Artificial Intelligence and Metaheuristic Optimization Strategies for Renewable Microgrid Sizing and Design: A Scoping Review

Abstract

Optimal sizing and design of renewable microgrids and hybrid renewable energy systems require balancing renewable resource variability, demand uncertainty, storage operation, reliability, and techno-economic constraints. Artificial intelligence and metaheuristic optimization strategies have been increasingly used to address these challenges; however, the evidence remains methodologically heterogeneous. This scoping review maps the literature on artificial intelligence, learning-based, metaheuristic, heuristic, and optimization-based strategies for renewable microgrid sizing and design. The review followed PRISMA-ScR guidelines. Searches were conducted in Scopus and the Web of Science Core Collection for research articles published between 2009 and March 2026. A total of 69 studies were included. Metaheuristics dominated the field, appearing in 63 studies, with particle swarm optimization and genetic algorithm-based strategies as the most frequent methodological families. Artificial intelligence and learning-based strategies were mainly used to support forecasting, surrogate modeling, uncertainty handling, and energy management. The most recurrent configurations involved photovoltaic, wind, and battery storage systems, often with diesel backup in stand-alone or off-grid contexts. The literature is strongly oriented toward metaheuristic sizing of PV–wind–battery microgrids, with emerging integration of AI-assisted prediction and decision-support strategies. Future studies should address reproducibility, uncertainty modeling, real-world validation, degradation assessment, explainability, and scalability.

1. Introduction

1.1. Background

The transition toward low-carbon energy systems has increased interest in renewable microgrids and hybrid renewable energy systems, particularly for rural, remote, island, and community applications [1]. Microgrids can be understood as low-voltage distribution systems that integrate distributed generation, energy storage, local loads, and control resources, with the capacity to operate either connected to the main grid or in island mode [2]. In this context, hybrid renewable energy systems combine resources such as photovoltaic generation, wind energy, biomass, battery storage, and conventional backup units to improve supply reliability, renewable penetration, and operational flexibility [3].

However, the integration of variable renewable energy sources introduces technical and operational challenges. Photovoltaic generation depends on solar irradiance and temperature, whereas wind power output is conditioned by wind speed and its temporal variability [4]. These fluctuations complicate the balance between generation and demand, especially in autonomous or weak-grid systems. Therefore, energy storage, load management, and the appropriate selection of installed capacities are central aspects of the design of renewable microgrids.

Optimal sizing aims to determine the capacity of components such as photovoltaic panels, wind turbines, battery banks, converters, backup generators, electrolyzers, fuel cells, and other distributed energy resources. These decisions directly affect total cost, reliability, autonomy, renewable energy utilization, and environmental performance. Thus, sizing is not only a technical selection task but also a constrained optimization problem involving multiple objectives, uncertain inputs, and operational requirements [5]. Advanced computational methods have been increasingly applied across different areas of energy engineering, including numerical modeling, sensitivity analysis, optimization, and artificial intelligence-based approaches. These methods have supported the analysis of complex nonlinear phenomena in both conventional and renewable energy systems [6,7].

In renewable microgrids, artificial intelligence and metaheuristic optimization methods have been widely used to address sizing, uncertainty, forecasting, and operational challenges [8]. Genetic algorithms, particle swarm optimization, multi-objective particle swarm optimization, and non-dominated sorting genetic algorithm II have been applied to sizing problems involving economic, technical, environmental, and reliability criteria [9]. These methods are suitable for nonlinear, restricted, and high-dimensional search spaces, where deterministic methods may face computational or formulation limitations [10]. In addition, learning-based methods have expanded the possibilities for renewable resource forecasting, load modeling, uncertainty representation, and energy management in microgrids [11]. Their integration with metaheuristics has led to hybrid frameworks that combine prediction, optimization, and decision support for renewable microgrid design [12].

1.2. Rationale

The optimal sizing of renewable microgrids is complex because it involves the interaction of renewable resource variability, demand uncertainty, technical constraints, investment costs, and reliability requirements. Solar and wind resources fluctuate at hourly, daily, and seasonal scales, which increases uncertainty in estimating available generation [13]. Load profiles also vary according to user type, application context, consumption patterns, and socioeconomic conditions [14].

This complexity is greater in autonomous or islanded systems, where supply continuity depends on the appropriate combination of renewable generation, storage, and backup resources [15]. Undersizing may lead to supply deficits, while oversizing increases investment costs, energy surplus, and underutilization of assets. Consequently, capacity definition must balance cost, reliability, autonomy, renewable energy utilization, and operational feasibility.

The problem is also inherently multi-objective. Sizing studies commonly consider economic indicators such as net present cost, levelized cost of energy, annualized cost, initial investment, and operation and maintenance costs. At the same time, they include technical and reliability metrics such as loss of power supply, loss of load probability, energy not supplied, battery state of charge, and supply continuity. Broader formulations also incorporate environmental objectives, including greenhouse gas emissions, carbon footprint, and other system-related impacts [16].

Technical and operational restrictions add further complexity. These include component capacity limits, power balance, voltage and frequency constraints, battery operating limits, depth of discharge, charge and discharge rates, backup availability and minimum reliability requirements [5]. The simultaneous presence of continuous, discrete, nonlinear and non-convex variables makes optimal sizing difficult to solve through classical approaches when system scale, technological diversity or scenario uncertainty increases.

In this context, metaheuristics and artificial intelligence-based methods provide methodological advantages. They do not require explicit gradient information, can handle discontinuous or non-convex search spaces, and can be adapted to single- and multi-objective formulations [10]. Nevertheless, the literature reports a wide diversity of algorithms, technological configurations, objective functions, constraints, computational tools, data sources, and application contexts. This heterogeneity makes it difficult to identify dominant trends, methodological roles, and research gaps in the field.

1.3. Why a Scoping Review

Despite the growing literature on artificial intelligence and metaheuristics for the optimal sizing of renewable microgrids, the field remains methodologically and technologically heterogeneous. Studies differ in terms of system configurations, optimization criteria, performance metrics, computational tools, data sources, and application contexts, which limits direct comparison and makes it difficult to identify dominant methodological patterns and research gaps [17].

A scoping review is appropriate for mapping the extent, nature, and characteristics of evidence in broad or heterogeneous research areas [18,19]. Unlike systematic reviews focused on estimating effects or comparing specific interventions, a scoping review allows concepts, methods, types of evidence, and knowledge gaps to be organized descriptively [20]. The PRISMA-ScR framework supports transparent reporting of the search, selection, data charting, and synthesis processes [21].

Accordingly, this review maps the literature on optimal sizing of renewable microgrids and hybrid renewable energy systems using artificial intelligence, learning-based methods, metaheuristics, and related optimization algorithms. It focuses on the methods used, technological configurations addressed, objective functions and constraints considered, tools and data sources reported, and application contexts covered.

1.4. Objectives and Review Questions

The objective of this scoping review is to analyze the scientific literature on artificial intelligence, learning-based methods, metaheuristics, and optimization algorithms applied to the optimal sizing of renewable microgrids and hybrid renewable energy systems. Specifically, the review seeks to identify methodological trends, technological configurations, objective functions, constraints, performance metrics, computational tools, data sources, and application contexts reported in the included studies. The main review question was:

How have artificial intelligence, learning-based methods, metaheuristics and optimization algorithms been applied to the optimal sizing of renewable microgrids and hybrid renewable energy systems based on photovoltaic, wind or PV–wind technologies?

The specific review questions were:

RQ1. What artificial intelligence, learning-based, metaheuristic, and optimization techniques have been used for sizing renewable microgrids and hybrid renewable energy systems?

RQ2. What technological configurations have been most frequently addressed, considering photovoltaic systems, wind systems, PV–wind architectures, batteries, hydrogen storage, grid connection, backup generators, and other distributed energy resources?

RQ3. What objective functions, constraints, and performance metrics are used in sizing studies, considering economic, technical, environmental, energy-related, and reliability criteria?

RQ4. What computational tools, simulation platforms, modeling environments, and data sources are reported in the literature?

RQ5. In which application contexts have these studies been developed, particularly in residential, community, rural, remote, off-grid, commercial, industrial, university, and hospital microgrids?

Unlike previous reviews that mainly classify optimization algorithms and technological configurations, this scoping review examines methodological limitations in renewable microgrid sizing studies. The analysis covers reproducibility, algorithm-parameter reporting, uncertainty treatment, degradation modeling, benchmark selection, code and data availability, and real-world validation. These dimensions support the critical interpretation of methodological consistency, reporting quality, and practical applicability of the included studies.

2. Materials and Methods

2.1. Protocol and Registration

This scoping review was conducted following methodological guidance for scoping reviews [20] and reported according to PRISMA-ScR guidelines [21]. The review procedures were defined before data extraction, but the protocol was not formally registered in a public repository.

2.2. Eligibility Criteria

Eligibility criteria were defined according to the review scope and are summarized in Table 1. The review included research articles published between 2009 and 2026 that addressed optimal sizing, capacity optimization, optimal design, or optimal configuration of microgrids or hybrid renewable energy systems with a renewable energy component. Eligible studies had to apply artificial intelligence, learning-based methods, metaheuristics, heuristics, or related optimization algorithms. Records outside the defined period, non-research articles, studies unrelated to microgrids or hybrid renewable energy systems, and works focused only on operation, control, dispatch, or forecasting without sizing or design were excluded.

Table 1. Eligibility criteria used in the review.

Criterion	Inclusion	Exclusion
Document type	Research articles	Reviews, book chapters, conference papers, editorials, errata
Period	2009–2026	Outside the defined period
Language	No initial restriction; mainly English corpus	Records with insufficient metadata
System	Microgrids or hybrid renewable energy systems	Microgrids or HRES
Technology	PV, solar, wind, PV–wind, or renewable energy	No renewable energy component
Method	AI, ML, DL, RL, metaheuristics, heuristics, or optimization algorithms	No relevant computational method
Problem	Optimal sizing, capacity optimization, optimal design, or optimal configuration	Operation, control, dispatch, or forecasting without sizing/design

Studies combining capacity sizing with dispatch, energy management, stochastic operation, demand response, or control were included when component capacities or system configuration were explicit optimization variables or reported outcomes. Studies addressing only operation, forecasting, dispatch, or control without a sizing or design component were excluded.

2.3. Information Sources

The bibliographic search was conducted in Scopus and the Web of Science Core Collection. These databases were selected because of their broad coverage of peer-reviewed literature related to microgrids, hybrid renewable energy systems, optimal sizing, metaheuristic optimization, and artificial intelligence applied to energy systems. Scopus was used as the main source, while the Web of Science Core Collection was used as a complementary source to improve coverage and reduce the limitations of relying on a single database. The final search was conducted on 30 March 2026.

2.4. Search Strategy

The search strategy was developed to retrieve studies on microgrids or hybrid renewable energy systems, optimal sizing or design, renewable energy technologies, and computational optimization methods. The same conceptual search structure was applied in Scopus and the Web of Science Core Collection, with database-specific syntax adjustments. The search was restricted mainly to title fields for the system, sizing, and method terms. Database-specific field options were used for the renewable-energy block: in Scopus, these terms were searched in the article title, abstract, and keywords, whereas in Web of Science, the All Fields option was used. The remaining search blocks were restricted to the title field. Screenshots of the database-specific field configurations are provided in Supplementary Figures S1 and S2. The Scopus search string used as the reference strategy was:

TITLE(microgrid* OR “micro-grid*” OR “hybrid energy system*” OR “hybrid renewable energy system*” OR HRES OR “stand-alone energy system*” OR “standalone energy system*” OR “off-grid energy system*” OR “islanded microgrid*” OR “autonomous microgrid*” OR “distributed energy system*”)

AND TITLE (“optimal sizing” OR “optimizing the sizing” OR “sizing optimization” OR “microgrid sizing” OR “system sizing” OR “component sizing” OR “capacity sizing” OR “capacity optimization” OR “optimal design” OR “design optimization” OR “optimal configuration”)

AND TITLE-ABS-KEY (photovoltaic OR PV OR solar OR wind OR “PV/wind” OR “PV-wind” OR “solar-wind” OR “wind-solar” OR “renewable energy”)

AND TITLE (“artificial intelligence” OR “machine learning” OR “deep learning” OR “reinforcement learning” OR “genetic algorithm*” OR “particle swarm optimization” OR PSO OR “grey wolf optimizer” OR GWO OR “whale optimization algorithm” OR WOA OR “ant colony optimization” OR ACO OR “differential evolution” OR DE OR “firefly algorithm” OR “bat algorithm” OR “cuckoo search” OR “harmony search” OR “simulated annealing” OR metaheuristic* OR “meta-heuristic*” OR heuristic* OR “multi-objective optimization” OR “NSGA-II” OR NSGAII OR MOPSO).

The title-field restriction was applied as a methodological criterion to improve precision and retain studies in which sizing, system design, and computational methods constituted the central research focus. A broader title–abstract–keyword search would have retrieved a larger number of records in which these elements appeared only as secondary aspects.

2.5. Selection of Sources of Evidence

The selection process followed the PRISMA flow structure, including identification, duplicate removal, title and abstract screening, full-text retrieval, and eligibility assessment. Records retrieved from Scopus and the Web of Science Core Collection were merged into a single database, and duplicates were removed before screening. Unique records were assessed by title and abstract according to the eligibility criteria. Potentially eligible reports were then sought for full-text retrieval, and unavailable documents were recorded as not retrieved. Full-text reports were assessed in detail, reasons for exclusion were documented, and studies meeting all criteria were included in the final synthesis.

Title and abstract screening and full-text eligibility assessment were conducted by two reviewers. Potential disagreements regarding study inclusion were resolved through discussion and consensus.

2.6. Data Charting Process

Data were extracted using a charting matrix designed for this scoping review and aligned with the review objective and questions. The matrix was preliminarily checked to ensure clear fields, adequate categories, and consistent coding before being applied to all included studies.

2.7. Data Items

Data items were defined according to the review questions and included bibliographic information, system configuration, technologies, optimization method, sizing problem, objective functions, constraints, performance metrics, computational tools, data sources, and application context.

2.8. Critical Appraisal of Individual Sources

A formal critical appraisal of methodological quality or risk of bias was not conducted, as the purpose of this scoping review was to map methods, technological configurations, optimization criteria, tools, application contexts, and research gaps rather than to assess intervention effects or rank evidence quality.

2.9. Synthesis of Results

Results were synthesized using a descriptive and narrative approach consistent with the purpose of a scoping review. Extracted data were organized into categories derived from the review questions, including methodological approaches, technological configurations, objective functions, constraints, performance metrics, computational tools, data sources and application contexts. Findings were presented through narrative synthesis, classification tables and descriptive figures where appropriate.

3. Results

3.1. Selection of Sources of Evidence

The search identified 133 records. After removing 38 duplicates, 95 records were screened by title and abstract, and 4 were excluded. Of the 91 reports sought for full-text retrieval, 15 could not be obtained in full text and were therefore not assessed against the eligibility criteria. The remaining 76 reports were assessed for eligibility. Seven full-text reports were excluded: 4 did not address sizing, design, or installed capacity; 1 was a review; 1 was a duplicate; and 1 was outside the technological scope of microgrids or hybrid renewable energy systems. Finally, 69 studies met the eligibility criteria and were included in the review, as shown in Figure 1. These 69 studies constitute the final evidence corpus analyzed in the scoping review; the larger reference list also includes methodological guidance, background literature, previous reviews, and other supporting sources.

Figure 1. PRISMA flow diagram of the identification, screening, eligibility, and inclusion process.

3.2. Characteristics of Included Studies

The included studies were published between 2011 and March 2026. Although the eligibility period began in 2009, no studies published in 2009 or 2010 met the inclusion criteria. Therefore, the final evidence corpus covers publications from 2011 to March 2026.

As shown in Figure 2, publication output was limited before 2020 and increased thereafter, with the highest concentration between 2022 and 2025. The largest number of studies was published in 2025, with 13 studies, followed by 2022 with 9 studies, and 2023 and 2024 with 8 studies each. Regarding database sources, 65 studies were retrieved from Scopus and 4 from the Web of Science Core Collection. Most studies were published in English, while one study was published in Chinese and one in Spanish.

Figure 2. Annual distribution of the included studies.

3.3. Bibliometric Overview

The 69 included studies were distributed across 55 publication sources, indicating a dispersed literature. As shown in Table 2, Energies was the most frequent source, with four records, followed by Energy, with three records. Several journals contributed two studies each, while most remaining sources contributed only one study.

Table 2. Main publication sources for the included studies.

Journal	n	%
Energies	4	5.8
Energy	3	4.3
Frontiers in Energy Research	2	2.9
Mathematics	2	2.9
Renewable Energy	2	2.9
Energy Reports	2	2.9
International Journal of Power Electronics and Drive Systems	2	2.9
Sustainability (Switzerland)	2	2.9
Solar Energy	2	2.9
Processes	2	2.9

Regarding geographical distribution, the analysis was based on the affiliation countries reported in the records and full-text documents. China showed the highest frequency, with 13 studies, followed by India with 9, Iran with 6, Egypt with 5, and Iraq with 4. Contributions from Morocco, Brazil, the United States, Malaysia, South Africa, and Nigeria were also identified. These counts refer to affiliation countries and should not be interpreted as the countries where the case studies were conducted, as shown in Table 3.

Table 3. Distribution of included studies by affiliation country.

Affiliation Country	n	% of 69 Studies
China	13	18.8
India	9	13
Iran	6	8.7
Egypt	5	7.2
Iraq	4	5.8
Morocco	3	4.3
Brazil	3	4.3
United States	3	4.3
Malaysia	3	4.3
South Africa	3	4.3
Nigeria	3	4.3

The keyword analysis showed that the most frequent terms were renewable energy, optimal sizing, particle swarm optimization, microgrid, multi-objective optimization, genetic algorithm and hybrid renewable energy systems. The strongest co-occurrences were observed between particle swarm optimization and renewable energy, optimal sizing and renewable energy, microgrid and renewable energy, and microgrid and optimal sizing. These patterns indicate that the corpus is mainly concentrated on the optimal sizing of renewable microgrids and hybrid renewable energy systems using metaheuristic approaches, particularly swarm-based and multi-objective optimization methods.

Figure 3 presents the overlay visualization of the keyword co-occurrence network derived from the author keywords of the included studies. The network shows that the most central and strongly connected terms were “optimal sizing”, “microgrid”, “renewable energy”, “optimization”, “multiobjective optimization”, and “particle swarm optimization”. These keywords form the core thematic structure of the reviewed literature and confirm the predominance of optimization-oriented approaches for renewable microgrid design. The temporal overlay further indicates that established topics such as particle swarm optimization, genetic algorithms, wind power, and solar energy were more frequent in earlier studies, whereas more recent publications increasingly incorporated terms related to energy management, learning algorithms, cost-effectiveness, and optimization algorithms. Overall, the map suggests a gradual evolution from conventional metaheuristic sizing approaches toward broader frameworks integrating data-driven methods, energy management, and techno-economic considerations.

Figure 3. Overlay visualization of the keyword co-occurrence network of the included studies.

3.4. AI, Metaheuristic and Optimization Methods for Optimal Sizing

3.4.1. Overview of Methodological Approaches Used for Optimal Sizing

Figure 4 summarizes the methodological approaches identified in the included studies. The methods were grouped according to their predominant role in the sizing process, including dominant metaheuristics, alternative and emerging metaheuristics, AI-assisted sizing frameworks, mathematical or stochastic optimization approaches, and software-based comparative tools. These categories provide an overview of how different techniques support optimization, prediction, uncertainty handling, benchmarking, and decision-making in renewable microgrid and hybrid energy system sizing.

Figure 4. Overview of methodological families and predominant roles in optimal sizing studies.

3.4.2. Artificial Intelligence, Metaheuristic and Optimization Techniques Used for Optimal Sizing

The findings for RQ1 show that optimal sizing of renewable microgrids and hybrid renewable energy systems was mainly addressed using metaheuristic algorithms. Of the 69 included studies, 63 used at least one metaheuristic, confirming the predominance of these methods over artificial intelligence, learning-based, and mathematical optimization approaches [22,23,24,25,26].

Particle swarm optimization (PSO) and its multi-objective variants were the most frequent family, appearing in 31 studies. As shown in Table 4, these methods were applied to a wide range of configurations, including PV–wind–diesel–battery systems [22,26,27], PV–wind–battery systems [25,28,29], hydrogen-based systems [24,30], biomass or biogas-supported systems [23,31], and electric vehicle-integrated microgrids [3,32]. Methodologically, PSO/MOPSO-based techniques were used mainly as core sizing algorithms [22,33,34], although some studies employed them for benchmarking [3,24], multi-objective optimization [25,26,35], forecasting-assisted sizing [29,36] or comparison with tools such as HOMER [37,38].

As shown in Table 5, GA/NSGA-II-based approaches and their variants represented the second most frequent methodological family for optimal sizing. These methods were identified in studies involving stand-alone hybrid renewable energy systems, PV–wind–biomass configurations, PV–wind–diesel–battery microgrids, hydrogen-storage systems, energy storage microgrids, and electric vehicle charging applications [23,24,25,27,39,40].

Table 4. PSO/MOPSO-based techniques and variants used for optimal sizing in the included studies.

Technique Role	PSO/MOPSO Variants	Typical Configurations	Studies
Main sizing	PSO, IMOPSO, FPSO, PSOPC, MPSO, BPSO	Off-grid, autonomous, PV–wind–battery, hydrogen, biomass	[24,33,35,41,42,43]
Multi-objective sizing	MOPSO, M-MOPSO, ALA-QPSO, SA-PSO	PV–wind–diesel–battery, EV-integrated, multi-resource systems	[27,28,34,37,39,44]
Forecasting/surrogate support	SVR-PSO, PSO with forecasting	PV–wind–battery and off-grid systems	[31,38]
Benchmarking/comparison	PSO, MOPSO	HOMER and inter-algorithm comparison	[26,40,45,46]
Sizing with control or placement	PSO-Fuzzy, PSO	Agricultural microgrids and DG placement	[40,47]

Table 4. PSO/MOPSO-based techniques and variants used for optimal sizing in the included studies.

Technique Role	PSO/MOPSO Variants	Typical Configurations	Studies
Main sizing	PSO, IMOPSO, FPSO, PSOPC, MPSO, BPSO	Off-grid, autonomous, PV–wind–battery, hydrogen, biomass	[24,33,35,41,42,43]
Multi-objective sizing	MOPSO, M-MOPSO, ALA-QPSO, SA-PSO	PV–wind–diesel–battery, EV-integrated, multi-resource systems	[27,28,34,37,39,44]
Forecasting/surrogate support	SVR-PSO, PSO with forecasting	PV–wind–battery and off-grid systems	[31,38]
Benchmarking/comparison	PSO, MOPSO	HOMER and inter-algorithm comparison	[26,40,45,46]
Sizing with control or placement	PSO-Fuzzy, PSO	Agricultural microgrids and DG placement	[40,47]

GA and NSGA-II were used in classical, multi-objective and hybrid forms, including improved NSGA-II, GA-VNS, MOGA, HGA, SAGA and GA combined with surrogate or metamodeling approaches. Their methodological role was mainly associated with main sizing, multi-objective optimization, and comparative benchmarking, particularly in problems involving cost, reliability, emissions, and component selection trade-offs [26,28,48,49,50]. Table 5 provides the study-level mapping of GA/NSGA-II-based techniques, system configurations, and methodological roles.

Table 5. GA/NSGA-II-based techniques and variants used for optimal sizing in the included studies.

Article	GA/NSGA-II or Identified Variant	Microgrid/HRES Type	Technique Role
[48]	Improved NSGA-II	DG + storage microgrid	Main sizing
[23]	GA	Stand-alone PV–wind–biomass HRES	Main/comparative
[49]	GA-VNS	HRES for substation auxiliary services	Sizing + uncertainty
[51]	SAGA	CHP microgrid	Multi-objective sizing
	GA + metamodel/DOE	Renewable energy microgrid	Surrogate-assisted sizing
[3]	GA	Off-grid microgrid + EV charging	Benchmarking
[52]	GA	Stand-alone HRES	Main sizing
[39]	HGA	PV–wind–diesel–battery microgrid	Main sizing
[24]	GA	Hydrogen-storage microgrid	Benchmarking
[36]	GA	Off-grid PV–wind–diesel–battery HRES	Forecasting + sizing
[53]	GA	Rural microgrid	Main sizing/economic operation
[54]	GA	Off-grid DC hybrid microgrid	Main sizing
[28]	MOGA	PV/BESS microgrid in remote area	Sizing + location
[27]	GA	Stand-alone PV–wind–battery HRES	Main sizing
[40]	MOGA + chaos optimization	Energy storage microgrid	Multi-objective sizing
[25]	NSGA-II comparison	Solar–wind–battery HRES	Comparative benchmark
[55]	GA	Hybrid energy systems	Main/comparative
[26]	NSGA-II	Stand-alone HRES	Multi-objective comparison
[56]	GA	Location-based HRES	Main/comparative

Table 5. GA/NSGA-II-based techniques and variants used for optimal sizing in the included studies.

Article	GA/NSGA-II or Identified Variant	Microgrid/HRES Type	Technique Role
[48]	Improved NSGA-II	DG + storage microgrid	Main sizing
[23]	GA	Stand-alone PV–wind–biomass HRES	Main/comparative
[49]	GA-VNS	HRES for substation auxiliary services	Sizing + uncertainty
[51]	SAGA	CHP microgrid	Multi-objective sizing
	GA + metamodel/DOE	Renewable energy microgrid	Surrogate-assisted sizing
[3]	GA	Off-grid microgrid + EV charging	Benchmarking
[52]	GA	Stand-alone HRES	Main sizing
[39]	HGA	PV–wind–diesel–battery microgrid	Main sizing
[24]	GA	Hydrogen-storage microgrid	Benchmarking
[36]	GA	Off-grid PV–wind–diesel–battery HRES	Forecasting + sizing
[53]	GA	Rural microgrid	Main sizing/economic operation
[54]	GA	Off-grid DC hybrid microgrid	Main sizing
[28]	MOGA	PV/BESS microgrid in remote area	Sizing + location
[27]	GA	Stand-alone PV–wind–battery HRES	Main sizing
[40]	MOGA + chaos optimization	Energy storage microgrid	Multi-objective sizing
[25]	NSGA-II comparison	Solar–wind–battery HRES	Comparative benchmark
[55]	GA	Hybrid energy systems	Main/comparative
[26]	NSGA-II	Stand-alone HRES	Multi-objective comparison
[56]	GA	Location-based HRES	Main/comparative

In addition to PSO/MOPSO- and GA/NSGA-II-based approaches, several other metaheuristics were identified for optimal sizing, as shown in Table 6. These included established algorithms such as the grey wolf optimizer, differential evolution, cuckoo search, harmony search, firefly algorithm, bat algorithm, artificial bee colony, and whale optimization algorithm, as well as more recent or less conventional approaches such as the artificial hummingbird algorithm, transient search optimization, coot optimization, child drawing optimization, and golden ratio optimization [24,57,58,59,60,61].

These techniques were applied to different microgrid and hybrid renewable energy system configurations, including PV–wind systems, grid-connected microgrids, hydrogen-based systems, PV–diesel–battery systems, wind–battery systems, and electric vehicle-integrated microgrids [41,51,62,63,64,65]. In most cases, these methods were used as main sizing algorithms, although some studies used them for benchmarking, comparison with other metaheuristics, or predictive/surrogate-assisted sizing [3,43,66,67]. Table 6 provides the study-level mapping of alternative and emerging metaheuristic techniques, system configurations, and methodological roles.

Table 6. Other metaheuristic techniques and variants used for optimal sizing in the included studies.

Article	Other Metaheuristic Techniques	Microgrid/HRES Type	Technique Role
[68]	BAT/BOA	Stand-alone microgrids	Main sizing
[57]	GWO-CS	Grid-connected microgrid	Main sizing
[62]	DGSHS/HS	PV-wind HRES	Main sizing
[69]	FA	Stand-alone hybrid AC/DC microgrid	Main sizing
[63]	RW-GWO-VMD	Wind-solar microgrid with HESS	Main sizing
[58]	DE	PV-WT HRES	Main sizing
[70]	GWO	Autonomous hybrid microgrid	Main sizing
[3]	AHA; BAT/BOA; CSA; DE; GWO	Off-grid microgrid + EV demand response	Benchmarking
[24]	ABC; HS	Hydrogen storage-based microgrid	Benchmarking
[53]	GWO; CSA (crow search algorithm)	Rural-area microgrid	Main sizing
[41]	MOSaDE	PV/wind/diesel hybrid microgrid	Main sizing
[66]	FA; HS	HRES	Benchmarking
[67]	DE variants	Autonomous microgrid	Predictive/surrogate + sizing
[59]	IMOWOA	Multi-microgrids with shared H2 storage	Main sizing
[60]	CDO; DCCS; GRO	Stand-alone HRES in remote region	Main sizing
[64]	GWO	PV-diesel-battery HRES	Main sizing
[71]	CSA (cuckoo search algorithm)	Remote HRES	Main sizing
[61]	TDO; COOT; GWO; BWO	Stand-alone HRES	Main sizing
[72]	IMOGWO	Island hybrid microgrid	Main sizing
[73]	FA	Off-grid HRES	Main sizing
[43]	DE comparison	HRES for smart cities	Proposed optimization + comparison
[56]	ABC	Location-based HRES	Main sizing
[65]	AHA	Isolated microgrid + EVs	Predictive/surrogate + sizing
[74]	IBA	Battery ESS for microgrid	Main sizing
[75]	HS	Wind-battery hybrid microgrid	Main sizing

Table 6. Other metaheuristic techniques and variants used for optimal sizing in the included studies.

Article	Other Metaheuristic Techniques	Microgrid/HRES Type	Technique Role
[68]	BAT/BOA	Stand-alone microgrids	Main sizing
[57]	GWO-CS	Grid-connected microgrid	Main sizing
[62]	DGSHS/HS	PV-wind HRES	Main sizing
[69]	FA	Stand-alone hybrid AC/DC microgrid	Main sizing
[63]	RW-GWO-VMD	Wind-solar microgrid with HESS	Main sizing
[58]	DE	PV-WT HRES	Main sizing
[70]	GWO	Autonomous hybrid microgrid	Main sizing
[3]	AHA; BAT/BOA; CSA; DE; GWO	Off-grid microgrid + EV demand response	Benchmarking
[24]	ABC; HS	Hydrogen storage-based microgrid	Benchmarking
[53]	GWO; CSA (crow search algorithm)	Rural-area microgrid	Main sizing
[41]	MOSaDE	PV/wind/diesel hybrid microgrid	Main sizing
[66]	FA; HS	HRES	Benchmarking
[67]	DE variants	Autonomous microgrid	Predictive/surrogate + sizing
[59]	IMOWOA	Multi-microgrids with shared H2 storage	Main sizing
[60]	CDO; DCCS; GRO	Stand-alone HRES in remote region	Main sizing
[64]	GWO	PV-diesel-battery HRES	Main sizing
[71]	CSA (cuckoo search algorithm)	Remote HRES	Main sizing
[61]	TDO; COOT; GWO; BWO	Stand-alone HRES	Main sizing
[72]	IMOGWO	Island hybrid microgrid	Main sizing
[73]	FA	Off-grid HRES	Main sizing
[43]	DE comparison	HRES for smart cities	Proposed optimization + comparison
[56]	ABC	Location-based HRES	Main sizing
[65]	AHA	Isolated microgrid + EVs	Predictive/surrogate + sizing
[74]	IBA	Battery ESS for microgrid	Main sizing
[75]	HS	Wind-battery hybrid microgrid	Main sizing

As shown in Table 7, artificial intelligence, machine learning, deep learning, and reinforcement learning techniques appeared less frequently than classical metaheuristics. These methods were mainly used as complementary components of the sizing process, particularly for forecasting, uncertainty modeling, surrogate modeling, energy management, and hybrid optimization support [29,36,76,77,78].

The identified techniques included deep reinforcement learning, multi-agent deep reinforcement learning, neural networks, long short-term memory models, support vector regression, random forest, gradient boosting models, fuzzy logic, and hybrid learning–optimization schemes. Only a few studies used these approaches as the main sizing algorithm [39,79,80], while most integrated them with metaheuristics or optimization models to improve input data quality, represent uncertainty or support decision-making [29,50,81]. Table 7 provides the study-level mapping of AI- and learning-based techniques, system configurations, and methodological roles.

Table 7. AI- and learning-based techniques used for optimal sizing in the included studies.

Article	AI/ML/DL/RL or Identified Technique	Microgrid/HRES Type	Technique Role
[79]	multi-agent deep reinforcement learning (MADRL)	HRES microgrid	Main sizing algorithm
[57]	fuzzy logic; hybrid GWO-CS	Grid-connected microgrid	Main sizing algorithm
[30]	fuzzy-based PSO (FPSO)	Stand-alone solar-hydrogen HES	Main sizing algorithm
[76]	ML-based uncertainty sets; neural networks (NN); VAR	Stand-alone HRES	Predictive/surrogate support + sizing
[49]	Monte Carlo simulation	HRES for substation auxiliary services	Predictive/surrogate support + sizing
[50]	metamodel/design of experiments	Renewable energy microgrid	Predictive/surrogate support + sizing
[80]	DDQN; deep reinforcement learning (DRL)	Grid-tied microgrid	Main sizing algorithm
[36]	DWT-LSTM meteorological forecast	Off-grid HES	Predictive/surrogate support + sizing
[77]	deep learning control; transfer learning	BESS-based microgrid configurations	Predictive/surrogate support + sizing
[67]	AI/ML-coded DE variants	Autonomous microgrid	Predictive/surrogate support + sizing
[81]	CatBoost; LightGBM; XGBoost	HRES with net-metering	Predictive/surrogate support + sizing
[78]	XGBoost; RF; SVR; ML models	Hydrogen storage-based microgrid	Predictive/surrogate support + sizing
[29]	SVR; hybrid SVR-HHO; hybrid SVR-PSO	Off-grid energy system	Predictive/surrogate support + sizing
[82]	two-stage PSO-fuzzy hybrid	Special-purpose microgrid	Main sizing algorithm

Table 7. AI- and learning-based techniques used for optimal sizing in the included studies.

Article	AI/ML/DL/RL or Identified Technique	Microgrid/HRES Type	Technique Role
[79]	multi-agent deep reinforcement learning (MADRL)	HRES microgrid	Main sizing algorithm
[57]	fuzzy logic; hybrid GWO-CS	Grid-connected microgrid	Main sizing algorithm
[30]	fuzzy-based PSO (FPSO)	Stand-alone solar-hydrogen HES	Main sizing algorithm
[76]	ML-based uncertainty sets; neural networks (NN); VAR	Stand-alone HRES	Predictive/surrogate support + sizing
[49]	Monte Carlo simulation	HRES for substation auxiliary services	Predictive/surrogate support + sizing
[50]	metamodel/design of experiments	Renewable energy microgrid	Predictive/surrogate support + sizing
[80]	DDQN; deep reinforcement learning (DRL)	Grid-tied microgrid	Main sizing algorithm
[36]	DWT-LSTM meteorological forecast	Off-grid HES	Predictive/surrogate support + sizing
[77]	deep learning control; transfer learning	BESS-based microgrid configurations	Predictive/surrogate support + sizing
[67]	AI/ML-coded DE variants	Autonomous microgrid	Predictive/surrogate support + sizing
[81]	CatBoost; LightGBM; XGBoost	HRES with net-metering	Predictive/surrogate support + sizing
[78]	XGBoost; RF; SVR; ML models	Hydrogen storage-based microgrid	Predictive/surrogate support + sizing
[29]	SVR; hybrid SVR-HHO; hybrid SVR-PSO	Off-grid energy system	Predictive/surrogate support + sizing
[82]	two-stage PSO-fuzzy hybrid	Special-purpose microgrid	Main sizing algorithm

As shown in Table 8, mathematical optimization, uncertainty analysis, multicriteria decision-making, and HOMER-based approaches appeared less frequently than metaheuristics, but they played relevant complementary roles in the sizing process. These methods were mainly used to formulate constraints, assess uncertainty, compare alternatives, support decision-making, and validate proposed configurations [3,32,49,83,84].

Table 8. Mathematical, uncertainty, multicriteria, and HOMER-based approaches used for optimal sizing in the included studies.

Article	Mathematical, Uncertainty, MCDM or HOMER-Based	Technique Role	Microgrid/HRES Type
[83]	Entropy weighting; TOPSIS	Main sizing algorithm	Microgrid with ESS
[38]	HOMER comparison	Proposed optimization + comparison	Microgrid + water heating
[76]	ML-based uncertainty sets; adaptive robust optimization	Predictive/surrogate support + sizing	Stand-alone HRES
[84]	MILP/PLEM	Mathematical optimization model	PV-battery-wind-diesel hybrid microgrid
[49]	Monte Carlo simulation	Predictive/surrogate support + sizing	HRES for substation
[51]	Triangular evaluation model	Main sizing algorithm	CHP microgrid
[85]	Multi-criteria optimization	Main sizing algorithm	HRES
[3]	Stochastic uncertainty quantification	Benchmarking/comparison	Off-grid microgrid + EV charging
[37]	HOMER Pro	Benchmarking/comparison	Residential HES
[32]	Monte Carlo/uncertainty scenario analysis	Predictive/surrogate support + sizing	HRES + EVs
[65]	Uncertainty quantification	Predictive/surrogate support + sizing	Isolated microgrid + EVs

The identified approaches included mixed-integer linear programming, robust or stochastic optimization, Monte Carlo simulation, multicriteria decision-making, TOPSIS, and HOMER/HOMER Pro. HOMER-based approaches were used mainly as simulation, benchmarking, or comparison tools rather than as advanced optimization algorithms [37,38]. Table 8 provides the study-level mapping of mathematical, uncertainty, multicriteria, and HOMER-based approaches, system configurations, and methodological roles.

3.4.3. Relationship Between Methodological Family and Technique Role

Figure 5 shows the relationship between methodological family and technique role in the included studies. Metaheuristics were mainly used as primary sizing algorithms, with 49 family–role co-occurrences. In contrast, AI- and learning-based methods were more frequently associated with forecasting, surrogate modeling, and predictive support. Mathematical, stochastic, and multicriteria approaches showed lower frequencies but covered a broader set of roles, including uncertainty assessment, formal optimization, and decision support. Software-based tools were mainly used for benchmarking or comparison. Overall, the heatmap confirms that metaheuristics were not only the most frequent methods but also the central methodological approach in optimal sizing studies.

Figure 5. Relationship between methodological family and predominant technique role in the included studies. Note. Cell values represent the number of family–role co-occurrences. A single study may include more than one methodological family or technique role; therefore, frequencies are not mutually exclusive and should not be interpreted as article counts.

3.5. Technological Configurations Addressed in Optimal Sizing Studies

The technological configurations identified in the included studies were mainly concentrated on hybrid renewable systems combining photovoltaic or solar generation, wind energy, and battery storage. As shown in Table 9, the most frequent configuration was PV/solar + wind + battery/BESS + diesel/backup, identified in 12 studies and mainly associated with stand-alone, off-grid, islanded or autonomous systems [22,26,34,39,47]. In these architectures, diesel backup was commonly incorporated to improve supply reliability and compensate for renewable resource variability.

Table 9. Main technological configurations identified in the included studies.

Configuration Pattern	Studies	Articles	Typical Context
PV/solar + wind + battery/BESS + diesel/backup	12	[22,26,34,36,39,41, 47,69,71,72,76,84]	Mainly stand-alone, off-grid, island or autonomous systems
PV/solar + wind + battery/BESS	8	[25,27,29,33,40,56,62,73]	Stand-alone, remote-area, rural/community or microgrid applications
PV/solar + battery/BESS + diesel/backup	3	[44,53,64]	Stand-alone or off-grid autonomous systems, often in rural/community contexts
PV/solar + battery/BESS + hydrogen/fuel cell	3	[24,30,68]	Stand-alone or emergency hydrogen-based hybrid systems
PV/solar + wind + battery/BESS + thermal/CHP/special load	3	[43,58,82]	Stand-alone systems with specific thermal or special-load applications
PV/solar + wind + biomass/biogas/waste	3	[35,66,67]	Rural/community and off-grid renewable systems using dispatchable bioenergy support
Wind + battery/BESS	3	[63,75,86]	Off-grid systems with hybrid or specialized storage configurations
Wind + battery/BESS + diesel/backup	3	[52,60,61]	Stand-alone/off-grid systems with diesel backup for reliability
PV/solar + wind	2	[55,87]	Stand-alone/off-grid renewable systems, sometimes with deferrable or thermal loads
PV/solar + wind + battery/BESS + biomass/biogas/waste + grid/distribution	2	[31,57]	Mixed architecture combining renewable generation, storage and distribution-network interaction
PV/solar + wind + battery/BESS + grid/distribution	2	[74,80]	Grid-connected or distribution-network microgrid context
Grid/distribution + storage unspecified/hybrid	2	[83,88]	Distribution-network context with storage-oriented configurations

The second most frequent configuration was PV/solar + wind + battery/BESS, reported in 8 studies [25,27,29,57]. This pattern reflects the central role of battery storage in balancing renewable generation, reducing supply deficits, and increasing autonomy in rural, remote, or community applications. Less frequent but relevant configurations included PV/solar + battery/BESS + diesel/backup [44,53,64], hydrogen- or fuel-cell-based systems [24,30,68], bioenergy-supported systems [35,66,67], wind–battery systems [63,75], and architectures integrating thermal loads, electric vehicles, or grid/distribution interaction [32,56,74,80].

Overall, the results show that optimal sizing studies have focused primarily on PV–wind–battery/BESS architectures, with diesel backup frequently used when reliability in isolated systems is prioritized. Hydrogen, biomass, thermal loads, electric vehicles, and grid-connected configurations appear as more specific or emerging extensions within the field. Table 9 provides the study-level mapping of the main technological configurations, article codes, and typical application contexts.

Representative architectures included off-grid PV–wind–battery–diesel systems, in which photovoltaic and wind generation supplied the primary renewable energy, batteries balanced short-term variability, and diesel generators provided backup during prolonged renewable deficits. PV–wind–battery systems operated without conventional backup and therefore required greater storage capacity or controlled load curtailment to maintain reliability. Hydrogen-based architectures incorporated electrolyzers to convert surplus renewable electricity into hydrogen, storage tanks for long-term energy storage, and fuel cells for electricity recovery during periods of low renewable generation. Grid-connected microgrids combined distributed renewable generation, battery storage, local loads, and bidirectional power exchange with the utility grid, which acted as an additional balancing and reliability resource.

3.6. Objective Functions, Constraints and Performance Metrics

Because several objective functions and performance metrics were repeated across multiple studies, Table 10 reports representative articles for each category. The identified objective functions, constraints, and metrics were grouped into six dimensions: economic, reliability, technical/operational, environmental, energy-related, and algorithmic. Economic criteria were among the most recurrent, indicating that optimal sizing was frequently formulated as a cost-minimization problem. Common indicators included net present cost, total net present cost, levelized cost of energy, cost of energy, annualized cost, life cycle cost, investment cost, and operation and maintenance cost [22,23,85].

Table 10. Main objective functions, constraints, and performance metrics identified in the included studies.

Dimension	Main Objective Functions/Metrics	Main Role in the Optimization Problem	Representative Articles
Economic	NPC/TNPC, LCOE/COE, ASC/ACS/TAC, LCC, investment and O&M cost	Mainly minimized as the primary objective	[22,23,39,41,46,49,68,85]
Reliability	LPSP, LOLP, LOLE, EENS, unmet load, interruption duration	Minimized or used as reliability constraint	[22,32,41,51,58,62,63,85]
Technical/operational	SOC, DOD, charge/discharge limits, energy balance, voltage/frequency stability	Used mainly as constraints or feasibility metrics	[28,38,46,49,63,80,83]
Environmental	CO2 emissions, GHG emissions, carbon footprint, environmental cost	Minimized in multi-objective formulations	[23,38,39,41,48,50,52,85]
Energy	Renewable fraction, renewable penetration, excess/dumped/curtailed energy	Used to assess renewable utilization	[23,39,41,51,57,70,80,85]
Algorithmic	Convergence curve, computation time, best fitness value, number of iterations	Used to compare optimization performance	[36,46,49,50,68,85]

Reliability was another central dimension, usually represented by loss of power supply probability, loss of load probability, loss of load expectation, expected energy not supplied, unmet load, and interruption duration [32,62,63]. Technical and operational criteria were mainly used as feasibility constraints, including battery state of charge, depth of discharge, charge/discharge limits, energy balance, and voltage or frequency stability [28,80].

Environmental and energy-related criteria appeared mainly in multi-objective formulations. These included carbon dioxide emissions, greenhouse gas emissions, carbon footprint, environmental cost, renewable fraction, renewable penetration, and excess, dumped, or curtailed energy [23,38,48,85]. In addition, some studies reported algorithmic metrics such as convergence curves, computation time, best fitness value, and number of iterations to compare optimization performance [49,50,68]. Overall, the reviewed literature shows a strong economic–reliability orientation, while environmental, energy-related, technical, and algorithmic criteria indicate a gradual transition toward broader multi-objective formulations. Table 10 summarizes the main categories, roles, and representative studies.

3.7. Computational Tools, Data Sources

Figure 6 shows the distribution of the main computational tools and platforms across the grouped technological configurations. MATLAB was the most widely reported environment and appeared across nearly all configuration groups, particularly in PV–wind–battery, PV–wind–battery–diesel, hydrogen/fuel-cell, and other hybrid systems. HOMER/HOMER Pro was the second most frequent platform and was mainly associated with techno-economic simulation, configuration comparison, and validation of hybrid renewable energy systems.

Figure 6. Computational tools, modeling environments, and analytical approaches reported across grouped technological configurations.

The use of computational tools varied according to system architecture. Deep-learning frameworks were concentrated in PV–wind–battery–diesel, PV–wind–battery, thermal or special-load, and other configurations, whereas Monte Carlo simulation appeared mainly in configurations requiring uncertainty or scenario analysis. Python, DOE/metamodel approaches, PVsyst, and Simulink were reported less frequently and were linked to more specific tasks, including surrogate modeling, photovoltaic performance assessment, dynamic simulation, and computational support.

Overall, the heatmap indicates that tool selection depends on both methodological requirements and technological configuration. MATLAB provides broad implementation flexibility, while HOMER/HOMER Pro offers a more standardized environment for simulation and benchmarking. The less frequent use of Python, DOE/metamodel, PVsyst, and Simulink suggests that customized, data-driven, or specialized modeling workflows remain comparatively limited in the reviewed literature.

Figure 7 shows how the main data-source categories were distributed across the grouped technological configurations. Cost and component databases, local meteorological data, measured or real load profiles, and Monte Carlo or scenario-based data were the most consistently reported sources across the reviewed configurations. Their prevalence confirms that optimal sizing studies depend primarily on techno-economic assumptions, renewable-resource characterization, demand representation, and uncertainty-related inputs.

Figure 7. Data-source categories reported across grouped technological configurations. Note. Each cell shows the number and percentage of studies within the corresponding configuration group reporting the data source. Percentages are calculated by row and are not mutually exclusive because a single study may report multiple data-source categories.

The distribution nevertheless varied by technological architecture. PV–wind–battery–diesel, PV–wind–battery, hydrogen/fuel-cell, and wind–battery systems showed broad use of meteorological, load, cost, and scenario data. In contrast, electricity tariff data, HOMER databases, water or thermal demand data, and NASA/NASA POWER were less frequent and were concentrated in specific configurations. Grid-connected systems relied comparatively more on measured load and scenario-based information, while thermal, electric-vehicle, and other special-load configurations showed a more heterogeneous combination of data sources.

Overall, the heatmap indicates that data-source selection is strongly conditioned by the technological configuration and application requirements. Conventional hybrid systems rely mainly on cost, meteorological, load, and scenario data, whereas specialized configurations require additional sources related to tariffs, thermal demand, electric mobility, or external databases.

Figure 8 shows the distribution of reported data types across the main technological configurations. Battery/SOC parameters and economic data were consistently reported in most configuration groups, confirming their central role in techno-economic sizing studies. Load profiles, grid/network data, and meteorological variables were also widely used, although their frequency varied according to the system architecture. Solar resource data was more common in PV-based configurations, whereas wind resource data was more prominent in wind-integrated systems.

Figure 8. Data types reported across grouped technological configurations.

Hydrogen/fuel-cell configurations showed broad use of economic, load, solar, environmental, and battery-related data, reflecting the larger number of components and operating assumptions required for their assessment. Thermal, water, and EV charging data remained concentrated in specialized configurations and were rarely reported in conventional PV–wind–battery systems. Overall, the heatmap shows that data requirements are strongly conditioned by technological architecture and that most studies still rely on conventional techno-economic, demand, storage, and meteorological inputs.

3.8. Reproducibility and Reporting Assessment

Table 11 shows the results of the reproducibility and reporting assessment across the nine evaluated criteria. Load profile resolution and component cost assumptions were the most frequently reported items, identified in 57 and 61 studies, respectively. Benchmark comparisons were also common, appearing in 52 studies. In contrast, algorithm parameters, convergence criteria, uncertainty treatment, and code or data availability were reported less consistently. Only 8 studies provided sufficient information on algorithm parameters, 20 reported explicit convergence or stopping criteria, and 23 incorporated formal uncertainty scenarios. Public code or data availability was identified in only one study, while 54 studies did not report any repository or availability statement.

Table 11. Reporting and reproducibility indicators across the included studies.

Criterion	Yes	Partial	No/Not Reported	Not Applicable	Yes (%)
Input data sources	16	42	11	0	23.2%
Load profile resolution	57	4	8	0	82.6%
Component cost assumptions	61	1	7	0	88.4%
Battery degradation	6	38	21	4	9.2%
Uncertainty scenarios	23	17	29	0	33.3%
Algorithm parameters	8	22	39	0	11.6%
Convergence criteria	20	13	36	0	29.0%
Benchmark comparison	52	0	17	0	75.4%
Code/data availability	1	14	54	0	1.4%

Battery degradation was explicitly modeled in 6 studies and partially addressed in 38 through battery lifetime, replacement, depth-of-discharge, or cycle-life assumptions. Twenty-one battery-based studies did not report degradation-related information, while four studies were classified as not applicable because their configurations did not include battery storage. These results indicate that the main reporting gaps are concentrated in algorithm reproducibility, convergence specification, uncertainty representation, degradation modeling, and open access to code and data.

3.9. Application Contexts

Figure 9 shows the distribution of the main application contexts reported in the included studies. Rural, remote, community, and residential applications accounted for the largest share of the reviewed literature, followed by industrial or critical-infrastructure settings. This distribution reflects the strong orientation of optimal sizing studies toward decentralized energy systems in which autonomy, supply continuity, renewable-resource utilization, and reduced dependence on the main grid are central design requirements.

Figure 9. Primary application contexts reported in the included studies.

Off-grid, islanded, and stand-alone applications were also prominent, confirming that much of the literature addresses systems operating under limited or absent grid support. In contrast, commercial buildings, electric-vehicle-integrated systems, grid-connected applications, thermal/water/CHP configurations, hospitals, and university campuses were reported less frequently. Their limited representation indicates that sizing methods have been evaluated less extensively in complex demand environments, sector-coupled systems, and critical-load applications, where operational constraints, temporal variability, and reliability requirements may differ substantially from those of conventional rural or residential microgrids.

4. Discussion

4.1. Principal Findings and Comparison with Previous Reviews

This scoping review mapped 69 studies on the optimal sizing of renewable microgrids and hybrid renewable energy systems using artificial intelligence, learning-based methods, metaheuristics, and related optimization algorithms. The evidence shows a clear increase in publications after 2020 and a strong concentration on PV–wind–battery/BESS architectures, off-grid or stand-alone systems, rural or community applications, and economic–reliability-oriented formulations.

The predominance of metaheuristics confirms that optimal sizing continues to be treated mainly as a nonlinear, constrained, and multi-objective optimization problem. PSO/MOPSO-based approaches were the most frequent family, followed by GA/NSGA-II and other bioinspired or evolutionary algorithms. This pattern is consistent with previous reviews that identified high methodological diversity in hybrid renewable energy systems and the widespread use of heuristic and metaheuristic methods for sizing and configuration problems [17,89]. However, the present review extends those contributions by mapping the methodological role of each technique, distinguishing between main sizing algorithms, benchmarking tools, forecasting-assisted approaches, uncertainty-support methods, and software-based comparisons.

The technological configurations also confirm the central role of hybrid PV–wind systems with battery storage. Earlier reviews highlighted the diversity of hybrid renewable architectures and the difficulty of comparing configurations because of differences in objectives, metrics, and assumptions [89]. The present review shows that, despite this diversity, most recent studies still prioritize a limited set of architectures, particularly PV–wind–battery/BESS systems with or without diesel backup. Diesel generators remain relevant in isolated systems because they provide reliability support under renewable variability, although their inclusion conflicts with decarbonization objectives.

Compared with previous reviews, the present study also provides a more detailed characterization of the scale and application context of the analyzed systems. The reviewed literature is strongly concentrated on stand-alone, off-grid, rural, remote, and community-scale HRES, whereas grid-connected, industrial, commercial, hospital, university-campus, hydrogen-based, and electric-vehicle-integrated configurations remain comparatively underrepresented. This distinction is relevant because methodological requirements, reliability constraints, demand profiles, and component interactions differ substantially across application scales and operating contexts.

The geographical distribution also shows an uneven evidence base. Most studies were associated with a limited group of countries, particularly China, India, Iran, Egypt, and Iraq, while contributions from Latin America, Europe, Sub-Saharan Africa, and other regions were less frequent. This concentration may limit the transferability of the reported configurations and optimization results, since renewable resources, demand patterns, technology costs, regulatory conditions, and infrastructure constraints vary considerably across regions. Therefore, future studies should extend validation to a broader range of geographical and socioeconomic contexts.

4.2. Methodological Implications

A central finding is that metaheuristics remain the dominant primary sizing methods, whereas AI and learning-based approaches are less frequent and mainly perform supporting functions. Most PSO/MOPSO, GA/NSGA-II and alternative metaheuristic approaches were used as core sizing algorithms rather than auxiliary methods. This suggests that the field still relies heavily on flexible search-based approaches capable of handling discrete and continuous decision variables, nonlinear constraints, multiple objectives and technology combinations.

PSO/MOPSO-based methods appear to have become reference approaches for renewable microgrid sizing, especially in PV–wind–battery/BESS, PV–wind–diesel–battery, hydrogen-based and electric vehicle-integrated systems [22,25,26,32,33]. GA/NSGA-II-based methods represent a second consolidated family, particularly in multi-objective formulations involving cost, reliability, emissions and component selection trade-offs [23,26,39,40,48]. Other metaheuristics, such as GWO, DE, HS, FA, BAT, ABC, WOA and more recent algorithms, show that the literature continues to explore alternatives for convergence improvement, search diversification and benchmarking [24,58,59,60,61].

From a computational perspective, no algorithmic family can be considered consistently superior across the reviewed studies because execution times, stopping criteria, hardware platforms, population sizes, and test systems were not standardized. PSO-based methods are relatively simple to implement and often converge rapidly, which makes them suitable for continuous sizing variables and conventional PV–wind–battery configurations; however, they may be sensitive to parameter selection and premature convergence. GA- and NSGA-II-based methods offer greater flexibility for discrete decisions and multi-objective formulations, but they generally require larger populations and more function evaluations.

Alternative metaheuristics may improve diversification or convergence in specific cases, although their reported advantages are difficult to generalize because benchmark conditions are heterogeneous and independent replication is limited. Mathematical optimization is more appropriate when the problem can be represented through linear or mixed-integer formulations, whereas stochastic, robust, and hybrid AI–optimization approaches are better suited to uncertainty-intensive or high-dimensional configurations, at the cost of greater computational and data requirements. Algorithm suitability therefore depends on problem dimensionality, variable type, objective structure, uncertainty treatment, and technological complexity.

In contrast, artificial intelligence, machine learning, deep learning and reinforcement learning methods appeared less frequently as standalone sizing methods. Their main contribution was complementary: forecasting renewable resources or demand, supporting surrogate modeling, representing uncertainty, assisting energy management or improving hybrid optimization frameworks [29,36,76,77,78]. This indicates that the field is moving gradually from isolated optimization algorithms toward integrated frameworks that combine prediction, optimization and decision support.

Economic and reliability indicators were reported across many studies; however, their values are not directly comparable without harmonized definitions and assumptions. NPC, LCOE, COE, LPSP, LOLP, EENS, renewable fraction, and emissions depend on system boundaries, load profiles, temporal resolution, discount rates, component lifetimes, replacement assumptions, reliability thresholds, battery models, and renewable-resource datasets. Variations in these inputs can materially affect the calculated values. Cross-study comparisons should therefore be limited to cases with equivalent formulations and boundary conditions.

MATLAB was the most frequently reported environment because it supports customized implementation of metaheuristics, hybrid models, and component-level simulations. However, reproducibility is limited when source code, algorithm parameters, initialization procedures, random seeds, and stopping criteria are not reported. HOMER and HOMER Pro provide more standardized techno-economic simulation and facilitate benchmarking across system configurations, but their predefined structure can restrict the implementation of customized objective functions, constraints, and control strategies. Therefore, the choice of computational tool involves a trade-off between modeling flexibility and reproducibility.

4.3. Implications for Microgrid Design and Application Contexts

The application contexts identified in this review indicate that optimal sizing has been developed mainly for rural, remote, community, residential, off-grid and stand-alone systems. In these contexts, the main design challenge is not only minimizing cost but also ensuring supply continuity under variable renewable generation and uncertain demand. Therefore, the frequent use of battery storage, diesel backup and reliability metrics such as LPSP, LOLP, LOLE and unmet load is consistent with the operational needs of isolated and weak-grid systems.

For rural and community microgrids, the results reinforce the importance of local data quality. Sizing outcomes depend strongly on load profiles, solar irradiance, wind speed, component costs, battery parameters and reliability assumptions. For residential systems, the main implications are related to the representation of hourly demand, self-consumption, storage operation, tariff structures and battery degradation. For industrial or critical infrastructure, sizing methods require stricter reliability constraints, reserve capacity, continuity requirements and, ideally, real or hardware-in-the-loop validation.

Battery storage dominates the reviewed configurations because it is technically mature, widely available, and supported by established cost and performance data. Hydrogen-based storage and fuel-cell systems appear less frequently because their sizing requires higher capital investment, lower round-trip efficiency, more complex operational constraints, and additional component models for electrolyzers, hydrogen tanks, and fuel cells. Their assessment is also affected by limited validated data on degradation, replacement, efficiency variation, and long-term operation. In addition, the economic value of seasonal or long-duration hydrogen storage depends strongly on the study horizon, renewable-resource variability, demand profile, and backup requirements, which complicates direct comparison with battery-based systems.

Less frequent contexts, such as hospitals, university campuses, commercial buildings, thermal/CHP systems, water-related loads and electric vehicle-integrated microgrids, represent emerging areas. These applications require broader formulations that integrate electricity, heat, water, mobility, grid interaction and demand-side management. Therefore, future sizing frameworks should move beyond conventional PV–wind–battery configurations and address multi-energy and sector-coupled microgrids.

4.4. Research Gaps and Future Directions

Several gaps were identified. First, reproducibility remains limited. Many studies report the algorithm and final configuration, but provide incomplete information on parameter settings, stopping criteria, random seeds, input datasets, code availability or simulation assumptions. Future studies should report computational settings in a standardized way and provide traceable data and code whenever possible.

Second, real-world validation remains insufficient. Most studies rely on simulations, case studies or historical data, while experimental validation, pilot microgrids, hardware-the-loop testing and long-term monitoring are less common. This limits the assessment of sizing solutions under real operating conditions, including forecast errors, battery degradation, equipment failures, control constraints and load variability.

Practical deployment also depends on the computational and hardware requirements of the proposed sizing and management frameworks. High-dimensional optimization, detailed component models, and real-time adaptive methods may require substantial processing capacity, memory, reliable communication infrastructure, and accurate sensing. Deployment in embedded controllers or resource-constrained microgrids may therefore be limited by computational latency, data availability, communication failures, and interoperability with existing energy management and control systems. These constraints should be considered when assessing the scalability and operational feasibility of optimized configurations.

Third, uncertainty treatment should be strengthened. Although some studies used Monte Carlo simulation, sensitivity analysis, stochastic formulations or robust optimization, many sizing models still rely on deterministic load, weather, cost or tariff assumptions. Future research should incorporate probabilistic and robust frameworks that account for climate variability, demand uncertainty, component degradation, energy price changes and contingency scenarios.

Fourth, explainability is becoming increasingly important. As learning-based and hybrid AI methods enter the sizing process, decision transparency becomes relevant, especially for critical infrastructure, hospitals, industrial facilities and isolated communities. Explainable artificial intelligence could help identify which variables, constraints and operating assumptions most influence the selected configuration.

Fifth, scalability and transferability remain open challenges. Many studies validate methods in a single case study, which limits evidence on performance across climates, load profiles, regions, scales and regulatory conditions. Future work should include cross-context validation, benchmark cases, multi-microgrid applications, distributed optimization, and transferable models.

Finally, future research should explore digital twins and real-time adaptive optimization for renewable microgrid sizing and operation. Digital twins can support continuous model updating, scenario evaluation, predictive maintenance, and validation using real operating data. Real-time adaptive optimization can allow sizing and energy management decisions to respond dynamically to changing demand, renewable generation, component degradation, and market conditions. These developments should be integrated with uncertainty-aware and explainable models to improve transparency, robustness, and operational applicability.

4.5. Limitations of the Review

This review has limitations. The search was restricted to Scopus and the Web of Science Core Collection; therefore, relevant studies indexed in other databases, institutional repositories, conference proceedings or grey literature may not have been retrieved. The search strategy focused on terms related to optimal sizing, optimal design, capacity optimization, microgrids, hybrid renewable energy systems, PV, wind, artificial intelligence, and metaheuristics. As a result, studies using alternative terminology, such as expansion planning, resource allocation, or configuration planning, may have been excluded.

The review focused on research articles and excluded reviews, book chapters, duplicated documents, and studies centered only on operation, dispatch, control, or forecasting without explicit sizing or design. In addition, data extraction depended on the information reported in each article. When tools, datasets, assumptions, constraints, or methodological roles were unclear, conservative coding was applied.

A formal critical appraisal of methodological quality or risk of bias was not conducted, which is consistent with the purpose of a scoping review. Finally, some categories were not mutually exclusive. A single study could include multiple optimization methods, configurations, objectives, data sources, or application contexts. Therefore, some frequencies should be interpreted as coded occurrences rather than exclusive article counts.

5. Conclusions

This scoping review mapped 69 studies on the optimal sizing of renewable microgrids and hybrid renewable energy systems using artificial intelligence, learning-based methods, metaheuristics, and related optimization algorithms. The evidence shows a clear growth of the field after 2020, with most studies focused on PV–wind systems, battery storage and stand-alone, off-grid, rural, or community applications.

Metaheuristics were the dominant methodological family. PSO/MOPSO- and GA/NSGA-II-based approaches were the most frequently used methods, while alternative bioinspired algorithms were also widely reported. In contrast, artificial intelligence and learning-based methods appeared less frequently as primary sizing tools and were mainly used to support forecasting, uncertainty modeling, surrogate modeling, and energy management.

Technologically, the most recurrent configurations were PV/solar–wind–battery/BESS systems, with or without diesel backup. This confirms that the literature has prioritized autonomy, reliability and the management of renewable variability, especially in isolated or weak-grid contexts. The main objective functions and performance metrics were concentrated on economic and reliability criteria, although environmental, energy-related, technical and algorithmic indicators are increasingly incorporated in multi-objective formulations.

MATLAB and HOMER/HOMER Pro were the most reported computational tools, while the most common data inputs were battery/SOC parameters, economic costs, load profiles, grid data, environmental factors, and meteorological variables. These findings show that optimal sizing depends strongly on the quality of storage, demand, cost, and renewable resource data.

The study-level assessment identified substantial methodological and reporting gaps. Battery degradation was explicitly modeled in only 6 studies and partially represented in 38 through battery lifetime, replacement, depth-of-discharge, or cycle-life assumptions; 21 studies did not report degradation-related information, while 4 were classified as not applicable. Formal uncertainty scenarios were incorporated in 23 studies, and benchmark comparisons were reported in 52. Only 8 studies provided sufficient algorithm-parameter information, 20 reported explicit convergence or stopping criteria, and one study provided publicly accessible code or data.

These findings support the need to improve reproducibility through open code, traceable datasets, and standardized reporting of algorithm parameters. Future research should improve reproducibility through open code, traceable datasets and standardized reporting of algorithm parameters. Further work is also needed to strengthen real-world validation, incorporate uncertainty and degradation more systematically, improve explainability in AI-assisted models, and assess the scalability and transferability of sizing methods across regions, applications, and technological architectures.