Systematic Literature Review of Heuristic-Optimized Microgrids and Energy-Flexible Factories

Abstract

Decentralized renewable energy generation and consumption through microgrids, coupled with short- and long-term storage systems and enhanced demand flexibility, represent a promising strategy for mitigating grid stress and reducing emissions in the industrial sector. However, transitioning into a sustainable industry often poses challenges in terms of economic feasibility. This review surveys current optimization approaches and simulation functionalities to enhance feasibility. It follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, covering 1066 studies from 2016 to 2023 across three research areas: optimal system sizing of microgrids (OSS), optimization of electrical energy distribution to storage systems and consumers (EED), and energy flexibilization of factories (EF). As a result, 24 filtered sources from these areas were analyzed. Quantitative analysis indicated that evolutionary and swarm-inspired metaheuristics are predominantly applied in OSS, whereas exact linear problem solvers are favored for EED and EF optimization. A range of functionalities is available, and approaches often prioritize individual functionalities, such as load forecasting, dynamic electricity pricing, and statistical representation of energy generation, rather than comprehensively integrating them. Furthermore, no current approach simultaneously integrates optimization and simulation models across all three research areas.

1. Introduction

Climate change is posing a growing threat to society. Previous measures have proven inadequate, suggesting that at the current pace, the 1.5 °C maximum temperature rise target will be significantly missed [1]. Given the Paris Climate Agreement and, as an example of an industrialized nation, Germany’s tightened climate targets, there is a substantial need for significant reductions in carbon dioxide (CO₂) emissions by 2030 (65% reduction compared to 1990) and the complete elimination of net emissions by 2045 [2]. Despite the steadily increasing share of renewable energy sources (RES) in recent years, emissions have not decreased proportionally [3,4]. This is partly due to an increase in grid volatility resulting from the rising share of fluctuating RES and suggests a saturation point, indicating asynchrony between the demand and availability of RES during peak periods [5]. Consequently, photovoltaic systems (PV) or wind turbines (WT) are increasingly curtailed during these periods [6,7].

Decentralized energy generation and consumption, coupled with energy storage and demand flexibility, offer a promising avenue to relieve strain on public grids and reduce emissions [8]. The industrial sector, which accounts for Germany’s largest electrical power consumption (44% in 2021 [9]) and has significant potential for PV integration on factory rooftops, has emerged as a key player in decarbonization efforts. However, small- and medium-sized enterprises (SMEs) in the manufacturing sector face financial challenges and barriers to implementing sustainability measures. This partial lack of economic viability has hindered the widespread adoption of decentralized energy supply and storage [8,10].

To initiate the transformation towards a CO₂-neutral industry, approaches for analyzing and evaluating sustainable investment measures are needed that increasingly incorporate sustainability aspects in addition to economic assessment. Measures that positively impact the environment but are economically precarious must be reassessed. Because SMEs face investment barriers and lack financial resources, they need support in their decision-making process through transparent cost analyses to ensure effectiveness and optimized investment measures [11].

Common battery energy storage systems (BESS) offer a solution for short-term storage of self-generated RES, but they are costly, have limited capacities, and suffer from self-discharge. As an alternative storage solution, hydrogen can be used to store excess energy over longer periods and in larger quantities. This allows companies to store their excess energy from sun-rich months and production-free days and utilize it during sun-poor times. Optimizing self-sufficiency, especially through short- and long-term storage, reduces electricity costs, emissions, and infrastructure requirements of the public grid [12].

Hydrogen storage components, particularly electrolyzers (EL) and fuel cells (FC), are currently experiencing a market ramp-up. Existing systems are still expensive and, in combination with FC as power-to-power (P2P) energy storage systems, have relatively low efficiencies [13,14]. Optimizing the adaptation to a company’s specific needs is essential for offsetting the low efficiency and high costs of P2P systems, thereby ensuring economic operation.

The field of operations research plays a crucial role in supporting the dimensioning process. Dimensioning, which involves determining the appropriate size or scale of resources, facilities, or systems, is essential for economic efficiency as it directly affects costs, resource utilization, and overall profitability [15].

In addition to utilizing energy storage systems, generating energy flexibility significantly contributes to achieving higher energy self-sufficiency. Energy flexibilization of factories allows the adjustment of consumer-side energy demand to fluctuations in energy supply. This pertains to both the utilization of decentralized RES and adaptation to volatile energy prices from the public grid [16,17].

However, current approaches that integrate these concepts only address aspects individually: the sizing of PV in combination with BESS without long-term storage (especially in residential areas) or the flexibilization of energy consumption in larger energy-intensive enterprises. Owing to the lack of a multi-criteria approach, a decentralized energy supply, particularly supported by P2P systems, is rarely economical; therefore, the widespread implementation of effective emission reduction is still pending.

Based on the introduction, this review will address the following research questions:

• What optimization methods are currently used for:
  • the optimal system sizing of microgrids (OSS);
  • the optimization of electrical energy distribution to storage systems and consumers (EED); and
  • the energy flexibilization of factories (EF)?
• What is the scope of the functionality of the models and optimization algorithms of the respective research approaches?
• To what extent do research approaches that integrate these three areas into a common use case already exist?

The remainder of this paper is organized as follows. Section 2 provides the foundations of microgrids (MGs), EED, and EF, as well as an introduction to the simulation models and optimization methods for OSS. Section 3 outlines the methodological approach for a systematic literature review (SLR). Section 4 presents the results of the descriptive analysis and evaluates the findings in the three research areas. This includes a comprehensive textual and tabular representation of the current research efforts, research gaps, and the need for action. Finally, Section 5 discusses the results and concludes the paper.

2. Foundations of Microgrids, Energy-Flexible Factories, and Optimization Concepts

2.1. Introduction to Microgrids

Diverse definitions of the term microgrid (MG) exist. Primarily, an MG is defined as a decentralized energy supply system driven by RES, notably PV, hydropower, or WT, and energy storage systems, such as BESS, pumped storage systems, or P2P systems [18]. Some definitions are seen as systems that supplement the public grid supply while achieving a certain degree of autonomy. The public grid can compensate for any MG deficit at any time. The larger the energy storage capacity of an MG, the more energy it can store during peak periods instead of supplying it to the grid. This stored energy can be made available during phases of higher demand [18].

The PV, BESS, and P2P systems are particularly suitable for manufacturing SMEs. This suitability is attributed to their comparatively simple installation process, lower space consumption, and less complex regulatory requirements, in contrast to more complex systems such as WT, hydropower, and pumped storage systems [19].

Figure 1 illustrates the schematic configuration of an MG consisting of an alternating current (AC) bus and a direct current (DC) bus connected by an inverter. The AC bus links a decentralized energy source with the energy storage systems. Before power is fed into the public grid or factory, it must be converted through an inverter. The integration of a bidirectional inverter allows the MG to store power from the public grid in its energy storage systems during periods of low-cost electricity.

Factories can be generically categorized into n Products, n Processes, and n Resources. This generic categorization is based on the product-process-resource (PPR) notation proposed in [20].

2.2. Electrical Energy Distribution Strategies

Various EED strategies exist to control and distribute energy among storage systems, energy sources, consumers, and the public grid. A commonly encountered strategy is the conventional one. In the case of excess energy capacity, energy is first directed to the most efficient storage systems, then to less efficient ones, and finally to the grid. In the case of a power deficiency, the procedure is reversed: energy is initially drawn from the most efficient storage unit, then from the less efficient units, and finally from the grid. It is important to consider not only the absolute energy quantity (often expressed in kWh) but also the coverage of electrical power (commonly denoted in kW). Storage systems possess a limited capacity and constrained input/output power capabilities [21].

Another strategy is the peak shaving strategy, which focuses on reducing the maximum power demand from the grid. Consumers frequently incur annual grid connection fees contingent on peak power consumption within a given year. The peak shaving strategy aims to mitigate peaks in energy demand by providing short-term power from storage units [21,22].

Additional EED strategies include the direct purchase/long-duration strategy, where electricity from the grid can be used to charge the P2P system during low-cost periods. The strategies presented are based on decision trees. However, it is also feasible to optimally schedule the electrical power between components for a specific time step using a heuristic optimizer approach. Depending on the season, application, and other conditions, various strategies may be more or less effective, thereby reducing additional emissions and costs associated with the acquisition of optimal MG dimensions [21].

2.3. Economic and Sustainable Optimization Objectives

Operations research is a multidisciplinary field that employs mathematical modeling, statistical analysis, and optimization techniques to support decision-making processes, such as determining the optimal dimensions of systems within factories. It encompasses a broad array of methods and tools designed to enhance efficiency, productivity, and overall performance across various sectors, including manufacturing, finance, and energy systems [15].

In the field of OSS, it is crucial to optimize objectives, such as profits and emission reductions, by dimensioning the MG components. A commonly used method for calculating the economic feasibility of investment measures is the net present value (NPV), as shown in Equation (1). The NPV is a financial metric used to calculate the present value of future cash flow [23].

$$\mathrm{N}\mathrm{P}\mathrm{V}={\sum }_{t=0}^n{\displaystyle \frac{{\mathrm{O}\mathrm{p}\mathrm{E}\mathrm{x}}_t}{{\left(1+r\right)}^t}}-\mathrm{C}\mathrm{a}\mathrm{P}\mathrm{e}\mathrm{x}.$$

(1)

In this context, ${OpEx}_t$ denotes the operational expenditure costs, which comprise the operation and maintenance (O&M) costs, electricity costs, and all other cash flows at time t. Furthermore, r denotes the discount rate, and n is the observation period in years. CaPex stands for the initial capital expenditure. A positive NPV indicates that investment increases the value of the capital employed, whereas a negative NPV suggests that investment is worth less than cost. The NPV is commonly used to compare different investment opportunities and assess the profitability of a system or project [23].

The electricity market is complex and depends on the electricity provider, the country, and the size of the company. Broadly, tariffs can be categorized into those with fixed energy prices from the futures market, spanning up to several years, and those with dynamic time-of-use (ToU) energy prices from the spot market. While the prices in the futures market in Germany in 2023 are approximately 0.40 € for purchasing (gross price) and 0.06 € for selling per kWh of electricity, ToU tariffs are characterized by significant fluctuations in electricity prices based on availability and demand in the spot market. These fluctuations can range from −0.10 € at night to 0.30 € during the day (net price), with variations also influenced by seasonal factors. Through the day-ahead market (DAM), hourly electricity prices can be retrieved 24 h in advance [24].

The gross electricity price comprises the net electricity price, including network charges, taxes, and levies. For large customers, the annual maximum peak loads are calculated separately [24].

To calculate the CO₂ emissions of electric power, the emissions per kWh, typically ranging from 0.4 kg to 0.5 kg in Germany, can be utilized [25]. This factor is then multiplied by the overall power consumption to estimate the overall CO₂ emissions. Hourly data sources can be used for more detailed calculations of the carbon footprint of electric power generation. Sources such as [26] compute the hourly emission factor of power consumption, allowing for a more granular assessment of overall CO₂ emissions. In this context, CO₂ equivalents for other emissions, such as sulfur dioxide (SO₂) and nitrogen oxides (NO_X), can also be considered. Equation (2) calculates the overall CO₂ emissions from the public grid electricity consumption. In this context, the self-sufficiency ratio (SSR), which represents the proportion of energy sourced from the MG, was subtracted.

$$\mathrm{Overall} \mathrm{C}{\mathrm{O}}_2 \mathrm{Emissions}={\sum }_{t=0}^n\left(\mathrm{Power} {\mathrm{Consumption}}_t-\left(1-SSR\right)\right)\cdot \mathrm{Emissions} \mathrm{per}{\mathrm{k}\mathrm{W}\mathrm{h}}_{\mathrm{t}}$$

(2)

2.4. Energy Flexible Factories

In addition to the OSS and EED, energy flexibility offers the potential to adapt to volatile power availability. EF refers to the ability of energy-consuming units to adjust their energy demand quickly without compromising product quality. To harness this capability, various measures must be integrated into the operational production planning and control of manufacturing systems, such as adjusting individual production control parameters or the entire production schedule. These measures are largely related to demand response programs, which involve planned and short-term operational changes to alter the electrical load profile. Demand response may involve temporarily shutting down individual processes, thereby affecting the overall productivity of the production system [27]. These programs can be price-based, time-based, behavior-based, or technology-based and contribute to enhancing grid reliability, reducing costs, and integrating RES by harnessing demand flexibility [28].

Binder et al. [29] present an approach to modeling and flexibilizing complex factories based on the PPR notation. It is founded on the Reference Architecture Model for Industry 4.0 (RAMI 4.0) [30]. The PPR notation allows for the consideration and subdivision of complex scenarios from various perspectives. Resources are the focal point and represent the main entities in production, including machines, robots, or conveyors modeled within the plant hierarchy. Resources execute processes or handle products, establishing a connection with the other two views. Products represent manufactured goods and undergo various processes ranging from raw material handling to intermediate product manufacturing. Processes encompass production processes as well as all sub-processes within the process chain, including the joining, transportation, or assembly of various sub-products [20,29].

The flow-shop scheduling problem is a fundamental issue in production planning in which a sequence of jobs or processes must be processed through a predetermined sequence of resources. Each process passes through all machines in the same fixed order [31]. The flexible job-shop problem extends this problem by allowing for more flexible job processing. Jobs or processes can be assigned to various resources, enabling the consideration of various combinations of resources for processing each job. Despite this added flexibility, both problems are central to enhancing the efficiency of production processes and present challenging optimization issues. By incorporating the energy consumption of individual resources and processes, constraints can be applied to optimize scheduling, even when considering energy availability [31]. The two scheduling problems can be modeled using PPR notation. Figure 2 illustrates the differences between these two.

2.5. Basics of Simulation Models

A simulation model mathematically mirrors the crucial aspects of a system or process, enabling the forecasting of future behaviors under various conditions. A simulation model typically encompasses a period and features a predefined step size. The development of a simulation model involves defining the process or system to be analyzed, identifying relevant variables, and precisely describing their relationships [32,33]. Simulation models in the context of MGs enable decision-makers to tailor an MG to specific application scenarios and optimize EED and EF. An exemplary simulation model was developed in a previous study in which various sub-models, referred to as agents, acted autonomously and communicated with each other in a multi-agent manner [34].

In modeling, there are three main types of simulation models: deterministic models consistently produce the same results given identical initial conditions because they rely on fixed mathematical relationships without random components. These models are particularly useful when the system parameters are precisely known and no uncertainties are present. Dynamic models account for the temporal evolution of a system and describe how the system’s state changes over time. They are suitable for analyzing processes that evolve continuously or discretely over time and can be either deterministic or stochastic. Stochastic models incorporate random variables and uncertainties into the modeling process. Unlike deterministic models, stochastic models do not always yield the same results under identical initial conditions, owing to the inclusion of random influences. These models are valuable for understanding and representing the impacts of uncertainties and variabilities within a system [35].

2.6. Introduction to Metaheuristic Optimization

Heuristics are problem-oriented approaches that are used to quickly find solutions to complex problems, although they are not guaranteed to be optimal. They are often based on experience, intuition, or simple rules, typically leading to satisfactory solutions but not necessarily the best or optimal ones. Furthermore, metaheuristics are abstract, high-level, problem-solving approaches based on heuristic techniques. They provide a general framework for solving various types of problems, particularly mathematical optimization problems. Metaheuristics are often flexible, adaptive, and efficient in adapting to various types of problems [36].

Metaheuristic-based multi-objective optimization considering two objectives within the context of MGs (i.e., cost-effectiveness and sustainability) poses a complex optimization problem owing to the wide range of input parameters [21]. These parameters include the dimensions of the MG components and the parameters for configuring the EED and EF strategies. Given the multitude of parameter combinations, exact and mathematically guaranteed optimization algorithms are not feasible due to the immense solution space and computational overhead. Instead, metaheuristics can be employed to find near-optimal solutions within a few (several hundred) simulation runs [37]. A well-established metaheuristic inspired by evolutionary theory is the non-dominated sorting genetic algorithm II (NSGA-II), introduced by Deb et al. [38] in 2002. By selecting the best results from a population P of individual simulation runs and randomly combining these results, better outcomes were generated from generation to generation (Figure 3). The outcome is the Pareto-optimal front (in red), which consists of multiple Pareto-optimal solutions representing the optimal results for multiple objectives.

An individual A dominates an individual B if the first and/or second objective is greater than that of individual B [38], as shown in Equation (3).

$$A\prec B\Rightarrow A(\mathrm{O}\mathrm{b}\mathrm{j}. 1)>B(\mathrm{O}\mathrm{b}\mathrm{j}. 1)\parallel A(\mathrm{O}\mathrm{b}\mathrm{j}. 2)>B(\mathrm{O}\mathrm{b}\mathrm{j}. 2)$$

(3)

A decision-maker can individually select a compromise between objectives (i.e., cost-effectiveness and sustainability) [37].

In contrast, swarm-oriented metaheuristics such as particle swarm optimization (PSO) model the behavior of a swarm of particles within a search space. Each particle possesses a position and velocity that are iteratively adjusted based on experience, and the best position is determined within the swarm. PSO promotes cooperation among particles by effectively sharing information regarding the best solution discovered within the swarm. In contrast, evolutionary metaheuristics often generate new solutions through a genetic exchange between individuals. Evolutionary metaheuristics can be applied to various areas, such as layout design and dimensioning problems. On the other hand, PSO metaheuristics are frequently employed for continuous optimization problems and parameter adaptation in complex systems where the search for a global optimum is crucial, such as typical scheduling problems [37].

In addition, there are exact solvers, such as the mixed integer linear problem (MILP) solver, which, in the context of deterministic linear simulations, can find the global optimum rather than the near-global optimum, as in metaheuristic approaches [37].

3. Implementation of the Systematic Literature Review

The SLR will be conducted based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines for the identification, selection, and analysis of the literature. PRISMA was published in 2009 and received a comprehensive update in 2020. This method assists authors in documenting SLRs in a structured and transparent manner. It has been applied to over 60,000 instances across more than 200 journals, demonstrating its effectiveness [39].

The PRISMA was slightly adapted to handle three different search pathways for this review. It was divided into a comprehensive collection of literature (Phase A) and a selection of sources (Phase B), each comprising several steps. Figure 4 shows a flowchart detailing the entire process. Based on the research questions, keywords and synonyms were identified, and relevant academic databases were selected. The use of Boolean operators and the establishment of inclusion criteria optimize the collection in Phase A. The collected literature was first subjected to the screening of abstracts, followed by the screening of full texts in Phase B.

Furthermore, the PRISMA was supplemented with subsequent steps to elucidate the handling of the results. These results are presented in Section 4, where the filtered literature was subjected to descriptive analysis. This is followed by an individual analysis in which the various approaches are summarized and compared in tabular form. The following research gap analysis evaluates individual functionalities, optimization methods, and model properties using quantitative statistics. Finally, research gaps were derived, followed by the identification of the need for action.

3.1. Phase A—Comprehensive Collection of Literature

Phase A, which consists of three steps, describes the procedure for a comprehensive collection of literature in the three research areas of OSS, EED, and EF. For each research area, an SLR search path with appropriate keywords and restrictions was created (SLR OSS, SLR EED, and SLR EF). Scopus was used as the primary bibliographic database. It offers extensive coverage of international journals, advanced search capabilities, Boolean operators for precise searching, filtering options, and citation statistics analysis [40].

Step 1 consists of defining keywords by linking them using Boolean AND and OR operators, as depicted in Figure 4. The keywords for each area were systematically selected to encompass a broad range of synonyms, effectively capturing important aspects, components, methods, and functionalities.

• For the SLR OSS search path, synonyms were chosen in the areas of hydrogen storage, decentralized energy supply, PV, BESS, and P2P systems, and methods for optimizing the MG component dimensions.
• For the SLR EED search path, functionalities related to load shifting, demand management, flexibility, dynamic pricing, and optimization were included.
• For the SLR EF search path, keywords were formulated in the areas of flexibility, resource scheduling, production planning, energy management, and factory decarbonization.

In addition, keywords can be excluded by using the NOT operator. These apply equally to all three search paths and include various synonyms for the description of decentralized grids with different meanings, such as nano grids or mini grids. Furthermore, excluding keywords was formulated in the context of fully independent, mostly rural off-grid systems, followed by synonyms for MGs in private household areas. This also includes some storage technologies and RES, such as hydro-pumping storage systems and biogas plants, which are not suitable for most SMEs. Advanced terminology from electrical engineering was excluded to focus on economic optimization. Finally, terms from the field of electromobility were excluded to maintain a focus on stationary processes.

Advanced keyword searches in the Scopus database involve searching for terminologies in the title, abstract, keywords, and full text of the sources. All keywords were also formulated in German to encompass German-language sources.

In Step 2, the inclusion criteria were formulated and applied. To ensure a certain level of currency, only the sources from 2018 to 2023 were considered. The sources had to be written in either German or English and freely accessible through the university library of Ruhr-University Bochum in Germany. Additionally, the sources were required to demonstrate a certain level of scholarly relevance; thus, only peer-reviewed sources with a minimum of two citations per year were selected.

In Step 3, the collection of other sources outside of Scopus was examined. Dissertations were specifically searched in the catalog of the German National Library [41]. Furthermore, relevant projects were identified by screening all German Fraunhofer websites [42]. Sources were also identified through participation in conferences and trade fairs, as well as through expert interviews. Finally, a backward search was conducted, considering sources from 2016 onwards. In total, 1066 sources were identified in Phase A.

3.2. Phase B—Selection of Sources

Phase B, comprising two steps, outlines the procedure for the selection of sources through the screening of abstracts and the screening of full texts. The outcome of this selection process is a suitable set of included sources, which will be thoroughly analyzed in the results of Section 4.

In Step 1, for the screening of abstracts, the sources should consider the following criteria. They should address both short- and long-term storage and be connected to the public grid. The application should also be transferable on an industrial scale as well as to SMEs. Furthermore, the approach should be based on a simulation model. The focus is on heuristic optimization methods, excluding sources involving artificial intelligence for optimization. 88 sources remain after the abstract screening (SLR OSS: 30 sources, SLR EED: 19 sources, SLR EF: 39 sources).

In Step 2, the sources underwent detailed screening of full texts to ascertain compliance with additional filter criteria. Simulation models should be capable of modeling within a reasonable time interval of 1–60 min, to enable simulations spanning multiple years. Emphasis is placed on ensuring that the models do not delve into detailed electrical engineering models, thereby maintaining a focus on the techno-economic level. The source should also incorporate a methodological approach for optimization. Finally, additional relevant sources, among other dissertations, were identified in the forward search. 24 sources were remaining (SLR OSS: 10 sources, SLR EED: 8 sources, SLR EF: 6 sources).

Particularly in forward and backward searches, relevant sources can be included even if a criterion is not directly met or contains an excluded keyword.

4. Results

4.1. Descriptive Analysis

To identify specific trends within each research area, sources underwent a descriptive analysis organized by publication year and research domain, including the 88 sources after the abstract screening phase (Figure 5). This analysis precedes full-text screening to allow for clearer and more general trends to emerge from the examination of multiple sources. Full-text screening subsequently filters out sources that may not contribute to the identification of these overarching trends. It should be noted that sources may pertain to multiple areas but are categorized under a single primary research domain for focused analysis.

In the area of OSS, there has been a significant increase in the number of sources since 2020, indicating rising scientific relevance. Compared to 2017 and 2018, the number of sources increased four- to ten-fold from 2020 onwards. A similar trend was observed in the EED area in 2022, with an earlier peak in 2019. The trend in the EED area was less steep, with a four- to five-fold increase between 2017 and 2023. In contrast, the number of sources in the EF area has remained relatively constant over the years, ranging from four to six sources per year, with a peak of eight sources in 2021. Overall, a clear upward trend in the relevance of these scientific areas is evident across all three areas.

In the context of RES, particularly PV, its effectiveness is significantly influenced by the climate and latitudinal positioning of the site. A further descriptive analysis examined the locations of the case studies from the selected sources (Figure 6). Given the axial tilt of the Earth, latitudes closer to the poles experience greater seasonal variations in sunlight availability. This analysis focused exclusively on sources identified through full-text screening to ensure precise identification of case study locations. It is observed that in temperate latitudes, compared to subtropical and tropical regions, less research is conducted on OSS and EED but more on EF, both in relative and absolute terms.

Considering all three research areas, the number of case studies is highest in Asian locations with eleven sources (five of which are from China) and in European locations with ten sources (six of which are from Germany). When exclusively examining the OSS research area, countries in subtropical and tropical latitudes, such as those in Asian locations, are relatively more represented with 55% sources compared to their total number of sources. In the EED area, Asia stands out with five sources, whereas Europe dominates the EF area with more than five sources.

After full-text screening, a total of 24 sources were included. The journals most represented were Applied Energy, with six sources, and Energy and the International Journal of Energy Research, with three sources each.

4.2. Individual Analysis

The following section provides an individual analysis of the filtered sources. Each SLR path is given its own subsection, which first presents a summary of the commonalities and differences of all approaches and then discusses the different approaches individually, followed by a comparative tabular overview.

4.2.1. Analysis of the Research Area of the Optimal System Sizing of Microgrids

In the following section, ten sources focusing on the OSS are presented (Table 1). All referenced MGs met the filter criteria of being grid-connected and including PV systems as decentralized RES. In addition, the MGs in [43,44,45,46,47] incorporated WT. All approaches combine BESS and P2P, whereas [21,43,48] consider BESS and P2P individually.

All approaches utilized a simulation time step of one hour and covered a simulation period of several years. Most models are characterized by deterministic or dynamic modeling paradigms, whereas [45] employs stochastic modeling paradigms. The focus is on optimizations during the design phase, sometimes coupled with the operations phase.

All approaches consider the NPV as a measure of total costs. The NPV often includes electricity costs, except for [44,47]. Refs. [43,46] consider not only the purchase price but also the fixed electricity selling price. The remaining approaches typically use ToU pricing for buying and selling electricity. The studies [21,48,49] included additional differentiated pricing components, such as variable and fixed grid connection fees and the pricing of the DAM. Furthermore, ref. [50] considered a general forecast for the increase in electricity prices over the next 25 years.

Sustainability models have not been considered as comprehensively as cost models. With two exceptions, the authors accounted for CO₂ emissions only indirectly through the SSR. Only [44] multiplied the purchased electricity by a fixed value for CO₂ emissions per kWh of grid electricity. In [43], the authors uniquely considered not only CO₂ emissions but also other greenhouse gases, multiplying these by a fixed factor for future social costs.

The selection of metaheuristics for OSS varies between evolutionary and swarm-based approaches. Authors employing multi-objective approaches consistently choose NPV and SSR as objectives [43,50], whereas [21,45,46,47,48,49] include the load loss ratio (LLR) as a third objective. The LLR represents the reliability of the MG, with a lower LLR indicating a more frequent reliance on the public grid. The remaining authors used a single-objective approach, focusing on the levelized cost of electricity (LCOE) per kWh. The authors of [49,51] used MILP solvers instead of metaheuristics.

All sources in the OSS initially utilized the conventional strategy for EED. Refs. [21,45,48,49,50,51] supplemented this with additional sub-strategies or developed entirely novel approaches. Only [48,49,50] also considered optimizing the EED using optimization algorithms, with source [21] uniquely addressing EF.

All authors examined scenarios in their respective case studies that can be applied to the dimensions of SMEs, considering warehouses, larger residential complexes, university buildings, and ports.

The following section presents the individual approaches in detail and highlights their unique features.

Ya. Zhang et al. [48] utilized an evolutionary-based genetic algorithm for OSS in their publication. They introduced a hybrid peak shaving strategy, which smooths out peak electricity demands through the targeted use of stored energy. When certain thresholds are reached, such as a specific hydrogen storage level or a particular season, the system switches from a conventional strategy to a peak shaving strategy. These thresholds can be optimized using a genetic algorithm as additional input parameters alongside the OSS.

Their peak shaving strategy is based on the MILP approach. At each time step in the simulation, a MILP solver was used to optimize the energy flows. The authors emphasized the use of EED strategies to further improve the economic performance of MGs and to tailor the OSS more precisely to specific application scenarios. Their approach is particularly suited to the seasonal fluctuations found in temperate European latitudes.

In his dissertation, Ya. Zhang [21] expanded his optimization approach from [48] by integrating EF. He considered the possibility of flexibilizing energy consumption as an additional condition in his MILP approach. The MILP solver can manage energy not only in and out of storage but also adjust the energy distribution to certain consumers through load shifting, provided that these consumers (regulated by constraints) allow for such adjustments within a specified timeframe. Although load shifting is only a theoretical consideration in his model, as no actual factories with their consumers are modeled, these constraints provide the MILP solver with some degree of flexibility for the individual energy consumers. Additionally, he presented an energy management system that enables a PV generation forecast, planning the next 24 h using the MILP Solver.

Akhavan Shams and Ahmadi [43] examined significantly more southerly latitudes in Iran, in contrast to [48]. Their EED relied solely on a conventional algorithm, and their deterministic simulation model was generally less complex than that in [48], which incorporated dynamic elements. In their case study, seasonal fluctuations in solar power availability were significantly smaller. However, unlike other authors, they assessed greenhouse gas emissions based on long-term social costs. For instance, in their model, the social costs for one ton of CO₂ were USD 2.86, SO₂ USD 521.5, and NO_X USD 171.5. Overall, this added an extra USD 0.19 per kWh of electricity to the initial cost of USD 0.05, given the Iranian mix.

Owing to the low energy costs of electricity, MGs in Iran were economically viable only when such social costs were considered, and the costs of the MG components further decreased.

Singh et al. [44], in contrast to [21,43,48], used a single-objective metaheuristic aimed at reducing the LCOE. The authors developed a novel metaheuristic based on the Artificial Bee Colony (ABC) algorithm, which drew upon the concept of collective intelligence within a bee swarm. This method allowed bees to communicate with each other to identify the optimal location for nectar, thereby achieving an optimal swarm configuration. The innovation in their approach was the incorporation of the predator effect, which enabled bees to locate an optimal nectar site more rapidly, thereby reducing the risk of entrapment in a local optimum. Benchmarking indicated that this algorithm performed marginally better than conventional algorithms, such as PSO.

Their case study was set in India, at a location with abundant sunlight throughout the year and relatively low energy costs. The authors primarily pursue an off-grid solution but also indicate that it is more economically viable if public grid electricity can occasionally be used.

Yi Zhang et al. [45] developed an MG simulation model based on stochastic paradigms. For time-varying variables, they used probability distributions instead of time series values, such as the beta distribution for PV yields and the Weibull distribution for wind energy yields. They also used a Gaussian distribution for energy consumption at a production site as well as a distribution for ToU electricity prices. They employed a slightly modified NSGA-II metaheuristic that outperformed other metaheuristics in benchmarking.

Crespi et al. [51] employed a single-objective MILP solver from Gurobi [52] to minimize the LCOE. Their analysis period spanned one year, although multiple years were also feasible. The LCOE incorporated costs such as NPV, replacement, and operational costs of components, which are proportionally scaled down on an annual basis. As in [48], they considered a strategy in which events trigger a switch from a conventional EED strategy to another. During periods of favorable ToU energy costs, hydrogen was generated directly through the power grid using a grid charging strategy. In addition, they addressed the start-up times of P2P systems to minimize frequent cycling in their strategy. Their optimization algorithm included thresholds as input parameters for optimizing their EED strategy, along with OSS.

Their case study aimed to sustain a continuous load of 1 MW through the MG. Through benchmarks, they determined that achieving an SSR of 40% necessitated BESS, particularly at the temperate latitudes of northern Italy. According to their study, an SSR exceeding 90% requires P2P utilization.

In a subsequent publication, building on their work in [51], Crespi et al. [49] investigated the DAM and ancillary service market (ASM) for further economic optimization. The analysis considered diverse options regarding electricity prices on the DAM and price-quantity combinations of accepted bids on the ASM. The greater the fluctuation in the electricity market, the greater the advantages of a P2P system, as it can sell energy asymmetrically in more favorable terms.

The provision of secondary frequency reserve services with an optimal bidding strategy and asymmetric bids enabled a reduction in average electricity costs by up to 15% compared to operating solely on the DAM. The goal was to decouple energy consumption and supply through storage, allowing for the advantageous purchasing and selling of energy. This became particularly relevant when solar energy is not available during nighttime and seasonal variations, as observed in European countries, where daily PV energy availability can vary by a factor of three between summer and winter.

Chen et al. [46] utilized a multi-objective PSO metaheuristic (MOPSO) in their study, focusing on the electricity and hydrogen supply for a port in China. Unlike other authors, the hydrogen produced can directly power mobile port equipment. They compared their MOPSO approach with the optimization software HOMER (Version 3.16.2) [53], which is considered the standard for OSS. A significant drawback of HOMER is its capability to perform only single-objective optimization. Therefore, their MOPSO approach delivers superior results, especially concerning sustainability aspects.

Xing et al. [47] introduced an approach in which a higher-level heuristic tunes the hyperparameters of an NSGA-II algorithm for OSS. They emphasized the crucial role of hyperparameter tuning in the quality of heuristic results. In this case, a MOPSO algorithm within an interleaved algorithm optimizes the two hyperparameters, crossover rate and probability rate for variation, to tune the NSGA-II. They referred to their approach as PSO and NSGA-II.

The methodology behind this dual optimization begins by applying NSGA-II to an optimization problem to generate a Pareto-optimal front. A sensitivity analysis identifies the most influential hyperparameters of the NSGA-II using the obtained Pareto-optimal front as a benchmark. Subsequently, the hyperparameters were optimized using MOPSO, again referencing the Pareto-optimal front as a benchmark. The NSGA-II can be effectively refined by preserving the optimal hyperparameters.

Le et al. [50] introduced a novel multi-objective firefly algorithm (MOMFA), inspired by the principles of PSO and the behaviors and light emission characteristics of fireflies, where the brightness of one firefly attracts others in its vicinity. They conducted benchmarks for their MOMFA metaheuristics and compared them with the NSGA-II. The MOMFA showed only marginally better robustness. They emphasized that NSGA-II can also achieve satisfactory results, but with more frequent simulation runs.

Furthermore, they developed a novel EED strategy comprising a conventional approach and an optimized long-term storage strategy that switches seasonally, bearing strong similarities to the grid charging strategy. Variable input parameters, such as start times for winter and summer months and thresholds for storage capacities, were simultaneously optimized alongside the OSS within their metaheuristic framework. If storage levels fall below a minimum and current electricity market prices drop below a specified threshold, hydrogen storage is supplemented directly from the grid beyond the PV inputs. This strategy yielded slightly better economic outcomes, particularly at temperate latitudes.

Table 1. Overview of the different SLR OSS approaches.

Source	Ya. Zhang et al. [48]	Ya. Zhang [21]	Akhavan Shams et al. [43]	Singh et al. [44]	Yi. Zhang et al. [45]	Crespi et al. [51]	Crespi et al. [49]	Chen et al. [46]	Xing et al. [47]	Le et al. [50]
OSS/EED/EF	●/◑/○	●/◑/◑	●/○/○	●/○/○	●/○/○	●/◑/○		●/○/○	●/○/○	●/◑/○
GRID/PV/WT	●/●/○		●/●/●	●/●/○	●/●/●	●/●/○		●/●/●	●/●/●	●/●/○
BESS/P2P	●/●		●/●	●/●	●/●	●/●	○/●	●/●	●/●	●/●
Model Type	Dynamic		Deterministic	Dynamic	Stochastic	Deterministic		Deterministic	Deterministic	Dynamic
Steps; Period	1 h; 25 a		1 h; 20 a	1 h; 20 a	1 h; 1 a	1 h; 1 a		1 h; 6 a	1 h; 30 a	1 h; 25 a
Stage	Design, Operation		Design	Design	Design	Design, Operation		Design, Operation	Design	Design, Operation
Cost Model	NPV		NPV	NPV	NPV	NPV		NPV	NPV	NPV
Electricity Model	ToU, DAM, Fees		Fixed	○	ToU (Buying)	ToU	ToU, DAM	Fixed	○	ToU
Sustainability Model	SSR		Social Costs	CO2 per kWh	SSR	SSR		SSR	SSR	SSR
Simulation Functionalities	○	* CHP	○	○	○	Ramp-Up		H2 Sale	○	Degradation
Stochastics	○	○	○	○	○	○		○	○	○
Metaheuristic	Genetic Algorithm		Genetic Algorithm [OSS]	ABC-PO [OSS]	NSGA-II [OSS]	○		MOPSO [OSS]	PSO and NSGA-II [OSS]	MOMFA [OSS/EED]
	* [OSS/EED]	[OSS/EED/EF]
Solver	MILP		○	○	○	MILP [OSS/EED]		○	○	○
	[EED]	[EED/EF]
Objectives	NPV, SSR, LLR		NPV, SSR	LCOE	NPV, SSR, LLR	LCOE	NPV, SSR, LLR	NPV, SSR, LLR	NPV, SSR, LLR	NPV, SSR
Optimizer Functionalities	MILP Integrated		○	○	○	○		○	Hyperparameter Tuning	○
EED Strategy	Conventional, Peak Shaving		Conventional	Conventional	Conventional, Grid Charging	Conventional, Grid Charging		Conventional	Conventional	Conventional, Grid Charging
EF Strategy	○	Load Shifting	○	○	○	○		○	○	○
Case Study	Multi-Apartment Building		University Building	University Building	○	1 MW Load	Factory	Port	Community	Warehouse
Location	Sweden		Iran	India	China	Italy		China	China	Vietnam

●—considered; ◑—partly considered (Each cited source has a specific focus (OSS/EED/EF). If an additional, less pronounced consideration of another area is included, it is rated as partly considered; ○—not considered; * CHP—Combined heat and power consideration; * [OSS/EED/EF]—indicates the research areas optimized by the metaheuristic or solver.

Table 1. Overview of the different SLR OSS approaches.

Source	Ya. Zhang et al. [48]	Ya. Zhang [21]	Akhavan Shams et al. [43]	Singh et al. [44]	Yi. Zhang et al. [45]	Crespi et al. [51]	Crespi et al. [49]	Chen et al. [46]	Xing et al. [47]	Le et al. [50]
OSS/EED/EF	●/◑/○	●/◑/◑	●/○/○	●/○/○	●/○/○	●/◑/○		●/○/○	●/○/○	●/◑/○
GRID/PV/WT	●/●/○		●/●/●	●/●/○	●/●/●	●/●/○		●/●/●	●/●/●	●/●/○
BESS/P2P	●/●		●/●	●/●	●/●	●/●	○/●	●/●	●/●	●/●
Model Type	Dynamic		Deterministic	Dynamic	Stochastic	Deterministic		Deterministic	Deterministic	Dynamic
Steps; Period	1 h; 25 a		1 h; 20 a	1 h; 20 a	1 h; 1 a	1 h; 1 a		1 h; 6 a	1 h; 30 a	1 h; 25 a
Stage	Design, Operation		Design	Design	Design	Design, Operation		Design, Operation	Design	Design, Operation
Cost Model	NPV		NPV	NPV	NPV	NPV		NPV	NPV	NPV
Electricity Model	ToU, DAM, Fees		Fixed	○	ToU (Buying)	ToU	ToU, DAM	Fixed	○	ToU
Sustainability Model	SSR		Social Costs	CO2 per kWh	SSR	SSR		SSR	SSR	SSR
Simulation Functionalities	○	* CHP	○	○	○	Ramp-Up		H2 Sale	○	Degradation
Stochastics	○	○	○	○	○	○		○	○	○
Metaheuristic	Genetic Algorithm		Genetic Algorithm [OSS]	ABC-PO [OSS]	NSGA-II [OSS]	○		MOPSO [OSS]	PSO and NSGA-II [OSS]	MOMFA [OSS/EED]
	* [OSS/EED]	[OSS/EED/EF]
Solver	MILP		○	○	○	MILP [OSS/EED]		○	○	○
	[EED]	[EED/EF]
Objectives	NPV, SSR, LLR		NPV, SSR	LCOE	NPV, SSR, LLR	LCOE	NPV, SSR, LLR	NPV, SSR, LLR	NPV, SSR, LLR	NPV, SSR
Optimizer Functionalities	MILP Integrated		○	○	○	○		○	Hyperparameter Tuning	○
EED Strategy	Conventional, Peak Shaving		Conventional	Conventional	Conventional, Grid Charging	Conventional, Grid Charging		Conventional	Conventional	Conventional, Grid Charging
EF Strategy	○	Load Shifting	○	○	○	○		○	○	○
Case Study	Multi-Apartment Building		University Building	University Building	○	1 MW Load	Factory	Port	Community	Warehouse
Location	Sweden		Iran	India	China	Italy		China	China	Vietnam

●—considered; ◑—partly considered (Each cited source has a specific focus (OSS/EED/EF). If an additional, less pronounced consideration of another area is included, it is rated as partly considered; ○—not considered; * CHP—Combined heat and power consideration; * [OSS/EED/EF]—indicates the research areas optimized by the metaheuristic or solver.

4.2.2. Analysis of the Research Area for the Optimization of Electrical Energy Distribution to Storage Systems and Consumers

Below, eight approaches in the SLR EED are presented (Table 2), with approaches [54,55] additionally incorporating EF through load shifting strategies, at least partially. All MGs included both PV and WT, except for [54,56]. Other than the approaches [54,57], BESS and P2P were considered. These approaches use P2P storage to replace BESS as both short- and long-term storage solutions.

Most approaches used a simulation time step of one hour, except for [54], which utilized a time step of 15 min. The simulation period typically spans 24 h. Only sources [54] with two weeks and [58] with over a year employed longer simulation periods. This is because of the need to calculate a greater number of parameters, specifically charging and discharging rates per time step, compared to approaches in OSS. Except for [59], all approaches employed either an exact solver or a metaheuristic to adjust the hourly power values. However, rule-based approaches, such as conventional or grid charging strategies, are more frequently used for optimization over longer periods, spanning several years. This becomes evident when comparing Table 1 and Table 2 regarding the employed EED strategy with the period.

In most cases, simulation models were deterministic or dynamic. Only [55] implemented stochastic approaches in their model, where the probability distributions for the electrical load, electricity prices, and uncertainties of RES are considered. The focus is exclusively on optimizations during the operational phase rather than the design phase.

In contrast to the consistent use of NPV as a cost model in SLR OSS, SLR EED primarily considered only O&M costs. The sources [54,56,60] additionally accounted for the ramp-up costs of P2P systems, while [56] also included CapEx. Furthermore, ref. [54] considered the costs associated with network connections for the peak loads. Only [58] incorporated NPV. Source [56] is the only one that considers costs for the LLR.

All approaches consider the dynamic buying and selling ToU costs of electricity. The approaches in [54,55] support DAM. Except for [60,61], all approaches consider CO₂ emissions per kWh of electricity from the public grid, with some sources including NO_X and SO₂ values.

Most sources use OpEx or LCOE as objectives, with [55,56,57] considering emissions as an additional objective and [54] also taking into account the maximum peak load. The simulation models do not have a fixed decision tree-based algorithm for their storage strategy but are controlled within the framework of metaheuristic optimization. Only [59] employs a conventional EED strategy because it does not utilize metaheuristics.

As case studies, most models assume a generic load that can represent various use cases. Only [54,56] specify their models based on a university or commercial building in Germany and Egypt, respectively. Source [59] examines a community in Malaysia.

Jaramillo and Weidlich [54] presented a multi-objective MILP approach for EED in their MG. Unlike most other authors, they employed an EL with slower alkaline technology, which requires up to 45 min to ramp up rather than a few minutes. To avoid frequent shutdowns, they incorporated ramp-up times and costs into their cost calculations. They also used the public grid to bridge short-term electricity deficits in their MG. In their case study, they utilized load profiles consisting of fixed and variable components. The fixed component was a standard load profile for an office building [62], combined with a flexible component in the form of an electric vehicle charging station (E-charging).

The MILP optimization algorithm was capable of planning, in 15-min intervals, not only the charging and discharging performances of the storage systems but also the energy consumption of the electric vehicle within certain constraints, thereby achieving a degree of energy flexibility. However, owing to the numerous configuration parameters for the power flows of storage systems in a 15-min interval, their algorithm was designed for a period of only two weeks. However, they emphasized that longer periods could be realized through a rolling-horizon approach, as demonstrated in [58]. In particular, a seasonal perspective over the course of a year is of interest to P2P. Additionally, they highlighted that the OSS significantly affects economic viability.

Khan et al. [59] based their approach on a multi-agent system, which comprised several autonomous agents, each responsible for managing a specific part of the MG. They distinguished between critical and non-critical loads. Their optimization algorithm included three main components: (a) prediction of energy generation and consumption based on forecast models utilizing historical data, (b) optimization of EED achieved through a simulation model, and (c) monitoring and control of the entire system. They presented an algorithm for energy distribution without optimizing the objectives using heuristics or linear solvers.

Shahryari et al. [55] considered uncertainties in market prices and loads, as well as WT and PV generation, integrating these into their EED strategy. They optimized their EED using an improved incentive-based demand response program to facilitate load shifting. For this purpose, they employed the multi-objective group search optimization (MOGSO) method to optimize two objectives: OpEx costs and emissions.

They developed an uncertainty modeling method that accounted for uncertainties related to wind and solar energy, ToU prices, and loads. The distribution function was created by overlaying the actual and forecast values.

Ruiming [57] presented an optimization approach for optimal day-ahead power dispatching, considering the operational and environmental aspects. He applied an improved NSGA-II to optimize the performance of EL and FC for each hour within a 24-h horizon. The two objectives pursued were the minimization of O&M costs and emissions. A Pareto-optimal Front is generated hourly.

Additionally, he developed an interactive search strategy to combine the results of the 24 Pareto-optimal Fronts in a manner that required minimal adjustments to the performance of EL and FC. His simulation model considered the degradation and ramp-up times of P2P systems. In a benchmark comparison, he demonstrated that his improved NSGA-II provided slightly better results than standard NSGA-II.

Mosa et al. [56] utilize the branch-and-reduce optimization navigator (BARON) solver to address a pseudo-multi-objective problem. Initially, the multi-objective problem was transformed into a single-objective problem by adding additional constraints to the original problem. The concept of the constraint approach is illustrated by combining it with BARON to generate a Pareto-optimal front between minimal cost and minimal emission functions.

Cambambi et al. [61] considered cycle costs and degradation in their O&M costs as part of their EED strategy. Their EED was based on the availability of energy and economic signals that were incorporated into a MILP solver approach. They used a rule-based, conventional hybrid strategy with fixed values. When the economic signals indicated that inexpensive energy could be drawn from the public grid and the energy storage systems were between 20% and 80% state-of-charge, they were charged. Conversely, energy was drawn from the storage systems when electricity could be sold at a high price.

Vaish et al. [60] investigated the optimization of BESS charging and discharging processes as well as cost optimization. They directly compared several (eight) physics-based metaheuristics as opposed to genetic algorithms. These include, for instance, the lightning search algorithm and black hole optimization.

Guo et al. [58] introduced a data-driven rolling-horizon algorithm for EED. The algorithm forecasts the future states of BESS and P2P storage systems using a two-stage approach that leverages reference curves for state-of-charge and load profiles from previous years. The fundamental idea is to decompose long-term operation into hourly sub-tasks and establish reference values for the P2P hydrogen level based on optimal storage level curves derived from historical scenarios. The approach analyzes the similarity between the current year and past years and subsequently formulates a multi-objective rolling-horizon optimization problem that balances cost reduction with adherence to reference storage level values.

Guo et al.’s approach is well suited for predictive synchronization between the time-bound availability of RES, forecasted energy storage levels, and projected load profiles. The data-driven approach provides predictive insights into short- and long-term storage states.

By utilizing historical data on renewable energy and loads, the scheduling system generates ex-post optimal state-of-charge sequences for seasonal energy storage for the upcoming operational year. During each real-time operation period, the correction module executes two steps: first, updating the reference state-of-charge for seasonal storage based on the ex-post optimal state-of-charge sequences and newly observed data; second, solving a bi-objective rolling-horizon optimization problem that minimizes current operating costs while steering the state-of-charge of seasonal storage towards its reference value.

Table 2. Overview of the different SLR EED approaches.

Source	Jaramillo et al. [54]	Khan et al. [59]	Shahryari et al. [55]	Ruiming [57]	Mosa et al. [56]	Cambambi et al. [61]	Vaish et al. [60]	Guo et al. [58]
OSS/EED/EF	○/●/◑	○/●/○	○/●/◑	○/●/○	○/●/○	○/●/○	○/●/○	○/●/○
GRID/PV/WT	●/●/○	●/●/●	●/●/●	●/●/●	●/●/○	●/●/●	●/●/●	●/●/●
BESS/P2P	○/●	●/●	●/●	○/●	●/●	●/●	●/●	●/●
Model Type	Deterministic	Dynamic	Stochastic	Dynamic	Dynamic	Deterministic	Deterministic	Dynamic
Steps; Period	15 min; 2 wk	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 a
Stage	Operation	Operation	Operation	Operation	Operation	Operation	Operation	Operation
Cost Model	O&M, Peak Load Fees,	O&M	O&M	O&M	O&M, CaPex, Load Loss	O&M	O&M	NPV
Electricity Model	ToU, DAM	ToU	ToU (buying), DAM	ToU	ToU	ToU	ToU	ToU
Sustainability Model	CO2 per kWh	CO2, NOX, SO2 per kWh	CO2, NOX, SO2 per kWh	CO2 per kWh	CO2, NOX, SO2 per kWh	○	○	CO2 per kWh
Simulation Functionalities	Standard Load Profile, Ramp-Up	Multi-Agent Approach	DAM	Degradation, Ramp-Up	DAM, PV Load Forecast, Ramp-Up	Degradation	CHP, Ramp-Up	SOC Forecast
Stochastics	○	○	RES, ToU, Loads	○	○	○	○	○
Metaheuristic	○	○	MOGSO [EED/EF]	NSGA-II [EED]	○	○	Physic-Based [EED]	Rolling Horizon [EED]
Solver	MILP [EED/EF]	○	○	○	BARON [EED]	MILP [EED]	○	MILP [EED]
Objectives	OpEx, Peak Loads, Emissions	○	OpEx, Emissions	OpEx, Emissions	OpEx	OpEx	LCOE	NPV, SSR
Optimizer Functionalities	○	○	Stochastic Modeling	Interactive Search	○	○	○	Data-Driven Scheduling
EED Strategy	Optimizer, Peak Shaving, Grid Charging	Conventional	Optimizer	Optimizer	Optimizer	Optimizer	Optimizer	Optimizer
EF Strategy	E-Charging	○	Load Shifting	○	○	○	○	○
Case Study	University Building	Community	Generic	Generic	Multi-Apartment Building	Generic	Generic	Generic
Location	Germany	Malaysia	Iran	China	Egypt	Brazil	India	China

●—considered; ◑—partly considered; ○—not considered.

Table 2. Overview of the different SLR EED approaches.

Source	Jaramillo et al. [54]	Khan et al. [59]	Shahryari et al. [55]	Ruiming [57]	Mosa et al. [56]	Cambambi et al. [61]	Vaish et al. [60]	Guo et al. [58]
OSS/EED/EF	○/●/◑	○/●/○	○/●/◑	○/●/○	○/●/○	○/●/○	○/●/○	○/●/○
GRID/PV/WT	●/●/○	●/●/●	●/●/●	●/●/●	●/●/○	●/●/●	●/●/●	●/●/●
BESS/P2P	○/●	●/●	●/●	○/●	●/●	●/●	●/●	●/●
Model Type	Deterministic	Dynamic	Stochastic	Dynamic	Dynamic	Deterministic	Deterministic	Dynamic
Steps; Period	15 min; 2 wk	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 d	1 h; 1 a
Stage	Operation	Operation	Operation	Operation	Operation	Operation	Operation	Operation
Cost Model	O&M, Peak Load Fees,	O&M	O&M	O&M	O&M, CaPex, Load Loss	O&M	O&M	NPV
Electricity Model	ToU, DAM	ToU	ToU (buying), DAM	ToU	ToU	ToU	ToU	ToU
Sustainability Model	CO2 per kWh	CO2, NOX, SO2 per kWh	CO2, NOX, SO2 per kWh	CO2 per kWh	CO2, NOX, SO2 per kWh	○	○	CO2 per kWh
Simulation Functionalities	Standard Load Profile, Ramp-Up	Multi-Agent Approach	DAM	Degradation, Ramp-Up	DAM, PV Load Forecast, Ramp-Up	Degradation	CHP, Ramp-Up	SOC Forecast
Stochastics	○	○	RES, ToU, Loads	○	○	○	○	○
Metaheuristic	○	○	MOGSO [EED/EF]	NSGA-II [EED]	○	○	Physic-Based [EED]	Rolling Horizon [EED]
Solver	MILP [EED/EF]	○	○	○	BARON [EED]	MILP [EED]	○	MILP [EED]
Objectives	OpEx, Peak Loads, Emissions	○	OpEx, Emissions	OpEx, Emissions	OpEx	OpEx	LCOE	NPV, SSR
Optimizer Functionalities	○	○	Stochastic Modeling	Interactive Search	○	○	○	Data-Driven Scheduling
EED Strategy	Optimizer, Peak Shaving, Grid Charging	Conventional	Optimizer	Optimizer	Optimizer	Optimizer	Optimizer	Optimizer
EF Strategy	E-Charging	○	Load Shifting	○	○	○	○	○
Case Study	University Building	Community	Generic	Generic	Multi-Apartment Building	Generic	Generic	Generic
Location	Germany	Malaysia	Iran	China	Egypt	Brazil	India	China

●—considered; ◑—partly considered; ○—not considered.

4.2.3. Analysis of the Research Area of the Energy Flexibilization of Factories

The following six sources focus on EF (

Table 3. Overview of the different SLR EF approaches.

Source	Fazli Khalaf and Wang [31]	Caro-Ruiz et al. [64]	Lombardi et al. [65]	Beier et al. [67]	Wanapinit et al. [66]	Küster et al. [63]
OSS/EED/EF	○/◑/●	◑/◑/●		○/○/●	◑/◑/●	○/◑/●
GRID/PV/WT	●/●/●	○/●/○	●/●/●	●/●/●	●/●/●	●/●/○
BESS/P2P	●/○	●/○	●/●	●/○	●/○	●/○
Model Type	Stochastic	Deterministic	Dynamic	Dynamic	Deterministic	Dynamic
Steps; Period	5 min; 1 d	1 h; 1 a		1 s; 1 h	5 min; 1 wk	1 s; 1 d
Stage	Operation	Design and Operation		Operation	Design and Operation	Operation
Cost Model	Electricity Costs	○		Electricity Costs	Electricity Costs	Electricity Costs
Electricity Model	ToU, DAM	○		Peak Fees	ToU, DAM	ToU
Sustainability Model	○	SSR		SSR; CO2 per kWh	○	SSR
Simulation Functionalities	○	BESS/Buffer Sizing Methodology, PPR Modeling		Throughput times, PPR Modeling, Multi-Agents, CO2 per product	PPR Modeling, Flexibilization Analysis, CHP	PPR Modeling
Stochastics	RES	○		○	○	○
Metaheuristic	○	○		○	○	Genetic Algorithm [EF]
Solver	SMILP [EED/EF]	MILP [EED/EF]	○	○	MILP [EED/EF]	○
Objectives	Electricity Costs	SSR	○	SSR	Electricity Costs	Fitness Function
Optimizer Functionalities	2 Stages: Deterministic and Stochastic	○		○	○	Scheduling Processes as Alleles
EED Strategy	Optimizer	Optimizer	Conventional	Conventional	Optimizer	Conventional
EF Strategy	Flow-Shop	Flow-Shop		Flow-Shop	Flexible Job-Shop	Flexible Job-Shop
Case Study	Factory	Factory		Factory	Factory	Factory
Location	USA	Germany		Germany	Germany	Germany

●—considered; ◑—partly considered; ○—not considered.

). Sources [31,63] also considered EED, and [64,65,66] partially incorporated OSS. All sources included PV in their MG, except for [64], and were connected to the public grid. Additionally, all except [63,64] integrate WT as an additional RES.

The simulation time steps varied significantly. Sources [64,65], similar to most sources in other SLR paths, use a time step of one hour and a simulation period of one year. However, in SLR EF, it is more common to simulate with a time step of one second over a period of one hour [67] or one day [63]. Furthermore, refs. [31,66] were simulated in 5-min intervals over a period of one day or one week.

Only [31] employed the stochastic modeling of RES, whereas others used either deterministic or dynamic modeling paradigms. The focus lay on optimizations during the operations phase, sometimes coupled with the design phase.

Except for [64,65], these approaches consider only electricity costs in their cost model, aiming to keep the throughput times of their products consistent, despite flexibility measures.

In their electricity models, refs. [31,63,66] accounted for dynamic ToU for both purchase and sale, with [31,66] additionally considering DAM. Source [65] independently examined the additional charges associated with peak loads.

Most approaches utilized only SSR in their sustainability models, whereas [65] also incorporated CO₂ emissions.

Only [63] employed a metaheuristic optimizer, whereas [31,64,65,66] used MILP solvers. The primary objectives were typically SSR, or electricity cost. The EED strategy was either an optimizer or a conventional strategy.

For the EF strategy, the flow-shop scheduling problem was considered, whereas [63,66] addressed the flexible job-shop scheduling problem.

All case studies were implemented in factories, predominantly in the USA and Germany, and are thus situated in temperate latitudes, except for [31].

The following sections detail each approach and highlight its unique features.

Fazli Khalaf and Wang [31] employed a MILP solver with stochastic elements (SMILP) to determine the most favorable electricity costs in a single-objective problem. The first stage of their two-stage flow-shop model addresses the flow-shop problem by considering the DAM and forecasted RES, thereby generating optimal work schedules and the optimal electricity demand curve.

In the second stage, the dynamics of ToU electricity prices and RES variability are considered for deviations in the forecasted renewable supply. The model aims to minimize the total cost of the real-time delivered electricity provided by the electricity market and maximize the SSR.

Each machine had a specific processing time and energy consumption. According to the flow-shop approach, the product must pass through machines in a predefined sequence. Their case study describes a manufacturing system consisting of ten machines designed to produce at least 30 products over a 24-h period. The simulation step size was divided into 5-min intervals, resulting in a total of 288 time steps. The buffer capacity between the machines was ten products, and the processing time for each machine was uniformly generated between one and ten steps. The electricity consumption for each machine was randomly generated and ranged between 30 kW and 50 kW.

Caro-Ruiz et al. [64] and Lombardi et al. [65] adopt an approach that creates additional flexibility options through the sizing of BESS and production buffers to synchronize energy availability and demand. Instead of utilizing metaheuristics or MILP solvers for production buffer and BESS sizing, they employed a methodological approach. They identified bottlenecks in energy supply or production buffers and determined the appropriate size to eliminate these bottlenecks. The goal was to achieve a completely emission-free factory. Subsequently, they conducted a sensitivity analysis to verify the accuracy of their determination. However, a disadvantage of this approach is insufficient cost efficiency. Complete decarbonization may not be economically viable owing to significant seasonal fluctuations in the energy supply at European latitudes.

Additionally, they used a generic approach for their production modeling based on [67], which is similar to the PPR notation and the value stream method [68], comprising a series of production processes, each connected with a buffer and arranged in series (flow-shop problem). Exclusively discrete processes can operate in three different modes: production, standby, and off, and are referenced with average energy consumption.

At the beginning of each control period, a forecast of PV generation was available. EED and EF are performed using a MILP solver [64,65].

Caro-Ruiz et al. and Lombardi et al. are among the few authors who partly integrate EF, EED, and OSS into a comprehensive implementation methodology. Unlike many other authors who employ metaheuristics, they rely on a more specific algorithm. Initially, they determined the storage size required to achieve a completely emission-free factory.

Beier et al. [67] did not present an optimization methodology but rather an approach to synchronize demand and supply in real-time without affecting throughput. They identified constraints in the material flow that must not be altered. This approach involved increasing the product’s value during periods of high or low energy availability and binding the energy within the product during times of low or expensive energy availability. Although neither BESS nor P2P is used, it is considered a future flexibility option.

The authors analyzed the production process in small periods of one hour with steps of one second, achieving a highly granular perspective. Their factory model resembles the PPR notation in many aspects. They considered two production classes: discrete and continuous. Machines have three process states: production, standby, and off. They use key performance indicators (KPIs) for the individual product, like the CO₂ emissions per product. They assume a simplified emission factor of 0.6 kg CO₂ per kWh of electricity from the grid and zero emissions from RES.

In addition, the average energy consumption per production process and machine was provided, with each station assigned a buffer or storage. Two main feedback loops were introduced: the execution loop for energy flexibility aimed to plan processes for better alignment between the system’s electricity demand and variable supply while maintaining constant throughput. The system adjustment loop modified the system parameters (e.g., buffer capacities and system utilization) to evaluate the effectiveness of the proposed energy flexibility control under various scenarios. The description began with the electricity supply module, which provides inputs for the other modules.

Wanapinit et al. [66] presented a template-based generic design aimed at minimizing the modeling effort for factory processes, particularly for SMEs. They introduced a method for quantifying flexibility potentials, in which time-dependent flexibility potentials and the costs of providing flexibility were determined. Their model consisted of two levels: physical and management.

Physical-level modeled components, as well as energy and material flow. At its core is the machine, which is represented as the converter of intermediate products and electricity and is an obligatory component of all processes. Similar to the approach in [64,65], a mandatory storage unit is assigned to each machine.

The management level represented constraints in operational planning, such as changes in production volumes or shifts in production times. Scheduling is planned, and flexibility measures are employed when actual operations deviate from the planned operations. An electric vehicle served as a flexible intermediate storage/consumer.

Küster et al. [63] presented a genetic algorithm for optimizing a flexible job-shop scenario, where, in brief, processes are represented as alleles of individuals. They plan processes using a weighted single-objective fitness function that incorporates energy consumption, peak shaving, and makespan. The optimization result included a schedule for each production cell and process.

Products consist of tasks that can be performed in various execution modes, with each mode potentially requiring a different machine or having a different energy consumption and duration. In addition to electrical energy, they considered limited resources, such as the state-of-charge of BESS, compressed air, or raw materials. Overall, their concepts share strong similarities with the PPR notation.

4.3. Research Gap Analysis

In the context of the first research question, metaheuristics are predominantly employed in OSS, which can be categorized into two main types: evolutionary- and swarm-based metaheuristics (

Table 4. Quantitative statistics of all three research areas considering optimization approaches.

Group	Feature	SLR OSS	SLR EED	SLR EF	Total
Metaheuristic/Solver	Evolutionary Based	5	1	1	7
	Swarm Based	4	1	-	5
	Linear Solver	3	4	3	10
Objectives	Single-Objective	2	3	5	10
	Multi-Objective	8	4	-	12
	Emissions-Related Objectives	9	4	2	15
	Costs-Related Objectives	12	7	2	21

). In contrast, approaches to EED and EF utilize MILP solvers more frequently. The use of single- and multi-objective approaches also varies: OSS predominantly employs multi-objective approaches, EED exhibits a balanced ratio, and EF predominantly employs single-objective approaches. Investment costs are the predominant objective, with emissions being considered to a lesser extent.

The NSGA-II algorithm is widely adopted for multi-objective optimization and is often customized for specific application scenarios. It is particularly suitable for OSS and threshold optimization of decision-tree-based EED and EF applications. Swarm-based metaheuristics excel at optimizing a variety of parameters, such as the charging and discharging profiles of multiple energy storage systems over extended periods. In contrast, MILP solvers are suitable for both OSS and threshold optimization, albeit they are restricted to linear simulation models that are less complex. Each optimization approach exhibits both strengths and weaknesses. No single approach is capable of simultaneously addressing all optimization potentials. Therefore, it is necessary to develop a novel approach that integrates the various optimization methods and consolidates their advantages.

Furthermore, a comparison was made between the objectives, categorized into two groups. Across all three SLRs, cost-related objectives are utilized significantly more frequently, with 21 occurrences, compared to emission-related objectives, which have 15 occurrences. This disparity is particularly pronounced in the SLR OSS and SLR EED, where cost-related objectives are employed nearly one-third and twice as often.

Regarding the second research question, it is noted that a variety of different approaches are often presented singularly (

Table 5. Quantitative statistics of all three research areas, considering functionalities.

Group	Feature	SLR OSS	SLR EED	SLR EF	Total
Step Size	s/min/h	-/-/10	-/1/7	2/2/2	2/3/19
Simulation Period	h/d/wk/a	-/-/-/10	-/6/1/1	1/2/1/2	1/8/2/13
Cost Model	Only Electricity Costs	-	-	4	4
	CaPex	-	1	-	1
	OpEx	-	8	-	8
	NPV	10	1	-	11
Electricity Model	DAM	3	2	2	7
	ToU Buying/Selling	5/3	7/6	3/3	15/12
	Peak Fees	2	-	1	3
	Electricity Price Trend	1	-	-	1
Sustainability model	CO2, NOX, SO2 Emissions	1	6	-	7
	Social Costs	1	-	-	1
	SSR	8	-	4	12
Simulation Model Functionalities	CHP	1	1	1	3
	Ramp-Up	1	2	-	3
	H2 Sale, Buying	1	1	-	2
	Degradation	1	2	-	3
	Standard Load Profile	-	1	-	1
	Multi-Agent	-	1	-	1
	Load Forecast	-	1	-	1
	RES Forecast	-	1	-	1
	SOC Forecast	-	1	-	1
	PPR Modeling	-	-	4	4
	Flexibilization Analysis	-	-	2	2
	KPIs per Product			1	1
	Stochastics	1	1	1	3
Optimization Model Functionalities	MILP Integrated	2	-	-	2
	Hyperparameter Tuning	1	-	-	1
	Stochastic Modeling	-	1	1	2
	Interactive Search	-	2	-	2
	Data-Driven Scheduling	-	-	1	1
	Scheduling Process as Alleles	-	-	1	1
EED Strategy	Conventional	10	1	3	10
	Grid Charging	7	-	-	7
	Optimizer	-	7	3	10
EF Strategy	Load Shifting	1	2	-	3
	Flow-Shop	-	-	4	4
	Flexible Job-Shop	-	-	2	2
Case Study	Factory/Building/Generic	4/5/-	-/3/5	6/-/-	10/8/5

).

Generally, the step size in the SLR OSS and SLR EED predominantly spans one hour, whereas the SLR EF approaches operate more frequently in the second and minute ranges. Concerning the simulation period, the EED and EF approaches also tend to focus on shorter durations, typically spanning days rather than years.

In terms of cost models, SLR OSS approaches extensively analyze NPV, whereas SLR EED and SLR EF approaches primarily consider the OpEx costs of the MG or only electricity costs.

In the electricity model, it is notable that across all research areas, nearly half of the approaches incorporated dynamic ToU pricing for purchasing and selling electricity. Conversely, only seven approaches included DAM considerations. Detailed energy costs, such as peak fees, are infrequently represented, with only one source examining the future trends in electricity costs over the next decades.

Sustainability models are noticeably less prevalent than cost models in the SLR OSS and SLR EF approaches. Most prominently, the SSR was considered in these models. Occasionally, the CO₂ emissions or additional emissions were calculated. Only one source incorporated the long-term social costs of emissions.

The additional functionalities of the simulation models are typically represented only once, to a maximum of four times. These functionalities include considerations such as ramp-ups and degradation of P2P systems, integration of CHP, and additional trading with hydrogen.

Only one source explicitly discussed the use of multi-agents, with the PPR model being relatively common in flexible factories. One author addressed KPIs per product, and individual authors delved into load forecasting, RES, and the state-of-charge. Only four authors have considered a stochastic perspective.

The functionalities of the optimization models, akin to the simulation model functionalities, were predominantly represented once. These included the integration of MILP solvers for EED within a metaheuristic optimization of the OSS. Other approaches included the stochastic modeling of distribution functions and the use of data-driven methods for optimized scheduling. In addition, notable functionalities included hyperparameter tuning and process scheduling as alleles in a genetic algorithm.

EED strategies were generally balanced between conventional, grid charging, and optimizer approaches. Conventional and grid charger strategies, which were optimized based on thresholds and simple decision algorithms, were most frequently employed in OSS. Optimizers were primarily used in the EED and EF strategies.

EF strategies were minimally integrated into OSS and EED, relying on basic load shifting and factory models. In contrast, EF approaches employed specific modeling strategies, such as flow-shop and flexible-job-shop.

Case studies typically involved factories, large buildings, and generic objects.

Regarding the third research question, it is observed that currently, there is no comprehensive research approach that integrates all three research areas. Furthermore, a comprehensive methodology for implementation in real-world enterprises is lacking, and there are no metaheuristics that simultaneously consider and synchronize all three research areas. Each individual approach exhibits distinct characteristics; however, no single approach combines all beneficial features.

It should be noted that there is less research conducted in the OSS area in temperate latitudes compared to more emphasis on EF research, particularly in subtropical and tropical regions. However, many highly developed industrialized countries are situated in northern latitudes, necessitating appropriate research approaches. Differences in sunlight hours underscore this need: while a city in Sweden experiences minimal daylight hours in winter, a city such as one in Iran enjoys relatively consistent daylight conditions throughout the year. These differences significantly influence the requirements and challenges of energy optimization and efficiency.

Nevertheless, sensitivity analyses and benchmarks in [45,50,51] demonstrated that a combination of short- and long-term storage that is particularly effective in temperate latitudes with pronounced seasonal sunlight availability fluctuations operates most efficiently. Further sensitivity analyses in [49,51] indicated that the selling price of electricity is a critical factor.

Without space constraints, maximizing PV capacity reduces the need for storage. Conversely, storage is crucial in a constrained space. Overall, exclusive PV generation without storage achieved the best economic outcome. The BESS enhances self-consumption efficiency, which can be further enhanced by hydrogen, creating tension between economic and sustainable factors.

Next, the identification of the need for action is derived from the following research gaps:

• Although energy pricing incorporated granular dynamic ToU prices, emissions were not considered in similar detail. A ToU consideration of emissions per kWh of electricity could shift focus from predominantly economic to sustainability considerations. Databases, such as [69], have already provided time- and location-dependent emissions of electricity.
• A holistic multi-agent simulation and optimization model covering all three research areas could not be identified. It is advisable to develop an overarching optimization algorithm that integrates various metaheuristics, thereby combining their respective advantages.
• The increasing complexity of optimization models, arising from the integration of numerous sub-models and functionalities into the optimization calculations, should not be underestimated. There is a risk of becoming trapped in local optima during the optimization process. However, the accumulation of experiential knowledge and the precise tuning of hyperparameters for various metaheuristics in relation to the specific application can enhance the results and improve the robustness of the methodology and models.
• Various functionalities have been introduced; however, an approach that implements all functionalities cannot be found. It is crucial to emphasize that not all functionalities can be applied within a single simulation and optimization model because of the potential complexity that could outweigh the benefits.
• Studies analyzing all functionalities through sensitivity analyses to determine their criticality are lacking, particularly for EED and EF strategies. Furthermore, an analysis should be conducted to determine which functionalities and model characteristics are better suited for the design and operation stages.
• A user-friendly modeling approach specifically tailored for factories, particularly a holistic implementation methodology with template-based generic sub-models designed for SMEs, is currently lacking.

5. Discussion and Conclusions

This review provides a comprehensive overview of the current research in three research areas within the context of microgrids (MGs): optimal system sizing of MGs (OSS), optimization of electrical energy distribution to storage systems and consumers (EED), and energy flexibilization of factories (EF). These areas collectively aim to optimize hydrogen storage-based MGs, which often operate on a narrow economic margin, covering optimization from energy generation through optimized distribution to energy consumption.

Furthermore, this review aims to outline the predominant optimization methods in each research area, the functionalities utilized by various optimization and simulation models, and the current approaches that integrate all three research areas.

Despite extensive literature research, this review acknowledges certain limitations due to the stringent exclusion criteria and keyword selection used to limit the volume of publications in these three areas. Source selection was conducted with rigor and adherence to principles of good scientific practice; however, completeness in capturing all relevant sources cannot be guaranteed. Moreover, there may be inherent biases in the source selection.

In this review, the focus was placed on metaheuristics and linear solvers. Consequently, artificial intelligence optimization approaches were explicitly excluded. Additionally, only approaches from the period between 2016 and 2023 were considered.

Compared to other SLRs in the field, this review is notably distinct in that none of the identified reviews integrate the domains of OSS, EED, and EF within a common consideration. While there are comprehensive SLRs available for each of these individual domains, no review encompasses all three areas together.

To provide an example, Agajie et al. [70] included over 200 sources in their SLR. The description of each source is only provided in a bullet-point format, and the sources are generally clustered. While they focused on OSS, they also incorporated a broader range of storage technologies, such as flywheels and hydropower. Additionally, they made distinctions between single- and multi-objective optimizations and presented various functionalities and metaheuristics.

To provide another exemplary review, Arsad et al. [71] focused on hydrogen-based microgrids. Their SLR is based on an analysis of citation counts, the software and methodologies used, as well as limitations and research gaps. The authors specifically address keywords such as “Vehicle-to-Grid applications”, “optimization algorithms for energy systems”, and “optimal scheduling for storage” and present these findings in quantitative analysis.

Another example in the field of EF is provided by Fernandes et al. [72], who focused their SLR on energy-efficient scheduling in job shop manufacturing rather than directly in the MG domain. Nevertheless, they also examined metaheuristics, differentiating between various types. The review considers objectives and distinguishes between single- and multi-objective approaches. Additionally, Fernandes et al. highlighted various functionalities and strategies relevant to the topic.

The reviews [73,74,75,76,77,78,79] also resemble the structure and scope of the aforementioned sources; however, their focus is exclusively on either OSS, EED, or EF. No review was found that examines and integrates these three research areas in equal measure. The SLR of this review reveals that most approaches tend to focus on one or two of these areas. Therefore, this review underscores the explicit need for further research within the context of MG optimization. Functionalities were examined across two phases: the operational and design phases, with their applicability contingent on specific use cases.

Due to the high initial investments required, hydrogen-based MGs have rarely proven to be economically viable for manufacturing SMEs. Given the currently low demand for P2P storage systems, a government-supported ramp-up of this technology is recommended. Increased demand could, in the long term, lead to a reduction in system costs through automated mass production. Additionally, with the market expansion, an improvement in the efficiency of these systems is also anticipated [80].

Therefore, it is advisable to establish a support program for the low-threshold introduction of MGs in manufacturing SMEs, with a particular focus on optimizing the design and operational phases to address concerns regarding economic viability and efficiency [81].

Furthermore, a heightened sanctioning of emissions at the production site could contribute to the full realization of the benefits of MGs, making them economically viable. The European Union is already taking a leading role in this regard by indirectly subsidizing companies with climate-friendly production through the EU Taxonomy and the Green Deal, in the form of reduced climate credits [82].

In our future research endeavors, we intend to leverage the insights from this review to develop a comprehensive, multi-agent-based simulation model encompassing the three research areas. We will conduct evaluations of the functionalities and sensitivity analyses and investigate which functionalities are most critical. This model encompasses a broad range of functionalities and serves as a software-based support system for MG sizing and operation.