Sizing with Technical Indicators of Microgrids with Battery Energy Storage Systems: A Systematic Review

Abstract

Worldwide, governmental organizations are restructuring energy policies, making them cleaner, encouraging transformation and energy transition by integrating renewable sources, engaging in environmental preservation, and, notably, meeting the growing demand for sustainable energy models, such as solar and wind energy. In the electricity sector, reducing carbon emissions is crucial to facilitating the integration of microgrids (MGs) with renewable sources and Battery Energy Storage Systems (BESSs). This work constitutes a systematic review that thoroughly analyzes the sizing of MGs with BESSs. The unpredictability and variability of renewable sources justify the complexity of this analysis and the loads connected to the system. Additionally, the sizing of a BESS depends primarily on the application, battery technology, and the system’s energy demand. This review mapped and identified existing computational and optimization methodologies for structured sizing in technical indicators of an MG with a BESS based on articles published between 2017 and 2021. A protocol was defined in which articles were filtered in multiple stages, undergoing strategic refinements to arrive at the final articles to address the Research Questions (RQs). The final number of articles was 44, and within these, technical indicators related to the RQs were addressed, covering the most relevant works and comparing them technically, including how each explains the objective and result of their work. The rejected articles did not meet the criteria established by the defined protocol, such as exclusion criteria, quality criteria, and RQs. In conclusion, studies employing the integration of machine learning coupled with optimization techniques exhibit a significant contribution to results, as historical data can aid machine learning for data prediction.

1. Introduction

Nowadays, about 63.3% of the world’s electrical energy is generated by burning fossil fuels [1,2,3]. Using renewable sources is one of the alternatives for reversing this scenario [4], supplying electrical loads [5], either for specific time intervals or continuously. The integration of Distributed Energy Resources (DERs) with a system’s loads is referred to as a microgrid (MG) [6], aiming for a better joint operation of these sources. Most MGs operate connected to the grid (on-grid), providing bidirectional energy flow [7] with energy generators and end-users, enabling better energy management. In a grid outage, the MG can operate in an isolated (off-grid) or autonomous mode [5], but both on-grid and off-grid modes are controlled and coordinated. The advantages of MGs include increased efficiency in improving the quality and reliability of electrical energy, reduced energy costs, the ability to generate revenue by injecting energy into the grid, the potential to provide ancillary services, reduced peak energy demand, lower emissions of pollutants, and the possibility of having multiple connected generation sources [8]. However, there are challenges in designing an MG, such as the appropriate selection of DERs and optimal sizing [1,2].

However, MGs need elements to ensure network stability and supply variable loads [9]. A typical example is diesel generators that support MGs; however, this alternative contributes to the emission of polluting gases. Fortunately, the Battery Energy Storage System (BESS) offers a solution to meet this demand while providing advantages when connected to renewable energy sources. These benefits go beyond complementing the variability of these resources [10]. Significant benefits can be expected from a BESS due to its flexible operation, such as demand control, acting when the load may exceed the contracted demand [11]. Additionally, a BESS facilitates energy shifting, storing energy during periods of excess supply and used during peak demand hours when the cost is higher [12]. There are opportunities to reduce costs for small- to medium-sized end consumers, especially during peak hours when energy tariffs increase compared to off-peak hours [13,14].

One of the challenges related to MGs is their Capital Expenditure (CAPEX) and Operation and Maintenance (O&M) costs. However, installing robust systems that provide reliability, such as BESSs, in applications with low response times offers more significant support to the network and, consequently, to the MG. For this reason, the optimal sizing of a BESS is necessary, as it reduces the operational costs of an MG in the long run, enabling better utilization of renewable energy resources such as wind and solar power [9]. Thus, the operation of BESS in an MG involves storing energy and discharging it according to the operator’s needs, with the system programmed for this purpose.

Having a BESS as part of the MG offers several advantages:

This article presents a Systematic Literature Review (SLR) focused on studies involving the installation of BESS in MG. This article aimed to map and identify existing methodologies related to MG sizing with BESSs, incorporating optimization techniques with or without machine learning. Several methods proposed in these studies are related to the optimal sizing of renewable energy sources and BESSs.

The study conducted in [15] utilizes machine learning techniques such as Elman Neural Network (ENN), Wavelet Neural Network (WNN), and the statistical method Auto-Regressive Integrated Moving Average (ARIMA) to overcome the unpredictability of generation, providing forecasts for the next hour and day. Additionally, it sizes the BESS based on the data. On the other hand, the author of [16] proposes an optimization technique using K-means clustering based on the elbow method (OKEM) to obtain typical load patterns of industrial users from their historical load data. Furthermore, an extension of the optimization function was necessary to determine the BESS’s ideal load pattern and energy storage capacity.

This SLR also discusses optimization techniques for finding the optimal value within an MG or a hybrid energy system for BESS sizing. The article [17] employs various optimization algorithms, such as Mixed-Integer Linear Programming (MILP), Mixed-Integer Nonlinear Programming (MINLP), and the Genetic Algorithm (GA). The study’s primary objective is to develop a model for energy customers with intelligent metering arrangements and limited historical consumption and generation data. The aim is to decide on the ideal size of the BESS. As observed, various types of analyses are conducted in different studies to determine the optimal BESS size within an MG. The MG can achieve better operation through system optimization while maximizing its intrinsic benefits, thus reducing energy costs.

The main contribution of this article lies in surveying the scientific research conducted on the sizing of BESSs in MGs. The analysis was conducted based on specific questions and selection criteria, utilizing a database constructed from major digital libraries, keywords, publication periods, and quality criteria, among other factors.

The article is structured as follows. Section 2 introduces essential concepts for better understanding the primary optimization techniques and machine learning methods. Section 3 presents the methodology, including the materials and methods used for the review. Section 4 details the research protocol that formed the basis of this SLR, including quantitative analysis from the initially selected articles to the finalists. Section 5 presents the research results, discussing the questions raised. Section 6 concludes the article.

2. Background

In recent decades, the optimal sizing of hybrid energy systems has emerged as a rapidly growing research field. This complex challenge involves integrating uncontrollable energy sources like solar, wind, and BESSs to meet demands economically and sustainably. In this context, various techniques have been explored, either individually or in hybrid forms. Among these, three approaches are the most prominent and promising: optimization techniques, machine learning, and statistical methods. Additionally, established software solutions in this domain are also discussed.

2.1. Optimization Techniques, Machine Learning, and Statistical Methods

2.1.1. Optimization Techniques

A solid mathematical foundation provides a rigorous framework for finding the ideal configuration of hybrid energy systems, considering a range of variables, physical and economic constraints, and specific objectives. Here are some of the most relevant optimization techniques applied in this context:

• Linear and Nonlinear Programming: Linear programming deals with optimization problems in which the objective function and constraints are linear. Nonlinear programming extends this concept to problems with nonlinear objective functions or constraints. Both approaches are widely applied in the optimal sizing of hybrid energy systems, considering costs, resource availability, and efficiency. Techniques such as Two-Constraint Linear Programming (TCLP) and Mixed-Integer Quadratic Programming (MIQP) are examples of linear programming and its variations [1,18].
• Evolutionary Algorithms: Inspired by the process of natural selection and evolution, these algorithms are used to find approximate solutions for complex optimization problems by exploring populations of candidate solutions and applying genetic operators such as selection, recombination, and mutation to enhance solutions over time. The methodologies include the Grey Wolf Optimizer (GWO), Particle Swarm Optimization (PSO), and the Genetic Algorithm (GA) [2,4,9,19].
• Multi-Objective Optimization: When it comes to hybrid energy systems, multiple objectives often exist, such as minimizing costs, maximizing efficiency, and reducing emissions. Multi-Objective Optimization deals with the search for solutions that balance these competing objectives, resulting in Pareto-efficient solutions representing trade-offs among the objectives. The techniques include Mixed-Integer Conic Programming (MICP) and Adaptive Mixed Differential Evolution (AMDE) [16,20].

2.1.2. Machine Learning

On the other hand, machine learning, with its ability to extract complex patterns from data and make adaptive decisions, provides a more flexible and data-driven approach to solving this problem. Here are some of the machine learning techniques relevant to hybrid energy systems:

• Neural Networks: Artificial Neural Networks (ANNs) are computational models inspired by the functioning of the human brain. They are used to learn complex patterns from data, particularly useful in predicting energy production from renewable sources such as solar and wind. Deep learning Neural Networks and Recurrent Neural Networks (RNNs) have also been applied to enhance the accuracy of predictions [15,20].
• Random Forests: Machine learning algorithms that combine multiple decision trees to create robust and accurate models. They can be used to optimize hybrid systems in real time, adapting to changes in operational conditions [17,21].
• Clustering: This is used to group similar data points into clusters or groups. In the context of hybrid energy systems, clustering is applied to identify behavior patterns of different system components. This methodologies include K-means Clustering (KC), Elman Neural Networks (ENNs), and Wavelet Neural Networks (WNNs) [15,17].
• Regression Model: Initially, regression analyses are commonly employed for prediction purposes, with their application closely overlapping with the domain of machine learning. Furthermore, regression analysis can be applied in specific cases to identify causal relationships between independent and dependent variables. Linear regression analysis can be divided into simple and multiple linear regression. Multiple linear regression is a statistical approach used to predict the outcome of a response variable by employing multiple explanatory variables. In contrast, simple linear regression isolates the influence of independent variables from the interaction among dependent variables [22,23].

2.1.3. Statistical Forecasting Procedures

In addition to optimization and machine learning techniques, statistical forecasting procedures play a fundamental role in analyzing and modeling hybrid energy systems which involves a considerable amount of time series interpretation. Some relevant statistical methods for this area are the following:

• Univariate Models: A statistical approach that deals with data collected over time, relying on only one historical series. In the context of hybrid energy systems, time series analysis is widely used to model historical behavior and make future predictions of energy production and consumption, as seen in the Auto-Regressive Integrated Moving Average (ARIMA) technique [15].
• Causal Models or Transfer Function Models: Future values of a series are not determined solely via their past values but can also be influenced by series that have some relationship with it. In the case of electricity load consumption, including the relative price as a correlated series can contribute to a more comprehensive explanation of this phenomenon [24].
• Multivariate Models: These models not only consider the autocorrelation of the main series but also incorporate values from external series that enhance the forecast and analysis of this series. These external series can provide evidence of linear or nonlinear causality or correlation, contributing to clarifying how the values of the main series develop over time. An example of such a model would be one capable of simultaneously predicting the energy consumption in various service-providing utilities in the country [24,25].

2.2. Utilization of Established Software Solutions

Continuing the analysis of optimized sizing for hybrid systems, it is important to note that many relevant articles in the literature also employ established software, incorporating the previously mentioned techniques. Examples include HOMER and MATLAB for analyses, simulations, and practical implementations. These tools are crucial in validating and applying proposed solutions in real-world scenarios.

• HOMER (Hybrid Optimization Model for Multiple Energy Resources): This is a tool designed to analyze and optimize hybrid energy systems. It enables the evaluation of various configurations of hybrid energy systems, considering renewable energy sources, energy storage, and other components. HOMER is widely employed to conduct economic and technical feasibility analyses for hybrid energy system projects [26].
• MATLAB: This is a numerical computing and programming platform that provides a flexible environment for implementing optimization and machine learning algorithms. It also allows for the integration of additional tools and the creation of custom models. MATLAB is a common choice for implementing and testing proposed solutions in hybrid energy systems [1,18,27].

3. Method and Data

The articles included in this SR were selected to map and identify existing methodologies for sizing MGs with BESSs. To achieve this, the general question (GQ) has the following approach:

• GQ: How is the methodology for sizing MGs with BESSs structured in a machine learning context?

3.1. Data Sources and Procedures for the Extraction of Articles

Based on the GQ, data sources and procedures for article extraction are selected. Therefore, the digital libraries that supported this SLR are as follows:

• Compendex (www.engineeringvillage.com, accessed on 5 September 2021);
• IEEE Xplore (www.ieee.org, accessed on 23 September 2021);
• Scopus (www.scopus.com, accessed on 11 October 2021);
• Science Direct (www.sciencedirect.com, accessed on 30 October 2021);
• Springer Link (www.springerlink.com, accessed on 15 November 2021).

Relevance and reliability in electrical and computer engineering were the selection criteria for these databases.

Thus, with the defined database, the review strategy involves selecting studies based on keywords and publication period.

• Keywords: (“battery energy storage system”) AND (“sizing” OR “dimensioning”) AND (“microgrid”) AND (“methodology” OR “method” OR “technique” OR “optimization” OR “approach”) AND (“artificial intelligence” OR “machine learning”);
• Period of search: 2017–2021.

3.2. Review Procedures

The review procedures are conducted based on two selection criteria: inclusion and exclusion. We present the details for each criterion as follows:

(I)

Inclusion Criteria:

• Works that report on the experience of the electric power sector;
• Works that include analysis with renewable sources;
• Works that show technical indicators;
• Works that contain financial indicators;
• Works using data to define the methodology.

(II)

Exclusion Criteria:

• Works that do not involve batteries;
• Works that are not in English;
• Works that are not available for consultation or download;
• Thesis, dissertation, poster, tutorial, and editorial;
• Duplicate and incomplete studies.

3.3. Filtering/Reviewing Process

During the article screening for the SLR, several steps are outlined to enhance the article selection process:

(1)

Application of the inclusion and exclusion criteria;

(2)

Evaluation according to the title, abstracts, and keywords to select studies that hold relevant information to this systematic review;

(3)

Improvement in evaluating the articles, reading them with more criticality. The list in

Table 1. Quality criteria.

ID	Quality Criteria	Answer
1	Are the objectives or questions of the study clearly defined?	Y/N/P
2	Does the study clearly and unambiguously report the results?	Y/N/P
3	Does the study list primarily or secondarily use the methodologies and sizing indicators?	Y/N/P
4	Does the study explicitly address approaches to any artificial intelligence techniques?	Y/N/P
5	Does the study reference renewable energy?	Y/N/P
6	Does the study refer to hybrid power systems with battery energy storage systems?	Y/N/P
7	Does the study report on hybrid power systems of battery technologies?	Y/N/P
8	Does the article evaluate more than one application or operating system mode?	Y/N/P
9	Does the article describe all the steps of the methodology used?	Y/N/P
10	Does the methodology used in the article have synergy with the reality of the Brazilian electricity sector?	Y/N/P

will support scoring and ranking the selected papers.

It follows the article scoring rule:

$$Yes\left(Y\right)=1.0\left|No\right(N)=0.0|Partially\left(P\right)=0.5$$

(1)

3.4. Strategy for Extracting and Summarizing Results

Finally, for each article, the following pieces of information must be extracted:

• Article title;
• Names of authors;
• Year;
• Country;
• Institution;
• Research base;
• Application context;
• Methodology procedures;
• Indicators employed;
• AI methods;
• Results;
• Advantages and disadvantages of the model.

With this extraction and summarizing of results, we defined three Research Questions (RQs) for the final filtering:

• RQ1: What methodologies for sizing hybrid energy systems with batteries are applied, and what applications are used? In which sector of the electrical system is it located: generation, transmission, distribution, or final customer?
• RQ2: How is battery technology impacting the hybrid power system sizing?
• RQ3: Which machine learning approaches of sizing hybrid power systems with batteries are employed?

The RQs will be discussed in Section 5 of this SLR.

4. Quantitative Analysis

This section presents the quantitative analysis of the SLR, covering the selection process of all articles that meet the initial criteria up to the final articles that address the RQs. The following subtopics will detail the number of articles from the databases, acceptance and rejection rates, the percentage from each database, the number of publications per year, and the structure of the SR study. The final refinement will also be presented, including excluding studies that did not address the RQs. This refined analysis will be presented step by step.

Quantity of Articles

The figures below describe the overview of the number of articles conducted in the SLR. Figure 1 presents the percentage according to each database from which the articles were initially selected. Out of the 5828 initially selected articles, the percentages related to each database are shown. This pattern repeats for the first and second refinements, represented by 157 and 44 articles selected, respectively. Among these selected articles, the percentages related to each database are presented (Figure 2 and Figure 3).

The first refinement consisted of excluding articles that fit the exclusion criteria, as described in Section 3.2, and the quality criteria outlined in Section 3.3. This refinement considered only the article title, abstract, and keywords.

The 157 articles that passed the first filter were read in full for the second refinement. Based on this analysis, articles that did not address at least one of the RQs were excluded.

The acceptance rate of articles in this SLR, during the first and second refinements, is illustrated in Figure 4 and Figure 5, respectively, referring to the total number without refinement, as shown in Figure 1, which includes the initial 5828 pre-selected articles.

The graphs in Figure 6 and Figure 7 show the exact number of articles selected from each database in the first and second refinements, respectively. These values correspond to the percentages in Figure 4 and Figure 5.

The graphs in Figure 8 and Figure 9 present the exact number of publications of the selected articles in the first and second refinements, respectively, organized by year. In addition, Figure 10 summarizes the structuring of studies focused on the SLR, starting from the total selection of 5828 articles to the finalists.

5. Discussion on Research Questions

The results described in this section are based on the protocol developed for this study. Subsections report the findings, research development stages, and this review’s limitations.

Many optimization techniques for sizing BESSs/MGs are found and widely used in the articles, while others bring the perspective of machine learning in a more limited manner. Consequently, various studies have employed optimization techniques for MG sizing, such as GA, PSO, MILP, and GWO. The rejected works include those that assume that one of the MG sources uses fossil fuels, in addition to optimizing its operation, without any technique for BESS sizing, studies needing more detailed technique development (only comparisons), and reviews.

Table 2. Comparative analysis of approved studies from the SLR.

Article	Energy Sector	Location of BESS Integration	BESS Applications	Battery Technology	Machine Learning Applied	BESS Impact on MG	Sizing Methodology
[1]	Generation and end-user	Microgrids	Balance between energy supply and demand	Uninformed	Not applicable	Reduction in transmission losses in energy, while ensuring the stability and reliability of the MG	Two-Constraint-Based Linear Programing (TCLP) and MATLAB
[2]	Generation and end-user	Microgrids	Load demand support and energy arbitrage	Uninformed	Not applicable	Increase in renewable energy self-consumption and efficiency of the hybrid MG system, but the payback period and return on investment are affected	Advanced Grey Wolf Particle Swarm Optimizer (AGWPSO) and GWO
[4]	Generation and distribution	Microgrids	Postponement of investment	Lead acid (PbA), sodium–sulfur (NaS), and lithium ion (Li-ion)	Not applicable	Helps improve the stability and reliability of the MG	GA and PSO
[7]	Generation, distribution, and end-user	Microgrids	Load shifting, peak shaving, and Backup	PbA and Li-ion	Not applicable	Improves energy efficiency and enhances the reliability of the electrical system	Mathematical model
[8]	Generation and distribution	Microgrids	Supports renewable generation and energy quality	Uninformed	Not applicable	Enhances MG benefits and provides reliability	Iterative technique
[9]	Generation and distribution	Microgrids	Postponement of investment, ancillary services, and energy arbitrage	Li-ion	Not applicable	Reduction in MG costs and improved efficiency	MILP
[11]	Generation and end-user	Residential	Backup and supply–demand balance	Uninformed	Not applicable	Reliable and sustainable energy supply for residential areas	PSO, GA, and Multi-Objective Particle Swarm Optimization (MOPSO)
[12]	Generation and end-user	Microgrids	Load shifting	Li-ion	Not applicable	Improvement in efficiency, reliability, and financial performance, but it depends on proper sizing and implementation of the BESS	Paired comparison and rating method—Multi-Objective Optimization
[13]	Generation	Microgrids	Power quality	Uninformed	Not applicable	Helps balance renewable energy generation with variable energy demand, improves the quality of supplied energy	Mixed-Integer Programming (MIP)
[14]	Generation, distribution, and end-user	Residential	Load demand support	Uninformed	Linear regression	Improved profitability of using BESS with the photovoltaic system	GA
[15]	Generation	Hybrid plants	Generation forecasting/compensation	Generic model	ENN and WNN	Fast response time (for firm generation curve)	NN with optimization and ARIMA
[16]	End-user	Industry	Power quality	Lead carbon (PbC) and Li-ion	Not applicable	Maximized benefits for end-users (industrial) and reduction in energy losses	OKEM, Adaptive Mixed Differential Evolution (AMDE), and Evolutionary Algorithms in MATLAB
[17]	Generation and end-user	Distributed generation	Load demand support	Uninformed	Linear regression	Minimizing the energy bill cost	Load and consumption forecasting using machine learning techniques
[18]	Generation and end-user	Microgrids	Peak shaving, power quality, and load shifting	Li-ion	Not applicable	Optimization of MG cost	Based on Mixed-Integer Quadratic Programming (MIQP) in MATLAB
[19]	Generation and distribution	Microgrids	Reactive control (frequency), to provide reliability and Backup	Li-ion—LFP	Not applicable	Improved stability and reliability of the electrical system during disturbances or power interruptions	GA
[20]	Generation and end-user	Microgrids	Power smoothing and load shifting	Uninformed	Set of RNNs enhanced with an LSTM	Generation and load balance	Not detailed
[21]	Generation and distribution	Microgrids	Load shifting	Uninformed	Scikit-learn in Python—Random Forest	Cost reduction	MILP
[26]	Generation, distribution, and end-user	Microgrids	Load shifting and Backup	Li-ion	Not applicable	Economic advantages, reduced dependence on the main grid (energy purchase)	Not detailed
[27]	Generation and distribution	Microgrids	Power smoothing, Backup, load shifting, and power quality	Li-ion	Not applicable	Reducing operating costs and increasing the reliability of energy supply to consumers	Determination of the optimal discharge range and battery capacity, considering the minimum operating cost of the system (MATLAB)
[28]	Generation and distribution	Microgrids	Backup, load demand control, and supply–demand balance	Valve-Regulated Lead Acid (VRLA) PbA	Not applicable	Cost reduction	PSO, HOMER Pro, and MATLAB
[29]	Generation and distribution	Microgrids	Load shifting, supply–demand balance, and Backup	Uninformed	Not applicable	Reduction in the total system cost and reliability	Combination of Fuzzy Logic and Grey Wolf Optimization Swarm (FL-GWO)
[30]	Generation	Microgrids	Supply–demand balance	Uninformed	Not applicable	Reduction in the total system cost and reliability	PSO
[31]	Generation and distribution	Microgrids	Backup and supply–demand balance	Uninformed	Not applicable	Improved performance and reliability of distributed renewable energy systems in MG	MILP
[32]	Generation, distribution, and end-user	Microgrids	Power quality	Uninformed	Not applicable	Reduction in operational costs of the MG	HOMER Pro and MILP
[33]	Generation and end-user	Microgrids	Backup, provide efficiency and reliability for the MG	Li-ion and PbC	Not applicable	Reduction in operational costs of the MG, improve efficiency and reliability in energy supply	Poli.NGR—MATLAB
[34]	End-user	Microgrids	Energy planning and Backup	Uninformed	Not applicable	Impact not detailed	HOMER Pro
[35]	Generation	Microgrids	Backup, supply–demand balance, and provide reliability	Uninformed	Not applicable	Provides an efficient and reliable renewable energy solution for remote areas without causing environmental pollution	Splitting Algorithm for three configurations of renewable hybrid energy systems
[36]	Generation, distribution, and end-user	Microgrids	Backup and seasonal demand	Li-ion	Not applicable	Reduction in costs and reliability	Optimization in the YALMIP—MATLAB toolbox
[37]	Generation	Hybrid power plant	Reactive power control and Backup	Li-ion—NMC	Not applicable	Not specified	Not detailed
[38]	Transmission	Grid	Contingency, ensuring system reliability, and ancillary services	Li-ion	Not applicable	Maintains electrical system stability in critical situations, reducing interruptions in power supply and improving the quality of energy provided to consumers	Maximum Minimum Flow Method
[39]	Generation and distribution	Microgrids (nanogrid)	Power quality	Uninformed	Not applicable	Reduction in costs	PSO
[40]	Generation and distribution	Microgrids	Ensure reliability in power supply	Li-ion	Not applicable	Reduction in greenhouse gas emissions, improvement in the reliability of electricity supply, and promotion of the transition to renewable sources	GA and Tabu Search Algorithm
[41]	Generation and distribution	Distributed generation	Energy planning	Uninformed	Backpropagation Algorithm—RNA	Impact on the future for distribution systems with DERs and BESSs participating in energy transactions in the decentralized market	Not detailed
[42]	Generation and end-user	Microgrids	Load demand support	Li-ion	Not applicable	Reduction in costs, demand response, renewable generation, taxes, and tariffs	Probabilistic—Monte Carlo
[43]	Generation and end-user	Distributed generation	Time-shift	Li-ion	Not applicable	Reduction in costs and maximizes the user’s investment value in electrical energy	Stochastic dynamic programming within a predictive control framework
[44]	Distribution	Solar PV power plant	Reactive power control and time-shift	Uninformed	Not applicable	Improvement in power quality; reduction in energy losses; increase in voltage stability in distribution networks; assists with load demand when there is a gap between generation and load	Calculation through the percentage of the total daily average energy yield from the PV plant
[45]	Generation, transmission, and distribution	Solar PV power plant	Load demand support	Li-ion	Not applicable	Improved microgrid performance, increased renewable energy penetration, reduced use of generators, and enhanced overall system efficiency	HOMER Pro
[46]	End-user	Microgrids (minigrids)	Backup and supply–demand balance	PbA	Not applicable	Enhances reliability in electricity supply and reduces operational costs	Stochastic simulation of time series using meteorological data
[47]	Generation, transmission, distribution, and end-user	Microgrids (nanogrids)	Reactive power control and Backup	Uninformed	Not applicable	Reduction in wasted energy generated from renewable sources, increasing the economic performance of the system	HOMER Pro
[48]	Generation and end-user	Microgrids	Backup	Li-ion	Not applicable	Reliable and efficient energy supply	Mixed-Integer Programming (MIP) Algorithm
[49]	Generation, distribution, and end-user	Microgrids	Load demand control and energy arbitrage	Li-ion	Generative Adversarial Network Method (GAN)	Minimizes energy consumption costs	Greedy, GA, and Deep Deterministic Policy Gradient (DDPG) Algorithms
[50]	Generation and distribution	Microgrids	Supply–demand balance and load demand support	NaS	Not applicable	Minimize the costs of the MG	MIP Algorithm
[51]	Generation and distribution	Microgrids	Backup	Uninformed	Not applicable	Minimizes energy consumption costs	MILP, linear programming (LP), GWO, PSO, Artificial Bee Colony (ABC), and GA
[52]	Generation, transmission, distribution, and end-use	Grid	Load demand support in an energy system	Li-ion—LFP, LMO, NMC, and LTO	Not applicable	Significant reduction in operational costs and required capacity size, along with improving the reliability of the electrical system	Mixed-Integer Convex Programming (MICP) and MATLAB

provides a summarized overview of the scientific articles accepted in this SLR. The analyzed parameters included the energy sector (generation, transmission, distribution, and end-user), BESS integration environment (MG, residential, photovoltaic (PV) plants, hybrid plants), BESS applications, battery technology, machine learning, BESS impact on MG, and sizing methodology. In the following subsections, the RQs from the SLR protocol are addressed. Related data for each question are presented, along with identifying the most relevant studies and their outcomes.

It is important to note that this methodology may lead to a discrepancy between the total number of themes identified and the number of articles included. Each identified theme represents a unique aspect addressed by one or more articles, and multiple themes can be covered within a single article, potentially resulting in a count exceeding the number of original articles. However, this approach enables the exploration of the breadth and depth of the research conducted, providing a more accurate and meaningful insight into the current state of knowledge in this SLR.

5.1. What Methodologies for Sizing Hybrid Energy Systems with Batteries Are Applied, and What Applications Are Used? In Which Sector of the Electrical System Is It Located: Generation, Transmission, Distribution, or Final Customer?

This RQ addresses three critical aspects of how BESS sizing, or MG sizing, is conducted. It also delves into system applications and the energy sector involved, all within the established time frame of the SLR protocol.

The most common sizing methods and the powerful software for sizing hybrid energy systems found in the studies were as follows:

• Evolutionary Algorithms;
• Mixed-Integer Linear Programming (MILP);
• MATLAB.

As seen in Figure 11, Evolutionary Algorithms were more frequently utilized among others, owing to their high efficiency and rapid convergence [28]. This set of methods, commonly referred to as heuristics or metaheuristics, seeks to find approximate solutions for optimization problems where finding the exact optimal solution is not possible or practical, as they draw inspiration from information that is difficult to represent. In the selected works of this review, mathematical models are established for the behavior of components within a hybrid plant or MG, providing a basis for creating the objective function. In this context, optimization methods work to minimize this objective function. This function can be related to energy or investment costs, the operation of a hybrid plant or MG, or the BESS, as well as maximizing the use of renewable resources [2,4,29].

For articles employing the Evolutionary Algorithms technique, it is evident that this approach can be applied to integrating renewables and sizing the sources with BESS. In the article, this is exemplified in [30], where the challenge is determining components’ size in an MG involving solar PV and BESS. The climatological dependency of generation introduces nonlinearity to the relationship between increased generation and the corresponding increase in reliability. Therefore, the most economical and consistent alternative is found through evaluations of various combinations. In this study, an enhanced version of the classic PSO, named Butterfly-PSO, is employed. Another study integrating renewable sources with a BESS in an MG is [28]. In this study, PSO was employed to address the problem of optimal sizing in DC MGs, which include PV, wind turbines, BESSs, and the grid. The algorithm-based technique brought optimal results for scaling under different operating modes of the system, and the operational objectives were achieved by solving the economic target functions of the system under relevant constraints.

Regarding articles based on the Mixed-Integer Linear Programming (MILP) model, these take into account the technical considerations of the system and optimization of the system’s capacity to achieve maximum efficiency with reduced costs. The study [31] addresses equipment degradation and battery lifespan in planning renewable energy hybrid systems. The model aims to minimize the total system cost while ensuring electricity demand is met within the MG. In the study presented in [32], the focus is on maximizing overall self-consumption and appropriately sizing batteries. This results in minimizing the total costs of the MG while meeting power quality constraints and other relevant considerations. The optimization problem solution determines the optimal size and temporal operation of batteries for load patterns and PV solar generation.

Another application in this SLR involves optimization techniques and consolidated software that employs advanced algorithms and precise models to optimize size systems. The most commonly used ones for this purpose were MATLAB and HOMER Pro. In the study in [33], MATLAB is the primary platform for developing the Poli.NRG software, which is a sizing tool for PV systems with BESS. Poli.NRG, developed within the MATLAB environment, allows for the robust sizing of a PV plant in conjunction with a BESS. It addresses the unique characteristics of rural environments, such as the unpredictability of energy sources and uncertainties in load consumption. The software incorporates appropriate component models and considers estimation errors during the design phase. In the study conducted in [34], HOMER Pro software is employed to model and simulate an MG system. This tool is utilized to implement MG’s energy management strategy, involving controlling adjustable energy resources, such as the grid and BESS, to meet hourly electrical load demand at the lowest possible cost. Additionally, HOMER Pro calculates each adjustable resource’s fixed and marginal costs and explores different production resource configurations to find the optimal arrangement that fulfills the MG’s energy needs. The software is also utilized to characterize each adjustable resource and determine the optimal production resource configuration that satisfies electrical load demand and operational reserve requirements while minimizing costs.

In the context of applications, the three most common perspectives emerge from the studies reviewed in this SLR (and Figure 12):

• Backup;
• Quality of energy (reduction in energy losses);
• Reliability.

The Backup application is predominantly used because it supports the load, MG, or hybrid plant. This application provides energy availability during emergencies or failures in the primary energy source, either the grid or renewable generation.

This application’s benefits include a power source for any disturbances in the primary grid, as observed in [19]. Additionally, automatic switching, aiming to trigger the BESS to supply stored energy to the MG or critical loads, is based on resource availability and load conditions, as discussed in [27]. This application also involves monitoring and management considerations to ensure the correct operation of the BESS, as seen in the article [26], which addresses dispatch strategies based on network behavior, load, and PV solar generation. Integrating a BESS with other systems can involve more complex functions, as demonstrated in [35], which presents an autonomous renewable energy hybrid system assisting in seawater desalination. It utilizes the BESS as a backup for high-demand loads and low renewable energy potential.

Regarding the energy sector in which BESS is implemented, the key findings are as follows (and Figure 13):

• Generation;
• Distribution;
• End-user.

BESSs can fulfill various functions in different energy sectors, including generation, transmission, distribution, and end-users. In the quantitative analysis of the energy sector, power generation is the most common area for BESS studies in this SLR. It can be utilized, in this context, for BESSs to store excess renewable energy. For instance, in the study [18], BESSs stand out as energy sources during extreme cases, such as power outages. Additionally, the work [36] employs BESSs to feed surplus energy back into the grid to accelerate payback, dependent on specific grid tariffs. Another approach in power generation involves BESSs responding rapidly to fluctuations in renewable energy sources, as demonstrated in the study [20], where BESS operations are unique to mitigating the variability in solar photovoltaic energy generation. This issue is also addressed in the article [37], which shows that BESSs can mitigate the intermittency of renewable generation. However, the study discusses that the larger the BESS, the smaller the rate of improvement in this context.

5.2. How Is Battery Technology Impacting the Hybrid Power System Sizing?

This research question addresses the pivotal point of the SLR, which is the sizing of BESSs for hybrid energy systems. The impacts of battery technologies can vary depending on the application, energy sector, system space allocation, cost, and lifespan, among other factors. To answer this question, data from each article in this SLR related to the technologies used were collected and illustrate the results in the graph shown in Figure 14.

From the graph, a significant number of articles do not use/address any specific technology, relying solely on the operational principles of a BESS. However, there is the utilization of the most common technologies currently, and that is reflected in the SLR:

• Lithium;
• Lead;
• Sodium.

The impact of sizing a BESS in a hybrid energy system can take various forms. In the SLR, this impact ranges from cost minimization to applications in MGs, reliability, performance improvement, and efficiency enhancement, among others. In the article [38], BESS helps maintain the stability of the electrical system in critical situations, reducing power supply interruptions and enhancing the quality of energy supplied to consumers. Moreover, the study mentions that an Energy Storage System’s (ESS’s) performance can be improved by selecting the optimal location and considering significant contingency impacts. Lithium ion batteries are used in this scenario, as they are suitable for immediate action applications. The study [39] shows a positive impact as batteries store energy generated by renewable resources in the nanogrid. There is an economy facilitated, without factoring in replacement costs, by the batteries in the nanogrid. Another important aspect is using BESSs to reduce diesel consumption, as mentioned in the article [40]. The BESS enables higher utilization of renewable sources, reducing greenhouse gas emissions. This study employs lithium ion technology, which synergizes well with renewable generation and can better address intermittency, providing more excellent stability to the grid. Another significant topic is BESSs’ participation in the energy market, as discussed in the article [41]. The article explores BESSs’ involvement in energy auctions, emphasizing optimal resource allocation and defined energy scheduling, i.e., energy transactions in the decentralized market. Furthermore, the study notes that the optimized system yields more benefits than conventional energy trading in the transmission network.

5.3. Which Machine Learning Approaches of Sizing Hybrid Power Systems with Batteries Are Employed?

Sizing BESSs within a hybrid, MG, or similar system presents a complex challenge. However, machine learning has become a crucial tool for sizing alongside optimization techniques. As observed in Section 5.1, optimization methods focus on finding approximate solutions concerning the size and capacity of hybrid system components, such as PV modules, wind turbines, and BESSs, among others, to achieve efficient and economical operation. Machine learning can be applied with optimization methods in specific process stages.

To address this research question, the data regarding machine learning approaches are illustrated in Figure 15:

From the graph, it is evident that many articles do not employ any machine learning techniques. However, at least three articles within the SLR specifically utilize Neural Networks (NNs).

The other machine learning approaches were found only once in articles within this SLR:

• Generative Adversarial Networks (GANs);
• Regression model;
• K-means clustering;
• Random Forest.

In the article [41], the third optimization step involves employing BESS to assist in energy trading, specifically in the distribution market. It simulates the auction bidding state, where all buyers and sellers can participate. Machine learning is utilized by training a Backpropagation (BP) Neural Network, where participants are classified into three categories. A sigmoidal function and an optimization algorithm are chosen to train the Neural Network. The Neural Network is fed, and the main characteristics are forward signal transmission and error backpropagation. In the study [15], the focus is on the significant impacts that may exist in determining the BESS size, specifically in the accuracy of meteorological data forecasting and how it impacts BESS sizing. Machine learning techniques are employed to mitigate the first mentioned impact, renewable generation forecasting. The study applies two variations of Neural Networks, namely, ENN and WNN, in addition to using the statistical technique of ARIMA for time series prediction. This prediction is made one day and one hour in advance, and the results are used for BESS sizing. The work [20] presents a sizing strategy for an isolated MG to find the best balance between generation and load. RNNs trained on 19 years of continuous hourly meteorological data are used to simulate weather uncertainty in PV production. These RNNs are enhanced with a Long-Short-Term Memory (LSTM) network to generate weather scenarios that exhibit long-term patterns, short-term autocorrelation, and a small random component. These generated data are used in the optimization problem to enhance MG robustness and reliability concerning weather uncertainty. The article [17] aims to predict energy consumption and PV generation based on historical consumption and energy generation data. These predictions determine the optimal battery size to maximize energy savings and reduce grid dependence. Machine learning techniques include the data clustering of net meter energy using K-means clustering, linear regression, and Random Forest. The K-means algorithm is applied differently for different seasons: summer, autumn, winter, and spring. Regression models are trained based on obtained seasonal cluster distributions and employed to predict seasonal cluster distributions of new net meter energy data for a specific period. The article [21] outlines an energy management strategy and optimal sizing for creating a 100% renewable MG. The Scikit-learn library in Python is utilized to implement machine learning algorithms like RF for wind speed, solar radiation, and PV energy generation prediction based on historical data.

6. Conclusions

This work reviewed the study of BESS sizing, or an MG with a BESS, using optimization techniques, with or without machine learning techniques. The protocol of this SLR selected 5828 articles, as shown in the graph in Figure 1, over four years. After removing duplicate articles, applying inclusion and exclusion criteria, and evaluating the quality criteria, as detailed in Section 3.3, 157 articles remained, and the study structure is detailed in Figure 10. After that, the articles were further filtered, needing to answer the RQs, as mentioned in Section 3.4, resulting in 44 articles, as seen in Figure 3. The main objective was to analyze the following questions: what is the methodology for sizing MG/hybrid systems with BESSs, which applications are used for BESSs, in which sector it is located, how the battery technology impacts sizing, and what machine learning approaches for sizing MG with BESSs are employed.

The optimization techniques for BESS sizing involve considering crucial parameters for achieving an optimal value for the studied system. Thus, mathematical modeling is carried out to estimate the behavior of each component within a hybrid system, resulting in the objective function. The parameters commonly addressed to attain an optimal value typically include minimizing energy or investment costs, the efficient operation of a hybrid plant/MG/BESS, and maximizing the use of renewable resources. Utilizing machine learning as part of the optimization process involves training an NN. The techniques employed are precise in forecasting meteorological data, directly impacting renewable energy generation. Furthermore, these techniques are essential for optimizing the charging and discharging operation according to energy market conditions and predicting consumption based on historical data. Thus, the integration of machine learning into the optimization process significantly contributes to finding efficient and economical solutions for hybrid systems with BESSs, as seen in [21].

For future research, it is advisable to delve deeper into the planning and operation of electrical systems with BESS, employing advanced optimization techniques in conjunction with Neural Networks. These approaches can be utilized to pinpoint more suitable locations for integrating BESSs into the electrical system, transforming them into a crucial component for reinforcing the grid during contingency situations. The optimization process would be instrumental in determining the ideal dimensions of the BESS. At the same time, Neural Networks would be employed to predict potential failures or contingencies that could impact the interconnected system. These predictions would be based on historical failure data, providing a more precise insight into the system’s vulnerabilities. This combination of advanced techniques could significantly enhance the reliability and resilience of electrical systems incorporating BESSs, contributing to a more stable and secure energy supply.