A Review of Optimization Strategies for Energy Management in Microgrids

Abstract

Rapid industrialization, widespread transportation electrification, and significantly rising household energy consumption are rapidly increasing global electricity demand. Climate change and dependency on fossil fuels to meet this demand underscore the critical need for sustainable energy solutions. Microgrids (MGs) provide practical applications for renewable energy, reducing reliance on fossil fuels and mitigating ecological impacts. However, renewable energy poses reliability challenges due to its intermittency, primarily influenced by weather conditions. Additionally, fluctuations in fuel prices and the management of multiple devices contribute to the increasing complexity of MGs and the necessity to address a range of objectives. These factors make the optimization of Energy Management Strategies (EMSs) essential and necessary. This study contributes to the field by categorizing the main aspects of MGs and optimization EMS, analyzing the impacts of weather on MG performance, and evaluating their effectiveness in handling multi-objective optimization and data considerations. Furthermore, it examines the pros and cons of different methodologies, offering a thorough overview of current trends and recommendations. This study serves as a foundational resource for future research aimed at refining optimization EMS by identifying research gaps, thereby informing researchers, practitioners, and policymakers.

1. Introduction

The International Energy Agency highlights significant growth in electricity demand, indicating the onset of a new “Age of Electricity”, driven by electrification in buildings, transportation, and industry, alongside rising needs for air conditioning and data centers [1]. In 2024, demand increased by 4.3% and is projected to grow at nearly 4% annually, totaling an increase of 3500 TWh over three years, similar to Japan’s total electricity consumption. In the Canadian case, energy demand is expected to rise significantly, escalating from 624 TWh in 2020 to over 819 TWh in 2050 [2].

While energy demand is increasing, it is accompanied by growing concerns [1]. First, if not managed effectively, this increase could lead to higher emissions, particularly from fossil fuel sources. Second, while Renewable Energy Sources (RESs) like wind and solar help mitigate emissions, their variability can introduce instability in the power supply, complicating emission reduction efforts if not properly balanced. Third, extreme weather events pose risks to power system reliability and resource adequacy, potentially exacerbating existing vulnerabilities and increasing dependence on less environmentally friendly backup generation.

MGs have emerged as innovative frameworks for integrating RESs and Energy Storage Systems (ESSs), providing a solution for the high dependency on fossil fuels. According to [3], an MG is a network of interconnected loads and Distributed Energy Resources (DERs), defined by clear electrical boundaries, that functions as a single controllable unit. It has the capability to operate in both grid-connected and standalone modes, allowing it to connect to or disconnect from the main grid as needed.

EMSs are essential for the optimal power sharing of MGs with more than one energy resource and ESS [4]. EMSs refer to software that optimizes energy production, distribution, and consumption while ensuring cost efficiency, safety, and reliability [5]. Additionally, it addresses the variability of RES and other objectives such as mitigation of environmental impact.

In recent decades, substantial research has been conducted on various aspects of EMS in MG, with an emphasis on mid-term planning and scheduling [6,7,8,9,10,11,12,13,14,15,16]. Furthermore, some research combines sizing and scheduling, addressing both mid-term and long-term objectives [17,18]. Other studies have emphasized power references for converters in real-time environments, focusing on short-term optimization. These approaches converge on the common optimization challenge despite differences in their time frames. Optimization EMS problems generally involve defining objectives, identifying constraints, and selecting appropriate solution methods.

Most of these studies identify objectives such as minimizing fuel consumption and emissions from Conventional Distributed Energy Resources (CDERs), maximizing the utilization of RES, or extending the lifespan of ESSs. The conflicting nature of these objectives and others highlights the need for Multi-objective Optimization (MOO), which has been integrated into the problem formulation of EMS [19,20,21,22,23,24]. Additionally, system limitations must be managed, including constraints such as power balance, battery State of Charge (SOC), Depth of Discharge (DOD), and power generation limits. Methods such as deterministic approaches [16,25] and metaheuristic techniques [26] are commonly employed to address these optimization problems.

Due to the uncertainties faced by MGs, particularly the intermittency of RES, variability of demand and market price, optimization techniques such as machine learning, robust optimization, and stochastic methods have been utilized to manage these uncertainties effectively [27,28].

According to the literature review, there are multiple gaps optimization EMS of MGs: the optimization horizon, the influence of weather and climate conditions, the lack of data for validation, and conflicting objectives in optimization problems.

First, most studies focus on mid-term scheduling and planning, leaving long-term and short-term optimization strategies under exploration. This limitation affects the ability to develop comprehensive optimization frameworks for varying operational timescales.

Second, despite the high dependence of MG performance on weather and climate conditions, few studies adequately consider seasonal variations in validation. For example, based on Canadian Climate Normals from 1991 to 2020 [29], temperatures in Val d’Or, a city in Canada, can vary from approximately $-{23 }^{\circ }\mathrm{C}$ in January during Winter to about ${25 }^{\circ }\mathrm{C}$ in July during Summer. This gap is particularly relevant, as weather fluctuations can significantly impact energy generation and demand patterns. Moreover, while reliability is a crucial factor in MG operation due to climate consequences, it is often overlooked, even though extreme weather conditions can exacerbate system vulnerabilities and compromise stability.

Third, the lack of clear information on the data sources for validation remains a challenge, hindering robust model verification and limiting the standardization of the proposed methodologies.

Finally, while the complexity and trade-offs inherent in MOO are widely acknowledged, they are often not adequately addressed. Many approaches do not integrate technical, economic, and environmental considerations within a single-objective formulation, highlighting the need for comprehensive MOO frameworks.

Although existing review articles provide valuable information on specific aspects of EMS of MGs, such as strategies focused on islanded MGs [30], detailed classifications of EMS architectures [31], or the application of specific heuristic optimization techniques [26], this review differs in both purpose and scope. The purpose of this article is to systematically review and critically interpret optimization EMS of MGs so that it can point out current trends, gaps, and future directions. Rather than comparing specific algorithms or methods, this exhaustive review allowed us to identify and categorize key elements reported in the literature. Thus, the review is categorized along two main axes, as illustrated in Figure 1: (1) MG concepts, including operation, size, components, typologies, and the influence of weather and climate on performance; and (2) optimization EMS, focusing on problem formulations, time frames, optimization methods, and uncertainty. Regarding uncertainty, this includes both uncertainty sources, such as renewable generation variability, fuel price fluctuations, and load-demand changes, along with approaches to handle these uncertainties, including stochastic modeling, robust optimization, and machine learning. This approach allows the article to highlight gaps in the literature, such as the limited exploration of climate impacts, the insufficient integration of Multi-Objective Optimization (MOO) strategies, and the lack of clear information regarding data sources for validation.

Figure 1. Review framework.

This paper aims to bridge these gaps by providing a comprehensive and critical review of the latest articles on EMS techniques in MGs.

The main contributions of this work can be summarized as follows:

• Explores the main aspects of MGs that incorporate RESs, such as sizes, operational components, types of MGs, and implications of weather and climate on performance. This includes an analysis of how weather can influence energy generation and demand, thereby affecting system performance.
• Provides an up-to-date review of published research on the fundamentals of optimization EMS of MGs from 2020 to 2025. Main aspects such as problem formulation, time frames, sources of uncertainty, techniques for managing uncertainties, and various optimization methods are thoroughly discussed. It also addresses MOO, detailing the main objectives and constraints based on MG components.
• Identifies contributions and limitations of existing research, along with recommendations for future work. This section highlights areas that require further exploration or methodological refinement, ensuring that new studies can build on the identified gaps and deficiencies in the current literature. This comprehensive examination not only informs current practices but also sets the stage for innovative developments in EMS.

The remainder of this literature review is structured as follows: Section 2 will detail the research methodology employed in identifying relevant literature, while Section 3 will discuss MG concepts and components. Subsequently, Section 4 will analyze optimization EMS and their implications for system performance. Section 5 will present a discussion of future trends, along with the contributions and limitations of the EMSs. Finally, Section 6 will conclude the paper by summarizing the main findings.

2. Literature Search Methodology

This literature search aimed to obtain publications that study EMSs using MG optimization methods, following the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) [32,33,34]. Scopus was selected due to its extensive coverage of peer-reviewed engineering and energy management journals, ensuring a comprehensive dataset. The review process consisted of four main stages, as illustrated in Figure 2 and described below:

Figure 2. Schematic representation of search.

Identification: a comprehensive bibliographic search was conducted across multiple publishers using IEEE Thesaurus to refine search terms [35]. The search equation was formulated as: “energy” AND “management” AND “system” AND “optimization methods” AND (“microgrids” OR “nanogrids”) AND “renewable energy sources”. The initial search on Scopus returned 31,647 related documents. This review is limited to the last five years (2020–2025) to emphasize recent advancements, particularly in optimization EMS, which has rapidly evolved with AI-based methods. However, when fundamental concepts, definitions, or background information were necessary, a small number of older studies, relevant books, and official government websites were considered in the Eligibility stage. This initial filtering step reduced the dataset to approximately 320 articles.

Screening: to refine the dataset, a filtering process was applied based on relevance to energy, engineering, and computer science. This refinement reduced the dataset to 306 articles. Next, manual screening was performed to ensure the studies explicitly focused on optimization EMS. Studies relying solely on rule-based EMS control were excluded, as they do not contribute to optimization-based advancements. Studies that lacked quantitative performance analysis of EMSs were also excluded. In total, 179 articles were considered out of scope and excluded from further review, leaving about 190 studies.

Eligibility: to ensure high relevance, a full-text review was conducted. Articles were excluded if they did not integrate RESs into MGs and instead focused primarily on CDERs. Moreover, studies were excluded if the optimization approach was not applied to EMSs. Following this review, about 190 articles met the eligibility criteria. Additional information on specific topics and definitions was drawn from eligible and cited articles, relevant books, and official government websites, as previously mentioned.

Inclusion: A final selection of high-quality, peer-reviewed journal articles was made, prioritizing the most recent and relevant work, which involved sorting through Scopus. In addition, information from official websites and books resulted in approximately 80 citations.

Having established the methodology for selecting the literature, we now focus on the foundational concepts of MGs, the systems for which the optimization EMS is developed.

3. Microgrids Concepts and Implications

This section outlines some aspects of MGs incorporating RESs, covering their operation, sizing considerations, component types, classifications of MGs, and the impact of weather and climate conditions on RES performance. Understanding these factors is crucial for identifying the most common MG classifications found in the literature and recognizing the significance of weather and climate impacts on MGs.

Operation: MGs can operate independently (standalone mode) or in connection with the utility grid (grid-connected mode) [5,24].

Size: MGs can vary significantly in size. They can be large and intricate networks with capacities of up to tens of MW and can operate in low-voltage and medium-voltage ranges, typically from 400 V to 69 kV. In contrast, smaller MGs may have hundreds of kW capacities, catering to only a few consumers [36]. This work considers small-scale with less than 10 kW capacity, medium-scale ranging from 10 kW to 1 MW, and large-scale with greater than 1 MW capacity.

Components: based on [30], the main components of an MG include DERs responsible for power production and energy storage, consisting of both conventional and non-conventional sources. CDERs, such as DGs and Micro Turbines (MTs), provide traditional power generation. Non-Conventional Distributed Energy Resources (NCDERs) refer to RESs, including PV systems and WP. ESSs play a critical role in MGs by storing electrical energy for backup power during periods of low production [26], with examples such as batteries, fuel cells, and supercapacitors. Loads are classified into controllable loads, such as air conditioners and heat pumps, and uncontrollable loads, such as industrial loads. Additionally, MGs incorporate power conversion devices, including converters, inverters, and other power electronic components [24], which ensure proper energy management and distribution.

Types of MGs: three main types of MGs are described: AC, DC, and Hybrid AC-DC MGs [24]. The types of MGs are illustrated in Figure 3. Here is a brief description of each type:

• AC MGs: power electronic devices connect DGs and ESSs to an AC bus. Depending on the on/off control at the Point of Common Connection (PCC), they can operate in grid-connected or standalone modes. AC MGs facilitate the integration of various RESs, supporting the existing AC power infrastructure.
• DC MGs: DC MGs supply AC and DC loads at different voltage levels via power electronic converters. They feature DERs connected to the DC bus, with ESSs linked to the DC network. DC MGs are more cost-effective when there are multiple DC power sources and loads, such as solar panels and batteries.
• Hybrid AC-DC MGs: these systems combine AC and DC buses, allowing them to serve both AC and DC loads. The DC grid connects to the AC bus through an inverter as a versatile power source. Hybrid MGs leverage the advantages of both AC and DC systems, effectively optimizing the feeding of various loads.

Figure 3. Schematic representation of Grid-Connected MGs.

Figure 3. Schematic representation of Grid-Connected MGs.

Weather and Climate Implications: RES supply depends on the weather, making considering local conditions and season impacts essential. In [37], it is noted that PV system efficiency decreases by 0.4–0.5% for each degree of temperature rise. Only about 20% of incident solar energy is converted into electricity, while the rest is lost as heat, meaning solar panels perform best in cold and low-temperature conditions. Additionally, solar radiation affected by cloud cover, air temperature, humidity, and wind speed directly impacts PV output [38].

Wind energy depends on weather, as turbines work best at certain wind speeds. In [39], a study on seasonal wind cycles in India found that when rainfall decreases, cooling demand rises, but wind power drops due to weaker winds. In contrast, stronger winds help generate more wind power during stormy periods with heavy rainfall.

Climate change further affects renewable energy. A study in [40] found that bioenergy availability may increase due to CO₂ fertilization, while hydropower and wind energy impacts vary by region. Solar power remains primarily stable. The shift to renewables is expected to reduce fossil fuel and nuclear reliance slightly, cutting CO₂ emissions by 1–2% by 2100, though climate-related risks to future energy systems are considered moderate.

Additionally, in [38] is made a study about the future of solar energy potential in a changing climate across the world where it is shown that seasonal changes further impact solar energy potential, with notable declines in boreal autumn and winter, particularly in India, China, and North America, while Europe remains stable. In boreal summer and austral autumn, Europe, South America, and Australia may see some PV gains.

The Dunkelflaute incidents in Northern Europe from 5 to 7 November and 11 to 12 December 2024 highlight the significant impact of seasonal weather on renewable energy generation. During these events, a substantial drop in combined wind and solar output strained electricity systems, particularly in Germany, resulting in tight supply conditions and sharp fluctuations in wholesale electricity prices [1].

Energy needs also vary with weather conditions. Energy demand peaks in winter due to heating and lighting requirements [41]. For instance, electricity is used for space heating and preventing water pipes from freezing in Arctic and sub-Arctic regions such as Alaska, Canada, Greenland, and Norway.

Thus, MG design and operation require careful consideration of climatic and weather conditions, which vary significantly across regions and impact EMS. Climate zone classifications are derived from the Köppen–Geiger climate classification system [42]. The five main groups are A (tropical), B (dry), C (temperate), D (continental), and E (polar). The temperature ranges from 1991 to 2023 are obtained from the Climate Change Knowledge Portal (CCKP) [43].

Tropical: Defined by consistently high temperatures and humidity throughout the year. MG impacts include reduced photovoltaic (PV) efficiency and increased cooling demands. Country and EMS examples include Nigeria [44], Colombia [45], Malaysia [18], and Egypt [46]. Example of temperature range: Colombia has an average minimum and maximum surface air temperature of ${20.22 }^{\circ }$C and ${30.6 }^{\circ }$C, respectively [42].

Dry: Characterized by low precipitation and high solar irradiance. MG impacts include limited water availability for PV cooling and dust accumulation, which leads to PV degradation. Country and EMS examples include Iran [47], Somalia [48], and Saudi Arabia [49]. Example of temperature range: Somalia has an average minimum and maximum surface air temperature of ${19.45 }^{\circ }$C and ${33.67 }^{\circ }$C, respectively [42].

Temperate: Typically experiences warm, humid summers and mild, wet winters. MG impacts include seasonal variability in solar generation and fluctuating heating and cooling needs. Country and EMS examples include Spain [50], Denmark [20], France [51], China [22,52], and Canada (Vancouver) [53]. Example of temperature range: France has an average minimum and maximum surface air temperature of ${7.32 }^{\circ }$C and ${25.23 }^{\circ }$C, respectively [42].

Continental: Characterized by large temperature variations between hot summers and cold winters. MG impacts include high heating demand during winter and high cooling demand in summer. An example location is Quebec, Canada [54]. Example of temperature range: Quebec has an average minimum and maximum surface air temperature of $-{25.19 }^{\circ }$C and ${18.45 }^{\circ }$C, respectively [42].

Polar: Defined by extremely cold temperatures and limited solar irradiance, particularly during winter. MG impacts include very high heating demands and challenges with battery performance and insulation. Country examples include Arctic and sub-Arctic regions such as Alaska, Northern Canada, Greenland, and Norway [41]. Example of temperature range: Greenland has an average minimum and maximum surface air temperature of $-{10 }^{\circ }$C and ${0.27 }^{\circ }$C, respectively [42].

With the presentation of MG concepts and their implications, the focus can now shift to the optimization of EMSs.

4. Optimization EMS

This section examines Optimization EMS in MGs, focusing on problem formulation, time frames, uncertainty, and optimization methods.

Before going into the main optimization aspects, Table 1 provides a comprehensive summary of various optimization EMS for MGs. As illustrated in Figure 4, there is a connection between the review classification for Table 1 and the EMS optimization process. The bottom part of Figure 4 represents a typical EMS optimization process, starting from data acquisition and ending with validation and decision-making, reflecting the transition from system state to actions or optimization policy. The top side summarizes the classification framework used in Table 1. The main elements are linked across both levels: MG information aligns with data acquisition, problem definition corresponds to problem formulation and time frame, optimization to optimization selection, and performance outcomes related to validation and decision-making. The curved arrow highlights this conceptual alignment.

Figure 4. Integrated presentation of EMS optimization process (bottom side) and the classification used for Table 1 (top side).

Thus, Table 1 categorizes studies based on MG components, operation type (grid-connected or standalone), system type (AC, DC, or hybrid), size, location, and weather conditions. Table 1 outlines the optimization objectives pursued in each study and the methodologies applied. It also identifies whether the studies incorporate MOO, validate results under weather variations, or model uncertainty. Furthermore, it provides insights into the EMS operational time frame (short-term, medium-term, or long-term). Additionally, the highlights column includes findings, such as cost reductions, emission minimization, and power stability improvements, which are summarized in Figure 5. These benefits are also quantified as a percentage of occurrence across all reviewed articles of Table 1. Notably, 50% of the studies address operational costs, 29% focus on energy cost reduction, and 18% target either fuel consumption or emission reductions. Other reported improvements include power stability (11%), ESS utilization (11%), reliability (11%), and power loss mitigation (11%). Less frequent but still relevant objectives involve demand response (7%), degradation reduction (7%), and economic dispatch (4%). Finally, only [44] validates their strategy under seasonal variations from the articles of this table.

Figure 5. Optimization results of MG EMS.

Figure 5 suggests that economic performance remains the dominant driver in EMS optimization and results, especially operational and energy cost minimization. The relatively low occurrence of aspects such as emissions reduction, battery degradation mitigation, reliability improvement, and seasonal adaptability does not mean these aspects are unimportant. Rather, integrating them into EMS design involves significant modeling and computational challenges. For example, modeling battery degradation requires detailed lifetime and aging models [55], while emissions calculations depend on accurate emission factors and clear policy information from governments [56]. Including these factors as additional optimization objectives increases the complexity of the problem. Reliability metrics often require probabilistic methods, such as the probability of loss of power supply [20], or failure simulations, such as the Monte Carlo method, which is used for reliability evaluation in [57]. Seasonal variability affects the performance of the MG forecasting model [58] but depends on climate data, which are not always available. These aspects remain an opportunity for work within the research literature because the increasing integration of renewable sources, the variability of climate change, and the demand for more robust systems make these aspects relevant. Addressing them could improve the adaptability, longevity, and environmental performance of MG systems.

Table 1. Optimization EMS.

Ref. Year	(MG) Information	Objective	Multi-Objective	Optimizer	Optimization Method	Timeframe	Uncertainty	Highlights
[44] (2020)	Photovoltaic (PV), Wind Turbine (WT), Battery Energy Storage System (BESS), Diesel Generator (DG); Standalone; Small Scale; Hybrid AC/DC; Nigeria, Yobe State; Tropical	Minimizing Cost of Energy (COE) and Deficiency of Power Supply Probability (DPSP)	Yes	Grasshopper Optimization Algorithm (GOA)	Metaheuristic	Long-Term	No	GOA—Reduced fuel consumption by 92.4%, Carbon Dioxide (CO2) emissions by 92.3%, and COE by 79.8%. Compared against Hybrid Optimization of Multiple Energy Resources (HOMER) software and traditional EMSs.
[50] (2021)	PV, WT, Flat-Plate Collector (FPC), BESS, Electric Water Heater (EWH); Grid-Connected; Medium Scale; Hybrid AC/DC; Spain, Public University of Navarre (UPNa); Temperate	Minimizing positive grid power peaks and negative grid power peaks, as well as Maximum Power Derivative (MPD), average power derivative (APD), and power profile variability (PPV)	No	Fuzzy Logic Control (FLC)	Heuristic	Mid-Term	Yes	FLC—Improved grid power stability, enhanced battery State of Charge (SOC) stability, and reduced power fluctuations. Compared against heuristic optimization.
[45] (2021)	PV, DG, BESS; Standalone; Large Scale; AC; Colombia; Tropical	Priority EMS reduces fuel consumption using predefined rules, while Optimal EMS optimizes total marginal cost for efficient Energy Management (EM).	No	Priority EMS (Rule-Based); Optimal EMS (Mixed-Integer Linear Programming (MILP)-Based)	Priority EMS: Heuristic; Optimal EMS: Deterministic	Mid-Term	No	Priority EMS vs. Optimal EMS—Priority EMS ensures simplicity and robustness, while Optimal EMS (MILP-based) improves BESS utilization and reduces diesel consumption.
[47] (2021)	PV, WT, BESS, Supercapacitor (SC); Both Grid-Connected and Standalone; Small Scale; DC; Iran, Shiraz; Arid	Particle Swarm Optimization (PSO) lowers maintenance costs and prevents battery life reduction, while Fuzzy Logic Control (FLC) minimizes peak current of the battery.	No	FLC and PSO	Metaheuristic	Short-Term	No	FLC + PSO—Improved stability against grid fluctuations in islanded and grid-connected modes.
[48] (2021)	PV, DG, BESS; Standalone; Medium Scale; Hybrid AC/DC; Somalia, Garowe; Arid	Minimizing operating costs by reducing DG power consumption.	No	Tunable Rule-Based Heuristic (TRBH) optimized using Evolutionary Algorithms	Heuristic	Mid-Term	No	TRBH + Evolutionary Algorithms—Achieved near-optimal performance while maintaining simplicity. Compared against MILP, Evolutionary Algorithms (SNO), and Rule-Based EMS.
[51] (2025)	PV, BESS; Grid-Connected; Small Scale; DC; France, École Polytechnique (Palaiseau); Temperate	Maintaining a specific SOC in the battery while minimizing electricity costs.	No	Scenario-Based Stochastic Model Predictive Control (MPC)	Stochastic	Short-Term and Mid-Term	Yes	Scenario-Based Stochastic MPC—Reduced operational cost by 6% and 22%. Compared against Chance-Constrained MPC, Rule-Based EMS, Single-Layer Scenario-Based MPC, and Two-Stage Stochastic Programming.
[59] (2024)	PV, WT, Fuel Cell (FC), Microturbine (MT), DG, BESS; Both Grid-Connected and Standalone; Medium Scale; AC; N/A; N/A	Minimizing main-grid energy costs, fuel and generator startup costs, renewable energy generation costs, greenhouse gas emissions costs, demand response incentive costs, and power loss costs.	Yes	Golden Jackal Optimization (GJO)	Metaheuristic	Mid-Term	No	GJO—Achieved 96% efficiency, significantly lowered operational costs. Compared against PSO, Artificial Bee Colony (ABC), and Tabu Search (TS).
[20] (2023)	DG, PV, WT, BESS; Standalone; Medium Scale; Hybrid AC/DC; Denmark, Sønderborg; Temperate	Optimizing system sizing while minimizing the Loss of Power Supply Probability (LPSP) and the Levelized Cost of Energy (LCOE).	Yes	multi-objective Moth Flame Optimization (MOMFO)	Metaheuristic	Long-Term	No	MOMFO—Improved reliability, increased renewable energy penetration, and reduced costs. Compared against NSGA-II, MOPSO, and MOSEO.
[21] (2022)	MT, WT, BESS, FC; Grid-Connected; Large Scale; N/A; N/A	Minimizing operational costs and emissions.	Yes	Oppositional Gradient-Based Grey Wolf Optimizer (OGGWO)	Metaheuristic	Mid-Term	Yes	OGGWO—Achieved efficient cost and emission reduction. Compared against NSGA-II, PSO, and Grey Wolf Optimizer (GWO).
[22] (2025)	WT, PV, BESS; Standalone; Medium Scale; N/A; China, Tianjin; Temperate	Minimizing the Tracking Assessment Cost of the Wind-PV Energy Storage System (ESS) and reduce Energy Storage charging and discharging costs.	Yes	State-Action-Reward-State-Action (SARSA) Algorithm	Machine Learning-Supported	Mid-Term	Yes	SARSA—Improved power tracking, enhanced storage efficiency, and reduced operational costs. Compared against Model Predictive Control (MPC) and Deep Q-Network (DQN) methods.
[23] (2022)	PV, WT, DG, MT, BESS; Standalone; Medium Scale; Hybrid AC/DC; N/A; N/A	Minimizing the optimization function by reducing generation, emissions, and maintenance costs.	Yes	Mixed-Integer Programming (MIP)	Deterministic	Mid-Term	No	MIP—Reduced operation costs, improved BESS utilization, and integrated demand response. Compared against Genetic Algorithm (GA).
[24] (2023)	PV, WT, DG, MT, BESS; Both Grid-Connected and Standalone; Medium Scale MG; All Types; Iran; Temperate	Minimizing operating costs and emissions.	Yes	Salp Swarm Algorithm (SSA)	Metaheuristic	Mid-Term	Yes	SSA—Achieved cost and emission reductions, ensuring feasible planning under uncertainty. The comparison focused on effectiveness in grid-connected and islanded scenarios.
[60] (2021)	PV, WT, BESS, DG; Grid-Connected; Medium Scale; Hybrid AC/DC; N/A; N/A	Minimizing fuel consumption, operation and maintenance costs, emissions, ES operation costs, and costs related to DG and grid interactions.	No	Column and Constraints Generation (C&CG)	Deterministic	Mid-Term	Yes	C&CG—Enhanced scheduling under Renewable Energy Source (RES) uncertainty, reducing operational costs. Validated through simulations on the IEEE-33 bus distribution network model.
[52] (2023)	Controllable Distributed Generators (CDG), PV; Grid-Connected; Large Scale; N/A; China, Nanjing; Temperate	Minimizing user load dispatch and emission dispatch.	Yes	Deep Reinforcement Learning (DRL)	Machine Learning-Supported	Mid-Term	Yes	DRL—Improved solving speed by 56.74% over NSGA-II and 15.62% over CVX, effectively managing renewable energy uncertainties. Compared against heuristic optimization.
[49] (2023)	PV, WT, BESS; Grid-Connected; Large Scale; N/A; Saudi Arabia, Wadi Alddawasir; Arid	Two-stage optimization to minimize operational costs and enhance self-consumption rates.	Yes	Harris Hawks Optimizer (HHO)	Metaheuristic	Mid-Term	Yes	HHO—Increased self-consumption (100% SCR), improved hosting capacity (8.863 MW to 10.213 MW), and reduced daily operating costs by 23.8%. Compared against standard optimization algorithms.
[53] (2023)	PV, BESS; Grid-Connected; Small Size; DC; Canada, Vancouver; Temperate	Maximizing power supply availability, minimizing Net Present Cost (NPC), and maximizing power efficiency.	No	GA	Metaheuristic Algorithm	Long-Term	Yes	GA—Achieved 93.76% energy efficiency, reduced costs by 14%, and provided 100 times lower downtime. Compared against HOMER software for MG design optimization.
[57] (2023)	DG, BESS, WT Farm; Standalone; Medium Scale; N/A; N/A	Minimizing total costs by considering unit operating costs, diesel generator emissions, ES degradation, and adjustment costs from wind power forecasting errors.	No	Mixed-Integer Second-Order Cone Programming (SOCP)	Deterministic	Mid-Term	Yes	SOCP—Improved reliability (99%) while balancing cost reductions. Compared against Deterministic and Stochastic Programming-Based models.
[61] (2022)	PV, WT Farm, DG, BESS; Grid-Connected; Medium Scale; N/A; Singapore; Tropical	Minimizing total economic energy costs.	No	Proximal Policy Optimization (PPO)	Machine Learning-Supported	Mid-Term	Yes	PPO—Achieved 32.8% cost savings and 28.5% carbon emission reduction. Compared to conventional grid-based power reliance.
[62] (2023)	RES, Combined Heat and Power (CHP), ESSs, Heat Energy Storage (HES); Both Grid-Connected and Standalone; Medium Scale; AC; N/A; N/A	Minimizing Cost of Energy Use (CET).	No	Myopic Optimization with Linear Programming (LP) and Reinforcement Learning (RL)	Hybrid Deterministic and Machine Learning-Supported	Long-Term and Short-Term	No	LP + RL—Improved operating cost by 90.98% compared to myopic model alone.
[63] (2023)	Exhaust Air WT, Concentrated Photovoltaic Thermal (CPVT), BESS, HES, DG; Grid-Connected; Medium Scale; AC; Iran; N/A	Minimizing the total CET and BESS degradation.	No	State-Action-Reward-State-Action (SARSA)-based Composite Differential Evolution (DE) Algorithm	Hybrid Machine Learning-Supported and Metaheuristic	Mid-Term and Short-Term	Yes	SARSA-Based Composite DE—Reduced energy and battery degradation costs, optimized renewable resource usage. Compared against Q-Learning and various DE-based methods.
[64] (2023)	MT, PV, WT, BESS; Both Grid-Connected and Standalone; Large Scale; AC; N/A; N/A	Minimizing operating costs and imbalance costs by penalizing power shortages and surpluses caused by renewable generation variability.	No	Data-Driven Robust Optimization Approach	Robust Optimization (RO)	Mid-Term and Short-Term	Yes	Data-Driven Robust Optimization—Reduced operational costs by 28% (MG1), 21% (MG2), and 19% (MG3). Compared against Traditional polyhedral set-based RO models.
[65] (2024)	BESS, FC, Natural Gas Storage (NGS), CHP Units; Both Grid-Connected and Standalone; Small Scale; AC; Iran (Zabol, Tabriz, Urmia); Continental	Minimizing the operational cost, including generation and reserve costs of MTs, ESS operation costs, and total trading payments. A two-stage RO model ensures feasibility against uncertainties by optimizing day-ahead scheduling and real-time adjustments, minimizing imbalance costs from renewable generation variations.	No	RO Techniques	RO	Mid-Term and Short-Term	Yes	RO—Achieved 67.54% reduction in load decrease, improved network performance from 15.48% to 69.85%, and enhanced recovery speed (61.59%).
[66] (2024)	PV, WT, BESS; Grid-Connected; Medium Scale; DC; India, Bhubaneswar; Tropical	Minimizing total operational costs, including EM of PV, wind, and batteries, maintenance expenses, and penalties for maintaining battery SOC within 20%-80%, while maximizing power availability to meet active and reactive power demands.	Yes	Modified MPC with Firefly Algorithm (FA)	Hybrid Model-Based and Metaheuristic Optimization	Short-Term	Yes	MPC + FA—Improved voltage stability, optimized active/reactive power, and reduced operational costs. Compared against Proportional-Integral PSO, Active Disturbance Rejection Control, and standard MPC variants.
[18] (2024)	PV, WT, BESS; Both Grid-Connected and Standalone; Small Scale; Hybrid AC/DC; Malaysia, Johor Bahru; Tropical	Maximizing reliability, minimize total costs, and reduce unutilized surplus power.	Yes	Iterative Pareto-Fuzzy (IPF) Method	Fuzzy Logic-Based	Mid-Term	Yes	IPF—Achieved a total cost of $5.12M, minimized surplus power to 201,364.45 kW, and ensured high reliability (>90% Index of Independence, >77% Reliability Test Efficiency). Compared against NSGA-II, Modified PSO, and GWO.
[46] (2024)	PV Plant, WT Farm, DG, BESS; Grid-Connected; Large Scale; DC; Egypt, Greater Cairo; Tropical	Minimizing operational costs by reducing total daily expenses, including grid electricity purchases, energy sales, battery operations, and diesel generator costs, while also minimizing deviations from target load profiles.	No	Honey Badger Algorithm (HBA)	Metaheuristic	Mid-Term	Yes	HBA—Achieving 4.56% cost reduction, and 2.3% savings in real-time operations under uncertainties. Comparison focused on real-time performance vs. theoretical optimal cost under perfect foresight.
[67] (2024)	WT, PV, BESS, Plug-in (EVs); Both Grid-Connected and Standalone; Medium Scale; N/A; N/A	Minimizing energy loss costs over 24 h, power purchase costs from the upstream grid, load curtailment costs, diesel DG operation costs, and battery and EV operational expenses.	No	Column and Constraints Generation (C&CG)	RO	Mid-Term	Yes	C&CG—Minimized total costs (86,978 vs. 93,936 for GA and 91,326 for PSO), reduced energy purchases and PV curtailment, and optimized peak charge. Compared against GA and PSO.
[68] (2024)	BESS, WT, PV; Grid-Connected; Small Scale; N/A; N/A	Minimizing the cost of energy trading for MGs, including buying, selling, and transportation costs.	No	Stochastic Dual Dynamic Programming (SDDP)	Stochastic	Mid-Term	Yes	SDDP—Improved MG EM efficiency, achieving cost reductions and optimizing decisions under uncertainty. Compared against Deterministic Equivalent Model (DEM).
[69] (2025)	DG, FC, WT, PV; Grid-Connected; N/A; Hybrid Ac/DC; Northwestern China; Dry	Minimization of economic dispatch cost.	No	Large-Scale Solvers like CONOPT and GUROBI	Deterministic	Mid-Term	No	Model of MGs to achieve a fully convex. Computational time reduction using the model III.

Table 1. Optimization EMS.

Ref. Year	(MG) Information	Objective	Multi-Objective	Optimizer	Optimization Method	Timeframe	Uncertainty	Highlights
[44] (2020)	Photovoltaic (PV), Wind Turbine (WT), Battery Energy Storage System (BESS), Diesel Generator (DG); Standalone; Small Scale; Hybrid AC/DC; Nigeria, Yobe State; Tropical	Minimizing Cost of Energy (COE) and Deficiency of Power Supply Probability (DPSP)	Yes	Grasshopper Optimization Algorithm (GOA)	Metaheuristic	Long-Term	No	GOA—Reduced fuel consumption by 92.4%, Carbon Dioxide (CO2) emissions by 92.3%, and COE by 79.8%. Compared against Hybrid Optimization of Multiple Energy Resources (HOMER) software and traditional EMSs.
[50] (2021)	PV, WT, Flat-Plate Collector (FPC), BESS, Electric Water Heater (EWH); Grid-Connected; Medium Scale; Hybrid AC/DC; Spain, Public University of Navarre (UPNa); Temperate	Minimizing positive grid power peaks and negative grid power peaks, as well as Maximum Power Derivative (MPD), average power derivative (APD), and power profile variability (PPV)	No	Fuzzy Logic Control (FLC)	Heuristic	Mid-Term	Yes	FLC—Improved grid power stability, enhanced battery State of Charge (SOC) stability, and reduced power fluctuations. Compared against heuristic optimization.
[45] (2021)	PV, DG, BESS; Standalone; Large Scale; AC; Colombia; Tropical	Priority EMS reduces fuel consumption using predefined rules, while Optimal EMS optimizes total marginal cost for efficient Energy Management (EM).	No	Priority EMS (Rule-Based); Optimal EMS (Mixed-Integer Linear Programming (MILP)-Based)	Priority EMS: Heuristic; Optimal EMS: Deterministic	Mid-Term	No	Priority EMS vs. Optimal EMS—Priority EMS ensures simplicity and robustness, while Optimal EMS (MILP-based) improves BESS utilization and reduces diesel consumption.
[47] (2021)	PV, WT, BESS, Supercapacitor (SC); Both Grid-Connected and Standalone; Small Scale; DC; Iran, Shiraz; Arid	Particle Swarm Optimization (PSO) lowers maintenance costs and prevents battery life reduction, while Fuzzy Logic Control (FLC) minimizes peak current of the battery.	No	FLC and PSO	Metaheuristic	Short-Term	No	FLC + PSO—Improved stability against grid fluctuations in islanded and grid-connected modes.
[48] (2021)	PV, DG, BESS; Standalone; Medium Scale; Hybrid AC/DC; Somalia, Garowe; Arid	Minimizing operating costs by reducing DG power consumption.	No	Tunable Rule-Based Heuristic (TRBH) optimized using Evolutionary Algorithms	Heuristic	Mid-Term	No	TRBH + Evolutionary Algorithms—Achieved near-optimal performance while maintaining simplicity. Compared against MILP, Evolutionary Algorithms (SNO), and Rule-Based EMS.
[51] (2025)	PV, BESS; Grid-Connected; Small Scale; DC; France, École Polytechnique (Palaiseau); Temperate	Maintaining a specific SOC in the battery while minimizing electricity costs.	No	Scenario-Based Stochastic Model Predictive Control (MPC)	Stochastic	Short-Term and Mid-Term	Yes	Scenario-Based Stochastic MPC—Reduced operational cost by 6% and 22%. Compared against Chance-Constrained MPC, Rule-Based EMS, Single-Layer Scenario-Based MPC, and Two-Stage Stochastic Programming.
[59] (2024)	PV, WT, Fuel Cell (FC), Microturbine (MT), DG, BESS; Both Grid-Connected and Standalone; Medium Scale; AC; N/A; N/A	Minimizing main-grid energy costs, fuel and generator startup costs, renewable energy generation costs, greenhouse gas emissions costs, demand response incentive costs, and power loss costs.	Yes	Golden Jackal Optimization (GJO)	Metaheuristic	Mid-Term	No	GJO—Achieved 96% efficiency, significantly lowered operational costs. Compared against PSO, Artificial Bee Colony (ABC), and Tabu Search (TS).
[20] (2023)	DG, PV, WT, BESS; Standalone; Medium Scale; Hybrid AC/DC; Denmark, Sønderborg; Temperate	Optimizing system sizing while minimizing the Loss of Power Supply Probability (LPSP) and the Levelized Cost of Energy (LCOE).	Yes	multi-objective Moth Flame Optimization (MOMFO)	Metaheuristic	Long-Term	No	MOMFO—Improved reliability, increased renewable energy penetration, and reduced costs. Compared against NSGA-II, MOPSO, and MOSEO.
[21] (2022)	MT, WT, BESS, FC; Grid-Connected; Large Scale; N/A; N/A	Minimizing operational costs and emissions.	Yes	Oppositional Gradient-Based Grey Wolf Optimizer (OGGWO)	Metaheuristic	Mid-Term	Yes	OGGWO—Achieved efficient cost and emission reduction. Compared against NSGA-II, PSO, and Grey Wolf Optimizer (GWO).
[22] (2025)	WT, PV, BESS; Standalone; Medium Scale; N/A; China, Tianjin; Temperate	Minimizing the Tracking Assessment Cost of the Wind-PV Energy Storage System (ESS) and reduce Energy Storage charging and discharging costs.	Yes	State-Action-Reward-State-Action (SARSA) Algorithm	Machine Learning-Supported	Mid-Term	Yes	SARSA—Improved power tracking, enhanced storage efficiency, and reduced operational costs. Compared against Model Predictive Control (MPC) and Deep Q-Network (DQN) methods.
[23] (2022)	PV, WT, DG, MT, BESS; Standalone; Medium Scale; Hybrid AC/DC; N/A; N/A	Minimizing the optimization function by reducing generation, emissions, and maintenance costs.	Yes	Mixed-Integer Programming (MIP)	Deterministic	Mid-Term	No	MIP—Reduced operation costs, improved BESS utilization, and integrated demand response. Compared against Genetic Algorithm (GA).
[24] (2023)	PV, WT, DG, MT, BESS; Both Grid-Connected and Standalone; Medium Scale MG; All Types; Iran; Temperate	Minimizing operating costs and emissions.	Yes	Salp Swarm Algorithm (SSA)	Metaheuristic	Mid-Term	Yes	SSA—Achieved cost and emission reductions, ensuring feasible planning under uncertainty. The comparison focused on effectiveness in grid-connected and islanded scenarios.
[60] (2021)	PV, WT, BESS, DG; Grid-Connected; Medium Scale; Hybrid AC/DC; N/A; N/A	Minimizing fuel consumption, operation and maintenance costs, emissions, ES operation costs, and costs related to DG and grid interactions.	No	Column and Constraints Generation (C&CG)	Deterministic	Mid-Term	Yes	C&CG—Enhanced scheduling under Renewable Energy Source (RES) uncertainty, reducing operational costs. Validated through simulations on the IEEE-33 bus distribution network model.
[52] (2023)	Controllable Distributed Generators (CDG), PV; Grid-Connected; Large Scale; N/A; China, Nanjing; Temperate	Minimizing user load dispatch and emission dispatch.	Yes	Deep Reinforcement Learning (DRL)	Machine Learning-Supported	Mid-Term	Yes	DRL—Improved solving speed by 56.74% over NSGA-II and 15.62% over CVX, effectively managing renewable energy uncertainties. Compared against heuristic optimization.
[49] (2023)	PV, WT, BESS; Grid-Connected; Large Scale; N/A; Saudi Arabia, Wadi Alddawasir; Arid	Two-stage optimization to minimize operational costs and enhance self-consumption rates.	Yes	Harris Hawks Optimizer (HHO)	Metaheuristic	Mid-Term	Yes	HHO—Increased self-consumption (100% SCR), improved hosting capacity (8.863 MW to 10.213 MW), and reduced daily operating costs by 23.8%. Compared against standard optimization algorithms.
[53] (2023)	PV, BESS; Grid-Connected; Small Size; DC; Canada, Vancouver; Temperate	Maximizing power supply availability, minimizing Net Present Cost (NPC), and maximizing power efficiency.	No	GA	Metaheuristic Algorithm	Long-Term	Yes	GA—Achieved 93.76% energy efficiency, reduced costs by 14%, and provided 100 times lower downtime. Compared against HOMER software for MG design optimization.
[57] (2023)	DG, BESS, WT Farm; Standalone; Medium Scale; N/A; N/A	Minimizing total costs by considering unit operating costs, diesel generator emissions, ES degradation, and adjustment costs from wind power forecasting errors.	No	Mixed-Integer Second-Order Cone Programming (SOCP)	Deterministic	Mid-Term	Yes	SOCP—Improved reliability (99%) while balancing cost reductions. Compared against Deterministic and Stochastic Programming-Based models.
[61] (2022)	PV, WT Farm, DG, BESS; Grid-Connected; Medium Scale; N/A; Singapore; Tropical	Minimizing total economic energy costs.	No	Proximal Policy Optimization (PPO)	Machine Learning-Supported	Mid-Term	Yes	PPO—Achieved 32.8% cost savings and 28.5% carbon emission reduction. Compared to conventional grid-based power reliance.
[62] (2023)	RES, Combined Heat and Power (CHP), ESSs, Heat Energy Storage (HES); Both Grid-Connected and Standalone; Medium Scale; AC; N/A; N/A	Minimizing Cost of Energy Use (CET).	No	Myopic Optimization with Linear Programming (LP) and Reinforcement Learning (RL)	Hybrid Deterministic and Machine Learning-Supported	Long-Term and Short-Term	No	LP + RL—Improved operating cost by 90.98% compared to myopic model alone.
[63] (2023)	Exhaust Air WT, Concentrated Photovoltaic Thermal (CPVT), BESS, HES, DG; Grid-Connected; Medium Scale; AC; Iran; N/A	Minimizing the total CET and BESS degradation.	No	State-Action-Reward-State-Action (SARSA)-based Composite Differential Evolution (DE) Algorithm	Hybrid Machine Learning-Supported and Metaheuristic	Mid-Term and Short-Term	Yes	SARSA-Based Composite DE—Reduced energy and battery degradation costs, optimized renewable resource usage. Compared against Q-Learning and various DE-based methods.
[64] (2023)	MT, PV, WT, BESS; Both Grid-Connected and Standalone; Large Scale; AC; N/A; N/A	Minimizing operating costs and imbalance costs by penalizing power shortages and surpluses caused by renewable generation variability.	No	Data-Driven Robust Optimization Approach	Robust Optimization (RO)	Mid-Term and Short-Term	Yes	Data-Driven Robust Optimization—Reduced operational costs by 28% (MG1), 21% (MG2), and 19% (MG3). Compared against Traditional polyhedral set-based RO models.
[65] (2024)	BESS, FC, Natural Gas Storage (NGS), CHP Units; Both Grid-Connected and Standalone; Small Scale; AC; Iran (Zabol, Tabriz, Urmia); Continental	Minimizing the operational cost, including generation and reserve costs of MTs, ESS operation costs, and total trading payments. A two-stage RO model ensures feasibility against uncertainties by optimizing day-ahead scheduling and real-time adjustments, minimizing imbalance costs from renewable generation variations.	No	RO Techniques	RO	Mid-Term and Short-Term	Yes	RO—Achieved 67.54% reduction in load decrease, improved network performance from 15.48% to 69.85%, and enhanced recovery speed (61.59%).
[66] (2024)	PV, WT, BESS; Grid-Connected; Medium Scale; DC; India, Bhubaneswar; Tropical	Minimizing total operational costs, including EM of PV, wind, and batteries, maintenance expenses, and penalties for maintaining battery SOC within 20%-80%, while maximizing power availability to meet active and reactive power demands.	Yes	Modified MPC with Firefly Algorithm (FA)	Hybrid Model-Based and Metaheuristic Optimization	Short-Term	Yes	MPC + FA—Improved voltage stability, optimized active/reactive power, and reduced operational costs. Compared against Proportional-Integral PSO, Active Disturbance Rejection Control, and standard MPC variants.
[18] (2024)	PV, WT, BESS; Both Grid-Connected and Standalone; Small Scale; Hybrid AC/DC; Malaysia, Johor Bahru; Tropical	Maximizing reliability, minimize total costs, and reduce unutilized surplus power.	Yes	Iterative Pareto-Fuzzy (IPF) Method	Fuzzy Logic-Based	Mid-Term	Yes	IPF—Achieved a total cost of $5.12M, minimized surplus power to 201,364.45 kW, and ensured high reliability (>90% Index of Independence, >77% Reliability Test Efficiency). Compared against NSGA-II, Modified PSO, and GWO.
[46] (2024)	PV Plant, WT Farm, DG, BESS; Grid-Connected; Large Scale; DC; Egypt, Greater Cairo; Tropical	Minimizing operational costs by reducing total daily expenses, including grid electricity purchases, energy sales, battery operations, and diesel generator costs, while also minimizing deviations from target load profiles.	No	Honey Badger Algorithm (HBA)	Metaheuristic	Mid-Term	Yes	HBA—Achieving 4.56% cost reduction, and 2.3% savings in real-time operations under uncertainties. Comparison focused on real-time performance vs. theoretical optimal cost under perfect foresight.
[67] (2024)	WT, PV, BESS, Plug-in (EVs); Both Grid-Connected and Standalone; Medium Scale; N/A; N/A	Minimizing energy loss costs over 24 h, power purchase costs from the upstream grid, load curtailment costs, diesel DG operation costs, and battery and EV operational expenses.	No	Column and Constraints Generation (C&CG)	RO	Mid-Term	Yes	C&CG—Minimized total costs (86,978 vs. 93,936 for GA and 91,326 for PSO), reduced energy purchases and PV curtailment, and optimized peak charge. Compared against GA and PSO.
[68] (2024)	BESS, WT, PV; Grid-Connected; Small Scale; N/A; N/A	Minimizing the cost of energy trading for MGs, including buying, selling, and transportation costs.	No	Stochastic Dual Dynamic Programming (SDDP)	Stochastic	Mid-Term	Yes	SDDP—Improved MG EM efficiency, achieving cost reductions and optimizing decisions under uncertainty. Compared against Deterministic Equivalent Model (DEM).
[69] (2025)	DG, FC, WT, PV; Grid-Connected; N/A; Hybrid Ac/DC; Northwestern China; Dry	Minimization of economic dispatch cost.	No	Large-Scale Solvers like CONOPT and GUROBI	Deterministic	Mid-Term	No	Model of MGs to achieve a fully convex. Computational time reduction using the model III.

The following discussion expands on the main optimization aspects, starting with problem formulation, which includes optimization objectives and constraints, addressing both multi- and single-objective discussion. Then, it examines the EMS time frames, approaches to handling the sources of uncertainty, and an analysis of optimization methods between studies.

4.1. Problem Formulation

This section explains the objective functions and constraints within the EMS, which are important for attaining the optimization goals.

4.1.1. Optimization Objectives

The objective functions in the optimization of the EMS are formulated based on the specific goals of the MGs. Table 2 presents the objective functions classified according to the key physical components of MG, including CDER, NCDER, and ESS. Another category is grid interaction and Demand-Side Management (DSM), which are related to the grid interaction and the demand. Finally, objectives related to system reliability and energy balance emphasize the importance of ensuring stability within the system. The objectives studied include operational cost (e.g., fuel consumption reduction, generator cycling minimization), environmental impact (e.g., emissions minimization), and system reliability (e.g., reducing power supply deficiencies, strengthening grid resilience, optimizing power flow and stability).

Table 2. Classification of Objective Functions in Optimization EMS by Common Components.

Objective Function	Description	Ref.
Conventional Distributed Energy Resources (CDERs)—DGs and MTs
Minimization of Fuel Consumption	Reducing dependency on fossil fuel generators to improve efficiency and sustainability.	[21,23,44,45,48,59,60]
Minimization of Generator Cycling	Reducing startup/shutdown cycles to extend generator lifespan and enhance operational efficiency.	[21,23,59,60]
Minimization of Emissions	Reducing greenhouse gas emissions, and integrating cleaner energy alternatives.	[15,21,23,24,59,60,70]
RES—PV and WT
Maximization of Renewable Energy Utilization	Increasing self-consumption of solar, wind, and hybrid renewable energy to reduce reliance on fossil fuels.	[21,23,49,57,59,66,67]
Mitigation of Renewable Intermittency	Smoothing power output variability from renewable sources to enhance grid stability.	[49,50,52,64,65,71]
ESS—Batteries, Hydrogen, and Supercapacitors
Optimization of Battery SOC and Charging Strategies	Managing SOC to maximize battery efficiency and lifespan.	[21,22,23,49,51,60,63,65,66,68]
Minimization of Battery Degradation	Reducing degradation by optimizing Depth of Discharge (DOD) and charge-discharge cycles.	[10,21,23,47,49,63,66]
Battery System Resilience	Ensuring energy storage readiness in emergencies or blackouts.	[49,65]
Optimization of Hydrogen Storage and Utilization	Enhancing hydrogen efficiency for energy storage and demand response integration.	[72]
Grid Interaction and Demand-Side Management
Reduction of Grid Energy Costs	Optimizing electricity purchases to minimize dependency on external energy sources.	[21,23,49,51,59,61,62,68,71,72]
Demand-Side Management	Implementing load shifting, demand response, and real-time energy balancing strategies, including emission-aware dispatching.	[21,23,49,52,59,65,66,71,72]
Enhancement of Energy Trading Mechanisms	Optimizing market participation, MG energy trading, and vehicle-to-grid (V2G) strategies.	[49,65,68,71]
System Reliability and Energy Balance
Minimization of LPSP	Ensuring reliable power supply and reducing the probability of outages.	[20,44]
Enhancement of Grid Resilience	Improving system reliability through advanced grid control and autonomous operation.	[18,20,21,23,49,53]
Power Flow and Stability Optimization	Ensuring stable voltage and frequency regulation across MG operations.	[21,23,49,50,59,64,65,66,71]

Although many optimization EMS focus on a single objective, such as minimizing fuel consumption or maximizing renewable energy utilization, they can also be formulated as MOO problems to account for multiple criteria simultaneously. The choice depends on the complexity of the system, priorities, and potential conflicts. In such cases, additional techniques, such as the Pareto framework, become essential for comprehensive assessment and optimization.

Some studies simplify the problem by aggregating multiple objectives into a single function, either by direct summation (e.g., [57,60,63,67]) or using weighting sum optimization, as seen in [53]. When applying the latter, normalization is recommended to handle objectives with different scales or units [73].

In contrast, MOO directly addresses diverse and often conflicting objectives. Unlike single-objective optimization, which prioritizes a single criterion, MOO seeks to optimize multiple objectives simultaneously [70].

In EMSs for MGs, conflicting objectives often arise, such as minimizing operational costs, reducing environmental impact, maximizing renewable energy utilization, or minimizing battery degradation. These objectives are interdependent, requiring careful trade-offs to achieve a desired solution.

In real-world applications, the Pareto front becomes a valuable decision support tool for visualizing trade-offs between objectives [74]. Once a set of non-dominated solutions is generated, system operators or policymakers can prioritize specific solutions based on contextual factors such as regulatory limits, carbon pricing, or operational budgets. For example, in environments with strict emissions constraints, solutions with lower emissions may be favored even if they involve slightly higher costs. In contrast, cost-sensitive decision-makers might opt for the solution with the lowest operational expense, tolerating a moderate environmental impact.

An example of these contrasting approaches can be seen in [44], where the authors initially solve a problem with two objectives: cost of energy (COE) and Deficiency of Power Supply Probability (DPSP). Initially, the problem is solved as a single objective with COE as the main objective and DPSP as the constraint. However, this approach only yields a single optimal solution, which may not represent a trade-off between objectives. Alternatively, a MOO approach using the Pareto front is explored. This method provides the optimum solution and a set of trade-off solutions in the Pareto front, giving decision-makers multiple options to choose from [14]. Here, the decision-makers can analyze this front to select solutions aligned with their priorities and constraints.

In [45], a MOO problem is defined for a hybrid MG with three objectives: hourly cost of using the generators, fuel cost, and battery degradation cost. It is referred to as the optimal strategy compared to the priority strategy, which employs a rule-based algorithm. The priority strategy aims to minimize diesel power generation but keep each DG running above its technical minimum power set point if possible; maximize PV power generation, curtailing if strictly necessary; and charge the BESS only with excess energy and discharge it only if this reduces diesel consumption. The results of this work present a comparison between both strategies, where the optimal strategy allows for higher PV penetration because of the consideration of the better adjustment of the charging and discharging operations of the batteries, which contributes to energy cost minimization since it reduces the DG operation.

In [47], a MOO strategy for hybrid MG is defined. The strategy focuses on reducing maintenance costs and preventing a reduction in battery life by minimizing the maximum battery current. In addition, the approach aims to optimize the maximum and minimum battery power to enhance the interaction between supercapacitors and batteries. These objectives are addressed separately using a MOO unit.

4.1.2. Constraints

Constraints are essential in EMS to ensure safety, maintain system stability, and address resource limitations. Table 3 presents the constraints identified in the reviewed literature, categorized by the type of components they affect. These can be categorized into three main types: technical, operational, and economic. Some studies also mention logical and environmental constraints [30].

Table 3. Categorization of EMS Constraints and Corresponding MG Components.

Constraint	CDERS	RES	ESS	Load	System	Ref.
Power Balance					X	[20,21,23,24,45,49,51,57,59,61,65,67,68,75]
SOC and DOD Limits			X			[20,45,47,48,49,50,51,62,63,67,68,76]
Charging and Discharging Constraints			X			[21,23,46,49,57,59,60,65,68,71]
Generation Power Limits	X	X				[22,23,44,45,47,48,49,52,57,60,63,64,67]
Spinning Reserve Requirements	X					[45,48]
Capacity Constraints	X	X	X			[18,20,21,22,24,59,65,67,68,71,76]
Grid Power Trade Limits					X	[51,67,68,71]
Renewable Energy Fraction		X				[20,44]
Voltage Limits					X	[18,47,49,65,67,71]
Load-Demand/Flow Constraints				X		[18,46,59,66,76]
Energy-Balance Constraints			X	X		[22,62,63]
Power-Flow Constraints					X	[61,66,71]
Demand-Response Constraints				X		[23,57]
Ramp-Rate Constraints	X	X				[21,57,60,61]
System-Imbalance Constraints			X		X	[64]
Thermal-Demand Constraints				X		[62,63]
Budget and Investment Limits					X	[53]
Power-Availability Constraints	X					[53]
Curtailment Constraints		X				[45]
Converter Limits					X	[53]
Energy-Demand Constraints				X		[62,63]
Energy-Flow Constraints					X	[62,63]
Power Supply				X		[44,67]

Technical constraints in EMSs define the physical limitations and interactions of components within the system. These include power balance, generation power limits, SOC and DOD limits, voltage limits, power ramp-rate constraints, and system-imbalance constraints. Operational constraints include charging and discharging constraints, Renewable Energy Fraction (REF) limits, demand response constraints, and spinning reserve requirements. Economic restrictions regulate budget limitations, investment constraints, and energy trading limits to ensure financial feasibility.

Finally, Figure 6 presents a conceptual map that illustrates some of the previously mentioned objectives along with categories of constraints that may be considered during the problem formulation stage. These objectives can be addressed individually within an EMS, resulting in a single-objective formulation or simultaneously, leading to a multi-objective optimization approach.

Figure 6. Conceptual map of EMS problem formulation.

4.2. Time Frame

In making predictions, it is crucial to establish a clear time frame for optimization [77,78]. This process involves considering two important aspects: the horizon, which defines the period during which predictions will be generated, and the resolution, which determines the level of detail in those predictions [30]. Since, depending on the application, the time horizon can vary from a few hours to several years. Here, we include the following categorization of time frames: Short-term ranges from seconds to some hours are primarily used for real-time control, load balancing, and immediate decision-making in EMS. Mid-term covers from one day to weeks, providing an essential basis for optimizing scheduling strategies, DSM applications, and forecasting energy needs. Long-term spans from months to years are essential for planning infrastructure investments, evaluating the feasibility of MG expansion, and studying long-term policy impacts.

For example, for operation management, the scheduling of MG EMS resources is typically dispatched with a one-hour scale resolution [49,52,57,60], meaning that predictions are made hourly. These forecasts extend over a “look-ahead time horizon” of 24 h, effectively providing a 24-h window into the future. This example is a mid-term frame. Another example is feasibility studies and sizing of MGs, which typically fall under a long-term time frame [20,44,53].

4.3. Uncertainty

MG EMSs are affected by various uncertainties that disrupt energy scheduling, cost optimization, and system reliability. These uncertainties arise from environmental, economic, and technical factors [79], and depending on the type of uncertainty, the optimization technique to adopt may be different [31]. Addressing these uncertainties is critical, as inaccurate predictions in renewable generation, energy prices, or load demand can result in inefficient power dispatch, higher operational costs, and system instability [80].

Table 4 presents studies that address uncertainties in MG EMS, detailing the reference, source of uncertainty, the method employed, time frame, and data source. The data sources include historical records, experimental MG datasets, meteorological databases, and simulated data, which are crucial for modeling and validating uncertainty-handling methods. The time frame of these studies is focused on short-term applications [18,22,46,49,50,51,57,61,62,63,64,65,66,67,68,71,72,76], this was the predominant scope observed in the reviewed literature.

Table 4. Uncertainty Considerations in Optimization EMS.

Ref.	Purpose/Source of Uncertainties	Method Category	Method	Data Source
[50]	Renewable power generation and load demand	Stochastic	Numerical Weather Prediction Model	MeteoGalicia THREDDS Server and Public University of Navarre dataset
[76]	Optimize demand response and hydrogen utilization under fluctuating renewable energy conditions	Machine Learning	Artificial Neural Network (ANN)-Based Load and Renewable Energy Forecasting	Historical operational data from the LabDER experimental MG
[51]	Quantify the effect of PV generation and load uncertainties on the performance of stochastic model predictive control	Stochastic	Scenario Generation with Stochastic Optimization	Historical PV and load profiles from NRLab (2016–2020)
[72]	To capture the uncertainty associated with prices, RES yields, and energy demand, a random model is employed	Stochastic	Random Model with Normal Distribution	Simulated renewable energy variability and demand fluctuations
[71]	Models various factors, including the number of plug-in electric vehicles, their arrival and departure times, prediction errors in hourly load and pricing, and power output variations from wind turbines and photovoltaic systems	Stochastic	Unscented Transform	N/A
[22]	Forecast EV charging demand and renewable generation variability for optimal scheduling	Stochastic	Monte Carlo Simulation	Predicted EV charging load and renewable energy forecast
[23]	The uncertainty of the load and renewable energy sources. Ensure power balance under uncertain demand and generation conditions	Interval Analysis	Standard Deviation Estimation with Spinning Reserve Adjustment	Synthetic data
[52]	The volatile nature of renewable energy generation	Stochastic	Probability Distribution Model	Solar Radiation Intensity Data from NREL, User Load Statistics Data
[49]	Modelling the volatility in market prices, RESs output, and electrical load using stochastic optimization techniques	Fuzzy	Fuzzy Clustering Method	N/A
[53]	Ambient Conditions Uncertainty and Load Uncertainty	Stochastic	Markov Chain and Degradation Models	Statistical Field Data
[57]	Manage uncertainties primarily related to both power generation and demand within the energy	Robust	Moment-Based Distributionally Robust Optimization	N/A
[61]	Determine the optimal schedule of the energy mix	Machine Learning	PPO	Weather Databases (Meteorological, from 12 September 2013 to 18 September 2013)
[62]	Energy storage charging and discharging decisions within an MG under specific constraints	Machine Learning	Reinforcement Learning	Open Energy Information Datasets, U.S. Department of Energy
[63]	Wind and solar power generation and the buying/selling costs of electrical and thermal energy	Machine Learning	SARSA Algorithm	Synthetic Data
[64]	Extracting probability distribution information in the uncertainty data of renewable generation	Stochastic	Robust Kernel Density Estimation (RKDE)	Synthetic Data
[65]	Addressing uncertainties that arise during planning processes	Robust	Robust Optimization Techniques	Synthetic Data
[66]	Uncertainty in PV irradiance, wind speed, and load variations	Stochastic	Modified MPC	Laboratory Data
[18]	Renewable energy uncertainty modeled using Gaussian noise for solar irradiance and Weibull noise for wind speed	Stochastic	Gaussian and Weibull Noise Models	Synthetic Data
[46]	Uncertainty in solar irradiance, wind speed, and grid tariff; actual demand load uncertainty up to 20%	Information Gap Decision Theory	Consideration of Forecasted Errors	N/A
[67]	Accounts for uncertainty in load demand, renewable generation, and influencing factors in optimization constraints	Robust	Robust Optimization Technique	N/A
[68]	Stochastic models for energy pricing and Stochastic representations of energy generation and demand	Stochastic	Monte Carlo Simulation	N/A

N/A: Not available.

Next, the sources of uncertainty in MG EMS and the different approaches used to address them are presented.

4.3.1. Sources of Uncertainty

Thus, studies categorize uncertainty in MG EMS into three main types: environmental, economic, and technical, as follows.

Environmental Uncertainty

• Solar and Wind Variability: RESs fluctuate due to weather conditions, making power generation unpredictable [18,50]. Stochastic weather models like Numerical Weather Prediction and Monte Carlo simulations incorporate these variations into EMS planning [22,68].

Economic Uncertainty

• Fuel Prices & Market Fluctuations: Variations in fuel and electricity prices impact MG cost efficiency [49,79]. Stochastic price modeling with Gaussian distributions integrates these uncertainties into optimization strategies [72].

Technical Uncertainty

• Load-Demand Variability: User consumption fluctuates based on behavioral patterns, seasons, and unexpected demand spikes [61,76]. Machine Learning (ML) techniques, such as Artificial Neural Networks (ANNs), enhance demand forecasting accuracy.
• Battery Performance & Degradation: The SOC and aging of batteries introduce complexity in EMS scheduling [51,57]. Robust optimization ensures EMSs remain feasible under worst-case battery degradation scenarios.
• Electric Vehicle Integration: The random arrival and charging demand of Electric Vehicles (EVs) create scheduling uncertainties [71]. Unscented Transform and Monte Carlo simulations help predict EV charging behaviors [22].

4.3.2. Handling Uncertainty

This subsection provides a concise overview of some approaches used to handle uncertainty in EMS. Each method is briefly introduced, accompanied by illustrative application examples drawn from recent literature.

• Stochastic modeling: models uncertainty in renewable energy generation, demand fluctuations, and pricing. Techniques include Monte Carlo simulations, probability distribution models, and scenario-based optimization [18,64,68]. It enables the analysis of a spectrum of potential future scenarios by employing probability functions [27].
• Fuzzy Logic Optimization: this technique confronts imprecise and linguistically defined uncertainties, mirroring human-like reasoning [31]. It handles imprecise and ambiguous data (e.g., user energy behavior and uncertain weather conditions). Methods such as fuzzy clustering and fuzzy rule-based models provide greater flexibility in EMS decision-making [27,49].
• Robust Optimization: ensures EMS remains feasible under worst-case deviations, particularly for fuel prices, demand spikes, and battery degradation [65,81]. This method defines an uncertainty set and seeks solutions that are resilient to all possible scenarios within that set, offering a degree of guaranteed performance [27].
• Model Predictive Control: it is a real-time control technique that optimizes energy dispatch and system operation by continuously adjusting decisions based on short-term forecasts. MPC is particularly effective in battery management, demand response, and optimizing hybrid energy systems, where it adapts to fluctuations in generation and consumption [27].
• Interval Analysis: a technique that models uncertainty by defining upper and lower bounds for unknown parameters rather than assuming a probability distribution [31]. It is useful when data are incomplete or lack precise statistical properties, ensuring that optimization accounts for all possible values within a given range [23].
• Information Gap Decision Theory: applied in extreme uncertainty scenarios where probability distributions are unknown, ensuring that solutions remain robust even with limited information [31], such as in [46,67]. This approach addresses uncertainty by starting with a predicted value for each uncertain parameter [31]. It then explores how much these parameters can deviate from the predicted value without causing the system to fail [31]. It also aids decision-makers by identifying the maximum uncertainty range within which the system remains reliable [27].
• Machine Learning Approaches: improve forecasting accuracy for energy demand, renewable generation, and real-time EMS control. This technique enables more responsive handling of uncertainties in EMS operation by identifying patterns in historical data [27]. Examples include reinforcement learning for ESS optimization [62] and deep learning for load forecasting [82].

By integrating these methods, MG EMS can improve resilience against uncertainties, minimize operational costs, and enhance the reliability of renewable-based energy systems.

Figure 7 provides a summary of the sources of uncertainty addressed in EMS design and the approaches employed to mitigate their impact, based on the studies reviewed in Table 4. Figure 7a shows the percentage of occurrence for each type of uncertainty: environmental uncertainties were the most frequently considered (76%), followed by technical uncertainties (71%) and economic uncertainties (29%).

Figure 7. Uncertainty summary. (a) Sources of uncertainty in MG EMS and (b) common handling approaches.

Figure 7b highlights the approaches used to manage these uncertainties. Stochastic modeling was the most prevalent approach (52%). Other approaches included robust optimization (14%), machine learning (19%), and, less frequently, fuzzy logic, interval analysis, information gap theory, and model predictive control (each accounting for 5%).

Figure 7 indicates a strong emphasis on addressing environmental and technical uncertainties in EMS design, reflecting their critical impact on renewable generation and load variability. However, the limited consideration of economic uncertainties points to a research gap. In particular, grid-connected MGs often rely on dynamic electricity prices, and ignoring such economic variability can result in revenue losses. The dominance of stochastic methods highlights a prevailing preference for probabilistic modeling. Nevertheless, the increasing complexity of MG operations under dynamic conditions calls for an adoption of diversified uncertainty-handling approaches. Notably, machine learning is also gaining interest due to its ability to learn complex patterns.

The following section details stochastic, deterministic, ML, and RO techniques in optimization EMS.

4.4. Optimization Methods

This subsection outlines some optimization methods and algorithms, referencing their application in the studies summarized in Table 1. To aid in the selection process, Table 5 presents a summary of these methods, highlighting their strengths and limitations. Additionally, examples of their use in MG applications are included.

Table 5. Optimization Methods: Pros, Cons, and Applications.

Method	Pros	Cons	MG Applications
Metaheuristic Optimization	- Handling complex, nonlinear problems. - Using stochastic search to explore solutions spaces. - Applicable to some variants of real-time control.	- Risk of becoming trapped in local minima. - It could present a high computational cost. - Sensitive to parameter tuning.	- Suitable for: Multi-objective optimization of complex MG operations with conflicting goals (e.g., cost vs. emissions). - Examples: Optimal dispatch of DERs in hybrid AC/DC MGs [50], ESS scheduling [59], demand response management [23].
Deterministic Optimization	- Guaranteeing global optimality (for linear/convex problems). - Well-suited for problems with clearly defined constraints.	- Struggling with nonlinearities. - Scalability issues for large, complex MGs. - Less effective for real-time control and highly dynamic or uncertain scenarios.	- Suitable for: MG problems that can be formulated linearly/convexly (e.g., sizing, economic dispatch in simpler systems). - Examples: Optimal sizing of MG components [20] and day-ahead scheduling [60].
Machine Learning-Supported Optimization	- ML can learn and improve over time, allowing systems to adapt autonomously, not just in the short term but over longer periods. - Handling worst-case uncertainty.	- High computational cost, especially in deep learning models. - Dependence on large, high-quality datasets. - Requiring significant processing power for training and real-time execution.	- Suitable for: MGs with abundant data, where forecasting and adaptive control are crucial. - Examples: Load forecasting [76], RES generation forecasting and adaptive ESS control [22].
Stochastic Optimization	- Effectively modeling uncertainties. - Providing a realistic representation of system behavior by considering probability distributions.	- High computational burden (scenario generation). - Being ineffective in smaller-scale scenarios.	- Suitable for: MGs with high RES penetration, for energy trading under uncertainty [68], and dispatch under weather variability. - Examples: Risk assessment, optimal bidding in electricity markets [68].
Robust Optimization	- Ensuring feasibility under worst-case scenarios (e.g., outages). - Not requiring probability distributions. - Computationally less demanding than stochastic.	- Potentially leading to overly conservative solutions. - Sensitivity to uncertainty set definition.	- Suitable for: MGs prioritizing reliability, for outage management [65]. - Examples: Sizing ESS for backup power, guaranteeing supply to critical loads.

Metaheuristic: these algorithms have gained traction in MG optimization because they can handle complex, nonlinear problems. Inspired by natural phenomena, these algorithms, such as Wind-Driven Optimization (WDO), Ant Colony Optimization (ACO), Genetic Algorithm (GA), and Particle Swarm Optimization (PSO), offer diverse strategies for solving optimization challenges [83]. Specific metaheuristic methods, like GA and PSO, incorporate stochastic elements such as random sampling and stochastic search techniques. These elements allow them to explore solution spaces and, in many cases, help avoid local optima [84].

However, despite these stochastic features, nature-inspired methods like PSO can still struggle with becoming trapped in local minima, especially when the problem landscape is complex, or the parameters are not well-tuned [30]. Alternative approaches, such as the Grasshopper Optimization Algorithm (GOA), have been explored to address this limitation. These methods demonstrate improved capability in escaping local optima [44].

PSO has been effectively applied to MOO problems in battery energy storage. As shown in [47], PSO was used to optimize battery maintenance costs and longevity, adapting to varying charging and discharging patterns caused by fluctuations in power generation. This flexibility allowed PSO to explore the solution space efficiently while maintaining reasonable computational speed, making it suitable for real-time control applications.

Similarly, GOA has been utilized to address MOO problems in hybrid MGs, specifically for minimizing the Cost of Energy (COE) and Deficiency of Power Supply Probability (DPSP) [44]. Compared to PSO and cuckoo search algorithms, GOA achieved better convergence, improving cost efficiency and power reliability.

Despite their advantages, metaheuristic methods have inherent limitations. Some of the concerns include high computational costs, long convergence times, sensitivity to parameter tuning, difficulties in handling larger problem sizes, and reliance on high-quality data for effective optimization [26].

Deterministic: deterministic optimizers are software tools or algorithms created to find the optimal solution to a mathematical problem. These problems can manifest in various forms: linear, nonlinear, integer, and Mixed-Integer Linear Programming (MILP). Among these, MILP is one of the extensively explored methods [85].

In [45], MILP was employed for the optimal EMS, which allows better use of the BESS based on the forecasted load demand and renewable energy generation. Additionally, it increases renewable penetration and reduces diesel generation. The strategy also considers multiple objectives, such as generator operation cost and battery degradation.

In addition to the last-mentioned deterministic methods, dynamic programming is also used. This problem-solving technique involves breaking down a complex problem into smaller, more manageable sub-problems. These sub-problems are then solved individually, and their solutions are stored to avoid redundant computations. By efficiently computing and storing solutions to these sub-problems, dynamic programming allows for calculating the overall optimal solution [86].

In [87], the primary objective of long-term EMS is to determine the optimal allocation of available storage device capacity, such as SOC for batteries and State of Health (SOH) for hydrogen tanks, during the upcoming time frame to achieve the defined energy management goals. The dynamic programming technique is employed in this context to optimize the management of battery SOC and hydrogen fuel cell SOH according to long-term energy plans.

Machine Learning-Supported: ML is a branch of Artificial Intelligence (AI) that enables systems to recognize patterns and make predictions without explicit programming [88]. In EMS, ML techniques enhance centralized and decentralized frameworks by improving forecasting accuracy, optimizing energy dispatch, and managing system uncertainties [5]. These algorithms assist in predicting solar radiation, air temperature, and potential power outages, ensuring optimal decision-making in hybrid MG operations [89].

Reinforcement Learning (RL) is an adaptive approach where an agent learns optimal control policies by interacting with the environment and maximizing long-term rewards [88]. Advanced RL techniques, such as Deep Reinforcement Learning (DRL), integrate neural networks to enhance real-time energy dispatch while managing uncertainties in renewable energy generation and load fluctuations [52]. In distributed EMS, RL facilitates autonomous decision-making, allowing MGs to adapt to market conditions, energy demand, and generation variability [5]. Proximal Policy Optimization (PPO) has been applied to minimize total economic energy costs, achieving significant cost savings and emissions reduction [61]. State-Action-Reward-State-Action (SARSA) algorithms optimize wind-PV energy storage management, reducing storage operation costs while improving power tracking [22]. Additionally, ANN combined with PSO has improved energy forecasting accuracy, reducing errors in multi-step predictions [76].

Despite its effectiveness, ML-based optimization faces challenges. High computational costs are a major limitation, especially in deep learning-based models that require significant processing power for training and real-time execution [88]. Additionally, ML algorithms depend heavily on large datasets, making their performance sensitive to the availability and quality of historical energy data [89].

Stochastic Optimization: Stochastic programming incorporates uncertainty into optimization by modeling random variables through Stochastic functions. It optimizes decision-making across multiple scenarios, typically structured in two stages: day-ahead planning using forecasted data and real-time adjustments based on actual conditions [27]. In MG systems, the variable nature of RESs due to changing weather conditions affects power generation, making stochastic approaches essential for energy management [90].

Scenario-based stochastic MPC has been applied to reduce operational costs and improve scheduling reliability in MGs [51]. Stochastic dual Dynamic Programming (SDDP) has been used for energy trading, managing cost variations due to uncertainties in power availability and demand [68]. By considering multiple scenarios and probability distributions, stochastic optimization provides a realistic representation of system behavior, helping EMS manage the risks associated with intermittent renewable generation [68].

However, the primary limitation of stochastic programming lies in high computational requirements, particularly in real-time applications. Stochastic dynamic programming offers a solution by incorporating uncertainties at each time step, improving adaptability in real-time optimization [27].

Robust: RO provides feasible solutions under uncertainty without requiring knowledge of probability distributions. Instead, it defines uncertainty sets to model parameter fluctuations such as RES output and load demand [80]. This approach ensures that decisions remain valid under worst-case scenarios, making it useful for MG EM, where variations in power production and demand must be accounted for [27]. Compared to stochastic optimization, RO reduces computational complexity by avoiding scenario-based modeling, allowing for faster decision-making [27].

In [64], a data-driven robust optimization approach was employed to minimize operational costs and power imbalances in MGs, addressing renewable energy variability. Similarly, a two-stage robust optimization model in [65] optimized day-ahead and real-time scheduling while reducing imbalance costs.

Despite its efficiency, RO often leads to overly conservative solutions, as it focuses on worst-case scenarios, potentially resulting in higher operational costs and missed optimization opportunities [80]. The accuracy of the uncertainty set is crucial; if they are not well-defined, the solution may either underestimate risks or be excessively cautious, leading to inefficiencies [27].

5. Discussion and Future Trends

Table 6 outlines the contributions and future work of the reviewed studies. Based on this table, the following subsections briefly discuss essential points, future trends, and recommendations based on the studies.

Table 6. Contributions and limitations of optimization EMS.

Ref.	Contribution	Limitations/Future Work
[44]	Develops a mathematical model for a hybrid PV-WT-Diesel-Battery MG with a rule-based EMS using queuing theory. Applies a nature-inspired optimization algorithm for capacity planning, outperforming PSO and Crow Search Algorithm. Demonstrates environmental and economic benefits over diesel generators and assesses system resilience to future uncertainties.	Enhances EMS with MG pooling and harmonic mitigation. Incorporates uncertainty analysis like robust programming and Monte Carlo simulations. Explores hydrogen storage, Direct DC MG feasibility, and advanced optimization algorithms to improve renewable energy system performance and resilience.
[50]	Develops an EMS framework using two 25-rule FLC to smooth power fluctuations in a residential electro-thermal MG. One optimizes grid power exchange; the other manages demand-side response. One-year simulations with real data validate superior power smoothing using quantified smoothing indexes.	Enhances online optimization for EMS with advanced FLC to improve power smoothing while managing complexity. Integrates economic variables to assess tariff policy impacts. Addresses long-term historical consumption data availability for more accurate demand-side management.
[45]	Compares rule-based Priority EMS and optimization-based Optimal EMS for isolated hybrid MGs. Optimal EMS enhances BESS utilization, boosts renewable penetration, and cuts diesel use, while Priority EMS provides a simpler, more robust solution with lower computational demands.	Analyzes forecasting errors that may impact the advantage of Optimal EMS over Priority EMS. Enhances real-time decision-making to improve EMS performance under uncertainties.
[47]	Proposes an intelligent EMS combining FLC and PSO to improve power distribution and voltage stability in hybrid AC/DC MGs. Optimizes battery usage, reduces maintenance costs, and prevents battery life degradation through PSO.	Refines forecasting accuracy, integrates economic factors into optimization, and assesses scalability under various grid configurations. Includes battery performance analysis and compares the proposed EMS against alternative strategies for enhanced evaluation.
[76]	Develops a self-adaptive EMS with cascade ANN optimized by PSO to improve decision-making in standalone hybrid MGs, enhancing forecasting accuracy and system efficiency.	Expands the model to incorporate more energy generation and storage subsystems, enhances ANN training with additional experimental data, and integrates an automated control system for improved efficiency.
[48]	Introduces a TRBH EMS for hybrid MGs, enhancing rule-based scheduling with predictive capabilities to achieve near-optimal performance comparable to MILP.	Evaluates TRBH adoption in real systems, extends comparisons with predictive methods, addresses forecast uncertainty, and implements stochastic training with scenario-based evaluation for improved performance.
[51]	Develops a two-layer scenario-based stochastic MPC for real-time EM of nanogrids, integrating an Alternating Direction Method of Multipliers (ADMM) to reduce the computational burden.	Extends to islanded nanogrids with programmable loads, integrates diverse energy sources like wind turbines and distributed generators, and explores advanced EM methods, including learning-based and state-of-the-art approaches.
[72]	Enhances energy generation and consumption by integrating fuel cells, hydrogen fueling stations, and multi-energy storage. Uses multi-objective optimization with the SSA to manage renewable uncertainties, balance electrical and thermal systems, and optimize load shifting through demand response, ensuring energy transitions, resilience, and adaptability.	Limited to a single MG configuration, restricting broader applicability. Future work explores scalability, integrates emerging technologies, and includes life cycle assessments for hydrogen production.
[71]	Develops a management system for multiple renewable sources and plug-in EVs for short-term MG scheduling. Optimized with the Modified Marine Predators Algorithm, it minimizes total network costs, addressing energy shortages, reliability, and power supply interactions with the main grid.	Integrating demand response, improving forecast reliability.
[59]	Introduces GJO for energy storage scheduling in MG management, overcoming computational inefficiencies of PSO, TS, and ABC. It ensures faster convergence, improved efficiency, and lower operational costs, validated through case studies with hybrid energy sources and BESSs.	Implements the multi-objective optimization in a real-world MG, integrating hybrid energy and battery storage systems, monitoring performance, and refining the model for validation.
[20]	Presents a multi-objective optimization for hybrid renewable MGs, balancing techno-economic and reliability goals. Combines the Taguchi method, moth flame optimization, and fuzzy decision-making to minimize costs and power losses while maximizing renewable energy utilization.	Enhances wind model accuracy in optimization by exploring Weibull and least squares models, improving precision over the cubic model used in this study.
[21]	Proposes the OGGWO for multi-objective optimal scheduling in grid-connected MGs, balancing operational cost reduction and emissions minimization.	Improves uncertainty modeling, compares with additional optimization techniques, and extends the approach to various MG configurations for broader applicability.
[22]	Proposes an online reinforcement learning-based EMS using the SARSA algorithm to optimize MG operation under uncertainty, reducing energy storage operation costs and improving power tracking efficiency.	Expands the approach to larger MG networks and evaluates its performance by comparing it with other reinforcement learning techniques.
[23]	Develops an advanced MG model incorporating an incentive-based demand response program and a Vanadium Redox Battery for optimal scheduling. Uses MILP to optimize scheduling and evaluates its impact on battery operation, comparing performance with case studies using a genetic algorithm.	Integrates a hybrid battery model combining VRB with lithium-ion or absorbed glass mat batteries to enhance performance and flexibility in MG energy storage.
[24]	Proposes a Salp Swarm Algorithm for optimal MG operation, handling in generation, load, and pricing uncertainties. It minimizes generation and emission costs for grid-connected and standalone modes, solving a mixed-integer optimization problem.	Requires accounting for real-world complexities beyond tested variables and investigating further optimization techniques.
[60]	Develops a coordinated scheduling method for seaport MGs, integrating renewable energy with the main grid. It uses Adaptive Robust Optimization and time-shiftable load scheduling to minimize operational costs, emissions, and uncertainties across multiple timescales while ensuring distribution network constraints.	Future improvements include incorporating multi-energy carriers (heating and cooling) and analyzing electricity pricing schemes on seaport operations.
[52]	Proposes a robust multi-objective Load Dispatch Model using DRL to handle uncertainties in renewable generation.	Incorporating ESSs (batteries or EVs) for controlled charging/discharging can mitigate fluctuations but introduces investment costs and complexity, warranting further research.
[49]	Introduces a Two-Stage Optimization Approach for MG performance and hosting capacity by optimizing BESS sizing and placement under uncertainty.	Future work suggests hybridizing BESS instead of relying solely on sodium-sulfur batteries.
[53]	Proposes a Vectorial Microgrid Optimization (VMO) method integrating power architecture selection and sizing for flexibility and performance. Uses Graph Theory and Stochastic models to automate MG architecture selection.	Limited reliance on initial performance metrics and environmental variability. Future work suggests extending VMO with additional renewable sources and adaptive strategies.
[57]	Develops a Distributionally Robust Joint Chance-Constrained Energy Management Model for standalone MGs. Uses a Moment-Based Ambiguity Set and Bonferroni Approximation for reduced conservativeness, leading to a SOCP formulation.	Future improvements include integrating additional operational constraints (voltage and power loss), exploring tractable reformulations for time-dependent scenarios, and alternative ambiguity sets for robustness.
[61]	Uses an RL approach with PPO to enhance energy dispatch in hybrid renewable energy systems, achieving a 32.8% cost reduction and 28.5% decrease in carbon emissions.	Enhancing scheduling by integrating sequential neural networks and off-policy learning methods, eliminating reliance on model simulators.
[62]	Integrates RL with Myopic Optimization for MG energy management, optimizing electrical, cooling, and heating power-flow decisions without relying on forecasting.	Model performance constrained by learning episodes and agent action space, requiring high computational resources. Future research to address limitations before deployment.
[63]	Introduces an Industrial Demand Response program for Industrial Multi-Energy MGs, optimizing power and heat supply while maximizing renewable energy utilization.	Future work includes expanding the IDR framework by integrating additional renewable sources and considering real-time market fluctuations.
[64]	Develops a Data-Driven Peer-to-Peer Trading Mechanism for networked MGs using Nash Bargaining Theory and robust optimization. Introduces RKDE for improved data reliability and employs ADMM and Augmented Lagrangian-Based Alternating Direction Inexact Newton (ALADIN) for computational efficiency.	Future improvements include MT Startup/Shutdown Integration, addressing privacy concerns in ALADIN, and shifting from single-machine to parallel computing for performance analysis.
[65]	Enhances Distribution Network Security by modeling severe incidents with a Security Measurement Index, utilizing distributed generation resources, storage, and demand management. Implements MG partitioning for power exchange and robust optimization to address uncertainties.	Future research will focus on advanced optimization techniques (ML, Power-to-Gas Units, EV Charging Integration, and Standardized Resilience Metrics).
[66]	Proposes an Energy Management Approach for hybrid MGs, minimizing power loss and maximizing generation. The Multi-MPC-Improved Firefly Algorithm (IFA) strategy combines MPC with improved IFA) for power regulation.	Future work will improve Hybrid Microgrid (HMG) control via Fault Management Integration (FMI), EV Charging/V2G Optimization, and adaptive methods like ML and DRL.
[18]	Introduces a Spectral Clustering Algorithm improving clustering via the Silhouette Coefficient. Proposes a multi-objective Sizing Framework for RESs and BESSs) using the IPF technique.	Future work to test scalability on larger Active Distribution Networks beyond IEEE 33-bus systems to address real-world applicability.
[46]	Proposes a Two-Level Energy Management Strategy for grid-connected MGs integrating Demand-Side Management for day-ahead scheduling and real-time rescheduling based on demand, weather, and tariff fluctuations. Uses a Hybrid Bat Algorithm for enhanced efficiency.	Enhancing model with additional renewable sources, testing under different climatic conditions, and incorporating advanced predictive algorithms.

5.1. Discussion and Key Points

This section summarizes the key insights obtained from the reviewed studies, providing a consolidated discussion of the main findings related to optimization EMS of MGs.

• The use of hybrid models incorporating multiple energy resources, such as PV and WT, besides ESS, emphasizes the versatility and complexity of modern optimization EMS. The studies show advancements in balancing various components, optimizing battery use, and enhancing overall system efficiency.
• This review analyzes various optimization EMS, covering essential aspects like objective functions, constraints, optimization techniques, uncertainty, and time frame. It emphasizes how EMS can enhance system efficiency, reliability, and cost-effectiveness while addressing challenges like renewable energy intermittency.
• There are notable contributions regarding the economic benefits of optimized EMS. Optimized EMS can significantly reduce dependence on fossil fuels by minimizing operational costs and maximizing renewable energy utilization, thereby promoting financial sustainability.
• The necessity of MOO is emphasized for effectively handling conflicting objectives, such as balancing cost reduction with environmental sustainability. This approach helps avoid biases in decision-making, allowing the EMS to consider multiple goals comprehensively, which enhances overall system performance.
• Various strategies are explored in the literature for managing conflicts and uncertainties, including robust optimization techniques that consider worst-case scenarios, ML for improved forecasting, and stochastic programming, which incorporates randomness into decision-making. These more complete techniques could better handle the objectives with better-informed decisions regarding the objectives of the MG.
• Various optimization EMS, including metaheuristic algorithms, machine learning, robust and deterministic optimization, have been explored. This diversity aims to achieve optimal solutions under different operational conditions, ensuring MGs can flexibly adapt to changing situations.
• The review concludes that it is essential to select the appropriate EMS carefully to ensure stable and reliable operations within MG systems. Tailoring the EMS to the specific requirements of the MG facilitates better integration of RESs, enhances operational reliability, and improves overall system performance.

5.2. Research Gaps and Future Trend

This section synthesizes the research gaps identified throughout the reviewed literature on optimization EMS of MGs. By analyzing both methodological limitations and practical challenges, the objective is to highlight the main aspects that require further investigation.

• A critical limitation highlighted in the analysis is the lack of validation for optimization EMS under different weather scenarios and real data. RES power production is closely linked to weather, thus seasonal patterns, making incorporating regional uncertainties and seasonal validation essential. Future research must focus on validations to assess the efficacy of optimization strategies across seasonal conditions and real data.
• The analysis of limitations reveals a significant challenge concerning data scarcity and clarity, as shown in Table 4, where most were not mentioned. Many studies depend on historical data that may not adequately represent future scenarios or unusual events. Improved data collection is necessary to enhance the EMS validation, create optimization results that match real-world conditions or account for uncertainties.
• Adaptive EMS is essential to ensure reliable MG performance, especially given the global impacts of climate change, the variability of renewable energy sources, and the seasonal differences in weather. Future research should focus on validating the performance of EMS in different climate zones, seasons and under failure conditions.
• Many current studies focus on mid-term time frames without addressing short- and long-term strategies. This creates a gap when applying EMS strategies to other operational needs. Future research should expand to include exploring sizing, real-time applications, power quality, and other EMS aspects, ensuring that optimization strategies extend beyond scheduling.
• The analysis indicates that many optimization EMS may not fully account for the dynamic nature of user behavior and technological evolutions. As energy consumption patterns change, particularly with the rise of EV integration and smart technologies, EMS must adapt accordingly.
• Some methods, while ensuring accuracy, introduce complexity that may limit their practical application. For example, stochastic, robust, and MPC approaches are often difficult to implement [91]. Future efforts should focus on validating these methods in real-world scenarios and comparing optimization strategies to ensure alignment with MG performance under realistic conditions.
• Dynamic Pricing models, such as the one proposed in [92], promote local energy use in MGs and reduce grid dependence. EMS must incorporate dynamic pricing as energy markets move toward this model. This integration remains an underexplored area in current research.
• Future research should also consider the impact of regulatory frameworks, such as grid interconnection policies, tariff structures, and government subsidies, on the deployment and operation of MGs. Without supportive government policies, renewable energy-based MG projects can struggle to be implemented, particularly in remote or developing regions, where upfront costs and logistical challenges often arise. These considerations are especially important in studies focused on system sizing and energy pricing, where regulatory constraints and incentives can influence design and economic outcomes.

6. Conclusions

This study reviews various optimization approaches in the optimization of EMS for MGs. Using the PRISMA methodology, it analyzes a range of sources, including review and research articles, to explore fundamental themes such as the components of optimization EMS, MOO, and the influence of weather and climate on MG performance. The integration of RES is emphasized as beneficial for improving system efficiency and addressing climate concerns. However, uncertainties arising from the high dependency on weather conditions, including the need for seasonal validation and reliability objectives, must also be considered to mitigate the challenges associated with this issue.

The work identifies challenges to the adoption of optimization EMS, including data scarcity, lack of clarity, and validation of strategies in various weather scenarios. Furthermore, it underscores the need for more research that goes beyond mid-term solutions to include short- and long-term strategies, addressing aspects beyond the scheduling.

As the demand for optimized EMS grows alongside the increasing integration of RES in the “Age of Electricity”, the significance of this study lies in its comprehensive analysis, providing a foundational framework for future research that can adapt to evolving technologies and consider the potential challenges posed by climate change.

This review focused specifically on the main aspects of identification and analysis of optimization-based EMSs; Comparisons between different approaches to handling uncertainty and optimization methods were not made, nor control strategies were analyzed, both of which are considered limitations of the present study and proposed for future work. In addition, the literature search was restricted to the Scopus database.