Flexible Energy Storage for Sustainable Load Leveling in Low-Voltage Electricity Distribution Grids with Prosumers

Abstract

The sustainability of the energy sector is linked today with the diminishing of the reliance on fossil fuels and on the large-scale adoption of renewable generation. Medium- and low-voltage electricity distribution grids see the proliferation of microgrids that supply consumers able to generate electricity with local installations of PV panels. These consuming and generating entities, called prosumers, use the local generation for their own consumption needs and are exporting the surplus in the grid, modifying the typical steady state operation conditions. For mitigating this inconvenience, local storage equipment can be used, which also helps the prosumers to reduce their costs and preserve the sustainable operation of the distribution infrastructure. The literature shows that by optimally using storage in microgrids, the deterioration in quality and security of supply can be minimized in the presence of prosumers. This paper presents a study regarding local storage management in prosumer-enabled microgrids, seeking to find the optimal configuration of community (shared) storage systems that charge batteries overnight, during low consumption hours, providing load leveling opportunities and energy loss minimization. A study case performed on a real low-voltage electricity distribution network (LVEDN) shows the performance of the proposed optimization.

1. Introduction

The growing awareness of today’s society about the adverse effects of human activity on the planetary ecosystem and climate has prompted the adoption of new policies and technological advances in multiple areas of concern. One of these areas is the energy sector, which is traditionally seen as a major pollution source, the energy supply being responsible, in 2010, with 35% of world’s greenhouse gases (GHG) emissions [1]. International initiatives such as the earlier Kyoto Protocol and the EU’s current Recovery and Resilience Plans started in response to the COVID-19 pandemic have stimulated the world’s countries to adopt and promote energy efficiency and the reduction of GHG emissions through, among others, generation and sharing of clean electricity by prosumers subsidized through government aid schemes [2].

The proliferation of small residential prosumers seen in the last years as an effect of these initiatives has revealed a series of negative consequences on the security and quality of supply, especially in older electric networks that were not prepared for this evolution. Because of climatic and environmental factors, the renewable sources of energy, particularly wind and photovoltaics (PV), have intermittent and unpredictable electricity generation, which results in adverse effects that become more obvious as they contribute with an increasing amount to the load flow in that network. The maintenance of transient and dynamic stability within the system may be hampered by the unpredictable changes in power produced by this renewable electricity [3]. Voltage transients, frequency drift, and harmonics are three power quality issues that are typically related to renewable energy sources [4]. At the consumer side, where the supply is often unbalanced, as one-phase consumers are supplied from three-phase grids, the unbalance in the grid can increase [5].

1.1. Load Leveling and Local Electricity Storage

Part of the solution for improving the quality of supply deteriorated by the presence of renewables in a local grid is the use of local electricity storage, with benefits for both parties involved: the consumer, which is able to defer the consumption of locally generated electricity to time intervals where the option to import from the grid is more expensive, and the network operator, which can mitigate, by leveling the daily demand profile, several problems such as load unbalance, voltage quality decrease, and energy loss increase in peak demand hours. Additional investment in network reinforcement can be deferred. All these benefits add up to a more sustainable operation of the electricity supply system.

Load leveling (Figure 1) is a technique for balancing the daily profile of electricity.

This can be performed by peak shifting, i.e., by advancing or delaying the peak load until the power supply system can accept additional load, or peak load shaving (reduction). This practice is not new, and it has been used to reduce peak demand and its associated energy expenditures in a variety of industrial and large-scale commercial facilities but in recent years has drawn significant attention due to the rapid growth of renewable energy facilities, since, as renewable energy continues to expand into a more prominent source of power in the grid, it becomes increasingly necessary to convert the variable and intermittent power output into a more steady and reliable source. If the balance between the local consumption and generation cannot be managed, resulting in significant reversed power flows and adverse effects on the local grid, energy storage systems (ESSs) must be implemented. Typically, the ESS is charged when the supply is only supporting a small load and when electricity costs are low, like at nighttime, resulting in filling the ‘valley’ in demand existing in this time interval. When the demand is rising, it is discharged to supply extra power during times of increased loading. Electricity bills can be reduced by using this method. Additionally, the aggregate demand of the electricity supply system is lowered, decreasing the need for supplementary generation capacity installed in large units and delivered over long distances.

1.2. Literature Review

As the prosumers have been considered lately as instrumental for the achievement of several goals of a sustainable society, such as the alleviation of energy poverty or advancing efficient energy distribution and Smart Grid technologies [6], ample research has been directed to the study of the efficient integration of local residential generation into the existing electricity distribution infrastructure. Because local storage can smooth this integration, with multiple economical and technical benefits, as described above, many of these studies consider the role and optimal operation of prosumers equipped with ESS capabilities.

Although the literature describes several types of energy storage [7], the preferred solution for small residential prosumers is the use of local batteries.

The scenarios in which the influence of storage on the behavior of prosumers is studied are diverse. The complexity of the systems under scrutiny begins from single households, where the local generation and use of storage is managed using Home Energy Management Systems (HEMSs). Paper [8] presents the case of a PV prosumer, while in [9], wind generation is analyzed, showing quantifiable benefits in using local storage. Small microgrid configurations consisting of a limited number of buses and load or generation points are sometimes investigated, as in [10], but entire networks that supply communities are preferred in case studies. These are modeled from generic electrical networks [11,12] or real distribution systems [13]. The impact of using storage is discussed from the technical standpoint, where parameters such as the analysis of transient states [13], voltage regulation [14], energy loss [15], or network security [16] are evaluated, but also from the economic perspective, modeled with indices such as the cost of electricity [8]. If the prosumer + ESS behavior is studied at the community level, often the study involves the creation and operation of a local electricity trading market. In this case, technical and economic or social objectives are combined. In [17], the minimization of the unfair benefit distribution that local market participants can gain through P2P energy trading is sought, while in [18], the focus is on the minimization of the cost of energy for the community. A separate entity acting as an aggregator is seen as having the potential to incentivize electricity trading between prosumers and passive consumers [16] or between the community and the wholesale market [19]. The use of storage also enables the communities of prosumers to participate in other services enabled by Smart Grid structures, such as collective or collaborative Demand Response [20,21] or virtual energy storage systems (VESS), where other energy carriers such as HVAC or thermal inertia of buildings can be used as collective storage [22,23]. In other studies, the local storage can be optimally used for improving the operation of EV station charging [24].

When discussing the presence of storage in grids with prosumers, one important aspect is the modeling of storage operation. In [25], battery degradation is predicted for storage operated in a microgrid. In [26], one of the objectives of an intelligent energy management system is the extension of life for the batteries by optimizing their charge–discharge cycles. In [27], the usage of the storage capacities is reduced to lower microgrid operation costs.

If aggregators are used to manage the collective energy surplus of a prosumer community, a similar entity can be considered for storage. While the classic approach is to let each prosumer decide how it uses its own storage reserve, increased benefits can be extracted from sharing this resource at the community level. Several studies, such as [28], discuss the alternatives of competitive and cooperative storage schemes that show advantages for the latter option. In [29], the individual batteries are aggregated under a virtual battery energy storage system (VBES), and a life cycle analysis shows quantifiable environmental benefits. In [30], the use of shared storage is shown to lead to increased self-consumption and self-sufficiency and lowered costs.

As the challenges encountered in integrating renewables and storage in electricity distribution systems can be often formulated as optimization problems, the tools used in solving them are diverse. They range from exact optimization techniques such as the gradient descent [17], mixed-integer linear programming [18], or the alternating direction method of multipliers [22], to simulation tools like MATLAB-Simulink [14], game theory [30,31], several types of artificial neural networks (ANNs) such as complex-valued ANNs [32], Radial Basis Function (RBF) [33], Long Short-Term Memory Networks (LSTM) [34], or layer-recurrent networks [27], or metaheuristic methods such as the Genetic Algorithm (GA), Particle Swarm Optimization (PSO) [11], Coyote Search [12], and Whale Algorithm [15], often using multiple stages of optimization for complex problems. For instance, paper [35] optimizes the microgrid planning for a scenario involving smart prosumers, electric vehicles, and storage optimization in three stages considering also a stochastic component, while a metaheuristic algorithm is used in [24] to find the number and power of charging points, the installed area of the PV panels, the size of required storage, and the power requirement from the grid in a fast-charging station for electric vehicles. In [34], a two-objective optimization problem (maximizing simultaneously the resilience and electricity generation of the microgrid) entails two preliminary steps, (1) predicting the occurrence of outages with a Monte-Carlo method and (2) wind and solar irradiance forecast using an LSTM ANN, followed by optimization with LSTM-aided PSO.

1.3. Contributions in This Paper

The research presented in this paper is based on the algorithm developed by the authors in [36] that investigated the optimization of storage placement in LVEDNs. The work used a genetic algorithm approach, determining the optimal placement of battery storage in three scenarios:

• A concentrated community storage system, managed by the Distribution Network Operator (DNO), installed at one bus, with unbalanced use of storage on the three phases, for minimizing the investment/operation cost and simplifying the storage management.
• A distributed community storage system managed by the DNO, with batteries placed in the buses and on phases optimally chosen by the algorithm.
• Batteries placed and managed individually by prosumers.

In all three cases, the algorithm optimized the configuration of the storage system with the objective of minimizing the active energy losses in the grid over a 24 h period and considered that the batteries were already charged at the start of the analysis (in scenarios 1 and 2) or that the batteries were charged using the available prosumer surplus at daytime (scenario 3).

This paper extends this stage by taking into consideration the effect of loading the batteries at nighttime, in the first two scenarios, where the charging is managed by the DNO. The main contributions of this paper are:

• The conceptualization of the mathematical model for night storage management.
• The adaptation of the GA used in [36] for the new assumptions used in the optimization problem.
• A comparison between the influence of day and night battery charging on the daily energy losses, using the results from [36] and the new assumptions.
• A case study performed in a real LVEDN from Romania.
• Discussions on the advantages and disadvantages of each storage solution investigated in the study.

The third scenario is used only for comparison purposes, as, in this case, the charging of the batteries takes place during daytime and has no influence over the valley period. Its characteristics were studied in the previous paper.

2. Materials and Methods

Low-voltage residential electricity distribution networks are operated with four-wire circuits that supply unbalanced demand in their nodes. The demand is unbalanced in space, as the one-phase consumers are unevenly distributed at connection points and in time because of the variable needs of each consumer. Lately, the local consumption is accompanied by generation from PV panels from prosumers that want to achieve independence from the grid or lower their electricity bills. In a specific network, several prosumers can be active at certain times and can generate a surplus of electricity that normally would be injected back in the grid. These supplementary power flows can cause additional energy losses that can be avoided if storage is used. The algorithm proposed by the authors in [36] used a fixed number of equal capacity one-phase storage batteries that were placed in the network according to three distinct scenarios:

Managed by the DNO:

• S1: one-bus, multiple phase storage, where all the batteries are concentrated in one bus and can be divided unevenly between the phases A, B, and C.
• S2: multiple bus, multiple phase storage, where individual batteries can be placed at different buses, on different phases, at the choice of the DNO.

Managed by the prosumers:

• S3: multiple bus, multiple phase storage, where the batteries are placed and managed by the individual prosumers, according to their needs.

In all these cases, the algorithm must find the optimal placement of the batteries so that the active energy losses occurring in the steady state operation of the LVEDN over a given time interval will be minimal.

The results presented in the study have shown that the charging of the batteries at daytime in scenario S3 resulted in a high increase of losses in the network in the charging interval, while the other two scenarios have shown minimal losses in the same time interval. Thus, the next step is to assess the effect of shifting the battery charging to nighttime. In this way, several advantages could be obtained:

• Load leveling by shifting the load associated to battery charging in valley intervals, where the aggregated demand in the LVEDN is minimal.
• Energy loss minimization over an interval of 24 h due to the load leveling.
• Cost reduction for the LVEDN operator by charging the batteries at nighttime.

2.1. The Storage Placement Modeled as an Optimization Problem

For an LVEDN with NB buses, NC consumers, and NPS prosumers, operated for H time intervals, the storage placement optimization problem has been formulated as follows: for each storage battery s_i, i = 1,…, NSS: find the bus b_i and phase ph_i of placement:

$$\left[\left\{b_1,p{h}_1\right\},\hspace{1em}\left\{b_2,p{h}_2\right\},\hspace{1em}\dots ,\hspace{1em}\left\{b_{NSS},p{h}_{NSS}\right\}\right]$$

(1)

to minimize the objective function:

$$\mathit{min}\left(\mathrm{\Delta }W\right)=\mathit{min}\left(\sum _{h=1}^H\sum _{b=1}^{NB}\mathrm{\Delta }{P}_{b,h}·t_h\right)$$

(2)

where ΔW—the sum of energy losses in the LVEDN, computed for each time interval h, h = 1, …, H, t_h—the time sample h, and ΔP_b,h are the active power losses for each time interval h and branch b = 1, …, NB:

$$\mathrm{\Delta }{P}_{b,h}=R_b·(I_{b,h}{)}^2+K_{b,h}$$

(3)

In (3), R_b is the resistance of branch b, I_b,h—the current flow on branch b at time interval h, K_b,h—the loss increase factor due to the current flow on the neutral wire on branch b at time interval h, computed according to the procedure described in [37].

The charging of the batteries placed in buses b_i, i = 1, …, NSS is considered to take place in a customizable time interval [h_j, h_k], j ≤ k. When charging, the batteries receive in each interval h a constant amount of energy computed so that the battery charging should be completed at time h_k:

$$W_{s,h}=\frac{SO{C}_{s,max}-SO{C}_{s,min}}{h_k-h_j}$$

(4)

The hourly active power demand at a generic bus i in interval h to charge the battery is modeled as an increase in the active load P_i,h in the respective bus.

The discharging of the batteries occurs after a time mark h_d, usually set at peak demand hours, to maximize the load leveling, and stops at midnight, or earlier if the battery reaches the minimum charging level SOC_min. The discharge of the batteries is modeled as a decrease in the instantaneous load in the respective bus.

The model described in (1)–(4) is subjected to several technical constraints:

• The state of charge (SOC) limits for the storage batteries should not fall below the minimal requirement SOC_min and should not exceed the maximum charging level SOC_max, for all the batteries s = 1, …, NSS, in each hour h in the interval of analysis, with h = 1, …, H:

  $$SO{C}_{s,min}\le SO{C}_{s,h}\le SO{C}_{s,max}$$

  (5)
• The voltage magnitude U_i,h should vary in the allowed range [U_i,min, U_i,max] in each bus i = 1, …, NN and in each hour h in the interval of analysis h = 1, …, H:

  $$U_{i,min}\le U_{i,h}\le U_{i,max}$$

  (6)
• The current flow I_b,h must not exceed the allowable ampacity I_b,max on all the branches from the LVEDN, b = 1, …, NB and in each hour h in the interval of analysis h = 1, …, H:

  $$I_{b,h}\le I_{b,\mathit{max}}$$

  (7)

2.2. The Adaptation of the GA to the Storage Management Problem

The GA was chosen because of its proven capabilities of solving NP-hard problems with discrete search spaces. Like other metaheuristic population-based algorithms, it performs a parallel search on multiple dimensions of solutions encoded as vectors of real or integer numbers. Being inspired by the biological behavior of passing desirable traits from parent to offspring by the recombination of chromosomes, its optimization mechanism is a ‘generational’ (iterative) search process that recombines the ‘genetic’ information already existing in the population to reach new combinations of ‘genes’ (values) into ‘chromosomal’ sequences (vectors that denote possible valid solutions). The ‘adaptation’ (optimality) of each solution is assessed by computing its ‘fitness function’. The best ‘chromosomes’ are preserved by ‘selection’, and the population is recombined through ’crossover’ to generate new solutions. The novelty and diversity in the population is assured by random limited ‘mutations’. The basic flowchart of the GA is presented in Figure 2.

The structure of a chromosome was adapted by the authors to consider the particularities of the problem. Since loss reduction is achieved by optimally placing batteries at certain buses and on desired phases, the length of the solution is 2*NSS, a gene being allocated for each battery for the bus, and another for the phase of connection. All these values are integer numbers. By using the GA, its crossover and mutation operators keep the nature of these values, eliminating the need of supplementary computational effort for validation. The structure of a solution/chromosome is provided in Figure 3.

This structure was chosen for the simplicity in which it can be used to simulate the three scenarios of storage management considered in the paper, as follows [36]:

• S1: (one-bus, multiple phase storage): bus 1 = bus i = bus NSS, ph j ∈ {1,2, 3}
• S2: (multiple-bus, multiple phase storage): bus i ∈ {1,2, …, NB}, ph j ∈ {1,2,3}
• S3: (multiple prosumer bus, multiple phase storage): bus i ∈ {PS₁, …, PS_NPS}, ph j ∈ {1,2,3}

From the multiple methods of selection, crossover, and mutation, the following were used in this paper, adapted to the particularities of each scenario:

• Tournament selection, where in each step, a number of c₁ chromosomes are selected randomly, only the first c₂ ranked according to their fitness are kept, where c₂ < c₁, c₁ < = population size, and several steps are repeated until a new population is completed.
• Uniform crossover, exemplified in Figure 4 for the bus section of a chromosome encoding 5 storage sources and random switch threshold r ≥ 0.5 that uses a random mask for exchanging genes between parents. The same procedure and random mask are used for the phases.
• Random mutation of one gene in a chromosome.

2.3. The Computation of the Fitness Function

In Figure 2, the COMPUTE FITNESS step requires in each iteration the decoding of a chromosome having the structure presented in Figure 3 and Equation (1), solving the problem (2–7) and identifying the active power losses that occur in the network using the storage solution and night battery loading scenario considered as initial assumption by the algorithm. The necessary steps are presented in Figure 5.

3. Results

For comparison validity reasons, the influence of the night battery charging was tested on the same LV four-wire electricity distribution network used in [36]. The network has 131 buses that supply 113 one-phase consumers, through 4-wire overhead feeders, 4840 m in length, out of which 2240 m is the length of the main feeder. The demand profiles of the consumers were measured by the local electricity distribution company using smart meters with 1 h profile sampling, installed at each consumer, and a benchmark summer day for the year 2018 was used in the study. The prosumers were modeled using typical profiles supplied by the company. The total demand of this network is of 219.85 kWh of active energy over 24 h. Out of this value, 31.73 kWh are found in the interval of valley load, used in the study to charge batteries overnight, and this is depicted in black in Figure 6. The discharge of the batteries is programmed in the peak load hours, depicted with violet bars in Figure 6, with a corresponding total load of 66.31 kWh. The one-line diagram of the LVEDN, where the prosumer buses are emphasized with black, and for each bus it is specified the number of existing consumers, is provided in Figure 7. There are 8 prosumers in this network, with a total generation of 122 kWh, out of which 75.38 kWh are surplus, distributed hourly as per Figure 8. In the absence of storage, this excess energy is injected back into the network.

If the network is operated without storage, the active energy losses will depend on two main factors: (1) the load patterns of the consumers, adjusted, for those who are also prosumers, with the contribution of the local generation and (2) the possible reversed power flows that will occur in the hours and on the phases where surplus will be generated and injected back into the grid. Another contributing factor can be the uneven load on the three phases due to the inherent unbalanced operating conditions found in the four-wire networks that supply one-phase consumers.

It should be noted that the low overall consumption in the network and the low number of consumers whose consumption is high enough to justify the prosumer generation for self-consumption is a common occurrence for the region to where the LVEDN belongs. If the loss pattern over 24 h is compared between the scenarios when prosumers are active or removed from the study, Figure 9 shows that the presence of the prosumer surplus leads to a significant reduction in the energy losses, as the available generation is used by the nearby consumers, decreasing the hourly network infeed and reducing branch power flows. The total 24 h losses amount to 15.98 kWh without prosumers and 8.74 kWh with prosumers.

As it was shown in [36], the installation of storage at selected network buses, set to discharge at peak hours, will lead to significant decrease in energy losses in the evening time interval if Scenarios 1 or 2 are used. However, in Scenario 3, the evening loss decrease is accompanied by loss increase in the day hours when the batteries are charged using the available prosumer surplus. This paper investigates the influence of night storage charging over the daily energy losses. The focus is on Scenarios 1 and 2, while Scenario 3 is used only for comparison. The results are compared with the findings from [36] using the same settings for the GA (

Table 1. The initial settings used for the algorithm.

Population Size	100
Iteration count	100
Battery storage capacity and type	4 kWh, one-phase
Battery stock	5
Initial state of charge for the batteries	20%
Maximum state of charge for the batteries	95%

). The algorithm was implemented in MATLAB R2021a and a Windows 11 workstation with 32 GB of RAM, and an AMD Ryzen 7 5700X 8-core processor was used to run the simulation. For the settings and the size of the network used in the case study, a typical run of the algorithm clocks at 0.8 s per iteration.

For both scenarios, two options regarding the charging time were investigated. The first considers extending the charging over most of the length of the night load valley interval, between hours 02:00 and 06:00, depicted with black bars in Figure 6. This case is referred to in the following as ‘slow charging’ (SC). The other case is the opposite, narrowing the charge to just one hour, for which very different results were obtained. This is ‘fast charging’ (FC), and the hour of charging was chosen between 02:00 and 03:00, where the demand is minimal. The best solutions obtained by the algorithm for each scenario and charging option are presented in

Table 2. The storage placement solutions.

Scenario	Solution, Buses					Solution, Phases					ΔW, kWh
S1, WNC	85	85	85	85	85	1	1	3	1	2	6.63
S1, SC	85	85	85	85	85	3	3	1	1	3	7.49
S1, FC	1	1	1	1	1	1	2	3	3	1	8.80
S2, WNC	85	119	119	85	56	1	2	2	1	1	5.62
S2, SC	1	85	37	85	85	3	3	3	1	1	7.03
S2, FC	86	1	85	3	1	3	3	1	2	1	8.25
S3, WNC	85	63	85	119	44	1	2	1	2	3	7.61
Reference (without storage)											8.74

, together with the associated energy losses. The change in losses for the night valley load and evening peak load times are compared in Figure 10, Figure 11, Figure 12 and Figure 13. The notation ‘WNC’ corresponds to the results obtained in [36], without considering night charging. A detailed interpretation of the results follows.

3.1. Scenario 1—Batteries Installed at the Same Bus

Scenario 1 is created to simulate the case of single community storage, where all the batteries will be placed at the same bus. If the SC option is chosen, as it is shown in Figure 10, the losses will increase in the night hours, and the loss increase is equal. The same bus is chosen as in the WNC case, where the batteries were considered already charged. The losses increase from 6.63 kWh to 7.49 kWh but remain significantly lower than in the reference case (8.74 kWh) or Scenario 3, prosumer-managed storage (7.61 kWh). The losses when batteries are discharged at evening peak time show a small increase compared to the WNC case (Figure 11) because the phase arrangement of the batteries is changed to better accommodate night charging with minimal losses.

However, if the fast charge scenario is used, the results show that the use of storage is not efficient. The algorithm places the batteries in bus 1, the closest to the source, and the batteries are not discharged in the evening hours. This choice is determined by two conflicting behaviors. First, if the batteries would be placed at a latter bus, the losses occurring on the supply path between the substation and the batteries would increase because of the need to supply a very large load (15 kWh needed to charge in one hour the batteries from SOC 20% to SOC 0.95). On the other hand, discharging the batteries located so close to the source to supply loads spread across the network would produce the same losses as when using the main source. Only one battery is discharged in this case from 0.95 to 0.79 to supply the local bus load. The losses increase for 1 h between 02:00 and 03:00 with a small amount, corresponding to the increase on the branch 01–02, and remain at the same high level as in the reference case in the peak interval, as the batteries are not used (Figure 11). The total losses increase compared to the reference case, from 8.74 to 8.80 kWh. Thus, the use of storage with fast charge is not recommended.

3.2. Scenario 2—Batteries Installed at Different Buses

At nighttime, when batteries will be charged from the source bus (the MV/LV substation), the losses increase, with a smaller amount in each hour when the SC option is used, and have a significant peak at hour 02:00, if the FC option is preferred, as Figure 12 shows. At evening peak time (battery discharge mode), the losses are higher than in the previous WNC case because the battery placement must consider the change in night losses also (Figure 13).

As in the first scenario, the use of fast charge will considerably increase the losses. If the night charging is included in the study, the losses will increase from 5.62 kWh to 7.03 kWh in SC mode and up to 8.25 kWh in FC mode. If slow charging is preferred, the use of storage remains more efficient than the case when the batteries are managed by the prosumers (Scenario 3). The use of fast charging will result in losses lower than in the reference case, but the difference is small (8.25 kWh FC, 8.74 kWh in the absence of storage).

4. Discussion

The injections of the prosumers with generation surplus can change the active power flows on the branches of an electrical network, affecting technical parameters such as energy losses and voltage levels. If local markets are organized to allow prosumers to sell electricity to neighboring consumers, storage can be used to defer supply and optimize the price of transactions. In this case, storage would usually be installed and managed by the prosumers to fulfill their own objectives. This paper examines another approach in using storage, in which the management of the batteries is made by the electricity distribution utility (network operator) and used to improve the technical parameters of the grid.

The first assumption of this paper is that the utility can augment its load profile leveling strategies by using the night load valley interval to recharge the storage batteries and later use the stored energy to reduce the peak load. In this case, the battery placement is a constrained optimization problem that can be solved to reach a desired objective.

The second assumption of this study is that in developing countries, the number of consumers that become prosumers and have the means to invest in individual high-capacity storage is low, and the same is often happening for electricity distribution utilities that need to modernize a significant number of local LV networks or microgrids. It would be an optimal approach to rethink the way in which the already existing limited local storage capacity is used and coordinated, with the aim of limiting or deferring the amount of new investment. Thus, the scenarios considered in this paper were simulated and analyzed. The case shown in this paper is frequently seen in the region from where the LVEDN originates: a structure consisting of long OHL feeders that supply mostly consumers with low daily demand, a variable batch of inactive consumers (closed homes), and a very small number of consumers for which the demand is high enough to justify the transition to prosumer status. These prosumers will often generate surplus, as their self-consumption can be low during daytime. The small number of consumers with high demand, if they are placed at the beginning or near the far end of a feeder, can increase the unbalance of the load on the three phases, contributing to the decrease in the quality of supply. One way to alleviate these inconveniences and make the operation of the LVED infrastructure more sustainable is to optimally use storage; but, as the results presented above have shown, the choice of the storage placement must be adapted to the structure and load pattern of each network. In the majority of the cases tested, the result was a decrease in losses that can be optimized by an adequate placement of the batteries at certain buses and phases. But, the study also found that operation scenarios can occur which can result in the opposite, increasing the losses in comparison with the no-storage use situation.

The third aspect that needs to be discussed is the method of choice, namely the use of the genetic algorithm. Along with the advantage of parallel search in discrete spaces, the GA could be adapted to use the measured data that a typical electricity distribution utility can provide. Also, it requires less input data compared to similar methods from the literature ([25,27,34]) involving Artificial Neural Networks, that typically need a large database for training, with multiple input variables, including voltage and frequency measurements, often unavailable in LV networks. The GA has, however, the disadvantage of working with load profiles that may not simulate accurately the full range of operating states that can occur in the LVEDN and, being a metaheuristic, provides no guarantee for reaching the global optimum. Careful consideration must be exercised in the choice of input parameters for the algorithm, but the initial setup is a problem encountered for other categories of methods available in the literature.

The method presented in this paper can be improved by considering a multi-objective optimization approach, which would consider the effect of supplemental storage capacity, the economic perspective (minimizing the price of electricity import), and the improvement of other supply quality parameters (PV generation smoothing or bus voltage optimization).

5. Conclusions

The study presented in this paper shows that different choices made in managing a limited storage resource can result in improving the operation conditions in local electricity distribution grids. By optimally choosing the placement and operation of storage batteries, the energy losses can be reduced and the existing resources can be used more efficiently, contributing to the sustainable development of the electricity distribution infrastructure in the transition phase towards smart grids.