**Efficiency and Sustainability of the Distributed Renewable Hybrid Power Systems Based on the Energy Internet, Blockchain Technology and Smart Contracts-Volume II**

Editors

**Nicu Bizon Mamadou Ba¨ılo Camara Bhargav Appasani**

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

*Editors* Nicu Bizon Faculty of Electronics, Communication and Computers University of Pitesti Pitesti Romania

Mamadou Ba¨ılo Camara Laboratoire GREAH Universite Le Havre ´ Normandie Le Havre France

Bhargav Appasani School of Electronics Engineering Kalinga Institute of Industrial Technology Bhubaneswar India

*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 Hybrid Power).

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 978-3-0365-6371-8 (Hbk) ISBN 978-3-0365-6370-1 (PDF)**

© 2023 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**


Reprinted from: *Sustainability* **2021**, *13*, 9502, doi:10.3390/su13179502 . . . . . . . . . . . . . . . . **197**

Synchronous Reluctance Motor for e-Vehicle Applications


Reprinted from: *Sustainability* **2021**, *13*, 8048, doi:10.3390/su13148048 . . . . . . . . . . . . . . . . **263**

## **Preface to "Efficiency and Sustainability of the Distributed Renewable Hybrid Power Systems Based on the Energy Internet, Blockchain Technology and Smart Contracts-Volume II"**

The very fast increase in the world's energy demand over the last decade, and the request for sustainable development, can be approached through micro- and nano-grids using hybrid power systems based on the energy internet, blockchain technology, and smart contracts.

This book is the second volume in these topics and includes innovative solutions and experimental research, as well as state-of-the-art studies, in the following challenging fields:











The climate changes that are becoming visible today are a challenge for the global research community. The stationary applications sector is one of the most important energy consumers. Harnessing the potential of renewable energy worldwide is currently being considered to find alternatives for obtaining energy by using technologies that offer maximum efficiency and minimum pollution. In this context, renewable energy sources, fuel cell systems and other energy generating sources must be optimally combined and connected to the grid system using advanced energy transaction methods.

As this book presents the latest solutions in the implementation of fuel cell and renewable energy in mobile and stationary applications, such as hybrid and microgrid power systems based on the Energy Internet, Blockchain technology, and smart contracts, we hope that they will be of interest to readers working in the related fields mentioned above.

> **Nicu Bizon, Mamadou Ba¨ılo Camara, and Bhargav Appasani** *Editors*

## *Review* **Blockchain-Enabled Smart Grid Applications: Architecture, Challenges, and Solutions**

**Bhargav Appasani <sup>1</sup> , Sunil Kumar Mishra <sup>1</sup> , Amitkumar V. Jha <sup>1</sup> , Santosh Kumar Mishra <sup>1</sup> , Florentina Magda Enescu <sup>2</sup> , Ioan Sorin Sorlei <sup>3</sup> , Fernando Georgel Bîrleanu <sup>4</sup> , Noureddine Takorabet <sup>5</sup> , Phatiphat Thounthong 5,6 and Nicu Bizon 2,3,4,\***


**Abstract:** The conventional electrical grid is undergoing substantial growth for reliable grid operation and for more efficient and sustainable energy use. The traditional grid is now metamorphosing into a smart grid (SG) that incorporates a diverse, heterogeneous blend of operating measures such as smart appliances, meters, and renewable energy resources. With better efficient results and dependability, the SG can be described as a modern electric power grid architecture. The SG is one of the greatest potential advances as a promising solution for the energy crisis. However, it is complex and its decentralization could be of tremendous benefit. Moreover, digitalization and integration of a large number of growing connections make it a target of cyber-attacks. In this sense, blockchain is a promising SG paradigm solution that offers several excellent features. There has been considerable effort put into using blockchains in the smart grid for its decentralization and enhanced cybersecurity; however, it has not been thoroughly studied in both application and architectural perspectives. An in-depth study was conducted on blockchain-enabled SG applications. Blockchain architectures for various applications, such as the synchrophasor applications, electric vehicles, energy management systems, etc., were proposed. The purpose of this article is to provide directions for future research efforts aimed at secure and decentralized SG applications using blockchain.

**Keywords:** smart grid; blockchain; smart contracts; cybersecurity; microgrids; electric vehicles; energy transactions; energy management; smart cities; advanced metering infrastructure; home automation; smart homes

### **1. Introduction**

The power grid is a complex engineering marvel, which is undergoing rapid changes due to the proliferation of renewable energy resources, high-speed signal processors, and intelligent sensors, etc. The present requirement involves bi-directional flow energy and information between the power generators and the power consumers. So, the traditional

**Citation:** Appasani, B.; Mishra, S.K.; Jha, A.V.; Mishra, S.K.; Enescu, F.M.; Sorlei, I.S.; Bîrleanu, F.G.; Takorabet, N.; Thounthong, P.; Bizon, N. Blockchain-Enabled Smart Grid Applications: Architecture, Challenges, and Solutions. *Sustainability* **2022**, *14*, 8801. https:// doi.org/10.3390/su14148801

Academic Editor: Thanikanti Sudhakar Babu

Received: 11 April 2022 Accepted: 14 July 2022 Published: 18 July 2022

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

**Copyright:** © 2022 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/).

[2].

power grid is evolving into a smart grid (SG), a grid that is capable of dynamically monitoring and controlling the flow of power, providing reliable power to the consumers [1]. quirements. These components include renewable and non-renewable energy sources, intelligent sensors, controllers, etc. The statistics on research publications related to the SG

power grid is evolving into a smart grid (SG), a grid that is capable of dynamically moni-

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 2 of 35

The SG connects heterogeneous components that vary in their functionality and re-

The SG connects heterogeneous components that vary in their functionality and requirements. These components include renewable and non-renewable energy sources, intelligent sensors, controllers, etc. The statistics on research publications related to the SG are shown in Figure 1. These statistics were obtained from the Scopus database. The various applications that the SG caters to are shown in Figure 2. In Figure 2, the share of research publications from the application's perspective is shown. From this figure, it can be observed that the main applications in an SG are the energy management systems (EMS), electric vehicles (EVs), microgrids (MGs), smart cities (SCs), home automation (HA), advanced metering infrastructure (AMI), and synchrophasor applications (SPAs) [2]. are shown in Figure 1. These statistics were obtained from the Scopus database. The various applications that the SG caters to are shown in Figure 2. In Figure 2, the share of research publications from the application's perspective is shown. From this figure, it can be observed that the main applications in an SG are the energy management systems (EMS), electric vehicles (EVs), microgrids (MGs), smart cities (SCs), home automation (HA), advanced metering infrastructure (AMI), and synchrophasor applications (SPAs) power grid is evolving into a smart grid (SG), a grid that is capable of dynamically monitoring and controlling the flow of power, providing reliable power to the consumers [1]. The SG connects heterogeneous components that vary in their functionality and requirements. These components include renewable and non-renewable energy sources, intelligent sensors, controllers, etc. The statistics on research publications related to the SG are shown in Figure 1. These statistics were obtained from the Scopus database. The various applications that the SG caters to are shown in Figure 2. In Figure 2, the share of research publications from the application's perspective is shown. From this figure, it can be observed that the main applications in an SG are the energy management systems

(EMS), electric vehicles (EVs), microgrids (MGs), smart cities (SCs), home automation

**Figure 1.** Publication statistics on SG.

**Figure 2.** Distribution of research related to SG.

**Others 2%**

**Energy Management Systems 36%**

**17%**

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 2 of 35

**Electric Vehicles**

**Figure 2.** Distribution of research related to SG. **Figure 2.** Distribution of research related to SG.

**22% Microgrids**

respectively in literature.

**Reference Blockchain from an SG Application**

SG enhances the reliability of power supply and materializes several applications at the cost of increased complexity [3]. In this complex network, at a given instance, there are several entities in the grid that carry out transactions. An important concern is validating a transaction between the various entities involved in a particular SG application. A promising and secure solution for this problem is the use of Blockchain technology. dating a transaction between the various entities involved in a particular SG application. A promising and secure solution for this problem is the use of Blockchain technology. Blockchain technology, first introduced by Satoshi Nakamoto, helps achieve consensus about the authenticity of a particular transaction and helps maintain trust between

SG enhances the reliability of power supply and materializes several applications at

are several entities in the grid that carry out transactions. An important concern is vali-

Blockchain technology, first introduced by Satoshi Nakamoto, helps achieve consensus about the authenticity of a particular transaction and helps maintain trust between various entities involved [4]. The number of papers published on blockchain technology every year is shown in Figure 3. Additionally, the corresponding number of papers published on Blockchain for SG is shown in this figure. The publication statistics were obtained from the Scopus database. various entities involved [4]. The number of papers published on blockchain technology every year is shown in Figure 3. Additionally, the corresponding number of papers published on Blockchain for SG is shown in this figure. The publication statistics were obtained from the Scopus database.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 3 of 35

**Figure 3.** Publication statistics on blockchain and blockchain for SG. **Figure 3.** Publication statistics on blockchain and blockchain for SG.

The statistics indicate that blockchain technology is not being exploited for SG applications. Only 3.5% of publications on the blockchain are related to the SG applications. The motivation for this review was to explore the research available on the blockchain for SG, categorize it based on the application, propose the blockchain architectures for the various SG applications, identify the challenges in this regard, and suggest suitable solutions. The review papers and surveys on blockchain for SG are summarized in Table 1. Contrary to these works, the present work presents a boarder perspective on different SG The statistics indicate that blockchain technology is not being exploited for SG applications. Only 3.5% of publications on the blockchain are related to the SG applications. The motivation for this review was to explore the research available on the blockchain for SG, categorize it based on the application, propose the blockchain architectures for the various SG applications, identify the challenges in this regard, and suggest suitable solutions. The review papers and surveys on blockchain for SG are summarized in Table 1. Contrary to these works, the present work presents a boarder perspective on different SG applications with the blockchain. Moreover, the present work also describes the architecture of the blockchain-enabled SG applications. A wide range of potential applications of SG is considered, such as EV, AMI, SPA, MGs, SCs, HA, and EMS.

applications with the blockchain. Moreover, the present work also describes the architecture of the blockchain-enabled SG applications. A wide range of potential applications of SG is considered, such as EV, AMI, SPA, MGs, SCs, HA, and EMS. **Table 1.** Existing reviews on blockchain for SG. "✓" and "✕" indicates "included" and "excluded" The paper is organized in the following sections, as represented in Figure 4. Section 2 discusses the basic concepts of a blockchain. It presents the terminology related to the blockchain and its general architecture. Section 3 presents a review of the blockchainenabled SG applications. Different applications are discussed, and their architectures are presented. The security concerns pertaining to these applications are discussed in Section 4, and Section 5 is the conclusion of this review.

**Perspective SG Applications Considered**

[4] ✕ ✕ ✓ ✕ ✕ ✕ ✓ ✕ ✕ ✓ [5] ✕ ✕ ✓ ✓ ✕ ✕ ✕ ✕ ✕ ✕ [6] ✕ ✕ ✕ ✕ ✕ ✕ ✕ ✕ ✕ ✓ [7] ✕ ✕ ✓ ✕ ✕ ✕ ✕ ✕ ✓ ✕ [8] ✕ ✓ ✕ ✓ ✕ ✕ ✕ ✕ ✕ ✓ [9] ✕ ✓ ✓ ✕ ✕ ✕ ✕ ✕ ✕ ✕


*Sustainability* **2022**, *14*, x FOR PEER REVIEW 4 of 35

**Table 1.** Existing reviews on blockchain for SG. "X" and "✕" indicates "included" and "excluded" respectively in literature. [10] ✕ ✓ ✓ ✕ ✕ ✕ ✕ ✕ ✕ ✕ [11] ✕ ✕ ✓ ✕ ✕ ✕ ✕ ✕ ✕ ✕

**Figure 4.** Organization of the paper. **Figure 4.** Organization of the paper.

### **2. Overview of Blockchain**

**2. Overview of Blockchain** In the past few years, blockchain technology has received tremendous attention worldwide. At the beginning of the technology's inception for application in digital currency, or cryptocurrency, blockchain was considered a cryptocurrency [23]. Bitcoin, the most popular cryptocurrency, was considered to be the blockchain. However, blockchain is the backbone of these cryptocurrencies. It is a distributed ledger for transactions in a decentralized network. Initially, the researchers were skeptical about this technology, but the popularity of Bitcoin changed their perception. This can be corroborated in the sudden In the past few years, blockchain technology has received tremendous attention worldwide. At the beginning of the technology's inception for application in digital currency, or cryptocurrency, blockchain was considered a cryptocurrency [23]. Bitcoin, the most popular cryptocurrency, was considered to be the blockchain. However, blockchain is the backbone of these cryptocurrencies. It is a distributed ledger for transactions in a decentralized network. Initially, the researchers were skeptical about this technology, but the popularity of Bitcoin changed their perception. This can be corroborated in the sudden growth in the number of published articles on the blockchain after 2016 as shown in Figure 3. Blockchain

healthcare, industries, etc. These various applications are depicted in Figure 5.

growth in the number of published articles on the blockchain after 2016 as shown in Figure 3. Blockchain is being considered in various other domains such as banking, healthcare,

is being considered in various other domains such as banking, healthcare, healthcare, industries, etc. These various applications are depicted in Figure 5.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 5 of 35

**Figure 5.** Applications of blockchain. **Figure 5.** Applications of blockchain. **Figure 5.** Applications of blockchain.

#### *2.1. Structure of Blockchain 2.1. Structure of Blockchain*

*2.1. Structure of Blockchain*  The blockchain comprises a series of blocks of transactions linked together in a chain, as shown in Figure 6. Client/server architecture is used in traditional client/server systems, and various administrators are in charge of them. On the other hand, blockchain is a distributed, decentralized peer-to-peer (P2P) network [24]. Each and every network participant can control the network. The network is made up of many connected computers or The blockchain comprises a series of blocks of transactions linked together in a chain, as shown in Figure 6. Client/server architecture is used in traditional client/server systems, and various administrators are in charge of them. On the other hand, blockchain is a distributed, decentralized peer-to-peer (P2P) network [24]. Each and every network participant can control the network. The network is made up of many connected computers or nodes, and the blocks in the chain cannot be changed without the network's approval. Each node in the network has its copy of the digital ledger. The blockchain comprises a series of blocks of transactions linked together in a chain, as shown in Figure 6. Client/server architecture is used in traditional client/server systems, and various administrators are in charge of them. On the other hand, blockchain is a distributed, decentralized peer-to-peer (P2P) network [24]. Each and every network participant can control the network. The network is made up of many connected computers or nodes, and the blocks in the chain cannot be changed without the network's approval. Each node in the network has its copy of the digital ledger.

**Transaction Counter Figure 6.** Structure of a blockchain. **Figure 6.** Structure of a blockchain.

as follows:

**TX TX TX TX TX TX TX TX TX TX** The main constituents of a blockchain and the associated terminology are described as follows: The main constituents of a blockchain and the associated terminology are described as follows:

**Transaction Counter**

**Parent Block Hash**

**Figure 6.** Structure of a blockchain. The main constituents of a blockchain and the associated terminology are described 1. Block: In a blockchain, pointers and linked list data structures are utilized to represent blocks. Using a linked list, the blocks are sorted in a logical order and aligned up with one another. A block is a data set containing transaction information like timestamps and links to previous blocks and is produced using a secure hash tech-1. Block: In a blockchain, pointers and linked list data structures are utilized to represent blocks. Using a linked list, the blocks are sorted in a logical order and aligned up with one another. A block is a data set containing transaction information like timestamps and links to previous blocks and is produced using a secure hash technique.

nique. The location of the next block is indicated via pointers. Every block is divided

into two sections: the block header and the block body.

(i.) Block version: specifies which set of block validation criteria should be used. (ii.) Merkle tree root hash: the sum of all transactions in the frame's hash value.

(i.) Block version: specifies which set of block validation criteria should be used. (ii.) Merkle tree root hash: the sum of all transactions in the frame's hash value.

up with one another. A block is a data set containing transaction information like timestamps and links to previous blocks and is produced using a secure hash technique. The location of the next block is indicated via pointers. Every block is divided

into two sections: the block header and the block body.

The block header has the following fields:

The location of the next block is indicated via pointers. Every block is divided into two sections: the block header and the block body. (iii.) Timestamp: from January 1, 1970, the current time is expressed in seconds in univer-

The block header has the following fields: sal time.

	- (iv.) nBits: a valid block hash's goal threshold. ber of transactions stored in a block is determined by the block size and the transaction
	- (v.) Nonce: a 4-byte field that starts with 0 and rises for each hash computation.
	- (vi.) Parent block hash: a 256-bit hash value that refers to the block before it. size.

A transaction counter and transactions make up the block body. The maximum number of transactions stored in a block is determined by the block size and the transaction size. 2 Public and Private keys: Blockchain is a constantly increasing network of interconnected and secured blocks using cryptographic processes [25]. To validate transac-


### *2.2. Types of Blockchain 2.2. Types of Blockchain*

The type of a blockchain depends on the nature of the application. There are three types of blockchains: public, private, and consortium [26]. These three types of blockchains are represented along with their properties in Figure 7. The type of a blockchain depends on the nature of the application. There are three types of blockchains: public, private, and consortium [26]. These three types of blockchains are represented along with their properties in Figure 7.

There is no control over a permissionless or public blockchain. Anyone may access the network and read or write data. Permissioned ledgers, on the other hand, are only accessible to network users who have been authenticated. Since they are encrypted with

**Figure 7.** Types of blockchains and their properties. **Figure 7.** Types of blockchains and their properties.

*2.3. Characteristics of Blockchain*

chains are combined in consortium blockchains.

There is no control over a permissionless or public blockchain. Anyone may access the network and read or write data. Permissioned ledgers, on the other hand, are only accessible to network users who have been authenticated. Since they are encrypted with a private key, everyone cannot read the blocks. The properties of public and private blockchains are combined in consortium blockchains. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 7 of 35 A blockchain is a decentralized network, and unlike a centralized system, the trans-

### *2.3. Characteristics of Blockchain* actions are validated by the nodes in the network [27]. The identity of the nodes in the

A blockchain is a decentralized network, and unlike a centralized system, the transactions are validated by the nodes in the network [27]. The identity of the nodes in the network remains unanimous, and once a transaction is validated by the nodes and added to the blockchain, it is impossible to reverse the transaction. Thus, the blockchain is immutable. The various other characteristics of a blockchain are depicted in Figure 8. network remains unanimous, and once a transaction is validated by the nodes and added to the blockchain, it is impossible to reverse the transaction. Thus, the blockchain is immutable. The various other characteristics of a blockchain are depicted in Figure 8.

**Figure 8.** Characteristics of a blockchain.

**Figure 8.** Characteristics of a blockchain. Although blockchain technology has gained traction in future Internet systems, several difficulties must be properly addressed. Expertise in blockchain technology is critical, as the technology is still in the nascent stages. Adoption of BCT provides promised benefits in various fields, but the high initial infrastructure costs are a big worry for businesses. Although blockchain technology has gained traction in future Internet systems, several difficulties must be properly addressed. Expertise in blockchain technology is critical, as the technology is still in the nascent stages. Adoption of BCT provides promised benefits in various fields, but the high initial infrastructure costs are a big worry for businesses. The deployment of blockchain technology is also influenced by privacy and security concerns. Scalability and legal requirements are also significant obstacles to its implementation.

#### The deployment of blockchain technology is also influenced by privacy and security con-**3. Blockchain for Smart Grid**

lished in journals.

cerns. Scalability and legal requirements are also significant obstacles to its implementation. **3. Blockchain for Smart Grid** Blockchain technology has much potential to transform applications by creating Blockchain technology has much potential to transform applications by creating more trust and increasing decentralization. Despite its rapid growth, its advantages are not being aggressively exploited by the SG applications. The number of articles published on blockchain from the perspective of the various SG applications is shown in Figure 9. These

more trust and increasing decentralization. Despite its rapid growth, its advantages are

on blockchain from the perspective of the various SG applications is shown in Figure 9. These statistics were taken from the Scopus database and considered only articles pub-

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 8 of 35

statistics were taken from the Scopus database and considered only articles published in journals.

### **Figure 9.** Publication statistics on blockchain for SG application. **Figure 9.** Publication statistics on blockchain for SG application.

Blockchain is widely adopted for energy management applications in an SG. Blockchain is also widely used for SCs, EVs, and MGs. SPAs, responsible for the wide area monitoring and control of the grid, are not employing blockchain technology for decentralizing the process. Only four conference articles reported the use of blockchain technol-Blockchain is widely adopted for energy management applications in an SG. Blockchain is also widely used for SCs, EVs, and MGs. SPAs, responsible for the wide area monitoring and control of the grid, are not employing blockchain technology for decentralizing the process. Only four conference articles reported the use of blockchain technology for SPA. In this section, blockchain technology will be explored from the perspective of these applications.

### ogy for SPA. In this section, blockchain technology will be explored from the perspective *3.1. Blockchain for Synchrophasor Application*

of these applications. *3.1. Blockchain for Synchrophasor Application* The major outages across the globe, such as those in Brazil in February 2011, the Pa-The major outages across the globe, such as those in Brazil in February 2011, the Pacific Southwest in September 2011, India in July 2012, Vietnam in May 2013, the Philippines in June 2013, Bangladesh in November 2014, etc., have necessitated the wide-area measurement system (WAMS) in the SG [28,29]. The WAMS is a comprehensive solution to monitor, control, and maintain the SG by incorporating the state-of-the-art infrastructure, emerging technology, and tools.

cific Southwest in September 2011, India in July 2012, Vietnam in May 2013, the Philippines in June 2013, Bangladesh in November 2014, etc., have necessitated the wide-area measurement system (WAMS) in the SG [28,29]. The WAMS is a comprehensive solution to monitor, control, and maintain the SG by incorporating the state-of-the-art infrastructure, emerging technology, and tools. Recently, synchrophasor technology emerged as a viable solution for the WAMS. The synchrophasor technology enables WAMS to monitor, control, and coordinate the SG in real-time and precisely [30]. The fundamental architecture of the synchrophasor measure-Recently, synchrophasor technology emerged as a viable solution for the WAMS. The synchrophasor technology enables WAMS to monitor, control, and coordinate the SG in real-time and precisely [30]. The fundamental architecture of the synchrophasor measurement system involves a phasor measurement unit (PMU), phasor data concentrator (PDC), and the communication network [31]. The PMUs are high-speed sensors that monitor the grid in real-time by measuring the grid voltages and currents. These measurements are time-synchronized using the global positioning system (GPS) and communicated to the PDC, which acts as an aggregator. The time-synchronized measurements of PMUs are referred to as synchrophasor data.

ment system involves a phasor measurement unit (PMU), phasor data concentrator (PDC), and the communication network [31]. The PMUs are high-speed sensors that monitor the grid in real-time by measuring the grid voltages and currents. These measurements are time-synchronized using the global positioning system (GPS) and communicated to the PDC, which acts as an aggregator. The time-synchronized measurements of The communication network acts as a backbone since it provides the infrastructure for communicating synchrophasor data between PMUs and PDCs [32]. The more generic architecture of WAMS comprises decentralized architecture where the devices are hierarchically arranged. The decentralized hierarchical architecture of the WAMS with three levels of hierarchy is shown in Figure 10. A local PDC may be located close to the microgrids, aggregating synchrophasor data from several PMUs in a power grid. Further, there may be a master PDC that aggregates data from several local PDCs. Finally, the data from several

the regional level, which is the highest level in the proposed hierarchy.

The communication network acts as a backbone since it provides the infrastructure

chically arranged. The decentralized hierarchical architecture of the WAMS with three levels of hierarchy is shown in Figure 10. A local PDC may be located close to the microgrids, aggregating synchrophasor data from several PMUs in a power grid. Further, there may be a master PDC that aggregates data from several local PDCs. Finally, the data from several master PDCs may be aggregated by a PDC known as a super PDC located at

PMUs are referred to as synchrophasor data.

master PDCs may be aggregated by a PDC known as a super PDC located at the regional level, which is the highest level in the proposed hierarchy.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 9 of 35

### **Figure 10.** Hierarchy in a WAMS. **Figure 10.** Hierarchy in a WAMS.

The data pertaining to the health of the grid can be used in WAMS for state estimation, stability analysis, situational awareness, etc., of the SG and its other operational-related functionalities. However, such data, typically referred to as synchrophasor data, can be exploited by cyber-attacks such as denial of service (DoS), distributed denial of service (DDoS), false data injection, spoofing, data tampering, etc. [33]. These attacks put the WAMS at risk, and its efficacy becomes questionable. The risk identification and assessment of smart grids is thoroughly discussed by Jha et al. in [34], where the authors considered risk assessment analysis of smart grid communication networks. The blockchain can be used with synchrophasor technology to mitigate the risk of cyber-attacks in a The data pertaining to the health of the grid can be used in WAMS for state estimation, stability analysis, situational awareness, etc., of the SG and its other operational-related functionalities. However, such data, typically referred to as synchrophasor data, can be exploited by cyber-attacks such as denial of service (DoS), distributed denial of service (DDoS), false data injection, spoofing, data tampering, etc. [33]. These attacks put the WAMS at risk, and its efficacy becomes questionable. The risk identification and assessment of smart grids is thoroughly discussed by Jha et al. in [34], where the authors considered risk assessment analysis of smart grid communication networks. The blockchain can be used with synchrophasor technology to mitigate the risk of cyber-attacks in a WAMS. Additionally, blockchain technology can simultaneously enhance the robustness, reliability, and integrity of the synchrophasor data by incorporating a decentralized peer-to-peer approach to communicate synchrophasor data in a WAMS.

### WAMS. Additionally, blockchain technology can simultaneously enhance the robustness, reliability, and integrity of the synchrophasor data by incorporating a decentralized peer-3.1.1. Blockchain Architecture for SPA

to-peer approach to communicate synchrophasor data in a WAMS. The blockchain architecture for the SPA in an SG will consist of three fundamental components:

	- The blockchain architecture for the SPA in an SG will consist of three fundamental 2. A shared ledger containing the synchrophasor data collected by all the member nodes.

1. The member nodes, which are the PMUs or the PDC. Each node generates its synchrophasor data and shares it using the IEEE C37.118-2 [35]. 2. A shared ledger containing the synchrophasor data collected by all the member nodes. 3. A peer-to-peer distributed network between the member nodes. The architecture of a blockchain for SPAs is shown in Figure 11. As shown in the The architecture of a blockchain for SPAs is shown in Figure 11. As shown in the figure, the PMUs are connected in a fashion to create a distributed peer-to-peer network where all PMUs are enabled as member nodes. Each PMU is responsible for collectively updating the shared ledger. The synchrophasor data from a PMU is referred to as a synchrophasor transaction. The synchrophasor transactions are generated by PMUs which can be verified using authentication methods such as the elliptic curve digital signature algorithm. Despite this authentication, it is quite possible that the false identity of a PMU can be created to obtain access to the network causing danger to the resources. Such an attack can be

figure, the PMUs are connected in a fashion to create a distributed peer-to-peer network where all PMUs are enabled as member nodes. Each PMU is responsible for collectively

chrophasor transaction. The synchrophasor transactions are generated by PMUs which can be verified using authentication methods such as the elliptic curve digital signature algorithm. Despite this authentication, it is quite possible that the false identity of a PMU can be created to obtain access to the network causing danger to the resources. Such an attack can be mitigated using device identity validation methods such as the Bloom filter-

based PMU identity validation approach.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 10 of 35

mitigated using device identity validation methods such as the Bloom filter-based PMU identity validation approach.

PDC

**Figure 11.** Blockchain architecture for SPA at the level of PMUs. **Figure 11.** Blockchain architecture for SPA at the level of PMUs. The PMUs are connected in a fashion to create distributed peer-to-peer network. Each

The PMUs are connected in a fashion to create distributed peer-to-peer network. Each PMU in the distributed peer-to-peer networks acts as a member which mines the block, where the synchrophasor transactions are included in a block. The contents of the block are shown in Figure 12. The PMUs are connected in a fashion to create distributed peer-to-peer network. Each PMU in the distributed peer-to-peer networks acts as a member which mines the block, where the synchrophasor transactions are included in a block. The contents of the block are shown in Figure 12. PMU in the distributed peer-to-peer networks acts as a member which mines the block, where the synchrophasor transactions are included in a block. The contents of the block are shown in Figure 12.

Each block is generated using the IEEE C37.118-2 standard. The block is encapsulated

approach can be used for consensus execution as it converges quickly without compromising the integrity of the synchrophasor transactions. Further, PMUs can follow consensus based on proof-of-work (PoW), where nonce is searched, which is a random number.

Each block is generated using the IEEE C37.118-2 standard. The block is encapsulated with other protocols for communication over the TCP/IP network. The PMUs are respon-

sensus execution and validation in blockchain technology. For SPA, the Markel tree-based approach can be used for consensus execution as it converges quickly without compromising the integrity of the synchrophasor transactions. Further, PMUs can follow consensus based on proof-of-work (PoW), where nonce is searched, which is a random number.

Encapsulation of block TCP/IP protocol suite

Encapsulation of block TCP/IP protocol suite **Figure 12.** Contents of the block in a blockchain-based SPA. **Figure 12.** Contents of the block in a blockchain-based SPA. **Figure 12.** Contents of the block in a blockchain-based SPA.

Each block is generated using the IEEE C37.118-2 standard. The block is encapsulated with other protocols for communication over the TCP/IP network. The PMUs are responsible for consensus execution and block validation. There are several approaches for consensus execution and validation in blockchain technology. For SPA, the Markel tree-based approach can be used for consensus execution as it converges quickly without compromising the integrity of the synchrophasor transactions. Further, PMUs can follow consensus based on proof-of-work (PoW), where nonce is searched, which is a random number. When all the synchrophasor transactions grouped in a block are validated and PoW is completed, only a block is considered successfully mined by the PMU. On validating with PoW, the newly created block is appended to the existing chain to update the blockchain. The first block in the blockchain is a genesis block, which a PMU in the network can generate. It is imperative that any PMU can validate any number of blocks and receives the whole existing blockchain from executing the consensus and PoW. The decentralization can also help remove the PDC, and the PMUs themselves can take commensurate actions based on the measurements available from other PMUs. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 11 of 35 When all the synchrophasor transactions grouped in a block are validated and PoW is completed, only a block is considered successfully mined by the PMU. On validating with PoW, the newly created block is appended to the existing chain to update the blockchain. The first block in the blockchain is a genesis block, which a PMU in the network can generate. It is imperative that any PMU can validate any number of blocks and receives the whole existing blockchain from executing the consensus and PoW. The decentralization can also help remove the PDC, and the PMUs themselves can take commensurate actions based on the measurements available from other PMUs.

#### 3.1.2. Challenges and Solutions for the Implementation of Blockchain-Based SPA 3.1.2. Challenges and Solutions for the Implementation of Blockchain-Based SPA

PMUs operate at a very high rate, typically 30–60 samples per second in a timesynchronized manner. Hence, the additional functionalities of creating the blocks and validating burden the device and hampers the granularity of its measurements. An alternative solution to this problem is implementing the blockchain at a higher level in the hierarchy, i.e., at the local PDCs. The architecture for implementing the blockchain at the level of local PDCs is shown in Figure 13. PMUs operate at a very high rate, typically 30–60 samples per second in a time-synchronized manner. Hence, the additional functionalities of creating the blocks and validating burden the device and hampers the granularity of its measurements. An alternative solution to this problem is implementing the blockchain at a higher level in the hierarchy, i.e., at the local PDCs. The architecture for implementing the blockchain at the level of local PDCs is shown in Figure 13.

**Figure 13.** Blockchain architecture for SPA at the level of local PDCs. **Figure 13.** Blockchain architecture for SPA at the level of local PDCs.

SPAs are mission-critical, so it becomes computationally intensive to validate all the transactions. A solution to this problem is to terminate the chain at periodic intervals and start a new chain. This reduces the secureness of the chain, so additional measures will be needed to maintain the security. Because of the problems of the mission-critical nature of the application and the high data rate of the PMU, not many works are available on this topic. SPAs are mission-critical, so it becomes computationally intensive to validate all the transactions. A solution to this problem is to terminate the chain at periodic intervals and start a new chain. This reduces the secureness of the chain, so additional measures will be needed to maintain the security. Because of the problems of the mission-critical nature of the application and the high data rate of the PMU, not many works are available on this topic.

A smart house is an integrated Internet of Things (IoT) domicile that provides users security, health, comfort, and a higher standard of life, among other benefits. People's life

attention of consumers and device makers. Although intelligent homes provide significant benefits to homeowners and other interested parties, they are vulnerable to harmful

*3.2. Blockchain for Home Automation*

### *3.2. Blockchain for Home Automation*

A smart house is an integrated Internet of Things (IoT) domicile that provides users security, health, comfort, and a higher standard of life, among other benefits. People's life and independent living are made easier with smart home solutions. They provide valuable capabilities such as behavior tracking and safety evaluations, which have drawn the attention of consumers and device makers. Although intelligent homes provide significant benefits to homeowners and other interested parties, they are vulnerable to harmful cyberattacks that risk users' safety and privacy [36]. Traditional solutions to such dangers exist, but they are extremely centralized and prone to large-scale attacks. As a result, the adaptability and scalability needed for effective utilization in the cutting-edge field of autonomous smart home applications and facilities are absent. Several clever technologies make life easier for individuals. Such programs generate enormous volumes of data. The archiving of this ever-changing material into repositories raises security problems. In cybersecurity technologies with remote connectivity and data transmission, blockchain has performed well. Thus, it is being employed for home automation applications. cyber-attacks that risk users' safety and privacy [36]. Traditional solutions to such dangers exist, but they are extremely centralized and prone to large-scale attacks. As a result, the adaptability and scalability needed for effective utilization in the cutting-edge field of autonomous smart home applications and facilities are absent. Several clever technologies make life easier for individuals. Such programs generate enormous volumes of data. The archiving of this ever-changing material into repositories raises security problems. In cybersecurity technologies with remote connectivity and data transmission, blockchain has performed well. Thus, it is being employed for home automation applications. 3.2.1. Blockchain Architecture for HA Home automation involves several smart devices, such as smart TVs, lights, etc.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 12 of 35

#### 3.2.1. Blockchain Architecture for HA These devices monitor and control the various parameters of the house, which operate

Home automation involves several smart devices, such as smart TVs, lights, etc. These devices monitor and control the various parameters of the house, which operate independently or are coordinated by a user. The interconnectivity of these smart devices is required to achieve the objective of HA. The interoperability challenges between the smart devices are handled using an IoT gateway. Users from one home cannot control the devices of another home to avoid a security breach. The service provider is responsible for providing necessary recommendations to the user for controlling smart devices based on prediction algorithms. The service provider can use machine learning algorithms for better recommendations or predictions. The blockchain network is used to connect different users and service providers to enhance security in the HA [37]. The blockchain network may be built using Ethereum or Hyperledger. The general architecture of blockchain for HA is shown in Figure 14. independently or are coordinated by a user. The interconnectivity of these smart devices is required to achieve the objective of HA. The interoperability challenges between the smart devices are handled using an IoT gateway. Users from one home cannot control the devices of another home to avoid a security breach. The service provider is responsible for providing necessary recommendations to the user for controlling smart devices based on prediction algorithms. The service provider can use machine learning algorithms for better recommendations or predictions. The blockchain network is used to connect different users and service providers to enhance security in the HA [37]. The blockchain network may be built using Ethereum or Hyperledger. The general architecture of blockchain for HA is shown in Figure 14.

The user within the house can control the entities within his home; he cannot have access to the entities present in another smart home. The various devices in the home can be directly connected to the blockchain network through the gateway. The data from the

mechanism of the blockchain. The service provider can analyze the data and send suggestions to the users, but he cannot directly control the devices in the smart home. This architecture can be customized based on the user's specific requirements by the service provider. The various devices in the home can be directly connected to the blockchain network through the gateway. The data from the devices can be placed into the blocks, which

are then chained together using the hashing mechanism of the blockchain.

**Figure 14.** The architecture of blockchain for HA. **Figure 14.** The architecture of blockchain for HA.

The user within the house can control the entities within his home; he cannot have access to the entities present in another smart home. The various devices in the home can be directly connected to the blockchain network through the gateway. The data from the devices can be placed into the blocks, which are then chained together using the hashing mechanism of the blockchain. The service provider can analyze the data and send suggestions to the users, but he cannot directly control the devices in the smart home. This architecture can be customized based on the user's specific requirements by the service provider. The various devices in the home can be directly connected to the blockchain network through the gateway. The data from the devices can be placed into the blocks, which are then chained together using the hashing mechanism of the blockchain.

### 3.2.2. State-of-the-Art on Blockchain for HA

The literature on blockchain for HA applications discusses access control mechanisms, homecare systems, utility payment services, etc. These works are summarized in Table 2.



### 3.2.3. Challenges and Solutions for the Implementation of Blockchain-Based HA

Various blockchain systems are being used for HA applications [46]. These systems have their specific data format, and their interoperability is challenging. Additionally, the consensus algorithms used by these systems are different. For seamless interaction, standardization of blockchain systems is required. Another challenge to implementing blockchain for HA applications is the real-time analytics of streaming data. The data have to be processed and analyzed in real-time. For example, an intruder detection system requires real-time face detection. Processing blockchains for real-time applications is challenging. A possible solution is to use a lightweight framework for this application.

### *3.3. Blockchain for Advanced Metering Infrastructure*

The heart of the AMI is a smart meter used to collect, monitor, and communicate the data related to energy consumption corresponding to every user. The meter data are used differently by different entities. For example, the grid operator can use this data for load forecasting and planning, and the market operator can use smart meter data for dynamic pricing and billing. On the other hand, the users can use such data to manage their electricity usage. Whereas AMI provides ample advantages, secure AMI data transaction is

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 14 of 35

challenging. The blockchain-based AMI plays an important role in achieving this objective. A generic framework for implementing AMI using blockchain is shown in Figure 15.

**Figure 15.** The architecture of blockchain for AMI.

**Figure 15.** The architecture of blockchain for AMI. The smart meters can be directly connected to the blockchain network through the gateway [47]. The data from the meters contains meter IDs and other utility-related information as per the IEC 62056 protocol. These meters are connected to the servers or nodes inside the blockchain network that create the blocks using the data received from the AMIs. These blocks are then shared with all other nodes inside the blockchain-enabled network. This network can only be accessed by the nodes related to the utility center and The smart meters can be directly connected to the blockchain network through the gateway [47]. The data from the meters contains meter IDs and other utility-related information as per the IEC 62056 protocol. These meters are connected to the servers or nodes inside the blockchain network that create the blocks using the data received from the AMIs. These blocks are then shared with all other nodes inside the blockchain-enabled network. This network can only be accessed by the nodes related to the utility center and so should be a private blockchain network. The private blockchain can be used for smart contracts and validations to provide energy utilization transparency without compromising security and privacy.

### so should be a private blockchain network. The private blockchain can be used for smart contracts and validations to provide energy utilization transparency without compromis-Challenges for the Implementation of Blockchain for AMI

ing security and privacy. Challenges for the Implementation of Blockchain for AMI Blockchain has not been used widely for this SG application despite its utility. Researchers have used it to enhance the security of AMI applications. In ref. [48], a lightweight blockchain-based framework was proposed to enhance AMI's security. The frame-Blockchain has not been used widely for this SG application despite its utility. Researchers have used it to enhance the security of AMI applications. In ref. [48], a lightweight blockchain-based framework was proposed to enhance AMI's security. The framework was secure against attacks, and its energy consumption was low. In ref. [49], blockchain was used to preserve the integrity of the customers using AMI. As with the blockchain for HA applications, the blockchain for AMI is also plagued with interoperability and real-time constraints.

#### work was secure against attacks, and its energy consumption was low. In ref. [49], block-*3.4. Blockchain for Electric Vehicles*

*3.4. Blockchain for Electric Vehicles*

their fleet of vehicles.

chain was used to preserve the integrity of the customers using AMI. As with the blockchain for HA applications, the blockchain for AMI is also plagued with interoperability and real-time constraints. The technological evolution of electric vehicles (EVs) and the rapid growth of the smart grid have led to the emergence of new connectivity structures—vehicle-to-grid (V2G) [50]. In the future, the importance of EVs using technologies such as the Internet of

smart grid have led to the emergence of new connectivity structures—vehicle-to-grid (V2G) [50]. In the future, the importance of EVs using technologies such as the Internet of Vehicles (IoV) [51] or the Internet of Things (IoT) [52] will increase, as it offers innumerable advantages, for example logistics companies provide fixed charging stations (CSs) for

Interconnectivity requirements with all technology systems in the real world have

led to the emergence of vehicle-to-everything (V2X) technology [53], using integrated ve-

Vehicles (IoV) [51] or the Internet of Things (IoT) [52] will increase, as it offers innumerable advantages, for example logistics companies provide fixed charging stations (CSs) for their fleet of vehicles. hicle sensor platforms that use the centralization of various functions through an integrated EV server, connected by a series of connectivity devices such as CAN, LIN, Wi-Fi,

Interconnectivity requirements with all technology systems in the real world have led to the emergence of vehicle-to-everything (V2X) technology [53], using integrated vehicle sensor platforms that use the centralization of various functions through an integrated EV server, connected by a series of connectivity devices such as CAN, LIN, Wi-Fi, and Bluetooth technology [54]. The results of V2X performances are based on a series of information on the collection and dissemination of multi-networks and technological capabilities between electric vehicles. and Bluetooth technology [54]. The results of V2X performances are based on a series of information on the collection and dissemination of multi-networks and technological capabilities between electric vehicles. The security factor, the speed of data transfer between interconnected vehicles, and the wide coverage of telecommunications systems led to the emergence of 5G networks and their distribution very quickly in the world [55]. The infrastructure of multi-networks

The security factor, the speed of data transfer between interconnected vehicles, and the wide coverage of telecommunications systems led to the emergence of 5G networks and their distribution very quickly in the world [55]. The infrastructure of multi-networks communication systems through 5G technology has the power to process applications at a superior level. The 5G network drives the V2X protocol, generating many scenarios for data management by promoting the development and integration of blockchain applications [56]. The implementation of blockchain systems in the vehicle-to-everything protocol tends to reinvent intelligent transport systems, leading to high efficiency of transport and road safety services [57]. communication systems through 5G technology has the power to process applications at a superior level. The 5G network drives the V2X protocol, generating many scenarios for data management by promoting the development and integration of blockchain applications [56]. The implementation of blockchain systems in the vehicle-to-everything protocol tends to reinvent intelligent transport systems, leading to high efficiency of transport and road safety services [57]. 3.4.1. Architecture of Blockchain for EVs

### 3.4.1. Architecture of Blockchain for EVs The general blockchain architecture for the EV application is shown in Figure 16. The blockchain-based EVs infrastructure requires regular nodes to capture mobile cars' dy-

The general blockchain architecture for the EV application is shown in Figure 16. The blockchain-based EVs infrastructure requires regular nodes to capture mobile cars' dynamics. These nodes are responsible for smart contracts and block validations, forming the basis of the blockchain. The mobile cars send their data to such regularly placed blockchain nodes or access points. The interconnectivity between mobile electric vehicles and nodes is through WiFi. An ID number uniquely identifies each EV. The data that an EV sends to the access point involve the battery status, vehicle status, bill payment for charging, etc. The data are placed into the blockchain network by the access points as blocks. The various nodes in the blockchain validate the transactions. The blockchain network is also accessible to the transport authority, who can continuously monitor the status of the EVs and send personalized recommendations or warnings to the EV user. However, the transportation authority cannot change the parameters of the EV. namics. These nodes are responsible for smart contracts and block validations, forming the basis of the blockchain. The mobile cars send their data to such regularly placed blockchain nodes or access points. The interconnectivity between mobile electric vehicles and nodes is through WiFi. An ID number uniquely identifies each EV. The data that an EV sends to the access point involve the battery status, vehicle status, bill payment for charging, etc. The data are placed into the blockchain network by the access points as blocks. The various nodes in the blockchain validate the transactions. The blockchain network is also accessible to the transport authority, who can continuously monitor the status of the EVs and send personalized recommendations or warnings to the EV user. However, the transportation authority cannot change the parameters of the EV.

**Figure 16.** Architecture of blockchain for EV applications. ing of the data, cybersecurity, handling of voluminous data, etc. Furqan Jameel et. al. [58] **Figure 16.** Architecture of blockchain for EV applications.

3.4.2. A panoramic Overview of Blockchain for EV

### 3.4.2. A panoramic Overview of Blockchain for EV

Despite abundant opportunities to incorporate blockchain in EV applications, some of the challenges are inherently present in the blockchain-enabled EV system, such as mining of the data, cybersecurity, handling of voluminous data, etc. Furqan Jameel et al. [58] provided a solution for unloading mining tasks in vehicle-to-everything cellular networks. A short block length transmission architecture has been proposed to meet the low-latency requirements for cybersecurity applications of EVs. In practice, finite block length architecture is a fairer approach to modeling blockchain networks. The inspiration for the theoretical application of adopting games defines a type of challenge for solving mining tasks and efficiently unloading them to clusters. The advantage of using blockchain databases ensures good data transfer rates and maintains the vehicles' fairness in the unloading process.

However, a significant disadvantage is the scalability of the data chains within the blockchain, which can be a design problem. Because data security is the main issue in conventional blockchain networks, the impact it has on the process of downloading data into electric vehicles is a real challenge. However, in ref. [59], the authors proposed a new coding sequence—Secure V2X—that capitalizes on the characteristics of the blockchain and the data networks protecting the confidentiality and security of the V2X protocol.

In addition to the benefits that blockchain initiates in low-security areas, confidentiality is the main issue in trading energy in a collective network type peer-to-peer (P2P) (E-trading). In recent years, electric vehicles have received worldwide recognition due to their potential in the green transportation system. The rapid development of technologies in smart communication networks has allowed EVs to relate to the environment. The electricity production costs are constantly decreasing through the implementation of renewable energy sources and smart grids [60]. Thus, the major challenge of peer-to-peer technology, E-trading and D-trading [61], and integration for electric vehicles is the development of a secure communication architecture that maintains data confidentiality and information anonymity. In addition, the objective of the blockchain is to mask trading relationships without compromising data integrity [62].

Various review papers in the literature focus on blockchain technologies applied in the Future Smart Grid [63,64]. Although the technology is considered one with a wide range of advantages, security needs to be assessed systematically to enhance reliability of the SG [65].

Motivated by previous development, Marina Dorokhova et al. [66] proposed integrating electric vehicle charging systems based on blockchain technology. The study is based on a popular blockchain platform, Ethereum, for interconnecting EV infrastructure and real-world infrastructure [67]. The advantage it offers is the crediting in the safety zone of the energy flows between the owners of electric vehicles and the companies that own charging stations. The only barriers that could be removed in the future are the limitations of the blockchain-high transaction costs due to network loads, high power consumption, or transactions that do not change in case of errors.

A case study by Shivam Saxena et al. [68] further demonstrated the need for technoeconomic evaluation of residential energy trading systems. The EV is a part of such system, which can be enhanced through the blockchain. Using blockchain in EVs not only improves the household's participation in the electricity markets but also drastically reduces the negative impact on the energy distribution network [69]. These seminal works are comprehensively summarized in Table 3.


**Table 3.** Summary of works related to blockchain for EVs.

3.4.3. Challenges and Solutions for the Implementation of Blockchain for EVs

The scalability of blockchain data chains, data security in the download process, and confidentiality are challenges that are yet to be addressed. The major challenge of peer-to-peer technologies is the processing of energy transactions and the anonymity of information. The high resource requirement and transaction cost in terms of energy consumption plagued the use of blockchain technology for EV applications with WSN infrastructure. Overcoming these limitations would make blockchain technology the main key factor for EVs. The development of lightweight blockchain algorithms for reaching consensus in real-time can be a probable solution.

### *3.5. Blockchain for Renewable Microgrids*

With every day passing, there is a continuous transition and evolution to a renewable grid that is based on various distributed energy resources such as photovoltaics, fuel cells, microturbines, batteries, etc. These transitions rely on the successful deployment of blockchain technology.

### 3.5.1. Architecture of Blockchain for MGs

The generalized blockchain architecture for the MG application is shown in Figure 17. In general, the power grid of a zone is sprawled over a large geographical area where different MGs are considered. The different MGs are interconnected using the blockchain network. The blockchain network aims to enhance security and privacy in the MG operation without hampering transparency and data integrity. The data block carries information regarding the energy generated, energy to be shared with other microgrids, etc. The data pertaining to the MG are grouped into the blocks where each newly generated block is validated using a consensus algorithm. The block is then placed onto the blockchain network and is added to the blockchain after being validated. The nodes in the blockchain need proper algorithms to reach a consensus on the energy being traded, the price at which the electricity is being traded, etc.

**Figure 17.** Blockchain for microgrids. **Figure 17.** Blockchain for microgrids.

3.5.2. A Panoramic Overview of Blockchain for MG

3.5.2. A Panoramic Overview of Blockchain for MG Early inquiries about the energy sector with the accent on the smart grid and mi-Early inquiries about the energy sector with the accent on the smart grid and microgrids are mainly found in refs. [70–73], where different requirements, technologies, architectures, trends, and cyber security issues are largely debated.

crogrids are mainly found in refs. [70–73], where different requirements, technologies, architectures, trends, and cyber security issues are largely debated. With rising social, economic, political, and environmental concerns and strategies such as increasing power consumption, dealing with the middleman, market liberaliza-With rising social, economic, political, and environmental concerns and strategies such as increasing power consumption, dealing with the middleman, market liberalization, pollution, etc., blockchain is seen as a promising solution in renewable microgrids for efficient operation such as complex point-to-point transactions between producers, traders, and users using elaborate algorithms in order to validate, secure, and record these transactions.

tion, pollution, etc., blockchain is seen as a promising solution in renewable microgrids for efficient operation such as complex point-to-point transactions between producers, traders, and users using elaborate algorithms in order to validate, secure, and record these transactions. The different authors reviewed blockchain in the context of microgrids from several The different authors reviewed blockchain in the context of microgrids from several perspectives. In ref. [74], the need for blockchain, benefits, and challenges was reviewed. In ref. [11], real solutions such as the Brooklyn Micro Grid based on the blockchain environment with the Proof of Work (PoW) mechanism were presented. Other comprehensive reviews can be found in ref. [75] that can serve as quality background research for those who want to propose and implement feasible solutions and methodologies for renewable microgrids based on blockchain technology.

perspectives. In ref. [74], the need for blockchain, benefits, and challenges was reviewed. In ref. [11], real solutions such as the Brooklyn Micro Grid based on the blockchain envi-On the other hand, many works propose different solutions and approaches that use blockchain technology to enhance and improve microgrids and their applications. To start, ref. [76] proposed an approach for using blockchain on the Dominican Republic's electricity

ronment with the Proof of Work (PoW) mechanism were presented. Other comprehensive

who want to propose and implement feasible solutions and methodologies for renewable

blockchain technology to enhance and improve microgrids and their applications. To start, ref. [76] proposed an approach for using blockchain on the Dominican Republic's electricity market, referred to as the main step in empowering automatic management of economic transfers with funds authenticating and supplier's guarantee. The approach presents an economic and energy blockchain-based flow to decentralize the current flows that involve total control through banking operations. Of course, such an approach must face serious economic interests, political regulations, and technological limitations to achieve its goals, but it is seen as a first step in applying blockchain in the electricity sector.

On the other hand, many works propose different solutions and approaches that use

In ref. [77], a local energy market model using private blockchain via home energy

management and demurrage mechanisms was presented. In the proposed model, there are three major actors: a small community (several microgrids) that uses photovoltaic systems as renewable energy as the prosumers; the consumers; and the main grid. Using

microgrids based on blockchain technology.

market, referred to as the main step in empowering automatic management of economic transfers with funds authenticating and supplier's guarantee. The approach presents an economic and energy blockchain-based flow to decentralize the current flows that involve total control through banking operations. Of course, such an approach must face serious economic interests, political regulations, and technological limitations to achieve its goals, but it is seen as a first step in applying blockchain in the electricity sector.

In ref. [77], a local energy market model using private blockchain via home energy management and demurrage mechanisms was presented. In the proposed model, there are three major actors: a small community (several microgrids) that uses photovoltaic systems as renewable energy as the prosumers; the consumers; and the main grid. Using Critical Peak Price (CPP) and Real-Time Price (RTP) schemes, simulations showed that costs were significantly reduced. Moreover, ref. [78] proposed a blockchain-based decentralized market mechanism to establish the price using auction methods. However, this is plagued with two limitations: difficulty in selling the oversupply of energy through auction and big challenges in ensuring privacy and security. Another solution for P2P energy trading was presented in ref. [79] by implementing the blockchain-based decentralized energy flexibility market for P2P transactions among prosumers. Two additional frameworks, first as decentralized blockchain-based and second as semi-decentralized, can be found in ref. [80], where the P2P energy trading subject was analyzed.

A more applied approach is in ref. [81], which proposed a method for an effective P2P blockchain-based energy market between a microgrid and the smart grid (IEEE 24-bus test system) where the function of distributed consensus algorithm was evaluated in the presence of Fault Data Injection Attack (FDIA). The main findings of this paper showed that the consensus process keeps running even if the case of a cyber-attack and the output response of the P2P market is very close to the centralized energy market. Maintaining the idea of the applied solution, the authors in ref. [82] suggested a model for an integrated energy management platform based on blockchain technology and, at the same time, implement a bilateral trading mechanism with simulation results showing significant optimization of the energy flow in a microgrid. Another model for blockchain-based energy management was suggested in ref. [83], where a Pythagorean fuzzy method was used in choosing the best solution for energy production, distribution, and waste control.

Further, in ref. [84], another P2P energy trading mechanism between microgrids based on the same technology using a fuzzy meta-heuristic approach as a pricing solution was presented with results showing increased profitability and reduced CO<sup>2</sup> emissions. Additionally, the fusion of the electricity market and blockchain was studied in ref. [85], where transactions were highlighted using multi-agent cooperation and sharing platform based on the Ethereum private blockchain, with results revealing several benefits such as transparent transactions and intelligent mutual trust.

Going deeper and deeper into the heart of the topic of this section, we arrive at the point where blockchain applications variates in terms of the constructive technology that microgrids are built on, this referring to AC microgrids, DC microgrids, or hybrid AC-DC MGs [86–90]. First, blockchain was used in ref. [86] to increase the security for interconnected hybrid AC-DC microgrids using a modified sine cosine algorithm to achieve the optimal decision in the shortest time and with high accuracy. The approaches in refs. [87] and [88] are based on the blockchain technology for energy management concerning DC and hybrid AC-DC microgrids using different strategies such as fuzzy logic control or the whale optimization algorithm. These seminal works are comprehensively summarized in Table 4.


### **Table 4.** Material summary—blockchain for renewable microgrids.

### 3.5.3. Challenges for Implementation of Blockchain for Microgrids

Like the numerous advantages, many challenges must be overcome in the blockchainbased renewable microgrids [91,92]. These challenges refer to technological constraints, economic aspects, social uncertainties, environmental concerns, political and institutional limitations, and law, regulations, norms, or end-to-end privacy and security.

A feasible and efficient balance between key features such as security, energy management, constraints, and costs is still challenging. Different consortiums operate different microgrids, so it is important to analyze and decide on the correct algorithm or methods to use, the best technology, the most suitable investor, and a very well-trained team.

### *3.6. Blockchain for Smart City*

With the development and use of blockchain technology, the Internet of Things (IoT), and Cloud Computing, rapid evolution can be observed in the smart city paradigm.

### 3.6.1. A panoramic Overview on Blockchain for SC

In refs. [93,94], some of the problems related to smart city transportation were debated. These works demonstrated that there are concerns in rethinking the transformations of

localities in terms of improvement of public transport and logistics [95,96], water supply [97], green energy [98], environment [99], health [100,101], education [102–105], and economics [106–109] by using the blockchain, which offers the possibility to use distributed stored data, and performs transactions without intermediaries between producers and beneficiaries [106,109] without data security problems [107]. The blockchain architecture [93,94] is the one that will strengthen the importance of using smart contracts in the development of transactions between the parties. These contracts are triggered by operations (agreements) between the parties or are determined by sensors, actuators, or IoT tags [97]. So, the blockchain and smart contracts are the ones that contribute to the transformation of localities into smart cities, finding the optimal adequacy in the development of logistics, energy, environment, water quality, health, etc. Some seminal results of the prospective of blockchain on health care are summarized in Table 5, whereas its applications in other smart city domains are summarized in Table 6.

**Table 5.** Summary of literature on blockchain for smart city health care system.


**Table 6.** Summary of literature on Blockchain for Smart City.



### **Table 6.** *Cont.*

### 3.6.2. Architecture of Blockchain for SCs

The prospects of the IoT determine the smart city architecture, the multitude of sensors and smart objects that help collect data collected from public infrastructure, public access to data, increasing the quality of services and costs of environmental protection, and economic development. The general architecture of blockchain for SCs is shown in Figure 18. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 23 of 35

**Figure 18.** Blockchain architecture for SC applications. ing. EMS aims to ensure reliable energy trading in real-time, including all energy market **Figure 18.** Blockchain architecture for SC applications.

to ensure the smooth operation of the services.

3.6.3. Challenges for Blockchain for SCs

networks in the SC is a challenge.

*3.7. Blockchain for Energy Management System*

chain technology can be used for this purpose.

3.7.1. Architecture of Blockchain for Energy Management System

quirements of a single service. Smart devices, i.e., smart transport vehicles, smart sensors in homes, smart monitoring devices in hospitals, etc., generate data that is put into a blockchain related to its service. Proper protocols and blockchain frameworks will be needed

Using the same blockchain network for all the services in the smart city is not feasible. Therefore, multiple blockchain networks have to be used depending on the size of the city

SC has many different entities. The blockchain network used by the SCs' various entities varies with the application type. These applications have diverse requirements. For example, in the case of smart transportation, the devices are changing their locations, and in the case of smart lighting, the devices are static. The blockchain architecture must be planned according to the nature of the application. Additionally, the entities are spread over a large geographical area, and to meet the criterion for real-time analysis, the blockchain must be fast and secure. Additionally, interoperability between different blockchain

Developing and implementing the distributed system, both in production and consumption and energy marketing, brought new benefits to producers and consumers. Moreover, the increasing energy use from wind turbines and photovoltaic panels necessitated changing the energy market's architecture and secure energy transactions. Block-

The blockchain has enormous potential in the transaction related to energy market-

Using the same blockchain network for all the services in the smart city is not feasible. Therefore, multiple blockchain networks have to be used depending on the size of the city and the nature of smart services provided. Each blockchain network may cater to the requirements of a single service. Smart devices, i.e., smart transport vehicles, smart sensors in homes, smart monitoring devices in hospitals, etc., generate data that is put into a blockchain related to its service. Proper protocols and blockchain frameworks will be needed to ensure the smooth operation of the services.

### 3.6.3. Challenges for Blockchain for SCs

SC has many different entities. The blockchain network used by the SCs' various entities varies with the application type. These applications have diverse requirements. For example, in the case of smart transportation, the devices are changing their locations, and in the case of smart lighting, the devices are static. The blockchain architecture must be planned according to the nature of the application. Additionally, the entities are spread over a large geographical area, and to meet the criterion for real-time analysis, the blockchain must be fast and secure. Additionally, interoperability between different blockchain networks in the SC is a challenge.

### *3.7. Blockchain for Energy Management System*

Developing and implementing the distributed system, both in production and consumption and energy marketing, brought new benefits to producers and consumers. Moreover, the increasing energy use from wind turbines and photovoltaic panels necessitated changing the energy market's architecture and secure energy transactions. Blockchain technology can be used for this purpose.

### 3.7.1. Architecture of Blockchain for Energy Management System *Sustainability* **2022**, *14*, x FOR PEER REVIEW 24 of 35

The blockchain has enormous potential in the transaction related to energy marketing. EMS aims to ensure reliable energy trading in real-time, including all energy market entities such as generation systems (both renewable energy sources and non-renewable energy sources), customer systems, grid operators, etc. The blockchain architecture for EMS is shown in Figure 19. entities such as generation systems (both renewable energy sources and non-renewable energy sources), customer systems, grid operators, etc. The blockchain architecture for EMS is shown in Figure 19.

The SG envisioned integrating renewable energy sources with conventional energy

On the one hand, it reduces the burden of the generation system, but on the other hand, it becomes vital to monitor the energy trading between users. Additionally, security and privacy in the energy trading market are equally important. To achieve this objective,

The blockchain aims to integrate all domains of the SG, such as the generation system, operation system, the consumer system, regulator, and control center, using the blockchain network as shown in Figure 19. The blockchain-based EMS ensures the security and privacy of energy transactions through its distributed approach, interoperability, and smart contracts. The private blockchain can implement data permissions and selective consortium access to ensure security and privacy in energy trading. Due to distributed approach, blockchain-based EMS augments the transparency without compromising pri-

The research on blockchain for EMS is gaining momentum and has been discussed in many recent works. As the amount of energy increases in trading, the greater the difficulties. So, this trading system needs to be controlled very carefully. An online energy transaction management model was proposed in ref. [111] where users can obtain information on their own trading and consumption through energetic transaction. For the transaction to be secure and fast, a payment plan was proposed based on the loan's value.

also fall in the consumer domain of the SG. However, the consumer domain entities act not only as the electricity consumer but also as the electricity producer. Such consumers can be referred to as prosumers. When surplus electricity is available at prosumers, it is

**Figure 19.** The architecture of blockchain-based EMS. Jiawei Yang et al., in ref. [112], propose a public pricing scheme based on the blockchain. **Figure 19.** The architecture of blockchain-based EMS.

blockchain can be integrated into the EMS.

vacy in peer-to-peer energy trading.

3.7.2. A Panoramic Overview of Blockchain for EMS

contributed to reducing the burden on the generation system.

The SG envisioned integrating renewable energy sources with conventional energy sources as generation sources. On the consumer side, there are individual homes, residential buildings, offices, market complexes, etc. In addition to these, the EV charging stations also fall in the consumer domain of the SG. However, the consumer domain entities act not only as the electricity consumer but also as the electricity producer. Such consumers can be referred to as prosumers. When surplus electricity is available at prosumers, it is contributed to reducing the burden on the generation system.

On the one hand, it reduces the burden of the generation system, but on the other hand, it becomes vital to monitor the energy trading between users. Additionally, security and privacy in the energy trading market are equally important. To achieve this objective, blockchain can be integrated into the EMS.

The blockchain aims to integrate all domains of the SG, such as the generation system, operation system, the consumer system, regulator, and control center, using the blockchain network as shown in Figure 19. The blockchain-based EMS ensures the security and privacy of energy transactions through its distributed approach, interoperability, and smart contracts. The private blockchain can implement data permissions and selective consortium access to ensure security and privacy in energy trading. Due to distributed approach, blockchain-based EMS augments the transparency without compromising privacy in peerto-peer energy trading.

### 3.7.2. A Panoramic Overview of Blockchain for EMS

The research on blockchain for EMS is gaining momentum and has been discussed in many recent works. As the amount of energy increases in trading, the greater the difficulties. So, this trading system needs to be controlled very carefully. An online energy transaction management model was proposed in ref. [111] where users can obtain information on their own trading and consumption through energetic transaction. For the transaction to be secure and fast, a payment plan was proposed based on the loan's value. Jiawei Yang et al., in ref. [112], propose a public pricing scheme based on the blockchain. The price is influenced by the share borne by the miners who are taxed with a part of the income for the power losses. The smart contract was created, although the testing was conducted with only 27 prosumers. The biggest problem when we talk about the price of energy in the trading process by using the blockchain is the high energy consumption used by this technology, which was resolved in ref. [113].

The security challenge was dealt with by Yi Zhang et al. in ref. [114] for users and energy flow. In ref. [115], S.N.G. Gourisetti et al. proposed an energy market framework using the online double auction. The authors explain the benefits and usefulness of blockchain technology and its use for transactional energy. The prognosis is that this technology and the implementation of smart contracts in stages can minimize and eliminate the challenge elements in key and certificate management. The authors stated that they expect that lower energy consumption will be achieved if users are more receptive.

Blockchain technology has allowed smart meters with enhanced security and privacy features. Further, a platform to monitor the energy generated from renewable sources by storing and trading energy between residents and network services of users was proposed in ref. [116]. The possibility of trading renewable energy generated by private producers using blockchain technology was shown in ref. [117]. The authors offer a high scalability solution based on smart contracts, which will not harm the decentralized system and data security. The costs of transactions made in this way will be lower compared with current blockchain costs. A cloud services platform for energy trading was proposed in ref. [118] by Lei Wang et al. Both users and suppliers participated in the platform, and the intelligent contract for trading between the parties was created. Antchain is used to make smart contracts, trade, and use the services offered by the cloud. The evolution over time of blockchain technology in the energy trading sector and the issues that stop the application of this technology were presented in ref. [119]. In ref. [120], the authors present the blockchain used by customers to pay for energy consumption. Some seminal work in this direction is comprehensively analyzed in Table 7.


### **Table 7.** Summary of works on blockchain for EMS applications.

### **4. Blockchain for Cybersecurity in SG**

The immediate need to incorporate renewable energy sources has necessitated considering a more diversified and distributed structure for the SG. This objective was achieved through distributed generation system and DER [123]. However, this has increased the complexity of the SG. Further, the SG's complex infrastructure comprises several devices such as the PMUs, smart meters, home automation sensors, remote terminal unit, spanning generation, transmission, distribution, customer, operation, marketing, and utility domains, etc. [124]. Situational awareness is vital to ensure the resiliency of such a marvelous SG infrastructure. The communication infrastructure and the communication protocols needed to support these applications vary. The core of the communication network is a wide area network (WAN). In addition to this, there exist other types of communication networks such as local area networks (LAN), home area networks (HAN), wireless sensor networks (WSN), neighborhood area networks (NAN), etc. These communication networks mostly use TCP/IP protocol suite for data communication. TCP/IP is not a secure protocol. Hence, the communication network of the SG applications can be easily attacked by exploiting its vulnerability. Despite the basic security measures such as firewall, intrusion detection, encryption, authentication, etc., which are already implemented in the SG, though it is still vulnerable to several cyber-attacks. An excellent survey on various detection algorithms was provided on false data injection in ref. [125].

The SG is a typical cyber-physical system [126]. As a cyber-physical system, cybersecurity is a vital parameter with three features: availability, confidentiality, and integrity. Availability is characterized as the property in which all data are available promptly. The cyberattack can compromise availability by blocking, delaying, and corrupting the data or even losing the data. The impact of cyber-attack on the availability of SG applications is huge. Confidentiality is characterized as the property of the system to protect the privacy and proprietary information from unauthorized access. The cyberattack on confidentiality can compromise the privacy and proprietary information of the SG application. Such incidents can grant illegal access to the application by stealing password-related information, causing enormous loss to the operation of the application. Integrity is characterized as the application's property to protect the system from unauthorized access to avoid any modification, alternation, and destruction of the data. The cyberattacks on integrity can modify the data to configure the application, resulting in an enormous loss. For example, the modification data can lead to misconfiguration of the sensors leading to failure of the SG application.

Blockchain is a distributed ledger that is immutable and does not depend upon any third party for its execution. This makes blockchain a secure method for data transactions and thus plays a vital role in SG applications. The blockchain can explicitly be used to mitigate the cyberattacks to strengthen the SG application's security. Among the different blockchains, the public blockchain is highly secure compared with the consortium and private blockchain due to the nature of the members and the consensus mechanism. The members of the public blockchain can be anonymous, whereas only the trusted nodes can be members of the consortium and private blockchains. The consensus mechanism followed in the public blockchain is proof-of-work, whereas multi-party voting in the consortium blockchain and strictly pre-approved nodes in the private blockchain are followed as a consensus mechanism. However, computational complexity is very high in the public blockchain. Thus, when security threats are fewer, and computation complexity is low, consortium and private blockchains are preferable to the public blockchain. The architecture of the blockchain for cybersecurity in SG applications is shown in Figure 20.

**Figure 20.** The architecture of SG cybersecurity using blockchain. **Figure 20.** The architecture of SG cybersecurity using blockchain.

The smart devices generate data communicated to the blockchain network server using the TCP/IP protocol. If the devices are computationally powerful, the hashing of the data and its encryption can be performed at the device itself, thereby creating the block, which is then placed into the blockchain network. This is the most secure architecture, as any data tampering after it leaves the device results in a change in the has function, leading to the invalidation of the block. However, this puts much computational pressure on the end devices, which are already over-burdened by other tasks. The other alternative is to send the data to the servers/nodes in the blockchain using TCP/IP and generate the blocks in the blockchain. This is less secure, but it is not a computationally powerful smart devices. In the latter case, using private and public keys for extra authentication can be beneficial. This architecture envisions maximizing the security since all participants in the consortium blockchain are trusted, and the consensus mechanism is based on multi-party voting with no scope for anonymity. The administrative and management authorities se-The smart devices generate data communicated to the blockchain network server using the TCP/IP protocol. If the devices are computationally powerful, the hashing of the data and its encryption can be performed at the device itself, thereby creating the block, which is then placed into the blockchain network. This is the most secure architecture, as any data tampering after it leaves the device results in a change in the has function, leading to the invalidation of the block. However, this puts much computational pressure on the end devices, which are already over-burdened by other tasks. The other alternative is to send the data to the servers/nodes in the blockchain using TCP/IP and generate the blocks in the blockchain. This is less secure, but it is not a computationally powerful smart devices. In the latter case, using private and public keys for extra authentication can be beneficial. This architecture envisions maximizing the security since all participants in the consortium blockchain are trusted, and the consensus mechanism is based on multi-party voting with no scope for anonymity. The administrative and management authorities select the member nodes acting as miners for the consortium and private blockchain. Next, the works related to blockchain for SG cybersecurity are comprehensively summarized in Table 8.

lect the member nodes acting as miners for the consortium and private blockchain. Next, **Table 8.** Summary of works on blockchain for SG cybersecurity.


[129] EMS and MG A novel blockchain hyperledger is proposed for

[130] MGs A master-slave mechanism is proposed to protect the

data against malicious attacks.

[81] EMS and MG

cyber-security of SG applications.

secure transactions on energy distribution.

A blockchain framework for P2P energy transactions

Blockchain technology for SG applications is still in the research phase and is gradually finding practical utility. Secure mechanisms are needed that can be implemented at the device level before the data leave the device. These mechanisms should be light and can be implemented in real-time.

### **5. Conclusions**

SG is evolving with the developments in storage and computational technologies. One such technology that can potentially transform the transactions amongst the various entities of the SG is the blockchain. The blockchain offers a decentralized and secure means of authorizing transactions, removing the need for a centralized authority. Despite its tremendous application in other domains, it has been underutilized for SG applications. This paper reviewed blockchain technology from a utility perspective for SG applications. General architectures were proposed for the important SG applications and identified challenges. The review is expected to enhance the research on developing novel technologies to meet the requirements of practical SG applications.

The blockchain-based applications are still in the nascent stage from various perspectives, which are seen as future research problems. Many SG applications operate in real-time, and the blockchain should not overburden the applications. The resource requirements for computation are a major challenge in blockchain-based systems. Blockchain must be developed to work on a lighter framework while retaining its security features. Additionally, regulatory bodies have to develop standardization procedures to make this technology interoperable and popular. Some of these research problems can be solved in the future, thoroughly revolutionizing blockchain-based applications.

**Author Contributions:** Conceptualization, B.A.; methodology, B.A.; software, S.K.M. (Sunil Kumar Mishra), S.K.M. (Santosh Kumar Mishra) and A.V.J.; validation, N.T., P.T. and N.B.; investigation, N.T., P.T., N.B. and B.A.; resources, S.K.M. (Sunil Kumar Mishra), S.K.M. (Santosh Kumar Mishra), A.V.J., I.S.S., F.M.E. and F.G.B.; data curation, F.M.E., N.B. and B.A.; writing—original draft preparation, S.K.M. (Sunil Kumar Mishra), S.K.M. (Santosh Kumar Mishra), A.V.J., I.S.S., F.M.E., B.A. and F.G.B.; supervision, N.B. and B.A.; project administration, N.B. and B.A.; formal analysis: N.T. and P.T.; funding acquisition: N.T. and P.T.; visualization: N.T., P.T., N.B. and B.A.; writing—review and editing: N.T., P.T., N.B. and B.A.; figures and tables: I.S.S. and A.V.J. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by the Framework Agreement between University of Pitesti (Romania) and King Mongkut's University of Technology North Bangkok (Thailand), in part by an International Research Partnership "Electrical Engineering—Thai French Research Center (EE-TFRC)" under the project framework of the Lorraine Université d'Excellence (LUE) in cooperation between Université de Lorraine and King Mongkut's University of Technology North Bangkok and in part by the National Research Council of Thailand (NRCT) under Senior Research Scholar Program under Grant No. N42A640328, and in part by National Science, Research and Innovation Fund (NSRF) under King Mongkut's University of Technology North Bangkok under Grant no. KMUTNB-FF-65-20.

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

**Informed Consent Statement:** Not applicable.

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

### **References**


## *Article* **Developing a Generalized Multi-Level Inverter with Reduced Number of Power Electronics Components**

**Hossein Shayeghi 1,2,\* , Ali Seifi 2,3, Majid Hosseinpour 1,2 and Nicu Bizon 4,5,6,\***


**Abstract:** Reducing the number of components of power electronic converters has been an important research topic over the past few decades. This paper introduces a new structure for a multi-level inverter based on reduced switch basic modules. The proposed basic module requires fewer switches and auxiliary devices. In addition, a lesser number of on-state switches for the synthesis of each voltage level results in less conduction losses, which enhances the converter efficiency. The proposed structure is capable of being implemented in both symmetrical and asymmetrical topologies. This is a merit feature for the proposed topology, which produces high voltage levels with a limited number of elements. The proposed structure is controlled using the fundamental frequency control scheme. The proposed basic module consists of six unidirectional switches and five DC voltage sources, generating five positive voltage levels. The performance of the recommended topology is analyzed from the various circuitry parameters, and a comprehensive comparison carried out with similar recent structures. The presented comparison reveals the advantage of the recommended inverter from different aspects of the circuitry parameters. The suggested structure is simulated using Matlab/Simulink software, and its performance is validated using a laboratory prototype. The results are reported for various steady-state and dynamic conditions.

**Keywords:** multi-level inverter; reduced switch basic modules; efficiency

### **1. Introduction**

Multi-level inverters are widely used in various applications due to their various features, such as low dv/dt stress, modularity, and high-power quality. In high voltage applications, equipment with low and medium voltage levels is often used [1]. Multilevel inverters are usually utilized in high power quality [2], FACTS devices [3], electric vehicles [4], variable speed drives [5], smart grids [6], high voltage applications [7], etc. The traditional multi-level inverters are segmented into three primary categories, which include Diode Clamped Multi-Level Inverter (DC-MLI) or Neutral Point Clamped Multi-Level Inverters (NPC-MLI), Flying Capacitor Multi-Level Inverter (FC-MLI), and Cascade H-Bridge Multi-Level Inverters (CHB-MLI) [8,9]. From the number of circuitry components view, the DC-MLI structure requires multiple diodes if the number of levels increases, making circuit control complex and tedious [1]. In the FC-MLI structure, the voltage balance problem of the capacitors can be solved by using additional switching modes. However, the number of passive components will increase which will be a threat to the

**Citation:** Shayeghi, H.; Seifi, A.; Hosseinpour, M.; Bizon, N. Developing a Generalized Multi-Level Inverter with Reduced Number of Power Electronics Components. *Sustainability* **2022**, *14*, 5545. https://doi.org/10.3390/ su14095545

Academic Editor: Pablo García Triviño

Received: 16 March 2022 Accepted: 27 April 2022 Published: 5 May 2022

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

**Copyright:** © 2022 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/).

reliability of the converter [10]. The CHB-MLI structure is modular and relatively simple compared to the other two topologies. Besides, it does not need additional circuits to match the voltage, but instead uses several isolated DC sources [11]. Despite the simplicity and modularity of CHB-MLIs, adding one module increases four switches and related devices. The need for more circuit devices in the classic topologies of multi-level inverters is a research motivation for new topologies by reducing the number of devices [12].

An advanced configuration was represented in [13] for a symmetrical voltage source multi-level inverter that the high voltage levels are synthesized by extending basic cells. This structure has fewer switches compared to the cascaded H-bridge inverter. However, the number of switches in this topology is still high. In [14], a sub-multi-level structure is introduced, and its cascaded structure was investigated, which requires many switches and drivers. In [14,15], a six-switch H-bridge configuration was utilized for an incremental combination of DC voltage sources. To generalize the voltage levels of this structure, bidirectional switches are considered on both sides of the basic structure. These structures also require a large number of switches. In [16], an extended cascaded structure is introduced, which was based on a basic module and level booster circuit. The number of power electronic devices is high in this structure. An improved symmetric generalized topology was introduced in [17]. This topology consists of two parts, the Level Generation Unit (LGU) and the Polarity Generation Unit (PGU), which uses bidirectional switches. This structure requires a high number of switches, and also the total blocking voltage is significant. In [18], a multi-level cascaded inverter structure was introduced with a new design in symmetrical and asymmetrical topologies, where the number of switches is relatively high. In [19], a bidirectional cascaded multi-level inverter topology was reported, where can operate in symmetrical and asymmetrical modes. This structure was designed by improving the other two structures, which have fewer switches. However, in this inverter, the number of switches and the blocking voltage are also high. In [20], a multi-level inverter for dynamic loads was suggested, which provides two separate structures for symmetrical and asymmetrical topologies. Its symmetrical structure consists of an advanced cascaded H-bridge with many semiconductors. The reference [21] introduces a symmetrical modular multilevel topology with a relatively low number of switches. However, the voltage sources number in this structure is a high value. The reference [22] introduced a switch-ladder structure for a new multi-level converter. This structure was implemented symmetrically and asymmetrically using unidirectional and bidirectional switches. A diode-containing bidirectional structure for a multi-level inverter was reported in [23], which solves the voltage spike problem in such diode-based structures. However, this structure uses a current sensor to realize the current polarity, complicating the control scheme. Also, a generalized diode-containing bidirectional structure for multi-level inverters was represented in [24]. This structure solved the current sensor problem in [23] yet has many switches and considerable blocking voltage. In [25], a modified K-type multi-level inverter structure was reported, utilizing the cascading method to generalize to higher voltage levels. This structure requires eight unidirectional switches to produce seven levels, while the number of switches is not small. A Cross-Switched T-Type (CT-Type) was presented in [26], which used a T-type inverter embedded on either side of the structure to yield more voltage levels. This inverter can operate symmetrically or asymmetrically, while the number of switches was not reduced.

This paper develops a novel configuration for multi-level inverters with symmetrical and asymmetrical topologies. Fewer power electronic devices and fewer switches in the current path of different voltage levels are the main design goals of the recommended topology. These goals make the suggested inverter useful for different applications. The main features of the proposed structure are as follows.

1. The proposed basic module has a lesser number of switches, which by generalizing the basic module, the proposed extended structure is realized.


The rest of this paper is organized as follows. Section 2 presents the configuration of the suggested MLI converter and its functionality. Section 3 compares the recommended structure with other topologies in terms of circuitry parameters, including switches and gate driver's number, and total standing voltage (TSV) for different voltage levels. Section 4 presents the mathematical equations and simulations related to losses and efficiency. The analysis of the simulation and laboratory results is reported in Section 5, and finally, the conclusions are presented in Section 6.

### **2. Suggested Structure**

### *2.1. The Reduced Switch Basic Module*

The proposed reduced switch basic module (RSBM) is depicted in Figure 1. It consists of five DC sources and six semiconductor switches along with their anti-parallel diodes. The proposed reduced switch basic module is responsible for producing positive levels. The switches (S1, S1), (S2, S2), (S3, S3) of the proposed RSBM operate complementarily. In other words, the conducting switches in the current path of any voltage level are always half of the total switches of the suggested RSBM. The recommended RSBM can produce five positive voltage levels.

**Figure 1.** (**a**) The suggested RSBM, (**b**) Vout = 0, (**c**) Vout = Vdc, (**d**) Vout = 2Vdc, (**e**) Vout = 3Vdc, (**f**) Vout = 4Vdc, (**g**) Vout = 5Vdc.

The reduced switch basic module consists of three switches S1, S2, S3, and their complementary pairs. According to Figure 1, if the switches S1, S<sup>3</sup> are turned on, the output voltage 0 is generated, and if the switch S<sup>3</sup> is turned on, the first voltage level is produced. To generate the second voltage level, the switches S1, S2, S<sup>3</sup> are turned on, and switches S2, S<sup>3</sup> are turned on to generate the third voltage level. The RSBM also generates the fourth voltage level when the switches S1, S<sup>2</sup> are turned on, and the fifth voltage level is generated when the switch S<sup>2</sup> is turned on. Table 1 demonstrates the switching pattern of the suggested basic module to produce voltage levels. In this table, 1 means on-state, and 0 means off-state for switches. Naturally, complementary switches behave inversely with the main switches.


**Table 1.** The switching logic of the RSBM.

### *2.2. Blocking Voltage of the Proposed Reduced Switch Basic Module*

Maximum blocking voltage (MBV) indicates the peak voltage across the off-state switch. Considering the MBV of all switches of the converter, the total blocking voltage (TBV) for the converter is obtained, which can be expressed as follows:

$$\text{TBV} = \sum\_{\text{Swiches}} \text{MBV} \tag{1}$$

The MBV and TBV parameters are an important challenge for multi-level inverters. The parameter MBV is the most important in selecting the voltage rating of switches. The price of the switches is proportional to their allowable voltage rating. Therefore, the lower the MBV value of the converter switches, the lower the total cost. The MBV value for each switch of the proposed basic module is given in the following relations.

$$\mathbf{S}\_{\mathbf{l}} = \overline{\mathbf{S}\_{\mathbf{l}}} = \mathbf{V}\_{\mathbf{dc}} \tag{2}$$

$$\mathbf{S\_2 = S\_2 = S\_3 = S\_3 = 2}\,\mathrm{V\_{dc}}\tag{3}$$

The MBV relations of the proposed RSBM reveal that this basic unit can generate five voltage levels using low-voltage rating switches.

### *2.3. The Proposed Generalized Inverter Structure*

The proposed reduced switch basic module can generate only positive voltage levels. So, a structure must generate positive and negative voltage polarity. The H-bridge module can be used for this purpose. Figure 2 displays the proposed multi-level inverter structure. According to this figure, the proposed structure consists of two parts. The first part is related to the level generator, and the second part is related to the polarity generator. In the proposed multi-level inverter, the reduced switch basic modules are connected in series and form the level generator. All switches of the proposed structure operate complementarily. This means that only half of the switches are in on-state at any output voltage level. In other words, the low number of on-state switches reduces the conduction losses of the switches. The size of voltage sources in the RSBMs can be designed and adjusted both symmetrically and asymmetrically.

In a symmetric topology, the size of voltage sources of the RSBMs is equal. If the voltage source size of the RSBMs is assumed to be Vdc, the following relations provide the various parameters of the proposed multi-level inverter in the symmetric topology.

$$\mathbf{V\_0 = V\_1 = V\_2 = \dots = V\_j = V\_{\mathbf{dc}}} \tag{4}$$

$$\mathbf{N\_{L}} = \mathbf{11j} + \mathbf{1} \tag{5}$$

$$\mathbf{N\_{IGBT}} = \mathbf{N\_S} = 6\mathbf{j} + 4 \tag{6}$$

$$\mathbf{N\_{GD}} = 6\mathbf{j} + 4 \tag{7}$$

$$\text{TBV} = \text{\textdegree 30\textdegree\textdegree V}\_{\text{dc}} \tag{8}$$

In the above relations, j shows number of RSBMs, V<sup>j</sup> shows the size of the voltage sources of the jth basic module, N<sup>L</sup> shows the number of output voltage levels that can be synthesized by the topology, NIGBT, and N<sup>S</sup> show the number of IGBT and switches, respectively. Since the bidirectional switch is not utilized in the structure, the number of switches and IGBTs are equal. NGD shows the number of gate drivers, and the TBV shows the total blocking voltage by the switches of the structure.

In an asymmetric topology, voltage source sizes can have different values using various algorithms. Table 2 presents the allowable algorithms for the size of voltage sources in asymmetric topology, where j represents the jth RSBM.

**Table 2.** Proposed algorithms for asymmetric topology voltage source size.


### *2.4. The Proposed Multi-Level Inverter Structure*

The structure shown in Figure 2 with a reduced switch basic module is assumed to be the basic cell. By connecting these cells in series, the cascaded structure of the proposed topology is realized. Figure 3 shows the proposed cascaded topology with n basic cells. The cascading method is also a method for generalizing the proposed structure to achieve high voltage levels by using a small number of sources. In addition, in the cascaded topology, the voltage range of H-bridge switches is limited, and it is possible to achieve medium and even high voltage and power levels using limited Maximum Voltage Blocking (MBV) switches.

**Figure 3.** The proposed cascaded topology.

In the cascaded topology, the output voltage (VL) is obtained from the output voltage of the multiple basic cells:

$$\mathbf{V\_L = V\_{\rm out,1} + V\_{\rm out,2} + \dots + V\_{\rm out,n}} \tag{9}$$

In the cascaded topology, different algorithms can be used to generate voltage levels. The cascaded topology can be implemented with the following algorithm.

In this algorithm, the voltage value of independent voltage sources is determined as follows:

$$\mathbf{V\_1 = V\_{dc} \cdot V\_2 = 11V\_{dc} \cdot \ \ \ \mathbf{V\_n = 5V\_{n-1} + 5V\_{n-2} + \dots + 5V\_1 + V\_{dc}} \tag{10}$$

In the main structure shown in Figure 2, the voltage sources simply have an incremental combination. However, the voltage sources in the cascaded topology, in addition to incremental combination, can also have a decreasing combination, which leads to a significant increase in the number of output voltage levels. The number of voltage levels, the number of IGBTs, and the number of drivers for the cascaded topology are presented in the following equations.

$$\mathbf{N\_L} = 2(\mathbf{5V\_1} + \mathbf{5V\_2} + \dots + \mathbf{5V\_n}) + 1\tag{11}$$

$$\mathbf{N\_{ICBT}} = \mathbf{N\_{S}} = 10\mathbf{n} \tag{12}$$

$$\mathbf{N\_D = 10n} \tag{13}$$

where n represents the number of basic cells.

One of the advantages of cascaded topology is the replacing capability for the basic cells. If one of the basic cells is damaged, it can be taken out of the circuit, and the structure will continue to work with fewer voltage levels.

### **3. Efficiency Calculation**

In this section, the calculation and simulation of the conduction losses and switching losses are presented to estimate the efficiency of an 11-level basic cell. The conduction losses are divided into two parts: switch conductance losses and anti-parallel diode conduction losses. A detailed analysis of the power losses of a power electronic converter is investigated in the following.

### *3.1. Conduction Losses*

Power electronic switches have conduction losses when conducting the current in ON-state. The conduction losses of the switch and its anti-parallel diode is calculated by the following equations:

$$\mathbf{P\_{c,S}(t)} = [\mathbf{V\_{S,ON}} + \mathbf{R\_S}i^{\alpha}(t)]\mathbf{i(t)}\tag{14}$$

$$\mathbf{P\_{c,D}(t) = [V\_{D,ON} + \mathcal{R}\_D \mathbf{i}(t)]i(t)}\tag{15}$$

where S indicates the switch and D indicates the diode. The voltages vs. and V<sup>D</sup> are the voltage drop across the switch and the anti-parallel diode in their conduction interval. The resistors R<sup>S</sup> and R<sup>D</sup> represent the equivalent resistance of the switch and its antiparallel diode, i(t) is the current flowing through the switch and the anti-parallel diode at the conduction moments. The parameter α is a switch constant that depends on the switch specifications, which is introduced by the manufacturer in the switch datasheet. The conduction losses are calculated from the sum of the conduction losses presented in Equations (14) and (15). The amount of conduction losses of a multi-level inverter depends on the number of switches conducted at different voltage levels. Considering N<sup>S</sup> as the conducting number of switches and N<sup>D</sup> as the conducting anti-parallel diodes in a time

interval, the average conduction losses of the converter in an output voltage period can be represented by (16):

$$\mathbf{P\_{c}} = \frac{1}{2\pi} \int\_{0}^{2\pi} \left[ \mathbf{N\_{S}(t)P\_{c,S}(t) + N\_{D}(t)P\_{c,D}(t)} \right] \mathbf{d(t)} \tag{16}$$

### *3.2. Switching Losses*

The switching losses are based on the energy losses due to the non-ideal switch performance. The energy losses include switch ON and OFF losses calculated by Equations (17) and (18):

$$\begin{split} \mathbf{E\_{ON,j}} &= \int\_{0}^{\mathbf{t\_{ON}}} \left[ \mathbf{v}(\mathbf{t}) \, \dot{\mathbf{t}}(\mathbf{t}) \right] \mathbf{d}(\mathbf{t}) \\ &= \int\_{0}^{\mathbf{t\_{ON}}} \left[ \left( \frac{\mathbf{V\_{S,j}}}{\mathbf{t\_{ON}}} \right) \left( -\frac{\mathbf{r}'}{\mathbf{t\_{ON}}} (\mathbf{t} - \mathbf{t\_{ON}}) \right) \right] \mathbf{d}(\mathbf{t}) = \frac{1}{6} \mathbf{V\_{S,j}} \mathbf{I}' \mathbf{t\_{ON}} \end{split} \tag{17}$$

EOFF,j = tOFF R 0 [v(t) i(t)] d(t) = tOFF R 0 h VS,j <sup>t</sup>OFF <sup>−</sup> <sup>I</sup> tOFF (t − t OFF ) id(t) <sup>=</sup> <sup>1</sup> <sup>6</sup>VS,jItOFF (18)

which EON,j and EOFF,j are the energy dissipation of the switch j at the moments of turning on and off, tON and tOFF are the time intervals required to turn a switch on and off, respectively. The parameters I and I 0 are the current that passes through the switch before turning it off, and after turning it on. VS,j is the reverse voltage across the switch after it is turned off. The switching power losses of switches in an output voltage period can be written as follows:

$$\mathbf{P\_{S}} = \sum\_{\mathbf{j=1}}^{\text{N}\_{\text{S}}} \left[ \sum\_{\mathbf{i=0}}^{\text{N}\_{\text{ON},\text{j}}} \mathbf{E\_{\text{ON},\text{j}}} + \sum\_{\mathbf{i=0}}^{\text{N}\_{\text{OFF},\text{j}}} \mathbf{E\_{\text{OFF},\text{ji}}} \right] \mathbf{f} \tag{19}$$

which NON,j and NOEE,j are the number of times that switch turns on and off in a cycle, and f is the output voltage frequency. Finally, the total losses are calculated by Equation (20):

$$\mathbf{P\_{Total}} = \mathbf{P\_c} + \mathbf{P\_s} \tag{20}$$

To investigate the efficiency of the proposed 11-level basic cell, power losses on proposed MLI at the output loads Z<sup>1</sup> = 50 Ω, Z<sup>2</sup> = 50 + j12.56 Ω, and Z<sup>3</sup> = 50 + j25.12 Ω are simulated with output voltage steps of 50 V. Figure 4a, b show the conduction and switching power losses for all three types of the loads, respectively. Besides, the efficiency and total losses are also displayed in Figure 4c. To compare the efficiency of the proposed 11-level structure, the efficiency curve for different output power is shown in Figure 5.

**Figure 4.** (**a**) Power losses and efficiency curves for the proposed 11-level basic cell topology at three types of loads, (**a**) Conduction losses, (**b**) Switching losses, (**c**) Temperature of switches, (**d**) Efficiency and total losses.

**Figure 5.** Efficiency comparison of the proposed 11-level basic cell with other topologies. [A] Alishah et al., 2021, [B] Jayabalan et al., 2017, [C] Ponraj et al., 2021.

### **4. Comparative Study**

In this section, the proposed topology is compared with other recent topologies presented in [13–31] to evaluate the validity and capability of the proposed MLI. For a fair comparison of the structures, a graph of the number of switches to different voltage levels is presented, and in addition, the NIGBT/N<sup>L</sup> ratio is calculated and presented in Table 3 for the proposed basic module and the comparative structures. Moreover, Table 3 lists other comparative parameters, including NIGBT (number of IGBTs), NGD (number of drivers), N<sup>L</sup> (number of voltage levels synthesized by structure), NDC (number of DC voltage source), TBV (total blocking voltage), N<sup>D</sup> (number of Diodes) and NIGBT, ON (number of conducting IGBTs in each voltage level) for the proposed topology and other symmetric topologies. According to this Table, the presented basic module utilizes fewer switches for various voltage levels.

To fairly compare the number of IGBTs in the proposed structure with other structures, a graph of the number of IGBTs relative to the number of voltage levels (NIGBT/NL) is evaluated. This ratio also provides cost-effectiveness of structures. The larger this ratio, the steeper the slope of the comparison curve, and the more IGBTs are used to achieve higher output voltage levels. Furthermore, the smaller this ratio, the lower the slope of the comparison curve, and the fewer IGBTs are required.

As shown in Figure 6a, using the proposed structure, a large number of voltage levels can be achieved with a smaller number of IGBTs, which produces high-quality output voltage with a smaller number of switches. Some structures use bidirectional commonemitter switches, which makes the number of gate drivers different from the number of IGBTs. For a fair comparison of the number of gate drivers of the proposed structure with other structures, the ratio of the number of gate drivers to the number of voltage levels (NGD/NL) is presented. Figure 6 demonstrates the (NIGBT/NL) and (NGD/NL) diagrams for the proposed and other structures. As shown in Figure 6a, the proposed topology has the lowest slope for (NIGBT/NL) diagram, which means the proposed structure utilizes the lowest number of switches to generate different voltage levels. Figure 6b also shows a comparison of the number of gate drivers, in which the proposed structure does not have the lowest curve slope regarding the number of gate drivers since it has not utilized a bidirectional switch. Nevertheless, the proposed structure still has a relatively good condition regarding the number of gate drivers compared to most comparative structures.


**Table 3.** Comparing the parameters of the proposed topology with other basic topologies.

**\*** Indicates product symbol.

**Figure 6.** *Cont*.

**Figure 6.** Comparative diagrams including: (**a**) NIGBT/NL, (**b**) NGD/NL. [A] Oskuee et al., 2015, [B] Alishah et al., 2021, [C] Jayabalan et al., 2017, [D] Ponraj et al., 2021, [E] Peddapati 2020, [F] Siddique et al., 2019, [G] Samsami et al., 2017, [H] Dhanamjayulu et al., 2017, [I] Gohari et al., 2019, [J] Alishah et al., 2016, [K] Hosseinpour et al., 2020, [L] Hosseini Montazer et al., 2021, [M] Selvaraj et al., 2020, [N] Meraj et al., 2019, [O] Ponraj et al., 2021, [P] Lee et al., 2017, [Q] Siddique et al., 2019, [R] Ali 2018, [S] Seifi et al 2020.

The total blocking voltage (TBV), the sum of the maximum blocking voltage (MBV) of the converter switches, is an essential parameter in comparing and evaluating structures because the voltage rating of required IGBTs for the structure is determined based on the MBV parameter, and the TBV parameter is directly related to the cost of the structure switches. Here, for a fair comparison of the TBV value, a graph of the ratio of this parameter to the number of output voltage levels (TBV/NL) is used, and this graph is plotted for different topologies, as shown in Figure 7. As Figure 7 displays, the proposed structure provides a relatively good TBV compared to other structures.

**Figure 7.** TBV/N<sup>L</sup> diagram for the proposed structure and other structures.\* Indicates product symbol. [A] Oskuee et al., 2015, [B] Alishah et al., 2021, [C] Jayabalan et al., 2017, [D] Ponraj et al., 2021, [E] Peddapati 2020, [F] Siddique et al., 2019, [G] Samsami et al., 2017, [H] Dhanamjayulu et al., 2017, [I] Gohari et al., 2019, [J] Alishah et al., 2016, [K] Hosseinpour et al., 2020, [L] Hosseini Montazer et al., 2021, [M] Selvaraj et al., 2020, [N] Meraj et al., 2019, [O] Ponraj et al., 2021, [P] Lee et al., 2017, [Q] Siddique et al., 2019, [R] Ali 2018, [S] Seifi et al., 2020.

### **5. Simulation and Laboratory Results of the Proposed Structure**

To investigate the performance of the proposed structure, a prototype containing two RSBM is simulated, and a laboratory prototype is implemented. The proposed structure containing two RSBM is investigated for the symmetric and asymmetric topologies. The laboratory sample hardware includes MOSFET IRFP460 power switches. The ARDUINO MEGA2560 microcontroller is used to generate gate pulses, which are isolated and amplified to drive the switches using the TLP250 optocoupler. The laboratory prototype and related instruments are shown in Figure 8.

**Figure 8.** (**a**) A laboratory prototype of the proposed structure, (**b**) The power circuit of proposed structure.

As it is known, multi-level inverter switching methods are classified into two categories according to the switching frequency, which can be low frequency or high frequency, respectively. From the first class it can be mentioned the staircase switching, fundamental frequency switching, active harmonic elimination, nearest level modulation (NLM) and selective harmonic elimination (SHE) technique. The second class includes sinusoidal PWM and space vector modulation (SVM) techniques. The modulation methods in both classes can be easily adapted and implemented to the topology proposed in this paper. The NLM method has been used to generate the switching pulses (see Figure 9a), using an integer that is close to the nearest voltage level as the reference signal. For example, if the voltage is in the range of 1.5 to 2.5, then the reference of 2Vdc will be generated. Figure 9b presents the NLM operation. The switching frequency is not well-known in NLM method. But it is low and is relatively near to output voltage frequency. The output voltage frequency is considered 50 Hz.

**Figure 9.** (**a**) NLM method, (**b**) The NLM operation.

In symmetric topology, the value of DC voltage sources is V<sup>0</sup> = V<sup>1</sup> = V<sup>2</sup> = 6 V. First, the inverter results for a purely resistive load Zload1 = 6.6 Ω are presented in Figure 10. The peak output voltage of the inverter with 11 levels of 6 V steps results in 66 volts. The peak load current, in this case, is 10A. The harmonic spectrum of the proposed topology is presented in Figure 10c, highlighting that the total harmonic distortion (THD) is 3.25%, with advantage in size and cost of the output filter.

**Figure 10.** The waveforms for resistive output load in symmetric topology, (**a**) laboratory sample voltage and current, (**b**) simulation voltage and current, (**c**) output voltage THD. **\*** Indicates product symbol.

Additionally, Figure 11 displays the results of the proposed structure in the case of symmetric sources for a resistive-inductive load Zload2 = 6.6 + j4.71 Ω. The peak load current in the case of R-L output load is 8.1 A, in which the inductance of the load filters, the current, and the load current is obtained similar to a sine wave.

**Figure 11.** Resistive-inductive load waveforms in symmetric topology, (**a**) laboratory sample voltage and current, (**b**) simulation voltage and current. **\*** Indicates product symbol.

To analyze and introduce the response of the proposed MLI in a dynamic situation, the modulation index changes abruptly, and the system response is evaluated. It should be noted that, this test is only open loop dynamic responses. As shown in Figure 12, for this purpose, the modulation index is once changed from 1 to 0.65 according to Figure 12a, and again according to Figure 12b, the modulation index is changed from 0.65 to 0.3. The proper performance of the proposed multi-level inverter in dynamic conditions is visible for changing the modulation index in Figure 12.

**Figure 12.** Voltage and current waveforms for changes of the modulation index: (**a**) from 1 to 0.65 and (**b**) from 0.65 to 0.3. **\*** Indicates product symbol.

The behavior of the proposed structure for a sudden change in load is evaluated as well. For this purpose, the load changes abruptly from resistive-inductive to pure resistive. Figure 13 illustrates inverter output voltage and current for a sudden change of the resistiveinductive load Zload2 = 6.6 + j4.71 Ω to a purely resistive load Zload3 = 4 Ω. The proposed structure performs well under dynamic load change conditions as well.

**Figure 13.** Voltage and current waveforms related to load's dynamic change, (**a**) laboratory result, (**b**) simulation result. **\*** Indicates product symbol.

Figure 14 displays the blocking voltage for switches of the proposed structure in the symmetric topology. Based on these curves, the maximum voltage across each switch is determined. These diagrams confirm the correct operation of the proposed structure.

To evaluate the performance of multi-level inverter in asymmetric topology, the value of DC voltage sources is selected based on the proposed fourth algorithm according to Table 2 and assumed as *V*<sup>0</sup> = *V*<sup>1</sup> = 2 V, *V*<sup>2</sup> = 12 V. The performance results of the proposed multi-level inverter with asymmetric topology for a resistive-inductive load are presented in Figure 15. The number of voltage steps increases from 11 levels in the symmetric topology to 36 levels of 2 V steps in the asymmetric topology. In this case, the peak output voltage results in 72 V. The peak load current equals 11.4 A in this condition. Figure 15b displays the zoomed staircase waveform of the output voltage. The proper performance of the proposed structure is clearly shown in this figure in the production of voltage steps. In addition, the THD value is decreased from 3.26% in the symmetric topology to 1.01% in the asymmetric topology. This THD value demonstrates that the asymmetric topology of the suggested MLI can operate properly without any output filter, which results in more reduction in overall size and cost.

Dynamic load testing is necessary to evaluate the capability of the proposed structure in the application of electric motor drives. Based on the simulation results and laboratory results of the prototype, it can be concluded that the performance of the proposed structure is satisfactory, and the proposed topology provides a desirable performance.

**Figure 14.** Blocking voltage of the proposed structure switches for symmetric topology.

**Figure 15.** The waveforms of Proposed MLI in asymmetric topology (**a**) voltage and current, (**b**) zoomed voltage, (**c**) output voltage THD. **\*** Indicates product symbol.

### **6. Conclusions**

In this paper, a novel configuration is proposed for multi-level voltage source inverters to reduce the number of switches. The proposed structure uses a reduced switch basic module to generate more output voltage levels. This structure is investigated for symmetric and asymmetric topologies. The efficiency of the structure is presented for different loads and compared with several other structures. The result is shown that the suggested inverter has higher efficiency. Besides, the comparison in terms of circuit devices reveals that the number of switches of the proposed structure is less than other reported structures. This difference in the number of switches is clearly evident at higher levels, and the graph slope of the number of switches to the number of levels is 0.6, which is less than other structures. Also, the number of gate driver, as well as the TBV parameter are in good situation and superior in comparison with most of similar structures. These results make that the proposed structure has a low cost and size. The performance of the proposed structure is confirmed by simulation and laboratory results. In the laboratory prototype, different loading conditions, including resistive load, resistive-inductive load as well as dynamic load, have been used to validate the proposed structure. Modulation index change is also performed for the proposed structure. Results evaluation of simulations

and laboratory samples, as well as the results of comparisons, indicate the appropriate performance of the proposed structure.

**Author Contributions:** Conceptualization, A.S., M.H. and H.S.; methodology, A.S., M.H. and H.S.; software, A.S. and M.H.; validation, M.H. and H.S.; investigation, M.H. and H.S.; resources, N.B.; data curation, N.B.; writing—original draft preparation, A.S., M.H. and H.S.; supervision, H.S. and N.B.; Funding acquisition: N.B.; Visualization: H.S. and N.B.; writing—review and editing: M.H., H.S. and N.B.; project administration, N.B.; Formal analysis: M.H., N.B. and H.S. 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.

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

### **References**


## *Article* **Design, Modeling, and Model-Free Control of Permanent Magnet-Assisted Synchronous Reluctance Motor for e-Vehicle Applications**

**Songklod Sriprang 1,2 , Nitchamon Poonnoy 2,\*, Babak Nahid-Mobarakeh <sup>3</sup> , Noureddine Takorabet <sup>1</sup> , Nicu Bizon <sup>4</sup> , Pongsiri Mungporn <sup>5</sup> and Phatiphat Thounthong 2,\***


**Abstract:** This paper describes the model-free control approaches for permanent magnet-assisted (PMa) synchronous reluctance motors (SynRMs) drive. The important improvement of the proposed control technique is the ability to determine the behavior of the state-variable system during both fixed-point and transient operations. The mathematical models of PMa-SynRM were firstly written in a straightforward linear model form to show the known and unknown parts. Before, the proposed controller, named here the intelligent proportional-integral (*i*PI), was applied as a control law to fix some unavoidable modeling errors and uncertainties of the motor. Lastly, a dSPACE control platform was used to realize the proposed control algorithm. A prototype 1-kW test bench based on a PMa-SynRM machine was designed and realized in the laboratory to test the studied control approach. The simulation using MATLAB/Simulink and experimental results revealed that the proposed control achieved excellent results under transient operating conditions for the motor drive's cascaded control compared to traditional PI and model-based controls.

**Keywords:** electric vehicle; inverter; permanent magnet-assisted synchronous reluctance motor; PMa-SynRM; model-free control; traction drive

## **1. Introduction**

By the end of 2021, the demand for electrical traction machines, including battery electric vehicles and hybrid electric vehicles (HEVs), surpassed two million units [1–4]. Electrical traction machines are also required to further develop more electric aircrafts (MEAs) [5–7]. For these reasons, several state-of-the-art machines have been developed in the last few years, such as synchronous reluctance motors (SynRMs) and especially permanent magnet-assisted synchronous reluctance motors (PMa-SynRMs). PMa-SynRMs can produce 75% of the torque of an interior permanent magnet synchronous motor (IPMSM) for the same size and liquid cooling technology [8,9]. In addition, state-of-the-art modern motors provide more desired characteristics for electric vehicle (EV) applications, in particular, high efficiency at low and high speeds. Therefore, PMa-SynRMs constitute a promising choice for these applications. However, PMa-SynRMs have a much more complicated structure, which affects the control system, and its model is strongly nonlinear. Therefore,

**Citation:** Sriprang, S.; Poonnoy, N.; Nahid-Mobarakeh, B.; Takorabet, N.; Bizon, N.; Mungporn, P.; Thounthong, P. Design, Modeling, and Model-Free Control of Permanent Magnet-Assisted Synchronous Reluctance Motor for e-Vehicle Applications. *Sustainability* **2022**, *14*, 5423. https://doi.org/ 10.3390/su14095423

Academic Editor: Mouloud Denai

Received: 17 February 2022 Accepted: 27 April 2022 Published: 30 April 2022

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

**Copyright:** © 2022 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/).

traditional control, such as field-oriented control (FOC) based on a proportional-integral (PI) controller, cannot accomplish high performance for all operating conditions of these modern machines.

Furthermore, in EV applications, safety, energy saving, and soft driving are mandatory and require improvement of the control performance of the motor drive system. Many studies have been conducted in the last few years regarding SynRMs and PMa-SynRMs, with special attention to machine design and optimization aspects. Multiple-flux barrier rotors and transversely laminated rotors were reported. Rotor laminations are made by traditional punching or wire cutting, resulting in easy and cheap construction [10,11]. Control characteristics have also been investigated [9,12,13]. In this regard, in the current control of PMa-SynRM drive systems, the essential objective is to ensure that the stator currents track the reference values with minimum errors in both transient and steady-state conditions. To design a robust controller with acceptable tracking performance, all the model-based control (MBC) approaches mentioned in the literature applied to PMa-SynRM require extensive knowledge about the dynamics and the model of PMa-SynRM systems. In addition, the MBC performance can be affected by unexpected dynamic variations of the system and parametric uncertainty, which are very common phenomena in industrial applications. To overcome the limitations of MBC approaches, some studies proposed model predictive control (MPC) as an appropriate current control scheme for electric motors, which ensures a fast dynamic and a remarkable safety factor [14–16]. This method's concept is based on predicting controlled variables in the next calculation step using the measured variables and a mathematical model of the controlled system. Then, the predicted results are analyzed using a cost function in terms of the difference between the desired trajectories and real outputs of the system. Compared to the previously mentioned control techniques, safety and fast dynamics are two remarkable features of the MPC method. Despite these advantages, the performance of MPC highly depends on the correctness of the model, given that a mathematical model is used in the prediction section [17]. When using the prediction at each sampling time of the MPC algorithm, some additional mathematical calculations are imported into the control algorithm.

Therefore, a control principle called model-free control (MFC) has been proposed to address the limitations of the abovementioned MPC and MBC techniques. MFC, also referred to as model-free tuning in the literature, uses a local linear approximation of the process model, which is valid for a small time window, and a fast estimator, which is employed to update the approximation [18,19]. The main advantage of MFC is that it does not require the process model in the controller tuning. Few experiments have been conducted on real-world control system structures concerning the tuning process. This paper introduces MFC development to control both torque and speed control of PMa-SynRMs. To verify the advantages of MFC, both simulations and experiments were carried out under several conditions.

This paper is organized as follows. A model-free control and control law are briefly introduced in Section 2. The main issues regarding the control of PMa-SynRMs, related state-of-the-art studies, as well as mathematical models are reviewed in Section 3, with a focus on MFC applied to PMa-SynRM drive systems. In Section 4, simulation and experimental results are provided to demonstrate the advantages of the proposed MFC. Sections 5 and 6 summarize and conclude the paper. A small-scale 1-kW test bench based on a PMa-SynRM with ferrite magnets was implemented to confirm the high performance of the designed control scheme in the laboratory [13].

### **2. Model-Free Control and Control Law (Brief Introduction)**

### *2.1. Model-Free Control*

The idea of model-free control accomplished for control system applications was originally proposed by Fliess et al. [20,21]. Many industrial applications have significantly changed with technology development and have become more complex. Accordingly, modeling the dynamic and process of these applications using mathematical models

becomes very difficult or at least time-consuming. In this case, using the MBC methods for these kinds of applications will be impossible. Conversely, almost all industrial applications generate and save a large number of process data that contain all the necessary information related to the system's operation. In this case, it is important to use these generated data, obtained online/offline, directly for designing the controller or other purposes. In this way, the model-free control (MFC) foundation is essential in controlling industrial applications. So far, the types of the modern control system can be roughly categorized by MBC and MFC, as in Figure 1. comes very difficult or at least time-consuming. In this case, using the MBC methods for these kinds of applications will be impossible. Conversely, almost all industrial applications generate and save a large number of process data that contain all the necessary information related to the system's operation. In this case, it is important to use these generated data, obtained online/offline, directly for designing the controller or other purposes. In this way, the model-free control (MFC) foundation is essential in controlling industrial applications. So far, the types of the modern control system can be roughly categorized by MBC and MFC, as in Figure 1.

The idea of model-free control accomplished for control system applications was originally proposed by Fliess et al. [20,21]. Many industrial applications have significantly changed with technology development and have become more complex. Accordingly, modeling the dynamic and process of these applications using mathematical models be-

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 3 of 22

**2. Model-Free Control and Control Law (Brief Introduction)** 

**Figure 1.** Control law's block diagram. **Figure 1.** Control law's block diagram.

*2.1. Model-Free Control* 

MFC is a control method that uses only the online data obtained from the controlled system to design the controller, without the additional need for information about the mathematical model or parameters of the studied system. Therefore, the MFC can be applicable for all nonlinear systems with complex or unknown structures. MFC is a control method that uses only the online data obtained from the controlled system to design the controller, without the additional need for information about the mathematical model or parameters of the studied system. Therefore, the MFC can be applicable for all nonlinear systems with complex or unknown structures.

The principle of model-free control is briefly introduced next. A nonlinear system can be described by a state-variable written as follows: The principle of model-free control is briefly introduced next. A nonlinear system can be described by a state-variable written as follows:

$$\begin{array}{l}\dot{\mathbf{x}} = f(\mathbf{x}, \boldsymbol{\mu})\\\mathbf{y} = h(\mathbf{x}, \boldsymbol{\mu})\end{array} \tag{1}$$

where where

$$\begin{array}{l} \mathbf{x} = [\mathbf{x}\_1, \mathbf{x}\_2, \dots, \mathbf{x}\_n]^T; \mathbf{x} \in \mathbb{R}^n\\ \mathbf{u} = [\boldsymbol{u}\_1, \boldsymbol{u}\_2, \dots, \boldsymbol{u}\_m]^T; \boldsymbol{u} \in \mathbb{R}^m\\ \mathbf{y} = [\boldsymbol{y}\_1, \boldsymbol{y}\_2, \dots, \boldsymbol{y}\_m]^T; \boldsymbol{y} \in \mathbb{R}^m \end{array} \tag{2}$$

, ,..., ; *T m u uu u u* = ∈ (2) where *x* is the state variable, *u* is the control variable, *y* is the output variable, and *n*, *m* N.

12 m 12 m , ,..., ; *T m y yy y y* = ∈ According to Equation (2), the system described by Equation (1) is flat. A control law of variable *u* can be expressed as follows [13]:

$$
u = 
u\_{\rm ref} + 
u\_{\rm feedback}(\varepsilon) \tag{3}$$

N. with *ε* = *y*ref − *y*.

ε

= − ref *y y*.

with

According to Equation (2), the system described by Equation (1) is flat. A control law of variable *u* can be expressed as follows [13]: ref feedback *uu u* = + ( ) ε (3) This control law is suitable for all systems with known parameters. However, if only some system parameters can be identified or the system described by Equation (1) cannot be identified, the controller needs to be modified as a partially-known model, replaced by a model-free control as follows:

$$
\mu = \frac{\pounds(y\_\prime \dot{y}\_\prime \ddot{y}\_\prime \dots \prime y^{(n)})}{b} + \frac{F}{b} \tag{4}
$$

where *α*ˆ( . *y*)/*b* is a known system, and *F* denotes an unknown part of the system. . .. . . ..

The difference between *α*ˆ(*y*, *y*, *y*, . . . , *y* (*n*) ), *α*ˆ( *<sup>y</sup>*), and . *y* is that the *α*ˆ(*y*, *y*, *y*, . . . , *y* (*n*) ) is the known part of the *α*(*y*, . *y*, .. *y*, . . . , *y* (*k*) ), the *α*ˆ( . *y*) is the only known part of the studied system, and the . *y* is the differential of the known part, respectively.

Alternatively, it can be rewritten and rearranged as a straightforward linear model as follows:

$$
\dot{y} = -F + b \cdot u \tag{5}
$$

### *2.2. Control Law*

Figure 2 represents the control law block diagram for the model-free control technique. The control law is defined as follows:

$$
\mu = \mu\_{\text{ref}} + \mu\_{\text{feedback}}(\varepsilon) + \frac{\widehat{F}}{b} \tag{6}
$$

where

$$\mu\_{\text{ref}} = \frac{\hbar \left( y\_{\text{ref}} \dot{y}\_{\text{ref}} \ddot{y}\_{\text{ref}} \dotsm y\_{\text{ref}}^{(\beta+1)} \right)}{b} \tag{7}$$

and \_ *F* is the estimated value of *F*, which is expressed as follows:

$$
\widehat{F} = b \cdot u - \dot{y} \tag{8}
$$

$$\text{The function } \boldsymbol{\mathfrak{a}} \text{ is a regular function [22,23].}$$

The feedback term *u*feedback can be described by applying the PI controller as follows:

$$
\mu\_{\text{feedback}} = \mathbf{K\_p} \cdot \varepsilon + \mathbf{K\_i} \cdot \int \varepsilon dt \tag{9}
$$

Substituting Equation (6) into Equation (5), and rearranging the expressions, Equation (5) can be expressed as follows: *Sustainability* **2022**, *14*, x FOR PEER REVIEW 5 of 22

$$\dot{y} = -F + b \cdot u\_{\text{ref}} + b \cdot u\_{\text{feedback}}(\varepsilon) + \widehat{F} \tag{10}$$

**Figure 2.** Control law's block diagram. **Figure 2.** Control law's block diagram.

### *2.3. Controller Design 2.3. Controller Design*

pressed as follows:

tion (12):

and

tively.

The observation term purposes to afford an estimated signal *F* so that *F F* → *t* → ∞ (under global convergence assumption for the estimation). Consequently, Equation The observation term purposes to afford an estimated signal \_ *F* so that \_ *F* → *F* as *t* → ∞ (under global convergence assumption for the estimation). Consequently, Equation (10) can be rewritten as follows:

*dt*

ε

$$
\dot{y} = b \cdot \mu\_{\text{ref}} + b \cdot \mu\_{\text{feedback}}(\varepsilon) \tag{11}
$$

ε

 ε

<sup>−</sup> +⋅ ⋅+⋅ = (12)

+⋅ ⋅+⋅ ⋅ = *bK bK* 0 (13)

(14)

= (15)

= (16)

ref feedback *y bu bu* =⋅ +⋅ ( )

Consequently, Equation (11) describes the dynamic of the closed-loop control system.

( ) <sup>0</sup> *ref dy y b K b K dt*

ε

Referring to the control law displayed in Figure 2, the controller coefficients can be

p i

<sup>2</sup> 2 0 *n n q qq* +⋅ ⋅ ⋅+ ⋅=

εε

 ω

*b*

2

ω

<sup>n</sup> *<sup>K</sup>*<sup>i</sup> *<sup>b</sup>*

The gain *b*∈ is a non-physical constant parameter. Instead of *α*, the *b* is present

where *ζ* and *ω*n are the tuning dominant damping ratio and natural frequency, respec-

in this paper, as shown in (4). It was chosen by the practitioner or obtained by trials and errors. *F*, which is continuously updated, subsumes the poor parts of the plant and the

various possible disturbances without distinguishing between them [24,25]

⋅ ⋅ ζ

n

ω

determined using the following expression obtained by taking time derivation in Equa-

Comparing Equation (13) to the 2nd order standard equation stated as follows:

ζω

p <sup>2</sup> *<sup>K</sup>*

p i

By substituting Equation (9) into Equation (11) and rearranging, Equation (11) can be ex-

(11)

as

the controller coefficients become:

Consequently, Equation (11) describes the dynamic of the closed-loop control system. By substituting Equation (9) into Equation (11) and rearranging, Equation (11) can be expressed as follows:

$$\frac{d(y\_{ref} - y)}{dt} + b \cdot \mathbf{K\_{P}} \cdot \varepsilon + b \cdot \mathbf{K\_{i}} \int \varepsilon dt = 0\tag{12}$$

Referring to the control law displayed in Figure 2, the controller coefficients can be determined using the following expression obtained by taking time derivation in Equation (12):

$$
\ddot{\varepsilon} + b \cdot \mathbf{K\_p} \cdot \dot{\varepsilon} + b \cdot \mathbf{K\_i} \cdot \varepsilon = 0 \tag{13}
$$

Comparing Equation (13) to the 2nd order standard equation stated as follows:

$$
\ddot{q} + \mathbf{2} \cdot \zeta \cdot \omega\_n \cdot \dot{q} + \omega\_n^2 \cdot q = 0 \tag{14}
$$

the controller coefficients become:

$$K\_{\rm p} = \frac{2 \cdot \mathcal{J} \cdot \omega\_{\rm n}}{b} \tag{15}$$

and

$$K\_{\rm i} = \frac{\omega\_n^2}{b} \tag{16}$$

where *ζ* and *ω<sup>n</sup>* are the tuning dominant damping ratio and natural frequency, respectively.

The gain *b* ∈ R is a non-physical constant parameter. Instead of *α*, the *b* is present in this paper, as shown in (4). It was chosen by the practitioner or obtained by trials and errors. *F*, which is continuously updated, subsumes the poor parts of the plant and the various possible disturbances without distinguishing between them [24,25]

### **3. Applying Model-Free Control to PMa-SynRM Drive**

### *3.1. Mathematic Model of PMa-SynRM/Inverter*

A variable speed drive (VSD), which powers the PMa-SynRM under study, is shown in Figure 3. Owing to the rotor geometries of the PMa-SynRM discussed in [26], the current control strategies in the literature differ from those applied to PMSM. The rotor geometries of PMa-SynRMs are given by the salient-pole, in which *L*<sup>d</sup> > *L*q. Its torque expression was given by Equation (17). In this case, the *i*<sup>d</sup> component should not be equal to zero to take advantage of the reluctance torque produced by the high saliency ratio. Therefore, the maximum torque per ampere (MTPA) control strategy was recommended for PMa-SynRMs. The main idea of this control was to develop the requested torque using the minimum value of the stator current magnitude:

$$T\_{\mathbf{e}} = n\_{\mathbf{p}} \left\{ \mathbf{Y}\_{\mathbf{m}} - (L\_{\mathbf{d}} - L\_{\mathbf{q}}) i\_{\mathbf{q}} \right\} \cdot i\_{\mathbf{d}} \tag{17}$$

The equations of a PMa-SynRM in the rotating *d*<sup>q</sup> reference frame and a mechanical equation are expressed by a state-space representation as follows:

$$\underbrace{\begin{bmatrix} \frac{d\boldsymbol{i}\_{\rm d}}{dt} \\ \frac{d\boldsymbol{i}\_{\rm d}}{dt} \\ \frac{d\boldsymbol{i}\_{\rm d}}{dt} \end{bmatrix}}\_{\begin{bmatrix} \frac{d\boldsymbol{i}\_{\rm d}}{dt} \\ \frac{d\boldsymbol{i}\_{\rm d}}{dt} \end{bmatrix}} = \underbrace{\begin{bmatrix} \left\{-R\_{\rm s}\dot{\rm{d}}\_{\rm d} + \omega\_{\rm e} \left(\mathcal{L}\_{\rm d}\dot{\rm{q}}\_{\rm d} - \mathbf{\mathcal{V}}\_{\rm m}\right) \right\}/\mathcal{L}\_{\rm q} \\\ \left[\begin{smallmatrix} -R\_{\rm s}\dot{\rm{q}}\_{\rm d} - \omega\_{\rm e}\mathcal{L}\_{\rm d}\dot{\rm{q}}\_{\rm d} \end{smallmatrix} / \mathcal{L}\_{\rm q} \end{bmatrix} / \mathcal{L}\_{\rm q}}\_{f(\boldsymbol{x})} + \underbrace{\begin{bmatrix} \frac{1}{\mathcal{L}\_{\rm d}} & 0 & 0 \\ 0 & \frac{1}{\mathcal{L}\_{\rm q}} & 0 \\ 0 & 0 & -\frac{1}{\mathcal{I}} \end{bmatrix}}\_{\mathbf{B}} \underbrace{\begin{bmatrix} \boldsymbol{v}\_{\rm d} \\ \boldsymbol{v}\_{\rm q} \\ \boldsymbol{v}\_{\rm d} \end{bmatrix}}\_{\mathbf{u}} \tag{18}$$

**Figure 3.** A three-phase inverter to control a PMa-SynRM prototype. **Figure 3.** A three-phase inverter to control a PMa-SynRM prototype.

**3. Applying Model-Free Control to PMa-SynRM Drive** 

A variable speed drive (VSD), which powers the PMa-SynRM under study, is shown in Figure 3. Owing to the rotor geometries of the PMa-SynRM discussed in [26], the current control strategies in the literature differ from those applied to PMSM. The rotor geometries of PMa-SynRMs are given by the salient-pole, in which *L*d > *L*q. Its torque expression was given by Equation (17). In this case, the *i*d component should not be equal to zero to take advantage of the reluctance torque produced by the high saliency ratio. Therefore, the maximum torque per ampere (MTPA) control strategy was recommended for PMa-SynRMs. The main idea of this control was to develop the requested torque using the

The equations of a PMa-SynRM in the rotating *d*q reference frame and a mechanical

<sup>L</sup> <sup>m</sup> p md d q qd f m

*<sup>v</sup> di*

− + −Ψ = −− + Ψ+ − − <sup>−</sup>

*T d n i L L ii B J*

*dt J*

*q*

{ }

ω

( ) <sup>1</sup> 0 0

sd e qq m <sup>d</sup>

sq e dd q

*Ri Li L v*

*d d*

*T n L Li i* e p m d qq d = Ψ− − ⋅ { ( ) } (17)

<sup>1</sup> 0 0

<sup>1</sup> 0 0

*q*

*u*

(18)

*3.1. Mathematic Model of PMa-SynRM/Inverter* 

minimum value of the stator current magnitude:

d

*di*

q

*x*

ω

*y*

=

1 0

010 001

**C**  d q m

*i i* ω

0

 

equation are expressed by a state-space representation as follows:

{ }

ω

*dt Ri Li L L*

( )

( )

ω

( )

**<sup>B</sup>**

*dt L*

*f x*

#### *3.2. Model-Free of Current and Speed Control Development 3.2. Model-Free of Current and Speed Control Development*

The control system of PMa-SynRMs proposed in this paper (Figure 4) had a case cascade construction consisting of two loops (i.e., inner current control loop and outer speed control loop). The inner current loop was much faster than the outer speed control loop, such that the model-free control for the current control was developed first. By defining *u* The control system of PMa-SynRMs proposed in this paper (Figure 4) had a case cascade construction consisting of two loops (i.e., inner current control loop and outer speed control loop). The inner current loop was much faster than the outer speed control loop, such that the model-free control for the current control was developed first. By defining *u* = [*u*<sup>1</sup> *u*2] <sup>T</sup> = [*v*<sup>d</sup> *v*q] T , y = [*y*<sup>1</sup> *y*2] <sup>T</sup> = [*i*<sup>d</sup> *i*q] T , and rearranging the first and second rows in Equation (18) in the form of Equation (5), the PMa-SynRM model is expressed as follows:

$$\begin{aligned} \frac{d\dot{i}\_{\rm d}}{dt} &= -\frac{R\_{\rm s}\dot{i}\_{\rm d}}{L\_{\rm d}} + \frac{\omega\_{\rm e}(L\_{\rm q}\dot{i}\_{\rm q} - \Psi\_{\rm m})}{L\_{\rm d}} + v\_{\rm d} \cdot \frac{1}{L\_{\rm d}}\\ \frac{d\dot{i}\_{\rm q}}{dt} &= -\frac{R\_{\rm s}\dot{i}\_{\rm q}}{L\_{\rm q}} - \frac{\omega\_{\rm e}L\_{\rm d}\dot{i}\_{\rm d}}{L\_{\rm q}} + v\_{\rm q} \cdot \frac{1}{L\_{\rm q}} \end{aligned} \tag{19}$$

According to the principle of the model-free as in [20,21], Equation (19) can be separated to identify the known and unknown terms as follows. The known terms are

 $\hat{\kappa}\_1 = \frac{\dot{y}\_1}{b\_1} = L\_\mathrm{d} \frac{di\_\mathrm{d}}{dt}$ 
$$\hat{\kappa}\_1 = \frac{\dot{y}\_2}{b\_2} = L\_\mathrm{q} \frac{di\_\mathrm{q}}{dt}$$

and the unknown terms are

$$\begin{aligned} F\_1 &= \left\{-R\_\text{s}i\_\text{d} + \omega\_\text{e}(L\_\text{q}i\_\text{q} - \Psi\_\text{m})\right\} \cdot \frac{1}{L\_\text{d}}\\ F\_2 &= \left(-R\_\text{s}i\_\text{q} - \omega\_\text{e}L\_\text{d}i\_\text{d}\right) \cdot \frac{1}{L\_\text{q}} \end{aligned} \tag{21}$$

According to the control law (Figure 2), the first term of the model-free control for inner current loop control is determined as follows:

 $\mu\_{1\text{ref}} = \frac{\dot{y}\_{1\text{ref}}}{\dot{b}\_1} = L\_\text{d} \frac{di\_\text{d}}{dt}$ 
$$\mu\_{2\text{ref}} = \frac{\dot{y}\_{2\text{ref}}}{\dot{b}\_2} = L\_\text{q} \frac{di\_\text{q}}{dt}$$

The estimation of unknown terms is expressed as follows:

$$\begin{aligned} \widehat{F}\_1 &= \frac{1}{L\_\mathrm{d}} u\_1 - \dot{y}\_1 = \frac{1}{L\_\mathrm{d}} v\_\mathrm{d} - \frac{d\dot{l}\_\mathrm{d}}{dt} \\\\ \widehat{F}\_2 &= \frac{1}{L\_\mathrm{q}} u\_1 - \dot{y}\_2 = \frac{1}{L\_\mathrm{q}} v\_\mathrm{q} - \frac{d\dot{l}\_\mathrm{q}}{dt} \end{aligned} \tag{23}$$

The feedback terms of *d*- and *q*-axis current control are obtained as follows:

$$\begin{aligned} b\_1 \cdot u\_{1 \text{feedback}} &= b\_1 \left( \mathbf{K\_{pd}} \cdot \varepsilon\_{\text{d}} + \mathbf{K\_{id}} \int \varepsilon\_{\text{d}} dt \right) \\ b\_2 \cdot u\_{2 \text{feedback}} &= b\_2 \left( \mathbf{K\_{pq}} \cdot \varepsilon\_{\text{q}} + \mathbf{K\_{iq}} \int \varepsilon\_{\text{q}} dt \right) \end{aligned} \tag{24}$$

Concerning the design procedure in the controller design, Equation (24) can be rewritten as follows: .. .

$$
\ddot{\varepsilon}\_{\rm d} + b\_1 \cdot K\_{\rm pd} \cdot \dot{\varepsilon}\_{\rm d} + b\_1 \cdot K\_{\rm id} \cdot \varepsilon\_{\rm d} = 0
$$

$$
\ddot{\varepsilon}\_{\mathbf{q}} + b\_2 \cdot \mathbf{K\_{pq}} \cdot \dot{\varepsilon}\_{\mathbf{q}} + b\_2 \cdot \mathbf{K\_{iq}} \cdot \varepsilon\_{\mathbf{d}} = 0
$$

The controller coefficients *K*pd, *K*id, *K*pq, and *K*iq are determined as follows:

$$\begin{aligned} K\_{\rm pd} &= \frac{2\tilde{\tau}\_{1}\omega\_{\rm n1}}{b\_{1}}, K\_{\rm id} = \frac{\omega\_{\rm n1}^{2}}{b\_{1}}\\ K\_{\rm pq} &= \frac{2\tilde{\tau}\_{1}\omega\_{\rm n1}}{b\_{2}}, K\_{\rm iq} = \frac{\omega\_{\rm n1}^{2}}{b\_{2}} \end{aligned} \tag{26}$$

The second model-free control for the outer speed control loop is developed here. The output of the speed control loop provides the torque reference of the MTPA algorithm, generating optimized *d*- and *q*-axis current references. Therefore, *T*<sup>e</sup> was chosen as a control variable of the outer speed control loop, such that *u*<sup>3</sup> = *T*eREF. Then, rewriting the mechanical equation of the PMa-SynRM represented by the third row in Equation (18) in the form of Equation (5) yields:

$$\frac{d\omega\_{\rm m}}{dt} = \left(-\mathcal{B}\_f \cdot \omega\_{\rm m} - T\_{\rm L}\right) \cdot \frac{1}{J} + T\_{\rm e} \cdot \frac{1}{J} \tag{27}$$

Separating this equation into the known and unknown terms, the known term is expressed as follows:

$$
\hat{\alpha}\_3 = \frac{\dot{y}\_3}{b\_3} = J \frac{d\omega\_\mathbf{m}}{dt} \tag{28}
$$

The unknown term is expressed as follows:

$$F\_3 = \left(-B\_f \omega\_m - T\_L\right) \cdot \frac{K\_l}{J} \tag{29}$$

Each part of the model-free control for the outer speed control loop is defined according to the following expression: .

$$
\mu\_{\text{3ref}} = \frac{\dot{y}\_{\text{3ref}}}{b\_{\text{3}}} = f \frac{d\omega\_{\text{m}}}{dt} \tag{30}
$$

The estimation of the unknown term is expressed as follows:

$$\widehat{F}\_3 = \frac{\mathbf{K}\_l}{J}\boldsymbol{\mu}\_3 - \dot{\boldsymbol{y}}\_3 = \frac{1}{J} \cdot T\_\mathbf{e} - \frac{d\boldsymbol{\omega}\_\mathbf{m}}{dt} \tag{31}$$

$$b\_3 \mu\_{\text{3feedback}} = b\_3 \left(\mathbf{K}\_{\text{p}\omega} \cdot \boldsymbol{\varepsilon}\_{\omega} + \mathbf{K}\_{\text{i}\omega} \int \boldsymbol{\varepsilon}\_{\omega} dt\right) \tag{32}$$

Regarding the controller design procedure, Equation (32) can be rewritten as follows:

$$
\ddot{\varepsilon}\_{\omega} + b\_1 \cdot \mathbf{K}\_{\text{p}\omega} \cdot \dot{\varepsilon}\_{\omega} + b\_1 \cdot \mathbf{K}\_{\text{i}\omega} \cdot \varepsilon\_{\omega} = 0 \tag{33}
$$

The controller coefficients *K*p<sup>ω</sup> and *K*i<sup>ω</sup> are determined as follows:

$$K\_{\rm pw} = \frac{2\zeta\_2 \omega\_{\rm n2}}{b\_3}, K\_{\rm iw} = \frac{\omega\_{\rm n2}^2}{b\_3} \tag{34}$$

where *ζ*<sup>2</sup> and *ω*n2 are the desired dominant damping ratio and natural frequency of the outer speed control loop, respectively. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 9 of 22

**Figure 4.** Control system of PMa-SynRM based on a model-free control diagram. **Figure 4.** Control system of PMa-SynRM based on a model-free control diagram.

### *3.3. Trajectory Planning*

expressed as follows:

*3.3. Trajectory Planning*  Finally, as presented in Figure 2, desired trajectory planning must be implemented to generate the input set-point *y*REF. A second order filter is often implemented to plan the desired trajectory for the controlled output. It permits limiting the derivative terms in the Finally, as presented in Figure 2, desired trajectory planning must be implemented to generate the input set-point *y*REF. A second order filter is often implemented to plan the desired trajectory for the controlled output. It permits limiting the derivative terms in the control law. The proposed trajectory planning for the two inner current control loops is expressed as follows:

control law. The proposed trajectory planning for the two inner current control loops is 2 *y*1REF *y*1COM = 1 ( *s <sup>ω</sup>*n3 <sup>2</sup> + 2*ζ*<sup>3</sup> *ω*n3 *s* + 1 ) (35)

$$\frac{\mathcal{Y}\_{\text{2REF}}}{\mathcal{Y}\_{\text{2COM}}} = 1 \Bigg/ \left\{ \left( \frac{s}{\omega\_{\text{n3}}} \right)^{2} + \frac{2\zeta\_{3}}{\omega\_{\text{n3}}}s + 1 \right\} \tag{36}$$

(35)

(36)

(37)

2 ζ where ζ<sup>3</sup> and ωn3 are the tuning dominant damping ratio and natural frequency, respectively. The trajectory planning of the outer speed loop is expressed as follows:

2REF 3

2COM n3 n3 <sup>2</sup> 1 1 *<sup>y</sup> <sup>s</sup> s y* ω ω = ++ *y*3REF *y*3COM = 1 ( *s <sup>ω</sup>*n4 <sup>2</sup> + 2*ζ*<sup>4</sup> *ω*n4 *s* + 1 ) (37)

where ζ3 and ωn3 are the tuning dominant damping ratio and natural frequency, respectively. where ζ<sup>4</sup> and ωn4 are the desired dominant damping ratio and natural frequency of the speed loop trajectory planning, respectively.

3REF 4

### The trajectory planning of the outer speed loop is expressed as follows: 2 ζ **4. Simulation and Experimental Validation of the Model-Free Control Applied to PMa-SynRM**

### *4.1. Experimental Setup*

3COM n4 n4 <sup>2</sup> 1 1 *<sup>y</sup> <sup>s</sup> s y* ω ω = ++ where ζ4 and ωn4 are the desired dominant damping ratio and natural frequency of the speed loop trajectory planning, respectively. **4. Simulation and Experimental Validation of the Model-Free Control Applied to PMa-SynRM**  *4.1. Experimental Setup*  A small-scale test bench 1-KW relying on the prototype PMa-SynRM was conceived in the laboratory, as shown in Figure 5. The prototype PMa-SynRM was supplied by a 3-kW 3-phase inverter (DC/AC) operating at a switching frequency of 16 kHz. Besides, the input DC grid voltage of the inverter was fed by a three-phase variable power supply combined with a three-phase diode rectifier. The PMa-SynRM was mechanically coupled with an IPMSM (interior permanent magnet synchronous motor) feeding a resistive load (see Figure 3). The measurements for the speed and rotor angle were acquired by a resolver placed on the rotor shaft. The developed control scheme relying on the model-free control was modeled in the Matlab/Simulink software, and then it was incorporated in the dSPACE 1202 MicroLabBox real-time interface to generate the gate control signals applied to the VSI.

A small-scale test bench 1-KW relying on the prototype PMa-SynRM was conceived

input DC grid voltage of the inverter was fed by a three-phase variable power supply combined with a three-phase diode rectifier. The PMa-SynRM was mechanically coupled with an IPMSM (interior permanent magnet synchronous motor) feeding a resistive load (see Figure 3). The measurements for the speed and rotor angle were acquired by a resolver placed on the rotor shaft. The developed control scheme relying on the model-free control was modeled in the Matlab/Simulink software, and then it was incorporated in the dSPACE 1202 MicroLabBox real-time interface to generate the gate control signals applied to the VSI. The main PMa-SynRM parameters are listed in Table 1, whereas the model-

free controller parameters are listed in Table 2.

The main PMa-SynRM parameters are listed in Table 1, whereas the model-free controller parameters are listed in Table 2. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 10 of 22

**Figure 5.** Experimental setup. **Figure 5.** Experimental setup.

**Table 1.** Specifications and parameters of the motor/inverter. **Table 1.** Specifications and parameters of the motor/inverter.


*V*dc DC bus voltage 400 V **Table 2.** Current/torque and speed regulation parameters.


*ω*n1q Natural frequency 1 2000 Rad.s−<sup>1</sup> *ζ*2 Damping ratio 2 0.7 *ω*n2 Natural frequency 2 107.1419 Rad.s−<sup>1</sup>

*ζ*3d Damping ratio 3 1 *ω*n3d Natural frequency 3 300 Rad.s−<sup>1</sup>

*ω*n3q Natural frequency 3 200 Rad.s−<sup>1</sup> *ζ*<sup>4</sup> Damping ratio 4 1


**Table 2.** *Cont. Sustainability* **2022**, *14*, x FOR PEER REVIEW 11 of 22

### *4.2. Simulations* The developed MFC algorithm for the PMa-SynRM drive was simulated under different operation conditions before its implementation. Figure 6 shows the simulation re-

one.

The developed MFC algorithm for the PMa-SynRM drive was simulated under different operation conditions before its implementation. Figure 6 shows the simulation results of the set-point tracking *d*-axis inner loop current control response using the model-free control. Interestingly, note that, during the transient response, the *d*-axis current tracked the reference very well, and there was no steady-state error. The simulation conditions were set as follows: for *d*-axis testing, the *q*-axis current command *i*qCOM was set to zero. sults of the set-point tracking *d*-axis inner loop current control response using the modelfree control. Interestingly, note that, during the transient response, the *d*-axis current tracked the reference very well, and there was no steady-state error. The simulation conditions were set as follows: for *d*-axis testing, the *q*-axis current command *i*qCOM was set to zero.

**Figure 6.** Simulation results: Dynamic response of the set-point tracking *d*-axis current control with the MFC. **Figure 6.** Simulation results: Dynamic response of the set-point tracking *d*-axis current control with the MFC.

Figure 7 shows the set-point tracking *q*-axis current control simulation results using the model-free control. Note that the control performance was satisfactory, with good setpoint tracking and zero steady-state error. The simulation conditions were set as follows: Figure 7 shows the set-point tracking *q*-axis current control simulation results using the model-free control. Note that the control performance was satisfactory, with good set-point tracking and zero steady-state error. The simulation conditions were set as follows: during *q*-axis testing, the *d*-axis command *i*dCOM was set to zero, and the load was the rated one.

during *q*-axis testing, the *d*-axis command *i*dCOM was set to zero, and the load was the rated

**Figure 7.** Simulation results: Dynamic response of *q*-axis currents with the MFC applied to the PMa-SynRM drive. **Figure 7.** Simulation results: Dynamic response of *q*-axis currents with the MFC applied to the PMa-SynRM drive.

Another simulation result is depicted in Figure 8. It shows the drive response to a step change on the speed reference from 0 to 1000 rpm. In this figure, Chs 1, 2, and 4 represent the speed command *n*COM, speed reference *n*REF, and measured speed *n*, respectively. Chs 3, 5, and 6 represent the torque reference *T*eREF and the *d*- and *q*-axis currents *i*<sup>d</sup> and *i*q, and Chs 7 and 8 represent the *d*-axis voltage and *q*-axis voltage, respectively. The parameters of the simulated drive are those of the test bench that will be later used for experimental validation. They are reported in Section 4.1. The MFC was designed to keep the torque within the range ±6 Nm. Note that the speed response was satisfactory with Another simulation result is depicted in Figure 8. It shows the drive response to a step change on the speed reference from 0 to 1000 rpm. In this figure, Chs 1, 2, and 4 represent the speed command *n*COM, speed reference *n*REF, and measured speed *n*, respectively. Chs 3, 5, and 6 represent the torque reference *T*eREF and the *d*- and *q*-axis currents *i*<sup>d</sup> and *i*q, and Chs 7 and 8 represent the *d*-axis voltage and *q*-axis voltage, respectively. The parameters of the simulated drive are those of the test bench that will be later used for experimental validation. They are reported in Section 4.1. The MFC was designed to keep the torque within the range ±6 Nm. Note that the speed response was satisfactory with small overshoot and without steady-state error.

small overshoot and without steady-state error. Although no torque sensor was employed in the experimental setup, the torque seemed to be limited to the allowed range. Moreover, *i*<sup>q</sup> and *i*<sup>d</sup> were generated on the basis of the MTPA algorithm discussed in [22].

Figure 9 shows the simulation of the disturbance rejection ability of the MFC applied to the PMa-SynRM drive. In this figure, Ch 4 represents the measured speed *n*, Ch2 represents the *d*-axis current *i*d, Ch3 represents the *q*-axis current *i*q, and Ch1 represents the torque reference *T*eREF. The simulation conditions were as follows: *n* = 1000 rpm; sudden increase of 3.7 Nm on the load torque *T*<sup>L</sup> at 0.3 s; and subsequent clearance of the load torque at 0.7 s. Note that, under the action of the proposed model-free control, when the load changed suddenly, the motor speed deviated slightly from its set-point, but it recovered very quickly. Figure 9 also shows the disturbance rejection capability of the MFC. As a result, the speed control performance was significantly improved, confirming the feasibility of the proposed MFC for this application.

**Figure 8.** Simulation results: Simulated drive response to a 0–1000 rpm reference speed pulse. From top to bottom: speed reference and response, *d*- and *q*-currents, *d*- and *q*-voltages, and active region number of the MFC controller. **Figure 8.** Simulation results: Simulated drive response to a 0–1000 rpm reference speed pulse. From top to bottom: speed reference and response, *d*- and *q*-currents, *d*- and *q*-voltages, and active region number of the MFC controller.

bility of the proposed MFC for this application.

of the MTPA algorithm discussed in [22].

**Figure 9.** Simulation results: Disturbance rejection of MFC applied to the PMa-SynRM drive. **Figure 9.** Simulation results: Disturbance rejection of MFC applied to the PMa-SynRM drive.

Although no torque sensor was employed in the experimental setup, the torque

Figure 9 shows the simulation of the disturbance rejection ability of the MFC applied

seemed to be limited to the allowed range. Moreover, *i*q and *i*d were generated on the basis

to the PMa-SynRM drive. In this figure, Ch 4 represents the measured speed *n*, Ch2 represents the *d*-axis current *i*d, Ch3 represents the *q*-axis current *i*q, and Ch1 represents the torque reference *T*eREF. The simulation conditions were as follows: *n* = 1000 rpm; sudden increase of 3.7 Nm on the load torque *T*L at 0.3 s; and subsequent clearance of the load torque at 0.7 s. Note that, under the action of the proposed model-free control, when the load changed suddenly, the motor speed deviated slightly from its set-point, but it recovered very quickly. Figure 9 also shows the disturbance rejection capability of the MFC. As a result, the speed control performance was significantly improved, confirming the feasi-

### *4.3. Experimental Validation of PMa-SynRM Drive Based on Model-Free Control*

*4.3. Experimental Validation of PMa-SynRM Drive Based on Model-Free Control*  The designed MFC for the PMa-SynRM drive was experimentally validated on a laboratory test bench. The experimental setup is depicted in Figure 5. The entire controller parameters of the current/torque and speed are presented in Table 2. The model-free control stability and response were easy to set compared to the FOC with PI controller. Thus, by defining and selecting the governing damping and natural frequency as mentioned in the literature [19], the controller coefficients of the PI controller for both the current and speed loops control may be calculated by (26) and (34). The PI controller was provided to The designed MFC for the PMa-SynRM drive was experimentally validated on a laboratory test bench. The experimental setup is depicted in Figure 5. The entire controller parameters of the current/torque and speed are presented in Table 2. The model-free control stability and response were easy to set compared to the FOC with PI controller. Thus, by defining and selecting the governing damping and natural frequency as mentioned in the literature [19], the controller coefficients of the PI controller for both the current and speed loops control may be calculated by (26) and (34). The PI controller was provided to deal with inevitable modeling errors and uncertainties. Therefore, the PI controller guaranteed the stability of the model-free control the ensure that the current and speed control achieved the steady-state error.

deal with inevitable modeling errors and uncertainties. Therefore, the PI controller guaranteed the stability of the model-free control the ensure that the current and speed control achieved the steady-state error. Figure 10 shows the current control test of the set-point tracking *d*-axis inner loop. In Figure 10 shows the current control test of the set-point tracking *d*-axis inner loop. In this figure, *d*-axis command *i*dCOM, *d*-axis reference *i*dREF, which is provided by the *d*-axis trajectory planning, and the actual *d*-axis current are represented. Ch5, Ch6, and Ch7 represent the measured stator phase currents A, B, and C, respectively. These results are similar to those obtained by simulation and confirm that the current control performance was satisfactory.

this figure, *d*-axis command *i*dCOM, *d*-axis reference *i*dREF, which is provided by the *d*-axis trajectory planning, and the actual *d*-axis current are represented. Ch5, Ch6, and Ch7 represent the measured stator phase currents A, B, and C, respectively. These results are similar to those obtained by simulation and confirm that the current control performance was satisfactory. The same test was conducted with the *q*-axis current while the *d*-axis current was regulated to zero. In this case, the motor was at a stand-still. Figure 11 depicts the experimental data, where Ch1 represents the *q*-axis current command *i*qCOM, Ch2 represents the *q*-axis current reference *i*qREF, and Ch3 represents the *q*-axis current measurement *i*q. Ch5, Ch6, and Ch7 represent the measured stator phase currents A, B, and C, respectively. Overall, the current control performance seemed satisfactory.

**Figure 10.** Experimental result: Set-point tracking *d*-axis current control response curve based on MFC. **Figure 10.** Experimental result: Set-point tracking *d*-axis current control response curve based on MFC.

The same test was conducted with the *q*-axis current while the *d*-axis current was regulated to zero. In this case, the motor was at a stand-still. Figure 11 depicts the experimental data, where Ch1 represents the *q*-axis current command *i*qCOM, Ch2 represents the *q*-axis current reference *i*qREF, and Ch3 represents the *q*-axis current measurement *i*q. Ch5, Ch6, and Ch7 represent the measured stator phase currents A, B, and C, respectively. Overall, the current control performance seemed satisfactory. Figure 12 depicts the speed startup of the PMa-SynRM drive using the MFC. In this figure, Chs 1, 2, 3, and 4 represent the torque reference *T*eREF, *d*-axis current *i*d, *q*-axis current *i*q, and measured speed *n,* respectively. Chs 5 and 6 represent the output *v*<sup>q</sup> and *v*d, chosen as the output of the MFC. Moreover, Chs 7 and 8 represent the estimated unknown terms of the *d*- and *q*-axis models. As expected, the torque was limited, and the speed response showed neither overshoot nor steady-state error. It is worth recalling that the torque reference generated *i*<sup>q</sup> and *i*<sup>d</sup> command references according to the MTPA algorithm.

**Figure 11.** Experimental result: Set-point tracking *q*-axis current control response curve based on MFC. **Figure 11.** Experimental result: Set-point tracking *q*-axis current control response curve based on MFC.

Figure 12 depicts the speed startup of the PMa-SynRM drive using the MFC. In this figure, Chs 1, 2, 3, and 4 represent the torque reference *T*eREF, *d*-axis current *i*d, *q*-axis current *i*q, and measured speed *n,* respectively. Chs 5 and 6 represent the output *v*q and *v*d, chosen as the output of the MFC. Moreover, Chs 7 and 8 represent the estimated unknown terms of the *d*- and *q*-axis models. As expected, the torque was limited, and the speed response showed neither overshoot nor steady-state error. It is worth recalling that the torque reference generated *i*q and *i*d command references according to the MTPA algo-Figure 13 shows the experimental validation of the disturbance rejection ability of the proposed MFC applied to the PMa-SynRM drive. In this figure, Chs 1, 2, 3, and 4 represent the torque reference *T*eREF, *d*-axis current *i*d, *q*-axis current *i*q, and measured speed *n*, respectively. Chs 5 and 6 represent the output *v*<sup>q</sup> and *v*d, chosen as the output of the MFC. Moreover, Chs 7 and 8 represent the estimated unknown terms of the *d*- and *q*-axis current models. The experimental conditions were set as follows: *n*REF = 1000 rpm, and sudden increase of the load torque (*T*L) to 3.7 Nm at 0.2 s. Note that the proposed

rithm.

model-free control compensated for the load torque variation and rejected its effect on the motor speed in a short time. This figure shows the effectiveness of the MFC in rejecting load torque disturbance and maintaining zero steady-state speed error. As a result, the speed loop control performance was good. This result confirms the feasibility of the proposed MFC for speed control of PMa-SynRM. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 17 of 22

**Figure 12.** Experimental result: Speed acceleration curve based on MFC. **Figure 12.** Experimental result: Speed acceleration curve based on MFC.

Figure 13 shows the experimental validation of the disturbance rejection ability of the proposed MFC applied to the PMa-SynRM drive. In this figure, Chs 1, 2, 3, and 4 represent the torque reference *T*eREF, *d*-axis current *i*d, *q*-axis current *i*q, and measured speed *n*, re-

Moreover, Chs 7 and 8 represent the estimated unknown terms of the *d*- and *q*-axis current models. The experimental conditions were set as follows: *n*REF = 1000 rpm, and sudden increase of the load torque (*T*L) to 3.7 Nm at 0.2 s. Note that the proposed model-free control compensated for the load torque variation and rejected its effect on the motor speed in a short time. This figure shows the effectiveness of the MFC in rejecting load torque disturbance and maintaining zero steady-state speed error. As a result, the speed loop

control performance was good. This result confirms the feasibility of the proposed MFC

for speed control of PMa-SynRM.

**Figure 13.** Experimental result: Disturbance rejection ability based on MFC. **Figure 13.** Experimental result: Disturbance rejection ability based on MFC.

### **5. Comparison of Traditional FOC with PI Controller, MBC, and Model-Free Control 5. Comparison of Traditional FOC with PI Controller, MBC, and Model-Free Control**

Traditional FOC based on the PI controller applied to PMa-SynRM was introduced in a previous study [24]. In addition, the MBC based on differential flatness-based control applied to PMa-SynRM was proposed in [13]. Thus, the comparison of the experimental results between the FOC with the PI controller and the MBC (the differential flatness-Traditional FOC based on the PI controller applied to PMa-SynRM was introduced in a previous study [24]. In addition, the MBC based on differential flatness-based control applied to PMa-SynRM was proposed in [13]. Thus, the comparison of the experimental results between the FOC with the PI controller and the MBC (the differential flatness-based control) is expressed below.

Figure 14a shows the current control test of the set-point tracking *d*-axis inner loop of the FOC with the PI controller, and Figure 14b illustrates the current control test of the

based control) is expressed below.

Figure 14a shows the current control test of the set-point tracking *d*-axis inner loop of the FOC with the PI controller, and Figure 14b illustrates the current control test of the set-point tracking *d*-axis inner loop of the differential flatness-based control applied to the PMa-SynRM drive system. In Figure 14a,b Ch1 is the current *i*dCOM, Ch3 is the measured current *i*d, Ch4 is the measured current *i*q, and Ch5 is the measured speed *n*. As shown in Figure 14a,b, in a transitory operation, the *i*<sup>d</sup> of the FOC with the PI controller exhibits a small overshoot compared to the differential flatness-based controller, and the *i*<sup>q</sup> of the FOC with PI controller shows oscillations. set-point tracking *d*-axis inner loop of the differential flatness-based control applied to the PMa-SynRM drive system. In Figure 14 a,b Ch1 is the current *i*dCOM, Ch3 is the measured current *i*d, Ch4 is the measured current *i*q, and Ch5 is the measured speed *n*. As shown in Figure 14a,b, in a transitory operation, the *i*d of the FOC with the PI controller exhibits a small overshoot compared to the differential flatness-based controller, and the *i*q of the FOC with PI controller shows oscillations. set-point tracking *d*-axis inner loop of the differential flatness-based control applied to the PMa-SynRM drive system. In Figure 14 a,b Ch1 is the current *i*dCOM, Ch3 is the measured current *i*d, Ch4 is the measured current *i*q, and Ch5 is the measured speed *n*. As shown in Figure 14a,b, in a transitory operation, the *i*d of the FOC with the PI controller exhibits a small overshoot compared to the differential flatness-based controller, and the *i*q of the FOC with PI controller shows oscillations.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 19 of 22

**Figure 14.** Experimental result: Comparison of the set-point tracking between (**a**) the FOC with the PI controller and (**b**) the MBC. **Figure 14.** Experimental result: Comparison of the set-point tracking between (**a**) the FOC with the PI controller and (**b**) the MBC. However, the differential flatness-based control was the model-based control (MBC), as mentioned in the introduction. Its performance depends on the system model. More clearly, the control laws of the model-free control and the differential flatness-based con-

However, the differential flatness-based control was the model-based control (MBC), as mentioned in the introduction. Its performance depends on the system model. More clearly, the control laws of the model-free control and the differential flatness-based control are shown in Figure 15. However, the differential flatness-based control was the model-based control (MBC), as mentioned in the introduction. Its performance depends on the system model. More clearly, the control laws of the model-free control and the differential flatness-based control are shown in Figure 15. trol are shown in Figure 15. The control law of the differential flatness-based control (See Figure 15a) has the inverse dynamic equation, which contains the system models including *R*s, *L*d, *L*q, and Ψm. In contrast, the control law of the model-free control (See Figure 15b) estimated all the system parameters through the unknown term, *F*.

The control law of the differential flatness-based control (See Figure 15a) has the in-

**Figure 15.** The difference between (**a**) the differential flatness-based control law and (**b**) the model-free control law.

(**a**) (**b**)

The control law of the differential flatness-based control (See Figure 15a) has the inverse dynamic equation, which contains the system models including *R*s, *L*d, *L*q, and Ψm. In contrast, the control law of the model-free control (See Figure 15b) estimated all the system parameters through the unknown term, *F*.

As a more concise summary, Table 3 shows a comparison of the advantages of traditional FOC+PI, differential flatness-based control, and model-free control.


**Table 3.** Comparison of three different control techniques applied to the PMa-SynRM drive system.

### **6. Conclusions**

In this study, we analyzed the application of an MFC for the current and speed control of motor drives. This novel control approach was applied to PMa-SynRMs for the combined control of the outer speed control loop and inner current control loop. After a brief introduction of the MFC fundamentals, the design approach was comprehensively described, providing a step-by-step procedure. Suggestions for extending the design to different drive controllers were also provided. Simulations and numerous experimental results highlighted the promising features and characteristics of MFC applied to electrical motor drives. Finally, the potential of MFC pointed out in this study should stimulate further exploration and analysis of this type of controller to achieve the expertise required to transfer the results to practical applications.

Interestingly, the proposed MFC provided high performance for the PMa-SynRM drives compared to FOC with the traditional PI controller. Besides, it had a higher dynamic performance than the PMa-SynRM drive using the differential flatness-based control.

In this study, the simulation and the experimental validation were performed by a prototype PMa-SynRM at GREEN Lab, Université de Lorraine. This machine can operate in constant torque and constant power regions if a proper field weakening control is applied. In summary, by applying MFC, the performance of the PMa-SynRM was improved not only in terms of the inner current control loop but also the outer speed control loop. Moreover, the controller coefficients of the proposed MFC are not complicated to define, and a unique design approach can be applied for the PMa-SynRM drive.

**Author Contributions:** Conceptualization, B.N.-M. and N.T.; methodology, S.S., N.P. and P.T.; validation, S.S., N.P. and P.T.; formal analysis, N.B. and P.M.; writing—original draft preparation, P.T.; writing—review and editing, N.B.; visualization, N.B.; supervision, B.N.-M. and N.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by the Framework Agreement between the University of Pitesti (Romania) and King Mongkut's University of Technology North Bangkok (Thailand), in part by an International Research Partnership "Electrical Engineering–Thai French Research Center (EE-TFRC)" under the project framework Lorraine Université d'Excellence (LUE) in cooperation with Université de Lorraine and King Mongkut's University of Technology North Bangkok and in part by the National Research Council of Thailand (NRCT) under Senior Research Scholar Program, Grant No. N42A640328, and in part by King Mongkut's University of Technology North Bangkok under Grant no. KMUTNB-64-KNOW-20.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors would like to express their gratitude to the GREEN laboratory at the University of Lorraine and King Mongkut's University of Technology North Bangkok (KMUTNB) for their constant support in boosting collaborations between France and Thailand. Besides, the authors would like to express their gratitude to the Rajamangala University of Technology Rattanakosin Wang Klai Kangwon Campus as an original agency of the first author (A25/2021).

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

### **References**


## *Article* **Simplified Super Twisting Sliding Mode Approaches of the Double-Powered Induction Generator-Based Multi-Rotor Wind Turbine System**

**Habib Benbouhenni <sup>1</sup> , Nicu Bizon 2,3,4,\* , Ilhami Colak <sup>1</sup> , Phatiphat Thounthong 5,6 and Noureddine Takorabet <sup>6</sup>**


**Abstract:** This work proposes a new indirect filed-oriented control (IFOC) scheme for double-powered induction generators (DPIGs) in multi-rotor wind turbine systems (MRWTS). The IFOC strategy is characterized by its simplicity, ease of use, and fast dynamic speed. However, there are drawbacks to this method. Among its disadvantages is the presence of ripples in the level of torque, active power, and current. In addition, the total harmonic distortion (THD) value of the electric current is higher compared to the direct torque control method. In order to overcome these shortcomings and in terms of improving the effectiveness and performance of this method, a new algorithm is proposed for the super twisting algorithm (STA). In this work, a new STA method called simplified STA (SSTA) algorithm is proposed and applied to the traditional IFOC strategy in order to reduce the ripples of torque, current, and active power. On the other hand, the inverter of the DPIG is controlled by using a five-level fuzzy simplified space vector modulation (FSSVM) technique to obtain a signal at the inverter output of a fixed frequency. The results obtained from this proposed IFOC-SSTA method with FSSVM strategy are compared with the classical IFOC method which uses the classical controller based on a proportional-integral (PI) controller. The proposed method is achieved using the Matlab/Simulink software, where a generator with a large capacity of 1.5 megawatts is used. The generator is placed in a multi-rotor electric power generation system. On the other hand, the two methods are compared in terms of ripple ratio, dynamic response, durability, and total harmonic distortion (THD) value of the electric current. Through the results obtained from this work, the proposed method based on SSTA provided better results in terms of ripple ratio, response dynamic, and even THD value compared to the classical method, and this shows the robustness of the proposed method in improving the performance and efficiency of the generator in the multi-rotor wind system.

**Keywords:** double-powered induction generator; indirect filed-oriented control; five-level fuzzy simplified space vector modulation; super-twisting algorithm; simplified STA; multi-rotor wind turbine systems

### **1. Introduction**

Traditionally, field-oriented control (FOC) is among the most widely used control methods in the field of controlling electrical machines, whether they are motors or electric generators, due to its simplicity and ease of implementation. This method is based on the

**Citation:** Benbouhenni, H.; Bizon, N.; Colak, I.; Thounthong, P.; Takorabet, N. Simplified Super Twisting Sliding Mode Approaches of the Double-Powered Induction Generator-Based Multi-Rotor Wind Turbine System. *Sustainability* **2022**, *14*, 5014. https://doi.org/ 10.3390/su14095014

Academic Editor: Miguel Carrión

Received: 28 March 2022 Accepted: 18 April 2022 Published: 22 April 2022

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

**Copyright:** © 2022 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/).

use of a classic proportional-integral (PI) controller, which gives this method a fast dynamic response. On the other hand, this method is based on the use of the traditional pulse width modulation (PWM) to control the inverter of the machine. There are two types of this method, the direct FOC method and the indirect FOC method [1]. In the field of control, this method was used to control several electrical machines such as the asynchronous motor [2,3] and the synchronous motor [4]. However, this method has been used in the field of renewable energies, as according to the reference [5], this method is among the most widely used methods in the field of generating electric power from wind.

In the FOC method, internal loops are used to control electrical machines, creating ripples in both the current and the torque level of the machine. Moreover, this method gives a slow dynamic response compared to both direct torque control (DTC) and direct power control (DPC) [6]. Among the disadvantages of this method is that the total harmonic distortion (THD) value of the current is much higher compared to both DTC and DPC [7].

In the field of scientific research, there are several solutions that have been suggested in order to improve the performance and effectiveness of the FOC method. Artificial intelligence methods and nonlinear methods are among the most widely used methods. The authors of [8] implemented a super twisting algorithm (STA) on a multi-phase induction motor system using the FOC method where the intention was to minimize the current and torque ripples. Despite the advantages that are achieved using the STA algorithm, compared to the FOC strategy with PI controllers, the main drawback of this proposed method is the unstable frequency and presence of torque and current ripples due to the use of the pulse width modulation (PWM) technique. Another nonlinear FOC strategy is proposed in [9] to control the active and reactive power of the induction generator (IG). According to the study that was completed in [10] with the use of the nonlinear method based on second-order sliding mode control (SOSMC), the results were compared with the classical FOC method based on PI controller, and significant improvements were noticed in both the current and the active power compared to the classical FOC method. The problem with the proposed method is that there is a problem of chattering. However, one of the major drawbacks of employing the SOSMC technique to control electrical machines is that the mathematical form of the studied system must be precisely known. Another nonlinear method was applied to an asynchronous generator controlled by the FOC method. In this proposed method, the third-order sliding mode control (TOSMC) technique is used to compensate for the PI and improve the quality of the active power. One of the results obtained from this work is that the control by TOSMC reduces the chatter problem present in the SMC in addition to reducing the ripples at the level of torque, current, and effective power of the asynchronous generator. In [11], two different methods were combined in principle, where the first method is represented by neural networks, which is characterized by precision and speed of response, while the second method is the TOSMC technique, which is characterized by durability and unaffected by external factors such as noise or change in the parameters of the studied system. The result of combining these two methods is to obtain a more durable method and get excellent quality of the electric current of the induction generator.

In the field of electric machine control, response speed or dynamic response is very important. This response is considered a criterion among the criteria that is selected and differentiated between the control methods. In [12], a new method is proposed under the name of terminal synergetic control in order to improve the performance and effectiveness of the direct FOC strategy for an asynchronous generator integrated into a wind turbine. This proposed method is a new method based on the use of nonlinear error instead of using linear error. The obtained results showed the effectiveness of the proposed method in improving the dynamic response of the induction generator while reducing the ripples of torque and the active power of the induction generator. In [13], a method of artificial intelligence based on the neural networks and fuzzy logic was used to improve the performance and effectiveness of the IFOC strategy of the asynchronous generator, where the classic PI controller was replaced by an intelligent controller under the name of the

neuro-fuzzy controller. The proposed IFOC strategy with the neuro-fuzzy controller is more robust compared to the traditional IFOC strategy. The results showed the effectiveness of the proposed method in improving the THD of stator current value compared to the classical method. Moreover, there is a fast dynamic response to the proposed method, where we find that the response time is much reduced compared to the PI controller. In [14], the author proposed a new FOC strategy based on an adaptive observer to control the induction generator-based wind turbine. The proposed FOC strategy with adaptive observer was experimentally tested on an 11 kW induction generator using a 27 kW DC motor to rotate the generator. The experimental results showed the effectiveness of the proposed FOC method in obtaining a fast dynamic response compared to the classical FOC method.

In this work, a new FOC strategy method is introduced using a new nonlinear method in order to obtain a more robust method and reduce the ripples of torque, active power, and electric current generated by the double-powered induction generator (DPIG). In addition, a new method for the STA algorithm is proposed, where this method is simplified more than the traditional STA method. Thus, a simpler method that can be easily developed and applied to any system regardless of its complexity is designed. The proposed nonlinear method is called the simplified STA (SSTA) algorithm. Among the advantages of this nonlinear method is the fact that the design of the method is not related to the studied system, regardless of what this system will be. This proposed nonlinear method is used in order to improve the performance and effectiveness of the STA method and on the other hand to improve the dynamic response speed of the generator placed in the multi-rotor wind electric power generation system.

The novelty and main contributions of this work accomplished in this paper are summarized in the following points:


This work is structured as follows. Section 2 presents the dynamic modeling of the multi-rotor wind turbine system. In Sections 3 and 4, the proposed new super twisting algorithm (STA) and the proposed multilevel fuzzy SVM technique are presented. Section 5 presents the traditional IFOC strategy with the PI controller. Section 6 presents the proposed IFOC strategy using the proposed simplified STA controller and five-level fuzzy SVM technique. Finally, Section 7 concludes the work by presenting the main findings and future directions of research, as well as some comments and recommendations.

### **2. Multi-Rotor Wind Turbine System**

The multi-rotor wind turbine is a new technology that has appeared in recent years in order to overcome the problems and defects of the old technology (single-rotor wind turbine). In this new technology, the output torque or power gained from the wind is greater than that gained in the case of a single-rotor wind turbine. In addition, this new technology surpasses the winds generated by wind farms, and thus the yield is greater compared to the single-rotor wind turbine [15]. Figure 1 represents the electric power generation system using the DPIG placed in the electric power generation system using a multi-rotor wind turbine.

**Figure 1.** Structure of the multi-rotor wind turbine system. **Figure 1.** Structure of the multi-rotor wind turbine system.

In the studied wind system, a large-power DPIG (1.5 megawatts) is used. This generator is fed from the electrical network using two different inverters. The first inverter is on the side of the electrical network to convert alternating voltage into constant voltage, while the second inverter, which is on the side of the DPIG, aims to convert the constant voltage resulting from the first inverter into alternating voltage. To rotate the generator, a In the studied wind system, a large-power DPIG (1.5 megawatts) is used. This generator is fed from the electrical network using two different inverters. The first inverter is on the side of the electrical network to convert alternating voltage into constant voltage, while the second inverter, which is on the side of the DPIG, aims to convert the constant voltage resulting from the first inverter into alternating voltage. To rotate the generator, a dual-rotor wind turbine is used.

dual-rotor wind turbine is used. The two-rotor wind turbine is used in this work in order to increase the energy gained from the wind. On the other hand, a dual-rotor wind turbine is two turbines linked together forming one turbine. The two turbines have the same axis of rotation, and the torque and energy resulting from them can be expressed by the following equations: The two-rotor wind turbine is used in this work in order to increase the energy gained from the wind. On the other hand, a dual-rotor wind turbine is two turbines linked together forming one turbine. The two turbines have the same axis of rotation, and the torque and energy resulting from them can be expressed by the following equations:

$$P\_t = P\_{ST} + P\_{LT} \tag{1}$$

$$T\_t = T\_{ST} + T\_{LT} \tag{2}$$

where *Tt* is the output torque of the dual-rotor wind turbine, *TLT* and *TST* are the output torque of the large and small wind turbines, Pt is the output mechanical power of the dualrotor wind turbine, *PLT* and *PST* are the output mechanical power of the large and small wind turbine torque. where *T<sup>t</sup>* is the output torque of the dual-rotor wind turbine, *TLT* and *TST* are the output torque of the large and small wind turbines, *P<sup>t</sup>* is the output mechanical power of the dual-rotor wind turbine, *PLT* and *PST* are the output mechanical power of the large and small wind turbine torque.

The term torque for both large and small turbines is shown in Equations (3) and (4), respectively [16]. The term torque for both large and small turbines is shown in Equations (3) and (4), respectively [16].

$$T\_{LT} = \frac{\mathbf{C}\_p}{2\lambda\_{LT}^3} \rho \cdot \pi \cdot \mathbf{R}\_{LT}^5 \cdot w\_{LT}^2 \tag{3}$$

$$T\_{ST} = \frac{\mathcal{C}\_p}{2\lambda\_{ST}^3} \rho \cdot \pi \cdot \mathbf{R}\_{ST}^5 \cdot w\_{ST}^2 \tag{4}$$

From the two equations, it can be seen that the torque of the two turbines is related to each of the air density (*ρ*), coefficient of power (*Cp*), the mechanical speed of the small and large turbines (*wST* and *wLT*), the blade radius of the small and large turbines *(RST*, *RLT*), and the tip speed ration of the small and large turbines (*λST* and *λLT*). From the two equations, it can be seen that the torque of the two turbines is related to each of the air density (*ρ*), coefficient of power (*Cp*), the mechanical speed of the small and large turbines (*wST* and *wLT*), the blade radius of the small and large turbines *(RST*, *RLT*), and the tip speed ration of the small and large turbines (*λST* and *λLT*).

The energy gained from wind by a dual-rotor turbine is related to the power factor, the value of which is related to pitch angle (*β*) and tip speed ratio (*λ*). This parameter can be expressed by the following equation:

$$\mathcal{C}\_{p}(\beta,\lambda) = \frac{1}{0.08\beta + \lambda} + \frac{0.035}{\beta^3 + 1} \tag{5}$$

The terms of the energy produced by both the small turbine and the large turbine are shown in Equations (6) and (7), respectively [17].

$$P\_{ST} = \frac{\mathbb{C}\_p(\boldsymbol{\beta}, \boldsymbol{\lambda})}{2} \boldsymbol{\rho} \cdot \mathbb{S}\_{ST} \cdot \boldsymbol{w}\_{ST}^3 \tag{6}$$

$$P\_{\rm LT} = \frac{\mathbb{C}\_p(\boldsymbol{\beta}, \lambda)}{2} \rho \cdot \mathbb{S}\_{\rm LT} \cdot w\_{\rm LT}^3 \tag{7}$$

The value of the tip speed ratios of the small turbine and the large turbine are given in Equations (8) and (9), respectively [15].

$$
\lambda\_{ST} = \frac{w\_{ST} \cdot R\_{ST}}{V\_{ST}} \tag{8}
$$

$$
\lambda\_{LT} = \frac{w\_{LT} \cdot R\_{LT}}{V\_{LT}} \tag{9}
$$

The values of the tip speed ratios for a large and a small turbine are related to the wind speed (*VST* and *VLT*), The speed of the large and small turbine (*wST* and *wLT*), and the blade radius of the small and large turbines (*RST, RLT*).

The wind speed between the large and small turbines is different from the wind speed before the large turbines. The wind speed can be calculated at any point between the turbines given the following relationship [17]:

$$V\_{\mathbf{x}} = V\_{LT} \left( 1 - \frac{1 - \sqrt{(1 - \mathbf{C}\_T)}}{2} \left( 1 + \frac{2\mathbf{x}}{\sqrt{1 + 4\mathbf{x}^2}} \right) \right) \tag{10}$$

The wind speed of the small turbine is related to the wind speed of the large turbine (*VLT*) and a constant value (*C<sup>T</sup>* = 0.9), as well as the separation distance (*x*) between the large and small turbine. In this case, this distance between the center of the large and small turbine is 15 m [15].

In this work, a multi-rotor turbine is used to drive an DPIG. The latter was used in this work, and this is due to the advantages that distinguish it compared to other generators. As it is known, the DPIG is characterized by durability, easy control, and low cost.

In order to give the generator a mathematical form, the park transformation is used. To give the generator the mathematical form, the equations of voltage, flux, and mechanical equation are used. In addition, the torque equation is given for the generator, since torque is related to both current and flux. Quadrature and direct rotor voltages are shown in Equation (11). The direct and quadrature rotor flux are related to the stator/rotor current and are represented in Equation (12) [18–20].

$$\begin{cases} \begin{array}{c} V\_{dr} = \mathcal{R}\_r I\_{dr} - w\_r \Psi\_{qr} + \frac{d}{dt} \Psi\_{dr} \\ V\_{qr} = \mathcal{R}\_r I\_{qr} + w\_r \Psi\_{dr} + \frac{d}{dt} \Psi\_{qr} \end{array} \end{cases} \tag{11}$$

$$\begin{cases} \; \Psi\_{dr} = L\_r I\_{dr} + M I\_{ds} \\ \; \Psi\_{qr} = M I\_{qs} + L\_r I\_{qr} \end{cases} \tag{12}$$

Equation (13) represents each of the direct and quadrature stator voltages. Through this equation, the tension is related to stator resistance (*Rs*), direct and quadrature stator current (*Ids* and *Iqs*), direct and quadrature stator (Ψ*qs* and Ψ*ds*), and electrical pulsation of the stator (*ws*). The direct and quadrature stator flux are represented in Equation (14). These two fluxes are related to both the inductance of the stator (*Ls*), direct and quadrature rotor current (*Idr* and *Iqr*), and direct and quadrature rotor current (*Ids* and *Iqs*).

$$\begin{cases} \begin{array}{c} V\_{qs} = R\_s I\_{qs} + w\_s \mathbf{Y}\_{ds} + \frac{d}{dt} \mathbf{Y}\_{qs} \\ V\_{ds} = R\_s I\_{ds} - w\_s \mathbf{Y}\_{qs} + \frac{d}{dt} \mathbf{Y}\_{sd} \end{array} \end{cases} \tag{13}$$

$$\begin{cases} \; \Psi\_{qs} = MI\_{qr} + L\_s I\_{qs} \\ \; \Psi\_{ds} = L\_s I\_{ds} + MI\_{dr} \end{cases} \tag{14}$$

The value of the generated torque (*Te*) is related to rotor current, the number of pole pairs (*p*), and stator flux and its expression can be given by Equation (15) [19].

$$T\_{\varepsilon} = 1.5 \text{ } p \frac{M}{L\_s} \left( -\Psi\_{sd} I\_{rq} + \Psi\_{sq} I\_{rd} \right) \tag{15}$$

The active power and the reactive power of the generator are represented in the Equation (16). The active power is related to stator current and stator voltage, and the same is related to the reactive power.

$$\begin{cases} \quad P\_s = 1.5(V\_{qs}I\_{qs} + I\_{ds}V\_{ds})\\ \quad Q\_s = 1.5(-I\_{qs}V\_{ds} + V\_{qs}I\_{ds}) \end{cases} \tag{16}$$

The mechanical part of the DPIG is represented by Equation (17). This equation gives the relationship between torque and speed (Ω).

$$T\_{\varepsilon} = f \frac{d\Omega}{dt} + f\Omega + T\_r \tag{17}$$

where *T<sup>r</sup>* is the load torque, Ω is the mechanical rotor speed, *J* is the inertia, *f* is the viscous friction coefficient.

### **3. Simplified STA Controller**

Super twisting algorithm (STA) is among the most widely used nonlinear methods in the field of electrical machine control, due to its durability and ease of implementation [21]. The use of the STA algorithm in automated systems gives great effectiveness in improving the performance and efficiency of electrical machines. On the other hand, the STA algorithm is a type of SOSMC technique. The STA algorithm reduces chattering problems compared to the traditional SMC technique [22]. Equation (18) represents the super twisting algorithm [23]. The classical STA algorithm can be illustrated in Figure 2.

$$\begin{cases} \ u = \mathcal{K}\_p |e|^r \text{Sign } (e) + u\_1 \\ \ \frac{du\_1}{dt} = \mathcal{K}\_i \text{Sign } (e) \end{cases} \tag{18}$$

where *e* is the error or surface, *r* is the exponent defined for the traditional STA regulator, and *K<sup>i</sup>* and *K<sup>p</sup>* are positive values.

In this work, a new look is given to the STA algorithm in order to further simplify the algorithm and increase its dynamic response. Equation (19) represents the proposed method for the traditional STA algorithm which has been called the simplified STA algorithm (SSTA). The Lyapunov theory is used in order to check the stability of this proposed SSTA controller. This proposed SSTA algorithm is simpler compared to the traditional STA technique.

$$w = K \times |e|^r \times \text{Sign}\,\left(e\right) \tag{19}$$

$$
\dot{e} \times \dot{e} < 0 \tag{20}
$$

**Figure 2.** Structure of the traditional STA algorithm. ×ሶ < 0 (20)

Figure 3 illustrates the working principle of the proposed SSTA controller. Through this figure, the proposed SSTA controller can be implemented easily and inexpensively and can be applied to any system. Moreover, this controller does not require the mathematical form of the studied system. Figure 3 illustrates the working principle of the proposed SSTA controller. Through this figure, the proposed SSTA controller can be implemented easily and inexpensively and can be applied to any system. Moreover, this controller does not require the mathematical form of the studied system.

These two fluxes are related to both the inductance of the stator (Ls), direct and quadra-

௦ = ௦௦ + ௦ௗ௦ +

ௗ௦ = ௦ௗ௦ − ௦௦ +

௦ = + ௦௦ ௗ௦ = ௦ௗ௦ + ௗ

The value of the generated torque (*Te*) is related to rotor current, the number of pole

The active power and the reactive power of the generator are represented in the Equation (16). The active power is related to stator current and stator voltage, and the

<sup>൜</sup> ௦ = 1.5(௦௦ + ௗ௦ௗ௦)

Ω

where *Tr* is the load torque, Ω is the mechanical rotor speed, *J* is the inertia, *f* is the viscous

Super twisting algorithm (STA) is among the most widely used nonlinear methods in the field of electrical machine control, due to its durability and ease of implementation [21]. The use of the STA algorithm in automated systems gives great effectiveness in improving the performance and efficiency of electrical machines. On the other hand, the STA algorithm is a type of SOSMC technique. The STA algorithm reduces chattering problems compared to the traditional SMC technique [22]. Equation (18) represents the super twist-

() + ଵ

The mechanical part of the DPIG is represented by Equation (17). This equation gives

 ௦

 ௦ௗ

(−௦ௗ + ௦ௗ) (15)

<sup>+</sup> Ω+T୰ (17)

= () (18)

௦ = 1.5(− ௦ௗ௦ + ௦ௗ௦) (16)

(13)

(14)

ture rotor current (*Idr* and I*qr*), and direct and quadrature rotor current (*Ids* and I*qs*).

൞

൜

= 1.5

same is related to the reactive power.

friction coefficient.

**3. Simplified STA Controller** 

and *Ki* and *Kp* are positive values.

the relationship between torque and speed (Ω).

pairs (*p*), and stator flux and its expression can be given by Equation (15) [19].

=

ing algorithm [23]. The classical STA algorithm can be illustrated in Figure 2.

=||

ଵ

where *e* is the error or surface, *r* is the exponent defined for the traditional STA regulator,

ቐ

 ௦

**Figure 3.** Structure of the proposed SSTA technique. **Figure 3.** Structure of the proposed SSTA technique.

### **4. Proposed Five-Level Fuzzy SVM Strategy 4. Proposed Five-Level Fuzzy SVM Strategy**

Traditionally, the SVM technique is among the most used methods for controlling reflectors and this is because of the results it offers compared to other methods such as the PWM technique [24]. In this method, the calculation of reference voltage and zones is relied upon, which makes this method more complicated, especially in the case of a multilevel inverter [25]. To overcome this problem, a new idea for this method was presented in [26], where the calculation of the maximum and minimum values of the three-phase feeding voltages was used in this proposed method. In this proposed method, the reference tension is not used or calculated as well as the regions where the reference tension is present. This proposed method in [26] is simpler and more intuitive compared to the classical SVM method. In this part, the proposed fuzzy SVM technique is used to control a five-level inverter for an asynchronous generator placed in a multi-turbine wind system. This proposed method of controlling a 5-level inverter is illustrated in Figure 4. The pro-Traditionally, the SVM technique is among the most used methods for controlling reflectors and this is because of the results it offers compared to other methods such as the PWM technique [24]. In this method, the calculation of reference voltage and zones is relied upon, which makes this method more complicated, especially in the case of a multi-level inverter [25]. To overcome this problem, a new idea for this method was presented in [26], where the calculation of the maximum and minimum values of the three-phase feeding voltages was used in this proposed method. In this proposed method, the reference tension is not used or calculated as well as the regions where the reference tension is present. This proposed method in [26] is simpler and more intuitive compared to the classical SVM method. In this part, the proposed fuzzy SVM technique is used to control a fivelevel inverter for an asynchronous generator placed in a multi-turbine wind system. This proposed method of controlling a 5-level inverter is illustrated in Figure 4. The proposed method was used in [27] in order to give the five-level SVM technique.

posed method was used in [27] in order to give the five-level SVM technique. In this proposed fuzzy SVM technique, twelve hysteresis comparators and four trigonometric signals are used. The use of hysteresis comparators in this proposed SVM method creates a signal at the inverter output of a non-fixed frequency. In order to overcome this drawback, fuzzy logic algorithm (FLA) is used instead of using hysteresis comparators. The use of the FLA leads to a signal at the output of the inverter with a fixed frequency, thus reducing the ripples of both the current and the active power. The SVM method based on the FLA technique proposed in this work is illustrated in Figure 5. With this figure, the proposed method of SVM for controlling the five-level inverter is simpler compared to using the classical five-level SVM method using the calculation of reference voltage and zones. In this method, twelve FLA techniques are used in order to obtain control signals for the inverter transistors (IGBTs).

**Figure 4.** Structure of the five-level SVM technique. **Figure 4.** Structure of the five-level SVM technique.

In this proposed fuzzy SVM technique, twelve hysteresis comparators and four trigonometric signals are used. The use of hysteresis comparators in this proposed SVM method creates a signal at the inverter output of a non-fixed frequency. In order to overcome this drawback, fuzzy logic algorithm (FLA) is used instead of using hysteresis comparators. The use of the FLA leads to a signal at the output of the inverter with a fixed frequency, thus reducing the ripples of both the current and the active power. The SVM method based on the FLA technique proposed in this work is illustrated in Figure 5. With The internal form of the FLA method is shown in Figure 6. From this figure, it is noted that the structure of the FLA technique is simple. This fuzzy controller has two inputs, the error and the change in error, and only one output. Three constant gains (K1, K2, and K3) are used to improve the response and adjust the response of fuzzy logic. The characteristics of the FLA method used to improve the performance and effectiveness of the proposed five-level SVM technique are shown in the bottom of the Figure 6. The type of fuzzy controller used in this work is the Mamdani controller.

this figure, the proposed method of SVM for controlling the five-level inverter is simpler compared to using the classical five-level SVM method using the calculation of reference voltage and zones. In this method, twelve FLA techniques are used in order to obtain

control signals for the inverter transistors (IGBTs).

**Figure 5.** Structure of the proposed five-level fuzzy SVM technique. **Figure 5.** Structure of the proposed five-level fuzzy SVM technique.

Seven membership functions (MFs) are used in the first entry (error) and seven MFs are used in the second input (change in error). These functions used to accomplish fuzzy

Using the Matlab/Simulink software, the surface for the fuzzy logic controller used in this paper is given in order to compensate for the traditional hysteresis comparators. This surface of the fuzzy logic controller is shown in Figure 8. The use of the fuzzy logic technique leads to improving the performance and effectiveness of the proposed five-level SVM technique and thus obtaining a good quality of the electric current, and this is the

Compared to the PWM technique, the proposed five-level fuzzy SVM (FSVM) method is more complex and contains 12 fuzzy logic controllers which make it not easy and costly compared to the traditional PWM technique. However, in terms of the results obtained, the proposed method is better than the PWM technique. The proposed five-level fuzzy SVM method gives the output of the inverter a high-quality electrical signal (cur-

**Figure 6.** Structure and parameters of the fuzzy logic technique. **Figure 6.** Structure and parameters of the fuzzy logic technique.

main objective of this work.

rent) with a small value of THD.

logic, the 7 × 7 rule is used, where these rules are represented in Figure 7c.

Seven membership functions (MFs) are used in the first entry (error) and seven MFs are used in the second input (change in error). These functions used to accomplish fuzzy logic are shown in Figure 7a,b. In order to get a good response and results for the fuzzy logic, the 7 × 7 rule is used, where these rules are represented in Figure 7c. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 11 of 24

**Figure 7.** Membership functions and fuzzy rules. **Figure 7.** Membership functions and fuzzy rules.

Using the Matlab/Simulink software, the surface for the fuzzy logic controller used in this paper is given in order to compensate for the traditional hysteresis comparators. This surface of the fuzzy logic controller is shown in Figure 8. The use of the fuzzy logic technique leads to improving the performance and effectiveness of the proposed five-level SVM technique and thus obtaining a good quality of the electric current, and this is the main objective of this work. **Figure 7.** Membership functions and fuzzy rules.

structure. Several scientific works dealt with this method [29–31], where four PI control-

indirect FOC method is more complex than the direct FOC method [32], but the indirect FOC method provides a better dynamic response than the direct FOC method. Moreover, the indirect FOC method reduces the ripples of torque, reactive power, and flux compared to the direct FOC method [33]. In order to control the generator inverter, the PWM tech-

lers are used in this method, which makes the dynamic response much faster. Thus, the indirect FOC method is more complex than the direct FOC method [32], but the indirect FOC method provides a better dynamic response than the direct FOC method. Moreover, the indirect FOC method reduces the ripples of torque, reactive power, and flux compared to the direct FOC method [33]. In order to control the generator inverter, the PWM tech-

nique is used. This method is based on the principle shown in Equation (21) [34].

The IFOC method is a type of FOC method which offers a fast dynamic response compared to the direct FOC method [28]. The indirect FOC method is a method that dif-

௦ = 0 and ௗ௦ = ௦ (21)

௦ = 0 and ௗ௦ = ௦ (21)

lers are used in this method, which makes the dynamic response much faster. Thus, the **Figure 8.** Control surface. **Figure 8.** Control surface.

**5. Traditional IFOC Strategy** 

Compared to the PWM technique, the proposed five-level fuzzy SVM (FSVM) method is more complex and contains 12 fuzzy logic controllers which make it not easy and costly compared to the traditional PWM technique. However, in terms of the results obtained, the proposed method is better than the PWM technique. The proposed five-level fuzzy SVM method gives the output of the inverter a high-quality electrical signal (current) with a small value of THD.

### **5. Traditional IFOC Strategy**

The IFOC method is a type of FOC method which offers a fast dynamic response compared to the direct FOC method [28]. The indirect FOC method is a method that differs from the direct FOC method in terms of the principle of work and in terms of internal structure. Several scientific works dealt with this method [29–31], where four PI controllers are used in this method, which makes the dynamic response much faster. Thus, the indirect FOC method is more complex than the direct FOC method [32], but the indirect FOC method provides a better dynamic response than the direct FOC method. Moreover, the indirect FOC method reduces the ripples of torque, reactive power, and flux compared to the direct FOC method [33]. In order to control the generator inverter, the PWM technique is used. This method is based on the principle shown in Equation (21) [34].

$$\Psi\_{qs} = 0 \text{ and } \Psi\_{ds} = \Psi\_s \tag{21}$$

Using Equation (21), the direct and quadrature stator voltages of the generator become as follows:

$$\begin{cases} \ V\_{ds} = V\_s = w\_s \Psi\_s\\ V\_{qs} = 0 \end{cases} \tag{22}$$

Using Equations (13) and (22), the direct and quadrature stator currents of the generator become as follows:

$$\begin{cases} \ I\_{ds} = -\frac{M}{L\_s} I\_{dr} + \frac{\Psi\_s}{L\_s} \\\ I\_{qs} = -\frac{M}{L\_s} I\_{qr} \end{cases} \tag{23}$$

Relying on Equations (11) and (23), the direct and indirect rotor voltages of the generator become as follows [34]:

$$\begin{cases} V\_{qr} = R\_{dr} I\_{qr} + \left( L\_r - \frac{M^2}{L\_s} \right) w\_r I\_{qr} + g \frac{MV\_s}{L\_s} \\ \qquad V\_{dr} = R\_{dr} I\_{dr} - w\_r \left( L\_r - \frac{M^2}{L\_s} \right) I\_{qr} \end{cases} \tag{24}$$

The direct and quadrature rotor flux of a generator can be expressed by the following equation [33]:

$$\begin{cases} \begin{array}{c} \Psi\_{dr} = \left(L\_r - \frac{M^2}{L\_s}\right) I\_{dr} + \frac{M}{w\_s L\_s} V\_s\\ \Psi\_{qr} = \left(L\_r - \frac{M^2}{L\_s}\right) I\_{qr} \end{array} \tag{25}$$

By Equation (25), direct rotor flux is related to both stator voltage and quadrature rotor current. As for the quadrature rotor flux, it is related to the quadrature rotor current of the generator.

From Equations (21)–(25), the internal structure of the indirect FOC strategy can be given in Figure 9. Through this figure, we note that this technique controls the reactive and active power by controlling the quadrature and direct rotor voltages (*Vqr*\* and *Vdr*\*).

technique.

**Figure 9.** Traditional indirect FOC strategy. **Figure 9.** Traditional indirect FOC strategy.

Figure 10 represents the total system used in this paper, where a multi-rotor wind turbine (MRWT) was used to rotate the generator (DPIG). The latter is controlled by the In order to estimate the active and reactive power, Equation (26) is used. In order to estimate the two values, we need to measure both rotor voltage and rotor current.

$$\begin{cases} Q\_s = -1.5 \left( \frac{\omega\_s \mathbf{v}\_s M}{L\_s} I\_{dr} - \frac{\omega\_s \mathbf{v}\_s^2}{L\_s} \right) \\\ P\_s = -1.5 \frac{\omega\_s \mathbf{v}\_s M}{L\_s} I\_{qr} \end{cases} \tag{26}$$

and flux compared to DPC and DTC. Moreover, it has a long dynamic response and gives a poor quality of electric current and active power. The reason for these shortcomings in this traditional indirect FOC strategy is due to the use of both the traditional PWM technique and conventional PI controller. The use of these classic methods (PI and PWM) makes the indirect FOC strategy not robust and char-Figure 10 represents the total system used in this paper, where a multi-rotor wind turbine (MRWT) was used to rotate the generator (DPIG). The latter is controlled by the traditional IFOC method. In this work, the reference value of the reactive power (*Qs*\*) is set to 0 Var. The reference value of the active power (*Ps*\*) is obtained using the MPPT technique. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 14 of 24

**Figure 10.** Traditional IFOC technique of the DPIG-based MRWT system. **Figure 10.** Traditional IFOC technique of the DPIG-based MRWT system.

FOC method is a change from the classic indirect FOC method, where the proposed SSTA controller is used in place of the traditional PI controller and the five-level fuzzy SVM technique is used instead of the PWM technique. This proposed indirect FOC method aims to control the active and reactive power of the generator placed in the multi-rotor wind turbine system. On the other hand, this proposed indirect FOC method reduces the ripples of torque, active power, and electric current compared to the classical indirect FOC method. The proposed indirect FOC strategy is shown in Figure 11. From this figure, to control the active/reactive power, two proposed SSTA controllers are used and the same

In this proposed indirect FOC method, the same equations used to estimate both the active and reactive power are used in the classical indirect FOC strategy. Therefore, it can be said that this designed indirect FOC strategy is more robust than the rest of the controls such as the classical indirect FOC strategy and DTC. The objective of this designed indirect FOC strategy is to obtain high-quality *Vqr*\* and *Vdr*\* from active and reactive power references for the inverter DPIG control. Controlling the latter very well leads to obtaining a high quality of stator current and active power. On the other hand, the reactive power reference is set to zero. As for the reference value of the active power, it is obtained using the maximum power point tracking (MPPT) technique. The system studied using the proposed indirect FOC method is represented in Figure 12, where almost the same structure

as the classical indirect FOC method is preserved.

**6. Proposed Indirect FOC Strategy** 

with the reactive power.

The indirect FOC strategy gives more ripples in the current, torque, active power, and flux compared to DPC and DTC. Moreover, it has a long dynamic response and gives a poor quality of electric current and active power.

The reason for these shortcomings in this traditional indirect FOC strategy is due to the use of both the traditional PWM technique and conventional PI controller. The use of these classic methods (PI and PWM) makes the indirect FOC strategy not robust and characterized by a long dynamic response compared to some methods such as the DTC strategy.

In order to improve the performance and efficacy of the traditional indirect FOC strategy, a novel scheme for the traditional indirect FOC strategy is proposed in Section 6. This proposed indirect FOC strategy is based on the use of both the proposed SSTA algorithm and the multilevel fuzzy SVM strategy.

### **6. Proposed Indirect FOC Strategy**

In this part, a new idea for indirect FOC strategy is presented based on the proposed simplified STA controller and the five-level fuzzy SVM strategy. The proposed indirect FOC method is a change from the classic indirect FOC method, where the proposed SSTA controller is used in place of the traditional PI controller and the five-level fuzzy SVM technique is used instead of the PWM technique. This proposed indirect FOC method aims to control the active and reactive power of the generator placed in the multi-rotor wind turbine system. On the other hand, this proposed indirect FOC method reduces the ripples of torque, active power, and electric current compared to the classical indirect FOC method. The proposed indirect FOC strategy is shown in Figure 11. From this figure, to control the active/reactive power, two proposed SSTA controllers are used and the same with the reactive power. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 15 of 24

**Figure 11.** Proposed indirect FOC strategy. **Figure 11.** Proposed indirect FOC strategy.

In this proposed indirect FOC method, the same equations used to estimate both the active and reactive power are used in the classical indirect FOC strategy. Therefore, it can be said that this designed indirect FOC strategy is more robust than the rest of the controls such as the classical indirect FOC strategy and DTC. The objective of this designed indirect FOC strategy is to obtain high-quality *Vqr*\* and *Vdr*\* from active and reactive power references for the inverter DPIG control. Controlling the latter very well leads to obtaining a high quality of stator current and active power. On the other hand, the reactive power reference is set to zero. As for the reference value of the active power, it is obtained using

In Table 1, a comparison is given between the classical indirect FOC (IFOC) technique and the proposed indirect FOC method. Through this table, the proposed indirect FOC strategy is more robust and reduces torque and current ripples compared to the classical indirect FOC technique. Moreover, the proposed indirect FOC method improves the rise time, response dynamic, THD value of current, and power quality compared to the classical indirect FOC technique. However, the proposed indirect FOC method is more complicated than the classical indirect FOC method due to the use of the five-level fuzzy SVM (FSVM) technique (instead, the classical indirect FOC method uses the PWM technique).

**Figure 12.** Proposed indirect FOC strategy of DPIG-based MRWT system.

the maximum power point tracking (MPPT) technique. The system studied using the proposed indirect FOC method is represented in Figure 12, where almost the same structure as the classical indirect FOC method is preserved. **Figure 11.** Proposed indirect FOC strategy.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 15 of 24

**Figure 12.** Proposed indirect FOC strategy of DPIG-based MRWT system. **Figure 12.** Proposed indirect FOC strategy of DPIG-based MRWT system.

In Table 1, a comparison is given between the classical indirect FOC (IFOC) technique and the proposed indirect FOC method. Through this table, the proposed indirect FOC strategy is more robust and reduces torque and current ripples compared to the classical indirect FOC technique. Moreover, the proposed indirect FOC method improves the rise time, response dynamic, THD value of current, and power quality compared to the classical indirect FOC technique. However, the proposed indirect FOC method is more complicated than the classical indirect FOC method due to the use of the five-level fuzzy SVM (FSVM) technique (instead, the classical indirect FOC method uses the PWM technique). In Table 1, a comparison is given between the classical indirect FOC (IFOC) technique and the proposed indirect FOC method. Through this table, the proposed indirect FOC strategy is more robust and reduces torque and current ripples compared to the classical indirect FOC technique. Moreover, the proposed indirect FOC method improves the rise time, response dynamic, THD value of current, and power quality compared to the classical indirect FOC technique. However, the proposed indirect FOC method is more complicated than the classical indirect FOC method due to the use of the five-level fuzzy SVM (FSVM) technique (instead, the classical indirect FOC method uses the PWM technique).

**Table 1.** A comparative study between the classical method and the proposed IFOC strategy.


The result of using the five-level fuzzy SVM technique in the proposed IFOC method makes the proposed IFOC method not simple and somewhat complicated compared to the classical IFOC method, where the latter uses PWM, which leads to problems in the case of achieving this proposed IFOC-SSTA-FSVM method.

To implement the proposed method (IFOC-SSTA-FSVM) empirically, there are several problems of implementation related to the large financial cost of completing this project due to the complexity of controlling the MRWT system. On the other hand, there is the complexity of the use of the five-level fuzzy SVM technique to control the inverter of the DPIG-based MRWT system. Moreover, the use of the maximum power point tracking technique increases the complexity and financial cost of the studied MRWT system. Nonetheless, given the results obtained in improving the quality of electric power and the importance of the MRWT system in improving the performance of classical wind turbines and in reducing the size of wind farms, this system is very necessary for the near future.

In the next part, the results are confirmed and the robustness of the proposed indirect FOC method is verified using the Matlab/Simulink software.

### **7. Results**

In order to verify the proposed indirect FOC method, the Matlab/Simulink software is used. The results of the proposed indirect FOC method are compared with the classical indirect FOC method in terms of the ratio of ripples at the level of torque, current, effective power, and reactive power. The two methods are also compared in terms of the THD value of the electric current.

In this work, a generator with the following data is used: 50 Hz, 380/696 V, *R<sup>s</sup>* = 0.012 Ω, *<sup>L</sup><sup>r</sup>* = 0.0136 H, *<sup>L</sup><sup>m</sup>* = 0.0135 H, *<sup>p</sup>* = 2, *<sup>J</sup>* = 1000 kg·m<sup>2</sup> , *Psn* = 1.5 MW, *R<sup>r</sup>* = 0.021 Ω, *L<sup>s</sup>* = 0.0137 H, and *f<sup>r</sup>* = 0.0024 N·m/s [35].

In this work, a multi-rotor wind turbine with the following data is used: *R*<sup>1</sup> = 13.2 m, *<sup>R</sup>*<sup>2</sup> = 25.5 m, *<sup>r</sup>*<sup>1</sup> = 1 m, *<sup>r</sup>*<sup>2</sup> = 0.5 m, *<sup>r</sup><sup>g</sup>* = 0.75 m, *<sup>J</sup>*<sup>1</sup> = 500 kg·m<sup>2</sup> , *<sup>J</sup>*<sup>2</sup> = 1000 kg·m<sup>2</sup> , *G*<sup>1</sup> *= r1/rg*, and *G*<sup>2</sup> *= r*2*/rg*.

In this work, the proposed indirect FOC method is tested in the case of two tests, the first test is a tracking test and the second test is to study the behavior of the proposed indirect FOC method in comparison with the classical indirect FOC method in the event of a change in the generator parameters. This is in order to know the robustness of the proposed indirect FOC method with the classical indirect FOC method.

### *7.1. First Test*

In this test, the behavior of the reference tracking is studied, for both the proposed IFOC-SSTA method and the classical IFOC method, where the obtained results are shown in Figure 13. Through this figure, the active and reactive power follow the references perfectly, and this is for the two IFOC methods with a preference for the proposed method in the dynamic response (see Figure 13a,b). The proposed IFOC-SSTA method gave better results in terms of ripples for both active and reactive power compared to the classical IFOC method (see Figure 14a,b). In Figure 13c, the generated torque has the same shape as the active power, where it can be seen that the increase in the active power corresponds to the increase in the torque. In addition, the proposed IFOC-SSTA method reduced torque ripples compared to the classical IFOC method (Figure 14c).

Regarding the current generated by the generator, it is shown in Figure 15d. Through this figure, the electric current takes the form of the active power, where it is noted that the behavior of the current is the same as the behavior of the active power. Moreover, the proposed IFOC-SSTA method gave excellent results in terms of electric current ripples and quality compared to the classical IFOC method (see Figure 14d).

Figure 13e,f represent the THD value of the proposed and classical IFOC method, respectively. Through these two forms, the proposed IFOC-SSTA method reduced the THD value of the electric current excellently compared to the classical IFOC method and the reduction ratio was about 96.72%. These results confirm the robustness of the proposed IFOC method in improving the quality of the effective power and electric current.

**Figure 13.** First test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current; (**e**) THD value of stator current (IFOC-SSTA strategy); (**f**) THD value of stator current (IFOC strategy). **Figure 13.** First test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current; (**e**) THD value of stator current (IFOC-SSTA strategy); (**f**) THD value of stator current (IFOC strategy). *Sustainability* **2022**, *14*, x FOR PEER REVIEW 18 of 24

the dynamic response (see Figure 13a,b). The proposed IFOC-SSTA method gave better results in terms of ripples for both active and reactive power compared to the classical IFOC method (see Figure 14a,b). In Figure 13c, the generated torque has the same shape as the active power, where it can be seen that the increase in the active power corresponds to the increase in the torque. In addition, the proposed IFOC-SSTA method reduced

Regarding the current generated by the generator, it is shown in Figure 15d. Through this figure, the electric current takes the form of the active power, where it is noted that the behavior of the current is the same as the behavior of the active power. Moreover, the proposed IFOC-SSTA method gave excellent results in terms of electric current ripples

Figure 13e,f represent the THD value of the proposed and classical IFOC method, respectively. Through these two forms, the proposed IFOC-SSTA method reduced the THD value of the electric current excellently compared to the classical IFOC method and the reduction ratio was about 96.72%. These results confirm the robustness of the proposed IFOC method in improving the quality of the effective power and electric current. In the next test, the proposed IFOC method is further confirmed if the parameters of

torque ripples compared to the classical IFOC method (Figure 14c).

and quality compared to the classical IFOC method (see Figure 14d).

the generator installed in the multi-rotor wind turbine system are changed.

**Figure 14.** Zoom in the first test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current. **Figure 14.** Zoom in the first test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Time (s)

Fundamental (50Hz) = 1513 , THD= 3.40%

<sup>0</sup> <sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>0</sup>

Frequency (Hz)

Time (s)

Qs(IFOC-PI) Qs(IFOC-SSTA) Qsref

Ias(IFOC-PI) Ias(IFOC-SSTA)

(**a**) (**b**)

Stator current Ias (A)




0

5 x 105

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

Time (s)

Fundamental (50Hz) = 1864 , THD= 0.11%

<sup>0</sup> <sup>500</sup> <sup>1000</sup> <sup>1500</sup> <sup>2000</sup> <sup>0</sup>

Frequency (Hz)

Time (s)

Ps(IFOC-PI) Ps(IFOC-SSTA) Psref



0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16

M ag (% of Fundam ental)


Torque Te (N.m)

0

5000

Active power Ps (W)

(**c**) (**d**)

Te(IFOC-PI) Te(IFOC-SSTA)

 (**e**) (**f**)

**Figure 15.** Second test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current;

(**e**) THD value of current (IFOC-SSTA strategy); (**f**) THD value of current (IFOC strategy).

M ag (% of Fundam ental) 5.96 5.98 6 6.02 x 105

Active power Ps (W)

Torque Te (N.m)

(**a**) (**b**)

Reactive power Qs (VAR)


1700

1720

1740

Stator current Ias (A)

1760



x 105

Qs(IFOC-PI) Qs(IFOC-SSTA) Qsref

Ps(IFOC-PI) Ps(IFOC-SSTA) Psref

(**c**) (**d**)

**Figure 14.** Zoom in the first test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator

0.794 0.795 0.796 0.797 0.798 0.799 0.8

0.3735 0.374 0.3745 0.375 0.3755

Ias(IFOC-PI) Ias(IFOC-SSTA)

Time (s)

Time (s)

current.

0.7982 0.7984 0.7986 0.7988 0.799 0.7992 0.7994 0.7996 0.7998

Te (IFOC-PI) Te (IFOC-SSTA)

Time (s)

0.7925 0.793 0.7935 0.794 0.7945 0.795

Time (s)

**Figure 15.** Second test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current; (**e**) THD value of current (IFOC-SSTA strategy); (**f**) THD value of current (IFOC strategy). **Figure 15.** Second test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current; (**e**) THD value of current (IFOC-SSTA strategy); (**f**) THD value of current (IFOC strategy).

In the next test, the proposed IFOC method is further confirmed if the parameters of the generator installed in the multi-rotor wind turbine system are changed.

The results obtained from this test are shown in Table 2. Table 2 represents the value of ripples in each of torque, current, active power, and reactive power, and this is for the two IFOC methods. Through this table, the proposed IFOC-SSTA method reduced the ratio of ripples of torque, active power, current, and reactive power by excellent ratios, and the ratios were as follows: 91.66%, 91.20%, 84.21%, and 93.75%, respectively.

**Table 2.** Comparative ripples obtained from the traditional IFOC with the proposed IFOC strategy.


In Table 3, the response time is extracted for each of the active power, torque, and reactive power of the two IFOC methods proposed in this work. Through this table, the proposed IFOC-SSTA method gave a small response time compared to the classical IFOC method, and the improvement ratio was about 97.54%, 97.54%, and 98.23% for the active power, torque, and reactive power, respectively. Moreover, the proposed method gave a good response time in comparison with both the DPC strategy and neuro-second order sliding mode control (NSOSMC) completed in [36] (see Table 4).

**Table 3.** Response time.


**Table 4.** Comparative analysis of response time.


### *7.2. Second Test*

The results of the second test are shown in Figure 15. In this test, the generated parameter values were changed in order to know the change in the behavior of the proposed IFOC-SSTA method compared to the classical IFOC method, as well as its robustness. In this test, Rs, Ls, Rr, Lm, and Lr were changed to the values 0.024 Ω, 0.00685 H, 0.042 Ω, 0.00675 H, and 0.0068 H, respectively. Zoom is given for torque, current, effective power, and reactive power in Figure 16. In Figures 15 and 16, there is a noticeable effect on the level of torque, current, active, and reactive power, where the classical IFOC method was affected by changing the generator parameters more than the IFOC-SSTA method. Moreover, active and reactive power keep following the references well in this test for both the proposed and the classical IFOC method (see Figure 15a,b). Both current and torque take the same form as active power (see Figure 15c,d) with ripples in both torque and current levels. The proposed IFOC-SSTA method reduced these ripples as compared to the classical IFOC method (see Figure 16c,d). Moreover, the proposed IFOC-SSTA method also reduced the ripples in both the active and reactive power compared to the classical IFOC method (see Figure 16a,b). The THD value of the electric current is shown in Figure 15e,f and this is for both the proposed and the classical IFOC method, respectively. Through the two figures, the proposed IFOC-SSTA method reduced the THD value of the electric current compared with the classical IFOC method.

The results obtained from this test are shown in Table 5. Through this table, the proposed IFOC-SSTA method reduced the ripples of torque, reactive power, current, and active power by 93.35%, 98%, 87.50%, and 83.33%, respectively.

The proposed IFOC-SSTA-FSVM method in this work provided excellent results compared to the classical IFOC method in terms of reducing the ripples of torque, current, and active power, as well as improving the quality of electric power. All of these factors help reduce malfunctions and thus reduce maintenance costs and help extend the life of the system as a whole.

current compared with the classical IFOC method.

**Figure 16.** Zoom in the second test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current. **Figure 16.** Zoom in the second test results. (**a**) Active power; (**b**) Reactive power; (**c**) Torque; (**d**) Stator current.

and this is for both the proposed and the classical IFOC method, respectively. Through the two figures, the proposed IFOC-SSTA method reduced the THD value of the electric



Reactive power ripple (VAR) Around 25,000 Around 500 98%

Active power ripple (W) Around 30,000 Around 5000 83.33% Stator current (A) Around 40 Around 5 87.50% The proposed IFOC-SSTA-FSVM method in this work provided excellent results compared to the classical IFOC method in terms of reducing the ripples of torque, current, and active power, as well as improving the quality of electric power. All of these factors help reduce malfunctions and thus reduce maintenance costs and help extend the life of The following is a comparison between the proposed IFOC-SSTA-FSVM method and some published works in terms of the THD value of the electric current. The results of the comparison are recorded in Table 6. Through this table, the proposed IFOC-SSTA-FSVM method significantly reduced the THD value compared to several published methods, which indicates the robustness of the proposed IFOC-SSTA-FSVM method and the effectiveness of the proposed IFOC-SSTA-FSVM method in improving the quality of electrical energy.

> the system as a whole. The following is a comparison between the proposed IFOC-SSTA-FSVM method and **Table 6.** Comparative results with other techniques.



**Table 6.** *Cont.*

### **8. Conclusions**

In this work, a new idea was given for the IFOC method based on both a simplified STA controller and a five-level fuzzy SVM technique. This proposed method was verified using the Matlab/Simulink software, comparing the results obtained with the traditional method. The application of the proposed method in the wind system led to the improvement of the dynamic response of the generator and the improvement of the quality of the electric current with the electric energy.

The points drawn from this work are illustrated in the following points:


In a future paper, to ameliorate the quality of the active power and current, the DPIG will be controlled using another robust control scheme, such as fractional order synergetic control and feedback PI controller [46,47].

**Author Contributions:** Validation, N.B., P.T. and N.T.; conceptualization, H.B.; software, H.B.; methodology, H.B.; investigation, H.B. and I.C.; resources, N.B., P.T. and N.T.; project administration, N.B.; data curation, N.B., P.T. and N.T.; writing—original draft preparation, H.B.; supervision, N.B. and I.C.; visualization: I.C. and N.B.; formal analysis: I.C. and N.B.; funding acquisition: N.B., P.T. and N.T.; writing—review and editing: I.C., N.B., P.T., N.T. and H.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported in part by the Framework Agreement between University of Pitesti (Romania) and King Mongkut's University of Technology North Bangkok (Thailand), in part by an International Research Partnership "Electrical Engineering–Thai French Research Center (EE-TFRC)" under the project framework of the Lorraine Université d'Excellence (LUE) in cooperation between Université de Lorraine and King Mongkut's University of Technology North Bangkok, in part by the National Research Council of Thailand (NRCT) under Senior Research Scholar Program under Grant No. N42A640328, and in part by National Science, Research, and Innovation Fund (NSRF) under King Mongkut's University of Technology North Bangkok under Grant no. KMUTNB-FF-65-20.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

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

## **References**


**Ephraim Bonah Agyekum 1,\* , Usman Mehmood 2,3 , Salah Kamel <sup>4</sup> , Mokhtar Shouran <sup>5</sup> , Elmazeg Elgamli 5,\* and Tomiwa Sunday Adebayo 6,7**


**Abstract:** Power distribution to decentralized and remote communities secluded from centralized grid connections has always been a problem for utilities and governments worldwide. This situation is even more critical for the isolated communities in Russia due to the vast nature of the country. Therefore, the Russian government is formulating and implementing several strategies to develop its renewable energy sector. However, very little information is available on the possible performance of solar photovoltaic (PV) modules under Russian weather conditions for all year round. Thus, this study has been designed to fill that research gap by assessing the performance ratio (*PR*), degradation, energy loss prediction, and employment potential of PV modules in the Sverdlovsk region of Russia using the PVsyst simulation model. A side-by-side comparison of the fixed tilted plane and tracking horizontal axis East–West were analyzed. According to the results, the annual production probability (P) for the fixed PV module for a P50, P75, and P90 is 39.68 MWh, 37.72 MWh, and 35.94 MWh, respectively, with a variability of 2.91 MWh. In the case of the tracking PV module, the annual production probability for the P50, P75, and P90 is 43.18 MWh, 41.05 MWh, and 39.12 MWh, respectively, with a variability of 3.17 MWh. A *PR* of 82.3% and 82.6% is obtained for the fixed and tracking systems, respectively, while the PV array losses for the fixed and tracking orientations are 15.1% and 14.9%, respectively. The months of May to August recorded the highest array losses due to the high temperatures that are usually recorded within that period.

**Keywords:** renewable energy; solar photovoltaic energy; degradation rate; PVsyst software; energy loss prediction

### **1. Introduction**

One challenge confronting the world today is how to generate energy in a more sustainable way to meet its energy needs while maintaining environmental security. The world's demand for electricity is growing due to increasing global population, improved lifestyle, and industrialization [1]. It has been estimated that fossil fuel forms about 80% of the world's primary energy, and energy consumption globally is expected to increase by 2.3% each year from 2015 to 2040 [2]. The concentration of atmospheric carbon dioxide (CO2) equivalent is said to have almost doubled since the inception of the Industrial

**Citation:** Agyekum, E.B.; Mehmood, U.; Kamel, S.; Shouran, M.; Elgamli, E.; Adebayo, T.S. Technical Performance Prediction and Employment Potential of Solar PV Systems in Cold Countries. *Sustainability* **2022**, *14*, 3546. https:// doi.org/10.3390/su14063546

Academic Editors: Nicu Bizon, Bhargav Appasani and Mamadou Baïlo Camara

Received: 11 February 2022 Accepted: 12 March 2022 Published: 17 March 2022

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

**Copyright:** © 2022 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/).

Revolution [3]. This has increased the average global temperature, which is negatively impacting global climate [4,5].

Solar photovoltaic (PV) technology can generate power by directly converting incident solar radiation to electrical power [6,7]. PV technology is one of the renewable energy (RE) options that can help to decarbonize the world to decrease greenhouse gas (GHG) emissions. The continual drop in the cost of PV systems and the formulation of policies by various governments to promote the development and use of RE technologies has led to the rapid growth of the PV industry [8]. The PV industry has witnessed a composite yearly growth rate of more than 40% during the last 15 years, making it one of the fastgrowing industries globally. This has necessitated the need to improve project designs and continuous monitoring and prediction of the performance of the PV systems that have either been installed or yet to be installed to ensure their reliability and performance [8].

Therefore, several researchers have performed studies that either assess the performance of already installed or yet to be installed PV power plants in several countries. Malvoni et al. [8] examined a 960 kWp monocrystalline silicon PV system's performance in southern Italy. They obtained a capacity factor (*CF*) and performance ratio (*PR*) of 15.6% and 84.4%, respectively. Kumar et al. [9] studied the performance, degradation, and energy loss of a 200 kW roof-integrated crystalline PV system in northern India. According to their results, the PV system is projected to operate with an annual *PR*, *CF*, and energy loss of 77.27%, 16.72%, and −26.5%, respectively. Ramanan et al. [10] evaluated the performance of two colocated grid-connected PV power plants, which consist of copper indium selenium (CIS) and polycrystalline silicon (p-Si) arrays. A yearly *PR* of 86.73% for the CIS and 78.48% for p-Si were obtained, while the capacity utilization factor ranges from 17.99% for the p-Si to 19.57% for the CIS systems. Similarly, Ameur et al. [11] analyzed and compared several indices that affect the performance of different grid-connected PV technologies, i.e., polycrystalline silicon (pc-Si), amorphous silicon (a-Si), and monocrystalline silicon (mc-Si) with capacities of 2 kWp each. The study's outcome suggests that the polycrystalline, monocrystalline, and amorphous technologies generated a four-year yearly alternating current (*AC*) energy average of 3239 kWh, 3246 kWh, and 2797 kWh with yearly *PR* of 77%, 77%, and 73%, respectively. Dahmoun et al. [12] explored and assessed the performance of a 23.92 MWp polycrystalline PV power plant located in El Bayadh in Algeria. Their study shows that the degradation of the *PR* during the study period is estimated to be 0.76% per year.

Furthermore, Kittner et al. [13] evaluated the economic investment, embodied energy, and CO<sup>2</sup> payback for amorphous silicon thin film and single crystalline systems. They reported that the amorphous silicon thin-film panels have higher net environmental and economic benefits. Padmavathi and Daniel [14] worked on a 3 MW grid-connected polycrystalline PV power plant in India. They evaluated normalized technical performance parameters for the system for the year 2011. The generated yearly average energy by the plant was 1372 kWh per kWp. Radue and van Dyk [15] reported losses of up to 30% for thin-film PV power plants sited in South Africa after a period of 14 months. Kymakis et al. [16] also estimated the performance of a 171 kWp polycrystalline silicon PV power plant installed on the island of Crete. After one year, the average annual *PR* and *CF* were 67.36% and 15.26%, respectively. Belmahdi and El Bouardi [17] evaluated the performance of a 1 MW solar PV power plant under Moroccan weather conditions using the PVsyst software. The optimal angle for the installation of the PV module for the study area in summer was identified to be 32◦ for fixed tilt and 48◦ for winter on seasonal adjustment tilts. A 60 kWp PV module was modelled for the Uttar-Pradesh area in India using the PVsyst software by [18]. Their system generated a total of 89.5 MWh per year with a performance ratio of 73.73%. In the Philippines, Dellosa et al. [19] assessed the technical and economic performance of 5 MWp PV system for that country using the PVsyst software. The results from their study revealed that the temperature of the PV panel accounted for the highest energy loss; it accounted for 8% of the loss. A payback period of 4.23 years was recorded for the system.

Finally, Chabachi et al. [20] assessed the performance of a poly-Si/6 MWp in Southwest Algeria. An average monthly efficiency for their PV array was 12.68% with a *PR* of 84%. Yadav et al. [21] employed the PVsyst software to examine the performance of a 295 Wp Si-poly PV module SYP295S under Nepal's weather conditions at Tribhuvan University. Their simulation results revealed that a total of 110 kWp of PV power plant's output would be enough for the whole campus. The designed system produced a surplus energy of 115.1 MWh/year, which can be exported to grid. A 2.4 kWp monocrystalline PV module was studied by [22] at the Mulhouse campus, France. The studied power plant generated a total of 5597.65 kWh of energy. Chandel and Chandel [23] conducted a performance assessment on a 19-MWp (17-MWac) PV plant installed with seasonal adjustable tilt (AT), fixed tilt (FT), and horizontal single-axis solar tracking (HSAT) configurations in India. Key findings from their study suggest that the studied FT system recorded an annual *PR* of 79% with a *CF* of 19%. The AT also recorded a *CF* of 20%, while the HSAT recorded 22%. Kumar et al. [24] also assessed a 10 kWp poly-Si PV power plant for the remote islands of Andaman and Nicobar in India. A yearly average *CF* and *PR* range of (13.73–14.61%) and (64.70–64.93%), respectively were obtained. Other studies, such as [25–27] investigated power plants with capacities ranging from 1.72–5 kWp and determined their performance parameters.

RE development and application in the Russian Federation are relatively lower than in other European countries. The country has mainly relied on fossil fuels and nuclear energy for its electricity and heating demands. Hydropower constitutes a large portion of its installed RE capacity in terms of renewables. The country's total installed RE capacity increased to 53.5 GW; hydro alone constitutes 51.5 GW, with bioenergy taking only 1.35 GW. The solar PV has only 460 MW, while onshore wind also has 111 MW as of 2015 [28]. The country's share of installed solar and wind energy capacities as of 2021 are 0.7 percent and 0.42 percent, respectively [29].

The government of the Russian Federation has therefore committed to the development of the entire RE sector. The government's target is to achieve a renewable energy share of 4.5% by 2030. The country's Energy Strategy stipulates that the percentage of RE in the energy mix must be at least 4.5% in the period from 2020–2030, producing 80–100 billion kWh/year [30,31]. However, there is a lack of detailed information on the performance of solar PV modules under Russian weather conditions for all year round. This does not help encourage individuals and private investors to use or invest in the sector. The objective of this study is to bridge this gap by assessing solar PV's technical performance and social aspect in the Sverdlovsk region of Russia. The current study predicts solar PV system's energy performance, degradation, and energy loss under different orientations, i.e., fixed tilted plane and tracking horizontal axis E–W using the PVsyst simulation software. This is the first time such a study has been conducted in the study area in the Russian Federation, to the best of our knowledge. Unlike the previously reviewed literature, this study assesses the employment potential of such systems in Russia. The results obtained in this study are expected to shape the technical and social aspects of solar PV energy in the country.

This work is organized into four sections; Section 2 covers the materials and method used for the study. The outcomes are obtainable in Section 3, the conclusion and future research recommendations are obtainable in Section 4.

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

The methodology adopted for the study, specifications of the PV modules studied, geographical information of the study area, and mathematical relations used for the evaluation of the various performance indicators are presented in this section.

### *2.1. PVsyst Simulation Model*

The PVsyst simulation tool is a commonly used software in designing solar power plants optimally and assessing the energy yields of the plants. It uses meteorological irradiation resources, extensive knowledge of PV technology, and PV system components

for the simulation. As a result, the PVsyst tool can assist researchers and engineers to comprehend the PV system's workings to improve the system's design. The proposed grid-connected PV system was simulated using the following steps [1]: prehend the PV system's workings to improve the system's design. The proposed gridconnected PV system was simulated using the following steps [1]: • Identification of the study area (location)

The PVsyst simulation tool is a commonly used software in designing solar power plants optimally and assessing the energy yields of the plants. It uses meteorological irradiation resources, extensive knowledge of PV technology, and PV system components for the simulation. As a result, the PVsyst tool can assist researchers and engineers to com-

• Identification of the study area (location) • Downloading of the weather data characteristics for the study area (i.e., solar irradi-

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 4 of 21

*2.1. PVsyst Simulation Model*

	- The discretionary choice to alter the values for the loss types.

### *2.2. Meteorological Data of the Study Area 2.2. Meteorological Data of the Study Area* The study area is Yekaterinburg, which is in the Sverdlovsk region in Russia. It is

The study area is Yekaterinburg, which is in the Sverdlovsk region in Russia. It is located on latitude 56◦50034.400 N and longitude 60◦39019.000 E. The weather data used for the analysis were obtained from Meteonorm 8.0. The study area has an average global horizontal irradiation (GHI) average of 2.82 kWh/m2/day and horizontal diffuse irradiation of 1.37 kWh/m2/day. The average annual temperature for the area is 2.6 °C, with an average wind speed of 3.7 m/s. The relative humidity of the study area is also 74.1%, with a global horizontal irradiation year-to-year variability of 3.7%. The weather characteristics of the study area are presented in Figure 1. The lowest temperature and solar irradiation usually occur during the winter; similarly, May–August record relatively high temperatures and insolation. The month of July generally receives the highest solar irradiation and temperatures. located on latitude 56°50′34.4″ N and longitude 60°39′19.0″ E. The weather data used for the analysis were obtained from Meteonorm 8.0. The study area has an average global horizontal irradiation (GHI) average of 2.82 kWh/m<sup>2</sup> /day and horizontal diffuse irradiation of 1.37 kWh/m<sup>2</sup> /day. The average annual temperature for the area is 2.6 ℃, with an average wind speed of 3.7 m/s. The relative humidity of the study area is also 74.1%, with a global horizontal irradiation year-to-year variability of 3.7%. The weather characteristics of the study area are presented in Figure 1. The lowest temperature and solar irradiation usually occur during the winter; similarly, May–August record relatively high temperatures and insolation. The month of July generally receives the highest solar irradiation and temperatures.

**Figure 1.** (**a**) GHI and temperature (**b**) wind speed characteristics of the study area (Meteonorm 8.0). **Figure 1.** (**a**) GHI and temperature (**b**) wind speed characteristics of the study area (Meteonorm 8.0).

*2.3. Solar PV Performance Assessment*

The system yields are classified into an array, a reference, and final yields. The yields show the actual array operations in relation to its rated capacity. The array yield *Y<sup>A</sup>* is said to be the output energy *DC* (direct current) produced from the *PV* array within a certain time frame normalized by the *PV* system's rated power [32]. It can be represented mathematically as indicated in Equation (1) [33].

$$Y\_A = \frac{E\_{D\mathbb{C}}}{P\_{PV,\,\,rated}} \left( \mathbf{kWh} / \!/\!/\!\_{\text{kW}\_P} \right) \tag{1}$$

where *EDC* is the *PV* array's *DC* energy output in (kWh), and *PPV*, *rated* is the *PV* system rated power in (kWp).

The alternating current (*AC*) energy (total) produced by the *PV* module for a specific time is divided by the installed *PV*'s system rated output power [32]. The final yield *Y<sup>F</sup>* can be defined as the inverter side output, which is *AC* energy produced daily *YF*, *<sup>d</sup>* or monthly *YF*, *<sup>m</sup>* by the system, which is normalized by its nominal or rated power of the installed *PV* array. It can be calculated using Equation (2) [34].

$$Y\_F = \frac{E\_{AC}}{P\_{PV,\text{rated}}} \tag{2}$$

where the *AC* energy output (kWh) is denoted by *EAC*.

The reference yield *Y<sup>R</sup>* describes the solar irradiance for the *PV* system; it is the total in-plane solar irradiance divided by the reference irradiance at standard test conditions (STCs). It is a function of weather conditions, array orientation, and location, as indicated in Equation (3) [9,33].

$$\mathcal{Y}\_R = \frac{G\_I}{G\_{STC}} \text{ (kWh/kW)}\tag{3}$$

where the total in-plane solar irradiance is denoted with *G<sup>I</sup>* (kWh/m<sup>2</sup> ), and the reference irradiance at STCs is represented by *GSTC* (1 kW/m<sup>2</sup> ).

The *PR* of a solar *PV* system measures the total outcome of losses on the *PV* system's rated output. The *PV* system's *PR* shows how close its performance approaches the ideal performance under real-life operations; it helps compare a *PV* system's independence of a location, orientation, tilt angle, and nominal rated power capacity. It can be calculated using Equation (4) [32,35].

$$PR = \frac{100 \times Y\_F}{Y\_R} \text{(\%)}\tag{4}$$

*CF* can be defined as the ratio of the *AC* energy that is generated by the *PV* system within a specified period of time (mostly one year) to the system's output energy, which would have been produced if the power plant were to operate at full capacity for the entire period. The yearly *CF* can calculated using Equation (5) [6].

$$\text{CF} = \frac{E\_{A\text{C}}}{P\_{PV,\text{rated}} \times 8760} \tag{5}$$

The array capture losses *L<sup>A</sup>* shows the losses occasioned by the array's operation, which highlights the array's failure to use the available irradiance [26] completely. The difference between the array yield and the reference yield is the array capture losses; this can be calculated using Equations (6) and (7) [32].

$$L\_A = Y\_R - Y\_A \text{ (kWh/kW}\_{\text{P}}\text{)}\tag{6}$$

where the losses *L<sup>S</sup>* are caused by losses in changing the inverter's *DC* output power from the *PV* system to *AC* power. This is mathematically represented as follows:

$$L\_S = Y\_A - Y\_F \left(\text{kWh/kW}\_P\right) \tag{7}$$

The thermal losses result from the effect of temperature on the performances of the *PV* modules. This can be expressed mathematically as indicated in Equation (8) [36].

$$E\_{therm} = E\_{PV} \left( 1 - \frac{1}{1 - \gamma (T\_{\odot} - T\_0)} \right) \text{(MWh)} \tag{8}$$

where the temperature coefficient of the maximal power is denoted by *γ*, the module temperature under STCs (25 ◦C) is represented by *T*0, and the module temperature is denoted by *TC*.

### *2.4. Energy Production*

The daily, monthly, and annually total energy (*AC* or *DC*) produced by the *PV* system can be attained through simple summations as shown in Equation (9) [37,38].

$$\begin{aligned} E\_d &= \sum\_{\substack{1 \\ n \\ E\_m = \sum\_{1} E\_d \text{ (kWh)}}}^{24} \\ E\_y &= \sum\_{1}^{12} E\_m \text{ (kWh)} \end{aligned} \tag{9}$$

where the number of days in a month is represented by *n*, and the hourly energy is denoted by *E<sup>h</sup>* . In addition, the daily, monthly, and yearly cumulative energy values are represented with *E<sup>d</sup>* , *Em*, and *Ey*, respectively.

### *2.5. Characteristics of the Various Components*

The *PV* module technology used for the analysis was Si-mono. The AE 300DGM6- 60 (1500) solar module manufactured by AE solar was used for the analysis. A total of 120 units were used to achieve the designed power output. The technical description of the *PV* module is presented in Table 1. Increasing the temperature of a *PV* cell decreases its output performance as a result of the increase in the rate of the internal recombination in the *PV* cell, which is caused by increased carrier concentrations. Both the electrical efficiency and the power output of the *PV* power plant relate linearly with its operating temperature. As a result, when the operating temperature of the *PV* module rises beyond 25 ◦C, it leads to a reduction in the semiconductor material's band-gap, which results in the reduction in the open circuit voltage [39,40]. The current-voltage (I–V) graph for the selected *PV* module for different cell temperatures is presented in Figure 2; in this graph, the negative effect of high temperatures on the performance of *PV* cells is clearly shown. The power–voltage graph, which also demonstrates the impact of solar irradiance on the output of the *PV* cell, is also represented in Figure 3.

**Table 1.** *PV* module system description.


**Figure 2.** I-V graph for the used *PV* module under different temperature conditions (Obtained from *PV*syst software). **Figure 2.** I-V graph for the used *PV* module under different temperature conditions (Obtained from *PV*syst software).

**Figure 3.** Power–voltage characteristics under different incident irradiance (Obtained from *PV*syst software). **Figure 3.** Power–voltage characteristics under different incident irradiance (Obtained from *PV*syst software).

In the inverter case, the Voltwerk model was used for the analysis; the model is 11 kW 400–800 V TL 50 Hz VS11 with a minimum MPP voltage of 400 V and a maximum MPP voltage of 800 V. It has a maximum efficiency of 98%. A total of 3 inverters were used to design the power plant. to design the power plant. In the inverter case, the Voltwerk model was used for the analysis; the model is 11 kW 400–800 V TL 50 Hz VS11 with a minimum MPP voltage of 400 V and a maximum MPP voltage of 800 V. It has a maximum efficiency of 98%. A total of 3 inverters were used to design the power plant.

Figure 4 depicts the schematic for the proposed *PV* system; it is made up of the *PV* array section where the *PV* modules are connected, the inverter section where the conversion of *DC* to *AC* is done, the load section, and the grid section. In the scheme, the denotes the *PV* array's energy output, while the inverter's energy output is represented by . denotes the energy used by the user load. Figure 4 depicts the schematic for the proposed *PV* system; it is made up of the *PV* array section where the *PV* modules are connected, the inverter section where the conversion of *DC* to *AC* is done, the load section, and the grid section. In the scheme, the *EArray* denotes the *PV* array's energy output, while the inverter's energy output is represented by *Eout inv*. *Eused* denotes the energy used by the user load.

**Figure 4.** Schematic diagram of the proposed *PV* system. **Figure 4.** Schematic diagram of the proposed *PV* system.

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 8 of 21

#### *2.6. Field Mechanisms Used in the Analysis 2.6. Field Mechanisms Used in the Analysis* **Figure 4.** Schematic diagram of the proposed *PV* system.

To take maximum advantage of the irradiation from the sun, *PV* panels are mostly tilted in the direction of the equator at an optimal tilt angle. The optimum tracking angle at a location is determined by both the latitude and the climatic conditions at the said location. There is the option to use the fixed tilted plane, which is commonly used; however, there is an option to use trackers. The single solar tracking and double-axis tracking option can be employed; this is subject to the degree of the freedom movement. The single solar tracking uses a single pivot point for rotation to track the sun's path from one point to another. In the case of the double-axis tracking, it tracks the sun's path in two different axes using two pivot points for rotation; it has both the horizontal and vertical axes [41,42]. In this study, two different field types were compared to assess their effect on the performance of the *PV* power plant at the study site. These are the fixed tilted plane and the tracking horizontal axis E–W. The optimum tilt angle for the fixed *PV* system is 45° for the study area. The two different mechanisms are represented in Figure 5. To take maximum advantage of the irradiation from the sun, *PV* panels are mostly tilted in the direction of the equator at an optimal tilt angle. The optimum tracking angle at a location is determined by both the latitude and the climatic conditions at the said location. There is the option to use the fixed tilted plane, which is commonly used; however, there is an option to use trackers. The single solar tracking and double-axis tracking option can be employed; this is subject to the degree of the freedom movement. The single solar tracking uses a single pivot point for rotation to track the sun's path from one point to another. In the case of the double-axis tracking, it tracks the sun's path in two different axes using two pivot points for rotation; it has both the horizontal and vertical axes [41,42]. In this study, two different field types were compared to assess their effect on the performance of the *PV* power plant at the study site. These are the fixed tilted plane and the tracking horizontal axis E–W. The optimum tilt angle for the fixed *PV* system is 45◦ for the study area. The two different mechanisms are represented in Figure 5. *2.6. Field Mechanisms Used in the Analysis* To take maximum advantage of the irradiation from the sun, *PV* panels are mostly tilted in the direction of the equator at an optimal tilt angle. The optimum tracking angle at a location is determined by both the latitude and the climatic conditions at the said location. There is the option to use the fixed tilted plane, which is commonly used; however, there is an option to use trackers. The single solar tracking and double-axis tracking option can be employed; this is subject to the degree of the freedom movement. The single solar tracking uses a single pivot point for rotation to track the sun's path from one point to another. In the case of the double-axis tracking, it tracks the sun's path in two different axes using two pivot points for rotation; it has both the horizontal and vertical axes [41,42]. In this study, two different field types were compared to assess their effect on the performance of the *PV* power plant at the study site. These are the fixed tilted plane and the tracking horizontal axis E–W. The optimum tilt angle for the fixed *PV* system is 45° for the study area. The two different mechanisms are represented in Figure 5.

**Figure 5.** The orientation used, (**a**) tracking horizontal axis E–W and (**b**) fixed tilted plane. **Figure 5.** The orientation used, (**a**) tracking horizontal axis E–W and (**b**) fixed tilted plane.

### **3. Results and Discussion**

According to a summary of the results from the simulations, the fixed tilted *PV* system will produce an energy of 39.5 MWh/year with an estimated specific production

of 1097 kWh/kWp/year and a *PR* of 82.3%, as shown in Figure 6. Similarly, the tracking horizontal axis E–W system will generate a total of 43.0 MWh/year, which is about 3.5 MWh/year more than that of the fixed tilted plane; it is also expected to record a specific production of 1194 kWh/kWp/year with a *PR* of 82.6%. As widely published in the literature, the performance of *PV* systems depends on several factors; some of these include the ambient temperature, clearness index, and the level of solar irradiation, among others. As reviewed in the introduction section of this work, the obtained *PR* in this study falls within the range of values obtained by studies, such as Malvoni et al. [8] under Mediterranean weather conditions, which obtained 84.4%. Eke and Demircan [43] obtained a *PR* of 72% under Turkey climatic conditions; similarly, Okello et al. [44] obtained 84.3% as the *PR* for a 3.2 kWp grid-connected *PV* system in South Africa. As presented earlier in this section, the slight differences in the *PR* values can be attributed to the factors that affect the *PV* system's performance. The high temperatures during the summer period affected the *PR* during those periods. The output performance of the *PV* decreases with an increase in *PV* temperature [39,45]. Hence the output performance of the system reduces to some level even if there is enough solar radiation. A malfunction in the *PV* system can also be detected based on the *PR* values. Months with lower *PR* can be ascribed to a malfunction in the inverter and the incorrect functioning of the system. The IEC norm describes the normalized production and represents the standardized parameter for the *PV* system's performance assessment. It can therefore be assessed to compare the characteristics of *PV* architectures that are constructed under similar climatic conditions [46]. The useful produced energy per installed kWp/day, system losses, and the collection losses for both orientations are estimated and presented in Figure 7. The arrays' temperature characteristics against effective irradiance are presented in Figure 8. It can be seen from the figure that the array's temperature ranged between −30 ◦C during the winter period to as high as about 65 ◦C in the summer period. tem will produce an energy of 39.5 MWh/year with an estimated specific production of 1097 kWh/kWp/year and a *PR* of 82.3%, as shown in Figure 6. Similarly, the tracking horizontal axis E–W system will generate a total of 43.0 MWh/year, which is about 3.5 MWh/year more than that of the fixed tilted plane; it is also expected to record a specific production of 1194 kWh/kWp/year with a *PR* of 82.6%. As widely published in the literature, the performance of *PV* systems depends on several factors; some of these include the ambient temperature, clearness index, and the level of solar irradiation, among others. As reviewed in the introduction section of this work, the obtained *PR* in this study falls within the range of values obtained by studies, such as Malvoni et al. [8] under Mediterranean weather conditions, which obtained 84.4%. Eke and Demircan [43] obtained a *PR* of 72% under Turkey climatic conditions; similarly, Okello et al. [44] obtained 84.3% as the *PR* for a 3.2 kWp grid-connected *PV* system in South Africa. As presented earlier in this section, the slight differences in the *PR* values can be attributed to the factors that affect the *PV* system's performance. The high temperatures during the summer period affected the *PR* during those periods. The output performance of the *PV* decreases with an increase in *PV* temperature [39,45]. Hence the output performance of the system reduces to some level even if there is enough solar radiation. A malfunction in the *PV* system can also be detected based on the *PR* values. Months with lower *PR* can be ascribed to a malfunction in the inverter and the incorrect functioning of the system. The IEC norm describes the normalized production and represents the standardized parameter for the *PV* system's performance assessment. It can therefore be assessed to compare the characteristics of *PV* architectures that are constructed under similar climatic conditions [46]. The useful produced energy per installed kWp/day, system losses, and the collection losses for both orientations are estimated and presented in Figure 7. The arrays' temperature characteristics against effective irradiance are presented in Figure 8. It can be seen from the figure that the array's temperature ranged between −30 °C during the winter period to as high as about 65 °C in the summer period.

According to a summary of the results from the simulations, the fixed tilted *PV* sys-

*Sustainability* **2022**, *14*, x FOR PEER REVIEW 9 of 21

**3. Results and Discussion**

**Figure 6. Figure 6.**  Monthly performance ratio for the ( Monthly performance ratio for the (**aa**) fixed plane and (**b**) tracking *PV* system. ) fixed plane and (**b**) tracking *PV* system.

107

(**a**)

tem.

(**a**) (**b**)

**Figure 7.** Normalized energy production per installed kWp for (**a**) fixed plane (**b**) tracking *PV* sys-

**Figure 8.** Array temperature vs. effective irradiance (**a**) fixed plane and (**b**) tracking *PV* system. **Figure 8.** Array temperature vs. effective irradiance (**a**) fixed plane and (**b**) tracking *PV* system.

The expected monthly energy injected into the grid from the *PV* system for both orientations is shown in Figure 9. It is evident that the most energy injected into the grid occurs during the summer period, obviously because of the longer period and high intensity of solar radiation during that time. The balances and main results are presented in Table 2; it includes the global horizontal, horizontal diffuse irradiance, effective global corresponding for shading (GlobEff), ambient temperature (T\_amb), global incident in collector plane (GlobInc), the energy output of the *PV* array (EArray), performance ratio The expected monthly energy injected into the grid from the *PV* system for both orientations is shown in Figure 9. It is evident that the most energy injected into the grid occurs during the summer period, obviously because of the longer period and high intensity of solar radiation during that time. The balances and main results are presented in Table 2; it includes the global horizontal, horizontal diffuse irradiance, effective global corresponding for shading (GlobEff), ambient temperature (T\_amb), global incident in collector plane (GlobInc), the energy output of the *PV* array (EArray), performance ratio

(*PR*), and the energy injected into the grid (E\_Grid for each month of the year as well as the overall energy out for the year. Although the highest solar irradiation at the study area occurs during the summer periods, it is clear from the Table that the *PR* at those periods

temperatures during those areas, which negatively affect the efficiency of the solar cells. It is clear from the energy values presented in Table 1 that both designs lost energy in the course of the conversion as there is a difference between the *DC* and *AC* sides of the energy produced. This difference is the energy lost due to the losses in the system. The fixed tilted

plane system lost 1.249 MWh against 1.277 MW for the tracking system.

**Global Horizontal, kWh/m<sup>2</sup>**

**Diffuse Horizonta l, kWh/m<sup>2</sup>**

(*PR*), and the energy injected into the grid (E\_Grid for each month of the year as well as the overall energy out for the year. Although the highest solar irradiation at the study area occurs during the summer periods, it is clear from the Table that the *PR* at those periods is relatively lower compared to the other periods. This may be due to the relatively high temperatures during those areas, which negatively affect the efficiency of the solar cells. It is clear from the energy values presented in Table 1 that both designs lost energy in the course of the conversion as there is a difference between the *DC* and *AC* sides of the energy produced. This difference is the energy lost due to the losses in the system. The fixed tilted plane system lost 1.249 MWh against 1.277 MW for the tracking system. *Sustainability* **2022**, *14*, x FOR PEER REVIEW 12 of 22

**Figure 9.** System output power distribution (**a**) fixed plane and (**b**) tracking system. **Figure 9.** System output power distribution (**a**) fixed plane and (**b**) tracking system.

**Energy Array,** 

**Fixed Tracking Fixed Tracking Fixed Tracking Fixed Tracking Tracking Fixed**

**MWh E\_Grid, MWh** *PR*

Jan 19.9 10.17 −14.57 58.0 68.5 60.9 71.3 2.087 2.446 2.032 2.384 0.929 0.926 Feb 42.8 16.23 −13.04 96.2 106.6 101.0 111.3 3.394 3.734 3.310 3.642 0.909 0.911 Mar 90.8 33.36 −4.61 144.6 148.5 152.4 156.4 4.880 4.994 4.754 4.865 0.864 0.867 Apr 117.8 52.17 3.92 137.9 141.2 146.1 149.9 4.467 4.579 4.333 4.446 0.824 0.824 May 152.6 77.64 12.05 147.5 162.6 156.4 172.6 4.634 5.136 4.489 4.991 0.803 0.797 Jun 168.1 76.44 16.66 154.5 178.0 163.9 188.6 4.725 5.481 4.576 5.329 0.785 0.776 Jul 163.9 78.29 19.14 154.9 174.1 164.3 184.7 4.680 5.291 4.531 5.140 0.773 0.766 Aug 125.8 69.21 16.66 130.7 137.1 138.6 145.6 4.014 4.222 3.885 4.093 0.781 0.779

**GlobInc,** 

**Table 2.** Balances and main results.

**kWh/m<sup>2</sup>**

**T\_amb, °C Global Eff,** 


**Table 2.** Balances and main results.

The *PV*syst also runs a probability distribution analysis for the total yearly energy produced from the system, which could be transferred into the grid system. The probability distribution variance for the plant's production forecast depends on several factors; some of these include inverter efficiency uncertainty, *PV* module modeling/parameters, meteo data, degradation uncertainty, and soiling and mismatch uncertainties [12]. The probability law supposes that during the many years of operation of the *PV* system, the annual yield distribution will follow a statistical law, and this law is assumed to be the normal or Gaussian distribution. The P50-P90 indicates the different levels of yield, for which the probability that a particular year's production is over this value of 50% and 90%, respectively [47]. The probability distribution function for the *PV* plant's energy generation forecast is as shown in Figure 10. According to the results from the simulations, the expected annual production probability for the fixed *PV* module for the P50, P75, and P90 is 39.68 MWh, 37.72 MWh, and 35.94 MWh, respectively, with a variability of 2.91 MWh. In the case of the tracking *PV* module, the annual production probability for the P50, P75, and P90 is 43.18 MWh, 41.05 MWh, and 39.12 MWh, respectively, with a variability of 3.17 MWh.

### *3.1. System Losses*

These are losses in the system that may occur during the conversion of the incident solar energy to electric energy. These losses enable users to know the energy converted into electricity by the system, which can be done by subtracting the total loss from the incident energy on the panel. The decrease in the performance of the *PV* module can be associated with these losses. The normalized production and loss factors for both orientations are presented in Figure 11. The *PV* array losses for both orientations, i.e., fixed and tracking, are 15.1% and 14.9%, respectively. The months of May to August recorded the highest array losses due to the high temperatures and relatively low wind speed during those periods. However, the tracking system positively impacted the *PV* module, as its system losses were relatively less than the fixed *PV* module. The produced useful energy (inverter output) for the fixed and tracked systems are 82.3% and 82.6%, respectively.

bility of 3.17 MWh.

**Figure 10.** Probability distribution for (**a**) fixed and (**b**) tracking PV modules. **Figure 10.** Probability distribution for (**a**) fixed and (**b**) tracking *PV* modules.

*3.1. System Losses* These are losses in the system that may occur during the conversion of the incident solar energy to electric energy. These losses enable users to know the energy converted into electricity by the system, which can be done by subtracting the total loss from the incident energy on the panel. The decrease in the performance of the PV module can be associated with these losses. The normalized production and loss factors for both orientations are presented in Figure 11. The PV array losses for both orientations, i.e., fixed and tracking, are 15.1% and 14.9%, respectively. The months of May to August recorded the highest array losses due to the high temperatures and relatively low wind speed during those periods. However, the tracking system positively impacted the PV module, as its system losses were relatively less than the fixed PV module. The produced useful energy (inverter output) for the fixed and tracked systems are 82.3% and 82.6%, respectively. The loss diagram for the two orientations is presented in Figure 12. The losses range from the incidence angle modifier (IAM), which is the reflection loss (optical effect) that corresponds to the weakening of the irradiation that is actually reaching the surface of the *PV* cell with respect to the irradiation under normal incidence [48]. Others include manufacture losses, ambient temperature, and ohmic wiring, etc. Soiling loss involves dirt or dust accumulation on the surface of the module, which produces a dimming effect on the incident solar irradiation. The results suggest that both orientations are expected to record a soiling loss of 3%. The soiling loss can be mitigated by periodically washing the surface of the *PV* module; it can, however, add to the operations and maintenance costs, which can affect the plant's economic viability. This is especially for sites with water scarcity. The IAM loss is 2.5% for both designs. The nominal array energy (at *STC* efficiency) is 45.72 MWh for the fixed module; it, however, increased significantly by some 3.92 MWh for the tracking system.

(**b**)

These are losses in the system that may occur during the conversion of the incident solar energy to electric energy. These losses enable users to know the energy converted into electricity by the system, which can be done by subtracting the total loss from the incident energy on the panel. The decrease in the performance of the *PV* module can be associated with these losses. The normalized production and loss factors for both orientations are presented in Figure 11. The *PV* array losses for both orientations, i.e., fixed and tracking, are 15.1% and 14.9%, respectively. The months of May to August recorded the highest array losses due to the high temperatures and relatively low wind speed during those periods. However, the tracking system positively impacted the *PV* module, as its system losses were relatively less than the fixed *PV* module. The produced useful energy (inverter output) for the fixed and tracked systems are 82.3% and 82.6%, respectively.

**Figure 10.** Probability distribution for (**a**) fixed and (**b**) tracking *PV* modules.

*3.1. System Losses*

**Figure 11.** Normalized production and loss factors (**a**) fixed and (**b**) tracking systems. **Figure 11.** Normalized production and loss factors (**a**) fixed and (**b**) tracking systems.

The loss diagram for the two orientations is presented in Figure 12. The losses range from the incidence angle modifier (IAM), which is the reflection loss (optical effect) that The module quality loss, the deviation between the nominal capacity indicated on the manufacturer's datasheet, and the real module capacity, signifies the loss of module quality. A module quality loss of 0.4% was recorded. The *PV* array mismatch loss for both designs is the same; they both recorded 3.66%. This power loss is known as electrical mismatch loss. According to a study by Koirala et al. [49], a mismatch loss of up to 12% in the series string may arise but can be reduced to between 0.4–2.4% using suitable series-parallel connections. Presorting according to the max power current is identified as the most effective method for optimizing *PV* array performance [50,51]. The module degradation loss for the two designs is the same for all; they all recorded 3.82%. A total of 1028 kWh/m<sup>2</sup> global horizontal solar irradiation was received during the analysis period.

The two *PV* system's predicted performances were compared with other literature to assess their performance. Results from the other works on other countries presented in Table 3 can be said to be relatively similar to what is presented in this study. The *PR* falls within the range obtained by most studies.


**Table 3.** Comparison with other studies.

### *3.2. Social Aspect*

Deployment of energy systems at any location is usually accompanied by job creation, as people will be needed from the construction stages to the operation and decommissioning phase [65]. As a result, the study considered the social aspect of the energy systems to assess their employment creation potentials for the study area. The estimated employment potentials for solar *PV* is estimated to be about 0.27549 <sup>×</sup> (10−7/kWh/year) [65,66]. According to the simulated results, the total energy exported to the grid per year for the fixed axis solar *PV* module is 39,480 kWh against 42,970 kWh for the tracking axis module. According to the mathematical computations, we obtained 0.0011 persons/year for the fixed *PV* module, while 0.0012 persons/year were obtained for the tracking system. Therefore, assuming both modules operated for a lifetime of 25 years, then the fixed module will have an employment potential of 0.028 persons against 0.030 persons for the tracking *PV* module. The employment potential between the two designs does not vary much. It is important to state that this analysis is only meant to assess the employment potential of the 36 kWp capacity *PV* modeled. Hence, it indicates the employment potential of a large-scale solar *PV* power plant under Russian weather conditions.

corresponds to the weakening of the irradiation that is actually reaching the surface of the *PV* cell with respect to the irradiation under normal incidence [48]. Others include manufacture losses, ambient temperature, and ohmic wiring, etc. Soiling loss involves dirt or dust accumulation on the surface of the module, which produces a dimming effect on the incident solar irradiation. The results suggest that both orientations are expected to record a soiling loss of 3%. The soiling loss can be mitigated by periodically washing the surface of the *PV* module; it can, however, add to the operations and maintenance costs, which can affect the plant's economic viability. This is especially for sites with water scarcity. The IAM loss is 2.5% for both designs. The nominal array energy (at *STC* efficiency) is 45.72 MWh for the fixed module; it, however, increased significantly by some 3.92 MWh

**Figure 12.** *Cont*.

for the tracking system.

(**b**)

**Figure 12.** Loss diagram for (**a**) fixed and (**b**) tracking systems. **Figure 12.** Loss diagram for (**a**) fixed and (**b**) tracking systems.

### The module quality loss, the deviation between the nominal capacity indicated on **4. Conclusions**

the manufacturer's datasheet, and the real module capacity, signifies the loss of module quality. A module quality loss of 0.4% was recorded. The *PV* array mismatch loss for both designs is the same; they both recorded 3.66%. This power loss is known as electrical mismatch loss. According to a study by Koirala et al. [49], a mismatch loss of up to 12% in the series string may arise but can be reduced to between 0.4–2.4% using suitable series-par-Russian weather conditions are considered harsh for large-scale solar power plants, especially due to the high negative temperatures during its long winter. Therefore, this study simulated two different designs (i.e., fixed tilted plane and tracking, horizontal axis E–W) to assess their performance, energy loss, and employment potential for potential largescale solar *PV* development in the Sverdlovsk region of Russia. The following conclusions are made from the study:


The two *PV* system's predicted performances were compared with other literature to assess their performance. Results from the other works on other countries presented in Table 3 can be said to be relatively similar to what is presented in this study. The *PR* falls to August are the periods within which much of the electricity will be generated due to the high solar irradiations recorded during those periods.


Russia is currently implementing measures that seek to promote, develop, and use its various RE resources to help cut down its GHG emissions. Therefore, this study is expected to serve as a reference material for government, interested parties, individuals, and policymakers in relation to small and large-scale solar *PV* development, using the performance of the two designs. The data provided in this study give useful information on the possible net energy output of such systems. Future studies can assess the economic viability of such projects on large-scale levels and possibly integrate them with other renewable energy resources, such as wind, to assess their viability for rural and far-to-reach areas in Russia. An environmental impact assessment can also be assessed for large-scale *PV* projects in the country to know the potential reduction in GHG emissions that such projects come with. Similarly, future studies can conduct experimental research to compare actual results and the simulated results to give a real understanding of the performance of *PV* modules under Russian weather conditions. This is because it is highly possible that the simulated results may differ from that conducted under real environmental conditions. It is also recommended to evaluate the performance of different *PV* technologies under Russian weather conditions. This would provide critical information on the optimum technology for Russian weather.

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

**Funding:** This research was funded by Taif University Researchers Supporting Project number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data used, and their sources are provided in the text.

**Acknowledgments:** The authors would like to thank Cardiff University/School of Engineering for accepting to pay the APC toward publishing this paper. In addition, the authors would like to acknowledge the financial support received from Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

## **Abbreviations**


## **References**


## *Article* **Bidirectional Interface Resonant Converter for Wide Voltage Range Storage Applications**

**Mouncif Arazi , Alireza Payman, Mamadou Baïlo Camara and Brayima Dakyo \***

Electrical Engineering, Faculty of Sciences and Technology, University Le Havre Normandie, 76600 Le Havre, France; moncef.arazi.10@gmail.com (M.A.); paymana@univ-lehavre.fr (A.P.); mamadou-bailo.camara@univ-lehavre.fr (M.B.C.)

**\*** Correspondence: brayima.dakyo@univ-lehavre.fr

**Abstract:** In this paper, a bidirectional zero voltage switching (ZVS) resonant converter with narrow control frequency deviation is proposed. Wide input–output voltage range applications, such as flywheel or supercapacitors storage units are targeted. Due to symmetrical topology of resonant circuit interfaces, the proposed converter has similar behavior in bidirectional operating mode. We call it Dual Active Bridge Converter (DABC). The proposal topology of the converter is subjected to multi resonant circuits which make it necessary to study with multiscale approaches. Thus, first harmonic approximation and use of selective per unit parameters are established in (2) Methods. Then, the forward direction and backward direction of power flux exchange are detailed according to switching sequences. Switching frequency control must be completed within a narrow range. So, the frequency range deterministic parameters are emphasized in the design procedure in (3) Methods. A narrow range of switching frequency and a wide range voltage control must be ensured to suit for energy storage units, power electronic devices capabilities and electromagnetic compatibility. A 3 kW test bench is used to validate operation principles and to proof success of the developed design procedure. The interest of proposed converter is compared to other solutions from the literature in (4) Results.

**Keywords:** bidirectional resonant converter; zero voltage switching; zero current switching; wide input voltage range; power losses

### **1. Introduction**

In the last decade, the converters used in electric vehicles, smart grids and renewable energies applications have had significant progress in term of electric performances [1,2]. In numerous cases, the performances are obtained by means of the integration of Energy Storage Systems (ESS) such as batteries, supercapacitors or flywheels in DC-bus using bidirectional DC/DC converters. The converters ensure the ESS charge and discharge operations according to the power balance [3]. Different topologies of bidirectional DC/DC converters are proposed in the literature [4–6]. Non-isolated bidirectional topologies were suggested to interface batteries and supercapacitors as described in [7]. Indeed, many specific systems require galvanic isolation especially for safety reasons. The operation with high switching frequencies allows reducing the size of the passive components [8]. Use of hard switching in turn off operations unfortunately increases the converter losses. To overcome this problem, the soft switching technique must be applied, which increases the efficiency of the converter [9]. Phase-shifted Dual Active Bridge (DAB) converter has been widely investigated in the literature [10,11]. In [12,13], the authors propose a phase shift control strategy combined with duty cycle control to extend the soft switching capability of the DAB converter.

The control strategy is based on the synchronized phase-shift control with dutycycle one. To improve the efficiency of the DAB converter, a resonant circuit is added

**Citation:** Arazi, M.; Payman, A.; Camara, M.B.; Dakyo, B. Bidirectional Interface Resonant Converter for Wide Voltage Range Storage Applications. *Sustainability* **2022**, *14*, 377. https://doi.org/10.3390/ su14010377

Academic Editor: Antonio Caggiano

Received: 16 November 2021 Accepted: 27 December 2021 Published: 30 December 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/).

to the basic topology to obtain a resonant converter [14,15]. Using this technique, the soft switching zone can be significantly extended. The series resonant circuit can ensure Zero Voltage Switching (ZVS) capability of the converter, but it can be subjected to fail for light load or no-load operating conditions [16,17]. Using parallel resonant circuits allows no-load operation, but this would come at the expense of a high resonant current almost independent of the load. Thus, exchanged internal energy and the conduction losses are unnecessarily large [18]. The combination of series and parallel circuits, such as LCC resonant converter [19,20] is expected to offer a converter with better characteristics. This topology can operate from rated power to no-load with small internal energy circulating. However, it causes high switching losses for wide voltage range applications. Among the different resonant DC/DC converters, LLC resonant converter has attracted the attention of researchers [21–24]. This topology can achieve the soft switching for both sides of the converter, and can operate in buck or boost mode. However, it is still a classic series resonant converter in backward mode because the inductances of transformer do not participate in the resonant operation. This reduces significantly the efficiency in the backward mode and penalizes LLC topology for bidirectional current applications [25], such as interfacing batteries and supercapacitors. CLLC resonant converters are proposed in [26–28]. These converters suffer from high reactive power when the switching frequency deviates from the resonant frequency during low load conditions. Modified LLC resonant converters with hybrid control are proposed in [29,30] to reduce the reactive power. However, they are still not suitable for wide voltage range systems and suffer due to high power losses in power semiconductors switch-off operations. A qualitative comparison between the results achieved by some bidirectional resonant converters in literature is presented in Table 1, and related behaviors are described in [31–33].

**Table 1.** Comparison of bidirectional resonant converters for wide voltage range applications.


The proposed topology aims to obtain the followings performs compared to referenced solutions in Table 1:


The paper is organized as follow: Section 2 gives the topology, the characteristic and operating principle of the novel converter. Modeling and global sizing based on the first harmonic approach is given. The details of operating sequences and per-unit variables definition are given to help for relevant analysis. The design procedure and soft switching (ZVS) performances of the converter are presented in Section 3. The experimental test bench developed in laboratory and results are presented and discussed in Section 4 to show the feasibility of the proposed solution. Concluding remarks are given in Section 5.

### **2. Operating Principle and Main Characteristics of the Proposed Novel Converter**

The topology of the proposed resonant converter is presented in Figure 1. It is composed of two active full bridges for bidirectional operations, a high frequency transformer and a symmetric resonant circuit. This last one includes two series inductances (*Lr*1, *Lr*2) and a parallel inductance-capacitance (*Lp*, *Cp*).

**Figure 1.** Proposed bidirectional topology of novel resonant converter.

### *2.1. Forward Mode Analysis*

In forward mode, the switches Q1–Q4 are controlled simultaneously using variable frequency PWM signal with a duty cycle of 50%. The switching signals of Q2–Q3 are in complementary logic with Q1–Q4 ones. In this mode, the MOSFET switches of Q5–Q8 and Q6–Q7 at the DC-bus side are turned OFF. So, only the diode rectifier operates in this case. Energy is then forwarded from the DC-source *Ve* to the DC-bus side *Vbus*. The voltage between A and B (*VAB*) of the Figure 1 is a square waveform. It is analytically expressed by Equation (1), where *n* is the harmonic frequency order, and *f* <sup>1</sup> is the fundamental frequency.

$$v\_{AB}(t) = \frac{4.V\_{\varepsilon}}{\pi} \sum\_{n=1,3,5\dots} \frac{1}{n} \sin(n.2\pi f\_1.t) \tag{1}$$

First Harmonic Approximation (FHA) is adopted assuming the active energy is mainly attached to the fundamental frequency. This condition is achieved by filtering the current nearby the resonant frequency of the LC–LL circuit. Fundamental of *vAB*(*t*) is given in (2). More information about the method can be found in [18].

$$v\_{\varepsilon1}(t) = \frac{4.V\_{\varepsilon}}{\pi} \text{sin}\,\left(\,\omega\_{1}.\text{t}\right) \tag{2}$$

The output voltage *vCD*(*t*) of the resonant circuit given in Figure 1 is also assumed as a square wave form, and its fundamental component is given in (3), where ϕ*<sup>v</sup>* is the phase angle.

$$v\_{s1}(t) = \frac{4.V\_{bus}}{\pi} \sin(\omega\_1 t - \varphi\_v) \tag{3}$$

The output impedance through the diodes bridge is reflected by an equivalent resistance *R<sup>e</sup>* expressed in (4). *R* is a resistance-like load on DC-bus [34], and *I*<sup>0</sup> is the average current from the rectifier. The FHA model of the converter is presented in Figure 2 and the corresponding analytical model is given in (5).

**Figure 2.** Equivalent model of the proposed converter in forward mode.

$$R\_{\varepsilon} = \frac{V\_{s1}}{I\_{s1}} = \frac{8}{\pi^2} \frac{V\_{bus}}{I\_0} = \frac{8}{\pi^2} . \text{R}, \ \omega\_1 = \omega \tag{4}$$

$$\begin{aligned} & \begin{bmatrix} V\_{\varepsilon1}(j\omega) \\ I\_{\varepsilon1}(j\omega) \end{bmatrix} \\ &= \begin{bmatrix} 1 & j\omega . L\_{r1} \\ 0 & 1 \end{bmatrix} \begin{bmatrix} 1 & \frac{j\omega . L\_{r2}}{m^2} \\ \frac{m^2}{j\omega . L\_p + \frac{1}{j\omega . L\_p}} & 1 + \frac{j\omega . L\_{r2}}{j\omega . L\_p + \frac{1}{j\omega . L\_p}} \end{bmatrix} \begin{bmatrix} \frac{1}{m} & 0 \\ 0 & m \end{bmatrix} \begin{bmatrix} V\_{s1}(j\omega) \\ I\_{s1}(j\omega) \end{bmatrix} \end{aligned} \tag{5}$$

The voltage gain (*G*) based on the equivalent model analysis is given in (6).

$$\begin{cases} \begin{array}{c} Z\_{\text{ill}} = j\omega\omega \, L\_{r1} + Z\_{\text{c}}\\ Z\_{\text{c}} = \frac{R\_{\text{c}}\left(j\omega\omega \, L\_{p} + \frac{1}{j\omega\omega \, C\_{p}}\right) + \frac{L\_{r2}}{C\_{p}} - \omega^{2}\, \omega \, L\_{r2}\cdot L\_{p}}{m^{2}\left(R\_{\text{c}} + j\omega\omega \, L\_{r2} + j\omega\omega \, L\_{p} + \frac{1}{j\omega\omega \, C\_{p}}\right)} \end{array} \tag{6}$$

$$G = \frac{V\_{s1}}{m.V\_{\varepsilon 1}} = \frac{Z\_{\varepsilon}}{Z\_{in}} \frac{R\_{\varepsilon}}{R\_{\varepsilon} + j.\omega \cdot L\_{r2}} \tag{7}$$

The FHA response of the proposed circuit can be better analyzed in a per-unit (p.u.) system with the adopted parameters described below:

$$Q = \frac{1}{R\_\varepsilon} \sqrt{\frac{L\_p}{C\_p}} \tag{8}$$

$$f\_{\rm n} = \frac{f}{f\_r} \text{ with } f\_r = \frac{1}{2\pi\sqrt{L\_p C\_p}}\tag{9}$$

$$\mathfrak{a}\_1 = \frac{m^2 \cdot L\_{r1}}{L\_p} \; ; \; \mathfrak{a}\_2 = \frac{L\_{r2}}{L\_p} \tag{10}$$

where, *Q* is the quality factor, *f<sup>n</sup>* is the normalized frequency, *f<sup>r</sup>* is the series coupled *Lp*-*C<sup>p</sup>* resonant frequency. *α*<sup>1</sup> and *α*<sup>2</sup> are respectively the normalized values of primary and secondary inductances ratio in per-unit. The resulting output–input DC voltages ratio (Hf) is given in (11).

$$\mathrm{Hf} = \frac{V\_{bus}}{V\_{\ell}} \approx \left| \frac{1}{\left[1 + \frac{a\_1 f\_n^2}{f\_n^2 - 1}\right] + \left[j.f\_n, Q.\left(a\_1 + a\_2 + \frac{a\_1 a\_2 f\_n^2}{(f\_n^2 - 1)}\right)\right]}\right| \tag{11}$$

For *α*<sup>1</sup> = *α*<sup>2</sup> = *α*, the voltage ratio Hf versus the normalized frequency *f<sup>n</sup>* calculated for different load conditions (i.e., *Q*) is shown in Figure 3. The favorite operation zone for ZVS is located in 0.72 < *f<sup>n</sup>* < 1.

**Figure 3.** Voltage gain of the converter in forward mode for *α* = 1.

Figure 4 shows the voltage gain versus the normalized frequency *f<sup>n</sup>* for *Q* = 1.9 and different values of inductances ratio *α*. As displayed in Figure 3, the converter has two potential operation zones (ZCS & ZVS). ZVS can be fully achieved by switching frequency *f<sup>s</sup>* = *f<sup>n</sup>* with (0.72 \**f<sup>r</sup>* < *f<sup>n</sup>* < *fr*).

**Figure 4.** Voltage gain of the converter in forward mode (*Q* = 1.9).

The analysis of Figures 3 and 4 shows the following characteristics: Voltage gain at the resonant frequency is zero; Maximum voltage gain decreases when *Q* and *α* increase; and Voltage ratio varies enough in a narrow frequency range.

### *2.2. Backward Mode Analysis*

In backward mode of Figure 1, the switches Q5–Q8 and Q6–Q7 are controlled by variable switching frequency. The MOSFET switches of Q1–Q4 and Q2–Q3 are here turned OFF. In this mode, the energy is transferred from the DC-bus to DC-source which is assumed to be reversible (supercapacitors energy storage unit). The equivalent model of the converter in backward mode is presented in Figure 5 and analytical model is given in (12).

**Figure 5.** Equivalent model of the converter in backward mode.

$$
\begin{bmatrix} V\_{b1}(j\omega) \\ I\_{b1}(j\omega) \end{bmatrix} = \begin{bmatrix} m & 0 \\ 0 & \frac{1}{m} \end{bmatrix} \begin{bmatrix} 1 & j\omega . I\_{r2} \\ \frac{1}{j\omega L\_P + \frac{1}{j\omega C\_P}} & 1 + \frac{j\omega L\_L}{j\omega L\_P + \frac{1}{j\omega C\_P}} \end{bmatrix} \begin{bmatrix} 1 & j\omega L\_{r1} \\ 0 & 1 \end{bmatrix} \begin{bmatrix} V\_{sb1}(j\omega) \\ I\_{sb1}(j\omega) \end{bmatrix} \tag{12}
$$

The input impedance of the resonant circuit is described in (13). The quality factor *Q<sup>b</sup>* is given in (14), and the voltage gain in backward mode (Hb) is presented in (15).

$$\begin{cases} \begin{array}{c} Z\_i = j\omega \, L\_{r2} + Z'\\ m^2 R\_b \left( j\omega \, L\_p + \frac{1}{j\omega \, C\_p} \right) + m^2 \left( \frac{L\_{r1}}{C\_p} - \omega\_s^2 \, \right. \, L\_{r1} \, \, L\_p \right) \end{array} \tag{13}$$

$$Q\_b = \frac{m^2}{R\_b} \sqrt{\frac{L\_p}{C\_p}} \tag{14}$$

$$Hb = \frac{V\_c}{V\_{bus}} = \left| \frac{1}{\left[1 + \frac{a.f\_n^2}{f\_n^2 - 1}\right] + \left[j.f\_n, Q\_b, \left(2.\alpha + \frac{a^2.f\_n^2}{(f\_n^2 - 1)}\right)\right]}\right|\tag{15}$$

Figure 6 shows the voltage gain of the backward mode versus *f<sup>n</sup>* for different values of the quality factor *Q<sup>b</sup>* . We can see that the converter in the backward mode has exactly the same behavior as the forward mode. This means similar operation capabilities in the same restricted frequency domain. If the same tuned values of *α*<sup>1</sup> = *α*<sup>2</sup> = *α* is assumed, only the quality factor *Q<sup>b</sup>* will determine the operation frequency *f<sup>s</sup>* within 0.72 ∗ *f<sup>r</sup> < f*<sup>1</sup> *< f<sup>r</sup>* .

**Figure 6.** Voltage gain of the proposed converter in backward mode.

### *2.3. Operation Principles Analysis*

The symmetric bidirectional resonant circuit ensures the same behavior of the converter in both forward and backward modes. Thus, only the principle of the forward operating mode will be discussed in this subsection. Operating waveforms in forward mode are illustrated in Figure 7. Switching conditions are also seen.

**Figure 7.** Waveforms of the converter in forward mode.

Referring to this figure, the operating states of the converter in forward mode can be divided into five states for a half switching period. The operating sequences are gathered in Figure 8, which shows the active circuits' paths participating to power exchanges.

State 1 (*t*<sup>0</sup> < *t* < *t*1): This mode begins at *t*<sup>0</sup> when Q<sup>2</sup> and Q<sup>3</sup> are turned off. The current in the inductance *Lr*<sup>1</sup> discharges the parallel capacitor of Q<sup>1</sup> and Q<sup>4</sup> and charges the parallel capacitor of Q<sup>2</sup> and Q3. This state ends when the voltages across Q<sup>1</sup> and Q<sup>4</sup> (VQ1, VQ4) go to zero and those of Q<sup>2</sup> and Q<sup>3</sup> (VQ2, VQ3) reach the input voltage.

State 2 (*t*<sup>1</sup> <*t* < *t*2): After the full discharge of the parallel capacitors of Q<sup>1</sup> and Q4, the negative resonant current *ir*<sup>1</sup> flows through the diodes of Q<sup>1</sup> and Q4. The energy stored in the resonant circuit is fed back to the input side, to discharge the capacitor *Cp*. The expressions of the currents and voltages of resonant components in this mode are defined as follows:

$$i\_{r1}(t) = \frac{V\_i - V\_{Lp}(t\_1) - V\_{\mathbb{C}p}(t\_1)}{L\_{r1}}(t - t\_1) + i\_{r1}(t\_1) \tag{16}$$

$$i\_{Lp}(t) = \left[\frac{-V\_s}{m} - V\_{\mathbb{C}p}(t\_1) + V\_{L\mathbb{Z}}(t\_1)\right] \sqrt{\frac{\mathcal{C}\_p}{L\_p}} m^2 \sin[\omega\_r(t - t\_1)] + i\_{Lp}(t\_1) \cos[\omega\_r(t - t\_1)] \tag{17}$$

$$V\_{\mathbb{C}p}(t) = \frac{-V\_{\mathbb{S}}}{m} - V\_{\mathbb{L}2}(t\_1) + \frac{1}{m^2} \sqrt{\frac{L\_p}{\mathbb{C}\_p}} \, i\_{\mathbb{L}p}(t\_1) \sin[\omega\_{\mathbb{V}}(t-t\_1)] + \left[\frac{V\_{\mathbb{S}}}{m} + V\_{\mathbb{C}p}(t\_1) - V\_{\mathbb{L}2}(t\_1)\right] \cos[\omega\_{\mathbb{V}}(t-t\_1)] \tag{18}$$

$$i\_{r2}(t) = \frac{m^2 \left[\frac{-V\_\\$}{m} - V\_{Lp}(t\_1) - V\_{\mathbb{C}p}(t\_1)\right]}{L\_{r2}}(t - t\_1) + i\_{r2}(t\_1) \tag{19}$$

where,

$$
\omega\_r = \frac{1}{\sqrt{L\_p \cdot \mathbb{C}\_p}}
$$

State 3 (*t*<sup>2</sup> < *t* < *t*3): At *t*2, the resonant current *ir*<sup>1</sup> changes the direction and the current flowing trough Q1 and Q4 becomes positive. Thus, Q1 and Q4 turn on with ZVS. The current *ir*<sup>2</sup> is negative. Hence, the diodes of Q6 and Q7 are conducting to transfer the energy from the DC-source to DC-bus side. The expressions of different resonant currents and voltages in this mode are given as follows:

$$i\_{r1}(t) = \frac{V\_i + V\_{Lp}(t\_1) - V\_{\mathbb{C}p}(t\_1)}{L\_{r1}}(t - t\_2) + i\_{r1}(t\_2) \tag{20}$$

$$i\_{Lp}(t) = \left[\frac{-V\_s}{m} - V\_{\mathbb{C}p}(t\_1) - V\_{L\mathbb{Z}}(t\_1)\right] \sqrt{\frac{\mathcal{C}\_p}{L\_p}} m^2 \sin[\omega\_r(t - t\_2)] + i\_{Lp}(t\_2) \cos[\omega\_r(t - t\_2)] \tag{21}$$

$$V\_{\mathbb{C}p}(t) = \frac{-V\_{\mathbb{S}}}{m} - V\_{\mathbb{L}r2}(t\_2) + \frac{1}{m^2} \sqrt{\frac{L\_p}{\mathbb{C}\_p}} i\_{\mathbb{L}p}(t\_2) \sin[\omega\_\mathbb{f}(t-t\_2)] + \left[\frac{V\_{\mathbb{S}}}{m} + V\_{\mathbb{C}p}(t\_2) + V\_{\mathbb{L}r2}(t\_2)\right] \cos[\omega\_\mathbb{f}(t-t\_2)] \tag{22}$$

$$i\_{l2}(t) = \frac{m^2 \left[\frac{-V\_s}{m} - V\_{Lp}(t\_1) - V\_{\mathbb{C}p}(t\_1)\right]}{L\_{l2}}(t - t\_2) + i\_{l2}(t\_2) \tag{23}$$

State 4 (*t*<sup>3</sup> < *t* < *t*4): This state begins when the resonant current reaches the magnetizing current. The secondary current *ir*<sup>2</sup> becomes zero and the capacitor *C*<sup>2</sup> supplies energy to the load. In the primary side, the resonant current *ir*<sup>1</sup> charges the resonant circuit. The expressions of the resonant currents and voltages are given in (24) and (25), respectively.

$$i\_{I1}(t) = i\_{Lp}(t) = \left[V\_{l} - V\_{\mathbb{C}p}(t\_{\mathbb{S}})\right] \sqrt{\frac{\mathcal{C}\_{p}'}{L\_{p}' + L\_{I}}} \sin\left[\omega\_{r}'(t - t\_{\mathbb{S}})\right] + i\_{I1}(t\_{\mathbb{S}}) \cos\left[\omega\_{r}'(t - t\_{\mathbb{S}})\right] \tag{24}$$

$$V\_{\mathbb{C}p}(t) = V\_i + \sqrt{\frac{L\_p' + L\_{r1}}{C\_p'}} \, i\_{r1}(t\_3) \sin \left[ \omega\_{r}'(t - t\_3) \right] + \left[ -V\_i - V\_{\mathbb{C}p}(t\_3) \right] \cos \left[ \omega\_{r}'(t - t\_3) \right] \tag{25}$$
 
$$\text{where,}$$

$$
\omega'\_{r} = \frac{1}{\sqrt{\left(L\_{r1} + L'\_{p}\right) . C'\_{p}}} ; \ \text{C'}\_{p} = \text{C}\_{p}.m^2 ; \ \text{L'}\_{p} = \frac{L\_{p}}{m^2}
$$

**Figure 8.** *Cont.*

**Figure 8.** Operating states of the converter in forward mode.

State 5 (*t*<sup>4</sup> < *t* < *t*5): At *t*4, the current *ir*<sup>2</sup> changes the sign and it becomes positive. Thus, the diodes of Q<sup>6</sup> and Q<sup>7</sup> turn *off* with ZCS and the body diodes of Q<sup>5</sup> and Q<sup>8</sup> turn *on* to deliver the energy to the load. The resonant currents and voltage in this mode are given by the following equations:

$$i\_{r1}(t) = \frac{-V\_i - V\_{Lp}(t\_4) + V\_{Cp}(t\_4)}{L\_{r1}}(t - t\_4) + i\_{r1}(t\_4) \tag{26}$$

$$i\_{Lp}(t) = \left[\frac{V\_s}{m} - V\_{\mathbb{C}p}(t\_4) + V\_{Lr2}(t\_4)\right] \sqrt{\frac{\mathbb{C}\_p}{L\_p}} m^2 \sin\left[\omega\_r(t - t\_4)\right] + i\_{Lp}(t\_4)\cos\left[\omega\_r(t - t\_4)\right] \tag{27}$$

$$V\_{\mathbb{C}p}(t) = \frac{V\_{\mathbb{S}}}{m} + V\_{\mathbb{L}r2}(t\_4) + \frac{1}{m^2} \sqrt{\frac{L\_p}{\mathbb{C}\_p}} \, i\_{\mathbb{L}p}(t\_4) \sin[\omega\_{\mathbb{T}}(t - t\_4)] + \left[\frac{-V\_{\mathbb{S}}}{m} - V\_{\mathbb{C}p}(t\_1) + V\_{\mathbb{L}r2}(t\_1)\right] \cos[\omega\_{\mathbb{T}}(t - t\_4)] \tag{28}$$

$$i\_{r2}(t) = \frac{m^2 \left[\frac{-V\_s}{m} + V\_{Lp}(t\_1) + V\_{\mathbb{C}p}(t\_1)\right]}{L\_{r2}}(t - t\_4) + i\_{r2}(t\_4) \tag{29}$$

The operating principles analysis for the next half cycle is the same as detailed above.

### **3. Converter Design Method**

The design of the converter must ensure essential criteria to have a better efficiency of the converter in both forward and backward mode. The criteria to be met are as follows:


To ensure these criteria, a design procedure will be presented for a 3 kW resonant circuit test bench. The proposed converter was designed to serve as energy path between a variable DC-source with a voltage range of 60~240 V and a 270 V DC-bus. The maximum power of the converter corresponds to the resistance *R* and the maximum quality factor is *Qmax*.

$$R = 24.3 \text{ } \Omega; \text{ } Q\_{\text{max}} = 1.9 \tag{30}$$

Referring to Figure 4, the maximum voltage gain decreases if *α* increases. The next step is to define the value of *α* that ensures the maximum voltage gain for *Qmax*. Based on Figure 4, the required maximum voltage gain can be achieved for *α* = 0.9 or less. The switching frequency range and the voltage gain rely on *α*. A small value of *α* involves a narrow switching frequency range. However, *α* should not be very small in order to not induce relatively large size of the inductance *Lp*. The converter must operate in the ZVS region to ensure soft switching by switches control. The minimum switching frequency is the frequency that gives the maximum voltage gain corresponding to the maximum delivered power. This frequency should be located in the ZVS region, as mentioned previously. The switching frequency increases from the minimum value and the voltage gain decreases until it reaches zero for no-load operation, theoretically corresponding to the resonant frequency *f<sup>r</sup>* . The DC-bus voltage level is controlled to 270 V while the DC-source voltage varies between 60 V and 240 V. So, the converter is with a voltage gain range of 1.125∼4.5 in forward mode. In backward mode, the converter is in buck mode depending on the changes of roles through transformer terminals and parameters of consequent quality factor *Q<sup>b</sup>* .

Figure 9 shows the operating regions of the converter in forward and backward modes. The parallel inductance *L<sup>p</sup>* should be chosen to ensure the required voltage gain. A small value of *L<sup>p</sup>* allows to have a high voltage gain with ZVS for all load conditions. However, a very small value of *L<sup>p</sup>* will increase the circulating energy and the conduction losses. Once *L<sup>p</sup>* is defined, *Lr*<sup>2</sup> is calculated using Equation (10). The critical parameter to respect is the capacitor current to reduce the volume of connected capacitors. For this reason, the parallel components *C<sup>p</sup>* and *L<sup>p</sup>* are placed on the secondary of the transformer with a high transformer turn ratio to decrease the current of the capacitor and to increase the voltage capability. Parallel capacitor is calculated using Equation (31), and the transformer turn ratio is set according to Equation (32). The series inductance *Lr*<sup>1</sup> is then calculated using Equation (10).

**Figure 9.** Proposed converter Forward and Backward operating regions.

$$\mathcal{C}\_p = \frac{1}{\mathcal{L}\_p \text{(2.\pi .f\_r)}^2} \tag{31}$$

$$m = \frac{V\_{bus}}{V\_{\varepsilon\_{\rm min}}} \tag{32}$$

The design procedure of the proposed converter is described in Figure 10.

**Figure 10.** Flowchart of the proposed design procedure.

### **4. Experimental Test Bench and Results**

In order to verify the design method and to test the proposed bidirectional resonant converter topology, a laboratory test bench with a 3 kW resonant circuit was built. It was integrated in a 12 kW Dual Active Bridge converter designed for the other laboratory works. The experimental test bench is shown in Figure 11. The characteristics of this test bench are given in Table 2 and the reference of the 800V/39A MOSFETs is IXFN44N80P.

**Table 2.** Characteristics of the built test bench.


**Figure 11.** Proposed resonant converter test bench.

Figure 12 presents the open loop control strategy, where the frequency *fs* is an input parameter. This switching frequency is adjusted through DS1103 controller board which generates the control signals to maintain the DC-bus voltage at a desired level. To verify the soft-switching over the full operating range of the converter, some experimental tests are done with various input voltage and for different loads powers in both forward and backward mode.

**Figure 12.** Open loop control strategy of the converter.

Figure 13 shows experimental waveforms in forward mode. It is seen that ZVS is reached for the primary switches of the inverter. Further, the rectifier diodes turn off with ZCS regardless the load variations and the input voltage value. We also notice the narrow variation of the switching frequency which is typically shown in Figures 3 and 4 with the variations of the load and the input voltage such as:


**Figure 13.** Experimental waveforms of the converter in forward mode: (**a**) Ve = 60 V, P = 3 kW (full load); (**b**) Ve = 140 V, P = 1.5 kW (half load); (**c**): Ve = 240 V at no load.

The narrow frequency range reduces internal energy circulating and the turn off losses. Figure 14 presents the *Vab* voltage at the primary side in forward mode, and the resonant current *ir*1. We can see that the current lags the voltage. Then, ZVS is always reached.

**Figure 14.** Experimental waveforms at P = 0.35 kW and Ve = 100 V.

Figure 15 shows experimental waveforms in backward mode for different load conditions and DC-source voltage. The active switches Q5~Q<sup>8</sup> are turned on with ZVS and the diodes of Q1~Q<sup>4</sup> are turned off with ZCS regardless load and voltage conditions. Note that the switching frequency range in backward mode is also very narrow, which proves also the analyzed behavior in Figure 6.

**Figure 15.** *Cont.*

**Figure 15.** Experimental waveforms of the converter in backward mode: (**a**) Ve = 60 V, P = 1 kW; (**b**) Ve = 140 V, P = 1.5 kW.

Figure 16 shows the experimental waveforms of *Vbus*, *ir*<sup>1</sup> and *vCp* during startup. We notice the soft startup of the converter with no voltage or current peaks. This is due to the zero-voltage gain at the starting resonant frequency which demonstrates another advantage of the proposed topology. Figure 17 shows the measured efficiency of the converter in forward mode versus the power for different input voltages. The maximum efficiency is about 96% at 3 kW when the DC-source voltage is more significant (240 V in this case). The converter efficiency reduces when the input voltage decreases because the input current increases.

**Figure 16.** Startup experimental waveforms: where *Vbus* is the DC-bus voltage, *ir*<sup>1</sup> is the primary resonant current and *vCp* is the capacitor voltage.

**Figure 17.** Efficiency of the converter in forward mode versus the power for different input voltages.

Note that the efficiency of the converter can be increased by using low power loss components with the electric wiring optimizing. Figure 18 presents the power losses distribution in the converter for a full load condition, where the losses in the transformer and the resonant circuit reach 49% of the total losses (120 W). On the other hand, the losses in the active inverter are about 27% of the total losses, while the losses in the diode rectifier are almost 15% of the total losses. The conduction losses are higher compared to those of switching, in order of 3.5 times for the inverter and 4 times for the rectifier. The conduction losses in the inverter reach 21% of the total losses and they are 12% of the total losses in the diode rectifier. Due to soft switching, the switching losses in the inverter and in the rectifier diodes are reduced. They are 6% and 3% of the total losses, respectively. Figure 18 gives also the characteristics of the converter compared to the other ones summarized previously in Table 1.


**Figure 18.** Losses distribution and criteria fulfillment of the novel resonant converter.

### **5. Conclusions**

This paper presents a bidirectional resonant converter for wide voltage range applications. Due to the symmetric resonant circuit, the converter has the same behavior in both energies transfer modes. The voltage gain varies from zero to a maximum value in the suitable ZVS region. So, ZVS for active switches of the inverter side and ZCS for the diodes of the rectifier side are achieved regardless the input voltage, the energy transfer direction and the load conditions. Proposed converter topology has the capability to operate under wide voltage range 60~240 V. Control of the power flow is carried out in a predetermined narrow fundamental switching frequency band which increases significantly the efficiency by

reducing the circulating energy and the turn off losses. In addition, the proposed converter has no voltage gain at the resonant frequency and so, no specific startup strategy is needed. The experimental results of the 3 kW resonant circuit test bench validate the feasibility of the proposed converter. Moreover, they prove the effectiveness of the proposed design procedure for the developed converter.

**Author Contributions:** Conceptualization, methodology, validation, formal analysis, investigation, data curation, writing—original draft preparation, writing—review and editing, visualization, M.A., A.P., M.B.C., B.D.; supervision, project administration, funding acquisition, A.P., M.B.C., B.D. All authors have read and agreed to the published version of the manuscript.

**Funding:** This article has been funded by the council of Normandy Region (France).

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding authors.

**Acknowledgments:** This work was supported by University of Le Havre Normandy and is funded by Normandy region in France.

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

### **References**


## *Article* **Frequency Stability of AC/DC Interconnected Power Systems with Wind Energy Using Arithmetic Optimization Algorithm-Based Fuzzy-PID Controller**

**Ahmed H. A. Elkasem <sup>1</sup> , Mohamed Khamies <sup>1</sup> , Gaber Magdy <sup>2</sup> , Ibrahim B. M. Taha 3,\* and Salah Kamel 1,\***


**Abstract:** This article proposes an intelligent control strategy to enhance the frequency dynamic performance of interconnected multi-source power systems composing of thermal, hydro, and gas power plants and the high penetration level of wind energy. The proposed control strategy is based on a combination of fuzzy logic control with a proportional-integral-derivative (PID) controller to overcome the PID limitations during abnormal conditions. Moreover, a newly adopted optimization technique namely Arithmetic optimization algorithm (AOA) is proposed to fine-tune the proposed fuzzy-PID controller to overcome the disadvantages of conventional and heuristic optimization techniques (i.e., long time in estimating controller parameters-slow convergence curves). Furthermore, the effect of the high voltage direct current link is taken into account in the studied interconnected power system to eliminate the AC transmission disadvantages (i.e., frequent tripping during oscillations in large power systems–high level of fault current). The dynamic performance analysis confirms the superiority of the proposed fuzzy-PID controller based on the AOA compared to the fuzzy-PID controller based on a hybrid local unimodal sampling and teaching learning-based optimization (TLBO) in terms of minimum objective function value and overshoots and undershoots oscillation measurement. Also, the AOA's proficiency has been verified over several other powerful optimization techniques; differential evolution, TLBO using the PID controller. Moreover, the simulation results ensure the effectiveness and robustness of the proposed fuzzy-PID controller using the AOA in achieving better performance under several contingencies; different load variations, the high penetration level of the wind power, and system uncertainties compared to other literature controllers adjusting by various optimization techniques.

**Keywords:** load frequency control (LFC); multi-source power system; fuzzy logic control (FLC); high wind energy penetration

### **1. Introduction**

Consistent with the noticeable increase in energy demand, it is necessary to establish new energy sources. However, most efforts concern with establishing renewable energy sources (RESs) instead of conventional energy sources (CESs) due to the negative and harmful effects (e.g., global warming) of CES on our community [1–3]. So, energy planners and researchers make great efforts and strive to establish RESs paralleling with electrical networks for reducing CESs' hazards. In addition, RESs are clean and safe energy which be friendly to the environment [4]. While the establishment of RESs decreases the system inertia and may negatively affect the system stability [5,6]. Based on the aforementioned observations, the modern power grids will face a great challenge in keeping the system's frequency and the tie-line power stable. Therefore, it is important to keep the system stable

**Citation:** Elkasem, A.H.A.; Khamies, M.; Magdy, G.; Taha, I.B.M.; Kamel, S. Frequency Stability of AC/DC Interconnected Power Systems with Wind Energy Using Arithmetic Optimization Algorithm-Based Fuzzy-PID Controller. *Sustainability* **2021**, *13*, 12095. https://doi.org/10.3390/ su132112095

Academic Editors: Nicu Bizon, Mamadou Baïlo Camara and Bhargav Appasani

Received: 18 September 2021 Accepted: 28 October 2021 Published: 2 November 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/).

during these previous conditions, this occurs by applying load frequency control (LFC) to maintain the frequency and tie-line power of the system at their specified values [7]. In this regard, numerous control techniques/strategies have been conducted to make progress in modern power system frequency stability.

One of these strategies is optimal control, which includes linear quadratic regulators that are applied to regulate the frequency of two-area power systems considering AC/DC tie-lines [8]. Also in [9], the linear quadratic regulator is applied to enhance the frequency of the two-area power system in the presence of electrical vehicles. Another strategy known as robust control techniques, such as the second-order sliding mode controller, has been applied to regulate the frequency in multi-area power systems [10]. In [11], one robust control technique known as the µ -synthesis approach is applied to regulate the frequency of the islanded micro-grid frequency containing (diesel engine generator, fuel cell, wind turbine, and PV array). In [12], µ-synthesis approach has been applied to recover the frequency fluctuations under uncertainty weighing selection in power plants. *H*<sup>∞</sup> Controller has been designed in [13] to regulate the frequency of the power system while accounting for uncertainties. There is another type of these strategies known as model predictive control, which has been applied to enhance the performance of three-area interconnected power system considering the penetration of wind turbines [14]. Also, there are intelligent control strategies such as (e.g., artificial neural network and fuzzy logic control (FLC)) have been applied to counteract the system's frequency deviations during disturbances in the presence of tidal power units, electrical vehicles, energy storage systems, and solar power systems [15,16] and so on. While all these strategies were successful in resolving LFC difficulties, they depended on the designer's knowledge and experience, experimenting, and trial and error procedures in finding controller parameters, and it takes a very long time to approximate their parameters.

On the other hand, the proportional-integral-derivative (PID) controller and its forms are still the most popular controller due to their good characteristics such as simplicity and low cost [17]. However, the PID controllers are sensitive to system parameter variations (i.e., system uncertainties) and nonlinearities [18]. Thus, the FLC strategy has been proposed to support the PID controller performance in enhancing the dynamic frequency stability of modern power systems. The main advantages of the fuzzy controller include its simplicity of execution, high sensitivity to system fluctuations, and the ability to safely handle changes in the operating point or system parameters due to online updating of the controller parameters [19]. The first attempt in using the FLC for solving the LFC problem with the PID controller was conducted in [20]. Additionally, the self-tuning Fuzzy-PID controller was proposed to enhance the frequency stability of the interconnected power system, consisting of two areas [21]. Also, the fuzzy-PID controller has been utilized to stabilize the frequency of a multi-source power system [22]. Where the proposed Fuzzy-PID controller parameters have been selected based on an optimization technique. In addition, the Fuzzy-PID controller has been utilized to improve the stability of power system frequency considering the flexible AC transmission system devices (i.e., the static synchronous series compensator, the unified power flow controller, and the interline power flow controller) effect in [23,24]. The FLC is presented by parameters (i.e., inputs, scaling factors, membership functions, and rule base) that have not specific rules which can be followed to detect their values. Generally, trial and error methods have been used to select the parametric values, but these methods may not give the best performance. In this regard, this study proposes selecting the scaling factors of the fuzzy-PID controller based on a new meta-heuristic technique known as the Arithmetic Optimization Algorithm (AOA).

Therefore, there is another choice to take care of the LFC issue: to utilize various optimization methods and these are successfully utilized to manage the nonlinear functions associated with the LFC design. These several optimization techniques are applied for finding the optimal controller parameters to overcome the LFC problem and achieve more system security. The utilized techniques by researchers in the LFC issue such as; grasshopper optimization algorithm [25], ant colony optimization technique [26], Jaya

algorithm [27], particle swarm optimizer [28], firefly algorithm [29], hybrid pattern search shuffled-frog leaping algorithm [30], multi-objective genetic algorithm [31], grey wolf optimizer [32], sine cosine algorithm [33], harris hawks optimizer and salp swarm algorithm [34], lightning-attachment procedure optimization (LAPO) [35] and improved LAPO [36]. However, these techniques achieve exceptional performance by ensuring effectual LFC design. They have a slow convergence rate, restricted search capability, as well as local optimum convergence. On the other hand, the AOA algorithm has been made to overcome previous limitations related to different optimization techniques. The superiority of AOA than other conventional optimization algorithms returns to the gradient-free mechanism and its capability to avoid the local solutions and obtain the global solution with little search agents. Also, many experimental results show that, AOA provides very promising results in solving real-world engineering design. So, this study applied the AOA algorithm for fine-tuning the proposed fuzzy-PID controller parameters due to its promising results in solving several real-world engineering design problems [37]. Also, the AOA recently gave a distinguished performance in medicine field (i.e., evaluate images of COVID-19) [38].

The interconnected multi-source power systems need strong links (tie-lines) utilized in constructural power exchange between different control areas. Also at abnormal conditions, these links provide a support to inter-area. Several studies have been maiden about the topic of LFC with the presence of AC transmission line only without HVDC line connection [39,40]. However, many problems related to the AC interconnection between areas in the power system, especially long-distance power transmission, remain. The problems associated with the AC interconnection can be mentioned as frequent tripping occurs at the instant of large power oscillations, an increase in fault current level, which leads to damage in the power system and transmission oscillations from one area to another, which causes deterioration in system dynamic performance. So this study proposes using of HVDC interconnection besides the AC tie line for transmitting the bulk power over the long distance to eliminate the demerits of the AC transmission lines and according to good features of HVDC transmission. The attractive features of HVDC links are summarized as: there are converters in HVDC lines that give the ability to fast controllability in power between interconnected areas. It can overcome the transient stability problems associated with AC transmission [41]. According to the point of obtaining stabilizing in the electrical power systems such as mentioned previously when adding HVDC lines, there are several studies deal with the issue of predicting processes of wind speed during participation in electrical power systems using different meta-heuristic techniques. These researchers seek to avoid fluctuations when wind speed exceeds the permissible limits [42–45]. On the other hand, there are off-shore wind turbines which characterized by a high average speed compared to on-shore wind turbines. Also, those off-shore wind turbines produce more electricity than on-shore wind turbines. Therefore, many researchers strive to choose the best site of off-shore wind turbines in coastal areas such as Turkey and USA to generate more electricity and link between these turbines and main electrical network to meet the need for citizens [46,47].

In terms of LFC issues, traditional controllers, such as the PID controller, have some challenges in parameter tuning and have not accommodated system stability in the face of uncertainties. Moreover, few studies applied fuzzy-PID controller to diminish the demerits of PID controller, but the parameters of the fuzzy-PID has been selected based in conventional and heuristic optimization algorithms. Furthermore, the renewables penetration effect has been considered in few studies, and not considered in other works [22,24,48]. Additionally, the effect of HVDC has been ignored in most works related to LFC studies [39,40]. So, this study proposes a robust control strategy based on Fuzzy-PID controller to keep the stability of systems involving different types of generating units in addition to renewable sources. In addition, the parameters of the fuzzy-PID controller have been selected based on a novel meta-heuristic algorithm known as AOA algorithm due to its merits. Unlike, previous works which have neglected the parameters variations effect [14,22],

the control design consider, system nonlinearities and system uncertainties have been considered during designing the proposed control strategy. Finally, the effect of HVDC has been considered to eliminate the demerits of AC transmission lines. Furthermore, Table 1 introduces a comparison between the motivation of this work and other studies.

**Properties [25] [39] [40] [49] [50] [50] This Study** Type of controller Fuzzy-PID controller Optimal PI-PD cascaded controller Optimal PID controller Optimal PID controller Optimal PID controller Fuzzy-PID controller Fuzzy-PID controller Adoption of controller design on Grasshopper optimization algorithm (GOA) Flower pollination algorithm (FPA) Grey wolf optimization (GWO) Differential evolution (DE) Teachinglearning based optimization (TLBO) Hybrid local unimodal sampling (LUS) with TLBO Arithmetic optimization algorithm (AOA) Penetration of renewable energy sources Not considered Not considered Not considered Not considered Not considered Not considered Considered with high penetration of wind energy Effect of system uncertainties considered Not considered considered Not considered Not considered Not considered considered Effect of HVDC link Not considered Not considered Not considered considered considered considered considered

**Table 1.** Comparison between this work and previous mentioned studies.

The main contribution of this study can be summarized as follows:


The remainder of this research is summarized as follows: the modeling and configuration of the studied interconnected power system considering wind energy are discussed in Section 2. Section 3 presents the proposed fuzzy-PID controller methodology, the proposed optimization technique AOA, and the construction of the proposed control strategy. Then, Section 4 presents the simulations and investigation results. Finally, the conclusion of this work is mentioned in Section 5.

### **2. Modeling and Configuration of the Studied System**

### *2.1. A dynamic Model of Two-Area Interconnected Power System*

This article discusses the LFC issue of interconnected multi-source power systems. Where the studied system is composite of two-areas which interconnected together by a tie-line. Three power plants (i.e., the reheat thermal unit, the gas unit and the hydro unit) are included in each area of the investigated power system. Moreover, each unit in both areas has its speed governing system, turbine, and generator. The capacity or rating power of the investigated system is 2000 MW [51]. Also, the system dynamic performance has been investigated in the presence and absence of an HVDC link. The fuzzy-PID controller is proposed to be equipped in both areas for each generation unit to minimize the oscillations in both area frequencies and tie-line power between them. The input signals of the proposed fuzzy-PID controller represent the area control error *(ACE*) and its derivative, while the output signal represents the secondary control action on each generated unit. Figure 1 shows the dynamic model of the studied two-area interconnected power system and the schematic diagram is shown in Figure 2. The nominal parameters' values of the studied power system are given in Table 2. The ACEs in both areas can be obtained according to formulas as follows in Equations (1) and (2):

$$ACE\_1 = \Delta P\_{tie1-2} + B\_1 \Delta f\_1 \tag{1}$$

$$ACE\_2 = \Delta P\_{\text{tie2}-1} + B\_2 \Delta f\_2 \tag{2}$$

where, ∆*Ptie*1−<sup>2</sup> and ∆*Ptie*2−<sup>1</sup> represent the tie-line power exchange at area 1 and area 2, *B*<sup>1</sup> and *B*<sup>2</sup> are the bias frequency factors of area 1 and area 2 respectively, ∆*f* <sup>1</sup> is the deviation in frequency waveform in area 1, and also ∆*f*<sup>2</sup> is the deviation in frequency in area 2.

**Figure 1.** Transfer function model of the two-area interconnected power system. (see Appendix A).

**Figure 2.** Schematic diagram of the studied AC/DC interconnected power system.

**Table 2.** The parameters of two identical interconnected areas with standard values [49].


### *2.2. The Wind Farm Configuration*

The model of the wind power has been built using the MATLAB/SIMULINK program. The random wind power is integrated with conventional units in both areas. According to the design of the wind power model, a white noise block is used to get a random speed which is multiplied by the wind speed as shown in Figure 3 [52]. The following equations illustrate the captured power from the wind by the rotor of the wind turbine [35].

$$P\_{wt} = \frac{1}{2} \rho A\_T v\_w^3 \mathcal{C}\_p(\lambda, \beta) \tag{3}$$

$$\mathbb{C}\_{p}(\lambda,\beta) = \mathbb{C}\_{1} \left( \frac{\mathbb{C}\_{2}}{\lambda\_{i}} - \mathbb{C}\_{3}\beta - \mathbb{C}\_{4}\beta^{2} - \mathbb{C}\_{5} \right) \times e^{\frac{-\mathbb{C}\_{6}}{\lambda\_{i}}} + \mathbb{C}\_{7}\lambda\_{T} \tag{4}$$

$$
\lambda\_T = \lambda\_T^{OP} = \frac{\omega\_T r\_T}{V\_W} \tag{5}
$$

$$\frac{1}{\lambda\_i} = \frac{1}{\lambda\_T + 0.08\beta} - \frac{0.035}{\beta^3 + 1} \tag{6}$$

where, *Pwt* represents the captured output power of wind turbine, *A<sup>T</sup>* is the swept area by the blades of turbine in *m*<sup>2</sup> , *ρ* is the air density (nominally 1.22 Kg/m<sup>3</sup> ), *V<sup>W</sup>* is the wind speed in m/s.

**Figure 3.** The wind power modeling using MATLAB/Simulink.

The climatic and geographical conditions where the studied wind turbine units located are the same. Thus, all previous parameters are applied to these units. The coefficient of rotor blades *C<sup>p</sup>* based on turbine coefficients *C*1–*C*<sup>7</sup> and it is a function on *λ<sup>T</sup>* which refers to the optimum tip-speed ratio (TSR) and pitch angle *β*. *λ<sup>T</sup>* is a function on the rotor speed (*ωT*) and the blade length of rotor radius (*rT*). Moreover, *λ<sup>i</sup>* is referring to the intermittent TSR. Table 3 shows the nominal parameter values of the wind turbine for the wind farm applied with the studied power system. Figure 4 shows the random output power of 130 wind turbine units of 750 KW which have been penetrated at both areas of the studied power system.



**Figure 4.** The wind power variation pattern.

### **3. Control Methodology and Problem Formulation**

According to RESs penetration, system nonlinearities, and system uncertainties, it is essential to design good controller to improve the system performance during abnormal conditions. Hence, this study proposes fuzzy-PID controller to overcome any deviations resulted from previous considerations. Moreover, the proposed controller parameters have been selected based AOA algorithm.

### *3.1. The Proposed Control Strategy*

In this article, three fuzzy-PID controllers are proposed as responsible for extracting extra active power from thermal, hydro, and gas turbines respectively, when load disturbance occurs. In this regard, several studies have applied the trial and error runs method to detect the fuzzy-PID controller's input and output scaling factors [53]. Therefore, it is difficult to obtain the optimal parameter values which enhance the system performance through these trial and error methods. Therefore, this paper proposes designing the fuzzy-PID controller with the optimized input and output scaling factors. The AOA has been selected in this work to obtain the optimized values of the proposed controller's input and output scaling factors. Furthermore, Figure 5 shows the structure of the fuzzy-PID controller of the thermal, hydro, and gas units. It has two inputs; ACE and the change in ACE, and one output. The input scaling factors are K<sup>1</sup> and K<sup>2</sup> and the output scaling factors are K<sup>3</sup> and K<sup>4</sup> which are optimized via the proposed AOA.

**Figure 5.** The proposed Fuzzy-PID controller diagram.

The fuzzy member ship may be triangular membership or Gaussian membership. Several studies applied the triangular membership due to its merits, where the triangular membership represents one of the attractive linear memberships of the fuzzy methodology, which is characterized by less computation time and simplicity [54]. The sensitivity increases when moving from linear membership functions to curvilinear membership functions (i.e., Gaussian-sigmoidal). Therefore, from the literature review of fuzzy logic,

the triangular membership is shown in more than 90% of practical applications of electrical systems which are used in input and output. In addition, this membership belongs to the first-order mathematical function, which is characterized by reducing the computational load. It is usually applied along with a PID controller and fully symmetric functions in input and output. The selection of fuzzy control parameters depends on the nature of the studied system and the knowledge of the designer of the system. So, the selection of fuzzy membership functions range is based on the prediction of the universe of discourse of input and output in the studied system [55]. Usually, the decision-maker is able to define the risk-free of input and output. Thus, the range intervals have been selected between [−1, 1] intervals, which is expected and doesn't need the input and output to go far away from this period to achieve more system stability. It is advisable to use the symmetric triangular MF with 50% overlap, and then apply a tuning procedure during which we can either change the left and/or right spread and/or overlap. This is to be continued till, getting satisfactory results [56]. In this work, five triangular membership functions of the fuzzy-PID controller are utilized namely negative big (NB), negative small (NS), zero (Z), positive small (PS), and positive big (PB). The triangular member function is shown in Figure 6, which is used for both inputs and output. Accordingly, the five memberships of the input and output variables of the fuzzy-PID controller, the generation of fuzzy output need 25 rules. These rules play an important role in the performance of the fuzzy-PID controller. These rules of the fuzzy logic controller are depicted in Table 4.

**Figure 6.** The Fuzzy Logic Controller membership functions of inputs and output.



### *3.2. The Proposed Optimization Technique (AOA)*

In 2020, a new meta-heuristic optimization technique namely the Arithmetic optimization algorithm (AOA) was invented by Laith et al. [57]. This technique is characterized by a high exploration search strategy, meaning that they can achieve the global optimum solution with few search agents. The distribution behavior of this technique is based on the main mathematical operations including; (Division (*D*), Multiplication (*M*), Addition (*A*), and Subtraction (*S*)). According to meta-heuristic techniques, the former of this method

can make a wide coverage of searching space by using the number of search agents of the algorithm to avoid the local solution but for obtaining the global one. The AOA follows a detected methodology to obtain the global solution mentioned in the next steps:

Step 1: the proposed controller parameters in this paper have upper and lower boundaries; the population of the AOA method according to the main mathematical operations has been generated between these boundaries to achieve the global goal. The population of the AOA method can be formulated in Equation (7) as follows:

$$s(N,d) = rand(N,d) \times \ (LB - LB) + LB \tag{7}$$

where; *N* refers to the number of utilized search agents, *d* represents the variable dimensions (controller parameters), *UB* and *L* represent the upper and lower value of variables.

Step 2: The AOA method begins with candidate solutions (S) generated randomly. The optimal obtained solution in each endeavor represents a solution near the global goal (target). The next matrix illustrates the position of solutions obtained.

$$s = \begin{bmatrix} s\_{1,1} & s\_{1,2} & \cdots & s\_{1,d} \\ s\_{2,1} & s\_{2,2} & \cdots & s\_{2,d} \\ \vdots & \vdots & \ddots & \vdots \\ s\_{n,1} & s\_{n,2} & \cdots & s\_{N,d} \end{bmatrix} \tag{8}$$

Step 3: achieving the fitness solution among the obtained ones as mentioned in step 2. The fitness solution can be formulated in Equation (9) as follows:

$$f\_{fitness} = \begin{bmatrix} f1 \ f2 \ f3 \dots \dots \dots \dots \dots \ f\_N \end{bmatrix}^T \tag{9}$$

Step 4: before the AOA role begins, the search phase (exploration and exploitation) must be detected through the next formulation in Equation (10) of Math Optimizer Acceleration (*MOA*):

$$MOA(\mathcal{C}\_{-}Iter) = \mathcal{C}\_{-}Iter \times \left(\frac{Max - Min}{M\_{-}Iter}\right) + Min \tag{10}$$

where; *MOA*(*C*\_*Iter*) refers to the value of function at the *t*th iteration, *C*\_*Iter* represents the current iteration. The current iteration is between among 1 and the maximum number of iterations (*M*\_*Iter*) dimensions (controller parameters), *Min* and *Max* are the minimum and maximum values of the accelerated function.

Step 5: the mathematical calculation processes as (Division (*D*) and Multiplication (*M*)) cannot reach the target easily according to their high dispersion. Accordingly, these strategies (*D* and *M*) are utilized in the exploration search process. The first operator *D* in this phase is conditioned by r2 < 0.5 (r2 is a random number) and the second operator *M* will be neglected until the first one ends its task in searching for a solution near to the goal. Otherwise, *M* will lead the process instead of the first operator *D.* The exploration process is modeled in Equation (11):

$$\text{Xi}\_{\prime}j(\text{C}\_{\text{ }Iter}+1) \left\{ \begin{array}{l} \text{best}\{\text{x}\_{\text{j}}\} \div (\text{MOP}+\text{c}) \times ((\text{L}\text{Bj}-\text{L}\text{Bj}) \times \mu + \text{L}\text{Bj}), \text{ } r2 < 0.5\\ \text{best}\{\text{x}\_{\text{j}}\} \times (\text{MOP}) \times ((\text{L}\text{Bj}-\text{L}\text{Bj}) \times \mu + \text{L}\text{Bj}), \text{ } \text{otherwise} \end{array} \right. \tag{11}$$

where, *Xi*(*C*\_*Iter* + 1) and *Xi*, *j*(*C*\_*Iter* + 1) are the *i*th solution in the next iteration and the *j*th position of the *i*th solution at the current iteration, *best xj* represents the *j*th position in the best-obtained solution, *e* is a small integer number, *µ* is the control parameter to make adjusting in the process of search which is fixed equal to 0.5.

Step 6: deep search (exploitation) in this strategy using the mathematical operators (Subtraction (*S*) and Addition (*A*)) has been applied to be near to the optimal solution and reach it after several iterations. The first operator *S* in this phase is conditioned by r3 < 0.5 (r3 is a random number) and the second operator *A* will be neglected until the first one

end its task in deep searching for obtaining the best solution. Otherwise, *A* will lead the process instead of the first operator *S.* The exploitation process is modeled in Equation (12):

$$\begin{aligned} \text{Xi},j(\text{C}\_{\text{L}}\text{Iter}+1) \left\{ \begin{array}{c} \text{best}\left(\text{x}\_{\text{j}}\right)-(\text{MOP}) \times \left(\left(\text{UBj}-\text{LBj}\right)\times\mu+\text{LBj}\right), \text{r3}<\text{0.5}\\ \text{best}\left(\text{x}\_{\text{j}}\right)+\left(\text{MOP}\right)\times\left(\left(\text{UBj}-\text{LBj}\right)\times\mu+\text{LBj}\right), \text{otherwise} \end{array} \right.\end{aligned} \tag{12}$$

Step 7: steps (3) to (6) are repeated until ending all iterations.

Step 8: The last step is to achieve the optimum solution which achieves the objective function.

Furthermore, Figure 7 illustrates the flow chart of the AOA which clarify the previous optimization steps.

**Figure 7.** The flowchart of the AOA technique.

### *3.3. The Proposed Fuzzy-PID Control Strategy Based AOA Algorithm*

The parameters of the proposed control strategy have been selected based on AOA algorithm to overcome any deviations related to the considered system. Moreover, the integral time absolute error (ITAE) function is utilized as an objective function ( *J* ) of the proposed optimization algorithm. Furthermore, Equation (13) formulates the objective function *J* to minimize the deviations in system related to the frequency and tie-line power. The ITAE has been selected in this work according to its merits like, it has an additional time multiplies with the error function which makes the system faster than using other

objective functions forms (e.g., the integral square error (ISE) and the integral absolute error (IAE)). Also, the ITAE performance index has the advantage of settling the system, which is more quickly compared to other objective functions [49].

$$J = \text{ITAE} = \int\_0^{\text{Tsim}} |\Delta f \mathbf{1}| + |\Delta f \mathbf{2}| + |\Delta p tie| ).t \times d\_t \tag{13}$$

where; *Tsim* is the simulation time. The flow chart of the proposed AOA is shown in Figure 8.

**Figure 8.** The flowchart of applying AOA technique with the proposed controller.

### **4. Discussion and Simulation Results**

The investigated system was built on a 2.60 GHz Intel (R) PC with 4.00 GB of RAM using the MATLAB/SIMULINK®software (R2019b) environment. In addition, the AOA has been written in the m file in order to tune the proposed controller parameters of the automatic load frequency control process. In this work, the performance of Fuzzy-PID and PID controllers that are applied to enhance the studied system performance using the AOA technique is measured according to the value of the best objective function over iterations. The initial values of the proposed AOA technique utilized in this work are; the number of search agents equals 30 and the number of maximum iterations equals 50. Also, the limitations of the proposed fuzzy-PID controller are in the range of [0, 10]. The convergence curve of the proposed fuzzy-PID controller compared to the PID controller using the AOA is shown in Figure 9. There is a clear difference between the performances of both controllers using the AOA technique. The behavior of the PID convergence curve can be summarized as follows: it begins with a high best function value (near to 0.08) and drops along iterations until it ends its career at the final iteration, reaching the best function value (near to 0.03). As for the Fuzzy-PID convergence curve behaviors, they start with a low best function value until it reaches a value (near to zero) to obtain an optimal objective function with more system stability. In general, the preference is for the Fuzzy-PID controller.

**Figure 9.** The convergence curve of the designed controllers based on the AOA.

All next simulation results ensure the effectiveness of the AOA in obtaining optimum controller parameters compared to other applied techniques as (PID controller based-DE [49], PID controller based-TLBO [50], and Fuzzy-PID controller based-LUS-TLBO [50]). For programming all mentioned optimization techniques, MATLAB/M-files matched with MATLAB /Simulink. All graphical and numerical numbers of obtained results are discussed in the next scenarios as follows:

### *4.1. Studied Power System Performance Considering AC-Lines Connection Only*

This scenario clarifies the dynamic system performance of the occurring deviation in frequency waveform at both areas and the tie-line power exchange between them with a 1%

step load perturbation (SLP) which applied to the first area only. The AC-line has tied both areas of the studied power system without any HVDC lines. Compared to other controllers applied previously by researchers, the proposed fuzzy-PID controller-based AOA proves its robustness in adjusting and stabilizing the power system frequency. Figures 10 and 11 show the frequency deviation performance at both areas of the studied system. The tie-line power exchange has been cleared in Figure 12. Table 5 indicates the optimal Fuzzy-PID controller parameters compared to other mentioned controllers with different optimization techniques. Additionally, the performance specifications; overshoot (OS), and undershoot (US) of the proposed Fuzzy-PID controller and followed controllers for the studied system are shown in Table 6. The percentage improvements in US and OS with different controllers are denoted in Table 7.

**Figure 10.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.1.

**Figure 11.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.1.

**Figure 12.** Dynamic response comparison results of ∆*ptie* for scenario 4.1.



**Table 6.** Performance evaluation of PID and Fuzzy-PID using all mentioned techniques for scenario 4.1.


**Table 7.** Percentage improvement in US and OS for all previous mentioned controllers with different techniques based on PID controller via DE for scenario 4.1.


### *4.2. Studied Power System Performance Considering AC-DC Lines Connection*

The effect of adding an HVDC lines connection in addition to the existing AC lines to transmit the power alternately between both areas in the studied system is introduced in this scenario with a 1% step load perturbation (SLP) is applied to the first area only. The behavior of frequency at both areas and the tie-line power between the both is shown in Figures 13–15 respectively. Table 8 indicates different obtained controller parameters of the studied system in the presence of an HVDC line connection. It is observed that from Table 9, the proposed fuzzy-PID controller-based AOA achieves more system stability (less oscillation) by monitoring the system performance specifications such as OS and US than other mentioned controllers. Table 10 illustrates the percentage improvement in US and OS of area frequencies and tie-line power with different controllers.

**Figure 13.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.2.

**Figure 14.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.2.

**Figure 15.** Dynamic response comparison results of ∆*ptie* for scenario 4.2.


**Table 8.** The optimal controllers' values for scenario 4.2.

**Table 9.** Performance evaluation of PID and Fuzzy-PID using all mentioned techniques for scenario 4.2.


**Table 10.** Percentage improvement in US and OS for all previous mentioned controllers with different techniques based on PID controller via DE for scenario 4.2.


### *4.3. Studied Power System Performance Considering AC-DC Lines Connection in Addition to Different Load Disturbances*

The perturbation in load at the first area has been increased to be a 5% SLP instead of 1% to ensure the validation of the obtained fuzzy-PID controller parameters that mentioned in Table 4 in stabilizing the studied system frequency. The frequency deviation of both areas and the exchanged power between them is shown in Figures 16–18 respectively. The OSs and USs values of oscillations at frequency waveform at area 1 and area 2 are mentioned in Table 11; in addition to values of tie-line power. Table 12 shows the percentage improvement in OS and US with different controllers in the case of increasing SLP to 0.05 p.u. It can be said that the studied system became more stable using the proposed fuzzy-PID controllerbased AOA compared to the fuzzy-PID controller-based LUS-TLBO and PID controller based on TLBO, DE, and AOA.

**Figure 16.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.3.

**Figure 17.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.3.

**Figure 18.** Dynamic response comparison results of ∆*ptie* for scenario 4.3.


**Table 12.** Percentage improvement in US and OS for all previous mentioned controllers with different techniques based on PID controller via DE for scenario 4.3.


### *4.4. Studied Power System Performance Considering the Effect of System Parameters' Variations*

The robustness of the studied power system has been tested by making a variation of system parameters such as thermal governor time constant at both areas, thermal turbine time constant at both areas and hydro governor time constant simultaneously in the range of +25% and −25% from their nominal values reported in Table 2. These variations ensure that, the obtained fuzzy-PID controller parameters based AOA that mentioned in Table 4 can efficiently damp and overcome the oscillations and achieve more system stability under system parameters variation. The dynamic performance of both areas frequencies and tieline power exchange after ±25% system parameters variation is illustrated in Figures 19–21 respectively.

**Figure 19.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.4.

**Figure 20.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.4.

**Figure 21.** Dynamic response comparison results of ∆*ptie* for scenario 4.4.

## *4.5. Studied Power System Performance Considering Wind Power Penetration* 4.5.1. Case A

The penetration of wind power at both areas of the studied system is tested in this scenario at nominal system parameters. Both wind farms have the same power rating and penetrated the studied system at the same time (t = 0 s) with the instant of 1% load variation. This scenario is applied to ensure the robustness of the proposed fuzzy-PID controller based AOA that mentioned in Table 4 in achieving system stability in existing wind energy. All dynamic system performance (both-area frequencies and tie power) have been plotted in Figures 22–24 respectively. Table 13 illustrates the OSs and USs behavior of both areas frequencies and tie-line power with penetrating of wind energy at the studied power system. Also, Table 14 indicates the percentage performance in US and OS with different controllers.

**Figure 22.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.5.1: case A.

**Figure 23.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.5.1: case A.

**Figure 24.** Dynamic response comparison results of ∆*ptie* for scenario 4.5.1: case A.

**Table 13.** Performance evaluation of PID and Fuzzy-PID using all mentioned techniques for scenario 4.5.1: case A.


**Table 14.** Percentage improvement in US and OS for all previous mentioned controllers with different techniques based on PID controller via DE for scenario 4.5.1: case A.


### 4.5.2. Case B

In this case, the studied power system is stabilized until load variation occurred at (t = 40 s) in the first area. Also, the wind farm at the first area shared generated power at (t = 300 s) then the wind energy from the wind farm at the second area enter to feed the system at (t = 600 s). It is observed that the obtained fuzzy-PID controller parameters based AOA that mentioned in Table 5 achieve more system stability than those obtained by LUS-TLBO. Also, the fuzzy-PID controller parameters based on the AOA recover the studied system and back it to nominal operation process faster than PID controller based TLBO, DE, and AOA. Figures 25–27 show the dynamic performance of area frequencies and the exchanged tie-line power when sharing wind energy to the studied system at different instants (t = 300 s, t = 600 s). Table 15 extract the effectiveness of the fuzzy-PIDbased AOA compared to other mentioned controllers by showing the OSs and USs of both

areas frequencies and the tie-line power. The percentage improvement in US and OS with different controllers via various techniques is mentioned in Table 16.

**Figure 25.** Dynamic response comparison results of ∆*f* <sup>1</sup> for scenario 4.5.2: case B.

**Figure 26.** Dynamic response comparison results of ∆*f* <sup>2</sup> for scenario 4.5.2: case B.

**Figure 27.** Dynamic response comparison results of ∆*ptie* for scenario 4.5.2: case B.

**Table 15.** Performance evaluation of PID and Fuzzy-PID using all mentioned techniques for scenario 4.5.2: case B.


**Table 16.** Percentage improvement in US and OS for all previous mentioned controllers with different techniques based on PID controller via DE for scenario 4.5.2: case B.


### **5. Conclusions**

In this paper, there are main points that have been included, which can be summarized as follows:


There are some points will be taken in consideration in the future work and can be summarized as follows:


**Author Contributions:** Conceptualization, A.H.A.E., M.K., G.M. and S.K.; data curation, I.B.M.T.; formal analysis, A.H.A.E., M.K. and G.M.; funding acquisition, I.B.M.T. and S.K.; investigation, A.H.A.E., M.K. and G.M.; methodology, I.B.M.T. and S.K.; project administration, A.H.A.E., M.K. and G.M.; resources, I.B.M.T. and S.K.; supervision, S.K. and I.B.M.T.; validation, A.H.A.E., M.K. and G.M.; visualization, A.H.A.E., M.K. and G.M.; writing—original draft, A.H.A.E., M.K. and G.M.; writing—review and editing, I.B.M.T. and S.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors would like to acknowledge the financial support received from Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

### **Nomenclature**




### **Appendix A**

**Table A1.** Transfer functions included in the studied system.


### **References**


## *Article* **An Improved Heap-Based Optimizer for Optimal Design of a Hybrid Microgrid Considering Reliability and Availability Constraints**

**Mohammed Kharrich <sup>1</sup> , Salah Kamel 2,\* , Mohamed H. Hassan <sup>2</sup> , Salah K. ElSayed <sup>3</sup> and Ibrahim B. M. Taha <sup>3</sup>**


**Abstract:** The hybrid microgrid system is considered one of the best solution methods for many problems, such as the electricity problem in regions without electricity, to minimize pollution and the depletion of fossil sources. This study aims to propose and implement a new algorithm called improved heap-based optimizer (IHBO). The objective of minimizing the microgrid system cost is to reduce the net present cost while respecting the reliability, power availability, and renewable fraction factors of the microgrid system. The results show that the PV/diesel/battery hybrid renewable energy system (HRES) gives the best solution, with a net present cost of MAD 120463, equivalent to the energy cost of MAD 0.1384/kWh. The reliability is about 3.89%, the renewable fraction is about 95%, and the power availability is near to 99%. The optimal size considered is represented as 167.3864 m<sup>2</sup> of PV area, which is equivalent to 44.2582 kW and 3.8860 kW of diesel capacity. The study results show that the proposed optimization algorithm of IHBO is better than the artificial electric field algorithm, the grey wolf optimizer, Harris hawks optimization, and the original HBO algorithm. A comparison of the net present cost with a different fuel price is carried out, in which it is observed that the net present cost is reduced even though its quantity used is mediocre.

**Keywords:** HRES; microgrid design and sizing; optimization algorithm; HBO algorithm; reliability

## **1. Introduction**

The implementation of hybrid microgrids is necessary due to their advantages. Many projects and studies have proven their essential ecological and economic effects. The literature has assessed the microgrid from all directions, including design, operation, optimization, control, and others. Literature reviews have provided more comprehensive studies. In [1], a comprehensive study on the optimization of microgrid operations has been presented. In [2], a review of AC and DC microgrid protection has been presented. Reference [3] presented a D.C. microgrid protection comprehensive review. Reference [4] presented a review on optimization and control techniques of the hybrid AC/DC microgrid, as well as the integration challenges. Reference [5] presented a comprehensive review of the planning, the operation, and the control of a DC microgrid. Reference [6] presented a review of microgrid sizing, design, and energy management.

The design and operation optimization of microgrids, considered the main objective of this work, has been presented in many papers. Reference [7] presented a design and assessment of the microgrid using a statistical methodology that calculates the effect of energy reliability and variability on microgrid performance. The paper used a REopt platform to explore the cost savings and revenue streams. In [8,9], the microgrid design has been investigated using several algorithms and configurations. In [10], a hybrid

**Citation:** Kharrich, M.; Kamel, S.; Hassan, M.H.; ElSayed, S.K.; Taha, I.B.M. An Improved Heap-Based Optimizer for Optimal Design of a Hybrid Microgrid Considering Reliability and Availability Constraints. *Sustainability* **2021**, *13*, 10419. https://doi.org/10.3390/ su131810419

Academic Editors: Nicu Bizon, Mamadou Baïlo Camara and Bhargav Appasani

Received: 9 August 2021 Accepted: 14 September 2021 Published: 18 September 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/).

simulated annealing particle swarm (SAPS) algorithm has been presented to determine the microgrid optimal size that is subject to the economic and reliable operation constraints and to subsequently boost power supply security and stability. The paper [11] presented a new compromise method based on the Six Sigma approach to compare several multiobjective algorithms. The new approach has been applied to microgrid sizing and design based on PV, wind turbine, diesel, and battery systems. Reference [12] presented a graphtheoretic algorithm known as P-graph which allows the identification of optimal and near-optimal solutions for practical decision making. This study proposed a multi-period Pgraph optimization framework for optimizing photovoltaic-based microgrids with batteryhydrogen energy storage. The proposed approach is demonstrated through two case studies. Reference [13] proposed a novel cash-flow model for Li-ion battery storage used in the energy system; the study considers the Li-ion battery degradation characteristic.

Optimization techniques are more competent in solving non-linear optimization problems, such as optimal reactive power dispatch (ORPD) [14], economic emission dispatch [15], intelligent energy management [16], and parameter estimation of photovoltaic models [17]. Reference [18] used an experimental validation of a lab-scale microgrid. Reference [19] concerns the undervoltage in smart distribution systems. The optimal power flow from attackers has been presented in [20].

The development of tools to design microgrids has become an important research area; the development of meta-heuristic algorithms begins a trend. In the literature, many papers presented different algorithms which have been applied to design a hybrid microgrid. In [21], an improved two-archive many-objective evolutionary algorithm (TA-MaEA) based on fuzzy decisions has been used to solve the sizing optimization problem for HRES. The simulation considered the following objective function: costs, probability of loss of power supply, pollutant emissions, and power balance. Reference [22] proposes an HRES of PV and fuel cells with an optimal total annual cost; the study used a new, improved metaheuristic called the amended water strider algorithm (AWSA). The reliability is considered, and the sensitivity analysis is applied. Reference [23] presents a microgrid design composed of PV, wind, an inverter, a rectifier, an electrolyzer, and a fuel cell. The paper used a modified seagull optimization technique to find the best cost of the optimal sizing. The proposed algorithm is compared with the original seagull optimization algorithm (SOA) and modified farmland fertility algorithm (MFFA). Reference [24] presents a new hybrid algorithm called IWO/BSA to resolve the microgrid design of any configuration, including PV/wind turbine (WT)/biomass/battery, PV/biomass, PV/diesel/battery, and WT/diesel/battery systems. The study's objective is to obtain the best system with optimal cost, pollution, availability, and reliability. Reference [25] presents an adaptive version of the marine predators algorithm (AMPA) to design a PV/diesel/battery microgrid system. The objective function minimizes the annualized cost, respecting the ecologic and reliability factors of the system. The results are compared with PSO and HOMER. Reference [26] proposed an improved version of the bonobo optimizer (BO) based on the quasi-oppositional technique to resolve the design problem of the HRES considering the PV, wind turbines, battery, and diesel. A comparison between the traditional BO, the new QOBO, and other optimization techniques is investigated to prove the efficacy of QOBO. Reference [27] proposed a deterministic approach to size a PV, battery, anaerobic digestion, and biogas power plant to meet a demand load in Kenya. The levelized cost of energy (LCOE) is considered the objective function, while the energy imbalance between generation and demand is considered.

The present paper proposes a new tool consisting of platforms using an improved version of the HBO algorithm called IHBO. The improvement of the HBO algorithm depends on enhancing the performance of the HBO algorithm using the velocity equation from the particle swarm optimization (PSO) algorithm. This equation improves the convergence capability behavior and enables different diversified solutions in the search space, which is necessary for such an algorithm and achieves the fitness function's optimal value. The proposed platforms design hybrid microgrid systems composed of PV, wind, diesel,

and batteries. Two configurations are presented, and four algorithms are used in the comparison. In summary, the paper addresses the following points:


The paper is organized as follows: the introduction occurs in Section 1; the modeling of HRES components is contained in Section 2; Section 3 presents the objective functions and constraints; Section 4 presents the new, improved algorithm, namely IHBO; the results and discussion are presented in Section 5; and the conclusion is presented in Section 6.

### **2. HRES Components Modeling**

*2.1. PV Panel Modeling*

The PV output power is calculated as follows [28,29]:

$$P\_{pv} = I\langle t\rangle \times \eta\_{pv} t \times A\_{pv} \tag{1}$$

where *I* represents the irradiation, *ηpv* represents the efficiency of PV, and *Apv* is the area of PV. The efficiency of PV can be calculated based on reference efficiency (*ηr*), the efficiency of MPPT (*ηt*), temperature coefficient (*β*), ambient temperature (*Ta*), PV cell reference temperature (*Tr*) and nominal operating cell temperature (NOCT), as follows:

$$\eta\_{\rm pw}(t) = \eta\_r \times \eta\_l \times \left[1 - \beta \times (T\_d \langle t \rangle - T\_r) - \beta \times I \langle t \rangle \times \left(\frac{\text{NOCT} - 20}{800}\right) \times (1 - \eta\_r \times \eta\_l)\right] \tag{2}$$

*2.2. Wind System Modeling*

The wind turbine output power can be calculated following these conditions [30]:

$$P\_{wind} = \left\{ \begin{array}{c} 0, \ v\langle t \rangle \le v\_{\rm ci}, \ v\langle t \rangle \ge v\_{\rm co} \\\\ a \times V\langle t \rangle^3 - b \times P\_{\rm r} & v\_{\rm ci} < v\langle t \rangle < v\_{\rm r} \\\\ P\_{\rm r} & v\_{\rm r} \le v\langle t \rangle < v\_{\rm co} \end{array} \right. \tag{3}$$

where *V* represents the wind velocity, *P<sup>r</sup>* is rated power, *vci* is cut-in, *vco* represents cut-out, and *v<sup>r</sup>* is the rated wind. *a* and *b* are constant values that expressed as:

$$\begin{aligned} a &= P\_r / \left( v\_r^3 - v\_{c\dot{l}}^3 \right) \\ b &= v\_{c\dot{l}}^3 / \left( v\_r^3 - v\_{c\dot{l}}^3 \right) \end{aligned} \tag{4}$$

The rated power of wind turbine can be calculated as:

$$P\_r = \frac{1}{2} \times \rho \times A\_{wind} \times \mathbb{C}\_p \times \upsilon\_r{}^3 \tag{5}$$

where *ρ* is the air density, *Awind* represents the swept area of the wind turbine, and *C<sup>p</sup>* is the maximum power coefficient (from 0.25 to 0.45).

### *2.3. Diesel System Modeling*

The diesel rated power can be calculated as [31]:

$$P\_{d\text{g}} = \frac{F\_{d\text{g}}\langle t\rangle - A\_{\text{g}} \times P\_{d\text{g,out}}}{B\_{\text{g}}} \tag{6}$$

where *Fdg* represents the fuel consumption, *Pdg*,*out* is the output power of the diesel generator, and *A<sup>g</sup>* and *B<sup>g</sup>* are two constant values represent the fuel linear consumption.

### *2.4. Battery System Modeling*

The battery capacity of the battery can be calculated as [31]:

$$\mathcal{C}\_{BESS} = \frac{E\_l \times AD}{DOD \times \eta\_i \times \eta\_b} \tag{7}$$

where *E<sup>l</sup>* is the load demand, *AD* is the autonomy of the battery which can lead power to the load on rainy days, *DOD* represents the depth of discharge, and *η<sup>i</sup>* and *η<sup>b</sup>* represent the inverter and battery efficiency, respectively.

### **3. Objective Function and Constraints**

### *3.1. Net Present Cost*

The NPC represents an economic factor, which is considered the objective function in this study. The goal of the paper is to minimize the NPC, which is the sum of all costs during the project lifetime. The NPC is calculated as [32,33]:

$$\text{NPC} = \text{C} + \text{OM} + \text{R} + \text{FC}\_{d\text{g}} \tag{8}$$

where *C* represent the capital cost, *OM* is the operation and maintenance costs, *R* is the replacement cost, and *FCdg* is the fuel cost.

### *3.2. LCOE Index*

The LCOE represents the price of energy and is a critical factor which is calculated as [31]:

$$\text{LCOE} = \frac{\text{NPC} \times \text{CRF}}{\sum\_{t=1}^{8760} P\_{load} t} \tag{9}$$

where CRF represents the capital recovery factor (obtained by converting the initial cost to annual capital cost), and *Pload* represents the power load. The CRF is calculated as:

$$\text{CRF}(ir, N) = \frac{i\_r \times (1 + i\_r)^N}{\left(1 + i\_r\right)^R - 1} \tag{10}$$

### *3.3. LPSP Index*

The loss of power supply probability (LPSP) is a technical index that ranges from 0 to 1. It is used to indicate the reliability of the microgrid system. The LPSP is calculated as follows [31]:

$$\text{LPSP} = \frac{\sum\_{t=1}^{8760} \left( P\_{load} \langle t \rangle - P\_{pv} \langle t \rangle - P\_{wind} \langle t \rangle + P\_{\text{dg},out} \langle t \rangle + E\_{bmin} \right)}{\sum\_{t=1}^{8760} P\_{load} \langle t \rangle} \tag{11}$$

### *3.4. Renewable Energy Index*

Renewable energy (RF) is calculated to determine the renewable energy percent that is penetrated into the microgrid system. The RF is expressed as [31]:

$$\text{RF} = \left( 1 - \frac{\sum\_{t=1}^{8760} P\_{\text{g},out} \langle t \rangle}{\sum\_{t=1}^{8760} P\_{r\varepsilon} \langle t \rangle} \right) \times 100 \tag{12}$$

where *Pre* represents the sum of renewable energy powers.

### *3.5. Availability Index*

The availability factor (Av) is assumed as an index of the customer's satisfaction; it measures the ability of the microgrid to convert the total energy to load charge. The availability is calculated as [33]:

$$\text{Av} = 1 - \frac{\text{DMN}}{\sum\_{t=1}^{8760} P\_{load} \langle t \rangle} \tag{13}$$

$$DMN = P\_{bmin} \langle t \rangle - P\_b \langle t \rangle - \left( P\_{pv} \langle t \rangle + P\_{wind} \langle t \rangle + P\_{dg,out} \langle t \rangle - P\_{load} \langle t \rangle \right) \times u \langle t \rangle \tag{14}$$

where *Pbmin* represents the battery min state, *P<sup>b</sup>* represents the battery power, and *u* is a fixed value which equals 1 when the load is not satisfied and which equals 0 otherwise.

### *3.6. Constraints*

Constraints are introduced to tune the microgrid system factors and help to improve the microgrid service quality. In this work, the constraints proposed are:

$$\begin{cases} \quad 0 \le A\_{pv} \le A\_{pv}^{max} \\ \quad 0 \le A\_{wind} \le A\_{wind}^{max} \\ \quad 0 \le P\_{dg} \le P\_{dg}^{max} \\ \quad 0 \le \mathcal{C}\_{BESS} \le \mathcal{C}\_{BESS}^{max} \\ \quad LPSP \le \mathcal{L} \text{PSP}^{max} \\ \quad \mathbb{R}\mathcal{F}^{min} \le RF \\ \quad Av^{min} \le Av \\ \quad AD^{min} \le AD \end{cases} \tag{15}$$

where LPSP*max* = 0.05, RF*min* = 70%, Av*min* = 90%, and AD*min* = 1 day. The sizing limit is different from configuration to the others. All other parameters are shown in Table A1.

### **4. Proposed Algorithm**

### *4.1. Heap-Based Optimizer (HBO)*

The heap-based optimizer algorithm (HBO) is inspired by the social behavior of human beings [34]. One sort of social interaction between human beings can be observed in organizations where people in teams are arranged in a hierarchy for achieving a specific target; this is known as corporate rank hierarchy (CRH). CRH is presented in Figure 1a. The HBO algorithm is based on CRH in a very distinctive manner. In this regard, the concept of CRH is to arrange the search agents based on their suitability in this hierarchy using a heap tree-based data structure to enact the implementation of priority queues. Figure 1b shows an example of 3 degrees (3-ary) of min-heap. Three types of employees' behaviors were used in the HBO algorithm. These types are: (i) the interaction of subordinates with their immediate head; (ii) the interaction between co-workers; and (iii) the self-contribution of individuals.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 6 of 26

**Figure 1.** Partial examples of corporate rank hierarchy (**a**) and 3-ary min-heap (**b**). **Figure 1.** Partial examples of corporate rank hierarchy (**a**) and 3-ary min-heap (**b**). **Figure 1.** Partial examples of corporate rank hierarchy (**a**) and 3-ary min-heap (**b**).

The mapping of the heap concept is divided into four steps: The mapping of the heap concept is divided into four steps: The mapping of the heap concept is divided into four steps:

*A. Modeling the corporate rank hierarchy A. Modeling the corporate rank hierarchy A. Modeling the corporate rank hierarchy* 

Figure 2 displays the procedure of CRH modeling through a heap data structure, wherein xi is the ith search agent of the population. The curve in the objective space describes the shape of the supposed objective function, and the search agents are drawn on the fitness shape according to their convenience. Figure 2 displays the procedure of CRH modeling through a heap data structure, wherein xi is the ith search agent of the population. The curve in the objective space describes the shape of the supposed objective function, and the search agents are drawn on the fitness shape according to their convenience. Figure 2 displays the procedure of CRH modeling through a heap data structure, wherein xi is the ith search agent of the population. The curve in the objective space describes the shape of the supposed objective function, and the search agents are drawn on the fitness shape according to their convenience.

**Figure 2***.* An illustration of the modeling of the CRH with min-heap. **Figure 2***.* An illustration of the modeling of the CRH with min-heap. **Figure 2.** An illustration of the modeling of the CRH with min-heap.

#### *B. Mathematically modeling the collaboration with the boss*  In a centralized organizational structure, the regulations and policies are enforced *B. Mathematically modeling the collaboration with the boss B. Mathematically modeling the collaboration with the boss*

from the upper levels, and subordinates must follow their direct manager. This can be mathematically described by updating the agent position of each search In a centralized organizational structure, the regulations and policies are enforced from the upper levels, and subordinates must follow their direct manager. In a centralized organizational structure, the regulations and policies are enforced from the upper levels, and subordinates must follow their direct manager.

as follows: This can be mathematically described by updating the agent position of each search as follows: This can be mathematically described by updating the agent position of each search as follows:

$$\mathbf{x}\_i^k(t+1) = \mathcal{B}^k + \gamma \lambda^k \left| \mathcal{B}^k - \mathbf{x}\_i^k(t) \right| \tag{16}$$

$$\gamma = \left| 2 - \frac{\left( t \bmod \frac{T}{C} \right)}{\frac{T}{4C}} \right| \tag{17}$$

$$
\dot{\lambda}^k = (2r - 1) \tag{18}
$$

 = (2 − 1) (18) where *t* is the current iteration, *k* is the *k*th component of a vector, *B* denotes the parent node, *r* is a random number from the range [0, 1], *T* is the maximum number of iterations, and *C* represents a user-defined parameter.

### *C. Mathematically modeling the interaction between the colleagues*

Colleagues cooperate and perform official tasks. It is assumed in a heap that the nodes at the same level are colleagues, and each search agent *x<sup>i</sup>* updates its location based on its randomly selected colleague *S<sup>r</sup>* as follows:

$$\mathbf{x}\_{i}^{k}(t+1) = \begin{cases} s\_{r}^{k} + \gamma \lambda^{k} \Big| s\_{r}^{k} - \mathbf{x}\_{i}^{k}(t) \Big|\_{\prime} & f(\mathcal{S}\_{r}) < f(\mathbf{x}\_{i}(t)) \\\ \mathbf{x}\_{i}^{k} + \gamma \lambda^{k} \Big| s\_{r}^{k} - \mathbf{x}\_{i}^{k}(t) \Big|\_{\prime} & f(\mathcal{S}\_{r}) \ge f(\mathbf{x}\_{i}(t)) \end{cases} \tag{19}$$

### *D. Self-contribution of an employee to accomplish a task*

In this phase, the self-contribution of a worker is mapped as follows:

$$
\mathfrak{x}\_i^k(t+1) = \mathfrak{x}\_i^k(t) \tag{20}
$$

The following part explains how exploration can be controlled with this equation.

### *E. putting all together*

The principal challenge is determining the selection probabilities for the three equations to balance exploration and exploitation. The purpose of the roulette wheel is to achieve a balance of possibilities. The roulette wheel is divided into three parts: *p*1, *p*2, and *p*3. The value of *p*<sup>1</sup> makes a population changes their position, and it is calculated from the following equation:

$$p\_1 = 1 - \frac{t}{T} \tag{21}$$

The selection of *p*<sup>2</sup> is computed from the following equation:

$$p\_2 = p\_1 - \frac{1 - p\_1}{2} \tag{22}$$

Finally, the selection of *p*<sup>3</sup> is calculated as follows:

$$p\_3 = p\_2 - \frac{1 - p\_1}{2} = 1\tag{23}$$

Accordingly, a general position-updating mechanism of the HBO algorithm is mathematically represented as follows:

$$\mathbf{x}\_{i}^{k}(t+1) = \begin{cases} \mathbf{x}\_{i}^{k}(t), p \le p\_{1} \\ \mathbf{B}^{k} + \gamma \lambda^{k} \Big| \mathbf{B}^{k} - \mathbf{x}\_{i}^{k}(t) \Big|, p > p\_{1} \text{ and } p \le p\_{2} \\ \mathbf{s}\_{r}^{k} + \gamma \lambda^{k} \Big| \mathbf{s}\_{r}^{k} - \mathbf{x}\_{i}^{k}(t), p > p\_{2} \text{ and } p \le p\_{3} \text{ and } f(\mathbf{S}\_{r}) < f(\mathbf{x}\_{i}(t)) \\ \mathbf{x}\_{i}^{k} + \gamma \lambda^{k} \Big| \mathbf{s}\_{r}^{k} - \mathbf{x}\_{i}^{k}(t) \Big|, p > p\_{2} \text{ and } p \le p\_{3} \text{ and } f(\mathbf{S}\_{r}) \ge f(\mathbf{x}\_{i}(t)) \end{cases} \tag{24}$$

where *p* is a random number in the range (0, 1).

### *4.2. Improved Heap-Based Optimizer(IHBO)*

In order to enhance the strength of the proposed IHBO algorithm for many highdimensional optimization problems, core aspects of one of the most used meta-heuristic algorithms, PSO, are utilized. The PSO algorithm is introduced by [35]. The velocity equation from the PSO algorithm is used in the proposed IHBO algorithm. This modification leads to the improvement of the ability of the global search and enhances the local search capabilities of the improved algorithm. This core equation is as follows:

$$V\_i^k(t+1) = w \cdot V\_i^k(t) + \mathsf{C}\_1 \cdot r\_1 \times (p\_{\mathrm{best}} - x\_i^k(t)) + \mathsf{C}\_2 \cdot r\_2 \times (g\_{\mathrm{best}} - x\_i^k(t))\tag{25}$$

$$\mathbf{x}\_i^k(t+1) = \mathbf{x}\_i^k(t) + V\_i^k(t+1) \tag{26}$$

where *C*<sup>1</sup> = *C*<sup>2</sup> = 0.5, as these values gave the best solution in [36]; *w* = 0.7; *r*<sup>1</sup> and *r*<sup>2</sup> are a random number in the range (0, 1); *pbest* is the best solution of an individual population, and *gbest* is the best solution so far. where ଵ = ଶ= 0.5, as these values gave the best solution in [36]; = 0.7; ଵ and ଶ are a random number in the range (0, 1); ௦௧ is the best solution of an individual population, and ௦௧ is the best solution so far.

The flow chart of the proposed IHBO algorithm is shown in Figure 3. The flow chart of the proposed IHBO algorithm is shown in Figure 3.

**Figure 3.** Flowchart of the proposed IHBO technique. **Figure 3.** Flowchart of the proposed IHBO technique.

Performance of the Proposed IHBO Algorithm The proposed IHBO algorithm's efficiency and performance are evaluated on differ-

Performance of the Proposed IHBO Algorithm

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 9 of 26

The proposed IHBO algorithm's efficiency and performance are evaluated on different benchmark functions, including statistical measurements, such as minimum values, mean values, maximum values, and standard deviation (STD) for best solutions obtained by the proposed IHBO algorithm and the other recent optimization algorithms. The results obtained with the proposed IHBO technique is compared with three well-known optimization algorithms, including the sine cosine algorithm (SCA) [37], salp swarm algorithm (SSA) [38], movable damped wave algorithm (MDWA) [39], and the original heap-based optimizer (HBO). Table 1 shows the parameters of all compared algorithms (SSA, MDWA, SCA, IHBO, and HBO). Qualitative metrics on F1, F4, F7, F9, F11, F12, F15, and F18, including 2D views of the functions, search history, average fitness history, and convergence curve, are presented in Figure 4. ent benchmark functions, including statistical measurements, such as minimum values, mean values, maximum values, and standard deviation (STD) for best solutions obtained by the proposed IHBO algorithm and the other recent optimization algorithms. The results obtained with the proposed IHBO technique is compared with three well-known optimization algorithms, including the sine cosine algorithm (SCA) [37], salp swarm algorithm (SSA) [38], movable damped wave algorithm (MDWA) [39], and the original heap-based optimizer (HBO). Table 1 shows the parameters of all compared algorithms (SSA, MDWA, SCA, IHBO, and HBO). Qualitative metrics on F1, F4, F7, F9, F11, F12, F15, and F18, including 2D views of the functions, search history, average fitness history, and convergence curve, are presented in Figure 4.

**Figure 4.** *Cont*.

**Figure 4.** Qualitative metrics on F1, F4, F7, F9, F11, F12, F15 and F18: 2D views of the functions, search history, average fitness history, and convergence curve. **Figure 4.** Qualitative metrics on F1, F4, F7, F9, F11, F12, F15 and F18: 2D views of the functions, search history, average fitness history, and convergence curve.



HBO sv = 100; degree = 3 Tables 2–4 tabulate the statistical results of the proposed IHBO algorithm and other well-known algorithms when applied for unimodal benchmark functions, named F1 to F7, multimodal benchmark functions, named F8 to F13, and composite benchmark functions, named F14 to F23, respectively. The best values, shown in bold, were achieved with the proposed IHBO algorithm, as well as MDWA and SCA, but the proposed IHBO technique achieves the best results for most of the benchmark functions. The convergence curves of all algorithms for the unimodal benchmark functions are shown in Figure 5 while Figure 6 shows the boxplots of each algorithm for these unimodal benchmark functions. Figure 7 displays the convergence characteristics curves of all algorithms for the multi-modal benchmark functions. The boxplots for each algorithm for these types of benchmark functions are presented in Figure 8. The convergence curves of all algorithms for the composite benchmark functions are displayed in Figure 9 while Figure 10 illus-Tables 2–4 tabulate the statistical results of the proposed IHBO algorithm and other well-known algorithms when applied for unimodal benchmark functions, named F1 to F7, multimodal benchmark functions, named F8 to F13, and composite benchmark functions, named F14 to F23, respectively. The best values, shown in bold, were achieved with the proposed IHBO algorithm, as well as MDWA and SCA, but the proposed IHBO technique achieves the best results for most of the benchmark functions. The convergence curves of all algorithms for the unimodal benchmark functions are shown in Figure 5 while Figure 6 shows the boxplots of each algorithm for these unimodal benchmark functions. Figure 7 displays the convergence characteristics curves of all algorithms for the multimodal benchmark functions. The boxplots for each algorithm for these types of benchmark functions are presented in Figure 8. The convergence curves of all algorithms for the composite benchmark functions are displayed in Figure 9 while Figure 10 illustrates the boxplots for each algorithm for these benchmark functions. The proposed algorithm reached a stable point for all functions. Also, the boxplots of the proposed IHBO technique are very narrow for most functions compared to the other algorithms.



The best values obtained are in bold.


**Function HBO IHBO SCA MDWA SSA** F13 Best **1.35** <sup>×</sup> **<sup>10</sup>**−**<sup>32</sup> 1.35** <sup>×</sup> **<sup>10</sup>**−**<sup>32</sup>** 5.71 <sup>×</sup> <sup>10</sup>−<sup>02</sup> 2.48 <sup>×</sup> <sup>10</sup>−<sup>06</sup> 1.51 <sup>×</sup> <sup>10</sup>−<sup>11</sup> Worst 1.84 <sup>×</sup> <sup>10</sup>−<sup>32</sup> **1.35** <sup>×</sup> **<sup>10</sup>**−**<sup>32</sup>** 3.59 <sup>×</sup> <sup>10</sup>−<sup>01</sup> 3.70 <sup>×</sup> <sup>10</sup>−<sup>04</sup> 1.10 <sup>×</sup> <sup>10</sup>−<sup>02</sup> Mean 1.37 <sup>×</sup> <sup>10</sup>−<sup>32</sup> **1.35** <sup>×</sup> **<sup>10</sup>**−**<sup>32</sup>** 2.16 <sup>×</sup> <sup>10</sup>−<sup>01</sup> 4.53 <sup>×</sup> <sup>10</sup>−<sup>05</sup> 1.65 <sup>×</sup> <sup>10</sup>−<sup>03</sup> std 1.10 <sup>×</sup> <sup>10</sup>−<sup>33</sup> **2.81** <sup>×</sup> **<sup>10</sup>**−**<sup>48</sup>** 7.48 <sup>×</sup> <sup>10</sup>−<sup>02</sup> 8.66 <sup>×</sup> <sup>10</sup>−<sup>05</sup> 4.03 <sup>×</sup> <sup>10</sup>−<sup>03</sup> The best values obtained are in bold.

**Table 3.** *Cont.*

**Table 4.** Results of composite benchmark functions.


The best values obtained are in bold.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 11 of 26

trates the boxplots for each algorithm for these benchmark functions. The proposed algorithm reached a stable point for all functions. Also, the boxplots of the proposed IHBO

trates the boxplots for each algorithm for these benchmark functions. The proposed algorithm reached a stable point for all functions. Also, the boxplots of the proposed IHBO

technique are very narrow for most functions compared to the other algorithms.

**Figure 5.** The convergence curves of all algorithms for unimodal benchmark functions (**a**) F1, (**b**) F2, (**c**) F3, (**d**) F4, (**e**) F5, (**f**) F6, and (**g**) F7. **Figure 5.** The convergence curves of all algorithms for unimodal benchmark functions (**a**) F1, (**b**) F2, (**c**) F3, (**d**) F4, (**e**) F5, (**f**) F6, and (**g**) F7. **Figure 5.** The convergence curves of all algorithms for unimodal benchmark functions (**a**) F1, (**b**) F2, (**c**) F3, (**d**) F4, (**e**) F5, (**f**) F6, and (**g**) F7.

**Figure 6.** Boxplots for all algorithms for unimodal benchmark functions (**a**) F1, (**b**) F2, (**c**) F3, (**d**) F4, (**e**) F5, (**f**) F6, and (**g**) F7.

F7.

F7.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 12 of 26

**Figure 7.** The convergence curves of all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**) F11, (**e**) F12 and (**f**) F13. **Figure 7.** The convergence curves of all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**) F11, (**e**) F12 and (**f**) F13. **Figure 7.** The convergence curves of all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**) F11, (**e**) F12 and (**f**) F13.

**Figure 8.** Boxplots for all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**) F11, (**e**) F12 and (**f**) F13.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 13 of 26

F11, (**e**) F12 and (**f**) F13.

F11, (**e**) F12 and (**f**) F13.

**Figure 8.** Boxplots for all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**)

**Figure 8.** Boxplots for all algorithms for multi-modal benchmark functions (**a**) F8, (**b**) F9, (**c**) F10, (**d**)

**Figure 9.** The convergence curves of all algorithms for composite benchmark functions (**a**) F14, (**b**) F15, (**c**) F16, (**d**) F17, (**e**) F18, (**f**) F19, (**g**) F20, (**h**) F21, (**i**) F22, and (**j**) F23. **Figure 9.** The convergence curves of all algorithms for composite benchmark functions (**a**) F14, (**b**) F15, (**c**) F16, (**d**) F17, (**e**) F18, (**f**) F19, (**g**) F20, (**h**) F21, (**i**) F22, and (**j**) F23. **Figure 9.** The convergence curves of all algorithms for composite benchmark functions (**a**) F14, (**b**) F15, (**c**) F16, (**d**) F17, (**e**) F18, (**f**) F19, (**g**) F20, (**h**) F21, (**i**) F22, and (**j**) F23.

**Figure 10.** *Cont*.

#### **Table 2.** Results of unimodal benchmark functions. **5. Project Implementation Location**

**Function HBO IHBO SCA MDWA SSA**  Best 8.81 × 10−<sup>65</sup> **3.11 × 10−<sup>86</sup>** 5.61 × 10−41 1.34 × 10−44 2.21 × 10−<sup>10</sup> The project was implemented in a small region in the west of Morocco called Terfaya, at coordinating latitude 27.932 and longitude −12.935.

#### F1 Worst 7.28 × 10−<sup>59</sup> **3.81 × 10−<sup>81</sup>** 1.24 × 10−28 3 × 10−39 8.48 × 10−<sup>10</sup> **6. Results and Discussion**

Mean 4 × 10−<sup>60</sup> **3.72 × 10−<sup>82</sup>** 8.72 × 10−30 3.01 × 10−40 5.63 × 10−<sup>10</sup> std 1.63 × 10−<sup>59</sup> **8.56 × 10−<sup>82</sup>** 2.92 × 10−29 8.19 × 10−40 1.99 × 10−<sup>10</sup> F2 Best 1.24 × 10−<sup>39</sup> **4.59 × 10−<sup>53</sup>** 6.81 × 10−27 1.14 × 10−22 2.69 × 10−<sup>06</sup> Worst 4.86 × 10−<sup>37</sup> **1.2 × 10−<sup>49</sup>** 1.84 × 10−19 2.86 × 10−22 1.54 × 10−<sup>05</sup> In this paper, the Terfaya region of Morocco is selected as the case study to implement an HRES platform based on an improved optimization algorithm called IHBO. The maps for the project location, the load charge, the annual ambient radiation, temperature, wind speed, and pressure are presented in Figures 11–15, respectively. *Sustainability* **2021**, *13*, x FOR PEER REVIEW 17 of 26

Mean 1.23 × 10−<sup>33</sup> **0.00** 3.12 × 10−01 2.93 × 10−05 5.74 × 10−<sup>10</sup> std 3.79 × 10−<sup>33</sup> **0.00** 1.35 × 10−01 2.67 × 10−05 1.60 × 10−<sup>10</sup> **Figure 11.** Load power. **Figure 11.** Load power.

**Figure 12.** Solar radiation.

**Figure 12. Figure 12.** Solar radiation. Solar radiation.

**Figure 11.** Load power.

**Figure 13.** Temperature. **Figure 13.** Temperature.

**Figure 14. Figure 14.** Wind speed. Wind speed.

**Figure 14.** Wind speed.

**Figure 15.** Pressure. **Figure 15.** Pressure.

The proposed HRES includes two renewable sources (PV and wind turbines), a diesel generator, and a battery storage system. According to the mathematical modeling of the mentioned systems, the PV output can be affected by the solar radiation data; otherwise, the output power of the wind is influenced by the wind speed data. The decision variables in this study are dedicated to the size of the HRES where: x(1) is the PV area (௩,), x(2) is the wind swept area (௪ௗ), x(3) represents the battery capacity (ாௌௌ) and x(4) is the rated power of the diesel generator (ௗ). In this paper, an analysis of fuel price variation is carried out. The proposed HRES includes two renewable sources (PV and wind turbines), a diesel generator, and a battery storage system. According to the mathematical modeling of the mentioned systems, the PV output can be affected by the solar radiation data; otherwise, the output power of the wind is influenced by the wind speed data. The decision variables in this study are dedicated to the size of the HRES where: x(1) is the PV area (*Apv*,), x(2) is the wind swept area (*Awind*), x(3) represents the battery capacity (*CBESS*) and x(4) is the rated power of the diesel generator (*Pdg*). In this paper, an analysis of fuel price variation is carried out.

### *6.1. Optimal HRES Design of PV/Diesel/Battery and PV/Wind/Diesel/Battery*  6.1.1. PV/Diesel/Battery HRES *6.1. Optimal HRES Design of PV/Diesel/Battery and PV/Wind/Diesel/Battery* 6.1.1. PV/Diesel/Battery HRES

The results of the optimal HRES design for the case study concerning the PV/diesel/battery HRES are summarized in Table 5. The table presents all used algorithms concerning the predefined constraints, including the LPSP, RF, and the availability. The algorithms are arranged as GWO, HBO, AEFA, HHO, and IHBO, with a net present cost of MAD 191,661, MAD 175,321, MAD 169,142, MAD 147,527, and MAD 120,463, respectively. The optimal system needs MAD 120,463, equivalent to an LCOE of MAD 0.13/kWh. The system designed respected the constraints very well, with a reliability (LPSP) of 3%, a renewable fraction of 95%, and power availability of 98%. Table 6 presents the optimal size of each algorithm; the best solution is then obtained by IHBO, with 1,673,864 m2 and 38,860 kW of diesel generator capacity. Table 7 presents the convergence time of all simu-The results of the optimal HRES design for the case study concerning the PV/diesel/ battery HRES are summarized in Table 5. The table presents all used algorithms concerning the predefined constraints, including the LPSP, RF, and the availability. The algorithms are arranged as GWO, HBO, AEFA, HHO, and IHBO, with a net present cost of MAD 191,661, MAD 175,321, MAD 169,142, MAD 147,527, and MAD 120,463, respectively. The optimal system needs MAD 120,463, equivalent to an LCOE of MAD 0.13/kWh. The system designed respected the constraints very well, with a reliability (LPSP) of 3%, a renewable fraction of 95%, and power availability of 98%. Table 6 presents the optimal size of each algorithm; the best solution is then obtained by IHBO, with 1,673,864 m<sup>2</sup> and 38,860 kW of diesel generator capacity. Table 7 presents the convergence time of all simulations.


HHO 147527 0.1187 0.0269 99.7289 98.5725 HBO 175321 0.1411 0.0383 95.5051 98.9342 IHBO 120463 0.1384 0.0389 95.3802 98.8665

lations. **Table 5.** Results of the PV/diesel/battery HRES.


**Table 6.** Sizing results of the PV/diesel/battery HRES.

**Table 7.** Convergence time of algorithms. IHBO 167.4 0 3.88


The convergence curve results for all scenarios are presented in Figure 16, in which the IHBO proves its efficacy to reach the optimal solution. IHBO 14,017

**Figure 16.** PV/diesel/battery. **Figure 16.** PV/diesel/battery.

6.1.2. PV/Wind/Diesel/Battery HRES

6.1.2. PV/Wind/Diesel/Battery HRES The second configuration used in this paper concerns the PV/wind/diesel/battery HRES. From Table 8, the results respect the constraints; then, the best algorithms results The second configuration used in this paper concerns the PV/wind/diesel/battery HRES. From Table 8, the results respect the constraints; then, the best algorithms results converge as HBO, GWO, AEFA, HHO, and IHBO, with an investment cost of MAD 461,233, MAD 226,559, MAD 221,694, MAD 215,371, and MAD 100,337, respectively. The best cost

1.0762 kW of diesel. Table 10 presents the convergence time of all simulations.

HRES; the curve shows that the IHBO algorithm gives better convergence results.

converge as HBO, GWO, AEFA, HHO, and IHBO, with an investment cost of MAD 461,233, MAD 226,559, MAD 221,694, MAD 215,371, and MAD 100,337, respectively. The

about 4%, the renewable fraction is near 100%, and the power availability is more than 99%. Table 9 presents the size results, which show that the best project needs 261.3031 m2 of PV area, 102.7114 m2 of swept area of the wind turbines, 23.2177 kWh of battery, and

Figure 17 presents the convergence curve of the NPC for the PV/wind/diesel/battery

needs MAD 100,337, equivalent to MAD 0.08/kWh; in this situation, the LPSP is about 4%, the renewable fraction is near 100%, and the power availability is more than 99%. Table 9 presents the size results, which show that the best project needs 261.3031 m<sup>2</sup> of PV area, 102.7114 m<sup>2</sup> of swept area of the wind turbines, 23.2177 kWh of battery, and 1.0762 kW of diesel. Table 10 presents the convergence time of all simulations.


**Table 8.** Results of the PV/wind/diesel/battery HRES.

**Table 9.** Sizing results of the PV/wind/diesel/battery HRES.


**Table 10.** Convergence time of the algorithms.


Figure 17 presents the convergence curve of the NPC for the PV/wind/diesel/battery HRES; the curve shows that the IHBO algorithm gives better convergence results.

### *6.2. Impact of Fuel Price Variation*

In the paper, if we suppose that the price of fuel is about MAD 0.41/L, then we can compare the total investment cost with the previous study that used the actual price, which is MAD 0.96/L.

From Table 11, it is clearly shown that the NPC of the HRES is reduced strongly while it is passed from MAD 120,463. Table 12 presents the optimal HRES size using all optimization algorithms. Figure 18 presents the convergence curve of the NPC for the PV/diesel/battery HRES, with a fuel price of MAD 0.54/L. This figure shows that the IHBO algorithm gives the better convergence results.

**Figure 17.** Convergence curve for the PV/wind/diesel/battery HRES. **Figure 17.** Convergence curve for the PV/wind/diesel/battery HRES.


**Table 8.** Results of the PV/wind/diesel/battery HRES.

**Table 9.** Sizing results of the PV/wind/diesel/battery HRES.

**(MAD/kWh) LPSP RF (%) Availability** 

AEFA 221694 0.1784 0.0440 99.4362 98.7671 GWO 226559 0.1823 0.0233 98.6033 99.5884 HHO 215371 0.1733 0.0076 99.9540 99.6287 HBO 461233 0.3712 0.0030 99.8870 99.9540 IHBO 165999 0.1336 0.0445 99.9133 99.0391

**Algorithm PV (m2) Wind (m2) Battery (kWh) Diesel (kW)**  AEFA 92.9 659.8 0.14 4.83 GWO 143.9 174 14.64 7.24 HHO 317.8 232.6 3.59 1.47 HBO 403.9 652.8 2.35 10.6 IHBO 261.3 102.7 23.2 1

**Algorithm Convergence Time (s)** 

AEFA 1176 GWO 1079 HHO 3931 HBO 680 IHBO 4942 **(%)** 

**Algorithm NPC (MAD) LCOE** 

**Table 10.** Convergence time of the algorithms.


**Table 12.** Sizing results of the PV/diesel/battery HRES.


In the paper, if we suppose that the price of fuel is about MAD 0.41/L, then we can compare the total investment cost with the previous study that used the actual price,

From Table 11, it is clearly shown that the NPC of the HRES is reduced strongly while it is passed from MAD 120,463. Table 12 presents the optimal HRES size using all optimization algorithms. Figure 18 presents the convergence curve of the NPC for the PV/diesel/battery HRES, with a fuel price of MAD 0.54/L. This figure shows that the IHBO algo-

AEFA 166303 0.1339 0.0324 96.1598 99.4268 GWO 107532 0.0865 0.0483 97.0846 98.0032 HHO 87394 0.0703 0.0496 99.6090 96.1823 HBO 125791 0.1012 0.0296 97.7468 98.8978 IHBO 68121 0.0548 0.1119 99.9999 88.8055

**Algorithm PV (m2) Battery (kWh) Diesel (kW)**  AEFA 216.7 3.7 6 GWO 163.7 0.4 2.24 HHO 161.8 0.7 0.29 HBO 194 0 2.79 IHBO 124.9 0 0

**(MAD/kWh) LPSP RF (%) Availability** 

**(%)** 

**Figure 18.** Convergence PV/diesel/battery with a fuel price of 0.54. **Figure 18.** Convergence PV/diesel/battery with a fuel price of 0.54.

### **7. Conclusions**

*6.2. Impact of Fuel Price Variation* 

rithm gives the better convergence results.

**Algorithm NPC (MAD) LCOE** 

**Table 12.** Sizing results of the PV/diesel/battery HRES.

**Table 11.** Results of the PV/diesel/battery HRES with fuel prices.

which is MAD 0.96/L.

This paper proposed a platform to design an HRES microgrid system based on two configurations, PV/diesel/battery, and PV/wind/diesel/battery. The platform is based on modeling, power management, and a cost optimization study using an improved IHBO algorithm. The proposed IHBO algorithm proved its efficacy in finding the optimal solution compared with many algorithms, including AEFA, GWO, HHO, and the original HBO. In the paper, we discussed the case of reducing fuel prices and its impact on the investment cost. The results show that the NPC is highly reduced when the use of diesel is small. Several systems, such as hydrogen storage and biomass systems, can be integrated in the microgrid. Future work will focus on developing configurations considering the degradation of battery characteristics.

**Author Contributions:** Conceptualization, M.K. and S.K.; methodology, S.K.E., M.H.H. and I.B.M.T.; software, M.K., S.K. and M.H.H.; validation, S.K.E. and I.B.M.T.; formal analysis, M.K., S.K. and M.H.H.; investigation, S.K.E. and I.B.M.T.; resources, M.K., S.K. and M.H.H.; data curation, S.K.E. and I.B.M.T.; writing original draft preparation, M.K., S.K. and M.H.H.; writing on characteristic, S.K.E. and I.B.M.T.; visualization, S.K.E. and I.B.M.T.; supervision, S.K.; project administration, M.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was funded by Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors would like to acknowledge the financial support received from Taif University Researchers Supporting Project Number (TURSP-2020/61), Taif University, Taif, Saudi Arabia.

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

### **Nomenclature**



### **Appendix A**

**Table A1.** Summary of the HRES parameters.


### **References**


## *Article* **Design, Modeling, and Differential Flatness Based Control of Permanent Magnet-Assisted Synchronous Reluctance Motor for e-Vehicle Applications**

**Songklod Sriprang 1,2 , Nitchamon Poonnoy 2,\*, Damien Guilbert <sup>1</sup> , Babak Nahid-Mobarakeh <sup>3</sup> , Noureddine Takorabet <sup>1</sup> , Nicu Bizon <sup>4</sup> and Phatiphat Thounthong 2,\***


**Abstract:** This paper presents the utilization of differential flatness techniques from nonlinear control theory to permanent magnet assisted (PMa) synchronous reluctance motor (SynRM). The significant advantage of the proposed control approach is the potentiality to establish the behavior of the state variable system during the steady-state and transients operations as well. The mathematical models of PMa-SynRM are initially proved by the nonlinear case to show the flatness property. Then, the intelligent proportional-integral (iPI) is utilized as a control law to deal with some inevitable modeling errors and uncertainties for the torque and speed of the motor. Finally, a MicroLab Box dSPACE has been employed to implement the proposed control scheme. A small-scale test bench 1-KW relying on the PMa-SynRM has been designed and developed in the laboratory to approve the proposed control algorithm. The experimental results reflect that the proposed control effectively performs high performance during dynamic operating conditions for the inner torque loop control and outer speed loop control of the motor drive compared to the traditional PI control.

**Keywords:** electric vehicle; inverter; permanent magnet assisted synchronous reluctance motor; differential flatness-based control; parameter observers; traction drive

### **1. Introduction**

Permanent magnet synchronous motors (PMSMs) are the most widespread motor technologies in transportation applications including more electric aircraft (MEA), electric vehicles (EVs or e-vehicle), and hybrid electric vehicles (HEVs) [1–5]. Indeed, this technology enables offering high torque, power density, and high efficiency; while providing an extensive speed range. Besides, due to their design, they are extremely versatile and can also be employed for low-power applications, offering high performance [1]. On the other side, these motors require the use of rare-earth metals to make permanent magnets (PMs) such as Nd-Fe-B (neodymium-iron-boron), located on the rotor. Due to the growing development of electric vehicles, the interest in rare-earth metals has been increasing; leading up consequently to high cost and environmental consequences for the extraction and refining of rare-earth elements.

As a result, to cope with these important issues, a new permanent magnet-assisted (PMa) synchronous reluctance motor (SynRM) has been conceived to reduce the size of PMs

**Citation:** Sriprang, S.; Poonnoy, N.; Guilbert, D.; Nahid-Mobarakeh, B.; Takorabet, N.; Bizon, N.; Thounthong, P. Design, Modeling, and Differential Flatness Based Control of Permanent Magnet-Assisted Synchronous Reluctance Motor for e-Vehicle Applications. *Sustainability* **2021**, *13*, 9502. https://doi.org/10.3390/ su13179502

Academic Editors: Marc A. Rosen and Lin Li

Received: 29 June 2021 Accepted: 18 August 2021 Published: 24 August 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/).

and to increase the use of ferrite magnet materials in the rotor part. Besides, the modern PMa-SynRM is more advantageous than the classic SynRM [6–8]. Its cost is reduced compared to the usual PMSM since ferrite magnets are cost-effective over rare-earth PMs. In summary, the PMa-SynRM is an emerging and attractive motor for the dissemination of the next generations of electric cars. The constitution of the proposed four-pole PMa-SynRM prototype relying on ferrite magnets, and the d-axis and q-axis are shown in Figure 1. It can be noted that the permanent magnets are positioned in the flux barriers of the rotor part. Hence, magnetization takes place along the negative q-axis. Given that the PMa-SynRM is relatively new, the achievement of high performance for large functioning conditions by controlling it remains a technical barrier. Indeed, its nonlinear features, parameter mismatch, and also parametric uncertainty make its control challenging.

**Figure 1.** The rotor's structure of the proposed PMa-SynRM.

To face this technical barrier from the control point of view, various control approaches have been reported and analyzed recently in the literature. First and foremost, Ion Boldea et al. [9] have studied a direct torque and flux control with space vector modulation (DTFC-SVM) drive control of PM-assisted reluctance synchronous motor/generator employed for mild hybrid vehicles applications. Peyman Niazi et al. [10] have conceived a maximum torque per ampere (MTPA) control strategy coupled with a parameter observer applied to a PMa-SynRM to face the change of motor parameters (inductances and PM flux density) and saturation effect due to the internal temperature. However, the tests have been performed when the PMa-SynRM operates under a constant torque region. By comparison, Elena Trancho et al. [11] have designed a robust torque-based control scheme addressed to the PM-Assisted Synchronous Reluctance Machine in EVs and HEVs. In this work, the authors have employed an arrangement of a second-order-based sliding mode control for inner current regulation and a look-up table/voltage restriction pursuit-based hybrid field weakening operation to cope with parameters deviation during operation. Despite these relevant works introducing and designing robust controllers, the achievement of high performance is still a challenging barrier. As emphasized in these works, this barrier is due to the variation of machine electrical parameters and the nonlinear properties. To meet these technical issues for the control of the PMa-SynRM, a nonlinear control named "Differential Flatness" is elaborated and has recently been suggested to reach the expected performances in controlling PMSM [12]. It has been demonstrated that differential flatness control presents higher performance than traditional control systems. Besides, adding the nonlinear observation for motor parameters estimating makes the control system more robust. This novel control approach has been applied in various nonlinear systems; for instance, for the energy management of a hybrid power plant (including a fuel cell and supercapacitors) [13], and the command of double fed inductor motor [14].

This paper presents the differential flatness control to control PMa-SynRM. As a result of this introduction reviewing the main issues for the control of PMa-SynRM and the current state-of-the-art, the mathematical models of PMa-SynRM are developed to prove the differential flat property in Section 2. Then, in Section 3, the intelligent proportional-integral (*i*PI) [15,16] controller is conceived as a control law to compensate for the torque and speed of the motor. A nonlinear estimator is introduced to estimate external torque disturbance. Finally, in Section 4, the comparison between differential flatness and conventional PI control is discussed to demonstrate the benefits of the proposed control algorithm. A small-scale test bench 1-KW relying on the PMa-SynRM with ferrite magnets has been realized to attest to the performance of the designed control scheme in the laboratory [17].

### **2. FEM-Based Magnetic Model**

Given that the inductances *L*<sup>d</sup> and *L*<sup>q</sup> play major roles in the overload capacity, the field-weakening operation has been enhanced, and an accurate regulation has been determined to make their calculation easier. The inductances in the d–q axis have been computed considering a nonlinear case where the saturation of the stator teeth and rotor ribs have been taken into account. Due to the effect of the internal temperature, the inductances can be saturated as emphasized in [10]. Hence, the saturation effects of the inductances *L*<sup>d</sup> and *L*<sup>q</sup> can be computed relying on the link between the flux linkage change and the small rise in the current of the d–q axis, as provided by the Equations (1)–(3) [18,19]. Furthermore, the cross-coupling effects due to saturation have been investigated as well for the PMa-SynRM. The parameters have been assessed in the d–q axis. The operating constraints of the current supply have been reproduced, and the flux linkages have been assessed by incorporating the magnetic vector potential. The d-q flux linkages linked to the d- and q-axes currents are illustrated in Figure 2. The d- and q-flux linkages related to based on FEM analysis are represented in Figure 2. Figure 2a depicts the lookup table of d-axis flux linkage Ψd(id,iq) (LUT1); while Figure 2b exhibits the lookup q-axis flux linkage Ψq(id,iq) (LUT2). In addition, Table 1 shows the given flux linkages of the PMs and d–q axes inductances of the control network design.

**Figure 2.** *Cont.*

**Figure 2.** The flux linkages Ψ<sup>d</sup> and Ψ<sup>q</sup> in the function of *i*<sup>d</sup> and *i*q. (**a**) Ψd(*i*d,*i*q). (**b**) Ψq(*i*d,*i*q).

$$L\_{\rm d} = \frac{\partial \Psi\_{\rm d}(i\_{\rm d}, i\_{\rm q})}{\partial i\_{\rm d}} = \left. \frac{\Delta \Psi\_{\rm d}(i\_{\rm d}, i\_{\rm q})}{\Delta i\_{\rm d}} \right|\_{i\_{\rm q} = \text{constant}} \tag{1}$$

$$L\_{\mathbf{q}} = \frac{\partial \Psi\_{\mathbf{q}}(i\_{\mathbf{d}\prime}i\_{\mathbf{q}})}{\partial i\_{\mathbf{q}}} = \left. \frac{\Delta \Psi\_{\mathbf{q}}(i\_{\mathbf{d}\prime}i\_{\mathbf{q}})}{\Delta i\_{\mathbf{q}}} \right|\_{i\_{\mathbf{d}} = \text{constant}} \tag{2}$$

$$L\_{\rm dq} = \frac{\partial \Psi\_{\rm d}(i\_{\rm d}, i\_{\rm q})}{\partial i\_{\rm q}} = \left. \frac{\Delta \Psi\_{\rm d}(i\_{\rm d}, i\_{\rm q})}{\Delta i\_{\rm q}} \right|\_{i\_{\rm d} = \text{constant}} \tag{3}$$

**Table 1.** Properties of the PMa-SynRM parameters relying on FEM analysis.


### **3. A Shot Briefly Differential Flatness Control and Control Law**

### *3.1. Differential Flatness Briefly*

The differential flatness-based control approach is crucial to control different types of systems [2,13,14]. A summary of differential flatness control theory is provided below. Considering a nonlinear system expressed by the following state-variable:

.

$$
\dot{\mathfrak{x}} = f(\mathfrak{x}, \mathfrak{u}).\tag{4}
$$

The system (4) is said to be "differentially flat" if a set of flat output equal to the number of inputs can be found. More precisely, the control output variable must be written as the function of the flat output and their derivatives as follow:

$$\mathbf{x} = \phi(y, \dot{y}, \dots, y^{(\beta)}),\tag{5a}$$

$$u = \psi(y, \dot{y}, \dots, y^{(\beta + 1)}),\tag{5b}$$

where *β* is the finite number of derivatives.

### *3.2. Control Law*

The control law's block diagram is provided in Figure 3. The trajectory planning, the feedback control (relying on two controller gains *K<sup>i</sup>* and *Kp*), the controller output *λ*, and the inverse dynamic equations are detailed below.

**Figure 3.** Control law's block diagram.

As shown in Figure 3, a controller output, *λ* can be defined as follow:

$$
\lambda = \mathfrak{u}\_{\text{ref}} + \mathfrak{u}\_{\text{feedback}}(\varepsilon), \tag{6}
$$

with

$$
\mu\_{\rm ref} = \psi(y\_{\rm ref}, \dot{y}\_{\rm ref}, \ddot{y}\_{\rm ref}, \dots, y\_{\rm ref}^{(\notin +1)}), \tag{7}
$$

$$
\varepsilon = y\_{ref} - y.\tag{8}
$$

As will be seen later, thank to "Inverse Dynamic Equation" (IDE), we will obtain.

$$
\dot{y} = \lambda.\tag{9}
$$

According to the control law's block diagram, combining (6)–(8) yields.

.

$$
\dot{y} = \dot{y}\_{ref} + \mathcal{K}\_p \varepsilon + \mathcal{K}\_i \int \varepsilon dt = 0. \tag{10}
$$

Taking time derivative (10) obtains.

$$
\ddot{y}\_{ref} + \mathcal{K}\_p \dot{\varepsilon} + \mathcal{K}\_i \varepsilon = 0. \tag{11}
$$

By comparing to the standard second-order equation, parameters *K<sup>p</sup>* and *K<sup>i</sup>* can define as follow:

$$
\ddot{q}(s) + 2\zeta\omega\_n\dot{q}(s) + \omega\_n^2 q(s) = 0,\tag{12}
$$

Consequently, the controller gains define as follow:

$$\begin{aligned} \mathbf{K}\_p &= \mathfrak{L}\widetilde{\zeta}\omega\_n\\ \mathbf{K}\_i &= \omega\_n^2 \end{aligned} \tag{13}$$

### **4. PMa-SynRM Modeling and Development of the Proposed Control Scheme**

*4.1. Mathematic Model of PMa-SynRM/Inverter*

The inverter shown in Figure 4 provides a symmetric sinusoidal three-phase AC voltage source for supplying to PMa-SynRM. In Figure 4, *VBUS*, *iBUS*, and *iA*, *i<sup>C</sup>* are respectively the input DC grid voltage, the inverter current, and the load motor phase current. According to Figure 2a,b, The electrical modeling equations of PMa-SynRM are discussed

by the nonlinear case. In Figure 2, the flux linkage of direct and quadrature axes may be defined according to d- and q-axes current id and iq as in the following equations.

$$\mathbf{Y\_d} = \mathbf{Y\_d}(i\_{\mathbf{d}\prime} i\_{\mathbf{q}}) \,\tag{14}$$

$$\Psi\_{\mathbf{q}} = \Psi\_{\mathbf{q}}(i\_{\mathbf{d}\prime} i\_{\mathbf{q}}).\tag{15}$$

**Figure 4.** A three-phase inverter to control the PMa-SynRM prototype.

By considering Equations (14) and (15) as well, as mentioned above, the rotating electrical modeling of PMa-SynRM is given by the following equations [10,17]:

$$v\_{\rm d} = R\_{\rm s} \cdot i\_{\rm d} + \frac{d\Psi\_{\rm d}(i\_{\rm d}, i\_{\rm q})}{dt} - \omega\_{\rm e} \cdot \Psi\_{\rm q}(i\_{\rm d}, i\_{\rm q}) \,, \tag{16}$$

$$v\_{\mathbf{q}} = \mathbf{R}\_{\mathbf{s}} \cdot i\_{\mathbf{q}} + \frac{d\Psi\_{\mathbf{q}}(i\_{\mathbf{d}\prime}i\_{\mathbf{q}})}{dt} + \omega\_{\mathbf{e}} \cdot \Psi\_{\mathbf{d}}(i\_{\mathbf{d}\prime}i\_{\mathbf{q}})\_{\prime} \tag{17}$$

where

$$
\omega\_{\mathbf{e}} = n\_p \cdot \omega\_{m\nu} \tag{18}
$$

where *v<sup>d</sup>* is the *d*-axis voltage, *v<sup>q</sup>* is the *q*-axis voltage, *ω*<sup>e</sup> is the electrical angular frequency, *n<sup>p</sup>* is the number of pole pairs, and *ω<sup>m</sup>* is the mechanical angular frequency. The important electromagnetic torque of the machine composed of torque produced with the interactivity between the magnet and the reluctance torque is expressed as follows.

$$T\_{\varepsilon} = n\_p \left( \Psi\_d i\_q - \Psi\_q i\_d \right). \tag{19}$$

The mechanical equation in the rotation moving is:

$$J\frac{d\omega\_{\rm m}}{dt} = T\_{\rm e} - B\_{\rm f}\omega\_{\rm m} - T\_{\rm L} \tag{20}$$

where *J* is the moment of inertia, *B*<sup>f</sup> is the viscosity, and *T*<sup>L</sup> is the load torque.

### *4.2. Differential Flatness Control of Current (or Torque) Loop Development*

By referring to Equations (5), (16) and (17), the Ψ<sup>d</sup> and Ψ<sup>q</sup> are determined as the state variables (*x*). The *v*<sup>d</sup> and *v*<sup>q</sup> are control variables (*u*). The flat output (*y*) candidates are the measured parameters, which are *i*<sup>d</sup> and *i*q. The systems can be seen as differentially flat if the control output variable must be noted according to the flat output, which are

$$w\_d = \mu\_1 = \mathbb{R}\_\delta y\_1 - \omega\_\ell \mathbb{1}\_\emptyset (y\_1, y\_2) + \frac{\partial \mathbb{1}\_d (y\_1, y\_2)}{\partial y\_1} \cdot \dot{y}\_1 + \frac{\partial \mathbb{1}\_d (y\_1, y\_2)}{\partial y\_1} \cdot \dot{y}\_2 = \psi\_1 (y\_1, y\_2, \dot{y}\_1, \dot{y}\_2), \tag{21}$$

$$w\_q = \mu\_2 = \mathbb{R}\_8 y\_2 + \omega\_\ell \mathbb{1}\_d(y\_1, y\_2) + \frac{\partial \mathbb{1}\_q(y\_1, y\_2)}{\partial y\_1} \cdot \dot{y}\_1 + \frac{\partial \mathbb{1}\_q(y\_1, y\_2)}{\partial y\_2} \cdot \dot{y}\_2 = \psi\_2(y\_1, y\_2, \dot{y}\_1, \dot{y}\_2). \tag{22}$$

The control scheme mentioned in Section 3.2 is applied to deal with some inevitable modeling errors and uncertainties. By referring to Equation (6), the control laws of current control can express as follow:

$$
\dot{y}\_1 = \dot{\Psi}\_{dref} + \mathcal{K}\_p \varepsilon\_d + \mathcal{K}\_i \int \varepsilon\_d dt \,\tag{23}
$$

$$
\dot{y}\_2 = \dot{\Psi}\_{qref} + \mathcal{K}\_p \varepsilon\_q + \mathcal{K}\_i \int \varepsilon\_q dt. \tag{24}
$$

Consequently, the output control variables yield as follow:

$$
\mu\_1 = v\_d = \dot{y}\_1 + IDE\_{d\nu} \tag{25}
$$

$$
\mu\_2 = v\_q = \dot{y}\_2 + IDE\_{q\prime} \tag{26}
$$

where the Inverse Dynamic Equations (IDEs) are

$$IDE\_d = -R\_s i\_d + \omega\_\varepsilon \Psi\_{q\prime} \tag{27}$$

$$IDE\_q = -R\_s i\_q + \omega\_\varepsilon \Psi\_d.\tag{28}$$

The controller parameters are

$$K\_{pd} = K\_{pq} = \mathcal{Z}\_1 \omega\_{n1\prime} \tag{29}$$

and

$$K\_{\rm id} = K\_{\rm iq} = \omega\_{\rm n1\prime}^2 \tag{30}$$

where *ζ*<sup>1</sup> and *ω*n1 are respectively the desirated governing damping ratio and natural frequency.

### *4.3. Differential Flatness Control of Speed Control Loop Development*

Figure 5 shows the proposed control diagram. The outer speed loop enables evaluating the torque reference value of the MPTA, which generates the current command for the inner current loop.

**Figure 5.** Schematic drawing of the designed control scheme.

For the MTPA algorithm, it has been proposed in [20]. So, the *T<sup>e</sup>* is defined as a command variable *u*3, and the flat output *y*<sup>3</sup> is *ω*<sup>m</sup> (or measured angular speed). The system is flat if the control variable is a function of flat output that is

$$\mathfrak{u}\_{\mathfrak{3}} = T\_{\mathfrak{e}} = \mathfrak{f} \cdot \dot{\mathfrak{y}}\_{\mathfrak{3}} + \mathfrak{B}\_{\mathfrak{f}} \mathfrak{y}\_{\mathfrak{3}} + T\_{\mathrm{L}} = \psi\_{\mathfrak{3}}(\mathfrak{y}\_{\mathfrak{3}}, \dot{\mathfrak{y}}\_{\mathfrak{3}}).\tag{31}$$

The control strategy of the speed control loop is

$$
\dot{y}\_3 = \dot{\omega}\_{ref} + \mathcal{K}\_{p\omega} \varepsilon\_\omega + \mathcal{K}\_{i\omega} \int \varepsilon\_\omega dt. \tag{32}
$$

The control variable of the speed loop can be express as follow:

$$
\mu\_3 = \mathbf{J} \cdot \dot{\mathbf{y}}\_3 + \mathbf{J} \cdot \text{IDE}\_{\omega\nu} \tag{33}
$$

where the inverse dynamic equation of the speed control loop (IDEω) is

$$IDE\_{\omega} = \frac{1}{J}(\mathcal{B}\_{\mathbf{f}} \cdot \mathcal{y}\_{\mathbf{3}} + T\_{\mathbf{L}}).\tag{34}$$

The controller parameters are defined as the following equation.

$$K\_{p\omega} = 2\mathbb{Z}\_2\omega\_{n2\omega} \tag{35}$$

and

$$K\_{i\omega} = \omega\_{n2\prime}^2\tag{36}$$

where *ζ*<sup>2</sup> and *ωn*<sup>2</sup> are, respectively, the desirated governing damping ratio and the natural frequency of the outer speed regulation loop.

Based on the current and speed control law development, the natural frequency setting of the designed controller is depicted in Figure 6. The switching frequency *f<sup>s</sup>* of the inverter shown in Figure 4 is equal to 16 kHz (*ω<sup>s</sup>* = 10<sup>5</sup> rad.s−<sup>1</sup> ) and it is reported in Figure 6. According to Figure 5, the speed control loop must be faster than the current control loop given that the outer speed loop enables assessing the torque reference value, and consequently the current. Considering the Nyquist-Shannon Theorem, the natural frequency *ωn*<sup>1</sup> must be chosen lower than a frequency equal to 10<sup>2</sup> rad.s−<sup>1</sup> (namely two times lower than the switching frequency *ωs*). Therefore, the natural frequency *ωn*<sup>1</sup> for the current control loop has been tuned at 2000 rad.s−<sup>1</sup> ; while for the speed control loop, a natural frequency *ωn*<sup>2</sup> has been set at 20 rad.s−<sup>1</sup> (100 times lower than *ωn*1). Both values have been reported in Figure 6, allowing defining the stable region included between these two values; whereas the unstable region is outside the natural frequency *ωn*1. Regarding the tuning of the damping ratios *ζ*<sup>1</sup> and *ζ*2, to guarantee underdamped transient behaviors with low overshoot and fast response, both parameters have been set at 0.7.

**Figure 6.** The natural frequency setting of the designed controller.

Note that the stability and response of the differential flatness-based control are easy to set compared to the traditional PI controller. By defining and selecting the governing damping and natural frequency [21,22], as shown in Figure 6, the controller parameters of current and speed loop control may be calculated by Equations (29), (30), (35) and (36).

### *4.4. Trajectory Planning*

The trajectory planning enables restricting the derivative terms. The reference inputs have been defined by trajectory planning utilized by the second-order low-pass filter. The trajectory planning of the current control loops are

$$\frac{y\_{1REF}(s)}{y\_{1COM}(s)} = \frac{y\_{2REF}(s)}{y\_{2COM}(s)} = 1 \Big/ \left( \left(\frac{s}{\omega\_{n3}}\right)^2 + \frac{2\zeta\_3}{\omega\_{n3}} + 1 \right). \tag{37}$$

In the speed control loop, the trajectory planning has been determined by the following equation.

$$\frac{y\_{3REF}(s)}{y\_{3COM}(s)} = 1 \left/ \left( \left( \frac{s}{\omega\_{n4}} \right)^2 + \frac{2\zeta\_4}{\omega\_{n4}} + 1 \right) \right. \tag{38}$$

where *ζ*3, *ω*3, *ζ*4, and *ω*<sup>4</sup> are the governing damping and natural frequencies of the secondorder low-pass filters, respectively.

### **5. Simulation and Experimental Validation**

### *5.1. Experimental Setup*

A small-scale test bench 1-KW relying on the PMa-SynRM has been conceived in the laboratory, as shown in Figure 7. Table 2 sums up the principal parameters of the studied machine. Table 3 outlines the controller parameters. The motor is supplied by

a 3-kW 3-phase inverter (DC/AC) operating at a switching frequency of 16 kHz. Besides, the input DC grid voltage of the inverter is fed by a 3-phase variable power supply combined with a 3-phase diode rectifier. The PMa-SynRM is mechanically coupled with an IPMSM (interior permanent magnet synchronous motor) feeding a resistive load (see Figures 4 and 7). Regarding the measurements both for the speed and rotor angle, they have been acquired by a resolver placed on the rotor shaft. The developed control scheme (see Figure 5) relying on the differential flatness controller has been modeled in the Matlab/Simulink software, and then it has been incorporated in the dSPACE 1202 MicroLabBox real-time interface to generate the gate control signals applied to the VSI.

**Figure 7.** The experimental setup.

**Table 2.** Specification and parameters of the motor/inverter.



**Table 3.** Current/torque and speed regulation parameters.

*5.2. Simulation and Test-Bench Results of the Speed Reversal Employing the Differential Flatness Controller*

For the first scenario, Figure 8 reports the obtained simulation results; whereas, Figure 9 exhibits the performed experimental tests to assess the dynamic performance of the designed controller when forcing the motor to reverse direction. In Figure 8, Ch1–10 are the command signal of the speed *n*COM, the reference signal of the speed *n*REF, the measured speed *n*, the command of q-axis current *i*qCOM, the reference of q-axis current *i*qREF, the q-axis current *i*q, the command of d-axis current *i*dCOM, the reference of d-axis current *i*dREF, and the d-axis current *i*d, respectively. In comparison, in Figure 9, Ch1–8 are the speed command *n*COM, the speed reference *n*REF, the measured speed *n*, the current *i*q, the current reference *i*dREF, the current *i*d, and the current reference *i*qREF, respectively. Firstly, the PMa-SynRM model has been tested by using Matlab/Simulink to support that the elaborated control system is appropriately conceived. Simulations and experimental tests have demonstrated that both simulation and experimental results are similar. Thus, the PMa-SynRM model is fit, and the controller parameters are suitably designed by choosing desired parameters. The experimental results indicate that the PMa-SynRM behaves in a good way when operating under the regenerative mode up to the speed the reference gets positive. Furthermore, it can be emphasized that the measured speed through the resolver enables tracking adequately the speed reference value. Afterward, the operation of the PMa-SynRM is shifted to the motoring mode up to the rotor speed comes to the speed command. The *d*-and *q*-axes currents reveal an appropriate behavior without surpassing the imposed limits. The dominant parameters of the PMa-SynRM enable being ensured, and the elaborated control offers worthwhile dynamic performance.

**Figure 8.** Simulation results: motor speed reversal.

**Figure 9.** Experimental results: motor speed reversal.

*5.3. Experimental Results of the Comparison between the Conventional PI Control and Differential Flatness Control*

For the second scenario, the performance of the system when the torque/current loop employs traditional PI control and differential flatness is compared to assess the benefits of the elaborated control scheme. Figure 10a represents the experimental results of the conventional PI control, and Figure 10b illustrates the experimental results of differential flatness control. In Figure 10a, Ch1 is the current *i*dCOM, Ch3 is the measured current *i*d, Ch4 is the measured current *i*q, and Ch5 is the measured speed *n*. In Figure 10b, Ch1 is the current *i*dCOM, Ch3 is the measured current *i*d, Ch4 is the measured current *i*q, and Ch5 is the measured speed *n*. As shown in Figure 10a,b, in a transitory operation, the *i*<sup>d</sup> of PI control exhibits a small overshoot, compared to the differential flatness controller, and the *i*<sup>q</sup> of the PI control shows oscillations.

**Figure 10.** Experimental results: (**a**) the traditional PI control, (**b**) the differential flatness control.

Although the PI controller has used decoupling and back-emf compensation, it demonstrates that the proposed nonlinear controller has a better transient current performance than the traditional linear controller. Furthermore, the speed response (Ch5 of Figure 10a) of the linear cascaded PI controller includes fluctuations, unlike the proposed nonlinear controller (Ch5 of Figure 10b).

For the third scenario, Figure 11a confirms the experimental results of the conventional PI controller, and Figure 11b indicates the experimental validation results of the differential flatness controller when the motor is forced to reverse direction from −1000 rpm to 1000 rpm. In Figure 11a, Ch1 is the speed command *n*COM, Ch2 is the speed *n*, Ch3 is the current command *i*qCOM, Ch4 is the measured current *i*q, Ch5 is the current command *i*dCOM, and Ch6 is the measured current *i*d. In Figure 11b, Ch1 is the speed command *n*COM, and Ch2 is the acquired speed *n*, Ch3 is the current command *i*qCOM, Ch4 is the measured current *i*q, Ch5 is the current command *i*dCOM, and Ch6 is the measured current *i*d. On one hand, as demonstrated in Figure 11a, during a transient process, the acquired speed *n* of the PI controller shows an overshoot, and the settling time is approximately 0.45 s. On the other hand, the differential flatness controller (Figure 11b), approximate time is around 0.15 s, as well as the measured *i*<sup>q</sup> of the PI controller, which fluctuates sharply (Ch4 of Figure 11a) in a transient process. The experimental results reflect that differential flatness has a better dynamic speed performance than the traditional PI controller.

**Figure 11.** *Cont.*

**Figure 11.** Experimental results: (**a**) the traditional PI control, (**b**) the differential flatness control.

For the fourth scenario, Figure 12a shows the test-bench results of the conventional PI controller; while, Figure 12b depicts the preliminary results of the differential flatness control when suddenly adding an external torque disturbance. In Figure 12a, Ch2 is the measured *d*-axis current *i*d, Ch3 is the measured *q*-axis current *i*q, Ch4 is the measured speed *n*, Ch5 is the measured phase-A current *i*a, and the trajectories of the transient stator current. In Figure 12b, Ch2 is the measured *d*-axis current *i*d, Ch3 is the measured *q*-axis current *i*q, Ch4 is the measured speed *n*, Ch5 is the phase-A current *i*a, and the path of the transient stator current. The experimental results are shown in Figure 12b, validating that differential flatness speed oscillation is roughly 113 rpm; whereas that for the PI controller is 221 rpm. The recuperation time of speed with the elaborated controller is also shorter than that with conventional PI control. These results corroborate that differential flatness control has better dynamic performance both for the torque/current and speed loop control and the external rejection ability.

**Figure 12.** Experimental results: (**a**) the traditional PI control (**b**) the differential flatness control.

### **6. Conclusions**

To cope with the control issues met in a permanent magnet-assisted synchronous reluctance motor (PMa-SynRM), various control approaches have been previously investigated. Nonetheless, the achievement of high performance is still a challenging barrier due to the nonlinear characteristics and parameter uncertainty conditions of this motor. In this work, a differential flatness control law has been elaborated and designed to control both current/torque and the speed of the PMa-SynRM. Furthermore, an intelligent proportionalintegral (*i*PI) has been combined with the nonlinear differential flatness controller to face unavoidable modeling errors and uncertainties for the torque and speed of the motor. This model-based approach requires an accurate model. In case the model is not perfectly known, the estimation of the unknown part is necessary to achieve the expected high performance. Through simulations and experimental tests performed on a small-scale test bench 1-KW including the PMa-SynRM, the dynamic performances of the system have been validated; while demonstrating the performance superiority of the differential flatness controller over the conventional PI controller from the overshoot and oscillation point of view. Furthermore, the results reflect that the dynamic recovery time response is faster using intelligent PI control than the field-oriented control (FOC) based on PI controller with approximately 0.15 s.

In the future work, another control approach will be tentatively applied to the control of PMa-SynRM. This approach, called the model-free control, does not require an accurate model. Indeed, only very limited knowledge of the controlled system is enough to regenerate the control action. Advantages and drawbacks of this controller will be discussed and its performance will be compared to the flatness-based controller in the next work.

**Author Contributions:** Conceptualization, B.N.-M. and N.T.; methodology, S.S., N.P., D.G. and P.T.; validation, S.S., N.P., D.G. and P.T.; formal analysis, N.B.; writing—original draft preparation, P.T.; writing—review and editing, S.S., D.G. and N.B.; visualization, N.B.; supervision, B.N.-M. and N.T. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was partially supported by the International Research Partnerships: Electrical Engineering Thai-French Research Center (EE-TFRC) between Université de Lorraine (UL) and King Mongkut's University of Technology North Bangkok (KMUTNB) and Framework Agreement between the University of Pitesti and King Mongkut's University of Technology North Bangkok through the Research Program Cooperation under Grant KMUTNB–61–GOV–01–67. Besides, this work was supported partly by the French PIA project «Lorraine Université d'Excellence», reference ANR-15-IDEX-04-LUE.

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

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are available on request from the corresponding authors. The data are not publicly available due to their current utilization for future works involving the authors of this paper.

**Acknowledgments:** The authors would like to express their gratitude to the GREEN laboratory at the University of Lorraine and King Mongkut's University of Technology North Bangkok (KMUTNB) for their constant support in boosting collaborations between France and Thailand.

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

### **References**


## *Review* **State-of-the-Art Review on IoT Threats and Attacks: Taxonomy, Challenges and Solutions**

**Ritika Raj Krishna <sup>1</sup> , Aanchal Priyadarshini <sup>1</sup> , Amitkumar V. Jha <sup>1</sup> , Bhargav Appasani <sup>1</sup> , Avireni Srinivasulu <sup>2</sup> and Nicu Bizon 3,4,\***


**Abstract:** The Internet of Things (IoT) plays a vital role in interconnecting physical and virtual objects that are embedded with sensors, software, and other technologies intending to connect and exchange data with devices and systems around the globe over the Internet. With a multitude of features to offer, IoT is a boon to mankind, but just as two sides of a coin, the technology, with its lack of securing information, may result in a big bane. It is estimated that by the year 2030, there will be nearly 25.44 billion IoT devices connected worldwide. Due to the unprecedented growth, IoT is endangered by numerous attacks, impairments, and misuses due to challenges such as resource limitations, heterogeneity, lack of standardization, architecture, etc. It is known that almost 98% of IoT traffic is not encrypted, exposing confidential and personal information on the network. To implement such a technology in the near future, a comprehensive implementation of security, privacy, authentication, and recovery is required. Therefore, in this paper, the comprehensive taxonomy of security and threats within the IoT paradigm is discussed. We also provide insightful findings, presumptions, and outcomes of the challenges to assist IoT developers to address risks and security flaws for better protection. A five-layer and a seven-layer IoT architecture are presented in addition to the existing three-layer architecture. The communication standards and the protocols, along with the threats and attacks corresponding to these three architectures, are discussed. In addition, the impact of different threats and attacks along with their detection, mitigation, and prevention are comprehensively presented. The state-of-the-art solutions to enhance security features in IoT devices are proposed based on Blockchain (BC) technology, Fog Computing (FC), Edge Computing (EC), and Machine Learning (ML), along with some open research problems.

**Keywords:** Internet of Things; security; threats; privacy; vulnerabilities; Blockchain

## **1. Introduction**

We live in a time when technology is an essential requirement for all humans, and the evidence is the increased dependence on technology in almost every aspect of our lives. Today's world is evolving with the rapidly growing Internet of Things (IoT)-based application [1]. The rise of the IoT has been a glorious phenomenon in recent years. The physical and virtual objects implanted with sensors, software, and other technologies are interlinked together in IoT [2]. It envisages communicating and sharing data with other devices and systems worldwide over the Internet. Further, IoT is like an array of network-enabled devices that exclude traditional computers such as laptops and servers.

IoT has sprawled everywhere, starting from the healthcare sector to the big industries. It is now implantable, wearable, and portable, resulting in a pervasive and interactive

**Citation:** Krishna, R.R.;

Priyadarshini, A.; Jha, A.V.; Appasani, B.; Srinivasulu, A.; Bizon, N. State-of-the-Art Review on IoT Threats and Attacks: Taxonomy, Challenges and Solutions. *Sustainability* **2021**, *13*, 9463. https:// doi.org/10.3390/su13169463

Academic Editor: Zubair Baig

Received: 30 May 2021 Accepted: 18 August 2021 Published: 23 August 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/).

world [3]. It modifies the physical objects around us into smart objects, creating an information environment that increasingly changes human living standards. For instance, IoT devices track and collect essential measurements (such as blood pressure, blood sugar level, pulse rate, etc.) in real time, allowing emergency alerts to improve the odds of a patient's survival [4]. Moreover, autonomous and self-driving vehicles prevent drivers from deviating from paths or accidents while providing them assistance to reach their destinations. In addition, those definitions are expanded to provide automatic emergency alerts of the closest road and medical assistance in the event of an accident. IoT also covers many aspects of modern industries, including manufacturing, assembly, packing, logistics, smart cities, and aviation industries [5]. Some of the essential IoT-based application domains in health, commerce, communication, and entertainment are shown in Figure 1. information environment that increasingly changes human living standards. For instance, IoT devices track and collect essential measurements (such as blood pressure, blood sugar level, pulse rate, etc.) in real time, allowing emergency alerts to improve the odds of a patient's survival [4]. Moreover, autonomous and self-driving vehicles prevent drivers from deviating from paths or accidents while providing them assistance to reach their destinations. In addition, those definitions are expanded to provide automatic emergency alerts of the closest road and medical assistance in the event of an accident. IoT also covers many aspects of modern industries, including manufacturing, assembly, packing, logistics, smart cities, and aviation industries [5]. Some of the essential IoT-based application domains in health, commerce, communication, and entertainment are shown in Figure 1.

devices and systems worldwide over the Internet. Further, IoT is like an array of network-

IoT has sprawled everywhere, starting from the healthcare sector to the big industries. It is now implantable, wearable, and portable, resulting in a pervasive and interactive world [3]. It modifies the physical objects around us into smart objects, creating an

enabled devices that exclude traditional computers such as laptops and servers.

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 2 of 47

Smart Transportation

**Figure 1.** Important IoT application domains. **Figure 1.** Important IoT application domains.

To implement IoT, the traditional technology had to undergo some major modifications. For example, to convert an isolated device into a transmitting device, there is a need to increase small computing devices' memory and processing capacity while dramatically reducing their size [6]. Further, the creation of various lightweight, secure protocols for communication between different IoT devices is equally important. The improvements to the conventional networks to help the operation of the IoT ecosystem have their own set To implement IoT, the traditional technology had to undergo some major modifications. For example, to convert an isolated device into a transmitting device, there is a need to increase small computing devices' memory and processing capacity while dramatically reducing their size [6]. Further, the creation of various lightweight, secure protocols for communication between different IoT devices is equally important. The improvements to the conventional networks to help the operation of the IoT ecosystem have their own set of consequences. However, the unprecedented growth of interconnected devices has crippled the IoT ecosystem. Consequently, there exists enough scope for threats and attacks in IoT-based applications.

In-flight services

The Global Vice President at New Net Technologies (NNT), Dirk Schrader, stated that IoT-based computers have become the crown jewels of cybercriminals. He also said that less than 42% of businesses can detect insecure IoT devices. Hence, for researchers to develop well-grounded solutions to trace and avert these threats, they must first understand the threats and attacks to make the IoT environment safe, secure, and reliable. There are three significant aspects to consider while examining the IoT from a security perspective. To begin with, there are a massive number of smart devices, possibly billions. This suggests that the IoT would be the most complex man-made system ever in terms of the number of entities involved [7]. Second, they are essentially heterogeneous, with respect to the functionality, protocol stacks, radios, operating systems (some objects do not even have one), energy sources, identities, and so on [8]. Third, each smart object is owned by a company or a person, and it is managed by the same or a different company or individual. Millions of businesses and individuals are in control of a subset of the smart objects in their management domains. From the standpoint of protection, privacy, and trust, how this control is technically upheld is a critical issue.

The attack surface in the IoT domain has increased significantly, as have the possible threats to the protection of these entities in the domain [9]. For example, the security threats to the autonomous and self-driving industry may lead to disastrous consequences. Autonomous vehicles are vulnerable to sensor-based attacks. By manipulating the sensors (e.g., linear acceleration sensor, magnetic sensor, etc.), attackers may collect data, transfer malware to it, or trigger a malicious activity [10]. Furthermore, smartphones and embedded systems contribute to a digital ecosystem for global communication that simplifies lives by being sensitive, flexible, and responsive to human needs. However, on the other hand, security cannot be assured due to vulnerabilities in IoT. When a user's signal is disrupted or intercepted, their privacy may be jeopardized, and their information may be leaked.

The state-of-the-art survey on various aspects of IoT, including security, privacy, and robustness, has been presented in [11] by Chen et al. The authors focused on specific issues of IoT interface positioning and localization. The development of lightweight block cipher algorithms has been proposed to be used in devices for data encryption and decryption [12]. A desktop review and qualitative analysis have been performed by Gamudani et al. in [13] to compute performance analysis of attacks. Cryptographic approaches have been discussed in [14] as a method of ensuring long-term security approaches. Different layer architectures of IoT and security issues associated with them have been discussed with possible countermeasures using Blockchain (BC) in [15]. The survey on security aspects of IoT has been presented by Alaba et al. in [16], covering the scope of security countermeasures in some other allied paradigms, including Machine-to-Machine (M2M), Cyber-Physical System (CPS), and Wireless Sensor Networks (WSNs). In [17], Abomhara et al. discussed various applications of IoT and the security threats related to them, including vulnerabilities, intruders, and some other attacks. The threats concerning security and privacy in IoT architecture have been presented without counter measuring techniques by Kozlov et al. in [18].

The organization of the paper is as follows. The state-of-the-art motivation and contributions of this research are presented in Section 2. The background of the IoT as the foundation to the security threats and attacks is presented in Section 3. Section 4 deals with the IoT reference model and the protocol stack. The state-of-the-art review on the vulnerabilities with threats and attacks taxonomy in the IoT paradigm is presented in Section 5. Security goals and a roadmap in IoT are presented in Section 6. Section 7 deals with the state-of-the-art security solution for IoT framework using ubiquitous technologies, such as BC, FC, EC, and ML. Some of the open research problems are discussed in Section 8. The last section deals with the conclusion of the article with future scope for research.

#### **2. State-of-the-Art, Motivation, and Contributions of This Research 2. State-of-the-Art, Motivation, and Contributions of this Research**

### *2.1. Trends in Literature and Motivation 2.1. Trends in Literature and Motivation*

There is an ample amount of work in literature focusing on the IoT from various perspectives. Particularly, aspects such as applications, architecture, protocols, and standards are extensively covered in the literature. However, threats and attacks in IoT are comparatively less explored. The analysis from one of the world's largest databases, i.e., SCOPUS, can be used to understand the relevance of the particular aspects of IoT. If we search the number of articles in the SCOPUS database that focus on IoT architecture, IoT architecture and threats, and IoT architecture and attacks, then it can be corroborated from the search results that the threats and attacks analysis of IoT architecture is sparsely explored in the literature, which can be validated from the SCOPUS statistics seen in Figure 2. Further, there is rapid growth in the interest of the researchers towards threats and attacks analysis in the IoT architecture. This can be corroborated from the number of articles pertaining to the threats and attacks analysis in IoT architecture, which was four and zero, respectively, in 2010, and rapidly increased to 73 and 157, respectively, up to the third quarter of the year 2021 (approximately). There is an ample amount of work in literature focusing on the IoT from various perspectives. Particularly, aspects such as applications, architecture, protocols, and standards are extensively covered in the literature. However, threats and attacks in IoT are comparatively less explored. The analysis from one of the world's largest databases, i.e., SCO-PUS, can be used to understand the relevance of the particular aspects of IoT. If we search the number of articles in the SCOPUS database that focus on IoT architecture, IoT architecture and threats, and IoT architecture and attacks, then it can be corroborated from the search results that the threats and attacks analysis of IoT architecture is sparsely explored in the literature, which can be validated from the SCOPUS statistics seen in Figure 2. Further, there is rapid growth in the interest of the researchers towards threats and attacks analysis in the IoT architecture. This can be corroborated from the number of articles pertaining to the threats and attacks analysis in IoT architecture, which was four and zero, respectively, in 2010, and rapidly increased to 73 and 157, respectively, up to the third quarter of the year 2021 (approximately).

**Figure 2.** Literature statistics on IoT architecture, IoT architecture and threats, and IoT architecture and attacks. **Figure 2.** Literature statistics on IoT architecture, IoT architecture and threats, and IoT architecture and attacks.

A similar trend can be seen with respect to the protocols and standards in the IoT paradigm. The publication statistics obtained from the SCOPUS database for the articles on IoT protocols, IoT protocols and threats, and IoT protocols and attacks are shown in Figure 3. The plotted statistics reveal that the threats and attacks analysis in IoT protocols were sparsely explored in the past 10 years. Nevertheless, these aspects are gaining rapid momentum, which can be corroborated from the published articles on threats and attacks analysis in IoT protocols, which were six and five, respectively, in 2010, and have increased to 160 and 254, respectively, by the third quarter of 2021 (approximately). In a nutshell, the increasing interests of the researchers in the paradigm of IoT architecture and A similar trend can be seen with respect to the protocols and standards in the IoT paradigm. The publication statistics obtained from the SCOPUS database for the articles on IoT protocols, IoT protocols and threats, and IoT protocols and attacks are shown in Figure 3. The plotted statistics reveal that the threats and attacks analysis in IoT protocols were sparsely explored in the past 10 years. Nevertheless, these aspects are gaining rapid momentum, which can be corroborated from the published articles on threats and attacks analysis in IoT protocols, which were six and five, respectively, in 2010, and have increased to 160 and 254, respectively, by the third quarter of 2021 (approximately). In a nutshell, the increasing interests of the researchers in the paradigm of IoT architecture and protocols,

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 5 of 47

tivating factor for the present work.

attacks.

which are sparsely explored from threats and attacks point-of-view, is the motivating factor for the present work. protocols, which are sparsely explored from threats and attacks point-of-view, is the motivating factor for the present work.

protocols, which are sparsely explored from threats and attacks point-of-view, is the mo-

**Figure 3.** Literature statistics on IoT protocols, IoT protocols and threats, and IoT protocols and **Figure 3.** Literature statistics on IoT protocols, IoT protocols and threats, and IoT protocols and attacks. attacks.

The other motivating factor for the present work is the threats and attacks analysis and possible solutions using ubiquitous technologies, such as BC, Fog Computing (FC), Edge Computing (EC), and Machine Learning (ML). The threats and attacks analysis and possible solutions in architecture, protocols, and standards have gained significant momentum in the past few years, corroborating the upwards trend in the published articles The other motivating factor for the present work is the threats and attacks analysis and possible solutions using ubiquitous technologies, such as BC, Fog Computing (FC), Edge Computing (EC), and Machine Learning (ML). The threats and attacks analysis and possible solutions in architecture, protocols, and standards have gained significant momentum in the past few years, corroborating the upwards trend in the published articles as per the SCOPUS database statistics shown in Figure 4. The other motivating factor for the present work is the threats and attacks analysis and possible solutions using ubiquitous technologies, such as BC, Fog Computing (FC), Edge Computing (EC), and Machine Learning (ML). The threats and attacks analysis and possible solutions in architecture, protocols, and standards have gained significant momentum in the past few years, corroborating the upwards trend in the published articles as per the SCOPUS database statistics shown in Figure 4.

as per the SCOPUS database statistics shown in Figure 4.

**Figure 4.** Literature statistics on IoT and BC, IoT and FC, IoT and EC, and IoT and ML from SCOPUS database. **Figure 4.** Literature statistics on IoT and BC, IoT and FC, IoT and EC, and IoT and ML from SCOPUS database. **Figure 4.** Literature statistics on IoT and BC, IoT and FC, IoT and EC, and IoT and ML from SCOPUS database.

IoT is far behind in realizing its true potential due to the lack of interoperability. The most comprehensive review on security standards and interoperability goals is presented by Lee et al. [19]. To comprehensively review the existing architectures, protocols, and standards is one of the promising means to address the interoperability issues in IoT and other challenges. If we look at the trend of the type of documents in the SCOPUS database, it can be seen that researchers are significantly contributing with Review articles (12.2%) being the third most in number behind Articles (50.8%) and Conference papers (29.1%). These statistics obtained from the SCOPUS database are shown in Figure 5. Conclusively, the present work is a review that comprehensively surveys the existing work and presents the possible solution in the context of threats and attacks pertaining to the architecture, protocols, and standards in the IoT paradigm. most comprehensive review on security standards and interoperability goals is presented by Lee et al. [19]. To comprehensively review the existing architectures, protocols, and standards is one of the promising means to address the interoperability issues in IoT and other challenges. If we look at the trend of the type of documents in the SCOPUS database, it can be seen that researchers are significantly contributing with Review articles (12.2%) being the third most in number behind Articles (50.8%) and Conference papers (29.1%). These statistics obtained from the SCOPUS database are shown in Figure 5. Conclusively, the present work is a review that comprehensively surveys the existing work and presents the possible solution in the context of threats and attacks pertaining to the architecture, protocols, and standards in the IoT paradigm.

IoT is far behind in realizing its true potential due to the lack of interoperability. The

*Sustainability* **2021**, *13*, x FOR PEER REVIEW 6 of 47

**Figure 5.** Literature trends from SCOPUS database. **Figure 5.** Literature trends from SCOPUS database.

*2.2. Comparison with Existing Surveys*

*2.2. Comparison with Existing Surveys* Several works have surveyed IoT, its architecture, its reference model, communication protocols, etc., from the perspective of security, threats, and vulnerabilities with possible countermeasure methodologies. In this section, some of the existing surveys are dis-Several works have surveyed IoT, its architecture, its reference model, communication protocols, etc., from the perspective of security, threats, and vulnerabilities with possible countermeasure methodologies. In this section, some of the existing surveys are discussed and compared with the present work.

cussed and compared with the present work. Many works in the literature cover the various aspects of the threats and attacks in IoT. Authors in [20–26] cover the taxonomy of the threats and attacks pertaining to the IoT. These works mainly focus on two broad categories: the architecture of IoT and protocols/standards in IoT. Despite covering the threats and attacks taxonomy, only a few of these works present the possible countermeasures. However, none of these works present countermeasures of threats and attacks based on ubiquitous technology such as BC, FC, EC, and ML. Such ubiquitous technologies in analyzing the threats and attacks have been Many works in the literature cover the various aspects of the threats and attacks in IoT. Authors in [20–26] cover the taxonomy of the threats and attacks pertaining to the IoT. These works mainly focus on two broad categories: the architecture of IoT and protocols/standards in IoT. Despite covering the threats and attacks taxonomy, only a few of these works present the possible countermeasures. However, none of these works present countermeasures of threats and attacks based on ubiquitous technology such as BC, FC, EC, and ML. Such ubiquitous technologies in analyzing the threats and attacks have been surveyed in scattered ways by the authors in [27–37]. A comprehensive review of all these technologies to combat IoT threats and attacks is not available.

surveyed in scattered ways by the authors in [27–37]. A comprehensive review of all these technologies to combat IoT threats and attacks is not available. The comprehensive survey on security and attacks with possible countermeasures solutions has been presented by Abosata et al. in [20], where authors consider the application-specific IoT architecture belonging to industrial IoT. Mann et al. in [21] presented the classification of attacks pertaining to the IoT environment. For attack classification, authors have considered a three-layer architecture comprising devices, gateway, and cloud with respect to the possible attack type. The countermeasures have also been discussed. Ogonji has comprehensively presented a threat taxonomy for the IoT environment in [22], including two broad categories: security threat and privacy threat. The authors of this survey presented taxonomy and countermeasures for the three-layer domain-specific IoT architecture. The state-of-the-art survey on intrusion detection for mitigating the impact of threats and attacks on IoT systems has been presented by Zarpelão et al. [23]. The The comprehensive survey on security and attacks with possible countermeasures solutions has been presented by Abosata et al. in [20], where authors consider the applicationspecific IoT architecture belonging to industrial IoT. Mann et al. in [21] presented the classification of attacks pertaining to the IoT environment. For attack classification, authors have considered a three-layer architecture comprising devices, gateway, and cloud with respect to the possible attack type. The countermeasures have also been discussed. Ogonji has comprehensively presented a threat taxonomy for the IoT environment in [22], including two broad categories: security threat and privacy threat. The authors of this survey presented taxonomy and countermeasures for the three-layer domain-specific IoT architecture. The state-of-the-art survey on intrusion detection for mitigating the impact of threats and attacks on IoT systems has been presented by Zarpelão et al. [23]. The authors proposed four attributes in the survey: intrusion detection placement strategy, detection method, security threats, and validation. A similar extensive survey by Hajiheidari et al. discusses the state-of-the-art intrusion detection system for IoT environment with a detailed

authors proposed four attributes in the survey: intrusion detection placement strategy,

dari et al. discusses the state-of-the-art intrusion detection system for IoT environment

taxonomy of the attacks responsible for intrusion in IoT at various layers [24]. The seminal survey using the top-down approach on various aspects of security in the IoT environment from application-specific IoT architecture has been presented by Kouicem et al. [25]. The authors have also presented the detailed taxonomy of security solutions covering several application-specific IoT architectures, and they also proposed software-defined networkingbased solutions to the security in IoT. Sun et al., in [26], focused on the physical layer of the IoT and presented a rich survey covering various security aspects of the protocols and standards, including the countermeasure methodologies. All these surveys are rich in content covering various aspects of threats and attacks in the context of IoT architectures or protocols. Despite presenting the possible countermeasures of threats and attacks, the countermeasures based on rapidly evolving technologies such as BC, FC, EC, ML, etc., have not been discussed to the authors' best knowledge. However, the concluding remarks of all these surveys identified some of the research gaps and provided a hint towards utilizing these technologies.

Elazhary [27] has presented an extensive survey on such computing technologies in the paradigm of IoT. Despite handling various aspects of IoT, particularly computation, processing, and analysis of voluminous IoT data, the security aspects of these data from the architecture point of view are not extensively covered in this survey. Taylor, P.J. et al., in [28], presents a seminal survey of using BC technology for providing the security countermeasures in IoT environment with some open challenges to incorporate other such technologies in the IoT for improving cybersecurity. BC as an infrastructure for IoT architecture with enhanced performance and security has been proposed by Memon et al. in [29]. In this survey, the authors have presented a comparative survey on cloud-based vs BC-based IoT architecture and identified some research gaps with some other similar technologies such as EC and FC. From the point of design objectives, a systematic survey on BC envisioning secure IoT infrastructure has been presented by Tran et al. in [30].

Fersi et al. developed a comprehensive survey in [31] about the scope of FC from the various aspects of the IoT, including enhanced data computing, network management, interoperability issues, security, etc. A similar review on FC from several perspectives, including threats and attacks countermeasures, has been presented by Atlam et al. [32]. Hamdan et al., in [33], comprehensively review the architecture of IoT based on EC. The survey is very rich from the architectural point of view; however, the threats and attacks analysis of such EC-based architecture is narrowly covered in this survey. Another pragmatic survey with EC-based architecture in IoT covering physical layer aspects is presented by Capra et al. in [34]. In this survey, the authors also cover the security aspects of hardware-based IoT architecture. Knowing the extraordinary effectiveness of the EC in IoT, the most seminal survey on the various simulator that can be used to validate the IoT model has been presented by Ashouri et al. in [35]. This survey is one of the best in its field, covering the EC-based simulation tools in IoT, which can even be exploited for modeling and analysis of threats and attacks in the IoT environment.

One of the most comprehensive surveys in the paradigm of ML to enable security and privacy in the IoT data ecosystem has been presented by Amiri et al. in [36]. This survey considers an ML-based approach for enhancing privacy in the IoT data ecosystem where a three-layer architecture comprising perception, network, and application layers of IoT has been considered. The authors also propose a similar approach of using BC with ML to enhance security on the IoT data ecosystem. The state-of-the-art review on the application of ML for intrusion detection in IoT environment has been presented by Adnan in [37]. The authors consider the three dominant attributes, namely, computational complexity, concept drift, and dimensionality, which are mitigated by integrating ML in IoT, envisioning the security of the IoT-based applications.

Some of the other seminal surveys in this context are summarized in Table 1.


**Table 1.** Some of the key literature surveys and research papers, and their scope.

These surveys are classified based on: (1) scope of threats and attacks analysis in IoT—architecture, protocols/standards, and general; and (2) possible technology adopted as a solution to the threats and attacks in IoT. The last entry of this table presents the scope of the present survey to highlight a clear comparative picture of the contributions of this survey.

### *2.3. Scope of the Present Survey and Contributions*

As discussed in the previous sections, the threats and attacks analysis in IoT is scattered and none of the surveys so far, to the best knowledge of the authors, covers the threats and attacks taxonomy covering architecture, protocols, and standards of IoT with possible countermeasures using rapidly evolving ubiquitous technologies, such as BC, FC, EC, and

ML, simultaneously. The threats and attacks were discussed in general without focusing on architecture and protocols [38–40]. Security concerns were addressed in [38] using EC and in [40] using BC. The research gaps were identified, and some open research problems were proposed in [38,40]. The analysis on threats and attacks based on protocols and standards without addressing security solutions was shown in [41]. On the other hand, [42] neither covers the architecture nor protocols for analyzing the threats and attacks. However, security countermeasures were discussed using BC with open research problems in [42]. In [43], threats and attacks were analyzed based on architecture without any security countermeasures. In [44], threats and attacks were discussed based on architectures with possible security countermeasures using FC and EC, but it does not identify the research gaps. Similar observations can be made throughout the seminal existing surveys discussed in Table 1. A comparative analysis reveals that none of these surveys analyze the threats and attacks covering all aspects, i.e., architecture as well as protocols and standards. In addition, the security countermeasures have not been discussed in any one of the existing surveys using all four ubiquitous technologies, i.e., BC, FC, EC, and ML, simultaneously. An extensive survey on threats and attacks analysis in the context of IoT, its challenges, taxonomy, and possible technological solutions covering the most important aspects of the IoT, such as architecture, protocols, and standards, is presented in this work. The vital contributions of the paper are highlighted below:


### **3. Elementary Overview of an IoT System**

The IoT is an evolving notion as a vast network of interconnected devices and services that store, share, and process data to dynamically adapt to the environment. IoT offers an ocean of opportunities, and so, many organizations aim to have IoT services integrated into their business processes. Before discussing the security threats, vulnerabilities, attacks, etc., it is pertinent to have a keen understanding of the layout of IoT. The emerging IoT technology typically consists of three levels of hardware which are integrated using software [76]. IoT devices, controllers, and peripherals constitute the first level of IoT, gateways and networks are associated with the second level, whereas cloud servers and control devices are part of the third level of IoT. Such a typical IoT system is depicted in Figure 6, followed by a brief discussion of each level.

**Figure 6.** Elementary overview of an IoT system. **Figure 6.** Elementary overview of an IoT system.

### *3.1. IoT Devices, Controllers, and Peripherals*

*3.1. IoT Devices, Controllers, and Peripherals* The first level consists of IoT devices, controllers, and peripherals consisting of sensors, actuators, transducers, etc. Their basic function is to capture real-time data of the outer world and convert them into information for further analysis. These devices can be connected to or implanted in any device that needs to be tracked or mounted in the envi-The first level consists of IoT devices, controllers, and peripherals consisting of sensors, actuators, transducers, etc. Their basic function is to capture real-time data of the outer world and convert them into information for further analysis. These devices can be connected to or implanted in any device that needs to be tracked or mounted in the environment to control the device indirectly.

ronment to control the device indirectly. The IoT devices are embedded devices capable of transmitting information across a network to improve interactions with people and with other smart objects. These smart devices make up the bottom layer of the basic IoT architecture. One of the most important features of IoT devices is their ability to use multiple sensors for different applications. Sensors in IoT gadgets are generally coordinated through the sensor hubs. A sensor hub is a single point of connection that gathers and sends data from multiple sensors to the system processing unit. Gathering data is the foremost step [77]. A sensor hub uses various transport mechanisms such as Inter-Integrated Circuit (I2C) or Serial Peripheral Interface (SPI) to transfer data between the sensors and the applications. A communication channel between sensors and applications is established by these transmitting mecha-The IoT devices are embedded devices capable of transmitting information across a network to improve interactions with people and with other smart objects. These smart devices make up the bottom layer of the basic IoT architecture. One of the most important features of IoT devices is their ability to use multiple sensors for different applications. Sensors in IoT gadgets are generally coordinated through the sensor hubs. A sensor hub is a single point of connection that gathers and sends data from multiple sensors to the system processing unit. Gathering data is the foremost step [77]. A sensor hub uses various transport mechanisms such as Inter-Integrated Circuit (I2C) or Serial Peripheral Interface (SPI) to transfer data between the sensors and the applications. A communication channel between sensors and applications is established by these transmitting mechanisms that accumulate sensor data through IoT devices [78].

nisms that accumulate sensor data through IoT devices [78]. The vulnerabilities associated with some of the sensors in the IoT paradigm are de-The vulnerabilities associated with some of the sensors in the IoT paradigm are described in Table 2.


scribed in Table 2. **Table 2.** A few sensor types and their vulnerabilities.

Position Sensors GPS Location Inference Eavesdropping False Data Injection Magnetic Sensor Sensors are vulnerable to numerous security attacks and threats which might be internal or external depending upon their features [79]. To name a few, information tampering, Man-In-The-Middle Attack (MITM), Distributed Denial of Service (DDoS), jamming, etc., are some of the notable threats to the IoT sensors.

### *3.2. Gateways and Networks*

A gateway for IoT is a system or software program that connects the cloud to controller development boards, actuators, and smart devices [80]. It builds a bridge between the cloud and IoT devices. It systematically connects the field to the cloud. An IoT gateway, either a software application or a hardware appliance, is responsible for transmitting data between the cloud and IoT devices. It serves as a network router, connecting IoT devices to the cloud. It is capable of handling both inbound and outbound traffic. Inbound traffic is used for system management tasks, including upgrading device firmware, while outbound traffic is used to transfer IoT data to the cloud. The IoT gateway provides services to safely accumulate, operate, and filter data for analysis. It aids in the secure and safe transport of confederated data produced by the systems and the devices from the edge to the cloud. Ethernet, Wi-Fi, or a 4G/3G modem are used to link the IoT gateway to the cloud [81]. For data exchange and command transfer, a two-way communication channel is developed with the cloud. In an IoT environment, sensors and devices must logically communicate with other devices through the gateway or redirect the necessary data to the cloud. Some of the key functionalities of the IoT gateway are enumerated below.


Glancing over the number of functions and responsibilities of the IoT gateway, one can easily quote that it is essential to have a secure gateway network to carry out all the enlisted functions safely and efficiently. The gateway is prone to several different kinds of attacks which can be classified into five categories [82]:


The state-of-the-art discussion on these attacks is comprehensively discussed later in this article.

## *3.3. Cloud Servers and Control Device*

Smart devices of the IoT are being deployed at a rapid rate. However, the amount of data they produce makes it difficult to store and process in the local platforms. The unstructured IoT data can be easily stored in a public cloud infrastructure [83]. The scalability provided by cloud computing offers a solution to this problem. Cloud computing provides flexible computing and storage tools that can be used to assist in data management. As a result, this technology can be used to analyze data generated by sensors and IoT devices. Many of the major cloud providers use object storage technology to offer low-cost, scalable storage systems. Cloud computing allows businesses to store and analyze data easily and in real time, enabling them to get the most out of their data. According to a survey conducted by Information Week [84], 65% of respondents said that "the opportunity to satisfy business demands easily" was one of the most significant factors for a company to migrate to the cloud. Since they have high-speed networks with no data ingress fees, the public cloud is an excellent place to store the vast quantities of IoT data generated by

businesses. However, the public cloud has plenty to do. Big data analysis applications that consume and process vast amounts of unstructured content have been added to the product offerings of cloud service providers. This enables companies that can potentially process data more efficiently than a private data center to build highly scalable IoT applications. Depending on the device's networking features, devices can connect to the cloud in a variety of ways. Some of these are cellular, satellite, Wi-Fi, Low Power Wide Area Networks (LPWAN) such as NB-IoT, and direct access to the Internet through Ethernet.

While the cloud has acquired universal popularity, and most IoT applications use cloud services for data storage and retrieval. However, questions about whether cloud technologies are genuinely safe and reliable are continuing to be debated. Nevertheless, cloud risks should also be addressed. The cloud is a public platform used by many people, and there could be malicious users on the cloud who pose a risk to IoT data. The cloud is vulnerable to several attacks such as SQL injection, DDoS, weak authentication, malicious applications, back doors, exploits, etc. [85]. An extensive survey on these aspects is discussed later in this survey.

### **4. IoT Reference Model and Protocol Stack**

### *4.1. Three-Layer Reference Model*

The mitigation of security threats and attacks in IoT can be achieved by understanding the IoT reference model and protocol stack in-depth. There is no widely agreed-upon framework for the IoT [86]. However, different architectures have been suggested by different researchers [87]. The most basic architecture being followed widely is the threelayer reference model consisting of perception layer, network layer, and the application layer, which is illustrated in Figure 7a. The functionality of each of these layers is briefly summarized below.


translation, fortune, medical, environmental monitoring, and global management for all intelligent applications.

### *4.2. Five-Layer Reference Model*

The architecture of IoT has been further improved by decomposing the responsibilities and functionalities of the existing three-layer architecture, resulting in a five-layer architecture [91]. A five-layer architecture consisting of a perception layer, network layer, service layer, operation layer, and application layer is proposed, which is different from that proposed by [92]. The pictorial representation of the five-layer reference model is as shown in Figure 7b. It is worthy to note that the application layer is segregated into three layers, namely, service layer, operation layer, and application layer. The functionalities of the service, operation, and application layers are briefly summarized below, whereas the perception layer and network layer hold the same responsibilities.

	- Operation layer: This is an important layer, especially from the business point of view in IoT. The supervision of services offered by IoT, creating business models, visualization of the data, decision-making, etc., are some of the key responsibilities of this layer. Ensuring QoS across all layers is one of the vital responsibilities associated with this layer. This layer is also responsible for real-time monitoring, control, and evaluation of various application-specific parameters in an IoT environment. plication layer is liable for offering types of assistance and decides a bunch of conventions for message passing at the application level. The application layer serves as a bridge between applications and end clients, allowing them to communicate. It defines the allocation of resources and computation in data production, processing, screening, and feature selection. The application layer is a client-driven layer that
	- Application layer: This layer is primarily responsible for providing service to the end-users related to particular applications. There exists a wide range of applications envisaged using IoT, viz., smart city, smart home, smart agriculture, industry 4.0, healthcare, environmental monitoring, etc. This is the layer through which end users usually interact and pay for the service provided to them. performs various tasks for the clients and offers customized assistance as per a client's pertinent requirements [90]. This IoT layer brings together the industries to create high-level intelligent application solutions such as disaster monitoring, health monitoring, translation, fortune, medical, environmental monitoring, and global management for all intelligent applications.

(**a**) Three-Layer Architecture. (**b**) Five-Layer Architecture.

**Figure 7.** Three-layer vs proposed five-layer architecture of IoT. **Figure 7.** Three-layer vs proposed five-layer architecture of IoT.

### *4.2. Five-Layer Reference Model 4.3. Seven-Layer Reference Model*

The architecture of IoT has been further improved by decomposing the responsibilities and functionalities of the existing three-layer architecture, resulting in a five-layer architecture [91]. A five-layer architecture consisting of a perception layer, network layer, service layer, operation layer, and application layer is proposed, which is different from that proposed by [92]. The pictorial representation of the five-layer reference model is as Even though the architectures of IoT are either application-specific or domain-specific, we propose a more generic IoT architecture that comprises seven layers. The seven-layer generic IoT reference model comprises a perception layer, abstraction layer, network layer, transport layer, computing layer, operation layer, and application layer. The representation of the seven-layer reference model is as shown in Figure 8. Further, the functionality of each of the layers is briefly described below.

shown in Figure 7b. It is worthy to note that the application layer is segregated into three layers, namely, service layer, operation layer, and application layer. The functionalities of the service, operation, and application layers are briefly summarized below, whereas the

• Service layer: This layer envisages facilitating the use of heterogeneous IoT devices, tools, testbeds, platforms, etc., for a wide range of IoT applications. The processing of the data from the network layers is also its responsibility. Generally, the data at this layer are voluminous, for which processing, computing, and analyzing are some

• Operation layer: This is an important layer, especially from the business point of view in IoT. The supervision of services offered by IoT, creating business models, visualization of the data, decision-making, etc., are some of the key responsibilities of this layer. Ensuring QoS across all layers is one of the vital responsibilities associated with this layer. This layer is also responsible for real-time monitoring, control, and evalu-

• Application layer: This layer is primarily responsible for providing service to the endusers related to particular applications. There exists a wide range of applications envisaged using IoT, viz., smart city, smart home, smart agriculture, industry 4.0, healthcare, environmental monitoring, etc. This is the layer through which end users

ation of various application-specific parameters in an IoT environment.

usually interact and pay for the service provided to them.

key challenges to be handled by this layer.

**Figure 8.** The proposed seven-layer architecture of IoT. **Figure 8.** The proposed seven-layer architecture of IoT.


ing for service discovery

Service discovery protocol mDNS, DNS-SD Domain name resolution, client pair-

with this layer. This layer is also responsible for real-time monitoring, control, and evaluation of various application-specific parameters in an IoT environment.

• Application layer: This layer is primarily responsible for providing service to the end-users related to particular applications. There exists a wide range of applications envisaged using IoT, viz., smart city, smart home, smart agriculture, industry 4.0, healthcare, environmental monitoring, etc. This is the layer through which end users usually interact and pay for the service provided to them.

### *4.4. IoT Protocols and Standards*

In the Internet of Things, the communication protocol is a bunch of rules set down for exchanging information between electronic gadgets. Since IoT devices are more resourcelimited/dependent than traditional network devices, the protocol stack in an IoT network must be different from the traditional OSI model. IoT protocols are supposed to be small and compact. The IoT protocol stack can be considered as an augmented version of the layered TCP/IP protocol stack [93]. In recent times, many standardization efforts have been seen to reduce the efforts of all stakeholders of the burgeoning IoT, such as service providers, developers, manufacturers, programmers, operators, etc. To this extent, although there are numerous players, some of the prominent organizations involved are EPC global, the European Telecommunications Standards Institute (ETSI), Internet Engineering Task Force (IETF), World Wide Web Consortium (W3C), and Institute of Electrical and Electronics Engineers (IEEE). The protocols can be broadly grouped into four categories: application protocol, service discovery protocol, connectivity and networking protocol, and other dominant protocols [94]. Some of the widely explored protocols under these categories are summarized in Table 3, whereas a detailed discussion can be found in the seminal work carried out in [41,57], of which we briefly describe some of the key protocols in the following subsections.


**Table 3.** Protocols at various layers of IoT architecture with key functionality.


• EXI: Efficient XML Interchange is abbreviated as EXI. This is an XML representation in a small package. To support XML applications on resource-constrained devices, EXI is described as a technique that uses less bandwidth and improves encoding/decoding efficiency. EXI compression aids in the reduction of document content by creating small tags internally based on the current XML schema, processing level, and context. It assures the tags are optimized for data representation. The document is in binary format, with all of the document's data tags encoded using event codes. Event codes are binary tags that keep their value only in the EXI stream where they are allocated.

### **5. IoT Vulnerabilities, Security Threats, and Attacks**

With the unprecedented growth in IoT devices with rapidly evolving technologies, the new generation IoT-based applications are at risk. Nevertheless, there is an increasing consciousness that the new age of cell phones, computers, and other gadgets might be powerless against malware and assault. Thus, the vulnerabilities, security, and attacks must be comprehensively analyzed to make envisioned IoT a reality.

### *5.1. Vulnerability*

Vulnerabilities are the defects in a framework's design or usefulness that permits the attacker to execute orders, access unapproved information, and launch distributed denial-of-service (DDoS) attack [107]. Attackers can utilize IoT gadgets with existing issues to infiltrate the networks. DNS rebinding attacks, which allow for the processing and ex-filtration of data from internal networks to new side-channel attacks, such as infrared laser inducted attacks against smart devices in homes and workplaces, are among the risks. In IoT systems, vulnerabilities can be found in several places [108].

Hardware and software systems are two central components of IoT frameworks, vulnerable to design flaws. Regardless of whether bugs are identified due to compatibility and interoperability of the equipment or efforts to remedy them, hardware flaws are very difficult to detect and even more difficult to repair. Computer bugs may exist in operating systems, programming software, and control software. Human elements and programming complexity are two factors that contribute to software configuration defects. Human flaws are normally the source of technical vulnerabilities [109]. Miscommunication between the developer and clients, lack of resources, skills, and experience, and a failure to manage and monitor the system can result from a poor understanding of the specifications introducing vulnerabilities in the IoT framework. Thus, vulnerability poses indispensable threats and attacks in the IoT environment. What follows next is the taxonomy of threats and attacks in IoT.

### *5.2. Taxonomy of Threats and Attacks in IoT*

A threat is an activity that exploits a system's security flaws and has a negative effect on it. Humans and the environment are the two main sources of security threats [110,111]. As an example, seismic tremors, typhoons, floods, and fires are all natural hazards that can cause serious damage to computer systems. Few shields can be used to protect against traumatic events since these naturally occurring events cannot be prevented. Backup and contingency planning, for example, are the best ways to protect stable infrastructures from common threats. Human threats are those that humans create, such as malicious threats that are either internal (someone has allowed access) or external (individuals or organizations operating outside the network) in nature and seek to damage or disrupt a system. Following are the different types of human threats:


from high-value information in industries such as manufacturing, banking, and national defense [112].

A taxonomy of threats posing a big concern from a security perspective in the IoT environment is shown in Figure 9.

Compared to the threat that can be intentional or unintentional, the attack is always intentional and malicious to cause damage. Several security attacks persist in the IoT framework, which can be analyzed with respect to the proposed IoT reference model. A taxonomy of attacks in IoT has been presented in Figure 10. These threats and attacks pose severe challenges to the IoT environment from a security perspective. The security concern due to various threats and attacks are categorically described in the following subsections.

### *5.3. Security Concern Due to Threats and Attacks at Different Layers*

### 5.3.1. Security Concern at Perception Layer

Since current sensor management systems and protection schemes are insufficient to protect the sensors, an attacker may use them in various ways. In general, sensor-based threats refer to passive and active malicious actions that are attempted by the manipulation of sensors for their malicious purposes. Different kinds of threats and attacks which cause serious security challenges at the perception layer are eavesdropping, battery drainages, hardware failure, malicious data injection, Sybil threat, disclosure of critical information, device compromise, node cloning, node capture, side-channel attack (SCA), tag cloning, Radio Frequency (RF) jamming, node injection, exhaustion, node outage, etc. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


of the chronicle [114].

• Sybil Attack: The malicious nodes in this can have multiple identities of a genuine node by either impersonating it or with a fake identity through duplication. One such malicious node may have several identities simultaneously or at different instances. • Disclosure of Critical Information: Sensors used in IoT gadgets can disclose sensitive information such as passwords, secret keys, credit card credentials, and so on. These details may be used to violate user privacy or to build a database for future attacks. One such example of this attack is eavesdropping. It is a kind of attack where a pernicious application records a discussion subtly by misusing sound sensors and extracts data from the discussion. An attacker can save the recorded discussion on a gadget or tune in to the discussion continuously. Soundcomber is one of the current instances of eavesdropping over the receiver of a cell phone. In this model, a pernicious application secretively records when a discussion is initiated from the gadget. Since the recording is carried out behind the scenes, a client is completely unaware

**Figure 9. Figure 9.**  Taxonomy of threats in IoT. Taxonomy of threats in IoT.


**Figure 10**. Taxonomy of attacks in IoT. **Figure 10.** Taxonomy of attacks in IoT.

be directed.

placement attacks.

afar.

• Side-Channel Attacks: The assailant gathers information and performs the reverse

• Malicious Data Injection: Attackers take advantage of flaws in communication protocols to insert data into the network [115]. The intruder will tamper with the information required to control the device if the protocol does not verify the integrity of the data. The injection attack may result in code execution or system control from

• Node cloning: In most cases, IoT devices such as sensor nodes and CCTV cameras are developed without hardware defects, given the lack of standardization of the IoT device design. Therefore, for unauthorized purposes, these devices can be easily forged and replicated. This is also known as the cloning of nodes. It can take place in either of the two phases, i.e., production and during operations. An internal attacker can replace an original device with an unauthorized, pre-programmed object in the former case. A node can be captured and cloned during the operational phase. Capturing nodes could further remove security parameters and substitute firmware re-

• Exhaustion attack: Jamming or DoS attacks that have been mentioned before could lead to attacks of exhaustion. In particular, energy consumption can affect the battery-operated devices if an assailant attacks the network continuously. Repeated retransmission attempts could cause collisions with IoT MAC protocols leading to

or ciphertext during the encryption process, but from the encryption devices. Sidechannel attacks the use of certain or all data to acquire the key the device uses. Some instances of such attacks include timing attacks, power or failure analysis, and electromagnetic attacks. The opponent uses data leaks and collects block cipher keys. In the event of the attacks, an intrusion prevention system such as Boolean masking can


### 5.3.2. Security Concern at Abstraction Layer

Different kinds of threats and attacks which cause serious security challenges at the abstraction layer are node replication, illegal access, device compromise, MITM, eavesdropping, spoofing, insertion of rogue devices, information theft, a threat to the communication protocols, data manipulation, device tampering, tag cloning, DoS, DDoS, SCA, traffic analysis, and sleep deprivation. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


the communication presents additional issues in IoT/CPS security design. Cellular technologies such as UMTS, GSM, and LTE, on the other hand, have their own set of security challenges. Because radio baseband stacks are implemented openly, mobile networks are vulnerable to hacking and cyber-attacks. Furthermore, aggressive attackers can use "IMSI Catching" to compromise GSM and UMTS networks.


## 5.3.3. Security Concern at Network Layer

Gateways and networking systems assist in the routing and networking of data packets to their intended destinations. If the gateway communicates using wireless protocols, the attacker will use wireless attacks to link to the gateway or internal network. As a result, the attacker will be able to carry out further attacks, such as ARP poisoning, MITM, packet injection, and sniffing. Different kinds of threats and attacks which cause serious security challenges at the network layer are illegal access, MITM, eavesdropping, spoofing, fragmentation, hello flood, network intrusion, device compromise, node replication, insertion of rogue devices, sinkhole attack, Sybil attack, clone ID attack, selective forwarding attack, blackhole attack, wormhole attack, traffic attack, and RPL exploits. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is presented in [49,57,65,67].


more likely to cause a major incident than a tempering attack, which involves a few affected nodes. As a result of sinkhole attacks, the whole infrastructure base could be controlled.


## 5.3.4. Security Concern at Transport Layer

Different kinds of threats and attacks which cause serious security challenges at the transport layer are jamming, eavesdropping, false data injection, unfair access, congestion, hello flood, DoS, DDoS, SCA, desynchronization, MQTT exploit, session hijacking, SYQflooding, timing attack, etc. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


### 5.3.5. Security Concern at Computing Layer

This part of the IoT infrastructure supports data storage and computer remote control. If cloud servers are not properly configured, they can then lead to the server and smart devices being exploited. Different kinds of threats and attacks which cause serious security challenges at the computing layer are malicious attack, SQL injection, data integrity, virtualization, software modification, illegal access, identity theft, flooding attack in cloud, cloud malware injection, access attack, false data injection, path-based DoS, hole attack, exhaustion attack, cloud outage, signature wrapping, storage attack, etc. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


### 5.3.6. Security Concern at Operation Layer

Different kinds of threats and attacks which cause serious security challenges at the operation layer are fake information, badmouthing, unauthorized access, users' privacy compromise, stealing users' critical information, MITM, secure on-boarding, firmware attack, software attack, illegal intervention, end-to-end encryption attack, interrogation attack, DoS, etc. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


## 5.3.7. Security Concern at Application Layer

The application layer manages the services offered to the clients. This layer serves applications such as telehealth, industrial automation, smart metering, and so on. This layer has its own set of security concerns that are unique to each program. Different kinds of threats and attacks which cause serious security challenges at the application layer are malicious code, software modification, data tampering, SQL injection, disclosure of critical information, cross-site script, identity theft, virus attack, malware attack, spyware attack, flooding, spoofing, code injection, intersection, message forging, DDoS attack, brute force attack, etc. Some of these security threats and attacks are briefly discussed below. Further, a detailed discussion on these threats and attacks is comprehensively covered in [49,57,65,67].


methodologies used to protect IoT applications from information burglary include data isolation, data encryption, privacy management, user and network authentication, etc.


To summarize, the different threats and attacks are reported in Table 4, along with their scope in IoT architecture and protocols, their impact, and references focusing on different detection, prevention, and mitigation strategies. With reference to this table, the following abbreviations are used: PL—Perception Layer, AbsL—Abstraction Layer, NL— Network Layer, TL—Transport Layer, CL—Computing Layer, OL—Operation Layer, AL— Application Layer, AP—Application Protocols, SDP—Service Discovery Protocols, RP— Routing Protocols, NLP—Network Layer Protocols, DLLP—Data Link Layer Protocols, CP—Connectivity Protocols, ODP—Other Dominant Protocols.

**Table 4.** The scope and panoramic view of threats and attacks with detection, prevention, or mitigation strategies in IoT architecture.



**Table 4.** *Cont.*


**Table 4.** *Cont.*

### **6. Security Goals and Roadmap in IoT**

There are certain security objectives that IoT must essentially meet to provide undisputed services. For smooth functioning, IoT applications require secure connections with proper authentication mechanisms and data confidentiality. To ensure information security, one needs to implement the CIA triad—data Confidentiality, Integrity, and Availability. Threats and violations in any of these areas can result in substantial damage to the system, compromise its integrity, and disrupt its activity. To be efficient in implementing effective IoT security, the following primary security objectives must be considered. These security objectives can be achieved with effective methodologies for detection, prevention, and mitigation of threats and attacks pertaining to the IoT ecosystem, described in the next section.


entity's behavior can also be traced through an accountability system, which can help determine the inside story of what occurred and who was ultimately responsible.

### **7. Scope of Security Enhancements in IoT with Burgeoning Technologies**

Now, we review the state-of-the-art methodology to enhance security and privacy in an IoT environment using a few of the ubiquitous technologies such as BC, FC, EC, and ML. Despite some other technologies such as cloud computing, Big Data, embedded system, digital twin, etc., the trend in the literature unanimously shows that BC, FC, EC, and ML have huge potential to answer the security concern in the IoT ecosystem. Further, these technologies are indispensable for the IoT ecosystem, which motivates researchers to address the security concern based on these ubiquitous technologies.

### *7.1. BC for IoT*

BC technology is a network of peer-to-peer nodes that stores transactional records, known as blocks; these blocks consisting of numerous public databases are known as the "chain". The fundamental principle of BC is based on a distributed ledger. IoT devices collect real-time data from sensors, and BC ensures the security of data by deploying a decentralized, distributed, and shared ledger [185]. Any transaction in this ledger is signed with the owner's digital signature, which verifies the transaction and protects it from tampering. As a result, the data in the digital ledger are extremely stable. The BC entries are both chronological and time-stamped. In the ledger, each entry is linked to the previous entry by applying cryptographic hash keys. Individual transactions are stored in a Merkle tree, and the tree's root hash is stored in the BC. Individual transactions are represented by T1, T2, T3, and Tn in the diagram. The cryptographically hashed transactions are stored on the leaf node represented as H1, H2, H3, and so on. The hashes of the child nodes are combined to create a new root hash. The BC stores the final root hash (i.e., Ha and Hb). It can be confirmed whether the transactions associated with the root hash are secure or not, by just verifying the root node. If a single transaction is modified, all hash values on that side of the tree will be affected. The miners verify all the transactions and then a key is produced that allows the most recent transaction to be included in the ledger. This procedure renders the most recent transaction available to all network nodes. It is very difficult and time-consuming for the attackers to hack the blocks as each block is secured using cryptographic hash keys [186]. The miners are only mining to gain their bonuses and have no personal stake in the transactions. The identity of the transaction's owners is unknown to the miners. Furthermore, several miners are working on the same collection of transactions, and they are in fierce competition to link the transactions to the BC. These characteristics enable the BC to serve as a safe, distributed, tamper-proof, and open data system for IoT data. The entire process of a transaction from its inception to its commitment to the distributed chain is elucidated in Figure 11.

In academia and industry, various platforms and frameworks are being built to support the development and maintenance of BC. Ethereum, Hyperledger Cloth, Ripple, and other platforms are examples of this kind [187]. Nevertheless, the simplified general architecture of the BC is as shown in Figure 12.

The following are the key characteristics of the BC that can be exploited to enhance security and privacy in IoT.


commitment to the distributed chain is elucidated in Figure 11.

entries are both chronological and time-stamped. In the ledger, each entry is linked to the previous entry by applying cryptographic hash keys. Individual transactions are stored in a Merkle tree, and the tree's root hash is stored in the BC. Individual transactions are represented by T1, T2, T3, and Tn in the diagram. The cryptographically hashed transactions are stored on the leaf node represented as H1, H2, H3, and so on. The hashes of the child nodes are combined to create a new root hash. The BC stores the final root hash (i.e., Ha and Hb). It can be confirmed whether the transactions associated with the root hash are secure or not, by just verifying the root node. If a single transaction is modified, all hash values on that side of the tree will be affected. The miners verify all the transactions and then a key is produced that allows the most recent transaction to be included in the ledger. This procedure renders the most recent transaction available to all network nodes. It is very difficult and time-consuming for the attackers to hack the blocks as each block is secured using cryptographic hash keys [186]. The miners are only mining to gain their bonuses and have no personal stake in the transactions. The identity of the transaction's owners is unknown to the miners. Furthermore, several miners are working on the same collection of transactions, and they are in fierce competition to link the transactions to the BC. These characteristics enable the BC to serve as a safe, distributed, tamper-proof, and open data system for IoT data. The entire process of a transaction from its inception to its

**Figure 11.** Basics of BC for enhancing security and privacy in IoT. **Figure 11.** Basics of BC for enhancing security and privacy in IoT.

**Figure 12.** The architecture of BC. **Figure 12.** The architecture of BC.

The following are the key characteristics of the BC that can be exploited to enhance security and privacy in IoT. The use of BC in IoT applications has several benefits. The followings are a summary of the main advantages of using BC in IoT applications.


• BC can be used to store the data from IoT devices: The IoT technologies incorporate

• Terminating the centralized cloud server system: BC boosts the security of IoT frameworks by removing the centralized cloud server and establishing a peer-to-peer network framework. Data pirates are mostly interested in centralized cloud servers. BC enables the distribution of data across all the nodes of the network and encrypts

• Forestalling illegal access: Several IoT applications necessitate a lot of contact between different nodes on a regular basis. Since BC communication is based on public and private keys, data can only be accessed by the intended party or node. If an unintended person accesses the data, the content will be nonsensical because it is protected with keys. As a result, the BC data system attempts to address a variety of

• A solution for resource-constrained devices: Because of the limited resources, IoT devices are unable to store large ledgers. There have been different works toward this path to work with the assistance of BC. One of the potential solutions for IoT devices to use BC is proxy-based architecture. By setting up the proxy servers, the data can

where. BC is a promising method for storing and protecting such an enormous amount of data. BC is an apt solution for storing and transmitting data regardless of

security problems that IoT applications face.

the layer in an IoT application.

them.

of the main advantages of using BC in IoT applications.

work framework. Data pirates are mostly interested in centralized cloud servers. BC enables the distribution of data across all the nodes of the network and encrypts them.


### *7.2. FC for IoT*

The Internet infrastructure is being challenged by an unprecedented amount of data generated by IoT. The integration of IoT and the cloud has led to the development of numerous new possibilities on how to process, store, manage, and secure data. These benefits do not fully address all of the problems associated with the IoT. Cloud computing and FC complement each other rather than replace each other [188].

Computing in the fog enables processing, storage, and intelligence control to come within the proximity of the data devices. It uses two frameworks, namely Fog-Device Framework and Fog Cloud Framework [189]. With the Fog-Devices framework, different services can be delivered to a user without involving any cloud servers. Whereas the simple decisions in the Fog-Cloud-Device framework occur at the fog layer, the complex ones occur at the cloud level [190]. The architecture of the Fog-Cloud-Device framework is shown in Figure 13.

The convenience and flexibility of this structure make it possible to offer cloud computing at the network edge. The result is a reduction in distance and improved efficiency while decreasing the amount of data required to be transported into the cloud for processing, analysis, and storage. Comparing the FC with cloud-only models, data traffic between the cloud and network edge is reduced by 90%, and response times for users are cut by 20% [191]. This flexible structure extends cloud computing services to the edge of the network. Thus, it reduces the distance across the network, improves efficiency, and decreases the amount of data needed to transport to the cloud for processing, analysis, and storage.

Using fog technology, data are collected at nodes referred to as fog nodes, and the nodes can process 40 percent [192]. It reduces the latency of IoT devices by offloading traffic from the core network. According to its time sensitivity, data are directed to the cloud, fog, or aggregation nodes. By providing cryptographic computations to IoT applications, fog nodes help secure communication [193].

**Figure 13.** An elementary overview of FC. **Figure 13.** An elementary overview of FC.

The convenience and flexibility of this structure make it possible to offer cloud computing at the network edge. The result is a reduction in distance and improved efficiency while decreasing the amount of data required to be transported into the cloud for pro-FC can provide some solutions to counteract certain security threats and attacks as discussed in the earlier section. More details are provided below to demonstrate how FC can counteract these threats.

be stored in an encrypted format and the encrypted resources can be downloaded

• Forestalling spoofing attack: Spoofing is a type of attack where a foreign node enters the IoT ecosystem and tries to emulate the existing nodes to be seen as a member of the original framework. This foreign node can monitor or inject malicious data into the network. The BC technology appears to be a potential solution for preventing such attacks. Each genuine client or gadget is enlisted on BC, and gadgets can un-

• Forestalling data loss: IoT devices acquire the danger of losing information. There is a possibility that the data are lost by the sender and the recipient due to natural environmental causes. The utilization of BC can forestall such losses as it is impossible

The Internet infrastructure is being challenged by an unprecedented amount of data generated by IoT. The integration of IoT and the cloud has led to the development of numerous new possibilities on how to process, store, manage, and secure data. These benefits do not fully address all of the problems associated with the IoT. Cloud computing and

Computing in the fog enables processing, storage, and intelligence control to come within the proximity of the data devices. It uses two frameworks, namely Fog-Device Framework and Fog Cloud Framework [189]. With the Fog-Devices framework, different services can be delivered to a user without involving any cloud servers. Whereas the simple decisions in the Fog-Cloud-Device framework occur at the fog layer, the complex ones occur at the cloud level [190]. The architecture of the Fog-Cloud-Device framework is

via proxy servers.

*7.2. FC for IoT*

shown in Figure 13.

doubtedly recognize and validate each other.

to eliminate a block once it is included in the chain.

FC complement each other rather than replace each other [188].


## *7.3. EC for IoT*

Both FC and EC share similar responsibilities, such as reducing latency, reducing the volume of the data sent to the cloud, enhancing computational efficacy, incorporating heterogeneity, etc., with a common objective to bring intelligence and computing possibly as close as to the data source. However, they are not the same. They differ in the way they operate and handle the data. For example, usually, FC takes place on the devices to which sensors are connected, such as switches, routers, gateways, access points, etc. On the other hand, EC takes place at the sensors themselves or devices which are at a one-hop distance from the sensor. Thus, the FC nodes are at more distance than the EC nodes.

Contrary to the EC, the data are transmitted from sensors to the FC nodes for processing and then sent back to the edge nodes for appropriate actions. Nevertheless, EC and FC are widely used by many companies as an extension of cloud computing. The main difference between cloud, fog, and edge stems from the location where intelligence and power computation are conducted. In the cloud, more data are processed, and users are comparatively located at a greater distance, requiring a much higher level of data processing [194]. EC uses a small edge server to overcome the problems associated with cloud computing, placed between the user and the cloud. *Sustainability* **2021**, *13*, x FOR PEER REVIEW 34 of 47 are comparatively located at a greater distance, requiring a much higher level of data processing [194]. EC uses a small edge server to overcome the problems associated with cloud computing, placed between the user and the cloud.

> Figure 14 shows the EC architecture's device components, which include edge devices, fog nodes, and cloud data centers [195]. The processing power and analytical capability are provided at the edge itself in an EC framework. An application comprises devices that communicate among themselves and collaborate to calculate data [196]. The IoT application can then minimize the amount of data sent to the outside, whether to cloud or fog nodes, and this will improve the application's security.EC reduces communication costs, as all the data do not have to be moved to the cloud. Figure 14 shows the EC architecture's device components, which include edge devices, fog nodes, and cloud data centers [195]. The processing power and analytical capability are provided at the edge itself in an EC framework. An application comprises devices that communicate among themselves and collaborate to calculate data [196]. The IoT application can then minimize the amount of data sent to the outside, whether to cloud or fog nodes, and this will improve the application's security.EC reduces communication costs, as all the data do not have to be moved to the cloud.

**Figure 14.** An elementary architecture of EC. **Figure 14.** An elementary architecture of EC.

data to a cloud service [199].

*7.4. ML for IoT*

Looking at the threats and attacks causing serious security concerns to the IoT system, the following are possible solutions that can be achieved by incorporating EC with Looking at the threats and attacks causing serious security concerns to the IoT system, the following are possible solutions that can be achieved by incorporating EC with IoT.


• Safety issues: Physical safety can be compromised even if there is just a slight delay

enable data processing to be conducted at the edge nodes rather than sending the

responses, devices can be deployed with EC to examine the abnormalities, process

In recent years, the field of ML has been of major interest. For their development, many domains use ML, and it is also used for IoT security. ML seems to be an excellent

the data, and send them to the data center.

• Safety issues: Physical safety can be compromised even if there is just a slight delay in responses. In the case of sensors that send all of their data and wait for the cloud to act, it may be too late to prevent injuries or deaths. Therefore, to achieve faster responses, devices can be deployed with EC to examine the abnormalities, process the data, and send them to the data center.

### *7.4. ML for IoT*

In recent years, the field of ML has been of major interest. For their development, many domains use ML, and it is also used for IoT security. ML seems to be an excellent way of protecting IoT devices against cyber assaults by offering an approach other than traditional methods to defend against attacks. ML refers to intelligent approaches that use example data or previous experience through learning to optimize performance criteria. Different ML algorithms have been developed to provide some non-traditional solutions to these challenges.

The basic requirement in IoT is the securing of all network-connected systems and devices. The role of ML is to use, train, and prevent data loss in IoT equipment to detect anomalies or to detect any unwanted activity in IoT systems. Consequently, ML provides a promising platform to overcome the problems in securing IoT devices.

Looking at the threats and attacks causing serious security concerns to the IoT system, the following are possible solutions that can be achieved by incorporating ML with IoT.


So far, from the discussion, it can be inferred that there is a huge potential for security enhancement in IoT using burgeoning technologies such as BC, FC, EC, and ML. The scope of possible security enhancements in IoT through the integration of these ubiquitous burgeoning technologies sprawling the appropriate layers is summarized in Table 5. Some of the research papers in literature focusing on security solutions in different capacities covering various aspects of IoT based on BC, FC, EC, and ML are shown in Figure 15. In this figure, three applications of IoT are considered, namely, healthcare, smart devices, and smart grid, for which some of the papers are presented from the literature which covers the security solutions in different capacities based on BC, FC, EC, and ML.

**Table 5.** Scope of security enhancement in IoT using burgeoning technologies. With abundant capabilities in processing, storing, managing the voluminous data, it


**Figure 15.** Some of the application domains of IoT and related work focusing scope for the security enhancement using burgeoning technologies. **Figure 15.** Some of the application domains of IoT and related work focusing scope for the security enhancement using burgeoning technologies [196,206–237].

### **8. Open Research Problems**

Despite a successful journey so far, the IoT has many technological challenges and research issues that are yet to be explored. Some of the prominent research challenges are enumerated below.


### **9. Conclusions**

The introduction of smart computing devices using IoT has made day-to-day lives more convenient. Data analytics, automation, and smart devices have all benefited from the introduction of IoT into human life. Nevertheless, the unprecedented growth in IoT has also been crippled with many vulnerabilities and challenges. Further, the IoT's heterogeneous design expands the attack surface and adds new challenges to an already vulnerable IoT network. The successful compromise of the system's security may have fatal consequences for users. The overall security of the device must be considered to ensure that critical vulnerabilities are mitigated. Policies and protocols must be enforced as much as possible to deter threats and attacks. In this paper, we have presented a most comprehensive survey on IoT from the perspective of security threats and attacks. Further, modern threats and attacks on the emerging IoT infrastructure, security flaws, and countermeasures are discussed in this paper. In addition, a roadmap of using ubiquitous technologies, viz., BC, FC, EC, and ML, for enhancing security in IoT are comprehensively discussed in this paper.

However, due to IoT devices' heterogeneous existence and limitations, any resolution would be ineffective and obsolete. Consequently, due to the evolving nature of technology, it is estimated that more countermeasures and vulnerabilities will be revealed in the near future. As future work, the authors are working on ML and IoT integration to enhance IoT-based applications' security under dynamically varying conditions.

**Author Contributions:** Conceptualization, A.V.J. and B.A.; methodology, A.V.J.; software, R.R.K.; validation, A.P., R.R.K. and A.V.J.; formal analysis, A.V.J. and A.S.; investigation, R.R.K.; resources, R.R.K.; data curation, A.P. and A.S.; writing—original draft preparation, R.R.K.; writing—review and editing, A.V.J. and N.B.; visualization, N.B.; supervision, B.A. and N.B.; project administration, A.V.J., B.A. and N.B.; funding acquisition, N.B. All authors have read and agreed to the published version of the manuscript.

**Funding:** There is no funding available for this research.

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

**Informed Consent Statement:** Not applicable.

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

### **References**


## *Article* **Performance Investigation of a Solar Photovoltaic/Diesel Generator Based Hybrid System with Cycle Charging Strategy Using BBO Algorithm**

**Anurag Chauhan 1,\*, Subho Upadhyay <sup>2</sup> , Mohd. Tauseef Khan <sup>1</sup> , S. M. Suhail Hussain <sup>3</sup> and Taha Selim Ustun 3,\***


**Abstract:** In the current scenario, sustainable power generation received greater attention due to the concerns of global warming and climate change. In the present paper, a Solar Photovoltaic/Diesel Generator/ Battery-based hybrid system has been considered to meet the electrical energy demand of a remote location of India. The cost of the energy of hybrid system is minimized using a Biogeographybased Optimization (BBO) algorithm under the constraints of power reliability, carbon emission and renewable energy fraction. Load following and cycle charging strategies have been considered in order to investigate the performance analysis of the proposed hybrid system. Further, different component combinations of specifications available on the market are presented for detail analysis. The minimum cost of energy of the proposed hybrid system is obtained as 0.225 \$/kWh.

**Keywords:** renewable energy; solar; diesel generator; battery; BBO

## **1. Introduction**

In the current scenario, renewable energy has been recognized as the most effective tool in addressing climate change and global warming [1,2]. The installation cost of solar and wind energy is decreasing day by day and becoming competitive with fossil fuels [3,4]. Additionally, the use of renewable energy technologies offers low carbon emission in the environment with reserves of fossil fuels. Further, integration of two or more renewable energy sources ensures continuous power supply and counterbalances the intermittent behavior of renewable sources [5–10].

Baruah et al. [11] investigated the techno-economic feasibility of a hybrid system which consisted of solar photovoltaic, biogas, wind turbine, syngas and hydrokinetic energy. They proposed this system in order to supply the demand of an academic township using HOMER Pro software. They have used Analytical Hierarchy Process in order to optimize the cost of energy generation and area of system. They have also conducted a sensitivity analysis of the system for the changes of different system parameters. Das et al. [12] minimized the net present cost of a hybrid system consisting of PV array, biogas generator, pumped hydro and battery storage using water cycle algorithm and moth-flame algorithms. They have also performed the comparison of statistical characteristics of the net present cost results obtained by water cycle algorithm, moth-flame algorithm and genetic algorithm.

El-houari et al. [13] designed a solar energy, wind energy and biomass-based hybrid system for ten houses located in remote villages in the Moroccan Fez-Meknes region. They found that the proposed hybrid system offered a reduction of 26.48 tons and 28.814 tons of carbon emission in comparison to the utility grid and diesel generator, respectively.

**Citation:** Chauhan, A.; Upadhyay, S.; Khan, M..T.; Hussain, S.M.S.; Ustun, T.S. Performance Investigation of a Solar Photovoltaic/Diesel Generator Based Hybrid System with Cycle Charging Strategy Using BBO Algorithm. *Sustainability* **2021**, *13*, 8048. https://doi.org/10.3390/ su13148048

Academic Editors: Nicu Bizon, Mamadou Baïlo Camara and Bhargav Appasani

Received: 7 June 2021 Accepted: 14 July 2021 Published: 19 July 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/).

Elkadeem et al. [14] suggested the different combinations of PV array, wind turbines, dieselbased generator and converter for agriculture and irrigation in Dongola, Sudan. They have also performed the sensitivity analysis to evaluate the effect of wind speed, diesel price, interest rate and solar radiation on system economic performance such as net present cost and cost of generation.

Kumar et al. [15] considered three types of battery such as lead acid, nickel iron and lithium ion during the design of hybrid system. They minimized annual cost using Salp Swarm Algorithm and compared the results with other algorithms in obtaining the best optimum solution. Ma et al. [16] proposed the sizing of a hybrid energy system comprised of PV array, wind turbines and battery with consideration of the saturation of renewable sources. They used the saturation factor changing from 0 (only wind system) to 1 (only solar system) in the step of 0.02. Jahangir et al. [17] investigated the economic and environmental assessment of a hybrid system comprised of PV array, wind turbines and biomass generator. They also performed the sensitivity analysis for the changing biomass price, inflation rate and biomass input and evaluated its impacts on the cost of energy and annualized system cost.

Many studies have been performed and investigated the size optimization of the hybrid system. However, many researchers have not considered the battery degradation model during the design of the hybrid system. Additionally, many authors have not accounted for the seasonal changes in the demand and sizes available in market. Renewable fraction and carbon emission in the hybrid system have also not been taken into account by many authors.

The system presented in the work involves a college premises, containing loads of the three hostels and one Sewage Treatment Plant (STP). The design focuses on developing a grid-independent system comprising only SPV and Battery, while diesel generator is working as a backup generator. The maximum number of SPV module is limited depending on the roof area available. The greater size of the SPV system helps to charge the selected size of batteries. The SPV selected through BBO helps to size batteries, which depend on the fulfillment of load primarily during the evening hours. Even then the cumulative usage of the SPV and the batteries are unable to supply the demand due to size constraints of SPV. Hence, sizing of diesel generator is selected to supply the deficit load in case of both load following and cycle charging strategies. In addition, in cycle charging strategy the excess amount of power is fed to charge the batteries. This further reduces the COE of the overall hybrid energy system. The results signify the importance of selecting batteries in place of a diesel generator during the overall operation of the system. This in turn will improve the renewable fraction and reduce the CO<sup>2</sup> emission.

The work also considers number of charging-discharging cycles of the batteries and investment cost, maintenance and operation cost and replacement cost. The parameters that are considered in the work are emission of CO2, variation of load during summer and winter seasons, Energy index ratio (i.e., a measure that load is fulfilled at all times), renewable fraction, fuel consumption, net present cost.

### **2. Hybrid System Configuration**

In the present paper, three hostels and one sewage treatment plant (STP) of the institute Rajkiya Engineering College Banda of India has been taken as the study area. This area is located at the latitude of 25.29◦ N and longitude of 80.57◦ E.

This area receives the yearly average daily solar radiation of 5.262 kWh/m<sup>2</sup> . Accordingly, a hybrid system comprised of photovoltaic array and diesel generator has been considered as presented in Figure 1. This system is proposed to supply the electrical energy requirements of three hostels and one STP of the institute. Power output of solar panels is connected to direct current bus, and power production of the diesel generator is linked to alternating current bus. A set of battery bank is also used in the system to store the additional electricity produced from the generating sources and supply the shortage of load

demand. A bidirectional converter has also been used in the system in order to convert AC to DC and vice-versa.

**Figure 1.** Schematic of Photovoltaic/Diesel Generator/Battery Bank based hybrid system.

A step wise method is required for the performance investigation of the hybrid energy system. It confirms the uninterrupted power supply at the consumer end at minimum system cost. The methodology develops an optimal hybrid system model by addressing all the operational constraints imposed by the user. A step wise description of methodology is summarized as follows:

Step 1: Estimate the electrical energy consumption of each appliance and further, calculate the hourly demand of the selected area.

Step 2: Develop the mathematical model of power output of each generating source and storage system.

Step 3: Choose different configurations of sources.

Step 4: Formulate a framework of objective function and operational constraints of the hybrid system.

Step 5: Take dataset of system components.

Step 6: Perform the simulation of the developed model of the hybrid system for a year. Step 7: Selection of power dispatch strategy.

Step 8: Show the best optimal configuration of the hybrid system with system sizes and cost parameters.

Step 9: Performance investigation of the best configuration.

A flowchart of methodology is prepared with the steps and depicted in Figure 2.

**Figure 2.** Flowchart of methodology adopted.

### **3. Mathematical Model**

Modelling is an important step before optimization as it gives the static and dynamic characteristics of the component. It relates to the output of the system component in terms of many input variables. The modelling of the hybrid system components is explained as below:

### *3.1. Model of Solar PV Array*

The selected area receives a good amount of daily average solar radiation and therefore PV array has the capability to meet the electricity demand of the area. Power output of a PV module is the function of open circuit voltage (VO), short circuit current (Is) and filling factor (F). It can be modeled as follows [18,19]:

$$\mathbf{P\_{FV}^d(t)} = \mathbf{V\_O^d(t)} \times \mathbf{I\_S^d(t)} \times \mathbf{F(t)}\tag{1}$$

Further, the open circuit voltage and short circuit current of PV module depend upon the different module parameters as provided by the manufacturer, and these can be estimated as follows:

$$\mathbf{I}\_{\rm S}^{\rm d}(\mathbf{t}) = \left\{ \mathbf{I}\_{\rm S,STC} + \mathbf{C}\_{\rm i} \left[ \mathbf{T}\_{\rm cell}^{\rm d}(\mathbf{t}) - 2\mathbf{5}^{\rm d} \right] \right\} \frac{\mathfrak{J}^{\rm d}(\mathbf{t})}{1000}. \tag{2}$$

$$\mathbf{V\_O^d(t) = V\_{O,STC} - C\_V \times T\_{cell}^d(t)}.\tag{3}$$

where IS,STC and VO,STC, respectively, represent the short circuit current and open circuit voltage of PV cell at standard test conditions, C<sup>i</sup> and Cv, respectively, represent the temperature coefficient of short circuit current and open circuit voltage, β is solar radiation and Tcell is PV cell temperature.

PV module cell temperature can be calculated with following equation as follows:

$$\mathbf{T\_C^d(t) = T\_A^d(t) + \frac{\text{NOCT} - 20^0 \text{C}}{800} \times \boldsymbol{\upbeta}^d(t)}.\tag{4}$$

where T<sup>A</sup> is ambient temperature and NOCT is nominal operating cell temperature.

### *3.2. Model of Diesel Generator*

Diesel generator is operated to serve the peak demand of the area. Fuel required (Q) in order to operate DG depends upon rated power of DG and can be modeled as follows [20,21]:

$$\mathbf{Q}\_{\mathbf{t}}(\mathbf{t}) = \mathfrak{a}\_{\mathbf{D}\mathbf{G}} \mathbf{P}\_{\mathbf{D}\mathbf{G}}(\mathbf{t}) + \mathfrak{B}\_{\mathbf{D}\mathbf{G}} \mathbf{P}\_{\mathbf{D}\mathbf{G},\mathbf{rated}} \cdot \tag{5}$$

where PDG is power yield of DG at a particular time, PDG,rated is rated power of DG, βDG (0.08145 l/kWh) and αDG (0.246 l/kWh) are constants of DG.

### *3.3. Model of Storage System*

The storage system is essential for energy balance in the hybrid system. It acts as a tool to mitigate the gap between energy generation and demand. In the present system, battery bank has been considered as a storage system. Battery capacity at a particular instant depends upon the previous capacity and difference between total generation and load demand.

A battery mostly works in two states viz. charging and discharging. In charging state, the battery stores surplus power supplied by sources and the current state of battery EB(t) can be calculated as follows [22–24]:

$$\mathbf{E}\_{\rm B}(\mathbf{t}) = \mathbf{E}\_{\rm B}(\mathbf{t} - \mathbf{1}) + \mathbf{E}\_{\rm CCO}(\mathbf{t}) \times \boldsymbol{\eta}\_{\rm CHG} \tag{6}$$

In discharging state, demand is more than the total electricity generation and the battery bank storage at hour 't' can be estimated as:

$$\mathbf{E\_{B}(t)} = (1 - \sigma) \times \mathbf{E\_{B}(t-1)} - \mathbf{E\_{Required}(t)}\tag{7}$$

$$\mathbf{E}\_{\text{Required}}(\mathbf{t}) = \frac{\mathbf{E}\_{\text{NL}}(\mathbf{t})}{\eta\_{\text{INV}} \times \eta\_{\text{DCHG}}} \tag{8}$$

$$\mathbf{E\_{NL}(t)} = \mathbf{E\_{Demand}(t)} - \left[\mathbf{E\_{SPVS}(t)} \times \eta\_{\rm Inv} + \mathbf{E\_{DG}(t)}\right] \tag{9}$$

where ERequired(t) is hourly energy required from the battery to meet the load (kWh), ENL(t) is net shortfall load, σ is hourly self discharge rate, ECCO is charge controller output, ηCHG and ηDCHG, respectively, represent charging efficiency and discharging efficiency of battery.

The *n S B* is the series connected batteries, which depends on the nominal DC bus voltage (VBUS) and individual nominal voltage of battery (Vnom). Here the VBUS is considered to be 48 V.

$$\mathbf{m\_{B}^{S}} = \frac{\mathbf{V\_{BUS}}}{\mathbf{V\_{nom}}} \tag{10}$$

The battery bank nominal capacity (Cn) is directly proportional to the number of batteries (NBAT) and nominal capacity of each battery (CB), indirectly related to the number of series connected batteries.

$$\mathbf{C}\_{\rm n} = \frac{\mathbf{N}\_{\rm BAT}}{\mathbf{n}\_{\rm B}^{S}} \mathbf{C}\_{\rm B} \tag{11}$$

Number of cycles of failure over minimum average depth of discharge for a period of battery bank is depicted in Figure 3. The battery bank replacement hours (NBR) can be calculated by determining the total number of cycles in which a battery (Ncycles) can be operated. Here, De is the yearly average minimum capacity of the battery bank achieved over a day and OBatt is the number of days for which the battery should be operated [25]. It is used to determine the life of the battery. A curve fitting toolbox is used to generate the coefficient values of the equation. Battery degradation model equations are described as:

$$\mathbf{O\_{Batt}} = \mathbf{a\_1} \times (\mathbf{D\_e})^{\mathbf{b\_1}} + \mathbf{c\_1} \tag{12}$$

$$\begin{array}{c} \text{N}\_{\text{BR}} = \text{n} \times \text{365Ncycles\\_DOD\%}\\ \text{a1} = 1.582 \times 105, \text{ b2} = -0.9964, \text{ c1} = -997.1 \end{array} \tag{13}$$

**Figure 3.** Number of cycles of failure over minimum average depth of discharge for a period of battery bank [25].

### *3.4. Mathematical Model of Charge Controller*

The charge controller makes energy balance among different system components and its model is described as:

$$\text{E}\_{\text{CCO}}(\mathbf{t}) = \text{E}\_{\text{EE}}(\mathbf{t}) \times \eta\_{\text{CC}} \tag{14}$$

where ECCO (t) and ECCI (t), respectively, represent the hourly output and hourly input to charge controller (kWh), EEE (t) is amount of excess energy from sources (kWh) after serving the demand and ηCC is charge controller efficiency.

### **4. Problem Formulation**

Minimization of system cost of energy is formulated and considered as an objective function for the present paper. Various constraints such as expected energy not supplied, carbon emission, renewable energy fraction and total net present cost have been incorporated during system optimization.

### *4.1. Objective Function*

Cost of generation is the fundamental financial parameter in order to evaluate the techno-economic feasibility of the hybrid system. It can be calculated with annual system cost (ASC) and energy generation (EGen) as:

$$\text{COE} = \frac{\text{Annual System Cost} \,(\text{ASC})}{\sum\_{t=1}^{8760} \text{E}\_{\text{Gen}}(\text{t})} \tag{15}$$

Annual system cost of hybrid system is a function of net present cost (NPC) and can be estimated as [26]:

$$\text{ASC} = \text{NPC} \times \left[ \frac{\text{dr} (1 + \text{dr})^{\xi}}{(1 + \text{dr})^{\xi} - 1} \right] \tag{16}$$

where dr is discount rate and ξ is project lifetime.

### *4.2. Operational Constraints*

System optimization has been investigated under operational economic and reliability constraints which are summarized as follows:

### 4.2.1. Power Reliability Constraint

At any time, when part of the load demand has not been met from the available generation, energy not supplied is calculated. It is the function of demand not met (NL) and duration of the period (T) as follows [27]:

$$\text{EENS} = \sum\_{i=1}^{8760} (\text{NL} \times \text{T}) \tag{17}$$

### 4.2.2. Economic Parameter Constraint

Constraints of economic parameters have been taken during system analysis. Total net present cost, total recurring cost (TCrec) and non-recurring cost (TCnon-rec) of the system have been evaluated. These parameters can be calculated as follows [28]:

$$\text{NPC} = \text{C}\_{\text{Inv}} + \text{TC}\_{\text{rec}} + \text{TC}\_{\text{non-rec}} \tag{18}$$

Total recurring cost of system changes with escalation rate (er) and discount rate. It can be estimated as follows [28]:

$$\text{TC}\_{\text{rec}} = \text{C}\_{\text{rec}} \frac{\left[\frac{1 + \text{er}}{1 + \text{dr}}\right] \left\{ \left[\frac{1 + \text{er}}{1 + \text{dr}}\right]^{\xi} - 1 \right\}}{\left[\frac{1 + \text{er}}{1 + \text{dr}}\right] - 1} \tag{19}$$

Total non-recurring cost of system can be calculated by using Equation (19) as follows:

$$\text{TC}\_{\text{non}-\text{rec}} = \sum\_{\mathbf{y}=1}^{\mathbf{y}=\text{n}\_{\text{rep}}} \text{C}\_{\text{Inv}} \left[ \frac{1+\mathbf{er}}{1+\mathbf{dr}} \right]^{\mathbf{y}\ast\mathbf{n}\_{\text{freq}}} \tag{20}$$

where nrep is total replacement of system component (in Nos.) and nfrep is year number of first replacement of system component.

### 4.2.3. Renewable Energy Fraction

Renewable energy fraction in total system generation ensures the sustainable generation of a hybrid system. It depends on energy generated by diesel generated (EDG) and total energy generation (TEGen). It can be calculated as:

$$\text{Renerable Energy Fraction}(\%) = \left[1 - \frac{\sum \text{E}\_{\text{DG}}}{\sum \text{TE}\_{\text{Gen}}}\right] \times 100\tag{21}$$

### 4.2.4. Total Carbon Emissions

Carbon emission is generated from the use diesel generator in the considered system. Total carbon emission (TECarbon) can be estimated as:

$$\text{TE}\_{\text{Carbon}} = \sum\_{\mathbf{t}=1}^{\text{T}} \sum\_{\mathbf{t}=1}^{\text{T}} \text{E}\_{\text{Carbon}} \times \text{P}\_{\text{DG}}(\mathbf{t}) \tag{22}$$

where ECarbon is carbon emission produced by 1 kWh electricity generation by DG.

### **5. Energy Management Strategy**

### *5.1. Load Following Strategy*

The load following strategy is initiated by determining the demand and solar power generation available in an hour. If the generated solar power in an hour is more than the demand, then the battery state of charge is checked. If the battery bank state of charge is found to be less than the maximum battery state of charge (SOC), the battery bank is charged, otherwise it is fed to the dump loads. In case of demand exceeding the generated solar power, the battery bank is operated until its SOC maximum value is not reached. If the battery bank is unable to fulfill the demand, the diesel generator is operated to supply only the net demand, without charging the battery bank. If any one of these conditions is matched, the iteration as unit time in an hour is updated and the process continuous until the year is complete. Figure 4 shows the methodology of operating load following strategy.

**Figure 4.** Working of load following strategy.

### *5.2. Cycle Charging Strategy*

In the cycle charging strategy, the hourly load is compared to the available solar power output. If the generated power becomes equal to the demand then the time is updated. If the solar power in an hour is more than the demand then the battery state of charge is checked. If the battery bank state of charge is found to be less than the maximum SOC, the battery bank is charged, otherwise it is fed to the dump loads. In case of demand exceeding the solar power generation, battery bank is operated to fulfill the net demand. If the battery bank is unable to supply the demand, diesel generator is operated. Here the diesel generator is operated to supply the demand as well as charge the battery bank, until the SOC maximum value has not been reached. If any one of these conditions is matched the iteration as unit time in hour is updated and the process continuous until the year is complete. The Figure 5 shows the methodology of operating cycle charging strategy.

**Figure 5.** Working of cycle charging strategy.

### **6. Biogeography-Based Optimization (BBO) Algorithm**

The biogeography-based optimization technique has unique features of longevity in the solutions survival nature as solutions in genetic algorithm die after each iteration. It also adds the merit of mutation, which removes the clustering effect seen while implementing BBO [29,30]. Due to these advantages, BBO is selected over GA and PSO for optimizing the cost of energy of the hybrid energy system.

BBO optimization procedure is as follows:

Step 1: Initialize the generation limit, population size and mutation rate.

Step 2: The fitness function values of all the individuals are evaluated.

Step 3: If the termination criterion is not met.

Step 4: Best habitats are saved in the temporary array.

Step 5: HSI (Habitat suitability index) is mapped using the number of ì, € e and S species for each habitat.

Step 6: Probabilistically choose the immigration island.

Step 7: SIV (Suitability index variables) are randomly migrated based on the island selected in Step 6.

Step 8: The population is randomly mutated.

Step 9: The fitness functions of all the individuals are evaluated.

Step 10: The population is sorted and arranged from best to worst.

Step 11: The best values of habitat as stored in temporary array replaces the worst values.

Step 12: Step 3 is again followed for continuing the next iteration.

Step 13: End the algorithm.

A flowchart of process of BBO algorithm is shown in Figure 6.

**Figure 6.** Biogeography based optimization process.

The parameters-chosen BBO technique have mutation probability of 0.4, population size is 50 selected due to better convergence of the objective function, and problem dimension is 3, while the available rooftop area limits the maximum number of solar modules to 500 modules. The variations of the total number of batteries and diesel generator to be selected are limited by 500 and 50, respectively.

### **7. Results and Discussions**

### *7.1. Input Technical and Economical Dataset*

Study area receives the annual average solar radiation of 5.26 kWh/m<sup>2</sup> per day. Peak solar radiation of 7.05 kWh/m<sup>2</sup> is found during the month of May and minimum solar radiation of 4.08 kWh/m<sup>2</sup> has been received during the month of December. A maximum temperature of 45 ◦C is recorded in the month of May and the lowest temperature of 12 ◦C is observed during the month of January. Monthly solar radiation and monthly temperature distribution of the selected site are depicted in the Figures 7 and 8, respectively.

**Figure 7.** Monthly solar radiation of the selected site.

Load profile on an hourly basis is shown in Figure 9. Peak electrical load of 89.80 kW and 59.60 kW are estimated during summer season and winter season, respectively. Meanwhile, a minimum load demand of 14.5 kW and 5.43 kW is observed during the summer season and winter season, respectively.

Detailed specifications of system components available in market have been considered in the present paper as given in Table 1. Three types of DG (20 kVA, 30 kVA and 50 kVA) have been taken during the analysis. However, diesel price of \$1.05 per Litre is same for all three DG. Two types of PV module (0.375 kW and 0.32 kW) have been used in the study. Capital cost of these modules is \$273.79 and \$205.34, respectively. Three types of battery sizes, 100 Ah, 150 Ah and 250 Ah have been considered during the performance analysis. Capital cost of these battery types is \$96, \$130 and \$219, respectively. Specifications of charge controller and bidirectional converter are also given in Table 1.

**Figure 9.** Hourly load demand of the study area.



*7.2. Results and Discussions*

The hybrid system model has been developed in MATLAB (R2019b, MathWorks, Natick, MA, USA) covering the specifications of individual components. Further, 18 combinations have been modelled and simulated for a year. The optimum configuration of these combinations is obtained using the BBO algorithm in MATLAB. Further, these combinations have been compared based on cost of energy, carbon emission, fuel consumption, renewable fraction and total operating hours. Descriptions of results are explained as follows.

### 7.2.1. Cost of Energy

For each combination of battery, solar photovoltaic module and diesel generator, the cost of energy (COE) is evaluated using cycle charging (CCS) and load following strategies (LFS). A comparison between the cost of energy for both the strategies is shown in Figure 10. Cycle charging strategy shows the minimum COE for each combination of components. The 13th combination provides the least COE of 0.225 \$/kWh, with COE of 17th combination at 0.229 \$/kWh and 7th combination as 0.231 \$/kWh.

**Figure 10.** COE for each combination of specifications using BBO with CCS and LFS.

### 7.2.2. Optimum Size for Each Combination

Each combination has been simulated and optimized using BBO algorithm in MAT-LAB. Results for each combination are given in Table 2. From the Table 2, it has been found that combination 13 offers minimum cost of energy generation. The net present cost of this combination is calculated as \$1,204,972. Combination optimum sizes consist of five numbers of DGs, 519 numbers of PV modules and 307 numbers of batteries.

### 7.2.3. Carbon Emission

The total CO2 emission is found to be at minimum about 155,184 kg/yr in case of the 13th combination as compared to the 7th and 17th combinations. This is due to the number of operating hours of diesel generator, about 732 for 13th combination as compared to 776 in case of the 7th combination. The carbon emission for each emission has been calculated and given in Table 3.

### 7.2.4. Renewable Fraction

Renewable fraction is directly dependent on the total output power of diesel generator and the renewable power output. This further depends on the overall fuel consumption by the diesel generator. Hence, higher fuel consumption of DG means minimum renewable fraction, the 13th combination has a maximum renewable fraction of 55% compared to 53% and 52% for the 7th and 17th combinations, respectively. Renewable energy for each combination is given in Table 3.

### 7.2.5. Total Fuel Consumption

The total fuel consumption of DG is minimal in the case of the 13th combination of 60,418 l, versus 7th and 17th combinations.


**Table 2.** Optimum size for each combination of hybrid system.

Note: Bold represents minimum cost of energy generation.

**Table 3.** Carbon emission, total fuel consumption and renewable fraction for each combination.


7.2.6. Total Operating Hours of DG and Battery Bank

The diesel generator needs to be operated 732 h and battery bank of 4772 h over a year in case of the 13th combination. The generators are operated for the minimum amount of time in this case while using cycle charging dispatch strategy. Operating hours for each combination are given in Table 4. Additionally, it has been found that replacement of battery bank is found to be minimal in case of CCS compared to LFS. The battery bank is to be replaced in every 3 years with CCS while battery banks are replaced after 2 years with LFS.


**Table 4.** Total operating hours of DG and battery bank.

### 7.2.7. Convergence of BBO Algorithm

Convergence plot using BBO algorithm is shown in Figure 11. The results attained using BBO for the 7th, 13th and 17th combinations, by considering COE as the minimization objective function. For the 13th combination the COE is 0.225 \$/kWh compared to 0.229 \$/kWh and 0.231 for the 17th and the 7th combinations, respectively.

**Figure 11.** COE convergence plot with BBO for the 7th, 13th and 17th combinations.

### **8. Conclusions**

In the paper, performance investigation of Solar Photovoltaic/DG/Battery-based hybrid system has been performed for the energy access of three hostels and one STP of an institute. In total, 18 combinations of system components sizes available on the market have been considered during the analysis. Cost of energy generation of hybrid system has been optimized using BBO algorithm.

After simulation, it has been found that the cycle charging strategy offers a lower cost compared to the load following strategy. All considered combinations have been compared based on COE, renewable fraction, carbon emission and operating hours of DG's. The optimum configuration offers minimum COE of 0.225 \$/kWh, a renewable fraction of 55% and carbon emission of 155,184 kg/yr. Further, government subsidy on the PV module can also further reduce the system cost. In future works, the addition of utility grid and waste utilization, collected from the campus to energy are proposed to make a more sustainable hybrid system.

**Author Contributions:** Conceptualization, A.C., S.U., M.T.K., S.M.S.H. and T.S.U.; methodology, A.C., S.U., M.T.K., S.M.S.H. and T.S.U.; software, A.C. and S.U.; validation, A.C., S.U., M.T.K. and S.M.S.H.; formal analysis, A.C., S.U. and M.T.K.; investigation, A.C. and S.U.; data curation A.C. and S.U.; writing—original draft preparation, A.C. and S.U.; writing—review and editing, M.T.K., S.M.S.H. and T.S.U.; visualization M.T.K. and S.M.S.H.; funding acquisition, T.S.U. 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.

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

### **References**


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

www.mdpi.com ISBN 978-3-0365-6370-1