Grid-Scale Battery Energy Storage and AI-Driven Intelligent Optimization for Techno-Economic and Environmental Benefits: A Systematic Review

Abstract

Grid-Scale Battery Energy Storage Systems (GS-BESS) play a crucial role in modern power grids, addressing challenges related to integrating renewable energy sources (RESs), load balancing, peak shaving, voltage support, load shifting, frequency regulation, emergency response, and enhancing system stability. However, harnessing their full potential and lifetime requires intelligent operational strategies that balance technological performance, economic viability, and environmental sustainability. This systematic review examines how artificial intelligence (AI)-based intelligent optimization enhances GS-BESS performance, focusing on its techno-economic, environmental impacts, and policy and regulatory implications. Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, we review the evolution of GS-BESS, analyze its advancements, and assess state-of-the-art applications and emerging AI techniques for GS-BESS optimization. AI techniques, including machine learning (ML), predictive modeling, optimization algorithms, deep learning (DL), and reinforcement learning (RL), are examined for their ability to improve operational efficiency and control precision in GS-BESSs. Furthermore, the review discusses the benefits of advanced dispatch strategies, including economic efficiency, emissions reduction, and improved grid resilience. Despite significant progress, challenges persist in data availability, model generalization, high computational requirements, scalability, and regulatory gaps. We conclude by identifying emerging opportunities to guide the next generation of intelligent energy storage systems. This work serves as a foundational resource for researchers, engineers, and policymakers seeking to advance the deployment of AI-enhanced GS-BESS for sustainable, resilient power systems. By analyzing the latest developments in AI applications and BESS technologies, this review provides a comprehensive perspective on their synergistic potential to drive sustainability, cost-effectiveness, and energy systems reliability.

1. Introduction

Scholars agree that the depletion of natural resources and environmental pollution have severe consequences in the power sector, where the most significant sources of emissions are electricity and heat production (∼26%) and transport (∼15%), along with hard-to-decarbonize sectors such as industry and aviation [1]. Decarbonization and grid modernization are crucial due to rising global energy demand and an aging grid, making transitioning to sustainable and RESs essential. Energy storage systems, which converts and stores various energy types with varying paths, efficiency, maturity, and scale, and particularly large-capacity storage, is critical for the effective use of renewable energy sources (RESs), and offers convenient access, coordinated control, mitigates volatility in new energy generation, and optimizes the power grid structure [2]. Hybrid renewable energy systems can further enhance clean energy adoption by integrating variable renewable energy (VRE), such as solar and wind, with ESS, thereby addressing intermittency [3] and supporting decarbonization goals [4]. Emerging technological advancements improve the efficiency, reliability, and affordability of RESs generation, distributed energy resources (DERs), and energy storage system (ESS), along with monitoring, automation, protection, control systems, and communication technologies, presenting significant opportunities for achieving a sustainable energy future. In the medium to long term, the grid is expected to shift away from carbon-based fuels, emphasizing greater integration of renewable DERs, storage, and E-transportation [5].

Electric power systems face challenges from technological advancements, environmental issues, weather dynamics, diverse consumer demands, and regulatory mandates, as well as aging infrastructure. To address these, IRENA has identified five technology pillars: green hydrogen, renewable generation, electrification, enhanced power system flexibility, and industry innovation in freight, shipping, and aviation. Energy storage is crucial for flexibility and reliability. Grid-scale batteries are less constrained than EV applications, facing lower energy density requirements, relaxed weight and volume limits, and simplified thermal and safety specifications. These characteristics enable aggressive cost optimization and greater resilience to raw material price fluctuations, largely due to the adoption of lithium iron phosphate (LFP) chemistry. Recent analyses highlight a structurally lower cost regime for stationary storage, enhancing its role in power system flexibility and the integration of large-scale renewables. According to the IEA (EV Outlook 2025), in 2024, the energy sector’s battery demand reached 1TWh, and Li-ion battery and pack prices dropped by about 20%, driven by low mineral costs and intense competition, with regional variation (e.g., ∼30% in China). This marked the most significant decline since 2017 and helped push battery pack prices toward or below the ∼$100/kWh threshold. However, these low prices may discourage investment and lead to supply shortages by 2030. Innovations and vertical integration are needed to stabilize prices and promote sustainable energy solutions [6].

The power grid is inefficient due to a mismatch between supply and demand, causing energy losses. BESS can reduce these inefficiencies, but sophisticated modeling tools are needed for precise monitoring, control, and optimal operation of storage assets. As the energy industry transitions to RESs, ESS, grid technologies, and the energy management system (EMS), achieving net-zero requires technological advancements and addressing gaps in energy system modeling. Power-to-X and energy storage are also crucial for integrating RESs into grid systems. Power-to-X technologies convert renewable electricity into clean fuels, transforming energy systems. While further research is needed into economic competitiveness and sustainability, ref. [7] presents a computationally efficient and easy-to-implement framework for designing BESS, providing a practical tool for system planning, and ref. [8] highlights future research directions in renewable energy planning, storage integration, grid modernization, and EMS development to support the transition toward net-zero energy systems.

Grid integration of intermittent power sources is crucial to the restructuring of power systems worldwide. Energy storage systems, particularly battery energy storage system (BESS), help mitigate the fluctuations and enable broader adoption of RESs. As storage capacity increases and BESS play ancillary services and support non-programmable RESs, there is growing interest in modeling these systems for power grid applications. BESS numerical models must offer high accuracy while maintaining limited computational effort [9]. Unpredictable RESs challenge power system operation, causing frequency fluctuations, reduced inertia, and insufficient voltage support. Peak shaving and energy transfer can be improved using coordinated control strategies; for example, studies on the Taiwan power system [10] and optimization-based multiple BESS control incorporating load forecasting, capacity limits, and security indices [11] demonstrated enhanced peak shaving performance.

Grid-Scale Battery Energy Storage Systems (GS-BESSs) are advantageous in electricity grids with or without RES integration [12]. They play a crucial role in maintaining the balance between electricity generation, distribution, and consumption while providing services such as peak shaving, load flattening, voltage support, frequency regulation, arbitrage, behind-the-meter, black start, and emergency backup. Real-world applications in China, the USA, and Australia demonstrate these benefits, employing diverse battery types and configurations to enhance grid stability, support renewable integration, and reduce curtailment [10]. Furthermore, ref. [13] discussed integrating lithium and flow batteries to increase VRE penetration into power grids, and ref. [14] optimized BESS location and size for mitigating voltage rise in low-voltage network. It suggests incentivizing BESS use to boost VRE share in weakly interconnected grids. Energy governance is critical for national security because it involves revenue management, stakeholder involvement, global politics, and societal values. Energy issues have a considerable impact on the geopolitical landscape [15].

Modern power systems are increasingly complex and uncertain due to their interconnected nature, requiring design, optimization, and control. Traditional grid management technologies, often based on linear models and manual intervention, struggle to handle nonlinear behavior, abrupt changes in energy consumption, and RESs [16]. To address these challenges, intelligent systems that mimic human cognitive processes knowledge acquisition, inference, and decision-making are essential for managing and optimizing grid operations. Heuristic rules, data, and quantitative analytical models must be combined. With significant ramifications for the global energy sector, artificial intelligence (AI) is quickly emerging as a primary business. The international energy agency (IEA) has conducted a comprehensive global analysis to understand the connections between energy and AI, including how to meet the demand for AI in the energy sector and how AI may change how energy is produced, used, and transported. Finding the right balance of energy sources, investing in grids, and enhancing communication between policymakers, the tech sector, and the energy industry are the three pillars the research recommends for countries to plan for, given the growing worldwide demand for electricity for AI. AI could also be a powerful tool for the energy sector, but collaboration between the public and private sectors is needed to seize the moment [17].

Traditional battery diagnostic methods struggle with large-scale datasets and variable conditions. AI improves energy efficiency, reduces costs, and drives innovation by enabling faster forecasting, real-time monitoring, and the discovery of new battery chemistries. In batteries, AI enhances charging cycles, detection, and asset lifespan, while machine learning (ML) enables continuous self-optimization and real-time adaptation to usage, environmental, and aging factors. Hybrid two-or-more neural network (NN) models further improve accuracy, training speed, and generalization, supporting preventative maintenance and system reliability [18]. As digital technology becomes central to the energy transition, AI’s role is expanding rapidly in areas such as supply forecasting, ML modeling, and energy system security [19]. AI-based optimization models improve BESS economic performance, minimize transition costs, and meet residential and industrial needs by considering battery lifecycle and renewable forecasts [20].

Inaccurate modeling of BESS results in financial and technical difficulties, erodes investor confidence in large-scale BESS initiatives, and impedes global carbon reduction efforts. Key barriers include high installation and maintenance costs [21], battery lifespan and efficiency, complex management, scalability, environmental impact, and safety risks. Advances in battery management and chemistry are expanding BESS applications from residential to industrial and utility sectors, enabling cost-effective and environmentally sustainable integration into global energy markets [22]. Recent studies review grid-service use cases and synergies [23], assess long-term usage patterns, and highlight AI-driven optimization, second-life battery applications, and innovative service models such as ESAaS and energy sharing [22]. Optimization algorithms prevent energy shortages or blackouts by balancing energy production, cost reduction, and environmental impact. Still, their computational complexity can hinder real-time implementation in large-scale systems with many variables. In dynamic contexts where quick decisions are needed for grid stability and energy reliability, this makes implementation more difficult [24]. However, BESS control remains challenging due to constantly changing energy markets, grid conditions, and complex system interactions. AI has emerged as a potential solution to improve ESS control methods, offering innovative, flexible approaches to these complex issues. This systematic literature review (SLR) explores the latest advancements in GS-BESSs enabled by AI-based intelligent optimization. It evaluates the practical applications of AI in energy markets and grid services and assesses their impact on techno-economic performance and environmental sustainability. The objective of this review is to systematically examine the state-of-the-art developments in GS-BESS technologies, including their technical features, research and development advancements, sophisticated computing systems, and deployment status. The main contributions are as follows:

• We provide a structured synthesis of AI applications for techno-economic and environmental optimization of GS-BESS.
• We review cutting-edge AI techniques for improving GS-BESS performance and sustainability.
• Also, we critically assess research trends, identify gaps and technical challenges, and outline future directions for intelligent GS-BESS optimization.

This SLR is organized as follows. Section 1 introduces GS-BESS and the role of AI in intelligently optimizing its performance. Section 2 outlines the methodology used for the systematic literature review, including search strategies, selection criteria, and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework employed to identify relevant studies. Section 3 analyzes the technological evolution and diverse applications of GS-BESS in modern power systems. Section 4 critically evaluates various AI techniques applied to enhance GS-BESS operation and control. Section 5 provides a comprehensive assessment of the techno-economic and environmental impacts arising from AI-optimized GS-BESS. Section 6 highlights policies and regulations, and Section 7 concludes the review with future research directions to advance next-generation intelligent energy storage solutions for sustainable and resilient power systems.

2. Methodology of the Systematic Review

2.1. Systematic Review Framework

This section covers methods and tools, focusing on PRISMA principles for producing accurate, transparent, and well-structured reports, enhancing evidence-based decision-making, and identifying studies. The review emphasizes Grid-Scale Battery Energy Storage Systems (GS-BESS) as a key enabler of intelligent, sustainable, and financially viable energy systems. It focuses on four interconnected pillars: AI, intelligent optimization, techno-economic analysis, and environmental impact assessment, as shown in Figure 1, from which four research questions (RQs) are derived as follows:

• How do GS-BESSs contribute to addressing key challenges in grid stability, RESs integration, and emission reduction in power systems?
• What are the current applications and most used AI-based approaches in GS-BESS?
• What are the key operational, technical, economic, and environmental challenges that GS-BESS faces?
• How are policy frameworks and regulatory structures evolving to support the deployment and integration of AI-enabled GS-BESSs?

2.2. Data Sources and Search Strategy

Recent research has shown growing interest in developing efficient and reliable BESS frameworks for power systems. To provide a comprehensive understanding, a historical and systematic review of scientific studies on Grid-Scale Battery Energy Storage Systems (GS-BESS) published in October 2025 was conducted.

Search Queries

To ensure reproducibility, the literature search was conducted using standardized Boolean search strings adapted to each database. The selected keywords with the search query were

(“grid-scale” OR “large-scale”) AND (“battery energy storage system” OR “BESS”) AND (“artificial intelligence” OR “machine learning” OR “deep learning” OR “reinforcement learning” OR “optimization”) AND (“techno-economic” OR “environmental” OR “policy”)

This query was applied across the following primary databases: Scopus, IEEE Xplore, ScienceDirect, Google Scholar, and Wiley Online Library. Additional databases, including the IET Digital Library, MDPI, SpringerLink, and IOPscience were grouped under “Others”, as shown in Figure 2. For Google Scholar, a modified query was used to include reviews, articles, and research papers while excluding results indexed in major databases, thereby minimizing duplication.

“(review OR article OR research) -site:scopus.com -site:ieeexplore.ieee.org -site:wiley.com -site:sciencedirect.com -site:wos.com -site:ieee.org -site:mdpi.com -site:springer.com -site:iopscience.iop.org”

Databases were selected for their comprehensive metadata and compatibility with reference management software such as Mendeley (v1.9.3), Zotero (v7.0.11), EndNote (v25), VOSviewer (v1.6.20), and Publish or Perish (v8.6), thereby facilitating bibliometric analysis.

2.3. Inclusion and Exclusion Criteria, PRISMA Flow Diagram, and Screening Process

Initially, a broad range of sources, including peer- and non-peer-reviewed journal articles, books, international conference proceedings, and grey literature such as PhD theses and reports from leading associations in the field, including IRENA, NREL, CIGRE, and the APEC Energy Storage Network, were considered. These were included in the first round of filtering to ensure comprehensive coverage of the field. In the sampling process, search terms were determined based on keywords derived from the research topic’s scope, and the number of documents obtained is presented in

Table 1. Records from databases.

Search Terms & Boolean Operators	Year	Google Scholar	IEEE Xplore	Science Direct	Scopus	Wiley	Others
(“grid scale” OR “large scale”) AND (“battery energy storage” OR “BESS”) AND (“optimization” OR “artificial intelligence”) AND (“techno-economic” OR “environmental” OR “policy”)	1950–1959	3				1
	1960–1969	3				1
	1970–1979	28		5		1
	1980–1989	47		12		9	4
	1990–1999	86		31	1	9	3
	2000–2009	319		47	21	32	7
	2010–2019	3670	29	848	1539	328	316
	2020–2025	13,100	135	5396	9267	882	1695
Total		17,256	164	6339	10,828	1263	2025

, recorded on 15 October 2025.

Exclusion criteria. Studies were excluded if they focused on AI applications in power grids without GS-BESS, AI methods unrelated to GS-BESS operation, control, or optimization, or addressed EV-only battery systems without clear relevance to grid-scale applications. In addition, articles that were not available for full-text review were excluded from the final analysis.

Following record identification, a filtering step was applied to two database fields: document type and language. Only peer-reviewed journal articles and review papers written in English were retained, without time restrictions. This step resulted in the removal of 32,937 records. Subsequently, 9893 duplicate records were identified and removed during the database import phase. Titles and abstracts were then screened to assess relevance to the research objectives. Studies unrelated to the research theme were excluded, resulting in the removal of 2014 articles and leaving 770 records for full-text assessment.

Inclusion criteria. At the eligibility stage, full-text articles were assessed based on predefined inclusion criteria. Eligible studies were required to be peer-reviewed journal articles or review papers focusing on GS-BESSs. In addition, studies had to address the following dimensions: artificial intelligence or optimization techniques, techno-economic analysis, environmental assessment, or policy and regulatory aspects related to GS-BESSs. Only English-language publications were considered.

Full-text articles were then reviewed against the inclusion and exclusion criteria, and those not meeting the requirements were excluded. This final screening resulted in 246 articles included in the SLR. The complete selection process is illustrated in the PRISMA flow diagram in Figure 3.

2.4. Risk of Bias and Methodological Quality Assessment

To comply with PRISMA 2020 guidelines, a tailored quality assessment framework was applied for AI-based GS-BESS studies. Each included study was evaluated using four key criteria:

-

Validation Strategy: Proper model validation using cross-validation, train/test splits, or real-world datasets.

-

Data Realism: Use of realistic, publicly available, or representative datasets.

-

Benchmark Comparison: Comparison with baseline methods or the prior literature.

-

Scenario Diversity: Consideration of multiple operational, environmental, or techno-economic scenarios.

A simple quality assessment scoring system was applied. Each criterion was evaluated using a three-level scale: ($✓=High/2 points$) if the criterion was fully met, (∼ = Medium/1 point) if partially met, ($\times =Low/0 points$) if not met, and N/A for non-applicable cases. Four research questions (RQ1–RQ4) and four quality criteria (validation, data realism, benchmark scenario, and Scenario Diversity) were considered. The total quality score for each paper was computed as the sum of applicable criteria, and studies were classified into high-, medium-, and low-quality levels based on percentage thresholds, as shown in Table 2. Out of the 246 studies included in this review, 10 papers (4.1%) were classified as low risk of bias, 174 papers (70.7%) as medium risk, and 62 papers (25.2%) as high risk of bias. The results indicate that the majority of the literature relies on simulation-based and analytical approaches, with relatively few studies providing comprehensive real-world validation and benchmarking.

Table 2. Risk-of-bias/methodological quality assessment of included studies (representative sample).

Paper	RQ Addressed	Validation	Data Realism	Benchmark Comparison	Scenario Diversity	Risk of Bias (Total Score)	Quality
Review [25]	RQ1	∼	∼	∼	∼	Medium (4)	High
Article [26]	RQ2	✓	×	✓	×	Low (4)	High
Article [27]	(RQ1) RQ3	∼	∼	∼	∼	Medium (4)	Low
Review [28]	RQ4	∼	∼	∼	∼	Medium (4)	High
Review [29]	RQ1, RQ2	∼	∼	∼	∼	Medium (4)	High
Article [30]	RQ1, RQ3	∼	×	×	×	High (1)	Low
Article [18]	RQ2, RQ3	✓	×	×	×	Medium(2)	Medium
Article [31]	RQ2, RQ3, RQ4	∼	∼	∼	∼	Medium (4)	High

Table 2. Risk-of-bias/methodological quality assessment of included studies (representative sample).

Paper	RQ Addressed	Validation	Data Realism	Benchmark Comparison	Scenario Diversity	Risk of Bias (Total Score)	Quality
Review [25]	RQ1	∼	∼	∼	∼	Medium (4)	High
Article [26]	RQ2	✓	×	✓	×	Low (4)	High
Article [27]	(RQ1) RQ3	∼	∼	∼	∼	Medium (4)	Low
Review [28]	RQ4	∼	∼	∼	∼	Medium (4)	High
Review [29]	RQ1, RQ2	∼	∼	∼	∼	Medium (4)	High
Article [30]	RQ1, RQ3	∼	×	×	×	High (1)	Low
Article [18]	RQ2, RQ3	✓	×	×	×	Medium(2)	Medium
Article [31]	RQ2, RQ3, RQ4	∼	∼	∼	∼	Medium (4)	High

Figure 3. PRISMA flow diagram. Source: modified from [32].

Figure 3. PRISMA flow diagram. Source: modified from [32].

2.5. Statistical and Bibliometric Analysis of Research Trends

A bibliometric and text mining analysis using VOSviewer^® software was employed to identify and visualize key research trends in GS-BESS studies. The co-occurrence analysis, presented in Figure 4, compares keyword associations (a) and spectral and (b) network visualizations after applying rigorous filtering criteria. The resulting visualization highlights three core dimensions of GS-BESS research: technological, economic, and environmental. Our analysis identifies “battery energy storage systems” as the central node in the co-occurrence network, underscoring its dominance in GS-BESS-related studies. Other frequently clustered terms include techno-economic modeling, machine learning algorithms, and battery management systems, revealing the interdisciplinary nature of the current research. This keyword mapping not only delineates the prevailing thematic focus areas but also informs our SLR categorization framework, ensuring a focused examination of the most impactful technological and methodological developments in the field.

Figure 5 presents the distribution of reviewed papers in Scopus by document type (a) and field (b). The article documents 60% in specific fields engineering (30%) and energy (27%). English dominates as the primary publication language (90.7%), followed by Chinese (9.0%) and other languages (0.3%). In terms of country contributions, China leads with over 7000 documents, followed by the USA with 2500, while other countries contribute less. This highlights both the global scope of the research and regional disparities in publication activity.

Scopus-indexed records confirm China’s dominance in this field, with the National Natural Science Foundation of China identified as the leading funding sponsor Figure 6a, while Figure 6b highlights the top source journals.

Figure 7 illustrates publication trends by decade and database. Research activity has grown significantly over time, especially since 2000. The number of papers surged from 374 in the 2000s to 13,660 in the 2010s and further to 34,307 in the 2020–2025 period. In terms of database sources, Google Scholar accounted for the largest share (26,547 papers), followed by Scopus (14,377) and ScienceDirect (5807), reflecting their central role in disseminating academic research.

The review provides insights into the geographical distribution of key contributors, publications, and top journals in the field. China leads the field, followed by the USA, which are ranked as the top countries based on the selected research keywords. The review also highlights the most frequently used terms in the field, providing an overview of the primary topics and concepts under investigation.

3. Grid-Scale Battery Energy Storage Systems (GS-BESSs)

Grid-Scale Battery Energy Storage Systems (GS-BESSs) are vital for ensuring grid adequacy, operational reliability, planning flexibility, and renewable energy integration. They require efficient storage and supply of electricity over various time scales, from seconds and hours to days, seasons, and even years, balancing supply and demand under varying conditions to ensure a sustainable and resilient energy system. This section aims to answer RQ1: How do GS-BESSs contribute to addressing key challenges in grid stability, renewable energy integration, and emission reduction in power systems?

3.1. Overview of Grid-Scale Battery Technologies

Electrochemical batteries have evolved from the “voltaic pile” to modern chemistries. Most batteries use galvanic cells, which produce electricity through spontaneous redox reactions, whereas electrolytic cells use an external power source for nonspontaneous chemical reactions. Advances in efficiency and cost have made large-scale BESSs practical for modern power systems [33]. ESSs can be classified by energy form: thermal, mechanical, chemical, electrochemical, electrical, magnetic, or hybrid, and by use, duration, and efficiency, each addressing specific time and space constraints [34]. A battery maintains the electric potential difference across its terminals, Figure 8.

Lead–acid batteries are low-cost, mature tech, have a limited cycle life, and are good for stationary backup. Sodium–sulfur (NaS) batteries offer high energy density, long duration, high efficiency, and proven long life. Redox flow (RF) batteries provide low energy density and economic promise for grid applications [36]. Lithium iron phosphate (LFP) batteries feature high thermal stability, safety, and long cycle life and are widely used in EVs and stationary storage, though they degrade over time due to lithium loss and active material aging [28]. Vanadium redox and zinc–bromine batteries show excellent durability, while NaS experiences higher self-discharge [37]; service lifetimes range from 5 to 15 years, as shown in

Table 3. BESS technologies [28,36,37,38,39,40].

Battery Type	Example	Energy Density (Wh/kg)	Round-Trip Efficiency (%)	Cycle Life (Cycles)	Lifetime (Years)
Li-ion	LFP (LiFePO4)	90–160	90–95	>1000–10,000 at 90% DoD	10–15
Na–S	Molten Na–S at ∼300 °C	150–240	∼80	∼4500	10–15
Flow batteries	Vanadium Redox Flow (VRFB)	15–30	70–85	>10,000	10–20
Flow batteries	Zn–Br2, Fe–Cr	30–70	65–80	5000–10,000	10–15
Lead–acid	AGM, Gel, Lead–Carbon	30–50	70–85	500–2000	3–8
Sodium–nickel chloride	Na–NiCl2	100–120	85–90	2000–4000	8–12
Aqueous Zn-based	Zn–Br2, Zn–Fe, Zn–MnO2, Zn–air	60–100	70–85	1000–5000	5–10
Solid-state Li-ion	Li metal anode, solid electrolyte	150–250	90–95	2000–7000	8–15
Na-ion	NaFePO4, NaMnO2 analogs	100–160	85–90	2000–5000	8–12
Iron–air/metal–air	Fe–air, Zn–air, Li–air	150–300	50–70	>10,000	15–25

. All-solid-state iron-air batteries are promising for large-scale, high-temperature energy storage but are limited by slow redox reactions; their development, materials, challenges, and strategies for performance improvement are reviewed in [38].

Battery performance and degradation are influenced by temperature, current, mechanical stress, and operational history. Its components include a battery pack, BMS, auxiliaries, a DC/AC inverter, a PLC, and a transformer [41]. The BMS, the central component of the BESS, tracks key variables in real time and calculates indicators to maximize battery performance, life, and safety [42]. Core parameters include the State of Charge (SoC) and Depth of Discharge (DoD), which quantify energy usage over time, while the State of Energy (SoE) tracks stored energy, accounting for charging/discharging power and efficiency. Estimation methods for SoC include Coulomb counting, model-based, and data-driven methods. Degradation is typically captured by semi-empirical capacity fade models, which relate cycle number and DoD to capacity loss, and State of Health (SoH) expresses remaining capacity as a percentage of nominal capacity, accounting for operating conditions and charge/discharge efficiency, ref. [43] developed a real-world, temperature-based SOH model showing 20–60% faster BESS degradation under high-temperature conditions. Thermal behavior is modeled using energy balance equations, where heat generation comes from reversible and irreversible electrochemical reactions and Joule heating, and the rise of temperature is determined by cell mass, specific heat, and convective heat transfer with convective heat losses [44].

Battery lifespan is estimated using Peukert-type relations that link life expectancy to usage patterns and DoD. These mathematical models provide a foundation for system design and operation optimization and form the basis for later ML-based prognostics.

$$j_v=a_v{i}_{ex}exp\left({\displaystyle \frac{a_a{F}_{\eta }}{RT}}\right)-exp\left({\displaystyle \frac{-a_c{F}_{\eta }}{RT}}\right)$$

(1)

For the Butler–Volmer equation, we are concerned with electrode/electrolyte interface current density. Exchange current density, $i_{ex}$, overpotential, $\eta$ and charge transfer coefficients for oxidation, $a_a$ and reduction, $a_c$.

$$\Delta T_{\sigma }={\displaystyle \frac{1}{C_{p}m}}\left(Q-hA(T_{\sigma }-T_{air}\right))\mathrm{or}\Delta T_{\sigma }={\displaystyle \frac{1}{C_{p}m}}\left(\int I(U_{oc}-U_t)dt+Q\right)$$

(2)

$\Delta T_{\sigma }$ is the temperature rise, $C_p$ is the specific heat capacity, m is the cell mass, Q denotes the heat generated, $T_{\sigma }$ is the temperature of the battery surface, $T_{air}$ is the ambient temperature, h is the convective heat transfer coefficient, and A is the cell surface area as follows:

$$Q=\int \left(I(U_{oc}-U_t)+I{T}_{air}{\displaystyle \frac{\partial U_{oc}}{\partial T_{air}}}\right)dt+Q_R$$

(3)

$U_{oc}$ and $U_t$ are the open circuit voltage and terminal voltage, respectively, I is the current, and $Q_R$ is the irreversible reaction heat.

$$E_{fade}=E_{0}exp\left({\left({\displaystyle \frac{n}{A}}\right)}^{\alpha }-{\left({\displaystyle \frac{DoD}{B}}\right)}^{\beta }\right)\mathrm{and}L{f}_s={\displaystyle \frac{N_{100}^{end}}{365.N_{100}^{eq.day}}}={\displaystyle \frac{N_{100}^{end}}{365{\sum }_t{(\Delta Do{D}_{t_s})}^{kp}}}$$

(4)

$A, B, \alpha , \beta$ are constant, specific to the battery configuration and chemistry, and n is the number of cycles used.

$$So{C}_t=So{C}_0+{\displaystyle \frac{{\int }_0^t{I}_c\left(t\right)dt}{C_n}}\mathrm{and}DoD(\%)=100-SoC(\%)$$

(5)

$C_n$ is the nominal capacity of the battery in Ah, $I_d\left(t\right)$ and $I_c\left(t\right)$ are the discharge and charge current, and Peukert is the lifetime constant, kp (0.8 to 2.1)

$$SoH(\%)={\displaystyle \frac{Q_C}{Q_N}}\times 100\mathrm{or}\Delta SoH=k.f\left(DoD,T,{\eta }^{{\displaystyle \frac{charge}{discharge}}}\right)$$

(6)

$$So{E}_t=So{E}_0-{\displaystyle \frac{{\int }_0^t{P}_d\left(t\right)dt}{E_n}}\mathrm{or}So{E}_t=So{E}_0-{\displaystyle \frac{{\int }_0^t{P}_d\left(t\right)dt}{E_n}}+{\eta }_r{\displaystyle \frac{{\int }_0^t{P}_d\left(t\right)dt}{E_n}}$$

(7)

$E_n$ is the nominal energy, and $P_d\left(t\right)$ and $P_c\left(t\right)$ are the discharge and charge power at time t.

Battery fault mechanisms exhibit time-varying, nonlinear, and nonuniform signaling. Overheating, overcharging, external impact, extrusion, and puncture can initiate safety incidents, with fault propagation varying in pathway and rate based on the mechanism [45]. The primary barrier to GS-BESS is its limited lifetime, which is affected by degradation from charge–discharge cycles, such as capacity fading and impedance rising, with higher degradation rates in early and late cycles; lifetime is defined by cycle life, the number of cycles until capacity drops below 80% of the initial rated capacity.

GS-BESS mainly uses Li-ions, supports frequency regulation [46], smooths renewable energy fluctuations, and provides backup energy, while a small BESS enhances flexibility in distributed generation. However, it functioned under ambient temperatures exceeding the suggested range for about 35% of the daily operational period [47]. A novel cold plate with self-crossing flow channels reduces maximum battery temperature by 7–25% and improves temperature uniformity, while optimal flow conditions using the Taguchi method further lower temperature and pump power by 12% and 30%, demonstrating enhanced heat dissipation and efficient thermal control [48]. Recycling and cascading use of used batteries are essential, yet China lacks a comprehensive recycling network, relying on hybrid models. Improving safety, efficiency, and economy, battery safety monitoring, residual value assessment, timely maintenance, and policy frameworks are crucial. Ref. [27] discusses cascade utilization, recycling, reverse logistics, policies, technologies, feasibility, and challenges as a reference for lithium battery health management systems. The growing focus on GS-BESS is driven by high capital costs associated with peak demand management, the need to invest to ensure grid reliability, and the increasing penetration of RESs. NaS, RF, and Li-ions are increasingly deployed, leveraging cost reductions and technological improvements [49,50,51]. Deployment depends on factors such as capacity, energy, and power density, self-discharge, efficiency, cycle life, scalability, eco-friendliness, modular design, raw material availability, temperature tolerance, safety, and toxicity. Li-ion currently dominates; others, including flow, zinc-based, sodium-ion, and solid-state batteries, are reviewed in [52].

3.2. BESSs in Power Grid Systems

GS-BESSs are deployed across all grid levels, shown in Figure 9. At the generation level, they support renewable integration through peak shaving and valley filling, ramp-rate control, frequency regulation, and spinning reserve. At the transmission level, they provide congestion relief, voltage support, dynamic stability, load leveling, black-start capability, grid power flow optimization, curtailment reduction, and deferral of grid upgrades. At the distribution level, they enable peak load management, power quality improvement, energy arbitrage, load shifting, microgrid and DER integration, coordinated control of distributed storage, and the construction of hybrid energy storage systems, ensuring a balance between generation, distribution, and consumption. Their modular design, rapid response, flexible installation, and short construction cycles, along with operational benefits such as availability and versatility, make them highly adaptable [53], and they play a role in sizing renewable loads and optimizing microgrid performance. Refs. [54,55] improved power quality and economic benefits by optimizing placement and capacity in distributed generation networks using an NSGA-II with entropy and AHP weighting. They also mitigated transmission congestion by shifting power flows and using virtual power lines and grid boosters, thereby reducing operational costs and improving transmission efficiency [56] more effectively than curtailment or demand response [57], and relaxing inertia constraints in grids with high renewable penetration [58].

The global shift to RESs underscores the critical role of storage in managing weather and seasonal variability and ensuring a reliable supply [34], and ESSs has increased from 1850 to 2022. Power electronics and converters allow energy to be stored when there is excess power and is released when needed. Ref. [59] examines their role in power conditioning, energy efficiency, and battery storage, emphasizing flexible active–reactive optimal power flow, grid-scale storage, and smart grid integration, promising battery technologies for long-duration, focusing on thermal and chemical storage media. Modular BESSs, including their electrical configurations, control, and performance, are reviewed in [60], while ref. [61] improves grid adaptability and reduces surplus generation, but challenges remain in data transparency and operational insights [23]. Figure 10 illustrates how energy storage supports diverse grid services across time scales, providing peak shaving, reducing renewable curtailment, and increasingly delivering ancillary services like frequency regulation and spinning reserves.

GS-BESS performs key functions ranging from seconds to hourly and longer-term, including frequency regulation, energy storage and dispatch, market participation, voltage support, deferral of infrastructure upgrades, grid reliability and resilience, and renewable energy integration. Paper [62] reduces investment costs by 18% by demonstrating the use of fast-acting energy storage systems in transmission expansion planning, and allowing for increased network usage and a deferral option for new transmission lines, while in Thailand, ref. [63] suggests the optimal battery size is influenced by load growth and energy needs, and ref. [64] improves the unit commitment utilizing a mixed-integer programming method. Operational studies explore BESS performance under different scenarios: lifespan and potential negative impacts [65], behavior under fluctuating demand and renewables [66], and quantifying their integration benefits and small-scale storage economically and technically. Comparisons with auxiliary devices, such as ultracapacitors, show the BESS’s effectiveness in damping oscillations and improving frequency response during sudden disturbances [67]. However, safety remains a concern, particularly with Li-ion batteries, due to risks of short circuits, overheating, and overcharging. Measures such as shutdown separators, fuses, positive-temperature-coefficient elements, and pressure release valves mitigate hazards. Predicting temperature rise is crucial; ref. [68] proposes online and two-step prediction methods to assess heat generation and maximum temperature under stress tests. The integration of BESSs with RESs and energy, generation, transmission, consumption, and storage components is crucial for improving reliability, efficiency, and environmental performance, as shown in Figure 11.

4. AI-Based Intelligent Optimization in GS-BESS

Studying the global shift towards sustainable energy addresses energy security, climate change, and socioeconomic issues, ref. [69] explores AI in optimizing energy systems, integrating RESs, and supporting UN Sustainable Development Goals (SDGs). This section reviews the ML-based intelligent optimization in GS-BESS to a focus on answering RQ2: What are the current applications and most used AI-based approaches in GS-BESS?

4.1. AI Approaches and Optimization Techniques for GS-BESS

Optimization is crucial for designers, saving time and cost across, and improving efficiency and system reliability in various fields, including energy system planning and operation. Problems can be categorized by type, such as discrete versus continuous, smooth versus non-smooth, static versus dynamic, constrained versus unconstrained, single objective versus multi-objective, and deterministic versus stochastic. Approaches include mathematical methods, gradient-based methods, linear and nonlinear programming, heuristic and rule-based methods, trial-and-error, and metaheuristic and evolutionary algorithms, including genetic algorithm (GA), particle swarm optimization (PSO), harmony search (HS), and simulated annealing (SA), and convex versus nonconvex problems. While these traditional optimization methods are powerful, many real-world problems, especially NP-hard or large-scale systems, require more adaptive and intelligent approaches [40].

Intelligent optimization is a crucial field in AI, operations research, control, and other technical domains. It involves modifying component specifications to achieve desired outcomes, especially when exhaustive searches are impractical. Traditional approach takes rule and data and give output, and AI takes data and out-put and give rules, shown in Figure 12. This approach minimizes resource consumption and emphasizes problems, constraints, optimization processes, and the objective function. Intelligent techniques like AI and database technology enable computer systems to perform complex tasks, such as acquiring knowledge, discovering hidden patterns, generating complex solutions, and automating routine tasks [70]. Hybrid optimization techniques manage complex interactions, aiming to maximize reliability, minimize costs, and reduce load loss probability, and ref. [71] highlights optimization methods for hybrid energy systems with battery storage, noting growing interest driven by sustainable development, environmental concerns, and energy efficiency. AI opens opportunities to tackle complex and uncertain problems by its ability to learn and adapt in diverse environments, offering high autonomy, enabling machines to learn, reason, and undertake decision-making decisions under uncertainty [72], and have goal-seeking performance, high abstraction, data fusion from multiple sensors, and learning and adaptation in heterogeneous environments, while traditional methods fail to solve them.

AI in GS-BESSs primarily consists of (I) ML, including supervised (SL), unsupervised (UL), reinforcement (RL), and deep learning (DL), (II) Expert Systems/Rule-Based, and (III) Fuzzy Logic methods, as shown in Figure 13. Common MLs are as follows:

• SL: AI-specific methods for classification and regression include SVM, decision trees (ID3, C4.5, CART), random forests, KNN, Naïve Bayes, XGBoost, and Gaussian Process Regression (GPR). Classical statistical regression methods (linear, polynomial, and exponential) are general tools that can be used within ML pipelines for feature modeling.
• UL: For clustering problems on unlabeled datasets and dimensionality reduction using K-means, hierarchical clustering, DBSCAN, PCA, and Isolation forests.
• RL: Agents maximize cumulative rewards using Q-learning, deep Q-network, policy gradient, and actor–critic algorithms.
• DL: ANN-based models like SLFNN, DNN, ELM; AE/VAE for feature extraction; CNN for representation learning; RNN/LSTM for time-series; transformers and GANs for forecasting and data generation [73].
• DRL: Combines DL and RL for complex state-action tasks, e.g., optimal scheduling and energy management.
• Hybrid learning: Integrates multiple paradigms for improved optimization in complex power systems.

ML follows key steps: data acquisition, selection, feature engineering (including preprocessing, conversion, and modelling), data processing, hyperparameter optimization, data regularization, normalization, and standardization. Performance evaluation is critical, employing key metrics and cross-validation techniques, summarized in Equations (8)–(15), and is used to assess model accuracy, robustness, and effectiveness. The most frequently adopted evaluation metrics across the reviewed studies are RMSE, MAE, and $R^2$, which are reported in approximately 41%, 23%, and 19% of the studies, respectively, as summarized in

Table 4. Frequency of performance metrics used in reviewed articles ($N=149$).

Performance Indicator	No. of Articles	Usage (%)
RMSE	61	41
MAE	34	23
$R^2$	28	19
MAPE	13	9
Others	12	8

. Model validation techniques include k-fold cross-validation, which splits data into k subsets for repeated training and testing; stratified k-fold, which preserves class distribution in each fold; leave-one-out cross-validation (LOOCV), which tests on a single instance at a time; time series cross-validation, which maintains temporal order; and nested cross-validation, which combines inner and outer loops for unbiased hyperparameter tuning and model selection.

In GS-BESS, reviewed studies show that AI approaches in Figure 13 supports a wide range of operational functions. ML methods are predominantly applied for SoC/SoH estimation, load and RES forecasting, and prediction. DL approaches provide enhanced capabilities for fault detection, anomaly identification, and degradation modelling. RL and DRL enable optimal control and autonomous decision-making for BESS scheduling and real-time energy management. For instance, DRL-guided observer-based control has demonstrated high state estimation accuracy and adaptability, with reported SOC RMSE values below 0.3% and enhanced robustness against modeling uncertainties and operational disturbances [74]. In addition, hybrid optimization, AI frameworks improve BESS sizing, scheduling, bidding strategies, and energy arbitrage [75].

Evaluation metrics for ML: Accuracy for classification quantifies the percentage of accurate predictions; it is less accurate for unbalanced datasets.

$$Accuracy={\displaystyle \frac{TP+TN}{TP+TN+FP+FN}}$$

(8)

Precision assesses the percentage of real positives among anticipated positives.

$$Precision={\displaystyle \frac{TP}{TP+FP}}$$

(9)

Recall, known as sensitivity, counts the percentage of actual positives accurately detected.

$$Recall={\displaystyle \frac{TP}{TP+FN}}$$

(10)

F1 Score combines precision and recall balancing performance in unbalanced datasets.

$$F1Score=2\times {\displaystyle \frac{Precision\times Recall}{precision+Recall}}$$

(11)

Root mean squared error (RMSE) expresses this in the target’s units

$$RMS=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{({\widehat{y}}_i-y_i)}^2}$$

(12)

Mean squared error (MSE) captures the average squared difference between predictions and actual values:

$$MSE={\displaystyle \frac{1}{n}}\sum _{i=1}^n{({\widehat{y}}_i-y_i)}^2$$

(13)

Mean absolute error (MAE) measures average absolute error with less sensitivity to outliers:

$$MAE={\displaystyle \frac{1}{n}}\sum _{i=1}^n|y_i-{\widehat{y}}_i|$$

(14)

$R^2$ (coefficient of determination) quantifies the proportion of variance explained by the model:

$$R^2=1-{\displaystyle \frac{{\sum }_{i=1}^n{({\widehat{y}}_i-y_i)}^2}{{\sum }_{i=1}^n{(y_i-{\widehat{y}}_i)}^2}}$$

(15)

BESS optimization involves determining the optimal sizing, placement, and control of battery energy storage systems to improve cost-effectiveness, energy savings, and demand management. Key factors include size, capacity, cost, lifetime, power quality, flow constraints, charging/discharging limits, system reliability, and environmental impact. Effective BESS sizing also considers aging, carbon emissions, power oscillations, abrupt load changes, and transmission or distribution interruptions. Optimization approaches include probabilistic, rule-based, deterministic, mathematical, dynamic programming, convex programming, second-order cone programming, heuristics, and others. Optimizing placement, sizing, scheduling, and control enhances overall network performance [76]. It applies Binary Grey Wolf optimization to weak grids with high-RES penetration. The Data Frequency Scheduling Optimization Framework offers real-time frequency support and improves energy efficiency, reaction time, and regulation accuracy up to 96.4%, making it suitable for smart grid management [77]. A CNN-LSTM-based adaptive framework further improves SOC estimation and reduces errors by integrating scenario decoupling and multi-time-scale feature fusion [78].

4.2. Role of AI in GS-BESS

In power systems, AI enhances efficiency by optimizing operations, planning, monitoring, predicting failures, and improving decision-making [79]. Specifically, AI assists in the design, optimization, cost reduction processes, and enhancing the efficiency of RES systems and storage technologies [80], as well as for catalysis, grids, and energy storage [81]. GS-BESS utilizes various technologies and emerging chemistries. Hybrid systems and solid-state batteries are gaining popularity due to improvements in safety and energy density. Battery innovation is a complex process involving materials, scales, operations, and achieving commercially feasible performance (energy density, rate, and cycle life) at reasonable prices. The advancement of materials like lithium cobalt oxide, aluminium oxide, and graphite has traditionally relied on experimental laboratories and trial and error, which are time-consuming, resource-intensive, and constrained by specific conditions, while computational methods rely on mathematical models and can be computationally demanding [82]. However, modern computational capabilities have enabled the use of computer simulations in materials design, combined with efficient sensing systems that monitor parameters such as voltage, current, temperature, and strain, providing a comprehensive understanding of energy storage devices. Atomistic simulations based on first principles offer insights into atomic- and electronic-scale processes independent of experimental data.

The integration of AI is revolutionary, reshaping the landscape of materials discovery predicting material properties, battery design, and intelligent energy storage system optimization in rechargeable battery technologies and assisting in selecting suitable algorithms for specific applications, batteries, phase-change and paving the way for a future where nanogenerators play a crucial role [83], by handling a diverse range of data types and scales, enabling material identification, testing, performance forecasting, production, BMS optimization, and end-of-life management. AI involves creating electrolytes, cathodes, and anodes, and [84] created novel AI-enhanced cylindrical Li-ion cell batteries with enhanced low-temperature operation, durability, and safety. SES AI introduced Molecular Universe MU-0, an AI- and physics-powered interactive software platform that enables large-scale discovery and analysis of battery-relevant molecules, aiming to accelerate materials innovation for next-generation energy storage [85]. Generative AI designs and screens new transition metal oxides, discovering stable and high-performance materials for next-generation multivalent-ion batteries [86]. In materials science, ML is categorized into four applications: material property prediction, device structure optimization, fabrication process reconstruction, and measurement data reconstruction [87,88,89]. The strengths and limitations of different ML approaches are compared in [88].

In power grids, AI optimizes energy usage, grid management, and sustainability, enhancing efficiency, reliability, and resilience. Applications include accurately forecasting consumption, demand prediction, and cybersecurity enhancement, enabling dynamic pricing, improving demand response, and ensuring grid stability by detecting anomalies and integrating RESs [16]. It detects, classifies, and addresses power quality issues improved by RES integration and power converters [90], in distribution systems [91], grid-connected power system [92], and forecasting RES [93], which reveals DNNs and ensemble strategies as best suited for handling fluctuating data, summarized in Figure 14. Large-scale multi-objective evolutionary algorithms achieve cost savings, reduced power losses, and improved voltage deviation [94], while ref. [95] develops an energy management model using lithium batteries and consumption data, incorporating battery cycle life, installation parameters, and grid demand, employing Deep Deterministic Policy Gradient, dynamic, and GA optimization techniques to reduce storage and operational costs.

GS-BESS significantly impacts power system operation and transient processes, necessitating their properties in mathematical models for real-dimension power system design. Storage modeling is categorized into three main types: (a) Bucket models, known as Energy Reservoir Models (ERMs) or Power-Energy Models (PEMs), represent the ESS as an ideal energy balance by tracking the state of charge through charging, discharging, and losses. They are simple, computationally efficient, and suitable for long-term system-level studies, but lack detailed technical accuracy for voltage/current or fast dynamics. (b) Electrical models capture voltage, current, and transient electrical behavior, making them useful for power quality and frequency response studies. They provide greater detail than Bucket Models but require extensive parameter data and are less flexible across applications. (c) Physical models, also called Electrochemical or Concentration-Based Models, simulate the underlying physical and chemical processes of storage systems, enabling high accuracy, degradation analysis, and new material design. However, they are data-intensive, computationally demanding, and best suited when detailed ESS-specific information is available [96]. A pseudo-2D lithium-ion battery model captures electrochemical and mechanical behavior along cell thickness and particle radius, and accounts for electrolyte concentration, voltage, current density, and temperature, electrode kinetics with the Butler-Volmer equation, and degradation phenomena such as SEI formation, cathode dissolution, nickel interactions, graphite particle fatigue, and SEI cracking, and considers electrode architecture, porosity, and tortuosity to realistically simulate ion transport and overall battery performance [97].

Optimal BESS deployment is vital for efficiency and cost-effectiveness, given high costs and limited lifespans. For greater reliability, lower costs and losses, improved voltage stability, and reduced environmental impact, placement and sizing are essential. Ref. [98] applies GA for placement and fuzzy logic for control, ref. [99] confirmed a significant impact on both cost and accuracy, and ref. [100] combines metaheuristics with greedy scheduling, addressing dynamic pricing, renewables. Ref. [101] uses PSO to minimize costs for voltage regulation, losses, and peak demand. Digital twins offer performance, safety, and maintenance benefits for storage systems, as explored in [102]. A RL-based preventive maintenance method optimizes BESSs’ maintenance strategies while balancing costs, capacity, and reliability using Monte Carlo tree search and a DNN to process state-of-health data and determine optimal maintenance actions [103].

Battery operation and safety modeling have made progress but accurately predicting the evolution of multi-physics systems remains challenging. BMSs are essential for battery performance and safety, lifespan, and for real-time monitoring and control, estimating SoC, SoH, and RUL while managing degradation, and data-driven estimation methods [104] but faces challenges from limited parameter measurements, feature extraction, and environmental uncertainties [105], and DL-enabled solutions are reviewed [106]. Battery aging prediction improves decision-making and grid reliability, with ML methods (SVM, ANN, RF, and DL) enhancing SoC and SoH estimation accuracy and adaptability [107], incorporating temperature effects through XGBoost improves RUL predictions [108,109]. The effectiveness of BMSs largely depends on accurate sensor measurements, particularly in lithium-ion batteries, which exhibit strong nonlinear dynamics. Small uncertainties or biases in sensor measurements can result in significant state-estimation errors, posing a major challenge for conventional BMS designs and adversely affecting battery safety and performance. Consequently, integrated strategies for sensor uncertainty detection and state estimation have been investigated to improve the robustness of BMSs under practical operating conditions [110].

Table 5. Comparative analysis of AI-control vs. conventional control in grid-scale battery energy storage systems.

Metric	Conventional Control	AI-Control	Practical Notes/Implications
Computational Cost	Low–moderate; rule-based/MPC	Moderate–high; training 1–5 h, online  10–20% faster	AI reduces real-time computation after training [21,111].
Accuracy/Error	RMSE  5–10%	RMSE  2–5%	Better SOC prediction, peak shaving, and dispatch reliability [111,112,113,114].
Economic Benefit	0–5% savings	5–15% savings	2.4× profit increase, 30% peak reduction [115]. Optimized charging/discharging improves ROI and battery life [18,109]. Optimized BESS/RES placement reduces operational cost up to 51% [94].
Scalability	Linear growth; simple	Moderate–challenging; retraining may be needed	Transfer learning and modular AI assist scaling [18,109,112,116].
Generalization	Limited; specific grids	Good; 90–95% performance across scenarios	Requires diverse datasets for reliable deployment [77,111,112,114].
Real-World Evidence	Few pilots	Multiple pilots/commercial demos	Confirms AI benefits and practical trade-offs [31,43,77,94].

summarizes the computational cost, accuracy/error, economic benefits, scalability, generalization, and real-world evidence, along with practical implications.

AI offers practical solutions for managing the complexity of GS-BESS: cloud-based ML predicts and classifies battery failures with 96.3% accuracy using voltage and temperature curves, while BiGRU-based methods estimate branch charging capacity and assess pack cell health [117]. Physics-informed ML with multi-fidelity models enables SoH estimation from a single charging segment, though online prediction requires significant storage [118]. Integrating domain knowledge into DL enhances Li-ion RUL prediction efficiency, robustness, and generalizability [26]. Time-series models such as LSTM and XGBoost are used for load, price, and RESs forecasting, while RL and DRL optimize energy dispatch, market participation, optimize load management, predict battery degradation, and schedule optimal dispatch. Hybrid models and anomaly-detection algorithms, such as autoencoders and isolation forests, are employed for cybersecurity and fault detection. Classical methods such as SVMs, decision trees, and regression are effective for linear datasets, while deep learning architectures such as CNNs, LSTMs, and hybrid models excel on complex, noisy, and temporal data [114]. Second-life batteries using modular BESS reduce the cost of GS-BESS [119]. Conventional methods are error-prone and labor-intensive, particularly for large datasets [120]. A Principal Component Analysis-based clustering achieved 83% fault detection and 88% accuracy. A flexible BESS with Proximal Policy Optimization-based control efficiently optimizes multi-terminal operation to enhance operational time and performance balance [121].

Challenges for ML include data acquisition, time and resource constraints, interpretation of results, and software development. Large volumes of data are needed for accurate training, while time and resources are required for algorithm development. Additionally, ML projects require costly software infrastructure and are highly susceptible to errors [88]. Battery degradation significantly impacts safety and sustainability, making data-driven models for SOH estimation gain attention. Ref. [104] reviews state-of-the-art models, benchmarks, and publicly available datasets. However, challenges with degradation and increased planning costs necessitate diagnosis and RUL determination. The ML models, specifically the XGBoost algorithm, improved predictions and contributed to optimizing the planning and operations of Li-ion batteries by estimating the RUL [108]. Grid operators play a crucial role in the energy transition, using advanced supervisory systems, AMI, DCS, SCADA, and high-speed technologies like 5G for real-time data acquisition, predictive diagnostics, proactive responses to anomalies, fault detection, and asset optimization. Effective AI deployment requires better data, preprocessing, and interpretable models, and integration of new data sources, including advanced sensors. Future directions include multi-task learning, USL, and specialized algorithms to enhance sustainable energy management.

5. Techno-Economic and Environmental Impacts of GS-BESS

This section addresses RQ3: What are the key operational, technical, economic, and environmental challenges that GS-BESS faces?

5.1. Technological Impact

The increasing penetration of VRE with dispatchable technologies like GS-BESS at both distributed and utility scales is transforming power system operations [122]. Widely used applications include FTM, Front-the-meter, which offer bulk energy services, ancillary services, grid support, and renewable energy integration, and BTM, behind-the-meter, accounts for 30% of the global Li-ion BESS market and provides customer energy management [123], reduces electricity expenses, increases self-consumption, and increases power stability [124], high upfront costs challenge economic feasibility and [125] identifies AI and LLMs as promising emerging approaches for load forecasting in BTM DERs.

BESS project parameters include battery technology, co-location coupling setup, capacity, and charge/discharge cycles. Co-located projects reduce costs, standalone projects offer flexibility, and hybrid PPAs enhance the financial viability of renewable projects. Li-ion and flow batteries are popular technologies for long-duration storage due to their high energy density, efficiency, and low self-discharge rates, but they have lower energy density and higher initial capital costs. The BMS regulates battery function through electrical, mechanical, and technical means, increasing reliability and lifespan through charge–discharge, temperature, cell potential, current, and voltage monitoring techniques [126]. Despite the promising potential of GS-BESS and their optimization through AI, several critical challenges hinder their widespread deployment and scalability. The lack of standardized regulations across jurisdictions complicates cross-border deployment, increasing costs and preventing the development of universally accepted benchmarks. AI integration in BESSs also complicates communication, leading to inefficiencies and grid instability. Emerging economies’ power grids lack flexibility, digitalization, and control for high-penetration BESS integration, necessitating infrastructure upgrades and adaptive regulatory frameworks. AI-powered BESS operations rely on large data volumes, raising concerns about data ownership, cybersecurity, and user privacy. The growing deployment of GS-BESS supports grid stability, frequency regulation, and system modernization and energy system evolution [127]. They maintain real-time generation and demand balance of power systems, and advancements in battery technologies have made them cheaper and inevitable for future power networks and a significant enabler of grid modernization, resilience, and energy system evolution [128].

5.2. Economic Impact

Technological advancements, declining battery prices, and policy discussions emphasize the economic appeal of BESS, driving the transition to sustainable energy. The increasing demand for sustainable energy and efficient energy storage drives the optimization of BESS operations across multiple markets, including wholesale and frequency restoration reserves, requiring new optimization techniques and business models. Volatility in electricity markets presents opportunities for BESS operators to charge and discharge during periods of renewable output [129]. However, uncertainty and risk must be considered for investment viability, necessitating innovative business models to ensure optimal operations and revenue maximization. BESS can operate across wholesale, ancillary service, capacity, and RES support markets, enhancing financial stability when individual revenue streams are insufficient. Ref. [130] evaluates the payback period, return on investment, and net present value. AI ensures secure bilateral energy trading and real-time optimal pricing for market participants [131]. A novel revenue stacking approach for Day-Ahead and automatic Frequency Restoration Reserve markets improved profits by 17.3% by modeling uncertainty and ensuring reliable real-time delivery [132]. Decision-Focused Learning, which integrates decision outcomes into training, enhanced profits by 9.5% over traditional Predict-Then-Optimize methods in the Belgian secondary reserve market [133].

Integrating AI with VRE and BESS enhances both economic and operational performance. Combining photovoltaics with storage increased low-cost energy purchases from 25% to 91% and reduced electricity bills by 55% without additional costs, while enhancing grid stability [134]. In deregulated markets, BESS enables energy arbitrage, and supports renewable energy integration, and ML market-oriented strategies, including roles as price taker or maker, fostering a policy–market symbiosis [135]. Findings in GS-BESS allocation and cost trends [136] reduce economic costs in Germany’s electricity system using agent-based simulations, and [137] identify Romania, Latvia, Lithuania, and Estonia as the most profitable markets, while Spain, Portugal, and Norway remain unprofitable from the allocation of 25 European countries, and using the wild geese algorithm [138], in Morocco [139], achieves 70% renewable energy share at a low levelized cost by comparing lead-acid, lithium-ion, and vanadium-redox flow batteries, and ref. [140] improves the financial benefits of nine industrial load profiles. Study [141] examines the role of block-chain technology and multi-microgrid coordination in energy markets. A block chain-based VPP framework integrating residential DERs and BESSs enables secure, transparent, and automated energy trading while optimizing storage utilization, with simulations showing up to 76% higher participant profits, 22% lower grid energy consumption, and 38% reduction in peak imports [142]. Virtual Power Plants (VPPs) aggregate DERs to enhance decentralized energy management [143]. The findings raise awareness of battery storage technology and promote grid-connected storage in deregulated markets.

Despite its critical role, BESS is often undervalued and overlooked across markets. A one-hour device provides nearly half the capacity value, while a four-hour device can achieve almost the full capacity value [144]. Market viability remains limited for many technologies, highlighting the need for technical advancement or market reform [145]. Operational strategies, market design, and high cycle costs influence GS-BESS profitability, while higher flexibility demand increases utilization time in the Spanish electricity market [146], and in long-term integration, it reduces costs and promotes renewable energy; it does not eliminate operational strategy optimization [147].

The cost breakdown of a BESS is categorized into the following: (a) CAPEX includes battery costs, power conversion systems, installation and construction, control and monitoring systems, electrical interconnection, ancillary equipment, and other associated costs like environmental studies and legal fees and (b) OPEX involves maintenance and repairs, operations and monitoring personnel costs, insurance and permits, grid connection fees, and costs for decommissioning and disposal at the end of life cycle [148]. The lifecycle of a battery is directly associated with the CAPEX and OPEX, with replacement included in the OPEX. Lifecycle considerations, particularly battery degradation, impact revenue and lifespan. Participation in contingency markets increases revenue with minimal aging effects [91]. Studies on financial and economic viability employ various decision-making criteria, as detailed in Equations (16)–(22) [149]. Some studies conducted comprehensive financial feasibility assessments using multiple approaches to inform investor decisions. Others focused on technical characteristics or other factors (environmental, technical, scale, and technology), incorporating economic variables or financial metrics. The payback period (PB) is used as a simpler alternative to Net Present Value (NPV), though it overlooks the time value of money and is prone to imprecision. Other considerations include cost/financial variables, cost–benefit ratios, Monte Carlo simulation (MCS), and real options (RO) theory. Among the reviewed literature, 36 articles explicitly addressed economic evaluation of GS-BESS. The most commonly adopted economic indicators include LCOS, NPV, and IRR. Specifically, 16 studies jointly employed IRR, LCOS, and NPV, while 15 studies focused on LCOS and NPV only, reflecting their dominance in cost–benefit and investment feasibility assessments. The remaining five studies utilized payback period or/and cost-based indicators. This distribution highlights a strong preference for LCOS- and NPV-based evaluation frameworks in GS-BESS techno-economic analyses.

Net present value (NPV):

$$NPV=-I+\sum _{j=1}^n{\displaystyle \frac{C{F}_j}{{(1+k)}^j}}$$

(16)

Net present cost (NPC), or lifecycle cost analysis (LCCA):

$$NPC=C_0+C_R+C_{O\&M}-C_{RV}=C_0+\sum _{j=1}^n{\displaystyle \frac{C_{R_j}}{{(1+k)}^j}}-{\displaystyle \frac{C_{R{V}_n}}{{(1+k)}^n}}$$

(17)

Internal rate of return (IRR):

$$0=-I+\sum _{j=1}^n{\displaystyle \frac{C{F}_j}{{(1+k_{IRR})}^j}}$$

(18)

Discounted cash flow (DCF):

$$DCF={\displaystyle \frac{C{F}_f}{{(1+k)}^f}}$$

(19)

Discounted payback (DPB):

$$DPB=L+{\displaystyle \frac{C}{A}}$$

(20)

Equivalent annual cost (EAC):

$$EAC=NPC{\displaystyle \frac{{(1+k)}^{n}k}{{(1+k)}^n-1}}$$

(21)

Levelized costs of electricity (LCOE):

$$LCOE={\displaystyle \frac{{\sum }_{j=1}^n\frac{C{F}_j}{{(1+k)}^j}}{{\sum }_{j=1}^n\frac{C_{T_j}}{{(1+k)}^j}}}$$

(22)

GS-BESS is growing due to cost reduction and flexibility, crucial for supporting the rapid increase in RESs in the power network, and for ancillary services, but their longevity and profitability depend on considering each available energy source and subordinate storage efficiency. Long-term techno-economic analyses indicate a break-even investment cost of 210 $/kWh for energy arbitrage applications. Li-ions offer high power, energy density, and low cost, and a data-driven framework for characterization [29,150,151,152], while enhanced frequency response and peak shaving strategies influence profitability and grid returns [153].

According to a Cognitive Market Research analysis, the energy storage and distribution market is expected to reach USD 230 billion by 2024, with Li-ion batteries dominating due to their efficiency and cost-effectiveness. Solid-state battery technology is also growing, with a 5% market share. The integration with RESs is increasing, with BESS in 30% of new solar installations and microgrids and smart grid technologies being implemented. The US leads the market, accounting for 60%, followed by China and Japan. The market is influenced by economic, social, technological, political, environmental, and legal factors. Key players include Tesla, LG Energy Solution, Samsung SDI, and Panasonic Corporation [154]. Financial feasibility is addressed in [155], which examines the long-term techno-economic potential of integrating BESS with a wind farm for wholesale energy arbitrage and wind curtailment mitigation, and in [156], introduces a Model Predictive Control method based on reinforcement learning to optimize multi-agent battery storage under price and demand uncertainty. Intelligent GS-BESS deployment has significant economic implications, with tools such as payback period analysis, storage-levelized cost, and cost–benefit analysis evaluating financial viability. Revenue streams can be generated through frequency regulation, demand response, and energy arbitrage; ML improves predictive accuracy and operational efficiency; and government subsidies and incentives promote BESS adoption. Financial modeling must now incorporate ML-driven operational strategies to reduce maintenance costs and accelerate investment returns.

5.3. Environmental Impact

Global energy demand in both developed and developing nations is driving efforts to cut carbon emissions and promote sustainable power generation and distribution through advanced storage technologies. GS-BESS is crucial to decarbonizing the energy sector by supporting the integration of renewables and replacing fossil-fuel-based plants. Ref. [157] highlights the potential for renewable energy to decarbonize 90% of the electricity industry by 2050. AI promotes energy transition and reduces carbon emissions. Ref. [158] shows using data from 69 countries from 1993 to 2019, with trade openness playing a mediating role, and it acknowledges the potential for counterarguments and suggests policymakers should consider energy consumption. However, environmental concerns persist with the extraction of raw materials, including lithium and cobalt, manufacturing, operation, and recycling [159], and ref. [160] showed that the cost per unit capacity of secondary utilization is lower than 63.99% of that of conventional batteries.

LCA methods, often augmented by AI, quantify environmental impacts and minimize ecological footprints, helping decision-makers select environmentally responsible solutions by assessing energy flows, material inputs, and emissions across a product’s lifecycle. Study [161] proposes six steps to achieve net-zero industrial carbon emissions by mid-century, drawing on lessons from low- and middle-income countries, including quintupling financing, expediting technology transfer, investing in human resources, and setting binding targets. A case study in Germany reveals that maximum-profit arbitrage increases system emissions by up to 7.5 tCO2 per MWh, with 60% avoided with minimal profit sacrifice [30]. A comparative study [162] finds Li-ions superior due to higher efficiency and lower environmental impact, while LAES offers flexibility that offsets its higher electricity consumption; cogeneration of LAES achieves the best environmental outcomes. LCC considers all expenses, from investment to recycling, and emphasizes the need to weigh each battery type’s social, environmental, and economic trade-offs [163]. Physics-informed neural networks offer a rapid and precise physics-based modeling solution, significantly surpassing conventional computational fluid dynamics and finite element method methods, facilitating real-time thermal management in high-power electronics and battery systems [116].

6. Policy and Regulations in GS-BESS

Supportive policy and regulatory frameworks are essential to make the shift to a smart, sustainable, and resilient energy system. This section addresses RQ4: How are policy frameworks and regulatory structures evolving to support the deployment and integration of AI-enabled GS-BESS? The rapid development of BESS is putting pressure on testing, certification, and standard development, influencing the evolving regulatory environment. GS-BESS regulations are complex, varying across jurisdictions and systems, and technological advancements and interoperability challenges can heighten risks in safety, fire prevention, and environmental regulations. To achieve this, policymakers should encourage effective regulations for innovative solutions in integrating GS-BESS into the grid, intelligent optimization, fair market access, cybersecurity, and adapting technologies [164]. Policies such as renewable energy targets, feed-in tariffs, net metering, and subsidies can accelerate adoption, while public–private partnerships can drive innovation, reduce costs, enable large-scale projects, support sustainable business models, and scale ML tools, and capacity-building programs can develop the next generation of energy professionals.

BESS and power grid systems are governed by international standards like IEC 61508 and IEC 62933, which focus on functional safety and safety integrity levels. Regional regulations, such as the EU Battery Regulation 2023/1542, aim to enhance battery sustainability, safety, and circularity throughout their lifecycle, including mandatory sustainability requirements, labeling, due diligence, producer responsibility, digital passports, and waste management targets. Countries are adopting diverse BESS policies such as [25], a three-phase policy framework, focusing on Asia, Europe, and the U.S (I) promotes BESS deployment with economic incentives and safety policies, (II) addresses waste management with recycling initiatives, and (III) focuses on long-duration BESS advancements, requiring policy refinement and adaptation. The evolution of policies and regulations supporting BESS development, utilization, and sustainability, highlighting environmental, economic, political, technological, and regulatory factors influencing their viability and growth reviewed in China, Japan, and South Korea [165], in Indonesia, Malaysia, the Philippines, Thailand, and Vietnam [166]; however, Thailand’s lack of a regulatory framework for BESS complicates interconnection, safety, and market participation [167].

Governments can encourage through incentives such as tariffs, tax credits, and compensation schemes, while advances in battery chemistry and a digitalized smart grid can address cost and sustainability challenges, extend lifespan, reduce environmental impacts, and maximize operations and revenue streams. The global focus on achieving net-zero emissions under SDG 13 and on providing affordable and clean energy under SDG 7 creates a favorable environment for BESS adoption. BESS can also be beneficial in regions with limited infrastructure, such as remote or island grids, by reducing diesel dependency, improving power reliability, and enabling higher RESs penetrations. However, there are several limitations to overcome, including regulatory and policy gaps, economic feasibility, materials and sustainability challenges, scalability for long-duration storage, and the difficulty in monetizing benefits like improved resilience and emission reductions [168], and high upfront costs, degradation, and lack of marketplaces. To overcome these, stakeholders must monitor, update products, participate in standards development, adjust taxation, and promote sustainability and environmental impact, including recycling and reuse of old batteries [169].

7. Conclusions and Future Directions

7.1. Conclusions

This review investigates how AI-based intelligent optimization in GS-BESS increased operational efficiency, lowering costs, minimizing environmental impacts, improving energy dispatch, forecasting, and real-time decision-making, and supporting decarbonization and the transition to a flexible and efficient energy system. These systems increase grid flexibility and resilience, enable critical services such as peak shaving and frequency regulation, reduce carbon emissions, and improve safety, chemistry, and system interoperability.

GS-BESSs are essential to modern energy systems, from micro-grids to utility-scale applications, providing grid stability, enabling renewable energy integration, load leveling, and emergency power. They address critical operational, economic, and managerial dimensions of challenges in modern power grids. Economically, they enable energy arbitrage, reduce peak demand charges, and defer costly infrastructure up-grades. Managerially, they support the integration of VRE sources, improve demand-side management, and offer flexibility in grid planning and operation. The evolution of service models like ESaaS, along with advancements in mobile and second-life storage technologies, expands their utility. Moreover, the integration of AI is opening new avenues to optimize BESS operations, enhancing safety, efficiency, and reliability. Deployment configurations of GS-BESS include standalone, integrated, aggregated, and virtual, and its integration with energy generation, consumption, and storage networks, with technical and economic advancements. In this review, AI-driven methods for GS-BESS are shown to significantly improve various metrics. Data-driven and hybrid AI approaches achieved state-of-charge (SoC) estimation accuracies over 98% with RMSE below 0.75%. Fault detection models reached up to 83% detection rates and accuracy above 88%, outperforming traditional methods. Additionally, hybrid AI-physics frameworks reduced prediction errors for remaining useful life (RUL) estimation to below 2% RMSE and 1% MAPE, highlighting the technical advantages of AI integration in energy storage systems.

Despite these advancements, challenges remain, including high capital costs, degradation issues, a lack of standards that limit their wider use, data quality, model generalization across many systems, and the computational scalability of advanced algorithms. AI-based models advance battery state estimation and performance prediction but lack an understanding of battery aging mechanisms, as they rely on electrochemical data from charge–discharge cycles. High-dimensional, small-sample datasets and complex battery operating environments lead to the “curse of dimensionality,” reducing model accuracy and generalization.

Achieving these requires robust, scalable AI frameworks that utilize real-time data, advanced thermal and degradation management, next-generation chemistries, predictive maintenance, and energy arbitrage. Advanced modeling, AI-driven control strategies, and adaptive systems are also crucial. As BESSs mature, their seamless integration with VRE supply and demand conditions will be crucial for the development of intelligent, sustainable energy infrastructure. Future research should focus on light-weight, interpretable, and adaptable models, high-quality input data, and integration of AI with high-throughput computational methods to accelerate R&D while ensuring safety and material practicality. Overcoming data transparency and standardized evaluation metrics improves the development of intelligent EMS and the accuracy of techno-economic modeling and promotes more robust and scalable BESS applications. Ultimately, GS-BESS, optimized through AI, holds the key to achieving a sustainable, resilient, and economically viable energy future.

• Key conclusions:

• AI-driven intelligent optimization significantly improves GS-BESS efficiency, cost-effectiveness, forecasting accuracy, real-time control, and environmental performance.
• GS-BESS is essential for grid flexibility and resilience, enabling renewable energy integration and services such as peak shaving and frequency regulation.
• GS-BESS provides strong economic and managerial benefits, including energy arbitrage, peak demand reduction, deferred infrastructure upgrades, and flexible grid operation.
• AI integration enhances system safety, reliability, interoperability, and optimization across diverse GS-BESS deployment configurations.
• Key challenges persist, including high capital costs, battery degradation, lack of standardization, data quality limitations, and scalability of AI models.
• Future research should prioritize interpretable and adaptive AI models, high-quality data, standardized metrics, and next-generation storage technologies.

7.2. Future Research Recommendations

Future research should integrate AI with physics-informed models and multi-agent control systems to improve performance and interoperability. Priorities include

• Developing advanced AI- and IoT-based models for accurate weather, demand, and VRE forecasting.
• Exploring new chemistries, hybrid ESS, and supercapacitors to extend lifespan, cut costs, and boost sustainability.
• Optimizing DER dispatch, enabling grid resilience, and supporting virtual power plant formation are also crucial.
• Data privacy and resilient autonomous operation are ensured through adaptive communication and secure control systems.
• Policy impacts on GS-BESS adoption and ML-driven optimization must be assessed, and regulatory frameworks must adapt to AI-optimized BESS in energy markets.
• Explore quantum computing, AI for autonomous grid management, and blockchain for decentralized storage and peer-to-peer trading.
• The convergence of GS-BESS and AI-based intelligent optimization is transforming modern energy systems, improving flexibility, resilience, techno-economic, and environmental benefits.