A5). Similarly, energy demand also had a positive effect on promoting the rural residential energyrelated CO2 emissions over the study period (Figure 6 and Table A6). In total, it contributed to 427.17 × 104 t CO2 emissions, with an annual average contribution rate of 11.96% (Figure 5b and Table 7). Compared to urban regions, the energy demand exerted a positive effect on increasing the rural residential energy-related CO2 emissions except Stage 1 (–25.50 × 104 t), which contributed to 233.69 × 104 t, 140.46×104 t and 78.52 × 104 t in Stage 2, 3 and 4 (Figure 5b and Table A4), respectively. This was because the energy demand decreased from 13.63% in 2000 to 13.13% in 2005, then it increased to 20.01% in 2010 and 27.12% in 2017 (Figure 7).

**Figure 7.** Trends of consumption expenditure, residence expenditure and energy demand in Jiangxi over the study period; Energy demand denotes the share of residence expenditure to consumption expenditure. **Figure 7.** Trends of consumption expenditure, residence expenditure and energy demand in Jiangxi over the study period; Energy demand denotes the share of residence expenditure to consumption expenditure.

Similarly, energy demand also had a positive effect on promoting the rural residential energy-related CO<sup>2</sup> emissions over the study period (Figure 6 and Table A6). In total, it contributed to 427.17 <sup>×</sup> <sup>10</sup><sup>4</sup> t CO<sup>2</sup> emissions, with an annual average contribution rate of 11.96% (Figure 5b and Table 7). Compared to urban regions, the energy demand exerted a positive effect on increasing the rural residential energy-related CO<sup>2</sup> emissions except Stage 1 (–25.50 <sup>×</sup> <sup>10</sup><sup>4</sup> t), which contributed to 233.69 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 140.46 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 78.52 <sup>×</sup> <sup>10</sup><sup>4</sup> t in Stage 2, 3 and 4 (Figure 5b and Table A4), respectively. This was because the energy demand decreased from 13.63% in 2000 to 13.13% in 2005, then it increased to 20.01% in 2010 and 27.12% in 2017 (Figure 7).

**Table 6.** Urban: Types and trends of various effects at different stages and average annual contribution rate in Jiangxi urban regions.


Note: <sup>a</sup> Stage 1, 2, 3 and 4 refer to 2000–2005, 2005–2010, 2010–2015, and 2015–2017, respectively; <sup>b</sup> is the average annual contribution rate of the total urban residential energy-related CO<sup>2</sup> emissions; + and – stand for positive and negative effect on the CO<sup>2</sup> emissions, respectively.

**Table 7.** Rural: Types and trends of various effects at different stages and average annual contribution rate in Jiangxi rural regions.


Note: <sup>a</sup> Stage 1, 2, 3 and 4 refer to 2000–2005, 2005–2010, 2010–2015, and 2015–2017, respectively; <sup>c</sup> is the average annual contribution rate of the total rural residential energy-related CO<sup>2</sup> emissions; + and – stand for positive and negative effect on the CO<sup>2</sup> emissions, respectively.

#### 3.2.3. Population (P)

It was apparent the effect of population was the most stable and negligible factor to promote the urban and rural residential energy-related CO<sup>2</sup> emissions (Tables A5 and A6), which was consistent with the previous studies [11,20]. Specifically, the average annual promoting impact of population effect (0.76% for urban, 1.70% for rural) was much lower than those of consumption expenditure per-capita effect and energy demand effect (Tables 6 and 7). See from urban and rural, at each of four stages, the additive decomposition effects were 19.25 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 20.06 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 18.29 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 12.14 <sup>×</sup> <sup>10</sup><sup>4</sup> t for urban residents (Figure 5a and Table A3), similarly, 14.09 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 19.62 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 16.11 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 10.77 <sup>×</sup> <sup>10</sup><sup>4</sup> t for rural regions (Figure 5b and Table A4). Over the study period, it could be found the population effect presented an inverted U-shape trend in urban and rural regions, implying the population effect would weaken in the future. The total population increased from 41.49 million to 46.22 million, with an annual growth rate of only 0.64% in Jiangxi over this 17-year period (Figure 8). In fact, to the best of our knowledge, population was a stable element for a region during a certain period, i.e., population would increase with a stable growth rate.

#### 3.2.4. Urbanization (U)

The additive decomposition results of urbanization in urban regions were 145.86 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 98.97 <sup>×</sup> <sup>10</sup><sup>4</sup> t, 125.77 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 55.32 <sup>×</sup> <sup>10</sup><sup>4</sup> t at each of four stages (Figure 5a and Table A3), indicating the effect of urbanization always had a positive influence on increasing the urban residential energy-related CO<sup>2</sup> emissions. The results were consonance with the results in Table A5. In total, the effect of urbanization made a specific contribution to the total urban residential energy-related CO<sup>2</sup> emissions growth (4.64%) (Table 6). Compare to urban regions, the additive decomposition results of urbanization effect in rural regions were –50.56 <sup>×</sup> <sup>10</sup><sup>4</sup> t, –66.74 <sup>×</sup> <sup>10</sup><sup>4</sup> t, –101.60 <sup>×</sup> <sup>10</sup><sup>4</sup> t and –55.66 <sup>×</sup> <sup>10</sup><sup>4</sup> t during the four stages (Figure 5b and Table A4), respectively. Eventually, the urbanization effect led to the total rural residential energy-related CO<sup>2</sup> emissions decrease at an annual average rate of 7.69% (Table 7). Overall, the results showed the urbanization effect was the inhibiting factor to the rural residential energy-related CO<sup>2</sup> emissions (Table A6). The difference between urban and rural regions could be explained by the followings. *Sustainability* **2020**, *12*, x FOR PEER REVIEW 16 of 27 growth (4.64%) (Table 6). Compare to urban regions, the additive decomposition results of urbanization effect in rural regions were –50.56 × 104 t, –66.74 × 104 t, –101.60 × 104 t and –55.66 × 104 t during the four stages (Figure 5b and Table A4), respectively. Eventually, the urbanization effect led to the total rural residential energy-related CO2 emissions decrease at an annual average rate of 7.69% (Table 7). Overall, the results showed the urbanization effect was the inhibiting factor to the rural residential energy-related CO2 emissions (Table A6). The difference between urban and rural regions could be explained by the followings.

As depicted in Figure 8, the population of urban residents increased from 11.49 million to 25.24 million with an annual growth rate of 4.74%, and the population of rural residents decreased from 30.00 million to 20.98 million with an annual change of –2.08%. Correspondingly, the urbanization rate steadily increased from 27.69% to 54.6% from 2000–2017. In light of the context mentioned above, the urbanization effect increased the urban residential energy-related CO<sup>2</sup> emissions, but decreased the rural residential energy-related CO<sup>2</sup> emissions. On average, when the urbanization increased by 1%, the urban residential energy-related CO<sup>2</sup> emissions might have a growth of about 15.83 <sup>×</sup> <sup>10</sup><sup>4</sup> t, while the rural residential energy-related CO<sup>2</sup> emissions might have a reduction of about 10.20 <sup>×</sup> <sup>10</sup><sup>4</sup> t. See from each of four stages, the annual average contribution rates were 5.40%, 3.61%, 3.54% and 2.99% in urban regions (Table 6), and –4.81%, –2.34%, –3.17% and –3.39% in rural regions (Table 7). The urbanization effect presented a decline trend on the whole. In the future, this effect might be offset with the further development of Jiangxi's urbanization. It was demonstrated by the contribution rate of the urbanization effect in Guangdong, the most developed province of China, experienced an increase trend to 53.6% in 2003, then declined to 17.3% in 2012 with a diminishing trend over time [21]. Furthermore, Fan, et al. [19] also pointed out the urbanization contributed to the increase to Chinese residential energy consumption but with a diminishing trend over time. As depicted in Figure 8, the population of urban residents increased from 11.49 million to 25.24 million with an annual growth rate of 4.74%, and the population of rural residents decreased from 30.00 million to 20.98 million with an annual change of –2.08%. Correspondingly, the urbanization rate steadily increased from 27.69% to 54.6% from 2000–2017. In light of the context mentioned above, the urbanization effect increased the urban residential energy-related CO2 emissions, but decreased the rural residential energy-related CO2 emissions. On average, when the urbanization increased by 1%, the urban residential energy-related CO2 emissions might have a growth of about 15.83×104 t, while the rural residential energy-related CO2 emissions might have a reduction of about 10.20×104 t. See from each of four stages, the annual average contribution rates were 5.40%, 3.61%, 3.54% and 2.99% in urban regions (Table 6), and –4.81%, –2.34%, –3.17% and –3.39% in rural regions (Table 7). The urbanization effect presented a decline trend on the whole. In the future, this effect might be offset with the further development of Jiangxi's urbanization. It was demonstrated by the contribution rate of the urbanization effect in Guangdong, the most developed province of China, experienced an increase trend to 53.6% in 2003, then declined to 17.3% in 2012 with a diminishing trend over time [21]. Furthermore, Fan, et al. [19] also pointed out the urbanization contributed to the increase to Chinese residential energy consumption but with a diminishing trend over time.

**Figure 8.** Trends of urban and rural population in Jiangxi over the study period. **Figure 8.** Trends of urban and rural population in Jiangxi over the study period.

#### 3.2.5. Energy Structure (ES) 3.2.5. Energy Structure (ES)

Energy structure adjustment played a minor role in increasing the residential energy-related CO2 emissions in urban and rural regions (Table A5 and Table A6), which was consistent with most relevant studies [21,51]. Over the study period, the aggregate changes of the energy structure effect were 94.29 × 104 t and 88.84 × 104 t (Figure 5), and the corresponding average annual contribution rates were 1.03% and 2.49%, respectively. Specifically, in the four stages, the corresponding annual average contribution rates were 6.34%, –2.23%, –0.82%, 0.72% for urban regions (Table 6), and 3.84%, 2.79%, Energy structure adjustment played a minor role in increasing the residential energy-related CO<sup>2</sup> emissions in urban and rural regions (Tables A5 and A6), which was consistent with most relevant studies [21,51]. Over the study period, the aggregate changes of the energy structure effect were 94.29 <sup>×</sup> <sup>10</sup><sup>4</sup> t and 88.84 <sup>×</sup> <sup>10</sup><sup>4</sup> t (Figure 5), and the corresponding average annual contribution rates were 1.03% and 2.49%, respectively. Specifically, in the four stages, the corresponding annual average contribution rates were 6.34%, –2.23%, –0.82%, 0.72% for urban regions (Table 6), and 3.84%, 2.79%,

energy-related CO2 emissions caused by oil and electricity rapidly increased, especially electricity. Furthermore, the share of natural gas was still low in Jiangxi, especially in rural regions. It was worth –1.13% and 0.31% for rural regions (Table 7). The usage of coal steadily decreased in both urban and rural regions, resulting in the share of the residential energy-related CO<sup>2</sup> emissions decreased to 5.90%, 27.52% in 2017 from 60.06%, 73.59% in 2000 (Figure 4). However, the share of residential energy-related CO<sup>2</sup> emissions caused by oil and electricity rapidly increased, especially electricity. Furthermore, the share of natural gas was still low in Jiangxi, especially in rural regions. It was worth mentioning that the source endowment of China (including Jiangxi) was dominated by coal in the long term, and the electricity generation was mainly from coal [9,52]. The renewable and sustainable energy, such as wind, solar, nuclear, and biomass energy, had not been widely used [36]. Thus, it still required a longer time to thoroughly shift the energy structure to the low-carbon pattern to reduce the residential energy-related CO<sup>2</sup> emissions.

#### 3.2.6. Energy Price (EP)

The energy price effect had the most important impact lowering the residential CO<sup>2</sup> emissions in urban and rural regions over the study period (Tables A5 and A6), which was consistent with the previous studies [10,21]. Specifically, the additive decomposition effects were –549.64 <sup>×</sup> <sup>10</sup><sup>4</sup> t, –152.25 <sup>×</sup> <sup>10</sup><sup>4</sup> t, –782.11 <sup>×</sup> <sup>10</sup><sup>4</sup> t and –128.61 <sup>×</sup> <sup>10</sup><sup>4</sup> t for urban regions (Figure 5a and Table A3), the corresponding average annual contribution rates were –20.34%, –5.55%, –22.02% and –6.95%, respectively (Table 6). Similarly, in rural regions, the energy price effect led to 195.41 <sup>×</sup> <sup>10</sup><sup>4</sup> t CO<sup>2</sup> emissions from 2000–2005. However, the decomposition effect decreased greatly to –413.33 <sup>×</sup> <sup>10</sup><sup>4</sup> t, –232.56 <sup>×</sup> <sup>10</sup><sup>4</sup> t and –63.31 <sup>×</sup> <sup>10</sup><sup>4</sup> t during Stage 2, 3 and 4 (Figure 5b and Table A4), respectively, and for each of the four stages, the average annual contribution rates were 18.60%, –14.48%, –7.27% and –3.85%, respectively (Table 7).

Overall, the energy price effect made the greatest contribution to the total annual rates of urban and rural residential energy-related CO<sup>2</sup> emissions mitigation (–17.55% for urban and –14.38% for rural) (Tables 6 and 7). This was mainly because the energy price increased from 2410.53 RMB/ ton-standard coal equivalent (SCE), 9855.43 RMB/ ton SCE in 2000 to 28078.53 RMB/ ton SCE, 16900.13 RMB/ ton SCE in 2017 for urban and rural regions, respectively (Figure 9), restraining the willing of people's consumption. In particular, it should be noted that the energy price rapidly grew from 9730.42 RMB/ ton SCE in 2012 to 24690.65 RMB/ ton SCE in 2015 because of the implementation of the "ladder electricity price plan (In November 2011, the National Development and Reform Commission (NDRC) issued the "residential ladder electricity price policy" to promote energy conservation and emission reduction as well as social equity. Specifically, residential electricity consumption was divided into three grades and a free grade for low-income households.)" in Jiangxi urban regions, making the energy price reached the peak value (–782.11 <sup>×</sup> <sup>10</sup><sup>4</sup> RMB/ton SCE). Moreover, during Stage 1, the energy price decreased from 9855.43 RMB/ ton SCE to 5226.89 RMB/ ton SCE, which stimulated the consumption of rural residents.

ton SCE, which stimulated the consumption of rural residents.

the residential energy-related CO2 emissions.

3.2.6. Energy Price (EP)

3.85%, respectively (Table 7).

mentioning that the source endowment of China (including Jiangxi) was dominated by coal in the long term, and the electricity generation was mainly from coal [9,52]. The renewable and sustainable energy, such as wind, solar, nuclear, and biomass energy, had not been widely used [36]. Thus, it still required a longer time to thoroughly shift the energy structure to the low-carbon pattern to reduce

The energy price effect had the most important impact lowering the residential CO2 emissions in urban and rural regions over the study period (Table A5 and Table A6), which was consistent with the previous studies [10,21]. Specifically, the additive decomposition effects were –549.64 × 104 t, – 152.25 × 104 t, –782.11 × 104 t and –128.61 × 104 t for urban regions (Figure 5a and Table A3), the corresponding average annual contribution rates were –20.34%, –5.55%, –22.02% and –6.95%, respectively (Table 6). Similarly, in rural regions, the energy price effect led to 195.41 × 104 t CO2 emissions from 2000–2005. However, the decomposition effect decreased greatly to –413.33 × 104 t, – 232.56 × 104 t and –63.31 × 104 t during Stage 2, 3 and 4 (Figure 5b and Table A4), respectively, and for each of the four stages, the average annual contribution rates were 18.60%, –14.48%, –7.27% and –

Overall, the energy price effect made the greatest contribution to the total annual rates of urban and rural residential energy-related CO2 emissions mitigation (–17.55% for urban and –14.38% for rural) (Table 6 and Table 7). This was mainly because the energy price increased from 2410.53 RMB/ ton-standard coal equivalent (SCE), 9855.43 RMB/ ton SCE in 2000 to 28078.53 RMB/ ton SCE, 16900.13 RMB/ ton SCE in 2017 for urban and rural regions, respectively (Figure 9), restraining the willing of people's consumption. In particular, it should be noted that the energy price rapidly grew from 9730.42 RMB/ ton SCE in 2012 to 24690.65 RMB/ ton SCE in 2015 because of the implementation of the "ladder electricity price plan (In November 2011, the National Development and Reform Commission (NDRC) issued the "residential ladder electricity price policy" to promote energy conservation and emission reduction as well as social equity. Specifically, residential electricity consumption was divided into three grades and a free grade for low-income households.)" in Jiangxi urban regions, making the energy price reached the peak value (–782.11 × 104 RMB/ton SCE).

**Figure 9.** Energy price in urban and rural Jiangxi. **Figure 9.** Energy price in urban and rural Jiangxi.

#### 3.2.7. Carbon Emission Coefficient (K) 3.2.7. Carbon Emission Coefficient (K)

As mentioned above, in this article, the carbon emission coefficient of electricity was not a constant value. The carbon emission coefficient effect mainly derived from the change of the carbon emission coefficient of power generation (Table A1). In addition, according to Ren, et al. [53], if the As mentioned above, in this article, the carbon emission coefficient of electricity was not a constant value. The carbon emission coefficient effect mainly derived from the change of the carbon emission coefficient of power generation (Table A1). In addition, according to Ren, et al. [53], if the contribution of the energy emission factor was negative, the residential energy-related CO<sup>2</sup> emissions decreased. Otherwise, the contribution of the energy emission factor was positive and the residential energy-related increased.

The carbon emission coefficient effect had the least influence on inhibiting the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions (Tables A5 and A6). In total, the average annual contribution rates were –1.17% and –2.82% for urban and rural regions (Tables 6 and 7), respectively. See from each of four stages, the corresponding contribution rates were 1.36% and 1.92% in urban and rural regions from 2000–2005, indicating the carbon emission coefficient effect had a minor influence on increasing the CO<sup>2</sup> emissions. This situation was likely that the carbon emission coefficient of electricity increased from 6.49 t-CO2/10<sup>4</sup> kwh to 7.99 t-CO2/10<sup>4</sup> kwh in Jiangxi (Table A1). However, in 2005–2010, 2010–2015, and 2015–2017, the average annual contribution rates decreased to –2.32%, –2.22%, –0.12% for urban regions, and –1.47%, –2.40%, –0.15% for rural regions, respectively. The results might be due to the improvement of power generation technology, updated generation equipment and other factors [53]. Specifically, the carbon emission coefficient of electricity decreased from 7.99 t-CO2/10<sup>4</sup> kwh to 4.84 t-CO2/10<sup>4</sup> kwh in 2017. On the other hand, the carbon emission coefficient effect represented an overall downward trend, resulting from the long-term coal-dependent power generation in Jiangxi (Average=80.91%) (Figure 10). Hence, the carbon emission coefficient effect eventually had a relatively marginal negative contribution to the residential energy-related CO<sup>2</sup> emissions reduction. To effectively enhance the effect of energy emission factor to mitigate CO<sup>2</sup> emissions, it is necessary to boost the share of cleaner energy types such as hydropower, wind power, and solar power in Jiangxi's electricity generation structure.

energy-related increased.

contribution of the energy emission factor was negative, the residential energy-related CO2 emissions decreased. Otherwise, the contribution of the energy emission factor was positive and the residential

The carbon emission coefficient effect had the least influence on inhibiting the residential energyrelated CO2 emissions in both urban and rural regions (Table A5 and Table A6). In total, the average annual contribution rates were –1.17% and –2.82% for urban and rural regions (Table 6 and Table 7), respectively. See from each of four stages, the corresponding contribution rates were 1.36% and 1.92% in urban and rural regions from 2000–2005, indicating the carbon emission coefficient effect had a minor influence on increasing the CO2 emissions. This situation was likely that the carbon emission coefficient of electricity increased from 6.49 t-CO2/104 kwh to 7.99 t-CO2/104 kwh in Jiangxi (Table A1). However, in 2005–2010, 2010–2015, and 2015–2017, the average annual contribution rates decreased to –2.32%, –2.22%, –0.12% for urban regions, and –1.47%, –2.40%, –0.15% for rural regions, respectively. The results might be due to the improvement of power generation technology, updated generation equipment and other factors [53]. Specifically, the carbon emission coefficient of electricity decreased from 7.99 t-CO2/104 kwh to 4.84 t-CO2/104 kwh in 2017. On the other hand, the carbon emission coefficient effect represented an overall downward trend, resulting from the long-term coaldependent power generation in Jiangxi (Average=80.91%) (Figure 10). Hence, the carbon emission coefficient effect eventually had a relatively marginal negative contribution to the residential energyrelated CO2 emissions reduction. To effectively enhance the effect of energy emission factor to

**Figure 10.** Energy types of power generation in Jiangxi. **Figure 10.** Energy types of power generation in Jiangxi.

#### *3.3. Decoupling Analysis of Residential Energy-Related CO2 Emissions at Four Stages 3.3. Decoupling Analysis of Residential Energy-Related CO<sup>2</sup> Emissions at Four Stages*

wind power, and solar power in Jiangxi's electricity generation structure.

As shown in Table 8, four decoupling states occurred in Jiangxi urban regions, namely SD, END, WD, and EC. From 2000–2005, the decoupling states switched every year, thereby implying that an unstable relationship between urban residential energy-related CO2 emissions and consumption expenditure. Eventually, the decoupling state showed WD, that is to say, the consumption expenditure increased a little faster than the CO2 emissions. After 2005, the decoupling states converged on the WD in most years of the period 2005–2016, except 2009–2010 and 2012–2013. Furthermore, the decoupling state was EC in the end, suggesting that there is no significant relationship between the CO2 emissions and consumption expenditure. Overall, the decoupling states were the WD, WD, and EC from 2005–2010, 2010–2015, and 2015–2017, respectively. Over the study period, the decoupling state was WD in Jiangxi urban regions. As shown in Table 8, four decoupling states occurred in Jiangxi urban regions, namely SD, END, WD, and EC. From 2000–2005, the decoupling states switched every year, thereby implying that an unstable relationship between urban residential energy-related CO<sup>2</sup> emissions and consumption expenditure. Eventually, the decoupling state showed WD, that is to say, the consumption expenditure increased a little faster than the CO<sup>2</sup> emissions. After 2005, the decoupling states converged on the WD in most years of the period 2005–2016, except 2009–2010 and 2012–2013. Furthermore, the decoupling state was EC in the end, suggesting that there is no significant relationship between the CO<sup>2</sup> emissions and consumption expenditure. Overall, the decoupling states were the WD, WD, and EC from 2005–2010, 2010–2015, and 2015–2017, respectively. Over the study period, the decoupling state was WD in Jiangxi urban regions.

Table 8 also presented the decoupling state of Jiangxi rural regions. Similarly, there are also four types of decoupling states: END, SD, WD, and EC. However, there are some differences. Specifically, Table 8 also presented the decoupling state of Jiangxi rural regions. Similarly, there are also four types of decoupling states: END, SD, WD, and EC. However, there are some differences. Specifically, the decoupling states were stable and converged on END from 2000–2005. Meanwhile most years of the period 2005–2015, the decoupling states mainly converged on WD and SD, except 2008–2009 (END). This caused the decoupling state shift from END in 2000–2005 to WD in 2005–2010 and 2010–2015. In the later period (2015–2017), the decoupling state was EC, resulting from the unsatisfactory decoupling states in 2015–2016 (EC) and 2016–2017 (END). Simply stated, the decoupling state in Jiangxi rural regions showed the END over the study period. Compared to the urban regions, the decoupling process needed further to accelerate in Jiangxi rural regions.

During Stage 1 (2000–2005), both urban and rural residents' income increased with the development of economy. Correspondingly, the consumption expenditure grew by 135% and 37% (Table 8), respectively. To ensure the quick development of economy, the demand for urban residents was restrained, which caused the residential energy-related CO<sup>2</sup> emissions increased by only 1%. Thus, the consumption expenditure increased much greater than the CO<sup>2</sup> emissions, leading to an overall WD (0.01) in Stage 1 although EC and END occurred in 2001–2003 and 2004–2005 (Table 8). See from rural, decline of energy price stimulated the residents' consumption desire which made the energy demand increase (Figure 7). Meanwhile, it caused the CO<sup>2</sup> emissions rose by 172% in 2000–2005 which was greatly higher than that of consumption expenditure. Consequently, the decoupling state was END in this stage.

During Stage 2 (2005–2010), the consumption expenditure rose by 114% (Table 8), which made urban residents pursuit more high-quality life and buy more high-carbon appliance and family cars (Table 4). Thus, the CO<sup>2</sup> emissions grew by 30%, presenting the WD (0.26) (Table 8) in 2005–2010. As to rural residents, on one hand, rural residential consumption expenditure further rose by 45%. On the other hand, traditional outdated energy gradually placed by the commercial energy in Jiangxi rural regions and the CO<sup>2</sup> emissions increased by 12%. As a result, the decoupling state shifted from END to WD in Stage 2.

During Stage 3 (2010–2015), both urban and rural regions presented WD. The growth rate of the consumption expenditure continued to slow down and grew by 89% in urban regions. However, it continued to rise and grew by 74% in rural regions (Table 8). This could be explained by the long-term inequality and lag between urban and rural regions in China's economic development (including Jiangxi), namely duality of urban–rural. Moreover, in order to save energy and reduce emissions, Chinese government (Jiangxi included) formulated and implemented the "ladder electricity price policy" after 2012, which effectively restrained residential energy consumption, especially in urban regions. Thus, the change rate of the CO<sup>2</sup> emissions in urban regions nearly had no obvious increase (30% from 2010–2015), and it grew by 28% in rural regions. Consequently, WD occurred.

**Table 8.** The decoupling state between residential energy-related CO<sup>2</sup> emissions and consumption expenditure in Jiangxi urban and rural regions over the study period.


Note: *C*% and *Y*% denote change rates of residential energy-related CO<sup>2</sup> emissions and consumption expenditure, respectively; D denotes the results of decoupling indicators.

During Stage 4 (2015–2017), the relationship between consumption expenditure and residential energy-related CO<sup>2</sup> emissions transformed from WD to EC in both urban and rural regions. Chinese economic development embarked on a new stage called "the new normal". Specifically, in this stage, China (including Jiangxi) shifted its development pattern from rapid growth to more inclusive and sustainable growth, including higher living standards, cleaner production [54]. Consequently, the growth rate of consumption expenditure and the CO<sup>2</sup> emissions declined, to some extent. Consumption expenditure rose by 23% and 14% in Jiangxi urban and rural regions (Table 8), respectively, which was slower than that of other periods. Additionally, the CO<sup>2</sup> emissions also showed a decline trend and grew by 20% and 16% in Jiangxi urban and rural regions, resulting from the improvement of energy efficiency. Therefore, they both presented EC in this stage, indicating there was insignificant decoupling between consumption expenditure and the CO<sup>2</sup> emissions.

#### **4. Conclusions and Policy Implications**

#### *4.1. Main Conclusions*

Presently, residential energy consumption has become the second-largest sector to the industrial energy consumption in China. However, little attention was paid to research the difference of the residential energy-related CO<sup>2</sup> emissions between urban and rural areas in the less developed and undeveloped regions. Thus, based on the urban–rural duality, this paper aims at exploring the difference of the residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions from 2000–2017. We first overviewed the changes of the residential energy-related CO<sup>2</sup> emissions in Jiangxi between urban and rural regions. Then, the LMDI method was introduced to distinguish the major factors affecting the residential energy-related CO<sup>2</sup> emissions. In addition, the Tapio decoupling model was used to analyze the relationship between residential energy-related CO<sup>2</sup> emissions and consumption expenditure. The main results were acquired as follows.

The residential energy-related CO<sup>2</sup> emissions rapidly increased after 2008 in both Jiangxi urban and rural regions. However, the annual growth rate of urban regions (7.47%) was faster than rural regions (6.03%). In addition, the gap between urban and rural regions became narrowed and stable. The energy structures of both urban and rural residential energy-related CO<sup>2</sup> emissions had shifted from coal-dominant to a multiple structure which consisted of oil, natural gas, and electricity. Specifically, the energy-related CO<sup>2</sup> emissions structure was electricity, oil, natural gas, and coal in sequence for urban; it was electricity, coal, oil, and natural gas in sequence for rural. As for the residential energy-related CO<sup>2</sup> emissions per capita, it showed a decline trend in urban regions, while an increasing tendency in rural regions over the study period. The gap in residential energy-related CO<sup>2</sup> emissions for urban and rural residents narrowed from 2000–2007.

The energy price effect played the most important role in reducing the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions. In addition, the carbon emission coefficient had a minor effect on decreasing urban and rural residential energy-related CO<sup>2</sup> emissions. The results showed that urbanization exerted a positive effect on increasing the urban residential energy-related CO<sup>2</sup> emissions, but decreasing the rural residential energy-related CO<sup>2</sup> emissions. Consumption expenditure-per-capita effect was the most important factor increasing the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions, followed by energy demand. From 2000–2017, population effect and energy structure played a minor role in increasing the urban and rural residential energy-related CO<sup>2</sup> emissions.

Over the study period, four decoupling status (END, SD, WD, and EC) occurred in both urban and rural regions when analyzed the relationship between consumption expenditure and residential energy-related CO<sup>2</sup> emissions. Overall, the decoupling state for urban and rural regions were the WD and END. This showed that residential energy-related CO<sup>2</sup> emissions in urban Jiangxi was less depend on consumption expenditure, but rural residential energy-related CO<sup>2</sup> emissions still depended on consumption expenditure.

#### *4.2. Policy Implications*

First, since the residential energy-related CO<sup>2</sup> emissions caused by coal was still high in rural regions, which accounted for 27.52% and was second only to electricity (53.78%) even in 2017. However, the corresponding proportion of natural gas was always low and took up less than 1% over the study period. Thus, the energy structure of rural regions had more space to optimize, and more efforts should be paid to further adjust the rural energy structure. For instance, promoting the proportion of natural gas and encouraging the use of renewable and clean energy in rural regions.

Second, since the rural residential energy-related CO<sup>2</sup> emissions per capita had exceeded that of the urban region after 2015, thereby suggesting that the lifestyles in rural regions were more extensive and lower energy efficiency compared with urban regions. On one hand, it is urgent and imperative for the government at all levels to strengthen related public education and guide residents to shift their high-carbon lifestyles to a green and low-carbon pattern, especially for rural residents. In addition, the government should accelerate the elimination of backward production capacity to improve the efficiency in rural regions. On the other hand, residents themselves should also initiatively adapt to the transformation and cultivate the awareness of environmental protection.

Third, consumption expenditure-per-capita effect became the most factor increasing the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions. With the improvement of economic growth and living standards, people had the desire and capacity to pursuit the quality of life and buy more energy-intensive electricity appliances and private transport tools. However, it is not wise to reduce the CO<sup>2</sup> emissions via restraining the consumption, this was because the consumption was a critical impetus of socio-economic development. Thus, the government could apply the fiscal and tax policy to cut down the residential energy-related CO<sup>2</sup> emissions and accelerate the decoupling. Specifically, the government should continue to impose the "Home Appliances Subsidy Program" and monitor its proper extent of implementation. For the high-carbon and extensive goods and services, the government could add the tax to restrain the CO2-intensive commodities' consumption. Meanwhile for the energy-saving appliances, reducing the tax could stimulate the low-carbon consumption. Moreover, more investment and efforts were needed to improve public transport infrastructures and the energy conservation technology.

Fourth, considering energy price played the most important role in decreasing the residential energy-related CO<sup>2</sup> emissions in both urban and rural regions, indicating tiered energy price was the effective and scientific pathway to achieve energy savings and emissions reduction. In light of the fact, the government needs further to uphold and ameliorate the energy price policy in the future. In particular, when applying the energy price mechanism, the duality as well as inequality of income level and economic development should be comprehensively considered in urban and rural regions.

Last, to the best of our knowledge, residential building energy consumption accounted for a large proportion of the total residential energy consumption, leading to enormous CO<sup>2</sup> emissions. Therefore, it is necessary to improve residential building energy efficiency, especially for rural regions. For urban regions, the intelligent energy management technology in household could be popularized and applied to achieve energy conservation.

#### **5. Limitations and Further Perspectives**

There are limitations to this study. Based on the perspective of end-use of energy, only the residential energy-related CO<sup>2</sup> emissions were estimated, while the indirect or embodied residential CO<sup>2</sup> emissions from production perspective were excluded. This could likely underestimate the residential CO<sup>2</sup> emissions in Jiangxi urban and rural regions. Therefore, the embodied residential CO<sup>2</sup> emissions need further study. Moreover, this paper only explored the factors affecting the residential energy-related CO<sup>2</sup> emissions and the decoupling relationship between consumption expenditure and residential energy-related CO<sup>2</sup> emissions in Jiangxi urban and rural regions, specifically carbon emission coefficient, energy structure, consumption expenditure per capita, energy demand, energy price, urbanization, and population. However, family size, floor space of residential buildings and different income levels (an aspect of social inequality) also had the nonnegligible effect on the residential CO<sup>2</sup> emissions. Thus, these influencing factors to the residential CO<sup>2</sup> emissions should be paid more attention in our future work.

**Author Contributions:** Y.Y., J.J. and C.C. conceived and designed this study; Y.Y. collected and analyzed the data; Y.Y. and J.J. wrote and revised the paper; C.C. contributed to progress of research idea. All authors read and approved the final manuscript.

**Funding:** This research was funded by Chinese National Science Foundation (Grant No. 71473113), and the Research Project of Humanities and Social Sciences in Jiangxi's Universities (Grant No. GL19225).

**Acknowledgments:** We thank the anonymous reviewers and editor for their constructive comments and suggestions to improve the quality of this article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Abbreviation**


#### **Appendix A**

**Calculation of the CO<sup>2</sup> emission coe**ffi**cient of electricity:** Based on Wang, et al. [21], the CO<sup>2</sup> emission coefficient of electricity can be calculated as follows:

$$f\_{\mathcal{E}} = \frac{\sum\_{\mathbf{k}} \sum\_{\bar{i}} E\_{i\mathbf{k}} \cdot f\_{i\mathbf{k}}}{T\_{\mathcal{E}}} = \frac{\sum\_{\bar{i}} E\_{i} \cdot f\_{\bar{i}}}{T\_{\mathcal{E}}} \tag{A1}$$

where the subscript *i* denotes the fuel type consumed in electric power generation, i.e., coal, oil and natural gas; *k* denotes the type of electricity; *T<sup>e</sup>* denotes the total amount of electric power generation; *E<sup>i</sup>* denotes the energy consumption consumed in electric power generation of *i* fuel type; *f<sup>e</sup>* denotes the CO<sup>2</sup> emission coefficient of electricity; *f<sup>i</sup>* denotes the CO<sup>2</sup> emission coefficient of *i* type fuel type.

According to Electricity Balance Sheet of JSY (2001–2018) [37], to date, the types of electric power generation include thermal power, hydropower, and wind power. Among these types of electric power generation, thermal power is the main source of CO<sup>2</sup> emissions, and other contribute negligible CO<sup>2</sup> emissions. Thus, here, we only consider the thermal power when calculating the CO<sup>2</sup> emission coefficient of electricity in Jiangxi. In addition, CO<sup>2</sup> emission coefficients of coal, oil, and natural are 2.7412, 2.1358 and 1.626 kg-CO2/ton-SCE, respectively. The results are listed in Table A1.

**Table A1.** CO<sup>2</sup> emission coefficient of electricity (t-CO<sup>2</sup> /10<sup>4</sup> kwh).



**Table A2.** CO<sup>2</sup> emission coefficients of different fuel types investigated in this study.

Note: <sup>a</sup> The value is the IPCC recommended value; <sup>b</sup> The unit is KJ/m<sup>3</sup> ; <sup>c</sup> The unit is t-CO2/10<sup>4</sup> m<sup>3</sup> .

#### **Appendix B**

**Table A3.** Urban, LMDI additive: Detailed additive decomposition results of the residential energy-related CO<sup>2</sup> emissions of urban residents in Jiangxi (10<sup>4</sup> t).



**Table A4.** Rural, LMDI additive: Detailed additive decomposition results of the residential energy-related CO<sup>2</sup> emissions of rural residents in Jiangxi (10<sup>4</sup> t).

**Table A5.** Urban, LMDI multiplicative: Detailed multiplicative decomposition results of the residential energy-related CO<sup>2</sup> emissions of urban residents in Jiangxi.



**Table A6.** Rural, LMDI multiplicative: Detailed multiplicative decomposition results of the residential energy-related CO<sup>2</sup> emissions of rural residents in Jiangxi.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Revisiting the Relation between Renewable Electricity and Economic Growth: A Renewable–Growth Hypothesis**

## **Minyoung Yang and Jinsoo Kim \***

Department of Earth Resources and Environmental Engineering, Hanyang University, 222 Wangsimni-ro, Seongdong-gu, Seoul 04763, Korea; yminy94@hanyang.ac.kr

**\*** Correspondence: jinsookim@hanyang.ac.kr; Tel.: +82-2-2220-2241

Received: 13 March 2020; Accepted: 10 April 2020; Published: 13 April 2020

**Abstract:** Global concern about the climate crisis has incited movements for switching to renewable electricity. Renewable electricity can contribute to economic growth as an input factor (electricity generation) and also as an industry (renewable manufacturing). We introduce a new hypothesis, the renewable–growth hypothesis, to investigate the role of the renewable manufacturing industry in the energy–growth nexus study. To test the hypothesis, we select a target country group using the market share of the renewable manufacturing industry and conduct the Granger causality test for solar photovoltaic and wind power. The autoregressive distributed lag bounds testing approach is applied for the causality test. The results show that renewable electricity Granger causes economic growth in target countries, which supports the renewable–growth hypothesis. However, the hypothesis did not hold in countries that export renewable power facilities more than they install them for domestic demand. We believe that the renewable–growth hypothesis would be secured soon if renewable electricity expands broadly over the world.

**Keywords:** renewable–growth hypothesis; renewable electricity; economic growth; renewable manufacturing; energy–growth nexus

### **1. Introduction**

International attention to global warming comprises the global effort for carbon reduction. However, if we keep our pledges at the current level, the world's temperature will increase to almost twice the limit referred to in the Paris Agreement by the end of the century. Climate change is now referred to as the "climate crisis" [1]. This phenomenon also draws attention to the topic of energy, particularly in the electricity industry. As part of that, many countries have displayed their transition to renewable energy from fossil fuels, which are considered the main source of carbon emissions. However, many governments, especially those in Asia, are still using coal-fired power stations. Even worse, the demand growth for gas as an alternative to coal is emerging [2]. Transitioning to renewable energy is a means for solving problems caused by the climate crisis.

Indeed, renewable energy accounted for an estimated 18.1% of total final energy consumption (TFEC) in 2017. Modern renewables composed 10.6% of TFEC, with an estimated 4.4% growth in demand compared to 2016. Particularly in the power sector, renewable energy has grown to account for more than 33% of the world's total installed power-generating capacity in 2018. Solar photovoltaics (PV) comprises 55% of renewable capacity, after the additional installation of around 100 gigawatts in 2018, and is followed by wind power (28%) [3].

Economic development and sustainability are important not only for carbon reduction but also for the global trend promoting renewable energies [4–6]. Developed economies promote renewable sources to strengthen the energy security of supply and control their greenhouse gas emissions [7]. Similarly, understanding the causal relationship between renewable energy and economic growth is significant for a country's economic development and as the basis for policymakers. Therefore, extensive research has been conducted on the relationship between them. In the 1980s, many studies investigated the relationship between energy consumption and economic growth [8]. "Energy–growth nexus" is a term referring to the link between energy consumption and economic growth. The energy–growth nexus has reached a more disaggregated level over time [9].

An analysis of renewable energy growth is needed considering the described circumstances. The facility cost of renewable electricity accounts for a higher percentage in power generation costs compared to non-renewable energy [10]. This means that, if we increase solar facilities, the demand for all sectors related to them also increases. This might lead to positive effects on economic growth. Furthermore, renewable electricity is capital intensive and has a value-added effect. According to Ernst & Young [11], the solar industry in the European Union 28 represented more than 81,000 full-time equivalents and more than EUR 4600 million gross value-added (GVA). They mentioned the installed capacity has a significant impact on job and GVA creation because there is a direct impact on manufacturing and services needed. Thus, the increasing demand for renewable electricity has a positive effect on such industries and is related to economic growth.

However, if a country generally imports equipment and produces its own electricity, renewable electricity demand has less or no positive impact on economic growth. Given the cost structure of the renewable electricity industry, growth is driven by demand growth in a related facility. It has a different growth mechanism when a country simply purchases equipment. If so, it is likely to have a positive impact on the economy of the country producing that facility. This means that if demand for renewable electricity increases, there is no induced effect like great value-added or additional demand in other industries or there is only an impact on the country's imports concerning the overall facility. Therefore, the relationship between renewable electricity demand and economic growth may not be evident in these countries.

There are four hypotheses on the energy–growth nexus considering the number of cases that could be the result of the analysis [12]. First, the "neutrality hypothesis" implies the absence of a causal relationship between energy consumption and GDP. The second one, the "conservation hypothesis," suggests that energy conservation policies may be implemented with little or no adverse effect on economic growth. In the third one, the "growth hypothesis" means that energy consumption is important for economic growth. The last one that implies bidirectional causality between the two factors is the "feedback hypothesis." It is possible to derive energy policy through an analysis of the above hypothesis. The energy conservation policy means reducing energy consumption for economic growth. If a specific country's energy–growth nexus follows the growth hypothesis, energy conservation policies may have an adverse impact on economic growth. However, if under the neutrality hypothesis, energy conservation policies may not have much impact. The main purpose of the energy–growth nexus study is to examine which hypothesis is investigated in a specific country or group of countries. The results of the energy–growth nexus study have yielded mixed results within hypotheses. Existing hypotheses are focused on energy itself and do not consider specific industries. Therefore, the existing hypothesis is not enough to cover the impact of renewable energy manufacturing.

Due to the feature originating from the renewable manufacturing industry, renewable electricity and economic growth will show a positive relationship. Thus, this study tries to fill the gap with a new hypothesis, which is the "renewable–growth hypothesis." We analyze the time-series data of countries to confirm the renewable–growth relationship. The main contributions of this study are as follows. First, this study investigates if the development of the renewable energy generation sector has a positive effect on economic growth considering the features of the renewable power sector. Second, a new perspective of renewable–growth hypothesis is proposed, and the national group supporting the analysis is established. This makes it possible to draw implications for the policy direction of fostering the renewable electricity industry. Furthermore, this study presents the issues to promote further studies.

The paper is organized as follows. Section 2 provides a review of the literature. Section 3 describes the countries to be analyzed and the data to be used for analysis and explains the model and methodology. Section 4 provides empirical results. Section 5 concludes the paper and provides remarks and policy implications.

#### **2. Literature Review**

#### *2.1. Overview of Energy–Growth Nexus*

The energy–growth nexus has been studied in order to confirm the direction of the causality and its hypothesis has been well developed via various studies (see Table 1). They can be divided into country-specific studies [13–16] and multi-country studies [17–19]. Some of the studies attempted to examine the causal relationship in the energy–growth nexus in both developed and developing countries or by using different period data [8–10,18,20–22]. A study can be conducted for a specific research purpose. Pao and Fu [23] tried to investigate the relationship between economic growth and energy consumption in Brazil. Contrary to the previous study, this study covered various types of energy. They found mixed results: A conservation hypothesis between non-renewable energy consumption and economic growth, a growth hypothesis between non-hydroelectric renewable energy consumption and economic growth, and a feedback hypothesis between economic growth and total renewable energy consumption. Others focused more on the causal relationship between economic growth and energy consumption [24,25]. Apergis and Danuletiu [24] examined the relationship between economic growth and renewable energy consumption for 80 countries using the long-run causality test. They concluded that there is a bidirectional causality between renewable energy consumption and economic growth in the long run. Kazar and Kazar [25] investigated the relationship between development and renewable electricity net generation values for 154 countries with a panel analysis. They found the presence of bidirectional causality in the short run and that the causal relationship differs both in the short run and long run depending on the human development level.

#### *2.2. Electricity and Economic Growth*

The energy–growth nexus has been studied in various countries and on a more disaggregated level [9]. In addition to the interest in climate change, many countries and policymakers have tried to implement an electricity conservation policy. The confirmed results of research have been used as a basis for implementing such a policy and to establish the right policy direction for the country. The electricity–growth nexus for a single country has been studied and developed for that reason. Ramcharran [26] studied the electricity–growth nexus in Jamaica and found that the country is energy dependent. Ghosh [27] and Narayan and Smyth [28] investigated energy consumption and economic growth in India and Australia, respectively. They found Granger causality from economic growth to electricity consumption in both countries. In addition, research is being conducted in various countries such as Korea [29], Bangladesh [30], Cyprus [31], China [32], Turkey [33], Malaysia [34], Lebanon [35], and Nigeria [36]. However, due to the omitted variable bias, the studies that use only energy consumption and economic growth as variables should be interpreted with caution. In that context, some studies have attempted to make econometric transformations by adding employment variables to the bivariate model [37] or setting up the model reflecting the structural beaks [38]. Lorde et al. [39] constructed a multivariate model using the neoclassical production model to examine the economic theory. In various studies, bivariate and multivariate models were analyzed, in order or simultaneously. The impact on economic growth has been studied for the implementation of national power-related policies until recently [40–43].

An electricity–growth nexus may be established by setting up a group of countries depending on the research purpose. Yoo [44] analyzed the relationship between electricity consumption and economic growth in ASEAN countries for similar characteristics in the electricity sector. This study tried to confirm the relationship between electricity consumption and economic growth within a

similar-featured group but with some differences in terms of investment in the power sector. Each of the four countries showed similar results with the two countries, respectively. The Organization of the Petroleum Exporting Countries (OPEC) is expected to be greatly influenced by energy conservation due to the characteristics of oil-producing countries. However, despite the obvious similarities, the results vary depending on the model and the time period [45,46]. This "no consensus" feature of the electricity–growth nexus was observed among the countries in the group even when the econometric method was modified. Acaravci and Ozturk [47] emphasized that they have results that conflict with the existing literature through the identification of the Granger causality of panel data. At the same time, they stated and emphasized that this issue deserves more attention and needs further research.

Some studies approach the characteristics of a country from an economic perspective. Apergis and Payne [48] analyzed a total of 88 countries using the panel vector error correction models based on the level of economic income. The study showed different results depending on whether a country was assessed in terms of the short run or the long run and also depending on its development level. In the long run, bidirectional causality exists in both high-income and upper-middle-income panels and lower-middle-income country panels. The growth hypothesis is satisfied in lower-middle-income country panels and low-income country panels in the short run. Even in the same group, mixed results were observed according to the short-run and long-run views [49], but some studies show a consensus between studies analyzing the same country [50].

Over time, research has been further disaggregated in various aspects, with national concerns shifting to energy transitions rather than just energy conservation. As a result, studies distinguishing renewable and non-renewable electricity from total electricity have begun [51–57]. Ibrahiem [51] mentions the limitations of the structure of the Egyptian electricity market, such as the crude-oil shortage, the need for renewable growth, or the transition on power mix. Apergis and Payne [52] conducted the panel causality test for the emerging market, including Egypt. Apergis and Payne [53] extend their work by examining the causal dynamics between renewable and non-renewable electricity consumption and economic growth in Central America. They show the negative bidirectional causality between renewable and non-renewable electricity consumption and conclude that the cause of this is that imported petroleum products raise concerns about the security of the region's energy supply. One of their contributions is their investigation of the negative bidirectional causality between renewable and non-renewable electricity consumption. They believe that this is due to imported petroleum products raising concerns about the security of the region's energy supply.

The analysis of renewable electricity is relevant to climate change and policies for global warming mitigation. Based on the results of the study, each country is recommended to increase its investment in renewable energy projects and vice versa (Table 1). According to the confirmed results of Al-mulali et al. [54], they recommended that Latin American countries should encourage not only the investment for renewable energy projects but also reduction in the role of non-renewable sources in electricity consumption. In recent years, these recommendations have been implemented in a way analysis of non-standard Granger causality [55–57]. In this way, it is possible to confirm the effect of specific energy sources [55] or conduct the analysis not only on aggregate electricity models but also those disaggregated into renewable and non-renewable models [57]. Despite the development of the research scope by the economic and econometric approach, the field of energy–growth nexus must still be investigated not only in terms of methodology but also in economics or policy.


 **1.**The previous study-related electricity growth nexus.

**Table**




Note: Y, EC, REC, NREC mean economic growth, electricity consumption, renewable electricity consumption, and non-renewable electricity consumption, respectively. EC → Y, growth hypothesis; Y → EC, conservation hypothesis; Y ⇔ EC, feedback hypothesis; Y == EC, neutrality hypothesis.

#### *2.3. Revisited Nexus Study*

This study tries to investigate the causal relationship between renewable electricity and economic growth. Given what previous studies have referred to as further studies [9,47,56], we "revisited" the nexus study. As mentioned in the previous "Literature Review" section, many studies refer to the existence of a consensus in the nexus field. Nowadays, to try to clarify the related points, several revisited studies are being conducted. Andrew and Bothwell [58] mentioned the limitation of using a bivariate or trivariate model within South Africa. They try to revisit the electricity–growth nexus by using a multivariate model considering the economic aspects of a country like export, employment, and consumer price index. Zortuk and Karacan [59] revisited the energy–growth nexus by selecting countries included in panel data. The ex-Soviet countries located in Central and Eastern Europe and the Caucasus are now in transition into free market economics. Therefore, analysis panel data containing the related countries can clarify the direction of the causal relationship. We also try to revisit the electricity–growth nexus in the same country as the previous study, but we will be using a different methodology and have a different purpose to them.

In this study, we have three distinctions from the previous study. First, the study's purpose is to confirm the renewable electricity–growth nexus by considering the characteristics of renewable energy. This feature is discussed in detail in the next section. Second, to reflect the characteristics that vary depending on the source, we built a model at the renewable source level. The differences in the characteristics of energy sources can affect the investigated findings [56]. There are cases in which previous studies have considered disaggregated level like sources [26,55]; however, only specific sources (steam, hydro, diesel, and gas turbine) were used for data availability and analytical purposes [26]. Halkos and Tzeremes [55] conducted their analysis using wind power, biomass, solar power, and geothermal data. However, the availability of the data used differed from country to country, and a non-parametric technique was applied. The last difference is that generation data is used for the application of the first and second implications mentioned above. The various differences between generation and consumption could affect interpretation in this study, which aimed at reflecting renewable characteristics rather than electricity [40,56]. This study is meaningful in that it analyzes through a revisiting flow in which the implication of the study is made by both the setting of the country and the variables.

#### **3. Data and Methodology**

#### *3.1. Renewable–Growth Hypothesis*

In this study, we analyze the renewable electricity and economic growth nexus considering renewable manufacturing industries. As mentioned above, the cost structure of renewable electricity is quite different from the conventional thermal power. In the renewable generation sector, much of the cost goes to equipment or facilities, which means a capital expenditure is much higher than an operational expense. On the other hand, the thermal power plants that use fossil fuels have substantial operational costs, although their capital cost is also considerable. Promoting renewable electricity could thus have a positive effect on economic growth through the production of renewable manufacturing industries such as wind turbine and solar panel manufacturing. In other words, renewable electricity can contribute to economic growth through industrial outputs as well as an energy input. We want to call this mechanism the 'renewable–growth' hypothesis in the context of the energy–growth nexus.

To test the renewable–growth hypothesis, we choose a country in which renewable electricity might lead to the development of related industries. We confirm that the renewable–growth nexus in a country includes companies that have developed in the renewable electricity sector. This study attempts to examine the positive relationship between economic growth and renewable electricity demand in such countries based on the market share of renewable manufacturing companies. At the moment, we are analyzing solar PV and wind power, which are expected to grow quickly [3]. We might expect to see a causal relationship by identifying the more disaggregated sectors.

Based on sales volume or revenue, we selected a global renewable energy company in solar [60] and wind [61] power. According to the collected data, we identified the country the company belongs to for analytical purposes. The analysis is conducted on solar and wind power, separately. Tables A1 and A2 show the companies and corresponding countries with the highest sales volume. For the solar PV, we have included the component in the plant and the market share of module companies. Four countries are analyzed for the solar PV: China (including Hong Kong), Canada, the USA, and Korea. A key component of the wind manufacturing industry is the turbine, which accounts for 60% of the total capital expenditure [62]. Thus, the standard for selecting the country included in the wind model is a global wind-turbine company [Table A2]. Wind models include six countries: China, the USA, Denmark, Germany, India, and Spain.

#### *3.2. Data and Estimation Procedure*

Consistent with the previous literature, we use real GDP as the dependent variable. Independent variables are electricity generation, gross fixed capital, and labor force. Due to the features of renewables, to transmission, to distribution losses, or even to theft, bias can exist in the investigated result [56]. This feature was also affected by country development level [40,63]. Therefore, we consider electricity generation data. Considering the connectivity to the power grid and intermittent characteristics of renewable energy, the detailed consumption data of each source is not available. To include this, the previous study used aggregate level consumption data. However, this study's purpose is to identify the effects of the nexus depending on the presence of manufacturing companies reflecting the manufacturing feature of each source. Reflecting on the characteristics of the energy source means it must be analyzed at a disaggregated level. Therefore, generation data is used for analysis, and solar PV and wind are configured separately.

We adopt the annual time series data of each country from 1980 to 2017. The data are obtained from the World Bank Development Indicators, World Energy Balances, and US Energy Information Administration and defined as follows: Real GDP (Y) in billion 2010 USD using purchasing power parity, fossil fuel electricity generation (NRE) defined in kilowatt hours, electricity generation from solar PV (RES) defined in kilowatt hours, electricity generation from wind (REW) defined in kilowatt hours, real gross fixed capital formation in current US dollars (K), and total labor force (L) in millions. We convert the unit of real gross fixed capital formation into constant 2010 US dollars using the GDP deflator from WDI. Additionally, all variables are converted to per capita and the natural logarithms are transformed.

#### *3.3. Estimation Strategy*

We assume that, due to the characteristics of renewables, renewable electricity generation and economic growth have a relationship. To confirm the causal relationship between them, we must check the stationarity of time series data. If data are not stationary, the estimated model might be a spurious regression and results may lack robustness. Common methods applied to unit root tests are the augmented Dickey–Fuller (ADF) [64], Phillips–Perron (PP) [65], and Kwiatkowski–Phillips–Schmidt–Shin (KPSS) [66]. If the null hypothesis of the ADF and PP tests are rejected, we conclude that time series data has a unit root. The KPSS test is applied to investigate the stationarity, and, if the null hypothesis is rejected, the data may be nonstationary time series data. The KPSS test is more suitable for testing small samples due to the lower lag truncation parameter, and it might complement the limitation of the ADF and PP tests [67]. Therefore, as a tool for cross-checking and for the robustness of results, we conduct ADF, PP, and KPSS tests together.

Before the Granger causality analysis, it is necessary to test for cointegration. There are several methodologies related to cointegration, and the Johansen test [68] is the most common method as it is more generally applicable than the Engle–Granger test [69]. Since this test is common and well known, we provide just a brief overview of this method. Johansen [68] modeled time series as a reduced rank regression, and we can trace test and maximum eigenvalue. The model is given as the following:

$$
\Delta Z\_t = \omega + \sum\_{i=1}^{q-1} \Delta Z\_{t-i} + \Pi \Delta Z\_{t-1} + \varepsilon\_t \tag{1}
$$

The *Z<sup>t</sup>* is a vector of p variables consisting of an (n × 1) column, and ω is an (n × 1) vector of constant terms. Γ means coefficient matrices, and ∆ is a difference operator. The ε*<sup>t</sup>* follows distribution as *N*(0, Σ). The Π means the coefficient is known as the impact matrix and contains information about the long-run relationships. By employing this method, it is possible to determine the number of the cointegrating vectors of the model.

We can also employ the cointegration test with the autoregressive distributed lag (ARDL) bounds test approach [70]. By using ARDL bounds testing, we can expect to take advantage of the following. First, it is possible to apply such a method irrespective of whether the regressors are I (0) or I (1). Furthermore, compared to the Johansen cointegration techniques, estimating with a smaller sample size is possible [71]. According to Narayan [72], the ARDL bounds test gives a reasonable critical value if the number of samples is between 30 and 80. Furthermore, the ARDL model can be derived in the form of an error correction model. Thi means it is possible to confirm that the long-run and short-run causality is the same as in VECM. In this study, we can establish the ARDL model as the following:

$$\begin{aligned} \Delta \text{GDP}\_{t} &= a\_{0} + \sum\_{i=1}^{q} \alpha\_{1} \Delta \text{GDP}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{2} \Delta \text{NRE}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{3} \Delta \text{RE}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{4} \Delta \text{K}\_{t-i} + \\\ \lambda\_{1} \text{GDP}\_{t-1} &+ \lambda\_{2} \text{NRE}\_{t-1} + \lambda\_{3} \text{RE}\_{t-1} + \lambda\_{4} \text{K}\_{t-1} + \mu\_{t} \end{aligned} \tag{2}$$

$$\begin{cases} \Delta \text{RE}\_{t} = \beta\_{0} + \sum\_{i=1}^{q} \beta\_{1} \Delta \text{RE}\_{t-i} + \sum\_{i=1}^{q} \beta\_{2} \Delta \text{NRE}\_{t-i} + \sum\_{i=1}^{q} \beta\_{3} \Delta \text{GDP}\_{t-i} + \sum\_{i=1}^{q} \beta\_{4} \Delta \text{K}\_{t-i} + \\\ \zeta\_{1} \text{RE}\_{t-1} + \zeta\_{2} \text{NRE}\_{t-1} + \zeta\_{3} \text{GDP}\_{t-1} + \zeta\_{4} \text{K}\_{t-1} + \mu\_{2t} \end{cases} \tag{3}$$

where GDP, NRE, RE, and K denote the logarithm form of real GDP, fossil fuel electricity generation, solar PV or wind electricity generation, and gross domestic capital formation, respectively. The parameters α, β are the short-run dynamic coefficient and λ, ζ are the corresponding long-run multipliers of each ARDL model. The µ represents the white noise error term.

To determine the existence of cointegration, we test the lagged value jointly by using the F test. However, as the F statistics from [70] are for several samples, we usually use the critical value from Narayan in the small sample size analysis [72]. The null hypotheses of each model are *H<sup>o</sup>* : λ<sup>1</sup> = λ<sup>2</sup> = λ<sup>3</sup> = 0 and *H<sup>o</sup>* : ζ<sup>1</sup> = ζ<sup>2</sup> = ζ<sup>3</sup> = 0. If the computed F statistics exceed the upper bound value, the null hypothesis is rejected. However, if the F statistics fall below the lower bound, we can conclude that we cannot reject the no cointegration hypothesis. However, if the value exists between the upper bound and lower bound, the results are inconclusive. Through the cointegration test, if we find the evidence for a long-run relationship between variables, we can conduct the test to check the existence of Granger causality.

According to the results from the cointegration test, there are two cases we must consider. If our variables are not cointegrated, we perform the test as a vector autoregressive (VAR) model in first differenced variable form. If we can confirm the existence of a long-run relationship, we can conduct the model with the error correction term. Thus, the model has a cointegration, and the Granger causality relationships are written as the vector error correction models (VECM) given below:

$$\begin{aligned} \Delta \text{GDP}\_{t} &= \alpha\_{0} + \sum\_{i=1}^{q} \alpha\_{1i} \Delta \text{GDP}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{2i} \Delta \text{NRE}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{3i} \Delta \text{RE}\_{t-i} + \sum\_{i=1}^{q} \alpha\_{4i} \Delta \text{K}\_{t-i} + \\ \lambda \text{ECT}\_{t-1} + \mu\_{1t} &\end{aligned} \tag{4}$$

$$\begin{aligned} \Delta RE\_{t} &= \beta\_{0} + \sum\_{i=1}^{q} \beta\_{1i} \Delta RE\_{t-i} + \sum\_{i=1}^{q} \beta\_{2i} \Delta NRE\_{t-i} + \sum\_{i=1}^{q} \beta\_{3i} \Delta GDP\_{t-i} + \sum\_{i=1}^{q} \beta\_{4i} \Delta K\_{t-i} + \\ \zeta \text{ECT}\_{t-1} &+ \mu\_{2t} \end{aligned} \tag{5}$$

The null hypothesis of the Granger causality from renewable electricity generation to GDP is *H<sup>o</sup>* : α<sup>1</sup> = α<sup>2</sup> = . . . α*<sup>q</sup>* = 0. To test the short-run causality of renewable electricity to GDP, we impose restrictions on all the lagged renewable electricity generation data using the F test. This is equivalent to the test of lagged first differences of the Granger causality from renewable electricity generation to GDP.

Whether or not they are integrated, the economic variables can be integrated into different orders. In that case, the Wald test will not have an asymptotic chi-square distribution, and VECM cannot be applied for the Granger causality test. Toda and Yamamoto [73] suggest the procedure for solving this problem. According to the standard stationary test, we can determine the order of the variables. Let m be the maximum order of integration, and l be the appropriate maximum lag length in VAR. Then take the preferred VAR model and add the m additional lags of each of the variables. We can test the Granger causality using that model. However, we must test the hypothesis with only the coefficients of the first l lagged values. Rejection of the null hypothesis implies the existence of Granger causality.

#### **4. Empirical Results**

#### *4.1. Unit Root Test Results*

Tables A3 and A4 present the result of a stationarity test for the solar PV case and the wind power case from ADF, PP, and KPSS test. We use Stata 14.0 for the analysis. The definition of GDP is real GDP, FOG is fossil-fuel electricity generation, RE is renewable electricity generation, and FXC is fixed capital formation. According to the model, RE is divided into solar PV (RES) and wind power (REW). In the process of examining stationarity, the maximum lag length was set to 4 considering the characteristics of the economic variables.

As mentioned above, ADF, PP, and KPSS tests were conducted simultaneously for robustness. If the differenced form does not reject the null hypothesis in more than two tests, the second differenced variable is tested. The first differenced variable is then analyzed based on the statistics considering the trend. Additionally, if we get the result that the first differenced is stationary in at least two tests, we do not write the statistics of the second differenced form in the table. Two or more tests indicate the same order of integration, and we set that value as the integration order of that variable. The results show that most variables in both models are in the form of I (1). Only the GDP variable of Spain in wind power has I (2). Since the order of integration of the variables is mixed, the analysis proceeds with Toda and Yamamoto's procedure.

Therefore, we constructed the VAR model with the variables to find the optimal lag length of the model setting the maximum lag as 4. Lag-order selection is based on Bayesian information criterion (BIC). In each model, the VAR model was constructed according to the optimal time difference obtained in each country, autoregression was checked, and the time difference was adjusted. The order of integration of variables and the optimal lag of the VAR model are summarized in Tables 2 and 3. Table 2 is for the solar PV, and Table 3 is for wind power. To investigate the Granger causality of each country and model, we reconstructed the VAR model with the sum of these two values as the lag length of the new model.


**Table 2.** Integration order and optimal lag results for solar photovoltaics (PV).

**Table 3.** Integration order and optimal lag results for wind power.


#### *4.2. Results of the Granger Causality*

Tables 4 and 5 summarize the results of the solar PV and wind power, respectively. The country shows that the growth hypothesis presented in the previous nexus study are Canada for the solar PV, and Germany, India, and Spain for wind power. China and the USA were included in both models, satisfying the conservative and feedback hypothesis, respectively. Denmark was the only country with a neutral hypothesis. The renewable–growth relationship is investigated in Canada, USA, and Korea

for the solar PV and the USA, Germany, India, and Spain for wind power. China in both models and Denmark in the wind power show different results based on our hypothesis.


**Table 4.** The results of the Granger causality in solar PV.

Note: \*, \*\*, and \*\*\* denote the rejection of the null hypothesis of no relationship at the 5% and 1% significance level, respectively.

**Table 5.** The results of the Granger causality wind power.


Note: \*, \*\*, and \*\*\* denote the rejection of the null hypothesis of no relationship at the 5% and 1% significance level, respectively.

In this study, the standard Granger causality test was also conducted to confirm the reliability of the results. The difference of the standard Granger causality is after the stationarity test, which means identifying the cointegration of the model. The Johansen method was used to confirm the cointegration, and the bounds test was also performed considering the small sample size. If cointegration exists, the VECM model is constructed to confirm the causality of the long run and short run. If the results indicate no cointegration in both tests, a VAR model with a differenced variable must be built, and the short run causality is checked using that model. The lag length of each model is the same as that of Toda and Yamamoto. Tables A5 and A6 summarize the standard Granger causality results. Only China, for the solar PV, shows a conflicting result in the cointegration test. The variables in this model are all I (1), so we follow the bounds test result due to the small sample size. Furthermore, because the lag lengths of Canada for solar PV and India for wind power are both 1, the short-run term disappears when constructing the VECM model. Therefore, both cases were analyzed only in the long-run relationship. However, due to the integration order of the variables, we used and analyzed the results using the Toda and Yamamoto procedure.

#### **5. Discussion**

The study aims to investigate the renewable–growth hypothesis of the countries with a renewable manufacturing industry. According to the results, the long-run relationship between renewable electricity generation and economic growth is present in all countries. However, in some countries, the results differ from our hypothesis. The results can be attributed to diverse factors like economic or socio-cultural points existing in different countries [45,47,52,74].

In China, the opposite of the renewable–growth hypothesis was found in both solar PV and wind power. China is the largest producer of both solar and wind components. The total installed solar PV was 175 GW in 2018, while the total for wind power was 185 GW in 2018 [75]. However, China's exports of manufactured goods to other countries are much higher than its domestic use [60,61]. Therefore, the causality test with renewable electricity generation in China may not reflect the positive effect of the renewable electricity industry. China's renewable electricity is also a reason for the ambiguous factor of the renewable–growth relationship. The renewable electricity industry induces economic growth in the indirect sectors, such as the development of the manufacturing sector and employment for facilities and the direct impact of the generation itself. Electricity curtailment due to grid issues is a chronic problem in China [76,77]. The resulting loss of generation has a negative effect on economic growth [60].

Denmark also has a similarity with China. In Denmark, renewable energy accounts for more than 50% of total electricity production [78]. However, the increase in installed wind energy capacity is the smallest among countries with wind power [75]. In the absence of increased domestic facilities, it is difficult to identify the growth hypothesis with given variables. Furthermore, Siemens Gamers has manufacturing facilities worldwide [79]. Vestas' manufacturing facilities are also located in other countries [80]. Thus, even if they satisfy the renewable–growth hypothesis, they may not be revealed under the given variables. In the rest of the target countries, we can examine the renewable–growth hypothesis.

The renewable electricity industry of countries showing a growth hypothesis has been more established in recent years. Germany, which shows the renewable–growth relationship between wind power and economic growth, had the largest wind power capacity in 2017. Germany leads the European Union (EU) in terms of power capacity with 56.1 GW, followed by Spain with 23.2 GW, which is also included in our analysis model. The manufacturing facility of Gamesa, which merged with Danish renewable company Siemens in 2016, still exists in Spain, and this could have a positive effect on the country. According to GWEC [81], one of the leading turbine suppliers for the EU is Siemens Gamesa. While installation numbers have slowed, the total capacity of India is 32.8 GW, taking the fourth position in the world's largest wind markets. Because the share of wind-generated electricity in India's total electricity consumption is still 4.35%, there is still growth potential in India's renewable electricity market [81]. The USA has the highest installed capacity after China among our target countries, both for solar PV and wind power. In 2018, the total installed capacity was 51 GW for solar PV and 94 GW for wind power [75]. In Korea, the installed solar PV capacity is 8 GW, but the number of related companies and employment is increasing due to recent aggressive renewable electricity policies. Canada has also installed a solar PV capacity of 3 GW, and while it is not growing much, it is considered to have high potential as the capacity continues to rise.

Under our hypothesis, the driving force of economic growth is the electricity industry itself, and other industries have induced effects from the increase of renewable generation. We used solar PV generation data and wind power generation data as variables to cover these points. However, these variables do not reflect the growth effect of the renewable manufacturing industry effectively in some countries. Development of relevant industries can lead to the growth of domestic products usage. So we set renewable electricity data as a variable in our model. This is a limitation of the study in that it does not consider the characteristics of trade. In the same line, the diverse factors that are inherent in countries need to be considered. The various political, economic, and socio-cultural factors could cause a bias in the nexus study. Therefore, further research to secure the renewable–growth hypothesis

is necessary. Developing variables to consider the export-focused characteristic and renewable energy policy of each country could also be the next step of the renewable–growth hypothesis. Furthermore, the existence of renewable growth may not be revealed in the results due to bias arising from the small sample size. It is necessary to overcome this limitation in the future with continuous research on specific energy sources. A country's characteristics may vary and be embodied in the results, so it also becomes the motivation for additional research. Lastly, our study is the first to present the renewable–growth hypothesis. However, it must be noted that while we use an analysis strategy by using Granger causality, it is unknown whether the relationship is positive or negative [82]. A more diversified approach could make it clear whether the hypothesis is correct or not.

#### **6. Conclusions**

The increasing concerns about the climate crisis cause trends such as energy transition from fossil fuels to renewable energy. According to the change in the power mix, the renewable electricity industry is affected both directly and indirectly. The difference between conventional electricity and renewable electricity sources lies in the weight of manufacturing-related parts. This is the case with solar PV and wind power, both comprising the world's highest growth in renewable electricity installation. Due to such characteristics, the increase in renewable energy generation has the additional effect of not only leading to advantages in the generation industry but also the development of related industries and increase in employment. As this will have a positive effect on economic growth, we have thus formulated the renewable–growth hypothesis, which is one step further from the growth hypothesis of the existing nexus studies.

In this study, we tried to investigate the relationship between renewable electricity and economic growth. We constructed and analyzed more disaggregated models like solar PV and wind power. Additionally, we established the target group based on global companies in manufacturing. It was expected that the renewable–growth relationship would be represented more accurately in those countries. The variables within the model are real GDP, fossil fuel electricity generation, renewable electricity generation, and fixed capital formation. Renewable electricity generation means the solar PV data for solar PV and wind power data in wind power. Because our purpose is to confirm the renewable–growth hypothesis, we used electricity generation as a variable, considering the difference between generation and consumption data.

The analysis showed that the renewable–growth hypothesis was satisfied except only China for solar PV and China and Denmark for wind power. These results are due to the characteristics of the country's power and power-related manufacturing industries. We can thus conclude that national and industry characteristics can influence the results of the analysis. Furthermore, we can provide the direction for further studies. First, renewable electricity time series data has a relatively small sample size. Therefore, if data is acquired over time, it is necessary to analyze a larger sample size. Doing this would determine the renewable–growth relationship more accurately. Second, considering additional variables to reflect the characteristics of industry or country would be necessary. However, as we can see from the results of other countries, the parameters were appropriate for verifying the hypothesis. Thus, while considering the general case, further research is necessary to reflect the characteristics of each country. The study's main contributions are its presentation of a new perspective in the form of the renewable–growth hypothesis and the establishment and analysis of a target group that reflects the characteristics of the renewable electricity industry.

**Author Contributions:** Conceptualization, J.K. and M.Y.; Methodology, J.K. and M.Y.; Validation, J.K.; Formal Analysis, M.Y.; Investigation, J.K. and M.Y.; Data Curation, M.Y.; Writing—Original Draft Preparation, M.Y.; Writing—Review & Editing, J.K.; Supervision, J.K.; Funding Acquisition, J.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the Human Resources Development program (No. 20194010201860) of the Korea Institute of Energy Technology Evaluation and Planning (KETEP) grant funded by the Korea government Ministry of Trade, Industry and Energy.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **Appendix A**


**Table A1.** The highest production company in solar PV (module).

**Table A2.** The market share of global wind power companies.


#### **Appendix B**

**Table A3.** Unit root test results for solar PV.



**Table A3.** *Cont.*

Note: \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, and 1% for ADF, PP. ', \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, 2.5%, and 1% for KPSS. <sup>1</sup> first difference form using trend when testing ADF and PP. <sup>2</sup> the result of second difference form is stationary.


**Table A4.** Unit root test results for wind power.


**Table A4.** *Cont.*

Note: \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, and 1% for ADF, PP. ', \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, 2.5%, and 1% for KPSS. <sup>1</sup> first difference form using trend when testing ADF and PP. <sup>2</sup> : the result of second difference form is stationary.

### **Appendix C**


**Table A5.** The results from the standard Granger causality test in solar PV.

Note: ', \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, 2.5%, and 1% for bound-testing. \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, and 1% for Granger causality. <sup>1</sup> the lag of this model is 1, so we can investigate the long-run relationship result only. <sup>2</sup> according to the bounds test result, we investigate the short-run relationship using only VAR.

**Table A6.** The results from the standard Granger causality test in wind power.


Note: ', \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, 2.5%, and 1% for Bound-testing. \*, \*\*, and \*\*\* indicate the level of significance at 10%, 5%, and 1% for Granger causality. <sup>1</sup> the lag of this model is 1, so we can investigate the long-run relationship result only. <sup>2</sup> this model does not have cointegration, so we can investigate the short-run relationship only. <sup>3</sup> while this model includes the I (2) variable, we applied the bounds test for reference.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Strategic Planning of O**ff**shore Wind Farms in Greece**

#### **Sofia Spyridonidou <sup>1</sup> , Dimitra G. Vagiona 1,\* and Eva Loukogeorgaki <sup>2</sup>**


Received: 9 January 2020; Accepted: 23 January 2020; Published: 26 January 2020

**Abstract:** In the present article, a new methodological framework for the efficient and sustainable exploitation of offshore wind potential was developed. The proposed integrated strategic plan was implemented for the first time at national spatial planning scale in Greece. The methodological approach is performed through geographical information systems (GIS) and Microsoft Project Server Software and includes five distinct stages: (i) definition of vision/mission, (ii) identification of appropriate areas for offshore wind farms' (OWFs) siting, (iii) determination of the OWFs' layout, (iv) calculation of the OWFs' (projects) total investment cost and, finally, (v) portfolio analysis. The final outcome of the proposed strategic planning is the prioritization of the proposed sixteen offshore wind projects based on their strategic value, as well as the estimation of the overall investment cost of the entire portfolio. High economic, socio-political and environmental benefits could be achieved through the implementation of only 60% of the total investment capital of the proposed strategic plan.

**Keywords:** strategic planning; site selection process; offshore wind farms; geographic information systems; portfolio analysis; Greece

### **1. Introduction**

In recent years, there has been a growing interest towards the installation of OWFs, due to the existence of multiple benefits related to the siting and operation of wind turbines offshore, such as existence of stronger winds of longer duration, availability of extensive free space for the construction of large-scale projects, reduction, and/or avoidance of noise and visual disturbances caused to the landscape by these structures, etc. Following the installation of the first OWF in Denmark in the early 1990s, a significant increase of the offshore wind industry was noted in the first decade of 2000, with the overall capacity doubling every 2–4 years [1].

On a global scale, according to statistical figures from Global Wind in 2014, over 90% of all offshore wind installations were implemented in European waters. Offshore wind energy in Europe reached the record figure of 3148 megawatt (MW) of total installed capacity in 2017, which corresponds to 560 new offshore wind turbines and 17 OWFs [1,2]. This particular record is two times higher than the figures of 2016 and 4% higher than the previous record of 2015 [2]. In the following year (2018), 409 new offshore wind turbines connected to the electricity grid across 18 offshore wind projects in Europe [3]. The referred amount corresponds to 2649 MW of net additional capacity, which is 15.8% lower than in 2017 [3]. Thus, Europe's cumulative offshore wind capacity reached 18,499 MW at the end of 2018, which corresponds to a total of 4543 grid-connected wind turbines across 11 European countries [3]. At present, the UK has the largest amount of offshore wind capacity in Europe with 44% of all installations in MW, followed by Germany with 34%, Denmark (7%), Belgium (6.4%) and the Netherlands (6%) [3]. Finally, in recent years focus was given on the development of OWFs in deeper waters [4–7], where floating support structures are preferable. For example, useful methodologies

have been developed for determining with accuracy all the relevant economic decision variables of floating OWFs [6], for proposing the best technological alternatives [5] and analyzing future wind resources in deeper waters [7].

An important process of the installation of OWFs is the determination of areas suitable for the deployment of the offshore energy systems. The site selection for utilizing OWFs corresponds to a multidimensional decision-making issue. Although several applications in various spatial planning scales can be found in the literature [8–24], the applications of OWF siting at national scale are really handful. Only five out of seventeen [8,10,14,17,19], refer to site suitability analyses on a national spatial planning scale. More specifically, [8] applied five factors and seven constraints using multi-criteria decision making and GIS models to provide a suitability map for offshore wind energy in Egypt. The analysis was conducted at large scale covering the whole of Egypt and its surrounding waters. The large-scale potential of China's offshore wind energy from the perspective of current technical, spatial, and economic constraints and its possible contributions to the nation's energy system was investigated in [10]. With the aid of a GIS-based tool, offshore wind potential was evaluated as a combination of wind resources, technical projections of wind turbines, economic costs and spatial constraints of offshore wind farms. Location-specific levelized production cost and cost supply curves of offshore wind energy were also developed. A two-step decision-making procedure was adopted in [14] to evaluate the locations for offshore wind farms in Greece. Unsuitable locations were initially rejected using GIS and three constraint factors, while the remaining sites were evaluated with the AHP method and five evaluation criteria. A spatial model for the assessment of offshore wind energy potential, production costs, and the identification of suitable areas based on GIS was presented in [17] and applied in a part of the Danish Exclusive Economic Zone (EEZ). Finally, a multi-criteria site selection analysis was performed by considering technical, social, and civil restrictions for finding the most suitable offshore wind farm locations in Turkey among the 55 coastal regions, including their technical power capacities [19].

Considering the tools and techniques that have been applied so far in the OWF siting literature, several researchers have applied multi-criteria techniques to rank OWF siting alternatives (e.g., [8,12–14]), indicating that multi-criteria decision analysis plays a crucial role in OWF siting. GIS presents also an important tool for the identification and the selection of suitable sites for the installation of wind farms either on land or in the marine environment [25]. It has been used globally in many countries for the site selection of offshore wind projects, such as Egypt [8], Ohio-USA [9], China [10], Greece [11–14], South Korea [15], Spain [16], Thailand [26], etc.

What is missing from the current literature and practice is the development of an integrated strategic plan for the efficient and sustainable exploitation of the offshore wind potential and the relevant deployment of OWFs. In the present paper a new Strategic Planning methodology to identify and prioritize suitable areas for offshore wind sites is introduced, which addresses a gap in knowledge in the offshore wind energy field. In order to do this, this work utilizes a countrywide case study (Greece) where the developed methodology is applied. It should be noted that no OWFs have been developed so far in Greece.

The methodology proposed and applied includes the implementation of five distinct stages corresponding to: (i) the definition of vision and mission of the strategic planning, as previously mentioned, (ii) the identification of appropriate areas for OWFs' siting based on specific exclusion criteria, (iii) the determination of the OWFs' layout in the aforementioned areas, (iv) the calculation of the OWFs' (projects) total investment cost by calculating the capital expenditure (CAPEX), the operating expenses (OPEX) and the decommissioning expenses (DECEX) and, finally, (v) a portfolio analysis based on seven assessment criteria.

A critical advantage of the proposed methodology is that it addresses existing gaps on renewable energy sources (RES) siting issues, by: (i) introducing a holistic, step-by-step, OWF siting methodology, which considers all the relevant critical issues that an OWF developer or/and an energy planner should analyze and resolve, (ii) providing a long-term planning approach (25 years, after the plan's

implementation), and (iii) recognizing a multi-disciplinary approach, as it considers legal, technical, economic, environmental, societal, and political issues. The final outcome is the assessment and ranking of all OWF project proposals considering their strategic value and cost constraints. The proposed methodology can be easily applied in other regions by following the abovementioned five stages. The novelty of the paper lies both on the integrated methodology itself (strategic spatial planning) and on the tools and criteria used in the analysis.

More specifically, GIS is used in a twofold way in the present paper: (i) for identifying the most suitable areas (SAs) for OWFs in Greece and (ii) for determining the layout and for the first time the precise location coordinates of the wind turbines in each OWF (OWFs' mapping and micro-siting determination) and, therefore, the energy capacity of the projects. The estimation of the precise location coordinates of the wind turbines in an OWF is a critical issue in the planning phase and should be accomplished before the construction phase of such large-scale projects. Moreover, the present paper proposes for the first time a prioritization of OWF projects (and, thus, an identification of optimum sites for OWFs' installation) through portfolio analysis. Portfolio analysis includes a collection of projects/proposals that will compete for selection based on their cost relative to their strategic value. The Microsoft Project Server tool, which integrates multi-criteria evaluation techniques and mathematical optimization, is used for the first time on the subject of OWFs' siting, in order to perform the required portfolio analysis and strategic scenarios in relation to the country' energy needs.

Finally, a total of twenty (exclusion and assessment) criteria are employed in the proposed methodology and the present site suitability support framework almost fully covers the economic, social, political, technical and environmental dimensions of the OWFs' siting problem in a national spatial planning scale. The proposed methodology deploys a number of criteria and restrictions of previous studies (e.g., wind velocity, water depth, distance from protected areas), while it introduces innovative criteria in relation to OWF siting issues. Exclusion criteria such as seismic hazard zones, landscape protection/visual and acoustic disturbance as well as assessment criteria (AC) such as electrical energy demand and distance from military exercise areas (firing fields and exercise locations) are applied for the first time at national planning scale.

This paper is structured in eight sections. Section 2 briefly presents the proposed methodology. Section 3 defines the exclusion criteria and the relevant exclusion zones and presents the sources and the processing method of the required spatial data. In Section 4, the main technical specifications of offshore wind turbines (e.g., most suitable support structure definition) and their layout characteristics considered in this study are presented. Section 5 describes the method applied for estimating CAPEX, OPEX, and DECEX during the life cycle of the projects and, thus, the total investment costs of the portfolio projects. Section 6 includes a description of the criteria used for assessing SA for OWF siting and of the portfolio analysis, while in Section 7 the results of all stages of the methodological framework are presented and discussed. Finally, in Section 8 the conclusions of the present study are cited.

#### **2. Materials and Methods**

In order to identify the most appropriate, sustainable, technically and economically viable solutions to site offshore wind projects in Greece the strategic planning methodology shown in Figure 1 is developed and applied in the present paper. The proposed methodological framework consists of five stages, which are analyzed below.

**Figure 1.** Proposed strategic planning methodology for offshore wind farms (OWFs') siting in Greece.

#### *Stage 1 (Stg1)—Vision and Mission of Strategic Planning*

In the first stage, the vision and mission of the strategic planning are defined, on which the next four stages are based. This stage is approached through a combination of proactive and empirical strategy [27]. It takes into account the current situation in the examined country, regarding the issue of energy independence, as well as the future demand for the production of a large number of public commodities, such as electricity, with the ultimate goal to export the latter and, thus, improve the country's current economic status. At this point, it is worth to mention that in Greece by the end of June 2018, the total wind capacity, generated only by onshore wind turbines, was 2690.5 MW, representing an increase of only 1.5% or 39.2 MW compared to the end of 2017 [28]. Thus, the abundant wind potential existing in the Greek marine environment remains still unexploited. In addition, the national effort to reduce GHG emissions is focused on the energy sector. The policy plan for Greece 'National Energy Plan: Roadmap to 2050' was posted by the Ministry for Energy, Environment and Climate Change in 2012 [29]. The roadmap aimed at a reduction of 60% to 70% of CO2 emissions from the energy sector by 2050 compared to 2005, with 85%–100% of electricity coming from RES [29].

#### *Stage 2 (Stg2)—Exclusion of Unsuitable Areas*

This particular stage is based on the use of the GIS mapping tool. It includes the exclusion of the areas deemed unsuitable for the siting of OWFs, through the application of various exclusion criteria, resulting to the definition of SA on a national level. The exclusion criteria are defined based on the special characteristics of the examined area, considering also the relevant provisions of the Greek Specific Framework for the Spatial Planning and Sustainable Development for the Renewable Energy Sources (SFSPSD-RES) [30].

#### *Stage 3 (Stg3)—Determination of Technical Specifications and Layout*

This stage deals with technical issues related to such projects, such as the selection of the model type of the wind turbine (rotor-nacelle-assembly), the selection of most suitable type of support structure, etc. The required technical specifications are determined by the following elements: (i) the specific characteristics of the suitable sites, that is, wind velocity, wind direction, water depth and the available surface area/shape of the proposed sites, and, also, (ii) studying similar projects that have been completed and are in full or partial operation to this day [31]. In addition, using GIS, the OWFs are sited within the SA identified in Stage 2.

#### *Stage 4 (Stg4)—Costing of OWFs*

This stage includes the estimation of CAPEX, OPEX and DECEX of all proposed projects, considering the available data of similar completed projects that are still in operation [31].

#### *Stage 5 (Stg5)—Assessment of SA and Portfolio Analysis*

In the final stage of the proposed methodology, the project portfolio is created for the strategic planning of OWFs in Greece and its analysis is carried out. The portfolio decision analysis approach combines multi-criteria evaluation and mathematical optimization and is characterized by the following elements: (i) formation of one portfolio of project proposals (basic goal), taking into account multiple objectives, interactions, and resource constraints; (ii) capturing of the decision makers' preferences regarding the objectives by utilizing a multi-attribute value function; (iii) implementation of integer optimization to obtain the feasible portfolio with the greatest overall strategic value and (iv) potential implementation of interactive "what-if" analyses to examine how the optimal portfolio of actions changes in response to changes in the model parameters or constraints [32]. The objective of this particular stage is to calculate the strategic value of the proposed projects based on specific AC and to prioritize the implementation of the portfolio projects. AC arise mainly from the special characteristics of the SA and their prioritization is achieved with the contribution of an expert group (EG) on issues related to the siting of RES and, more specifically, OWFs, through a questionnaire survey. The portfolio analysis is carried out using Microsoft Project Portfolio Server software. Various energy policy scenarios are formulated for the country based on economic restrictions.

#### **3. Exclusion Criteria and Data Collection**/**Digitization**

#### *3.1. Exclusion Criteria*

The study area is defined by the EEZ of Greece and any area outside of it, is legally excluded. The EEZ of a country or otherwise the National Territorial Waters, particularly in recent years, is formally taken into consideration, as a siting criterion, [16,20,33–35]. The exclusion criteria considered in this paper are discussed below.

#### Wind Velocity

Wind velocity is a significant criterion for the site selection of an OWF, as it is directly linked to the economic feasibility of the project. Therefore, an accurate and detailed analysis of wind data is crucial for a potential wind energy assessment of the proposed suitable sites. In this study, wind velocity data are provided based on measurements made at the height of 80 m on an hourly basis and includes measurements for 10 years (2009–2018). In the present site suitability analysis, marine areas, where annual average wind velocity is smaller than 6 m/s at a height of 80 m above the mean water level, are considered unsuitable for the siting of OWFs [16].

#### Water Depth

Water depth is one of the key criteria for OWFs' siting, as it significantly contributes to the determination of the investment cost of such projects [1]. Specifically, the water depth affects the selection of the wind turbine's support structure, as well as the CAPEX and OPEX of an OWF project, which increase significantly in deeper waters. For example, according to [36], it can be assumed that with water depth the costs increase due to mooring, anchoring, and cabling costs in deeper waters. In the present investigation, the maximum limit of water depth is set to 500 m [16,36,37].

#### Military Zones

These marine areas are officially used by the National Army either for training purposes or as firing fields and therefore cannot be considered for any other use. The present criterion is taken into consideration by [8,15,33,34,37].

#### Seismic Hazard Zones

The seismic hazard factor should be considered generally in the site selection process to reduce construction cost. Greece corresponds to one of the most seismically active countries worldwide. Therefore, all infrastructures should be adequately designed against earthquake. In the case of OWFs, this fact may lead to special designs of the wind turbines' support structure and, therefore, to larger construction costs. Thus, the areas belonging to the Seismic Hazard Zone III (0.36 g) in Greece are excluded. The present criterion has not been considered so far in any other study of OWFs' siting internationally, whereas it has been proposed as a criterion for selecting sites suitable for OWF developments in the South Korea by [38], but it was not considered as a site selection criterion in their study.

#### Underwater Cables

This exclusion criterion refers to the cables that already exist on the seafloor and serve either for electricity transmission or for telecommunication purposes (e.g., [8,12,33,34]). It is important to consider the underwater routes of those cables, in order to avoid any damage to them during the installation process of OWF developments.

#### Distance from Ports

The distance of an OWF project from a port presents an important factor affecting the total investment cost, since it has a direct impact on the installation costs, the operation and maintenance costs, as well as the decommission costs of the OWF [39]. Specifically, the total investment cost decreases as the location of an OWF is closer to an existing port, while, moreover, the proximity of the installation area to a port simplifies the overall project management (e.g., no need to install a substation within the marine environment). This criterion has been considered in the site selection of hybrid offshore wind and wave energy systems in Greece [37]. The selected limit of the distance from a domestic port is set in the present paper at 100 km and marine areas that are further away from 100 km are excluded.

#### Distance from High Voltage Electricity Grid

The distance of an OWF from the national electricity grid and particularly from a high voltage grid is extremely important for technical and economic reasons. A connection to the high voltage grid is selected, because in the opposite case (connection to a medium or low voltage grid) there might be a serious risk of cable destruction due to overloading of the electricity grid [20,26,36]. There are studies that set the distance of the candidate siting areas from the electricity grid at a limit of 200 km [8,39], while there are others that reduce this limit to 60 km [20] or even to 40 km [26]. In the present paper, the limit of 100 km from the existing and the potential officially approved high voltage electricity grid was selected. Moreover, one of the most important factors for the development of OWFs is the evaluation of capacity of the grid. In Greece, the Independent Power Transmission Operator (IPTO) S.A. undertakes the role of transmission system operator for the Hellenic Electricity Transmission System (medium and high voltage grid). In 2018, IPTO published an approved future plan for the spatial development of medium and high voltage grids in Greece as a target for 2027 [40]. In this plan, the majority of the islands that are both close and far away of the mainland lack medium and high voltage grids. Therefore, it is impossible and economically not viable to find locations for the development of OWFs in a long distance from the mainland in the near future. A more detailed analysis of the capacity of grid could be useful for the OWFs' development, but this analysis is out of the scope of the present study.

#### Landscape Protection/Visual and Acoustic Disturbance

The present criterion is related to the distance of an OWF from the coast and it has been used to ensure landscape protection, avoid visual and acoustic disturbances, and ensure the social acceptance of an OWF [15,17]. In the present study, marine areas with a distance from the coast smaller than 20 km are considered unsuitable for OWFs' siting and are excluded from further analysis. This limit is defined based on [36]. Moreover, the 20 km ensure a distance of at least 130 times of the total height of the selected offshore wind turbine, in order to avoid the visual and acoustic impacts of the project.

#### Distance from Shipping Routes

The existence of safe navigation routes that connect the plethora of Greek islands with the mainland is an extremely important issue. In order to ensure the protection of shipping movement either for trade or tourism, a safety distance of approximately 5 km (3 miles) from the referred routes is selected [9,13].

#### Distance from Marine Protected Areas

In this paper, marine protected areas correspond to Sites of Community Importance (SCI) of Natura 2000, national marine environmental parks, coastal bathing waters monitored and assessed in the framework of the Monitoring Programme of Bathing Water Quality according to the provisions of the Directive 2006/7/EC and swimming beaches awarded with the Blue Flag. In this paper, the minimum distance from marine protected areas is selected equal to 2 km, as according to previous studies [11,13,17] the relevant distance limit is set at 1–2 km.

#### Distance from Wildlife Refugees and Migration Corridors

This criterion includes migration corridors and wetlands of international importance, as defined according to the Ramsar Convention. The specific criterion is considered in order to reduce the potential risk of birds' collision on the wind turbines, mainly during the migratory period. The installation of OWFs should be avoided within the boundaries of the referred areas which are hosting a variety of birds. An exclusion zone of 3 km is taken into account [10,13].

#### Distance from Residential Network

*3.2. Data Collection/Digitization* 

an official map found in [40].

[31].

*4.1. Definition of Wind Turbine Model* 

database of the official European website "EMODnet" [45].

**4. Technical Specifications and OWF Siting Layout** 

*4.2. Selection of Wind Turbines' Support Atructure* 

According to the national legislative framework (SFSDSP-RES) [30] minimum distances from residential settlements and from traditional settlements equal to 1 km and 1.5 km respectively are taken into consideration.

Based on all the above, Table 1 summarizes the exclusion criteria considered in the present paper and their incompatibility zones.


#### **Table 1.** Exclusion criteria and incompatibility zones.

In order to identify and analyze all the environmental, economic, technical, legal, and political characteristics of the EEZ of Greece, it was essential to collect and appropriately digitize, if necessary, certain geographical information data from national institutes, research centers, services, and official international and national websites that provide officially approved cartographic data. More specifically, the digital data used in the present study in correspondence with the responsible entity/source are as follows: (i) Water depth data obtained from the Hellenic Navy Hydrographic Service [41]. (ii) Wind velocity data provided by the Hellenic Centre for Marine Research [42]. (iii) Data of the EEZ of Greece, the Mediterranean Sea and Greece gathered from the electronic database of the European Statistical Service [43]. (iv) Data of SCI, national marine environmental parks, coastal bathing waters, swimming beaches and wetlands of international importance obtained from the "GEODATA" official national website, which has been characterized as the national gate of geographical information data of Greece [44]. (v) Data of the underwater telecommunication cables within the EEZ of Greece, which were collected from the electronic

Except of the above, the following data were identified, collected, and mapped: (i) The verified shipping routes of the whole country were digitized through the basemaps of the cartographic tool ArcGIS, using the same projected coordinate system. (ii) The military zones used for training purposes and as firing fields, which were provided in analog format by the Hellenic Navy Hydrographic Service [41] and they were, then, appropriately digitized. (iii) The migration corridors, which were mapped by obtaining a corresponding map (in image format) from the Hellenic Ornithological Society [46]. (iv) The domestic ports, which were mapped by providing information of their locations and their names [47]. Only the officially designated ports of the country were mapped. (v) The seismic hazard zones of the country, which were digitized through the official seismic hazard map, collected as an image from the Technical Chamber of Greece [48]. (vi) The data related to the underwater cables of the electricity grid, the locations of the 400 kilovolt (kV) high voltage centers and the 150 kV high voltage substations, obtained from the IPTO, through

In the present study, the generic 5 MW turbine, which was developed by the National Renewable Energy Laboratory (NREL) [49] is selected. This wind turbine model has been also used in several previous studies [39,50–52], while a large number of existing and fully or partially operational OWFs globally, deploy offshore wind turbines with the same nominal power (ten in Europe and six in Asia)

settlements)

#### *3.2. Data Collection*/*Digitization*

In order to identify and analyze all the environmental, economic, technical, legal, and political characteristics of the EEZ of Greece, it was essential to collect and appropriately digitize, if necessary, certain geographical information data from national institutes, research centers, services, and official international and national websites that provide officially approved cartographic data.

More specifically, the digital data used in the present study in correspondence with the responsible entity/source are as follows: (i) Water depth data obtained from the Hellenic Navy Hydrographic Service [41]. (ii) Wind velocity data provided by the Hellenic Centre for Marine Research [42]. (iii) Data of the EEZ of Greece, the Mediterranean Sea and Greece gathered from the electronic database of the European Statistical Service [43]. (iv) Data of SCI, national marine environmental parks, coastal bathing waters, swimming beaches and wetlands of international importance obtained from the "GEODATA" official national website, which has been characterized as the national gate of geographical information data of Greece [44]. (v) Data of the underwater telecommunication cables within the EEZ of Greece, which were collected from the electronic database of the official European website "EMODnet" [45].

Except of the above, the following data were identified, collected, and mapped: (i) The verified shipping routes of the whole country were digitized through the basemaps of the cartographic tool ArcGIS, using the same projected coordinate system. (ii) The military zones used for training purposes and as firing fields, which were provided in analog format by the Hellenic Navy Hydrographic Service [41] and they were, then, appropriately digitized. (iii) The migration corridors, which were mapped by obtaining a corresponding map (in image format) from the Hellenic Ornithological Society [46]. (iv) The domestic ports, which were mapped by providing information of their locations and their names [47]. Only the officially designated ports of the country were mapped. (v) The seismic hazard zones of the country, which were digitized through the official seismic hazard map, collected as an image from the Technical Chamber of Greece [48]. (vi) The data related to the underwater cables of the electricity grid, the locations of the 400 kilovolt (kV) high voltage centers and the 150 kV high voltage substations, obtained from the IPTO, through an official map found in [40].

#### **4. Technical Specifications and OWF Siting Layout**

#### *4.1. Definition of Wind Turbine Model*

In the present study, the generic 5 MW turbine, which was developed by the National Renewable Energy Laboratory (NREL) [49] is selected. This wind turbine model has been also used in several previous studies [39,50–52], while a large number of existing and fully or partially operational OWFs globally, deploy offshore wind turbines with the same nominal power (ten in Europe and six in Asia) [31].

#### *4.2. Selection of Wind Turbines' Support Atructure*

The selection of the wind turbines' support structure is related to the water depth of the SA (Stg2, Figure 1). These areas are located at a water depth of over 50 m (see Section 7.2 below) and, therefore, floating platforms as support structures are preferable [16,39,51]. In the present paper, for water depths of 50–200 m, the Tension Leg Buoy (TLB) is selected as the floating support structure of each wind turbine. TLB systems can serve as sustainable solutions for the exploitation of RES [39,53]. In the North Sea, where harsher wind and wave conditions compared to the Greek marine environment exist, it has been demonstrated that TLBs present the most cost-effective systems for water depths from 50 to 200 m [52]. As for water depths of 200 to 500 m, the Hywind concept is selected. The simple design of Hywind offers many important advantages, such as evidence-based technology, simple support structure construction with the possibility of standardization and lower fabrication cost, as well as robustness and suitability in case of harsh environmental conditions [54]. The Hywind floating system, although slightly more expensive than the SWAY system in terms of cost per megawatt hour (MWh) [39], corresponds to a floating platform that has been used with success globally [54,55].

#### *4.3. Definition of the Wind Turbines' Layout Within the Suitable Areas*

The layout of the wind turbines depends on various factors (e.g., cost, wake effects, etc.) and its determination corresponds to an optimization problem [56,57]. In this paper, an oriented approach that fulfills the aim of the present investigation is developed which, moreover, takes into consideration the European best practices. Specifically, the distance between two successive turbines at a line parallel and perpendicular to the prevailing wind direction is denoted as dx and dy respectively, while Drotor denotes the rotor diameter. In existing, fully-commissioned European OWFs (e.g., Nysted OWF Denmark, Eneco Luchterduinen OWF Netherlands, Kentish Flats OWF United Kingdom, Belwind OWF Belgium), the values of dx are between 4.6~12.1Drotor, while the corresponding range for dy is 3.2~8Drotor [58]. In addition, according to [59] NREL recommends values of dy correspond to 5~10Drotor. In the present study, the wind turbines' layout is determined with the use of 'Advanced Editor Tools' in ArcGIS and according to the following elements: (i) main wind direction of each specific site, (ii) Drotor of the selected wind turbine model, (iii) shape of each OWF site and, also, (iv) the referred European standards, in order to minimize the array losses in the proposed OWF projects. Considering all the above, in this study, the defined dx and dy values for the wind turbines layout are 7Drotor and 7Drotor respectively.

#### **5. Cost of OWFs**

#### *5.1. Estimation of CAPEX*

The components of CAPEX considered in the present study are: (i) development and consenting, (ii) construction phase insurance, (iii) rotor-nacelle-assembly costs, (iv) production costs (including tower and support structure), (v) mooring costs (including installation for the case of floating wind turbines), (vi) grid costs (including installation), and (vii) installation of the whole wind turbine system [39,51]. According to [16,60–62], CAPEX is estimated taking into account the water depth and the distance from the shore. In addition to these factors, the CAPEX of such projects depends upon: (i) the wind turbine support structure deployed, which can be either fixed to the sea bed or floating [16,39,51,52,60] and (ii) the nominal power of the wind turbine.

#### *5.2. Estimation of OPEX*

OPEX of OWFs are a major part of the total investment costs and they are directly linked and largely affected by the distance from ports and by the water depth [39,51]. In addition, according to [16,39,51], OPEX per year for floating support structures correspond to approximately 3% of CAPEX (€/MW). In [39,51] OPEX was considered equal to 3.7% of CAPEX (€/MW) for wind turbines with a TLB floating platform and equal to 3.44% of CAPEX (€/MW) for wind turbines with a Hywind floating platform. After calculating OPEX for all OWFs for their first year of operation, OPEX during their total life cycle, which is defined equal to 25 years [62], can be estimated. For this purpose, the formula of the present value of annually allocated expenses (PVAAE) (Eq 1.) was used [63]:

$$PVAAE = \frac{(1+i)^n - 1}{i(1+i)^n} \tag{1}$$

where n are the years of operation (25 years) and i the interest rate, which is taken equal to 2.5%.

#### *5.3. Estimation of DECEX*

DECEX of an OWF project may correspond to 0%–4% of the total investment costs [16,51,62,64]. In the present paper, for each project, DECEX is taken equal to 2% of the corresponding total investment cost, since it is considered that although a significant amount is collected from recycling the salvaged construction materials, this amount does not suffice to cover the DECEX in total.

#### **6. Assessment of Suitable Areas and Portfolio Analysis**

#### *6.1. Assessment Criteria*

The AC used for the portfolio analysis include: (i) wind velocity (AC.1) [8,12–14,20,26,36], (ii) water depth (AC.2) [8,26,36], (iii) electrical energy demand (AC.3), (iv) distance from ports (with water depth >10 m in terms of draft requirements) (AC.4) [36], (v) distance from high voltage electricity grid (AC.5) [8,9,12], (vi) distance from Marine Protected Areas (AC.6) [12–14] (vii) distance from military exercise areas (firing fields and exercise locations) (AC.7). It is worth to note that "electrical energy demand" and "distance from military exercise areas (firing fields and exercise locations)" have not been used before as AC for the siting of OWFs.

#### *6.2. Prioritisation of AC*

The prioritization of AC is based on a suitably designed questionnaire, which was sent via e-mail to an EG on siting of RES, and, more specifically, of OWFs. The group consists of forty experts from universities, institutes, research centers and companies around the world (USA, Europe, and Asia). These experts were carefully selected, considering the different backgrounds of the participants, so that their distinct opinions reflect different strategic orientations of the present RES siting problem and in order to emphasize the complexity of such siting problems. Out of the forty questionnaires sent out, seven were successfully completed, and were answered by professors, experienced scientific researchers and spatial analysts of RES from various European countries (Greece, Spain, Italy, The Netherlands, etc.). The experts prioritized the selected criteria in the questionnaires based on their own high experience and their own different preferences; the majority of the experts have over seven years of experience on such topics. Figure 2 presents the final prioritization of the AC. The consistency ratio calculated by Microsoft Project Server reached 100%.

**Figure 2.** Relevant weights of assessment criteria (AC).

#### *6.3. Portfolio Analysis*

This particular step involves the actual selection process, where all project proposals are examined in conjunction with the AC and economic restrictions. More specifically, the portfolio analysis arrives at a prioritization of the project proposals, depending on their performance in relation to the AC. Thus, if a project has a major impact on several parallel AC, it is placed high on the priority list and receives a high strategic value score.

In this paper, the portfolio analysis is carried out using Microsoft Project Server software and include the following main steps: (i) input of AC in the library of the influencing factors, (ii) input of all projects (OWFs) and of their estimated total investment cost in the project center, (iii) assessment of the strategic impact of each OWF, based on the particular features of the projects related to the specific AC using a 5-point scale (Table 2), (iv) input of the relevant weight of each AC (Figure 2) according to the EG and (v) creation of scenarios based on economic restrictions (50%, 60%, and 75% of the total portfolio).



### **7. Results and Discussion**

#### *7.1. Strategic Planning Vision and Mission*

The Strategic Planning vision and mission involves a clear constant declaration of purpose that describes the values and priorities of the country, as regards the implementation of OWFs and it is addressed to several different stakeholders, people or groups that are directly or indirectly linked to the implementation of the proposed Strategic Plan. In this paper the vision/mission is defined as follows:

"To acquire national energy independence, by considering sustainable development as well as social responsibility and acceptance as a top priority. The main strategic goal is to promote integrated solutions for the siting, technical characteristics and decision-making issues of OWFs".

#### *7.2. Identification of Suitable Areas*

For identifying marine areas suitable for OWFs' siting in Greece the 13 exclusion criteria, EC.1~EC.13, (Table 1) are taken into account. Figure 3a–d show indicatively the developed for this purpose thematic maps of the exclusion criteria corresponding to: (a) wind velocity (EC.2), (b) marine protected areas (EC.11), wildlife refuges and migration corridors (EC.12), (c) water depth (EC.3) and (d) seismic hazard zones (EC.5). By creating, editing, and managing three different linear models with the use of geoprocessing tools in the 'ModelBuilder' in ArcGIS software, the SA for OWFs' siting in the Greek marine environment emerge and are shown in Figure 4. Specifically, applying the referred models, all prohibited and unsuitable areas erased of the map, according to the limitations that defined on Stg 2.

The number of SA is sixteen (16). Certain very small sites, less than 2.5 km<sup>2</sup> , are not taken into account, since the installation of wind turbines in these small areas does not fulfill any financial purpose. Among the exclusion criteria considered, the water depth and the lack of a high voltage electricity grid represent the two most limiting factors. More specifically, with regard to latter exclusion criterion, there are very few high voltage centers of 400 kV in the south of the country, which are actually situated close to the shore. It should be also noted that for all SA, the water depth is larger than 50 m, advocating the use of floating support structures.

**Figure 3.** Thematic maps of: (**a**) wind velocity (EC.2), (**b**) marine protected areas (EC.11), wildlife refuges and migration corridors (EC.12), (**c**) water depth (EC.3) and (**d**) seismic hazard zones (EC.5).

### *7.3. Identification of Suitable Areas*

The mapping of the 16 OWFs is carried out, using the 'Advanced Editor Tools' in ArcGIS, and the exact number of 5 MW wind turbines is calculated, along with the capacity of each OWF (Table 3). Indicatively, the micro-siting configuration of OWF15 and OWF4 is presented in Figure 5.

**Figure 4.** Suitable areas for OWFs siting in Greece.



<sup>1</sup> TLB is considered for water depths 50–200 m, while Hywind for water depths 200–500 m (see Section 4.2).

#### 405

**Figure 5.** Micro-siting configuration of OWF15 and OWF4.

According to European Network of Transmission System Operators for Electricity [65], the net electricity generating capacity of Greece was in total 16,392.43 MW for 2017 and specifically the amount of 8220.02 MW was produced from RES (25.33% from onshore wind energy, 29.78% from solar energy, 0.73% from biomass energy, 41.35% from renewable hydro energy and 2.81% from other renewable energy sources) and the amount of 8172.41 MW was produced from non-RES. Consequently, with the development and the implementation of the proposed strategic plan, the results from the applied methodology reveal that 45.34% of the electrical energy produced from all the projects in the above table would suffice for the country to achieve energy independence.

In order to keep the total investment costs (e.g., operation and maintenance costs) to minimum and to deploy the suitable marine sites to maximum, the large OWF projects (e.g., OWF1, OWF2) should not be separated into smaller projects. It is suggested that the referred large projects should be implemented in phases, such as 'Gode Wind OWF' in Germany (Project Capacity: 582 MW) [66,67].

#### *7.4. Investment Cost of OWFs*

According to [39], CAPEX of offshore 5 MW wind turbines with a TLB platform and Hywind platform, located at a distance of 200 km from the port and at a water depth of 200 m, are estimated equal to 3.537 <sup>×</sup> <sup>10</sup><sup>6</sup> €/MW and 3.807 <sup>×</sup> <sup>10</sup><sup>6</sup> €/MW respectively. Moreover, it has been shown in [1] that an increase of 10% in water depth or of the distance from the shore will lead to an increase of 1% of investment costs, related mainly to installation and grid connection costs, as well as support structure costs (including installation). By using this rule-of-thumb and considering the aforementioned values of CAPEX per MW as a basis for CAPEX calculations, the CAPEX of OWFs for a given water depth and distance from shore are finally estimated. For example, for a 5 MW wind turbine with a TLB floating platform installed at a water depth of 100 m and at a distance of 50 km from the shore, CAPEX is estimated equal to 3.095 <sup>×</sup> <sup>10</sup><sup>6</sup> €/MW (a 12.5% reduction of the 3.537 <sup>×</sup> <sup>10</sup><sup>6</sup> €/MW base value has been assumed, due to decrease of both water depth and distance from the shore). Regarding OPEX and DECEX, these quantities are expressed as percentages of CAPEX and of total investment cost respectively, as explained in Sections 5.2 and 5.3 above. Table 4 includes the total investment cost of the

16 OWFs, which is used as input in the portfolio analysis. The proposed projects, shown in Tables 3 and 4, could be financed by European Investment Bank (EIB), as the EIB Board (2015) approved support for strategic infrastructure investment totaling nearly € 7 billion for new OWFs in the UK and Belgian coasts [68]. Moreover, the proposed OWF projects could be financial supported by both private (e.g., Greek private banks) and public funds (e.g., Greek public bank).


**Table 4.** Investment cost and OWFs' characteristics in relation to the AC.

#### *7.5. Results of Portfolio Analysis*

The characteristics of all OWFs in relation to the AC required as input in the portfolio analysis are presented in Table 4.

By taking into account the relevant weights of the AC (Figure 2) and the data of Table 4, the portfolio analysis is implemented. Table 5 presents the prioritization of the project portfolio based on the calculated strategic value. The implementation of the whole portfolio achieves a strategic value of 100%, since all proposed projects are implemented, and the investment cost of the whole portfolio amounts to 111.214 <sup>×</sup> <sup>10</sup><sup>9</sup> €.


**Table 5.** Strategic value of portfolio projects.

Next, three possible scenarios for the implementation of portfolio projects are examined. The first scenario corresponds to an investment of 50% of the total portfolio (i.e., 55.607 <sup>×</sup> <sup>10</sup><sup>9</sup> €), the second one to an investment of 60% of the total portfolio (i.e., 66.729 <sup>×</sup> <sup>10</sup><sup>9</sup> €), while the third one to an investment of 75% of the total portfolio (i.e., 83.411 <sup>×</sup> <sup>10</sup><sup>9</sup> €). By implementing the first scenario, eleven OWFs with a 70.08% total strategic value can be realized including OWF3, OWF5, OWF16, OWF8, OWF7, OWF13, OWF11, OWF4, OWF15, OWF12, and OWF14 (in descending order). Most of these OWFs are located Southeast of Euboea (OWF3, OWF5, OWF8, OWF11, Figure 4) and in the Northern Aegean (OWF4, OWF14, OWF15). The total investment cost of these eleven OWFs is equal to 55.207 <sup>×</sup> <sup>10</sup><sup>9</sup> €. On the other hand, in the case of the second scenario, the following twelve OWFs with a 76.92% strategic value and a total investment cost of 66.181 <sup>×</sup> <sup>10</sup><sup>9</sup> € are implemented (Figure 4): OWF3, OWF1, OWF16, OWF8, OWF7, OWF6, OWF13, OWF11, OWF10, OWF15, OWF12, and OWF14 (in descending order). Finally, the third scenario leads to the implementation of the following fourteen OWFs with 87.79% strategic value: OWF3, OWF5, OWF16, OWF8, OWF7, OWF6, OWF13, OWF11, OWF4, OWF10, OWF2, OWF15, OWF12, and OWF14 (in descending order). The total investment cost of these projects amounts to 82.160 <sup>×</sup> <sup>10</sup><sup>9</sup> €, while most of these OWFs are located Southeast of Euboea (OWF3, OWF5, OWF8, OWF11, Figure 4) and in the Northern Aegean (OWF2, OWF4, OWF14, OWF15).

It is worth noting that any of the three scenarios selected for implementation will be particularly beneficial for the country, since the strategic value of all scenarios is over 70%, and national energy independence is ensured. The optimum choice naturally is the implementation of the whole portfolio, since this entails a strategic value of 100% and all strategic goals of this plan are reached. However, such an investment would be particularly costly. Therefore, the second scenario is recommended for implementation, since it requires a little less than 60% of the total investment capital, while, at the same time, it fully satisfies the vision and mission of the present strategic plan (i.e., national energy independence).

Through the strategic planning performed, it has been proven that the combination of GIS with Microsoft Project Server software can, therefore, be seen as a powerful tool for solving complex siting and decision-making issues of OWFs. Moreover, the GIS-Microsoft Project Server integration could be used as an excellent tool for displaying the results in useful maps and, also, in practical portfolio selection scenarios. The proposed methodological approach is essential, as it could be utilized efficiently and easily from academics, scientific researchers, renewable energy planners, developers, government agencies, consultants, and potential investors for the efficient and sustainable exploitation of the offshore wind potential and the relevant deployment of OWFs at different spatial planning scales. The need and the importance of such a strategic planning increase significantly over time, as the increasing population and improving living standards produce an increment on the energy demand. The present study could be further extended with a field investigation analysis of the proposed OWF sites, in order to verify their overall appropriateness. Thus, future work could include a thorough individual assessment of the proposed sites in conjunction with field investigation, in order to make the final selection/prioritization of sites for OWFs installation. Finally, it would be interesting to consider the utilization of energy storage systems in the proposed strategic plan for improving the penetration of RES into the electricity supply mix and maximizing the reliability of the OWF projects regardless weather conditions.

#### **8. Conclusions**

The present paper aims at the development of an integrated strategic planning methodology for OWFs in Greece, where portfolio analysis is used for the first time in relation to the siting of RES. More specifically, it focuses on a strategic plan that will ensure national energy independence for 25 years following its partial implementation, with significant economic, socio-political and environmental benefits for Greece.

A total of twenty criteria (exclusion and assessment) are employed for the siting of OWFs, which almost fully cover the economic, socio-political, technical and environmental issues that are related to such installations and applications. The majority of the criteria are in line with the international literature; however, there are also criteria proposed and applied for the first time in relation to siting issues, such as "seismic hazard zones" (exclusion criterion—on RES issues) and "electrical energy demand" (assessment criterion—on RES issues). In addition, "landscape protection/visual and acoustic disturbance" criterion is introduced, facilitating the combined use of two, relevant, previously suggested siting criteria ("distance from shore" and "visual and acoustic impacts").

Wind turbines of 5 MW capacity are taken into account, while the types of support structures selected are the TLB for depths of 50–200 m and the Hywind for depths of 200–500 m. For each OWF the precise mapping of the wind turbines is carried out by setting specific in-between distances in directions parallel and perpendicular to the prevailing wind direction. Then, the total investment cost for each project is calculated by estimating CAPEX, OPEX and DECEX.

In order to perform the portfolio analysis, the seven selected AC are evaluated by an EG. The highest weight is assigned to the "wind velocity" and "distance from high voltage electricity grid" criteria. Therefore, according to different expert participants the strategic policy orientation of the current plan focuses on the technical and economic dimension of the planning issue, based on the own high-experience. Next, the strategic impact of each OWF is evaluated on each AC and three different management scenarios are realized involving economic restrictions (investing 50%, 60%, and 75% of the total portfolio). After the application of the economic restrictions on the three scenarios, 11, 12 and 14 OWF projects are respectively selected for implementation. In all three scenarios, OWF3 presents the highest strategic value out of the selected projects and OWF14 the lowest, while the commonly selected projects in all 3 portfolios are OWF7, OWF8, OWF11, OWF12, OWF13, OWF14, OWF15, and OWF16. It is worth noting that, in all three cases, the energy independence of the country is achieved. The second scenario is recommended for implementation, since it requires less than 60% of the total investment capital and fully satisfies the vision and mission of the present strategic plan, which ensures national energy independence.

The proposed strategic planning for OWFs in Greece serves as a guide for the development of OWFs in Greece, according to criteria that are based on: (i) the special characteristics of the study area and the individual characteristics of each suitable site, (ii) the relevant provisions of the SFSPSD-RES, (iii) indisputable global experience, (iv) experts' opinion and (v) on extensive international literature review, which specializes on such topics. For this reason, the present paper does not simply recommend certain sites for the installation of OWFs, but areas that have been examined according to several parameters within a broader field of study, which is based on a guideline with clear strategic goals. Consequently, the current strategic planning framework can bridge important gaps between research, development and implementation on the complex RES siting and development issues. The relevant methodology includes distinct stages and can be applied in relation to various study areas and diverse spatial planning scales.

**Author Contributions:** Conceptualization, S.S. and D.G.V.; methodology, S.S., D.G.V. and E.L.; formal analysis, S.S. and E.L.; investigation, S.S.; data curation, S.S.; writing—original draft preparation, S.S., D.G.V.; writing—review and editing, S.S., D.G.V. and E.L.; visualization, S.S.; supervision, D.G.V. and E.L. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** The authors would like to thank Paraskevi Drakopoulou, GIS Specialist of the Lab of Geographical Information Systems at the Institute of Oceanography of the Hellenic Centre for Marine Research (HCMR) for providing wind velocity data, as well as all professors, researchers and spatial analysts who participated in the questionnaire survey.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

1. Ng, C.; Ran, L. *O*ff*shore Wind Farms: Technologies, Design and Operation*; Woodhead Publishing: Duxford, UK, 2016.


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## **Thermo-Poroelastic Analysis of Induced Seismicity at the Basel Enhanced Geothermal System**

#### **Sandro Andrés <sup>1</sup> , David Santillán 1,\*, Juan Carlos Mosquera <sup>2</sup> and Luis Cueto-Felgueroso <sup>1</sup>**


Received: 8 November 2019; Accepted: 29 November 2019; Published: 4 December 2019

**Abstract:** Geothermal energy has emerged as an alternative to ensure a green energy supply while tackling climate change. Geothermal systems extract the heat stored in the Earth's crust by warming up water, but the low rock permeability at exploitation depths may require the hydraulic stimulation of the rock fracture network. Enhanced Geothermal Systems (EGS) employ techniques such as hydro-shearing and hydro-fracturing for that purpose, but their use promotes anthropogenic earthquakes induced by the injection or extraction of fluids. This work addresses this problem through developing a computational 3D model to explore fault reactivation and evaluating the potential for earthquake triggering at preexisting geological faults. These are included in the model as frictional contacts that allow the relative displacement between both of its sides, governed by rate-and-state friction laws and fully coupled with thermo-hydro-mechanical equations. We apply our methodology to the Basel project, employing the on-site parameters and conditions. Our results demonstrate that earthquakes which occurred in December 2006 in Basel (Switzerland) are compatible with the geomechanical and frictional consequences of the hydraulic stimulation of the rock mass. The application of our model also shows that it can be useful for predicting fault reactivation and engineering injection protocols for managing the safe and sustainable operation of EGS.

**Keywords:** geothermal energy; induced seismicity; fault; Basel; poroelasticity

#### **1. Introduction**

The 2030 Agenda for Sustainable Development, adopted by world leaders in September 2015, was established by the United Nations and comprises 17 goals and 169 targets to be fulfilled by the ratifying countries. The Agenda aims to overcome all forms of poverty, while tackling climate change and environmental protection. Goal 7—"Affordable and Clean Energy"—aims to ensure access to affordable, reliable, sustainable, and modern energy for all [1]. Goal 7 goes hand-in-hand with Goal 13, "Climate Action", which aims to take urgent action to combat climate change and its impacts. Geothermal energy emerges as an alternative renewable energy to reach both goals [2,3], as it is affordable and clean. Geothermal systems extract the heat stored in the Earth's crust by warming up water or a mixture of water and gas. The fluid is circulated down through injection wells, heated by the contact with rocks, and returned to the surface through production wells to form a closed loop [4,5]. Hot water or steam is then transformed into a marketable product, such as electricity. Nevertheless, in most geothermal reservoirs, rock permeability at exploitation depths is very low, rendering geothermal projects economically infeasible. The challenge of permeability enhancement has been addressed by the so-called Enhanced Geothermal Systems (EGS) [2,6,7].

EGS enhance rock permeability through hydraulic stimulation of the rock mass fracture network. Two approaches are widely used: hydro-fracturing, which creates new fracture networks, and

hydro-shearing, that reactivates preexisting joints [8,9]. A major environmental issue for these techniques is the risk of induced seismicity as a result of water injection and production [2]. Some major EGS experiences include the Soultz-sous-Forêts project (France) [10], the Cooper Basin project (Australia) [11], the Fenton Hill project (New Mexico, USA) [12], the Rosemanowes project (UK) [13,14], or the Ogachi project (Japan). Furthermore, a number of EGS demonstration projects have recently been launched in the US [15].

Hydro-fracturing, or hydraulic fracturing, enhances rock permeability by opening preexisting fractures and creating new ones. Water is injected within the rock mass under high pressure to open and create the new fractures by tensile failure [16–19]. This technique is broadly used for oil-well stimulation, although it is currently under debate due to environmental concerns [2]. Hydro-shearing increases rock permeability by initiating shear failure. Water is injected under high pressure, reducing normal stress across them and eventually triggering shear failure [9]. Since fault roughness guarantees slip is permanent after injection stops, permeability changes are also permanent [20]. The magnitude of seismic events triggered during hydro-shearing operations is typically small because fractures slip without increasing their aperture as the pore pressure remains bellow the minimum principal stress. Nonetheless, hydro-shearing is widely used on a global scale when conducting field experiments [21] and numerical simulations [8,22,23] to assess permeability evolution.

Most of the scientific community accepts that induced earthquakes may be triggered by water injection into the subsurface. Some examples are the disposal by injection into deep wells of waste water from oil and gas production [24,25], CO<sup>2</sup> sequestration in deep aquifers [26,27], or EGS facilities [28,29]. Earthquakes are the result of a fast slip event on a fault [30]. The onset of the slip depends on the shear stress on the fault and the frictional resistance, both of them affected by fluid injection [31]. On one hand, the increase in pore pressure puts effective stress on the fault decrease, whereas on the other hand, thermo-poro-mechanical effects increase shear stress on the fault [32,33]. The experience gained over the years on EGS projects has shown the usefulness of numerical models for simulating the nucleation and rupture of earthquakes under given injection protocols. Numerical simulations of induced earthquakes require coupling fluid flow, rock deformation, heat transfer, and a fault frictional response [34–36]. Highly nonlinear frictional laws and the disparity in time scales pose a major challenge for computational models. Over the past decade, there have been numerous contributions to thermo-hydro-mechanical modeling of induced seismicity [33,37]. Faults are typically simulated as 3D failure zones with slip-weakening rheology [38–40], or as frictional contact surfaces [41–46]. We adopt the latter approach, since it is consistent with the observed structure of faults [47]. We simulate faults as contact surfaces whose friction evolves according a rate-and-state friction law [48] that incorporates terms depending on the evolution of effective normal stress [49,50]. Our results suggest that injection protocols are relevant so they can be designed to minimize seismic risks [51].

A paradigmatic case is the Deep Heat Mining (DHM) project in Basilea (Switzerland) [52,53], a milestone EGS power station with an energy output capacity of 6 MW of electricity and 17 MW of heat. The station would provide electricity and heat for 10,000 dwellings [54]. The project required the drilling of two wells, Basel-1 at 5.000 m deep to inject cool water, and another well to return hot water back to the surface [55]. Once the former was built, hydraulic stimulation was performed in December 2006. However, it led to perceivable induced seismicity up to a local event magnitude of *M<sup>L</sup>* 3.4, which exceeded the acceptable levels in the Basel urban area [56]. The seismic events led to the premature halt of the project, and finally, its withdrawal.

Here, we develop a fully implicit and monolithically coupled finite element model to simulate fault reactivation. Our model encompasses fully coupled heat transport, rock deformation, and fluid flow processes. Moreover, we describe fault as interfaces whose friction is governed by a laboratory derived rate-and-state friction law incorporating an effective normal stress-rate dependence. We analyze the effects of hydraulic stimulation on seismic risk through a Coulomb failure analysis. Measure of tendency to slip has proven very useful to understand some of the geomechanical challenges posed by subsurface energy technologies (e.g., [46,57–64]). We aim to analyze the hydro-shearing effects during

the stimulation at the Basel-1 well using our three-dimensional (3D) model. We use, as input, the same stress-field and rock properties estimated for the main hydrogeological units at Basel-1, and simulate the stimulation using a transient well-head pressure and flow evolution similar to the values reported at the study site. The proposed model is a useful tool for engineers and practitioners to answer the essential question for the development of the geothermal energy as an affordable, clean, and safe renewable energy source: what is the optimal exploitation protocol that minimizes seismic risk and maximizes economic and energy performance?

#### **2. Materials and Methods**

In this section, we describe our thermo-hydro-geomechanical model, which fully couples heat transport, rock deformation, and fluid flow; faults are described as interfaces with friction governed by the rate-and-state law. We perform numerical simulations of the hydraulic stimulation operations conducted at the DHM project in December 2006 to show the ability of our model to characterize fault reactivation. We adopt as inputs the recorded well-head pressure and flow evolutions, as well as on-site material parameters.

#### *2.1. Frictional Strength and Resistance of Faults*

We employ the Amontons–Coulomb theory as a constitutive model of rock friction. The frictional strength, *τ* ∗ , that impedes sliding of a static, cohesionless contact interface is given by *τ* <sup>∗</sup> = *µ*|*σ*| [65–67]. In general, the relation between these Coulomb magnitudes and the shear stress acting on the contact plane, *τ*, depends on the sliding regime. A static interface satisfies *τ* ≤ *τ* ∗ and for sliding under quasi-static conditions, the relationship *τ* ≈ *τ* ∗ holds.

In the fluid-saturated media, frictional strength is defined using the *effective* normal stress, *σ* 0 = *σ* + *p*, where *σ* is the *total* normal stress acting on the contact, and *p* is the pore pressure of the fluid. In the above and following expressions, tensile stresses are positive, and pore pressure is positive when above the atmospheric value. We assume that effective normal stresses remain compressive on contact surfaces.

#### Rate-and-State Models for Interfaces

Faults are often assumed to be well-oriented for failure but locked prior to injection, in such a way that the onset of slip on an inactive fault—reactivation—is essential to understand the geomechanics of induced seismicity.

Rate-and-state formulations gather the traditional concepts of static and dynamic friction by including the dependence of *µ* on the slip velocity and history of sliding [66,68–70]. These models were derived from laboratory experiments of unidirectional slip in the double-direct shear configuration, and account the response of *µ* after step changes in slip velocity or normal stress [71]. For a frictional interface that is sliding at velocity *V*, the definition of *µ* reads:

$$
\mu = \mu^\* + a \ln\left(\frac{V}{V^\*}\right) + b \ln\left(\frac{\theta}{\theta^\*}\right),
\tag{1}
$$

where *µ* ∗ is the steady-state coefficient at the reference slip velocity *V* ∗ , *a* is the direct-effect parameter, and *b* is the friction evolution parameter. *θ* is the state variable, and *θ* ∗=*Dc*/*V* ∗ is its steady-state value, where *D<sup>c</sup>* is the characteristic slip memory distance over which *τ* ∗ evolves once the system is perturbed [72].

Several definitions for *θ* have been proposed according to the rate-and-state friction models. Deep physical understandings and theoretical analyses [70,73,74], as well as comparisons with experiments of velocity steps [75,76], shear stress steps [77], and normal stress steps [78–81] have allowed us to elucidate the relative advantages and disadvantages of the proposed formulations. Our study focuses on those models that incorporate a dependence on the effective normal stress rate. In such a sense, the Linker–Dieterich model [49] generalizes the basic rate-and-state laws by including a term in the state evolution equation, applicable to both the "slip" [69] and "aging" [68] laws [73]:

$$\begin{cases} \frac{d\theta}{dt} = -\frac{V\theta}{D\_c} \ln \frac{V\theta}{D\_c} - \frac{a\theta}{b\sigma'} \frac{d\sigma'}{dt} & \text{(Slip law)} \\\\ \frac{d\theta}{dt} = 1 - \frac{V\theta}{D\_c} - \frac{a\theta}{b\sigma'} \frac{d\sigma'}{dt} & \text{(Aging law)} \end{cases} \tag{2}$$

In the above equations, the empirical parameter *α* controls the stressing-rate effect on the state variable, ranging from 0 to *µ*<sup>0</sup> [82]. Since we focus on the reactivation of faults which are initially at rest, we adopt a regularization of the rate-and-state models in the limit of zero slip speed, *V* = 0, proposed by Yang et al. [83]. *µ* is then defined as:

$$
\mu(V,\theta) = \mu\_0 + a \ln\left(\frac{V + V^\*}{V^\*}\right) + b \ln\left(\frac{\theta V^\*}{D\_c}\right),\tag{3}
$$

with *µ*<sup>0</sup> being the initial friction coefficient. We implement in our model the equation for the aging law:

$$\frac{d\theta}{dt} = 1 - \frac{\theta \left(V + V^\*\right)}{D\_c} - \frac{a\theta}{b\sigma'} \frac{d\sigma'}{dt},\tag{4}$$

which is equivalent to the regularized model used by Tal et al. [84] with a threshold velocity *V th* = *V* ∗ .

#### *2.2. Thermo-Hydro-Mechanical 3D Model of Fault Reactivation*

The rock is modeled as a thermo-poroelastic saturated material with single-phase flow. We adopt the classical theory of linear poroelasticity [85,86] and solve for the combination of fluid pressure, rock deformation, temperature, and frictional contact on the fault [31,32]. The solid and mass conservation, as well as mechanical equilibrium are coupled using the effective stress. The quasi-static Biot equations for a porous medium read:

$$
\rho\_f S \frac{\partial p}{\partial t} + \rho\_f \alpha\_B \frac{\partial \varepsilon\_v}{\partial t} = \nabla \cdot \left(\rho\_f \frac{k}{\eta\_f} (\nabla p - \rho\_f \mathbf{g})\right), \tag{5}
$$

$$
\nabla \cdot \sigma = \mathbf{0},
\tag{6}
$$

where *α<sup>B</sup>* is the Biot coefficient, *ε<sup>v</sup>* = tr(*ε*) (with *ε* being the infinitesimal strain tensor) is the volumetric strain, *k* is the intrinsic permeability of the porous medium, *η<sup>f</sup>* is the fluid dynamic viscosity, *ρ<sup>f</sup>* is the fluid density, *p* is the pressure field, and *σ* is the total stress tensor. We consider a linear poroelastic material under small deformations, as well as plane strain conditions. Then, the effective stress tensor, *σ* <sup>0</sup> = *σ* + *α<sup>B</sup> p***I**, is a linear function of strains, *σ* <sup>0</sup> = 2*Gεel* + *λ*tr(*εel*)**I**, where *λ* and *G* are the Lamé constants, *εel* = *ε* − *εth* = 1 2 ∇**u** + ∇**u** *T* − *εth* is the elastic strain tensor, the result of subtracting the thermal strains to the total strain tensor *ε*, with **u** being the displacement field. The storage coefficient, *S* = *φχ<sup>f</sup>* + (*α<sup>B</sup>* − *φ*)*χ<sup>s</sup>* , depends on the rock porosity, *φ*, and on the fluid and solid matrix compressibilities, *χ<sup>f</sup>* and *χ<sup>s</sup>* , where *<sup>χ</sup><sup>s</sup>* <sup>=</sup> (<sup>1</sup> <sup>−</sup> *<sup>α</sup>B*)/*K*, and *<sup>K</sup>* <sup>=</sup> *<sup>λ</sup>* <sup>+</sup> <sup>2</sup> <sup>3</sup>*G* is the bulk modulus of the porous matrix.

Conservation of energy reduces to the heat equation [87]:

$$\begin{aligned} (\rho c)\_{sat} \frac{\partial T}{\partial t} + \rho\_f c\_f \mathbf{v} \cdot \nabla T + \nabla \cdot \mathbf{q} &= Q\_f \\ \mathbf{q} &= -\kappa\_{sat} \nabla T, \\ (\rho c)\_{sat} &= \phi \rho c\_s + (1 - \phi) \rho\_f c\_{f'} \\ \kappa\_{sat} &= \phi \kappa\_s + (1 - \phi) \kappa\_{f'} \end{aligned} \tag{7}$$

where *c<sup>s</sup>* is the heat capacity of the rock, *c <sup>f</sup>* is the fluid heat capacity, *κ<sup>s</sup>* is the thermal conductivity of the rock, and *κ <sup>f</sup>* is the fluid thermal conductivity. (*ρc*)*sat* y *κsat* are the saturated values of the product of density by heat capacity and thermal conductivity. *Q* includes the source or sink terms. Temperature changes propagate by diffusion and convection throughout the rock mass and the fluid. This changes produce thermal strains controlled by the thermal expansion coefficient *αT*:

$$
\mathfrak{e}\_{\mathsf{tf}} = \mathfrak{a}\_T (T - T\_0)\_\prime \tag{8}
$$

where *T* is the temperature field, *T*<sup>0</sup> is the reference initial temperature, and *α<sup>T</sup>* is the solid thermal expansion coefficient. We consider that the fluid keeps in a liquid state and its properties do not change with temperature or pressure, due to the high temperature and pressure conditions at the usual depths of the EGS projects. This couples the diffusive part of heat transport, while the convective part is modeled by introducing the Darcy velocity field **v** in Equation (7).

#### *2.3. Case Study: The Deep Heat Mining Project in Basel, Switzerland*

The DHM project in Basel was a milestone in geothermal energy. The know-how gained after the DHM project and the seismic events of 2006 boosted the development of EGS systems [52,53]. The first stage of the project was drilling the 5000 m depth Basel-1 well. The well reached a crystalline granitic rock basement at a temperature of 200 ◦C. The well was also employed for the hydraulic stimulation of the reservoir and the field characterization [56].

The number of fractures between 4629 m and 5000 m depth was between 0.2 and 0.3 fractures per meter [56]. The preferred fracture direction and orientation were NW–SE to NNW–SSE with dips greater than 60◦ , although the measures of the hypocenter locations during seismic events detected new families of fractures [88]. The events with greatest magnitude which rolled around in 2006 took place on a family with a deviation of 10◦ with respect of the maximum principal stress direction and a dip of 80◦ . The orientation of the principal stresses was deduced from acoustic geophysical studies within the Basel-1 well. The minimum principal stress, *σh*,*min*, had an orientation of 54 ± 14◦ and the maximum one, *σh*,*max*, 144 ± 14◦ [88]. These orientations are consistent with the in situ stress state in the upper Rhine Basin deduced from previous seismic events in the crystalline rocky massif [56].

The magnitude of the principal stresses were also quantified. The tectonic ratio of the minimum principal stress, *σh*,*min*, to the vertical one is 0.7, and the ratio of the maximum principal stress, *σh*,*max*, to the vertical one is 1.6, in such a way that the vertical stress is the intermediate principal stress [56]. Mechanic boundary conditions are defined by the expressions:

$$\begin{aligned} \sigma\_{\upsilon}(d) &= (\rho(1-\phi) + \rho\_f \phi)g d\_\prime \\ \sigma\_{h,\min}(d) &= \text{Tect}\_{\min} \cdot \sigma\_{\upsilon} = 0.7 \sigma\_{\upsilon\prime} \\ \sigma\_{h,\max}(d) &= \text{Tect}\_{\max} \cdot \sigma\_{\upsilon} = 1.6 \sigma\_{\upsilon\prime} \end{aligned} \tag{9}$$

where *d* is the depth. In these expressions, compressive stresses are assumed positive. The imposed stresses at the boundaries are assumed constant in time, which is a feasible hypothesis even if simulation time exceeds one decade [89]. The temperature at the bottom of the well is between 190 and 200 ◦C, with a thermal gradient of 40 ◦C/km [55].

#### *2.4. The Basel 3D Model*

The geometry of the 3D model domain is a 1.5 km<sup>3</sup> cube that is supposed to be homogeneous porous rock mass. In the center of the cube, we situated a sphere divided into two hemispheres with their intersection plane representing the main fault. The injection well is modeled as a cylinder with a diameter of 1 m and a height of 380 m whose geometry is subtracted from the solid cube. The well is vertical and located at a distance 50 m away from the center of the fault (Figure 1). The diameter of the simulated well is higher than the real one due to mathematical issues, where the diffusion of the injected fluid should not be simulated with elements which are too fine. The geometry of the model is described in Figure 1.

**Figure 1.** Scheme of the 3D Basel EGS model. In (**a**) we show the domain that is a 1.5 km<sup>3</sup> cube with the fault plane oriented 10◦ with respect to the *x*-axis and dip 80◦ to the SW. The domain is located between 4050 and 5550 m depth, while the injection section of the Basel-1 well extends from 4629 to 5000 m depth. In (**b**) we plot the injection protocol (left vertical axis) and injection pressure (right vertical axis) measured at the Basel-1 well, obtained from [56].

We focus on the depth range between 4050 and 5550 m, and the injection takes place through the well between 4620 to 5000 m depth. The *x* axis is parallel to the maximum principal stress (*σ<sup>x</sup>* = *σh*,*max*), the *y* axis to the minimum principal stress (*σ<sup>y</sup>* = *σh*,*min*), and the *z* axis to the intermediate principal stress. We apply vertical stress to simulate overburden strata according to *σv*(*z*) = (*ρ*(1 − *φ*) + *ρ<sup>f</sup> φ*)*g*(5500 − *z*), where the *z* coordinate ranges between 0 and 1500 m in our model.

The complexity of the fracture network requires the adoption of some simplifications to define the orientation of the fault plane, based on the hypocenter locations estimated from the motion data records of the historical seismic activity [88]. We adopt the most unfavorable fracture family as the preferred sliding plane, which has a direction deviated approximately 10◦ with respect to the direction of the maximum principal stress, which dips 80◦ towards the SW. Most of the detected hypocenters do not separate more than 50 m with respect to the upper injection section of the well [55]. For that reason, the fault plane of the model is oriented 10◦ with respect to the *x*-axis and dipping 80◦ to the SW, with the injection well-located at a distance 50 m away from its center. We refine the mesh of the model in the area of the central sphere, as shown in Figure 2, and force mesh conformity throughout the fault plane. We use tetrahedral elements—quadratic for the solid mechanics equations and linear for fluid flow and heat transport. Modelo Elementos Finitos 3D • Parámetros y condiciones de contorno

**Parámetros de fricción** <sup>0</sup> = 0.55 <sup>0</sup> = 10−9 / = 0.005 = 700

Material elástico: = 20 ; = 0.25 Densidad sólido: = 2700 /<sup>3</sup>

Permeabilidad intrínseca: = 10−15 <sup>2</sup>

Porosidad: = 0.1

 <sup>∗</sup> = 0

= 7 ∙ 105

**Condiciones del modelo**

Densidad del agua: = 1000 /<sup>3</sup>

Conductividad térmica del sólido: = 2.4 /( ∙ ) Calor específico del sólido: = 800 /( ∙ ) Conductividad térmica del agua: = 0.6 /( ∙ ) Calor específico del agua: = 4200 /( ∙ )

ℎ, = 1,6, ℎ,= 0,7

Tensiones de confinamiento <sup>=</sup> ℎ

Gradiente geotérmico local 40º/ Presión hidrostática <sup>0</sup> = ℎ Profundidad del modelo ℎ = 4050~5550

**Parámetros del modelo**

Coeficiente de Biot: = 1 Viscosidad del agua: = 0.00024 ∙

Compresibilidad del agua: <sup>=</sup> <sup>4</sup> <sup>∙</sup> <sup>10</sup>−10 Coeficiente de dilatación térmica: −1 <sup>=</sup> <sup>8</sup> <sup>∙</sup> <sup>10</sup>−6

= 0.2

= 0.03

**Figure 2.** 3D finite element mesh and mechanical boundary conditions applied. At the exterior boundaries with no stresses applied, we impede the displacement in its normal direction.

Mechanic boundary conditions are equal to tectonic stresses and are imposed on three faces of the rectangular domain (Figure 2). We impose a reference hydrostatic pressure, *p* = *ρ<sup>f</sup> gd* = *ρ<sup>f</sup> g*(5550 − *z*), along all external boundaries, where *d* is the depth, and the temperature is set at the geothermal gradient T = 0.04 [C◦/m]·*d* = 0.04 [C◦/m]·(5550 − *z*).

We impede the displacements in the normal direction and impose no-flow (thermal and hydraulic) conditions on the other ones. At the fault plane, we impose the contact condition, allowing relative tangential displacements between its edges. We consider the fault is almost impermeable and has the same thermal properties as the rest of the domain, resulting in thermal continuity. We simplify the fluid flow around the injection well by imposing a volumetric flux along the boundary of an effective injection region (the lateral surface of the cylinder). During the simulations, temperature is also fixed in the injection well.

Based on the in-site measurements, we define feasible parameters for each physical process of the model.

The Young modulus, the Poisson ratio, and the density of the solid skeleton are usual values for crystalline rock formations of granitic type. The parameters related to the fluid, such as density or compressibility of water, are also usual values, unlike viscosity, whose value of 2.4 <sup>×</sup> <sup>10</sup>−<sup>4</sup> Pa·s, is lower than the viscosity of water at ambient temperature (*η<sup>f</sup>* = 10−<sup>3</sup> Pa·s) due to the high temperatures at such depths [23]. Permeability and porosity have been chosen according to the characteristics of the rock mass. Lastly, thermal parameters have been taken from [90] that collects data of thermal conductivities and heat capacities from different materials. Coupling between the flow and mechanical problems have been included using the Biot coefficient *α<sup>B</sup>* = 1 and between the thermal and mechanical physics with the thermal expansion coefficient *<sup>α</sup><sup>T</sup>* = 8 <sup>×</sup> <sup>10</sup>−<sup>6</sup> .

The parameters of the "rate-and-state" model have been chosen within a feasible range to emulate the fault reactivation at a similar time scale. We consider the fault is arrested before reactivation, being the slip speed *V* = 0. Hence, Equations (1) and (2) can be simplified, so it is only necessary to define the parameters *α*, *b*, *Dc*, and *V* ∗ . We adopt the aging law and the rate-and-state parameters, *b* = 0.03, *D<sup>c</sup>* = 700 µm, *V* <sup>∗</sup> = 10−<sup>9</sup> m/s, and *µ*<sup>0</sup> = 0.55. Table 1 lists the parameters of the model.


**Table 1.** Parameters of Basel 3D model.

The constitutive laws for fault strength are given by Equations (1) and (2), and the frictional contact on the fault is modeled using an Augmented Lagrangian formulation [91]. We solve, in a monolithically-coupled fashion, the field Equations (5)–(8) and the rate-and-state aging law (1) and (2) with the frictional contact variables [31,32].

#### **3. Results and Discussion**

#### *3.1. Calibration*

The hydraulic stimulation of the DHM geothermal project required the injection of 11, 600 m<sup>3</sup> of water, following the protocol depicted in Figure 1b [56]. This volume was injected prior to the 8 December 2006 earthquake sequence. Since we simulated the on-site conditions during the injection operations, we imposed an inflow velocity at the injection well of the model that is the result of dividing the flow rate by the lateral surface of the cylinder *qiny*=*Q*0/(2*πrh*), where *r* is the radius of the cylinder and *h* is the height where injection takes place. We assumed that the injection temperature remained constant and equal to ambient rock temperature.

Figure 3a,b shows a plot of the simulated evolutions of pressure and injected flow rate computed with our model, and the registered data at the Basel-1 injection well in 2006. We adopted as the flow boundary condition the measured injected flow rate at the injection well (see Figure 3a). Our computed injected pressure evolution initially differs from the measured values on-site, Figure 3b. The difference between pressure observed in real data and model results arises from the weakening and fracturing of the rock in the vicinity of the well, allowing water to flow through the rock matrix and its fractures. After 5 days of injection, when the largest earthquakes occurred, the difference between our simulated results and the measured data were drastically reduced, showing that our model is properly adjusted.

**Figure 3.** 3D model calibration results. In (**a**) we show the injection flow rate pattern used in the model (red line) that is similar to the real injection protocol from the Basel-1 data (blue line). In (**b**) we plot the computed injection pressure (red line) and the values measured in 2006 (blue line).

The results of pore pressure and temperature fields are included in Figure 4. We define a reference horizontal plane at 4800 m depth (*z* = 750 m), where the changes of pore pressure and temperature cause by the injection of cold water are displayed. We show in Figure 4a the locations of the reference plane, the fault plane, and the injection well. We also include the buildup of pore pressure around the injection well and near the fault plane at the time that reactivation occurs (*t* = 5.5 days). The injection cools down a small area around the well which does not reach the fault (Figure 4b). Since the time for heat diffusion is higher than the one for pressure diffusion, pore pressure changes around the well are much faster than temperature changes. As fault temperature remains constant during the stimulation phase (Figure 4b), pore pressure increases on the fault plane (Figure 4c). We explain in the next section how pore pressure on the fault plane controls the fault reactivation and the frictional properties of the contact.

**Figure 4.** 3D model results at the horizontal reference plane. In (**a**) (left) we show the reference plane inside the model, the fault plane, and the injection well. The increase of pore pressure (right) in that plane shows the results of the pore pressure increase caused by the injection at the instant of fault reactivation (among 26 MPa). In (**b**) we display the increment of temperature due to the injection at the reference plane, with an inset that zooms the surroundings of the injection well and shows that the temperature diffusion is much slower than pressure propagation. In (**c**) we show the results of the pore pressure increase at the fault plane. The vertical axis of the image corresponds to the maximum slope line of the fault plane and the horizontal axis corresponds to a horizontal direction in the 3D model deviated 10◦ with respect to the x-axis.

#### *3.2. Fault Reactivation*

Fault reactivation is the onset of fault slip. It depends on the variables involved in fault stability, such as frictional strength *µ*|*σ* 0 | or shear stress *τ*. We quantify fault stability through the Coulomb Failure Function, defined as *CFF* = *µ*|*σ* 0 | − *τ*. The *CFF* equals to zero when the fault is at rest, and when *CFF* is less than zero the fault reactivates, given that the shear stress *τ* exceeds the frictional strength of the contact *µ*|*σ* 0 |. Changes in Coulomb Failure Function can be used as a proxy for fault weakening (∆*CFF* < 0) or strengthening (∆*CFF* > 0). In that sense, we show in Figure 5 the increase in the failure function ∆*CFF* = *CFF*(*t*) − *CFF*(*t* = 0) on the fault plane, which indicates how the fault weakens due to the effect of fluid injection.

**Figure 5.** 3D model results on the fault plane 5.5 days after the onset of injection (the instant of fault reactivation). We display the increase in Coulomb Failure Function on the fault plane due to the injection ∆*CFF* = ∆(*µ*|*σ* 0 | − *τ*) = *CFF*(*t*) − *CFF*(*t* = 0). The results show the fault weakening (red color, ∆*CFF* < 0) due to the injection of cold water.

We include in Figure 6 our computed results with the 3D model for the variables involved in the frictional stability on the fault plane. The distribution of effective normal stress, |*σ* 0 |, acting on the fault plane (Figure 6a) at the onset of the slip shows that there is a decrease in fault effective compressions. Moreover, the spatial distribution is symmetric with respect to a vertical axis. The decrease in effective normal stress is almost the same as the increase in pore pressure (Figure 4c). Differences arise from poroelastic and thermal effects, and indicate that pore pressure changes dominate over thermal and poroelastic effects.

We plot the modulus of the shear stress *τ* on the fault at reactivation time in Figure 6b. Shear stress increases on the north side of the fault and decreases on the right side. This response is caused by the increase in pore pressure, as well as by the poroelastic effects accounted in our model [16–19,31,32]. The poroelastic effects are also coupled with the fault orientation and tectonic stresses, which contribute to the asymmetry of the results.

We show the value of the friction coefficient, *µ*, in Figure 6c, computed with the rate-and-state law (Equatons (3) and (4)). Since the fault is initially at rest (*V* = 0), the observed evolution of the friction coefficient is attributed to the decrease in the effective stress through the Linker-Dieterich term and the *α*-parameter in Equation (2) [79]. The Coulomb Failure Function *CFF*, Figure 6d, indicates that fault reactivation occurs after 5.5 days of injection. The asymmetry in the *CFF* distribution remarks the influence of shear stresses, in contrast with the symmetry of effective normal stress and friction coefficient that are directly pore-pressure dependent. This pattern of symmetry and asymmetry of stresses before and at reactivation also influence on the nucleation and rupture phases of the earthquake.

**Figure 6.** Display of 3D model results at the 2D fault plane 5.5 days after the injection starting (the instant of fault reactivation). In (**a**) we show the effective normal stress |*σ* 0 | (positive values of effective normal stresses are compressive). In (**b**) we display the modulus of the shear stress *τ*, in (**c**) the friction coefficient *µ* obtained from the rate-and-state equations, and in (**d**) the Coulomb Failure Function *CFF*.

We illustrate the evolution of stresses and frictional variables up to the fault reactivation at a control point on the fault. The point is located at the left-half part of the fault, and it is the first point where the ratio of the acting shear stress to the effective normal stress equals the frictional strength—that is, it is the first fault point at which slip occurs (blue area where *CFF* = 0 in Figure 6d).

Figure 7 shows the evolution of the friction at the control point up to the reactivation. The mobilized friction (green line) is the ratio *τ*/|*σ* 0 | that normalizes the shear stress acting on the fault with the effective normal stress, and the friction coefficient *µ* (blue line) is a dimensionless frictional strength. When the mobilized friction equals frictional strength at 5.5 days (*τ*/|*σ* 0 | = *µ*), the fault reactivates.

We observe that due to the injection protocol that systematically increases the flow rate, and consequently, the pore pressure, the slopes of both the friction coefficient and the mobilized friction curves increase. Moreover, the friction coefficient changes from *µ*<sup>0</sup> = 0.55 to 0.61 at the reactivation instant due to the effect of the variation of the effective normal stress on the friction coefficient. It delays the fault reactivation, which changes from 4.6 days if effective normal stress rate is disregarded, and to 5.5 days if the rate is accounted for. Therefore, the effect of normal stress rate on friction coefficient needs to be taken into account, as it partially controls the reactivation time. These results verify that our models properly reproduce the on-site reactivation of the reservoir's representative fault after 5.5 days of injection, elapsed from December 3 to 8.

Our methodology may be a useful tool for engineers, practitioners, and stakeholders to assess fault reactivation under real conditions of natural stresses, temperatures, and injection protocols. The application of our model to the DHM project in Basel has shown its ability to predict fault reactivation and demonstrated that the earthquake sequence occurred in December 2006 may have been caused by the hydraulic stimulation of the rock mass. Our model can also be useful for assessing new injection protocols, as well as stimulating and managing the operation of EGS system in the short and long term.

**Figure 7.** Evolution of the friction variables at the fault control point. The blue line represents the evolution of the friction coefficient *µ* as derived from the rate-and-state equations. The green line plots the evolution of the ratio *τ*/|*σ* 0 |. The slopes of both curves are related to the different flow-rate levels of the injection protocol.

#### **4. Conclusions**

Geothermal energy emerges as an alternative renewable energy to ensure access to affordable, reliable, sustainable, and modern energy for all the world. In most of the geothermal deposits, rock permeability at exploitation depths is very low, rendering geothermal projects economically infeasible. This drawback has been solved by the so-called Enhanced Geothermal Systems (EGS), where rock permeability is enhanced through the so-called rock stimulation. One of the most widely used techniques is hydro-shearing, which reactivates preexisting joints by initiating shear failure. Water is injected under high pressure, reducing normal stress across them and eventually triggering shear failure. A major environmental issue for these techniques is induced seismicity as a result of water injection.

Here, we have presented a finite element model for the simulation of fault reactivation in poroelastic media with rate-and-state friction law. Our model monolithically couples fluid flow, rock mechanics, heat transport, and rate-and-state friction. We have conducted three-dimensional simulations of fault reactivation with frictional strength governed by a Linker–Dieterich law, embedded in a poroelastic homogenous media, and driven by fluid injection. The Linker–Dieterich law accounts for the effect of effective stress rate on the friction evolution, and is able to explain the observed delays in fault reactivation.

We applied our model to simulate the hydro-shearing effects during the stimulation at the Basel-1 well at the Deep Heat Mining geothermal project in Basilea (Switzerland). We adopted as input the stress field and rock properties estimated for the main hydrogeological units at Basel-1, and simulated the stimulation phase using a transient well-head pressure and flow evolution similar to the values

reported at the study site. Our three-dimensional model satisfactorily reproduced the registered injection flow rate and pressure injection, as well as the time of the fault reactivation.

Our simulated results indicate that thermal effects are negligible during the stimulation phase. Temperature changes occur in a small area around the injection well and do not reach the fault plane. Nevertheless, pressure changes, together with poroelastic effects, weaken the fault, leading eventually to fault reactivation and the onset of slip. Our model was able to reproduce the instant at which fault reactivation occurred at the Basel-1 site, demonstrating that the earthquake sequence that occurred in December 2006 in Basel was caused by the hydraulic stimulation of the geothermal reservoir.

Our methodology emerges as a useful tool for engineers, practitioners, and stakeholders to assess fault reactivation under real conditions of natural stresses, temperatures, and injection protocols. The application of our model to the DHM project in Basel has shown its ability to predict fault reactivation in a real case. Our model can also be useful for assessing new injection protocols, as well as stimulating and managing the operation of an EGS system in the short and long term.

#### **Author Contributions:** Investigation, S.A., D.S., J.C.M., and L.C.-F.

**Funding:** This research was supported by the Universidad Politécnica de Madrid through the special programme for young scientists under grant VJIDOCUPM19DSS ("Programa Propio de I+D+I de la Universidad Politécnica de Madrid. Convocatoria de ayuda dirigida a jóvenes investigadores doctores para fortalecer sus planes de investigación"). DSS gratefully acknowledges funding from the Fundación BBVA though Becas Leonardo a Investigadores y Creadores Culturales 2019 (Proyecto realizado con la Beca Leonardo a Investigadores y Creadores Culturales 2019 de la Fundación BBVA).

**Conflicts of Interest:** The authors declare no conflict of interest. The founders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Sustainability* Editorial Office E-mail: sustainability@mdpi.com www.mdpi.com/journal/sustainability

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18