Next Article in Journal
Parametric Study on Effective Thermal Conductivity of Dispersed Disks with Internal Heat Sources
Previous Article in Journal
Implications of Spatial Reliability Within the Wind Sector
Previous Article in Special Issue
Novel Artificial Intelligence Applications in Energy: A Systematic Review
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Artificial Intelligence Applications for Energy Storage: A Comprehensive Review

Department of Electrical and Electronic Engineering, Imperial College London, London SW7 2AZ, UK
*
Author to whom correspondence should be addressed.
Energies 2025, 18(17), 4718; https://doi.org/10.3390/en18174718
Submission received: 16 July 2025 / Revised: 22 August 2025 / Accepted: 1 September 2025 / Published: 4 September 2025
(This article belongs to the Special Issue Optimization and Machine Learning Approaches for Power Systems)

Abstract

The integration of artificial intelligence (AI) and machine learning (ML) technologies in energy storage systems has emerged as a transformative approach in addressing the complex challenges of modern energy infrastructure. This comprehensive review examines current state of the art AI applications in energy storage, from battery management systems to grid-scale storage optimization. We analyze various AI techniques, including supervised learning, deep learning, reinforcement learning, and neural networks, and their applications in state estimation, predictive maintenance, energy forecasting, and system optimization. The review synthesizes findings from the recent literature demonstrating quantitative improvements achieved through AI integration: distributed reinforcement learning frameworks reducing grid disruptions by 40% and operational costs by 12.2%, LSTM models achieving state of charge estimations with a mean absolute error of 0.10, multi-objective optimization reducing power losses by up to 22.8% and voltage fluctuations by up to 71%, and real options analysis showing 45–81% cost reductions compared to conventional planning approaches. Despite remarkable progress, challenges remain in terms of data quality, model interpretability, and industrial implementation. This paper provides insights into emerging technologies and future research directions that will shape the evolution of intelligent energy storage systems.

1. Introduction

The global energy transition represents humanity’s shift from fossil-fuel-based energy systems to renewable, sustainable alternatives, driven by climate imperatives and technological advances. This transition fundamentally challenges existing energy infrastructure designed for centralized, dispatchable generation. Unlike traditional power plants that can adjust output on demand, renewable sources like solar and wind are inherently variable and intermittent, creating unprecedented challenges for grid stability and reliability. Energy storage emerges as the critical enabler of this transition, serving multiple essential functions.
First, it decouples energy generation from consumption, allowing excess renewable energy to be stored during periods of high generation and released during peak demand or low generation. Second, storage provides grid stabilization services, including frequency regulation, voltage support, and spinning reserves, traditionally supplied by fossil fuel plants. Third, it enables microgrids and distributed energy resources to operate independently or in coordination with the main grid, enhancing resilience and enabling energy access in remote areas. Fourth, storage facilitates demand-side management by allowing consumers to shift their energy use patterns without sacrificing service quality. The complexity and scale of modern energy storage applications exceed the capabilities of traditional control systems.
Conventional controllers, designed for predictable, linear systems, struggle with the nonlinear dynamics of battery chemistry, the stochastic nature of renewable generation, and the multi-objective optimization required for economic operation. AI becomes essential for several reasons. Machine learning algorithms can capture complex patterns in historical data to predict renewable generation and demand with accuracy unattainable by physical models alone. Deep learning methods can model the nonlinear aging processes in batteries, enabling precise state estimation and predictive maintenance. Reinforcement learning can optimize real-time decisions in uncertain environments, balancing multiple objectives like cost minimization, battery life extension, and grid stability. Furthermore, AI’s ability to process vast amounts of data from distributed sensors enables coordinated control of thousands of distributed storage units, creating virtual power plants that would be impossible to manage with traditional methods. Within this context, AI-enabled energy storage represents not merely a technical improvement but a cornerstone of energy transition. It transforms storage from a passive buffer to an intelligent system capable of predictive, adaptive, and autonomous operation. This transformation is essential for achieving renewable energy targets, as studies indicate that reaching 80% renewable penetration requires storage capacity and intelligence far beyond current capabilities. The convergence of AI and energy storage thus emerges as one of the most critical technological challenges and opportunities of our time.
The global transition towards sustainable energy systems has intensified the demand for efficient and reliable energy storage solutions. The mounting pressure for efficient and sustainable energy solutions has driven AI adoption in contemporary energy systems [1]. Energy storage systems play a crucial role in integrating renewable energy sources, managing grid stability, and providing backup power during outages. However, the complexity of modern energy storage systems, particularly lithium-ion batteries (LIBs), presents significant challenges in terms of performance optimization, safety management, and lifecycle prediction [2].
The development of affordable, environmentally acceptable energy storage devices is required to address the present energy problem and to offer a viable solution for renewable energy sources with intermittency. The inherent variability of renewable energy sources, such as solar and wind power, poses significant challenges for grid integration and stability. Energy storage systems play a pivotal role in mitigating these challenges by storing excess energy during periods of high generation and releasing it during periods of low generation or high demand. Using electricity/energy storage technologies to stabilize the output of solar and wind power could increase their share in power generation and ease their integration in the electric grid [3]. This capability is essential for achieving higher renewable energy penetration rates and maintaining grid reliability.
Artificial intelligence (AI) is revolutionizing the development and optimization of LIBs, which are critical in modern technologies like energy storage systems and electric vehicles (EVs) [4]. The application of AI techniques in energy storage has gained substantial momentum in recent years, driven by the need to address challenges such as battery degradation prediction, state estimation, thermal management, and system optimization.
The integration of AI in energy storage systems offers numerous advantages including enhanced predictive capabilities, real-time optimization, improved safety monitoring, and extended system lifespan [5]. AI techniques, including machine learning models like ensemble methods, support vector machines, and neural networks, have been instrumental in predictive maintenance, state of charge (SoC) and state of health (SoH) estimation, and materials discovery [6]. These applications enable more accurate predictions of battery behavior, optimize charging and discharging cycles, and facilitate the development of advanced energy management strategies.
This work provides a comprehensive review of optimization techniques using AI for energy storage systems within renewable energy setups. The scope of AI applications in energy storage extends beyond individual battery cells to encompass entire energy storage systems, microgrids, and large-scale grid applications. This comprehensive integration requires sophisticated algorithms capable of handling multiple variables, uncertainties, and real-time constraints.
The implementation of AI techniques in energy storage systems presents significant technical and practical difficulties that must be acknowledged. The inherent complexity of battery systems, with their nonlinear electrochemical behaviors, multi-scale physical phenomena, and degradation mechanisms that span years, creates fundamental challenges for AI model development and deployment. Real-world energy storage systems operate under highly variable conditions with incomplete and noisy sensor data, making it difficult to develop models that generalize well across different operating scenarios. Furthermore, the safety-critical nature of energy storage applications demands extremely high reliability and interpretability standards that many current AI approaches struggle to meet. The computational constraints of embedded battery management systems, the need for real-time decision making, and the requirement for models to remain accurate over the entire system lifetime add additional layers of complexity. These difficulties are compounded by the limited availability of the high-quality, long-term operational data needed to train robust AI models, as battery degradation studies often require years of continuous monitoring. Understanding and addressing these challenges is essential for advancing from laboratory demonstrations to practical industrial deployments.
The motivation for this review stems from the rapid evolution of AI technologies and their increasing adoption in energy storage applications. This literature review consolidates evidence from more than 20 recent studies on AI-based approaches for renewable energy and smart grid management. It discusses AI methods like machine learning, deep learning, reinforcement learning, and optimization techniques applied in energy forecasting, load management, fault detection, and demand response. Understanding the current state of research, identifying gaps, and exploring future directions are essential for advancing the field and maximizing the potential of AI-driven energy storage solutions.
This comprehensive review examines 155 peer-reviewed publications, with particular emphasis on recent developments in the field. The review spans literature from established foundational works to the most current research published in 2025, capturing the rapid evolution of AI applications in energy storage systems. The review methodology involved systematic searches across major scientific databases including IEEE Xplore, ScienceDirect, SpringerLink, and MDPI, using keywords combining AI techniques (machine learning, deep learning, reinforcement learning, optimization) with energy storage applications (battery management, grid integration, thermal control). Papers were selected based on their technical contribution, methodological rigor, and practical relevance, with priority given to studies demonstrating quantitative results and real-world applications. The temporal distribution of reviewed papers reflects the accelerating pace of research in this field, with a significant concentration of publications in recent years corresponding with advances in both AI capabilities and energy storage deployment. The depth of analysis ranges from detailed examination of novel AI architectures and their performance metrics to comparative assessments of different approaches for specific applications, ensuring both breadth of coverage and technical depth where innovations warrant closer scrutiny.
While several reviews have examined specific aspects of AI applications in energy storage, significant gaps remain in the literature. Existing reviews typically focus on narrow technical domains, such as battery state estimation or grid integration, failing to provide a holistic view of how AI technologies interconnect across the entire energy storage ecosystem. Moreover, previous reviews have not adequately addressed the critical intersection between technical AI advances and practical implementation challenges, nor have they examined the economic implications through frameworks such as real options theory. The rapid evolution of AI technologies, particularly in areas such as federated learning, explainable AI, and physics-informed neural networks, has outpaced existing reviews, creating a knowledge gap for researchers and practitioners seeking current, actionable insights. Additionally, no existing review provides a unified framework that bridges the gap between fundamental AI techniques and their specific adaptations for energy storage applications, from cell-level battery management to grid-scale optimization. This comprehensive review addresses these gaps by providing an integrated analysis that spans multiple scales, technologies, and application domains, while also examining implementation challenges, economic considerations, and future research directions in a manner accessible to both AI specialists and energy storage practitioners.
This review makes several novel contributions that distinguish it from existing literature. First, it provides the first comprehensive integration of economic valuation frameworks, specifically real options theory (Section 9), with technical AI applications in energy storage, revealing how option value can reach £12.9 billion in national-scale deployments and demonstrating why traditional NPV approaches systematically undervalue flexible storage technologies. Second, unlike previous reviews that focus on either algorithms or applications, we present bidirectional mapping between AI techniques and energy storage challenges, enabling both AI researchers to identify relevant applications and energy storage practitioners to select appropriate AI tools. Third, we provide critical analysis of model transferability and generalization challenges, identifying this as a fundamental barrier to commercial deployment. Fourth, this review uniquely examines the complete innovation pipeline from materials discovery through grid-scale deployment, revealing critical gaps where laboratory successes fail to translate to industrial applications. Fifth, we provide the first comprehensive analysis of safety-critical AI applications in energy storage, including thermal management and fire prevention systems—areas largely overlooked in previous AI-focused reviews. Finally, by encompassing 155 recent publications with quantitative performance metrics, this review offers the most current and comprehensive assessment of the field’s rapid evolution, particularly capturing the transformative developments in federated learning, physics-informed neural networks, and explainable AI that have emerged since 2023.
Despite significant progress, several critical gaps remain in the application of AI to energy storage systems. First, model generalization deficiencies persist, with existing AI models demonstrating limited ability to transfer learning across different battery chemistries, requiring extensive retraining for each new battery type. Second, edge computing bottlenecks continue to hinder real-time deployment, as current AI models are often too computationally intensive for implementation in resource-constrained battery management systems, creating delays in practical applications. Third, there is a notable lack of multi-physics coupling in current AI approaches, with slow progress in developing models that can simultaneously handle the complex interactions between electrical, thermal, and mechanical phenomena in battery systems. Addressing these gaps is crucial for the next generation of AI-enabled energy storage technologies.
This review addresses four key questions to advance the understanding of AI applications in energy storage systems. First, regarding what will be reviewed, we examine AI techniques spanning machine learning, deep learning, and reinforcement learning as applied to battery management systems, energy storage optimization, grid integration, and safety applications, with particular focus on practical implementations that have demonstrated quantitative improvements. Second, concerning source selection and analysis, we analyze peer-reviewed publications selected from major scientific databases based on their technical rigor, practical relevance, and contribution to advancing real-world applications, with emphasis on studies published between 2020 and 2025, in order to capture recent technological advances. Third, the purpose of this review is to bridge the gap between AI research and energy storage applications by providing both researchers and practitioners with a unified framework for understanding which AI techniques are most effective for specific energy storage challenges, while critically evaluating their limitations and implementation constraints. Fourth, this review’s contribution is threefold: it provides the first comprehensive mapping of AI techniques to specific energy storage applications across multiple scales (from cell-level to grid-scale); it offers a critical comparative analysis of different approaches highlighting practical trade-offs often overlooked in individual studies; it identifies specific technical gaps and future research directions needed to transition from laboratory demonstrations to industrial deployment. By addressing these questions systematically, this review serves as both a technical reference and a strategic guide for advancing AI-enabled energy storage systems. Critical future work directions include developing robust multi-physics AI models that can handle coupled electrical–thermal–mechanical phenomena, creating federated learning frameworks for collaborative model development while preserving data privacy, advancing explainable AI techniques for safety-critical applications, and establishing standardized benchmarks and datasets to enable fair comparison of different approaches. The review concludes with a comprehensive roadmap for transitioning from current laboratory demonstrations to industrial-scale deployments.
To provide a structured overview of the extensive literature in this field, we have developed a comprehensive classification framework for AI applications in energy storage systems, as illustrated in Figure 1. This classification organizes the reviewed works into five main categories based on their primary application domain, with further subdivisions reflecting specific technical approaches and methodologies.

Review Framework and Methodology

This narrative review employs a structured analytical framework to evaluate AI applications in energy storage systems. While not following systematic review protocols, our approach ensures comprehensive coverage and critical analysis through six guiding questions:
Effectiveness Assessment: How effective are AI techniques in improving energy storage performance? We evaluate effectiveness through reported quantitative metrics including prediction accuracy (MAE, RMSE), optimization improvements (cost reduction, efficiency gains), and operational benefits (reduced downtime, extended lifetime). Each technique is assessed within its application context rather than through direct comparison, recognizing that effectiveness is application-specific.
Limitations Analysis: What limitations constrain AI deployment in energy storage? We examine technical limitations (computational complexity, data requirements), practical constraints (sensor accuracy, embedded system resources), and fundamental challenges (generalization across battery chemistries, long-term reliability). This analysis draws from both reported failures and implicit limitations evident in experimental conditions.
Contextual Factors: What conditions influence AI performance? We identify key factors including data quality and availability, system scale (cell- versus grid-level), operational environment (laboratory versus field conditions), and safety requirements. The review highlights how these factors often determine success more than algorithm choice.
Comparative Evaluation: How do AI approaches compare with traditional methods and with each other? Rather than declaring universal superiority, we examine trade-offs between accuracy and complexity, development time versus performance gains, and robustness versus optimality. Comparisons acknowledge that different applications may favor different approaches.
Contribution Assessment: What unique contributions and redundancies exist? We map novel contributions while identifying areas where multiple studies address similar problems with marginal differentiation. This helps distinguish genuine advances from incremental variations.
Research Gaps: What future research is needed? We systematically identify gaps based on unsolved problems, scalability challenges, and barriers to industrial adoption. Priority is given to gaps that block practical deployment rather than theoretical limitations.
Article Selection Process: Papers were identified through searches in IEEE Xplore, ScienceDirect, SpringerLink, and MDPI, using combinations of AI and energy storage terms. Selection prioritized studies with quantitative results, real-world validation, and clear methodological descriptions. The final 155 papers represent diverse AI techniques, energy storage applications, and geographic regions; they have been selected to provide comprehensive coverage rather than exhaustive enumeration. The temporal emphasis on recent publications (2020–2025) reflects the rapid evolution of both AI capabilities and energy storage deployment.
This framework structures our analysis throughout the review, with Section 2, Section 3, Section 4, Section 5 and Section 6 presenting the state of the art, Section 7 addressing limitations and comparisons, Section 8 identifying research gaps, and Section 9 examining economic considerations through option value theory.

2. AI Techniques in Energy Storage Systems

Before examining specific AI techniques, it is helpful to provide an overview of how AI is being applied across different domains within energy storage systems. This review examines AI applications ranging from cell-level battery management to grid-scale integration. Table 1 presents a structured overview of the AI applications covered in this review, organized by application domain.
As shown in Table 1, the reviewed literature demonstrates that AI applications in energy storage span multiple scales and domains, from fundamental battery cell management to complex grid-scale optimization. While this review primarily focuses on lithium-ion battery systems due to their market dominance and the extensive research base, the AI techniques discussed are increasingly being adapted to other storage technologies as they mature.
To complement the application domain perspective provided in Table 1, Table 2 presents a technique-centric view that maps specific AI methods to their primary objectives within energy storage systems. This mapping helps researchers and practitioners identify which AI techniques have proven most effective for particular challenges.
This technique-centric mapping reveals that certain AI methods are particularly suited to specific challenges in energy storage systems. Time-series methods like LSTM dominate in prediction tasks, while reinforcement learning excels in control and optimization applications. The choice of AI technique often depends on the specific constraints of the application point, such as computational resources in embedded systems or the need for interpretability in safety-critical applications.

2.1. Machine Learning Fundamentals

Machine learning (ML) has emerged as a cornerstone technology for energy storage applications, offering a powerful tool for pattern recognition, prediction, and optimization. ML mechanisms have been classified into supervised, unsupervised, and deep learning approaches [63], which have been practically reviewed with respect to renewable and hybrid energy demand applications. The results indicate that supervised learning is mainly applied to classification regression, and unsupervised learning to clustering. The application of ML in energy storage systems leverages various algorithmic approaches to address specific challenges and requirements [64].
Supervised learning techniques have found extensive applications in energy storage systems, particularly for state estimation and performance prediction [65]. ML techniques enable accurate and effective data-driven predictions in certain situations [14]. These methods use labeled datasets to train models that can predict battery parameters such as state of charge, state of health, and remaining useful life [66]. Support vector machines, random forests, and neural networks are among the most commonly employed supervised learning algorithms in this domain.
Unsupervised learning techniques play a crucial role in discovering hidden patterns and anomalies in energy storage system data [67]. Results indicate that supervised learning is mainly applied to classification regression, and unsupervised learning to clustering. These methods are particularly valuable for fault detection, system monitoring, and identifying operational patterns that may not be immediately apparent through traditional analysis methods [68].

2.2. Deep Learning Applications

Deep learning has revolutionized the field of energy storage by providing advanced capabilities for handling complex, high-dimensional data and nonlinear relationships [61]. In the initial stage, three deep learning (DL) models, stacked long short-term memory networks (stacked LSTMs), gated recurrent unit (GRU) networks, and stacked recurrent neural networks (SRNNs) are developed based on the training of six input features. These architectures are particularly well-suited for time-series prediction tasks that are common in energy storage applications [69].
While various deep learning architectures have been applied to energy storage applications, each offers distinct advantages and trade-offs. GRUs provide similar performance to LSTMs but with reduced computational complexity due to their simplified gating mechanism, making them more suitable for edge computing applications in battery management systems. Temporal convolutional networks (TCNs) have emerged as a promising alternative for time-series modeling in energy applications, offering parallel processing capabilities and longer effective memory compared to recurrent architectures.
TCNs are a class of convolutional architectures specifically designed for sequence modeling tasks. Unlike recurrent networks that process sequences step by step, TCNs use dilated causal convolutions to capture temporal dependencies across different time scales simultaneously. The dilated convolutions exponentially increase the receptive field with network depth, allowing TCNs to model very long-range dependencies while maintaining computational efficiency through parallelization. In energy storage applications, TCNs are particularly attractive for high-frequency data analysis and edge deployment due to their deterministic inference time and lower memory requirements compared to recurrent architectures.
Table 3 provides a comprehensive comparison of these deep learning architectures for energy storage applications.
The selection of architecture depends on specific application requirements. For applications requiring accurate long-term predictions with sufficient computational resources, LSTMs remain the preferred choice. However, for real-time applications with limited computational capacity, such as embedded battery management systems, GRUs or TCNs offer more practical solutions due to their computational efficiency and faster inference times.
LSTM networks have proven highly effective for battery state estimation and energy forecasting applications. The daily electricity demand for future load forecasting used the LSTM technique in order to analyze the appropriate size of the battery energy storage system (BESS) for residences. The ability of LSTM networks to capture long-term dependencies in sequential data makes them ideal for modeling battery behavior over extended periods.
The paper in [70] introduces a hybrid model integrating LSTM with the Coot bird search algorithm to optimize energy storage for wind power producers. The approach improves wind power prediction accuracy and enhances storage system scheduling, leading to more efficient renewable energy integration.
The authors in [7] propose an LSTM-based model to estimate the state of charge (SOC) and state of energy (SOE) in lithium-ion batteries. Their model enhances real-time prediction performance and reliability without relying on complex battery internal models.
The study in [36] presents a combined model using LSTM and particle swarm optimization for short-term load forecasting and energy management. This integration improves load prediction accuracy and enables smarter decisions for demand response and battery scheduling.

2.3. Reinforcement Learning for Energy Management

Reinforcement learning (RL) has emerged as a powerful approach for energy storage system control and optimization, particularly in dynamic environments where traditional control methods may be insufficient [62]. This paper focuses on employing RL algorithms to control energy flow in an AC microgrid. By incorporating AI and ML into the energy management system (EMS), the paper aims to optimize costs and facilitate the integration of renewable energy sources [32]. The RL agent is designed to trade energy with the main grid, taking advantage of the energy storage system and achieving cost savings.
The application of RL in energy storage systems enables autonomous decision-making and adaptive control strategies. The RL agent is tested using real spot prices data in Norway from a simulation model that combines Python 3.12 and MATLAB-Simulink 2025 software programs as an efficient co-simulation technique.
Emerging technologies, such as reinforcement learning and federated learning, show great promise for addressing these obstacles, enabling the dynamic optimization of charge cycles and the collaborative development of more generalized AI models [28]. The continuous learning capability of RL systems makes them particularly valuable for adapting to changing operational conditions and optimizing performance over time.
The study in [47] presents a comprehensive study on the application of a distributed reinforcement learning (DRL) framework for the optimization of grid-scale energy storage systems within the context of renewable energy integration, with the objective of evaluating the framework’s impact on grid stability, energy efficiency, and operational costs. Results from a 12-month analysis indicated that the DRL framework significantly improved grid stability, reducing grid disruptions by 40% and decreasing the average duration of these disruptions by 44%, contributing to enhanced reliability and fewer maintenance requirements. In terms of energy efficiency, the DRL framework showed a 5.2% improvement, increasing from 86.5% to 91.7% compared to the conventional baseline, accompanied by a 10.5% reduction in energy purchase costs and a total operational cost reduction of 12.2% over the 12-month period.

2.4. Optimization Algorithms

Advanced optimization algorithms play a critical role in energy storage system design and operation. The variables are microgrid optimal location and the capacity of the HMG components in the network, which are determined through a multi-objective improved Kepler optimization algorithm (MOIKOA) modeled on Kepler’s laws of planetary motion, piecewise linear chaotic map, and using the FDMT [42]. These algorithms address complex multi-objective optimization problems that arise in energy storage applications.
Generally, these multi-objective optimization problems can be formulated to find a vector of decision variables, x, optimizing a set of conflicting objective functions. A typical formulation is as follows:
min x F x = f 1 x ,   f 2 x ,   ,   f k x T
Subject to
g j x 0 ,   j = 1 , , m
h l x = 0 ,   l = 1 , , p
x m i n x x m a x  
Here, x is the vector of decision variables, which can include the rated power and capacity of the BESS, its location within the network, and operational setpoints. The F x is the vector of k objective functions to be minimized simultaneously. Based on the literature [43], these objectives often include the following:
f 1 x : Minimization of economic costs, such as the total investment and operational cost of the energy storage system.
f 2 x : Minimization of technical losses, such as the total power loss in the distribution network.
f 3 x : Minimization of voltage fluctuations to improve grid stability and power quality.
The problem is subject to a set of m inequality constraints g j x and p equality constraints h l x . These typically represent the physical and operational limits of the power system, such as power balance equations, thermal limits of lines, voltage magnitude limits at each bus, and the operational constraints of the BESS itself (e.g., state of charge limits, charge/discharge power limits). The final constraint defines the feasible range for the decision variables. Algorithms like MOIKOA and multi-objective particle swarm optimization (MOPSO) are employed to solve such problems by finding a set of Pareto optimal solutions, from which a final decision can be made using higher-level criteria.
Genetic algorithms and particle swarm optimization have been widely employed for energy storage system sizing and placement optimization. Advanced modeling and simulation techniques, such as stochastic optimization and genetic algorithms, are crucial for managing renewable energy variability [53]. These metaheuristic approaches are particularly effective for handling discrete variables and non-convex optimization landscapes commonly encountered in energy storage problems.
We delve into optimization strategies, highlighting dynamic programming’s role in decision making, reinforcement learning’s adaptability to environmental changes, and genetic algorithms’ explorations of optimal charging/discharging strategies [71]. The combination of multiple optimization techniques provides comprehensive solutions for various aspects of energy storage system management.
The paper in [43] addresses the integration of hybrid electric–hydrogen energy storage systems (HESSs) in distribution networks to mitigate adverse effects from renewable energy integration. The authors develop a multi-objective optimization framework using MOPSO, which considers the economic indicators of BESSs and hydrogen energy storage systems, power loss, and voltage fluctuation as fitness functions. The optimization approach employs the technique for order preference by similarity to ideal solution, based on information entropy weight, in order to select optimal solutions from the Pareto non-dominated solution set. Testing on extended IEEE-33 and IEEE-69 systems with 20% and 35% renewable energy penetration rates, the results demonstrate that power loss can be reduced by 7.9–22.8% and voltage fluctuation can be reduced by 40.0–71%, proving that the MOPSO-optimized energy storage system locations and capacities significantly improve distribution network stability and economy.
The effectiveness of optimization algorithms in energy storage systems is significantly enhanced by advances in multi-physics sensing technologies. While traditional approaches rely primarily on electrochemical sensors, emerging non-invasive joint monitoring technologies offer new possibilities for real-time optimization. Temperature dynamics, in particular, play a critical role in battery performance and safety, yet conventional temperature sensors often suffer from drift in extreme conditions. Recent developments in ultrasonic reflection wave technology have demonstrated a capability for simultaneous state of charge and temperature estimation in lithium-ion batteries, achieving RMS errors of ≤1.5% over a temperature range of −20 °C to 60 °C [72]. This multi-parameter fusion measurement approach addresses the sensor drift challenges in low-temperature conditions and provides optimization algorithms with more reliable real-time data. The integration of such advanced sensing technologies with optimization frameworks enables more accurate constraint handling and improved decision making, particularly for thermal management objectives in multi-objective optimization problems. These non-invasive techniques also reduce sensor-related degradation and maintenance requirements, contributing to the long-term reliability of optimization strategies in energy storage systems.

3. Battery Management Systems and State Estimation

3.1. State of Charge Estimation

State of Charge (SoC) estimation represents one of the most critical applications of AI in energy storage systems, as accurate SoC determination is essential for optimal battery performance and safety. Specific to BMS, these advanced concepts enable a more accurate prediction of battery performance such as its state of health (SoH), SoC, and state of power (SoP) [8]. Traditional methods for SoC estimation often suffer from accuracy limitations, particularly under varying operating conditions and battery aging [9].
ML approaches have demonstrated significant improvements in SoC estimation accuracy across different battery chemistries and operating conditions. Battery parameters such as SoC and SoC need precise measurement and calculation [10]. Monitoring them directly is a difficult task. In the present work, methodologies and approaches for estimating the batteries parameters using AI methods are investigated. The complexity of directly measuring these parameters necessitates the use of AI-based estimation techniques.
The study in [11] explores the use of a LSTM neural network for estimating the SoC in battery systems. The model is tested in the context of peak demand reduction, showing superior accuracy over traditional approaches such as Coulomb counting and Kalman filters. The LSTM achieved an MAE of 0.10 and RMSE of 0.12, outperforming other methods including feedforward neural networks. However, the original study did not report confidence intervals or statistical significance testing for these performance comparisons—a common limitation in the current literature that makes it difficult to assess the statistical reliability of reported improvements.
The comprehensive review in [12] explores the integration of physics-based models with AI methods for SoC estimation. It categorizes techniques into model-based (MB), data-driven (AI), and hybrid approaches, emphasizing the benefits of blending domain knowledge with learning-based systems for robust and adaptive SoC prediction.
Further, the paper in [13] discusses the implementation of AI-powered battery management systems (BMSs) for electric vehicles, focusing on SoC estimation and battery health monitoring. It details the use of ML algorithms like support vector machines, neural networks, and decision trees to enhance prediction accuracy, extend battery life, and improve system safety.
A critical observation across the reviewed SoC estimation studies is the general absence of rigorous statistical validation. While many papers report performance metrics such as MAE and RMSE, few provide confidence intervals, standard deviations, or statistical significance tests (e.g., paired t-tests with p < 0.05) to validate the claimed improvements. This lack of statistical rigor makes it challenging to determine whether reported performance differences are statistically significant or within the margin of experimental error. Future research should adopt standardized statistical validation protocols, including reporting 95% confidence intervals and conducting appropriate hypothesis testing, in order to enable more reliable comparisons between different SoC estimation approaches.

3.2. State of Health Monitoring

State of health (SoH) monitoring is critical for predicting battery lifecycle and scheduling maintenance activities. Accurate lithium-ion battery SoH evaluation is crucial for correctly operating and managing battery-based energy storage systems. SoH degradation is a complex phenomenon influenced by multiple factors including temperature, charge/discharge rates, depth of discharge, and calendar aging.
AI techniques have proven highly effective for SoH estimation by identifying subtle patterns in battery behavior that indicate degradation. These AI approaches enable more accurate predictions of battery degradation and failures, optimizing charge cycles and improving real-time diagnostics. ML models can detect early signs of capacity fade, internal resistance increase, and other degradation mechanisms before they become critical.
The development of explainable AI models for SoH monitoring has become increasingly important for industrial applications. In the present paper, an optimized explainable artificial intelligence (Ex-AI) model is proposed to predict the discharge capacity of the battery. Ex-AI is applied to identify relevant features and to further optimize Ex-AI operating parameters, and the jellyfish metaheuristic optimization technique is also considered. Ex-AI provides insights into model decision-making processes, enhancing trust and facilitating troubleshooting.
Recent advances in battery health prediction have highlighted the importance of considering voltage loss recovery mechanisms in improving prediction accuracy. The degradation mechanism of batteries involves complex voltage behaviors that traditional models often overlook. By incorporating dynamic recovery factor correction models, which account for voltage relaxation phenomena after charge/discharge cycles, researchers have achieved significant improvements in lifetime prediction accuracy. These models recognize that voltage loss is not merely a static parameter but exhibits dynamic recovery characteristics that vary with battery age, temperature, and usage patterns. The integration of voltage recovery dynamics into AI-based prediction models has demonstrated substantial reductions in lifetime prediction errors, as detailed in recent studies [73]. This approach provides a more complete understanding of battery degradation mechanisms and enables more accurate long-term performance predictions.
The paper in [15] reviews the application of ML for estimating the SoC and SoH in batteries used in electric vehicles, with a focus on 12 V hybrid energy storage systems.
The authors in [16] present developments in battery degradation modeling using ML techniques for SoH and remaining useful life (RUL) estimation across energy storage technologies.
The paper in [17] discusses deep learning techniques in smart battery management systems for predicting SoC, SoH, and RUL, including practical challenges and system integration aspects.

Technical Challenges in SoH Monitoring

Despite the promising advances in AI-based SoH monitoring, several critical technical challenges limit the practical deployment of these systems. Current Ex-AI models exhibit significant limitations in predicting capacity decay inflection points, with high failure rates in identifying these critical transitions. This high failure rate stems from the complex nonlinear dynamics that occur at inflection points, where traditional feature extraction methods fail to capture the subtle precursors of rapid degradation. Early decay feature extraction presents another fundamental challenge. The false alarm rate for detecting early-stage capacity decay remains high, creating reliability concerns for preventive maintenance applications. This issue arises from the difficulty in distinguishing between normal capacity variations due to operational conditions and genuine early-stage degradation signals. The low signal-to-noise ratio in early degradation phases compounds this problem, as measurement uncertainties can mask critical degradation indicators. Transfer learning between different battery applications reveals substantial limitations in model generalizability. When automotive battery models are migrated to grid energy storage applications, prediction errors increase despite similar battery chemistries. This degradation in performance highlights significant differences in operational profiles, including charge/discharge patterns, depth of discharge variations, and thermal management conditions between applications. The failure of transfer learning approaches indicates that current models capture application-specific patterns rather than fundamental degradation mechanisms. Perhaps most concerning is the vulnerability of SoH prediction models to adversarial conditions and sensor perturbations. Even minor temperature sensor perturbations can result in errors in SoH prediction, revealing the fragility of current models to input variations. This sensitivity poses serious concerns for real-world deployments where sensor drift, calibration errors, and environmental noise are common. These vulnerabilities suggest that current AI models may be overfitting to training data patterns rather than learning robust physical relationships, necessitating new approaches that incorporate uncertainty quantification and robust optimization techniques.

3.3. Predictive Maintenance

Predictive maintenance strategies leveraging AI techniques have revolutionized battery management by enabling proactive maintenance scheduling and failure prevention. AI techniques, including ML models like ensemble methods, support vector machines, and neural networks, have been instrumental in predictive maintenance, state of charge and state of health estimation, and materials discovery. These AI approaches enable more accurate predictions of battery degradation and failures, optimizing charge cycles and improving real-time diagnostics.
The integration of multiple data sources and advanced analytics enables comprehensive predictive maintenance strategies. The use of AI for predictive maintenance, load prediction, as well as real-time optimization of power flow is particularly beneficial for ensuring the efficient integration of renewable energy sources while maintaining system stability [18]. These approaches consider not only battery-specific parameters but also the system-level factors that influence battery performance and longevity.
ML models for predictive maintenance must balance accuracy with computational efficiency, particularly for real-time applications. These models forecast energy demands and system behavior, facilitating proactive maintenance and system efficiency improvements. The integration of ML not only enhances predictive maintenance and charging protocol optimization but also addresses challenges related to data scarcity, model generalizability, and interpretability. The development of lightweight models suitable for edge computing applications is an active area of research.
The review in [19] provides an overview of AI applications—particularly ML models—for predictive maintenance in renewable energy, with a focus on optimizing energy storage systems such as batteries.
The paper in [20] highlights the synergy between AI technologies and electric vehicle battery storage, showing how ML and deep learning enhance predictive maintenance and prolong system reliability.
The authors in [21] focus on smart home applications, demonstrating how AI models predict failures in Photovoltaic (PV) and battery systems to enhance operational performance and maintenance.
Critical Assessment: The reviewed predictive maintenance studies reveal a fundamental disconnect between academic metrics and industrial needs. While references [18,19,20,21] report high accuracy in pattern recognition, none provide actionable prediction horizons, nor the confidence bounds necessary for maintenance scheduling. More concerning, these studies assume pristine sensor data and stable operating conditions that rarely exist in industrial settings. The field’s focus on algorithmic sophistication obscures the core challenge: maintenance decisions require not just detecting degradation but predicting failure windows with sufficient lead time and confidence for intervention. Until research addresses these practical constraints, industrial adoption will remain limited.

3.4. Fault Detection and Diagnosis

AI-based fault detection and diagnosis systems provide critical safety functions in energy storage applications. Applications of cutting-edge ML techniques can improve system reliability with fault detection diagnosis (FDD), automation agent-based reinforcement learning, flexibility model predictive controls, and so on. The early detection of faults, such as thermal runaway, internal short circuits, and cell imbalances, is essential for preventing catastrophic failures.
Advanced pattern recognition techniques enable the identification of subtle anomalies that may indicate developing faults. Moreover, these technologies enable the development of self-healing grids that can detect, as well as respond to, faults autonomously, reducing downtime and enhancing the resilience of the grid. The ability to autonomously detect and respond to faults represents a significant advancement in energy storage system safety and reliability.
The study in [22] explores predictive fault diagnosis techniques in energy storage systems using AI and big data analytics. It emphasizes how AI technologies can forecast system failures and improve reliability and safety.
The paper in [23] delves into fault detection methods in energy storage by combining ML with large-scale sensor data, offering a comprehensive prognostics and health management (PHM) approach.
The authors in [24] present a comprehensive review of ML methods applied to the fault diagnosis of lithium-ion batteries. It evaluates supervised and unsupervised learning models and highlights gaps in real-time implementation.

4. Energy Storage Optimization and Control

4.1. Charging Optimization Strategies

Optimal charging strategies are crucial for maximizing battery lifespan, minimizing energy costs, and ensuring system safety. These AI approaches enable more accurate predictions of battery degradation and failures, optimizing charge cycles and improving real-time diagnostics. Furthermore, AI enhances the design of safer and more efficient battery components by accelerating materials research, thus improving lithium-ion battery capacity and safety profiles. AI-driven charging optimization considers multiple factors including battery chemistry, temperature, state of charge, and grid conditions. Dynamic programming and reinforcement learning have emerged as powerful approaches for charging optimization.
The review in [5] investigates the integration of AI, particularly reinforcement learning, in optimizing charging protocols for electric vehicle batteries. It highlights how traditional methods lack adaptability under dynamic driving and grid conditions. AI-based control systems, including predictive analytics and real-time learning, show promise for enhancing battery longevity, charging speed, and grid compatibility. The paper also maps out challenges such as computation overhead and integration complexity in embedded automotive systems.
The paper in [29] provides a broad overview of AI-driven techniques in optimizing BESSs. It supports deep reinforcement learning (DRL) for intelligent scheduling and decision making. The authors present comparative results of model-based versus model-free optimization algorithms, showing how DRL can adaptively learn charging policies from historical and simulated environments. Use cases include peak shaving, load balancing, and dynamic price responsiveness in energy markets.
The study in [30] applies a deep neural network to control a home microgrid that integrates photovoltaics and batteries. The system forecasts solar generation and consumption patterns, adjusting battery charge/discharge accordingly. The optimization goal is to minimize energy costs while maintaining energy independence and reducing grid dependency. It showcases significant gains in energy efficiency and economic returns, especially in regions with variable solar resources.
The work in [31] focuses on optimizing the sizing of battery storage systems based on predicted household energy consumption, particularly with electrical vehicle charging in mind. AI-based load forecasting techniques like LSTM are employed to predict daily and seasonal energy needs. These predictions inform a multi-objective optimization that balances cost, space, and grid impact. The study proves that intelligent sizing improves reliability and reduces overinvestment in oversized systems.
Critical Assessment: Reviewed charging optimization methods [5,29,30,31] consistently overlook a fundamental constraint: human behavior. These studies assume complete control over charging decisions, yet field studies show users routinely override ‘optimal’ schedules for convenience. Furthermore, the optimization objectives—minimizing cost or maximizing battery life—often conflict with user priorities like ensuring sufficient charge for unexpected trips. No reviewed study addresses what happens when thousands of vehicles follow the same ‘optimal’ strategy, potentially creating new demand peaks. These omissions suggest current research optimizes mathematical abstractions rather than real-world systems.

4.2. Energy Management Systems

Intelligent energy management systems (EMSs) represent the integration of multiple AI techniques to optimize overall system performance. By incorporating AI and ML into the energy management system, the goal is to optimize costs and facilitate the integration of renewable energy sources.
The paper in [33] focuses on the development of a methodology for the energy management, combining photovoltaics and storage systems, considering as the main case study a multi-story building characterized by a high density of households, used to generate data which allow feasibility foresights. The physical model of the algorithm is composed by two main elements: the photovoltaics modules and the battery energy storage system. In addition, to gain information about the real-time consumption a machine learning module is included in our approach to generate predictions about the near future demand.
The review in [34] explores the integration of AI techniques in optimizing renewable energy systems. It examines how AI methods—particularly ML, neural networks, and optimization algorithms—can enhance energy forecasting, resource assessment, grid integration, and operational control. The paper discusses key challenges in deploying AI in Renewable Energy Sources (RES), such as data heterogeneity and model transparency, and emphasizes the importance of emerging AI trends like explainable AI and edge computing. The authors envision a future where AI empowers autonomous energy management and smart grid resilience for more sustainable and efficient energy solutions.
The paper in [35] presents a comprehensive review of the integration of hybrid renewable energy sources into microgrids, emphasizing the challenges stemming from the intermittent and unstable nature of RESs. It classifies various integration schemes, communication challenges, and control strategies, with a particular focus on how AI methods can improve system performance. The study reviews optimization techniques—such as combining AI with artificial neural networks and particle swarm optimization—to enhance prediction accuracy and reduce operational costs. A case study demonstrates that using AI significantly improves control accuracy, achieving a normalized mean square error of just 1.10%.
The paper in [30] presents a deep-learning-based energy management system for home microgrids that incorporate photovoltaic systems and battery energy storage systems. It employs a bidirectional long short-term memory (Bi-LSTM) model combined with an optimization algorithm to schedule energy usage, aiming to reduce electricity costs under time of use pricing while considering daily peak demand penalties. The model integrates user behavior, BESS characteristics, and renewable generation variability to determine optimal dispatch strategies. Simulation results demonstrate the system’s effectiveness in lowering household electricity expenses and enhancing energy management efficiency.

4.3. Load Forecasting and Demand Prediction

Accurate load forecasting is fundamental to effective energy storage management, enabling optimal scheduling of charging and discharging operations.
In [31], an ML approach using a multilayer perceptron artificial neural network (MLP-ANN) has been used to forecast solar radiation, wind speed, temperature, and load data. Advanced forecasting techniques consider multiple variables including weather conditions, historical patterns, and seasonal variations. The daily electricity demand for future load forecasting used the LSTM technique in order to analyze the appropriate size of the battery energy storage system for residences. For this research, load forecasting will be presented to find the appropriate size of BESSs by considering the minimum daily load over the month, which is equal to 102.67 kWh, which can determine the size of the BESS to be 17.84 kWh. The integration of load forecasting with storage sizing optimization demonstrates the practical value of AI-driven approaches.
ML models for load forecasting must adapt to changing consumption patterns and external factors. Here, an ML-based Gaussian process regression (GPR) is used for predicting the volatile RES power output while the uncertain energy demand is estimated with a Monte Carlo simulation (MCS) [37]. The combination of different prediction techniques provides robust forecasting capabilities under uncertainty.
The paper in [38] reviews how forecasting is used to optimize energy storage systems in renewable energy contexts, especially given the stochastic nature of sources like solar and wind. It emphasizes that, although forecasting is often central to managing storage operations, its actual value varies by application and is not always systematically evaluated. The authors assess various forecasting approaches and compare them with baseline methods that do not use forecasting, revealing that the benefits are highly context-dependent. They propose that future research should focus on tailoring forecasting efforts to storage applications that demonstrably benefit from it.
The study in [39] proposes a novel two-phase framework for efficient short-term electricity load forecasting, which is critical for smart grid energy management. In the first phase, raw data undergo pre-processing, and, in the second, a deep residual convolutional neural network (CNN) extracts spatial features, which are then fed into a stacked LSTM network to capture temporal dependencies. This hybrid CNN–LSTM model is tested on the IHEPC and PJM datasets, showing significant improvements in prediction accuracy compared to traditional models like linear regression, LSTM, and GRU. The model notably reduced RMSE by 3.4% on the PJM dataset, showcasing its robustness and accuracy across various metrics.
The paper in [40] surveys the role of computational intelligence (CI) methods—primarily ML—in forecasting energy load within smart energy grids. The authors analyze over 50 research studies, classifying them into single and hybrid CI-based forecasting techniques. Their evaluation shows that hybrid approaches, especially those using metaheuristic algorithms, outperform standalone models in prediction accuracy. The study highlights the potential of intelligent load forecasting in reshaping energy consumption patterns, enhancing demand-side management, and addressing future challenges in real-time smart grid operations.
Wang et al. [41] investigate the optimal allocation of customer energy storage systems using power big data and an enhanced LSTM forecasting model. They propose a demand response framework that encourages regional customers to participate in the energy market by identifying profitable energy storage configurations. Their method evaluates user behavior responses to incentives, determines suitable users through predictive analytics, and optimizes battery capacity allocations accordingly. The study also introduces a model for iron phosphate batteries, showing that coupling storage systems with time of use pricing can significantly reduce costs and increase economic viability.

4.4. System Sizing and Placement Optimization

Optimal sizing and the placement of energy storage systems requires sophisticated optimization algorithms that consider multiple objectives and constraints. The paper in [31] presents the optimization sizing of a battery energy storage system for residential use from load forecasting, using AI. When comparing the size of the BESS from actual load values with the load from the forecast, it can significantly reduce the size and cost of the BESS. AI-driven optimization can significantly reduce system costs while maintaining performance requirements.
The study in [44] presents an optimization framework for determining the optimal location and size of a BESS to minimize power losses in electrical distribution networks. It introduces the whale optimization algorithm (WOA), a nature-inspired metaheuristic, as the primary tool for solving this problem. Two strategies are evaluated: one that separates placement and sizing into two stages, and another that optimizes both simultaneously. The WOA is benchmarked against firefly algorithm and particle swarm optimization, demonstrating superior performance in reducing losses and improving network efficiency.
Fossati et al. [45] present a method based on genetic algorithms to optimally size energy storage systems within microgrids. Their approach aims to minimize the overall operating cost by determining the ideal power and energy capacities of the energy storage system. A fuzzy expert system, tuned by the genetic algorithm, manages energy flow within the system, allowing the simultaneous optimization of both the energy management strategy and storage sizing. Importantly, the method incorporates an aging model for batteries to more accurately estimate lifecycle costs, and it also includes the unit commitment problem to ensure operational reliability. The effectiveness of the method is validated through two case studies.
This review paper [46] comprehensively explores the methodologies used for determining the optimal placement and sizing of distributed generators (DGs) and energy storage systems in microgrids. The authors first break down the components and operational modes of microgrids in environments with distributed energy resources. They then analyze the planning and functioning of both distributed generators and energy storage systems individually, followed by an integrated perspective. The study highlights prevalent objective functions, constraints, and mathematical formulations used in existing models. Finally, it identifies research gaps and suggests directions for developing new models that enhance the financial and operational performance of microgrids.

5. Grid-Scale Applications and Smart Microgrids

5.1. Grid Integration and Stability

The integration of large-scale energy storage systems with power grids presents unique challenges that AI techniques are uniquely positioned to address. The integration of AI and ML in electrical power systems and smart grids has the ability to greatly enhance their efficiency, reliability, along with sustainability. With the increasing complexity of modern power grids as well as the growing reliance on RESs, AI and ML provide advanced solutions for optimizing operations, enhancing grid stability, and addressing challenges such as intermittent energy generation, energy storage, and fault detection.
AI-driven grid integration strategies must consider the complex interactions between storage systems, renewable generation, and grid infrastructure. The increasing numbers of distributed energy resources incur new challenges for the energy supply due to the volatilities and uncertainties of renewable energies [48]. To utilize the benefits of Distributed Energy Resources (DER), a combination with storage systems and intelligent controls is necessary. The coordination of multiple distributed resources requires sophisticated control algorithms capable of real-time decision making. The grid energy storage system has been widely used in smart homes and grids [49]; its performance and working conditions directly affect the safety and reliability of the power grid. AI techniques play a crucial role in ensuring the safe and reliable operation of grid-connected storage systems.
Recent comprehensive reviews have examined the application of ML and AI techniques specifically for smart grid stability analysis [74], highlighting how these methods address the increasing complexity of modern power systems. The challenge of uncertainty in smart grid operations has led to sophisticated optimization approaches for energy storage management that explicitly account for stochastic variables in both generation and demand [75,76].
The authors in [50] explore the integration of AI/ML techniques in managing and optimizing energy storage systems. They outline how ML enhances energy prediction, battery management, and decision making in real time, especially within renewable energy frameworks. Emphasis is placed on the transition toward intelligent grid systems and how AI can resolve operational challenges like energy fluctuations and demand–supply gaps.
Yun et al. [51] propose an integrated energy system that combines compressed air energy storage (CAES) with a solid oxide fuel cell (SOFC), aiming to optimize the system’s performance using ML techniques. The SOFC generates electricity, part of which powers the CAES, while the flue gases from the SOFC are repurposed for domestic heating. ML models—based on regression—accurately predict system behavior, achieving R2 values above 98% and predicted R2 values mostly exceeding 99%. The optimized configuration yields a 63.4% energy efficiency and 32.5% exergy efficiency, highlighting the system’s viability and the critical role of AI in its optimization.

5.2. Microgrid Management

Smart microgrids represent a key application area for AI-driven energy storage management, requiring autonomous operation and optimization capabilities [52]. This generates various new opportunities, one of which is the capacity to monitor, control, and regulate the energy flow inside and outside of the microgrid. By making microgrids distributed energy resources and energy storage components economically viable with AI and ML incorporated into the cost-optimization process, the demand for and growth of these technologies will be accelerated significantly.
The development of autonomous microgrid operation requires sophisticated energy management algorithms. The key findings highlight the integration of emerging technologies, like AI, the Internet of Things (IoT), and advanced energy storage systems, which enhance microgrid efficiency, reliability, and resilience [53]. AI and ML optimize real-time microgrid operations, enhancing predictive analysis and fault tolerance. These capabilities enable microgrids to operate independently while maintaining optimal performance.
As the integration of PV inverters and battery energy storage systems gradually increases in the distribution network, the rapid fluctuation and random nature of these distributed generators put forward an urgent demand for real-time Volt–VAR optimization (VVO) [54]; to address this issue, this paper proposes a VVO strategy learning method via ML framework. In this context, real-time optimization capabilities are shown to be essential for managing the dynamic nature of microgrid operations.
The review in [55] examines how AI techniques are being applied in microgrids to address design and operational challenges caused by load uncertainties, renewable energy variability, and multi-source energy management. It highlights how AI can enhance efficiency, stability, and security in microgrid operations through applications in energy management, forecasting, control, and cybersecurity. The authors categorize AI tasks (e.g., regression, classification) and discuss methods like ML, neural networks, fuzzy logic, and support vector machines. Additionally, the paper outlines the benefits, limitations, and future trends of AI deployment in microgrids.
Wu and Wang [56] explore the integration of AI into the operation and control of microgrids, which increasingly rely on distributed energy resources and require flexibility across operational modes. They highlight the complexities of microgrid management due to limited system inertia, generation uncertainty, and diverse network topologies (AC, DC, hybrid). The authors emphasize the capabilities of deep learning and DRL in addressing decision-making challenges, noting their potential in enhancing microgrid performance. The paper also analyzes unique control architectures and configurations in microgrids and presents AI-driven approaches for system optimization and control under dynamic conditions.
The survey in [57] explores how AI and ML techniques enhance energy management systems in microgrids, which are vital components in the push for grid decentralization and renewable integration. The authors categorize energy management system architectures into centralized, decentralized, and distributed models, examining how AI can improve each. Techniques such as artificial neural networks, federated learning, long short-term memory, recurrent neural networks, and reinforcement learning are reviewed for tasks like economic dispatch, optimal power flow, and energy scheduling. Despite the benefits of AI, like improved reliability and scalability, the paper highlights challenges such as data privacy, explainability, and security. It concludes with proposed future research directions for real-world deployment.

5.3. Renewable Energy Integration

The integration of renewable energy sources with storage systems presents significant challenges due to the intermittent and unpredictable nature of renewable generation [58]. AI techniques enable better prediction and management of renewable energy variability through advanced forecasting and control strategies. These capabilities are essential for maximizing the utilization of renewable energy while maintaining system stability. In this context, the development of robust control strategies that handle uncertainty is crucial for successful renewable integration.
The paper in [59] discusses how ML and AI, combined with data science, are revolutionizing renewable energy systems. It highlights the use of predictive analytics to enhance solar and wind energy efficiency by forecasting outputs using weather and historical data. AI algorithms also improve grid stability by balancing energy supply and demand, reducing waste, and integrating diverse sources. Furthermore, data science facilitates predictive maintenance, fault detection, and energy storage optimization. The study offers a literature review, methodological insights, and policy considerations for a more sustainable, AI-driven energy infrastructure.
The comprehensive review in [60] discusses the integration of renewable energy sources into smart grids, emphasizing their importance for achieving sustainability and reducing greenhouse gas emissions. The paper explores cutting-edge smart grid technologies including advanced metering infrastructure, distributed control systems, and Supervisory Control and Data Acquisition (SCADA) systems, highlighting how these enhance grid reliability and efficiency. It also examines the role of ML in optimizing energy management within smart grids. Finally, the study underscores the importance of continued technological advancement and supportive policy frameworks for a resilient and sustainable energy future.
Synthesis and Gaps: While studies [58,59,60] present sophisticated integration strategies, they universally assume institutional arrangements that rarely exist. Renewable generation and storage typically have different owners with conflicting objectives, yet the reviewed methods assume coordinated optimization. Moreover, the reported improvements lack context—percentage gains mean little without baseline performance levels. The real integration challenge is not technical but institutional—a matter of how to align incentives across multiple stakeholders. Until AI methods can handle multi-party optimization with conflicting objectives and incomplete information sharing, practical impact will remain limited.

6. Thermal Management and Safety Applications

6.1. Thermal Modeling and Control

Thermal management represents one of the most critical aspects of energy storage system safety and performance, and is an area where AI techniques have demonstrated significant value. Their focus has been on higher energy efficiency, improved thermal performance, and optimized multi-material battery enclosure designs. The integration of simulation-based design optimization of the battery pack and battery management system is evolving and has expanded to include novelties such as AI/ML to improve efficiencies in design, manufacturing, and operations for their application in electric vehicles and energy storage systems.
AI-driven thermal management systems enable the real-time monitoring and control of battery temperatures, preventing thermal runaway and optimizing performance. The complex thermal–electrical interactions in battery systems require sophisticated modeling approaches that AI techniques can provide.
ML models for thermal prediction must consider multiple factors including ambient conditions, charge/discharge rates, and battery aging. ML/DL models are developed for various classes of problems involving large nonlinear parameters from energy usage patterns in power distributions, including renewable power sources, building energy consumption, management to improve energy consumption, efficiency in optimized design, and predictive analysis of thermal heat management systems in electronics and battery storage systems [25]. Thus, the integration of thermal management with overall system optimization is crucial for achieving optimal performance.
The comprehensive review in [26] explores the integration of AI into thermal energy storage (TES) systems to enhance their performance and efficiency. It surveys a variety of AI methods, including particle swarm optimization, artificial neural networks, support vector machines, and adaptive neuro-fuzzy inference systems, highlighting their applications in modeling, optimizing, and managing TES. The paper evaluates the effectiveness of these AI tools in achieving various energy-related objectives and emphasizes their accuracy and adaptability. Additionally, it provides practical recommendations and outlines future research directions, aiming to drive further innovation in sustainable energy systems.
The paper by [27] presents a bibliometric analysis of the application of AI, including ML and deep learning, in the design of thermal energy storage tanks. It analyzes publications from the Scopus database, and uses VOSviewer 1.6.20 for keyword mapping, in order to identify research trends, knowledge gaps, and future opportunities in AI-enabled TES tank design. The study finds a sharp rise in related publications since 2020, with a dominant focus on optimization techniques such as genetic algorithms and materials like phase change materials. It emphasizes AI’s potential in improving energy efficiency and sustainability in TES tank design while noting the need for further research in underexplored areas.

6.2. Fire Prevention and Suppression

The development of AI-driven fire prevention systems for energy storage applications has become increasingly important as system sizes and energy densities increase. These systems must integrate multiple detection modalities and provide rapid response capabilities to prevent catastrophic failures. Advanced sensor fusion techniques enable comprehensive monitoring of fire precursors including temperature gradients, gas emissions, and electrical anomalies. The integration of these diverse data sources requires sophisticated AI algorithms capable of real-time pattern recognition and decision making.
The paper by Lee et al. [77] presents a transformative approach to fire detection in gas-to-liquids facilities using AI. Traditional detection systems often struggle in such high-risk environments due to delayed responses and environmental interference. The authors propose integrating AI, IoT, and multispectral imaging to improve real-time fire detection. AI algorithms analyze complex thermal and multispectral data, while IoT networks enable constant monitoring and rapid responses. The study emphasizes the importance of sensor fusion, adaptive model training, and edge computing for high-precision detection in operationally complex settings.
The study by Ekunke et al. [78] reviews emerging technologies in fire detection and suppression tailored to the unique hazards of oil refineries. It highlights the limitations of traditional methods like water sprinklers and foam, emphasizing newer alternatives such as AI-powered IoT sensors, water mist systems, and eco-friendly fluorine-free foams. The paper also explores innovations including firefighting drones, robots, and VR/AR-based emergency training that improve both response times and training effectiveness. Notably, sensor-fusion systems stand out in detection performance, while robotic and drone-based suppression systems show promise for practical deployment. The authors conclude by calling for further research, industrial partnerships, and regulatory support to scale these technologies sustainably.

7. Challenges and Limitations

7.1. Data Quality and Availability

One of the primary challenges in implementing AI techniques for energy storage applications is ensuring adequate data quality and availability. However, despite these advancements, challenges like data quality, model interpretability, and the integration of AI models into existing industrial frameworks, persist [4]. The performance of AI models is fundamentally dependent on the quality and quantity of training data, which can be challenging to obtain in real-world applications.
This paper proposes a new method to model batteries with low-quality data. First, it designs a data cleaning method for GESS battery operating data, including missing data filling and outlier data repair. Then, the repaired data are used to model the battery. Data preprocessing and cleaning are crucial steps in developing robust AI models for energy storage applications.
The challenge of data scarcity is particularly pronounced in long-term studies where battery degradation patterns must be captured over extended periods. The integration of ML not only enhances predictive maintenance and charging protocol optimization but also addresses challenges related to data scarcity, model generalizability, and interpretability. Developing methods to handle limited data scenarios is essential for practical AI implementations.
Recent advances in generative AI and self-supervised learning offer promising approaches in addressing data scarcity challenges in energy storage applications. Generative adversarial networks (GANs), which have shown remarkable success in computer vision tasks such as image dehazing, can be adapted to synthesize realistic battery operational data under various conditions. By training a generator network to produce synthetic battery cycling data that is indistinguishable from real measurements, GANs can augment limited datasets while preserving the statistical properties of actual battery behavior. Similarly, self-supervised learning methods, which learn useful representations from unlabeled data without requiring extensive manual annotation, can extract meaningful features from raw sensor data streams. These techniques have demonstrated success in learning robust representations from partial or noisy data in image processing applications and show potential for battery state estimation under data-constrained conditions. The adaptation of such methods from computer vision to energy storage represents a promising research direction, particularly for scenarios where obtaining labeled training data is expensive or time-consuming. However, careful validation is required to ensure that synthetically generated data accurately captures the complex electrochemical dynamics and degradation patterns of real battery systems.

Self-Supervised Learning and Generative Approaches for Data Scarcity

The data scarcity challenge in energy storage applications can be effectively addressed through self-supervised learning (SSL) and generative approaches. SSL has emerged as a particularly powerful paradigm for battery management systems, as it can leverage the vast amounts of unlabeled operational data that are continuously collected but rarely annotated.
Self-supervised learning methods create pretext tasks from the data, such as predicting future voltage profiles from historical measurements or reconstructing masked sensor readings. For battery applications, SSL can learn rich representations from raw time-series data including voltage, current, and temperature measurements without requiring labeled degradation states. Recent implementations have shown that SSL pretraining on unlabeled battery cycling data can reduce the labeled data requirements for downstream tasks, like SoH estimation, by up to 70%, while maintaining comparable accuracy to fully supervised approaches.
The key advantage of SSL for energy storage lies in its ability to exploit the inherent structure and physics of battery systems. Contrastive learning methods can identify similar charge/discharge patterns across different batteries, while predictive SSL approaches can learn the temporal dynamics of degradation processes. These learned representations transfer effectively across different battery chemistries and operating conditions, addressing generalization challenges.
GANs complement SSL by synthesizing realistic battery operational data for rare or safety-critical scenarios that are expensive or dangerous to collect experimentally. For instance, GANs can generate synthetic thermal runaway data or extreme fast-charging profiles that would damage real batteries. However, validation remains crucial—generated data must accurately reflect electrochemical constraints and degradation physics, not just statistical distributions.
The combination of SSL for representation learning and GANs for data augmentation presents a promising path forward, potentially reducing the data collection burden by an order of magnitude while improving model robustness across diverse operating conditions.

7.2. Model Interpretability and Trust

The black box nature of many AI models presents significant challenges for industrial adoption, particularly in safety-critical applications. However, despite these advancements, challenges like data quality, model interpretability, and the integration of AI models into existing industrial frameworks, persist. Users and regulators require an understanding of how AI models make decisions, especially when these decisions impact system safety and reliability.
In the present paper, an Ex-AI model is proposed to predict the discharge capacity of the battery. Ex-AI is applied to identify the relevant features and further optimize Ex-AI operating parameters, and the jellyfish metaheuristic optimization technique is also considered. The development of Ex-AI techniques is crucial for building trust and enabling adoption in industrial applications.

7.3. Computational Requirements and Real-Time Constraints

Many AI techniques require significant computational resources, which can be challenging to provide in energy storage applications with real-time constraints. Commonly used control methods are based on model predictive control (MPC) with forecast and optimization, which require computational capabilities at the DER level, on which the distribution system operator (DSO) has no access [79]. This paper proposes a control algorithm based on data, combining a model-based and an AI-based approach to utilize benefits from both methods while compensating for their drawbacks.
The development of lightweight AI models suitable for edge computing applications is an active area of research. These models must balance accuracy with computational efficiency to enable real-time decision making in resource-constrained environments.

7.4. Critical Comparison of AI Approaches

A critical examination of the reviewed AI techniques reveals important trade-offs that are often understated in individual studies. While deep learning methods demonstrate superior accuracy for certain tasks, such as the LSTM achieving an MAE of 0.10 for SoC estimation, as reported in reference [11], these gains must be weighed against increased computational complexity and implementation challenges in resource-constrained battery management systems. The effectiveness of different AI approaches varies significantly with application context. The distributed reinforcement learning (RL) framework in reference [47] achieved a 40% reduction in grid disruptions for grid-scale applications, demonstrating RL’s strength in complex optimization scenarios. However, such approaches may be less suitable for embedded systems with limited computational resources or safety-critical applications requiring guaranteed performance bounds. A critical gap in the current literature is insufficient attention to practical deployment constraints. While some studies report performance metrics such as accuracy and error rates, few address computational requirements, memory footprint, or real-time processing capabilities. This omission makes it difficult for practitioners to assess whether reported improvements are achievable in operational systems. The field also exhibits a concerning lack of robustness analysis. Most studies evaluate performance under ideal conditions without examining sensitivity to sensor noise, model degradation over time, or performance under off-nominal conditions. Given that real-world energy storage systems operate in variable environments with imperfect sensors, this represents a significant limitation in assessing the practical viability of the proposed approaches.

7.5. Strategic Comparative Analysis and Technology Outlook

Based on our comprehensive review, we present a strategic analysis comparing AI techniques across key performance dimensions and their suitability for different energy storage applications.
Comparative Performance Assessment: The reviewed literature reveals clear trade-offs between accuracy and computational complexity. LSTM networks, as demonstrated in reference [11], with an MAE of 0.10 for SoC estimation, provide high accuracy for time-series prediction but require substantial computational resources. Traditional machine learning methods like support vector machines and Random Forest offer more computationally efficient alternatives, though the literature lacks consistent quantitative comparisons of computational requirements across methods. For optimization tasks, the DRL framework in reference [47] achieved 40% reduction in grid disruptions, demonstrating RL’s effectiveness in complex dynamic environments, while genetic algorithms prove valuable for static optimization problems, as shown in system sizing applications [45].
Scalability Considerations: Federated learning, discussed in references [28,57], emerges as a promising approach for distributed systems by enabling collaborative learning while preserving data privacy. However, quantitative scalability limits remain unexplored in the reviewed literature. Edge computing implementations face the fundamental challenge of balancing model complexity with embedded system constraints, though specific performance trade-offs require further investigation.
Industrial Implementation Barriers: The transition from laboratory demonstrations to industrial deployment faces several challenges identified throughout the review. The lack of standardized benchmarks prevents meaningful comparison between approaches. Model robustness under real-world conditions remains largely unaddressed, with most studies conducted under idealized laboratory settings. Ex-AI, while critical for safety applications, as noted in reference [14], adds complexity that may conflict with real-time processing requirements.
Application-Specific Insights: Different energy storage applications favor different AI approaches. Grid-scale applications with substantial computational resources can leverage complex deep learning and reinforcement learning methods, as demonstrated in references [47,48,49,50,51]. Embedded battery management systems require computationally efficient approaches, though the literature provides limited guidance on specific method selection. For emerging battery technologies with limited operational data, physics-informed approaches, as discussed in reference [12], offer a path forward by incorporating domain knowledge.
To provide concrete guidance based on the reviewed evidence, Table 4 presents method rankings by application domain, synthesizing performance metrics from the 155 reviewed publications.
These rankings reflect reported performance in the reviewed literature. For instance, LSTM’s top ranking in SoC/SoH estimation is supported by its MAE of 0.10, as reported in reference [11], while DRL’s leading position in grid-scale optimization is justified by the 40% reduction in disruptions demonstrated in reference [47]. However, practitioners should note that these rankings are based on different studies using varied datasets and conditions, as direct comparative studies remain rare in the literature.
Strategic Research Priorities: Moving forward, the field must shift focus from pursuing marginal accuracy improvements to developing robust, deployable solutions. Key priorities include developing standardized benchmarks reflecting real-world conditions, quantifying computational requirements alongside accuracy metrics, and creating adaptive frameworks that maintain performance as systems age. The integration of physics-based knowledge with data-driven approaches represents a promising direction, as does the development of uncertainty quantification methods, essential for safety-critical applications.

7.6. Root Causes and Pathways for Overcoming Persistent Challenges

The persistent technical challenges identified throughout this review—data scarcity, model opacity, limited generalization, computational constraints, and lack of standards—remain unresolved, not due to lack of awareness but due to fundamental structural barriers that require coordinated action to overcome.
Why Data Quality Problems Persist: The data challenge stems from three interlocking factors. First, battery manufacturers guard operational data as proprietary competitive advantages, creating data silos that prevent collaborative model development. Second, the long timescales of battery degradation (3–10 years) mean that comprehensive lifecycle datasets require sustained investment with delayed returns, discouraging both academic and industrial efforts. Third, the diversity of battery chemistries, operating conditions, and applications means that even available data often lack transferability. Practical implications include models that fail catastrophically when deployed on new battery types or operating regimes, as evidenced by the 22% performance degradation when transferring between applications. Overcoming this requires establishing pre-competitive data sharing consortiums, similar to semiconductor industry models, where competitors collaborate on fundamental challenges while competing on implementation.
The Interpretability Paradox: Model opacity persists because the most accurate models (deep neural networks) are inherently the least interpretable, creating a fundamental tension. Current explainable AI methods add 30–40% computational overhead while providing only partial insights. This opacity has severe practical implications: regulatory bodies cannot certify black box models for safety-critical applications, and operators cannot trust systems they do not understand, leading to manual override of AI recommendations. Progress requires developing new model architectures that build interpretability into their fundamental structure rather than adding it post hoc, potentially through physics-informed designs that constrain models to physically meaningful representations.
Generalization Failures: Models fail to generalize because they learn spurious correlations rather than causal relationships. Laboratory-trained models capture the specific patterns of controlled environments—stable temperatures, clean power supplies, precise sensors—that disappear in field deployment. The practical result is systems that work perfectly in demonstrations but fail unpredictably in real applications, undermining confidence and adoption. Addressing this requires a paradigm shift from accuracy-focused to robustness-focused development, including adversarial training, uncertainty quantification, and systematic testing under distribution shift.
Computational Constraints: The mismatch between AI computational requirements and embedded system capabilities persists due to divergent evolution paths. AI research prioritizes accuracy on powerful hardware, while battery management systems use decades-old microcontrollers for reliability and cost. This gap means that sophisticated AI models must be drastically simplified for deployment, losing much of their advantage. Solutions require co-design of algorithms and hardware, developing AI methods specifically for edge deployment rather than adapting desktop methods, and potentially new computing paradigms like neuromorphic chips designed for AI inference.
Standards and Validation: The absence of benchmarks reflects deeper disagreements about what constitutes success in energy storage AI. Academic metrics (MAE, RMSE) poorly correlate with industrial needs (safety, reliability, total cost of ownership). Without agreed metrics, comparing approaches becomes impossible, slowing progress. Developing meaningful standards requires unprecedented collaboration between academia, industry, and regulators to define multi-dimensional performance metrics that capture real-world requirements.
Path Forward: Overcoming these challenges requires recognizing that they are not independent technical problems but interconnected systemic issues. Data sharing enables better models, which justify the investment in interpretability, which enables regulatory approval, which drives industrial adoption, which generates more data. Breaking this cycle requires coordinated intervention: public funding for pre-competitive research infrastructure, regulatory frameworks that incentivize transparency while protecting IP, and new academic–industrial partnerships focused on deployment rather than publication. The energy transition’s urgency demands that we move beyond identifying problems to implementing solutions, even if imperfect, while continuously improving through real-world learning.

8. Future Research Directions

8.1. Emerging AI Technologies

Several emerging AI technologies show promise for advancing energy storage applications beyond current capabilities. Hybrid physics–AI models represent a significant trend, combining domain knowledge with data-driven insights. While reference [12] discusses the integration of physics-based models with AI methods, the field is moving toward more sophisticated physics-informed neural networks that embed conservation laws and electrochemical principles directly into learning architectures. This approach addresses the fundamental limitations of pure data-driven methods, particularly their inability to extrapolate reliably beyond training conditions.
Ex-AI is gaining importance, as demonstrated by reference [14]’s work on Ex-AI for battery discharge prediction. As AI systems move towards safety-critical applications, the need for interpretable models becomes paramount. Current research explores various interpretability methods, including attention mechanisms and feature importance techniques, though balancing interpretability with performance remains an open challenge.
Digital twin technology emerges as another promising direction, creating virtual replicas of energy storage systems updated in real-time through sensor data. While not extensively covered in the current literature, digital twins offer potential for predictive maintenance and optimization without risking physical systems. The integration with IoT infrastructure and edge computing enables new paradigms for system monitoring and control.
The maturity of these technologies varies considerably. While some physics-informed approaches are approaching industrial deployment for specific applications, others remain in the early research phases. Regulatory frameworks are beginning to address AI in critical infrastructure, with implications for how these technologies can be deployed. Future adoption will likely depend on demonstrating clear value propositions, developing appropriate standards, and addressing concerns about reliability and safety in operational environments.
Federated learning, mentioned in references [28,57], continues to show promise for collaborative model development while preserving data privacy, which is particularly relevant for distributed energy storage systems. Quantum machine learning, while still largely theoretical, may eventually enable the solution of optimization problems that are currently intractable with classical computing methods. These emerging technologies collectively point toward a future where AI systems become more robust, interpretable, and capable of handling the complex challenges inherent in energy storage applications.

8.2. Integration with Digital Technologies

The integration of AI with other digital technologies is creating new opportunities for energy storage applications. The integration of new technologies enables real-time monitoring with an inclination towards Industry 4.0 [80]. With motivation from the above research outcomes, this study aims to review the significance of advancements such as wireless sensor networks (WSNs), the IoT, AI, cloud computing, edge blockchain, and twin machine learning. Finally, this article suggests significant recommendations regarding the computation of AI model-based devices, the customization of IoT-based hybrid models, ML-based twins modeling, and blockchain data sharing.
The development of comprehensive digital ecosystems that integrate AI, IoT, cloud computing, and blockchain technologies is enabling more sophisticated energy storage management capabilities. These integrated systems provide enhanced monitoring, control, and optimization capabilities while ensuring data security and system reliability.
Future research directions include advanced technologies, enhanced energy management system capabilities through AI/ML, and more highly developed smart infrastructures [81]. Policy recommendations stress that regulatory support stakeholder collaboration drives innovation and scale deployment, ensuring sustainable future. The convergence of multiple technologies is creating new possibilities for intelligent energy storage systems.

8.3. Sustainability and Circular Economy

The integration of AI techniques with sustainability considerations and circular economy principles is becoming increasingly important. These methodologies collectively contribute to sustainable energy practices and resource conservation, marking significant advancements in the field [71]. AI can play a crucial role in optimizing resource utilization, extending system lifespans, and enabling effective recycling and reuse strategies.
However, critical areas need further exploration, such as real-time forecasting, long-term storage predictions, hybrid neural networks for demand-based generation forecasting, and the evaluation of various storage scales and battery technologies [58]. In conclusion, this study emphasizes the need for continued research and the development of new algorithms to address existing limitations in the field [82,83,84]. Addressing these research gaps is essential for realizing the full potential of AI in sustainable energy storage applications.

9. Energy Storage and the Option Value Concept

9.1. Introduction to Real Options in Energy Storage Planning

The integration of energy storage systems into modern power networks presents unique investment challenges characterized by deep uncertainty about future demand patterns, renewable generation deployment, and technological evolution. Traditional net present value (NPV) approaches to investment appraisal fail to capture the strategic value of managerial flexibility [85]—the ability to defer, scale, relocate, or abandon investments as uncertainty resolves over time [86]. This limitation has led to the adoption of real options theory in energy storage planning, which explicitly values the flexibility embedded in smart grid technologies.
The concept of option value in energy storage stems from the recognition that many smart grid assets, including battery storage systems, can be deployed rapidly and adaptively compared to conventional network reinforcements [87]. Unlike traditional infrastructure investments that require long lead times and represent irreversible commitments, energy storage and associated smart technologies create managerial flexibility that allows planners to ‘wait, learn, and adapt’ as uncertain conditions unfold [88]. This flexibility translates into quantifiable economic value that conventional deterministic planning approaches systematically overlook.

9.2. Energy Storage as a Real Option

Energy storage systems exhibit several characteristics that make them particularly suitable for real options analysis. First, their modular nature enables incremental deployment [89], allowing network operators to start with pilot installations that reveal information about system performance and demand patterns [90]. Second, the relatively short commissioning times for battery storage—typically months rather than years for conventional reinforcements—provide temporal flexibility to respond to emerging network constraints [91]. Third, the operational flexibility of storage systems, including their ability to provide multiple services simultaneously, creates additional option value through operational adaptation [92].
The option value of energy storage manifests in multiple dimensions. At the strategic level, storage deployment can defer or displace costly and irreversible transmission line upgrades, which is particularly valuable under uncertain load growth trajectories. Operationally, storage systems can be dynamically repositioned across different network locations or repurposed for various services as system needs evolve [93]. This multi-dimensional flexibility is particularly valuable in contexts with high renewable energy penetration, where the location and magnitude of network constraints may shift significantly over time.
The study in [94] proposes an optimal investment strategy expanded NPV using a real options approach that accounts for technical types. The real options approach is defined according to sum of total values, including all conjunction with the extended binomial tree, based on economic theory.
The study in [95] investigates the economic viability of compressed air energy storage (CAES) systems, using real options analysis under volatile energy market conditions. It models future prices for electricity, natural gas, and reserve demand using Monte Carlo simulations. Three configurations of diabatic and adiabatic CAES are evaluated, revealing that the most economical configuration is a diabatic CAES used for load leveling. The study highlights the challenges of high capital costs and market exposure for CAES investment.
The paper in [96] investigates how battery storage can be valued in deregulated electricity markets using real options theory. It models a storage unit’s operation as a perpetual American swing put option, where the operator makes sequential decisions about buying power from an energy imbalance market (EIM) and selling incremental balancing reserves. The EIM price is treated as a stochastic diffusion process, and the paper derives optimal strategies for timing these decisions. The model is applied using data from Germany’s Amprion EIM to showcase both operational and economic insights.
The paper in [97] explores how energy storage systems can help address peak-valley load differences and enhance grid reliability, especially with increased integration of wind and solar energy. The authors evaluate the economic feasibility of deploying energy storage systems—specifically lithium-ion, vanadium redox flow, and sodium-sulfur batteries—at wind farms, where investment decisions are challenged by irreversibility and uncertainty. Leveraging real options theory and a binary tree option pricing model, they assess the viability of such investments under uncertain environments. Their findings support the application of real options analysis as a valuable decision-making tool in planning ESS investments amidst policy and market volatility.
The study in [98] analyzes how price volatility—particularly under Korea’s Renewable Portfolio Standard scheme—affects investment decisions in energy storage integrated with solar PV. Using real options analysis, the authors determine that increased volatility in renewable energy certificate prices significantly raises the required investment threshold for justifying such projects. Specifically, the trigger price rises by 10.5% under price volatility, and doubles in volatility would require a 26.6% higher investment price. The paper proposes a hybrid auction scheme combining feed-in tariff and Renewable Portfolio Standard advantages to reduce subsidy risks and improve market efficiency.

9.3. Quantification Methods and Modeling Frameworks

The quantification of option value for energy storage requires sophisticated modeling approaches that can capture both the stochastic nature of future uncertainties and the sequential decision-making process of network planning. Multi-stage stochastic optimization has emerged as the predominant framework for option valuation in this context. These models represent uncertainty through scenario trees that branch at decision points, allowing planners to make adaptive investment decisions based on revealed information.
A critical distinction in option valuation models is between exogenous and endogenous uncertainty. Exogenous uncertainties, such as demand growth or renewable generation patterns, evolve independently of planning decisions [99]. In contrast, endogenous or decision-dependent uncertainties, such as consumer participation in demand response programs or the success of storage technology pilots, are resolved through the act of investment itself. This distinction is particularly relevant for energy storage, where pilot deployments can reveal crucial information about technical performance, cost trajectories, and operational constraints [90,100,101,102,103,104].
Advanced decomposition techniques, particularly Benders decomposition and its variants, have proven essential for making these large-scale stochastic optimization problems computationally tractable [105,106]. These methods enable the evaluation of option value for realistic network sizes while maintaining the granularity needed to capture operational constraints and multi-period dynamics.
The paper in [107] explores the vital role of energy storage technologies in facilitating the transition to a low-carbon energy system. It highlights how these technologies decouple supply and demand and can be integrated across supply, transmission, distribution, and end-use stages of the grid. Services range from long-term storage to short-duration reserves, with pumped hydropower representing the dominant form, accounting for over 99% of installed global capacity (~141 GW). The authors emphasize the need for supportive policy frameworks to ensure compensation for the diverse services energy storage provides.
The study in [108] presents a novel decision-making framework using multi-attribute value theory, a type of multi-criteria decision analysis, to evaluate various energy storage options. Six energy storage projects in Cornwall, UK—including power to gas, liquid air, and battery systems integrated with solar PV or wave energy—are assessed based on technical attributes and stakeholder perspectives. The framework allows for a transparent and flexible evaluation, factoring in both technical feasibility and local priorities. The battery storage project integrated with PV and Cornwall Airport Newquay demand ranked highest, highlighting multi-attribute value theory’s utility in supporting nuanced, location-specific energy planning.
The review in [109] explores the integration of energy storage systems into power networks to support renewable energy deployment and maintain grid reliability. It evaluates mechanical, electrochemical, chemical, and thermal storage technologies from techno-economic and environmental perspectives, based on 91 reviewed publications. The study highlights that the levelized cost of energy (LCoE) declines with longer storage durations due to economies of scale. Additionally, round-trip efficiency, cycle life, and depth of discharge significantly impact the cost-effectiveness and environmental footprint of these technologies.
The study in [110] provides a thorough review of various energy storage technologies crucial for integrating renewable energy sources into sustainable energy systems. It evaluates mechanical, electrochemical, chemical, thermal, and electromagnetic storage methods based on key performance metrics like system balancing, energy arbitrage, environmental impact, and power quality. The review highlights that, while some technologies are mature and commercially viable, others remain in the early development stages. Ultimately, the paper offers strategic recommendations for advancing smart energy storage solutions to meet diverse end-user needs and to enhance sustainability.

9.4. Empirical Evidence and Case Studies

Extensive empirical evidence demonstrates the materiality of option value in energy storage planning across diverse contexts and scales. At the distribution level, studies have shown that treating battery storage as a real option can reduce expected system costs by 45–81% compared to conventional reinforcement-only strategies. For instance, in an 11 kV network case study, the option value of a portfolio, including storage and other smart technologies, reached £4.31 million over 6 years, representing an 81% saving compared to traditional planning approaches.
The option value of storage becomes even more pronounced at the transmission level and in national-scale studies. Research on India’s transmission expansion planning (2020–2060) revealed that treating large-scale battery storage as a real option yielded an option value of approximately £12.9 billion, derived primarily from £27.5 billion in operating cost savings and £2 billion in deferred or avoided line investments [111]. Similarly, studies of Great Britain’s power system demonstrated that the option value of storage-enabled electric vehicle smart charging could reach £10.8 billion over 40 years when bidirectional (V2G) capabilities are considered [112].
The magnitude of option value varies significantly with technology characteristics and deployment contexts. Sensitivity analyses consistently show that option value increases with the following: (i) greater operational flexibility of the storage technology, (ii) lower capital costs, (iii) higher uncertainty in future demand or generation patterns, and (iv) longer planning horizons [113,114]. Conversely, option value diminishes rapidly as storage costs increase or operational constraints limit flexibility [115].

9.5. Portfolio Effects and Technology Interactions

Energy storage rarely operates in isolation but, rather, as part of integrated smart grid portfolios that may include a demand-side response (DSR), dynamic line rating (DLR), soft open points (SOPs), and smart electric vehicle charging infrastructure [116]. The option value of such portfolios often exceeds the sum of individual technology option values due to complementarity effects. For example, combining storage with SOPs and smart electric vehicle charging in distribution networks can lift portfolio option value to £324–337k, nearly equal to the entire conventional reinforcement budget.
These portfolio effects arise from several mechanisms. First, different technologies hedge different types of uncertainties—storage addresses temporal mismatches while SOPs handle spatial load imbalances. Second, operational synergies enable more efficient utilization of each asset when deployed together [117]. Third, the information revealed by deploying one technology can inform decisions about others, creating compound option structures [118].
Carmona and Ludkovski in [119] propose a stochastic control framework to value energy storage facilities, such as natural gas domes and hydroelectric pumped storage. They model the storage operation as an optimal switching problem, where operators face a constrained compound American option on the temporal spread of commodity prices. To address the challenge of path-dependency from inventory levels, they design a robust Monte-Carlo-based numerical method that accommodates general Markovian price processes and operational constraints. Their method outperforms traditional quasi-variational approaches and is demonstrated through various practical examples.
The study in [120] assesses how energy storage can support decarbonization by improving operational flexibility and enabling greater integration of variable renewable energy sources. Using a capacity expansion model for a Texas-like grid, the authors find that storage helps reduce investments in gas and nuclear capacity while improving overall asset utilization. However, the marginal value of short-duration (2 h) storage exceeds current costs only under stringent emissions constraints, whereas longer-duration (10 h) storage aligns with the economics of pumped hydro. Overall, storage is most critical when relying heavily on wind and solar, but less so with a diversified low-carbon energy mix.
The report in [121] evaluates how energy storage contributes to the power grid by simulating various storage configurations in a utility system in the western U.S. It found that storage offers relatively low value for load-leveling but higher value for services like spinning and regulation reserves. The authors measured system-wide operational cost savings with and without storage, and estimated potential market revenues for merchant storage. However, in restructured markets, actual revenues may fall short of system benefits due to price suppression and partial benefit capture, highlighting deployment challenges in deregulated systems.
The study in [122] presents a novel investment appraisal methodology for electrical energy storage systems using real options analysis. It highlights how real options analysis captures flexibility in decision making under uncertainty, enabling investors to delay investments or stage them based on market conditions. Applied to the UK energy market, the model shows that this approach enhances the economic performance of energy storage systems in price arbitrage and reserve services, although modest policy incentives are still needed to ensure viability. The method allows the reassessment of project value throughout its development, thus mitigating financial risks effectively.
The authors in [123] develop a model to determine the optimal timing of investments in battery storage systems under uncertainty in future revenues and costs. They show that relying solely on the spot electricity market does not justify the investment due to insufficient revenue. However, when battery systems also participate in ancillary (balancing) service markets, the NPV becomes positive, and the real options value increases further, emphasizing the strategic value of flexible timing. Their results confirm that incorporating market duality and uncertainty significantly enhances the economic viability of energy storage investments.

9.6. Implications for Planning and Policy

The recognition of option value fundamentally alters optimal investment strategies for energy storage. Rather than pursuing deterministic ‘build-everything-now’ approaches, option-aware planning favors staged deployment strategies that preserve flexibility [124]. This often means investing in storage and other smart technologies earlier than NPV analysis would suggest, as their value includes not just immediate operational benefits but also the strategic flexibility they create for future decisions [125].
However, current regulatory frameworks and planning standards often fail to recognize or remunerate option value, leading to systematic underinvestment in storage and overinvestment in conventional assets. This misalignment stems partly from institutional factors—utilities operating under rate of return regulation may prefer capital-intensive conventional investments—and partly from methodological limitations in standard planning tools [126].

9.7. Future Directions and Research Needs

Several areas require further research to fully realize the option value of energy storage. First, methods for incorporating multiple interacting uncertainties, including policy and regulatory changes, need development [127]. Second, the integration of machine learning techniques [128] with stochastic optimization could enable more sophisticated modeling of endogenous uncertainties and learning effects [129,130,131]. Third, practical implementation frameworks that can be adopted by system operators while maintaining computational tractability remain a priority [132,133].
The evolution towards increasingly decentralized and renewable-dominated power systems will likely amplify the importance of option value in storage planning [134]. As uncertainty deepens and the pace of change accelerates, the strategic flexibility provided by energy storage becomes ever more valuable. Quantifying and capturing this value through appropriate planning methods and regulatory frameworks represents a critical challenge for achieving cost-effective and resilient energy transitions.

10. Discussion

10.1. Synthesis of Key Findings

Our analysis of 155 publications reveals a field characterized by rapid technical progress but significant implementation gaps. The reviewed literature demonstrates that machine learning methods consistently outperform traditional approaches for prediction tasks, with LSTM networks emerging as the dominant architecture for time-series applications, such as the SoC estimation achieving an MAE of 0.10, as reported in reference [11]. However, these accuracy gains often come at computational costs that prohibit embedded deployment, explaining why simpler methods persist in industrial applications. A critical finding is the disconnect between laboratory and field performance. The literature shows a strong bias toward reporting positive results, with limited discussion of failures or negative outcomes. This publication bias obscures the true state of technology readiness. The absence of failure analysis in the literature represents a significant gap, as understanding why AI methods fail is crucial for improvement.

10.2. Comparative Analyses of Key Studies

Comparing findings across different research groups reveals concerning inconsistencies. Studies on similar applications report widely varying performance metrics that cannot be explained by methodological differences alone. This suggests that unstated factors—data quality, preprocessing methods, or test conditions—dramatically influence outcomes. Regional research emphases also differ, with varying foci on algorithm development versus robustness and regulatory compliance, reflecting different energy system challenges and regulatory environments. The theoretical frameworks employed diverge significantly. Pure machine learning approaches dominate the reviewed literature, while hybrid physics–AI methods remain underexplored despite showing promise for improved generalization, as suggested by reference [12].

10.3. Evaluation of Strengths and Weaknesses

Current AI approaches excel in pattern recognition and multi-objective optimization—areas where traditional methods struggle. The ability to capture nonlinear aging dynamics and optimize across competing objectives represents genuine advancement. RL’s success in grid-scale applications, such as the 40% reduction in operational disruptions reported in reference [47], demonstrates AI’s transformative potential when properly applied. However, fundamental weaknesses persist. Model sensitivity to input perturbations, as discussed in our analysis of SoH monitoring challenges, reveals insufficient robustness for safety-critical deployment. The requirement for extensive training data conflicts with the realities of new battery technologies, where operational data are scarce. Most critically, the black box nature of deep learning models prevents the mechanistic understanding necessary for certification and trust.

10.4. Gaps and Future Research Priorities

Our analysis identifies several critical gaps requiring immediate attention. First, the absence of standardized benchmarks prevents meaningful comparison between approaches. Each study uses different datasets, metrics, and test conditions, making meta-analysis impossible. Second, the lack of uncertainty quantification in the vast majority of reviewed studies is concerning, given the safety implications. AI systems must communicate confidence levels, not just point predictions. Third, the field lacks research on failure modes and edge cases. Understanding when and why AI methods fail is essential for safe deployment. Fourth, economic analysis remains disconnected from technical development. While Section 9 documents option values reaching billions of pounds, few technical studies consider implementation costs or economic viability.

10.5. Implications for Research and Practice

For researchers, our findings suggest redirecting efforts from marginal accuracy improvements to robustness and interpretability. The field has achieved sufficient predictive accuracy; it lacks the reliability and transparency for industrial adoption. Future work should prioritize adversarial testing, uncertainty quantification, and explainable architectures. For practitioners, the review reveals that AI readiness varies dramatically by application. Grid-scale optimization with abundant computational resources and tolerable failure consequences represents the nearest-term opportunity. Conversely, embedded battery management requires fundamental advances in edge AI before widespread deployment. A staged adoption strategy, beginning with advisory systems that augment human decision making rather than replacing it, offers a pragmatic path forward. For policymakers, the findings highlight the need for proactive regulatory frameworks that balance innovation with safety. Current regulations, designed for deterministic systems, inadequately address probabilistic AI behaviors. New standards must accommodate uncertainty while ensuring minimum safety guarantees. Public investment in shared research infrastructure—standardized datasets, testing facilities, and validation protocols—could accelerate progress while maintaining competitive innovation. The energy transition’s urgency demands that we move beyond academic exercises towards deployable solutions. This requires acknowledging current limitations while working systematically to address them, fostering collaboration between domains traditionally separated, and maintaining focus on real-world impact rather than publication metrics.

11. Conclusions

This comprehensive review has examined the current state and future prospects of artificial intelligence (AI) applications in energy storage systems. Analysis of the recent literature reveals significant progress in multiple areas, from battery management systems to grid-scale applications, demonstrating the transformative potential of AI techniques in this critical domain.
AI techniques, including machine learning models like ensemble methods, support vector machines, and neural networks, have been instrumental in predictive maintenance, state of charge and state of health estimation, and materials discovery. These AI approaches enable more accurate predictions of battery degradation and failures, optimizing charge cycles, and improving real-time diagnostics. These technical achievements demonstrate the maturity of AI applications in critical energy storage functions. AI models indeed have practical utility for precise state estimation and prediction tasks.

11.1. Current Limitations and Challenges

Despite significant progress, several challenges remain that limit the widespread adoption of AI in energy storage applications:
Data and Model Challenges: Challenges like data quality, model interpretability, and the integration of AI models into existing industrial frameworks, persist. These fundamental challenges require continued research and development efforts.
Infrastructure Requirements: Although it has the potential to transform, the use of AI in energy systems is confronted with various challenges such as high infrastructure expenditure, data needs, system integration problems, and regulatory issues [1]. The cost and complexity of implementing AI solutions remain significant barriers to adoption.
Research Gaps: Critical areas need further exploration, such as real-time forecasting, long-term storage predictions, hybrid neural networks for demand-based generation forecasting, and the evaluation of various storage scales and battery technologies. These research gaps represent important opportunities for future development.

11.2. Future Outlook

The future of AI applications in energy storage appears promising, with several emerging trends and technologies poised to drive further advancement:
Emerging Technologies: Emerging technologies, such as reinforcement learning and federated learning, show great promise for addressing these obstacles, enabling the dynamic optimization of charge cycles and the collaborative development of more generalized AI models. These advanced AI techniques offer new capabilities for addressing current limitations.
Digital Integration: Future directions include advancements in technologies, enhanced energy management system capabilities through AI/ML, and the development of smart infrastructures. The convergence of AI with other digital technologies is creating new possibilities for intelligent energy storage systems.
Sustainability Focus: These methodologies collectively contribute to sustainable energy practices and resource conservation, marking significant advancements in the field. The integration of sustainability considerations with AI optimization is becoming increasingly important.

11.3. Recommendations for Future Research

Based on this comprehensive review, several recommendations emerge for future research in AI applications for energy storage:
  • Data Quality and Standardization: Develop standardized datasets and benchmarks for evaluating AI models in energy storage applications. Address data quality issues through improved sensor technologies and data preprocessing techniques.
  • Explainable AI Development: Advance research in explainable AI (Ex-AI) techniques specifically tailored for energy storage applications to improve model interpretability and industrial acceptance.
  • Real-Time Implementation: Focus on developing lightweight AI models suitable for real-time applications with limited computational resources while maintaining high accuracy.
  • Integration and Interoperability: Research methods for seamlessly integrating AI techniques with existing energy storage systems and developing interoperable standards.
  • Long-Term Studies: Conduct extended studies to validate AI model performance over complete battery lifecycles and under diverse operating conditions.
  • Multi-Scale Modeling: Develop AI techniques capable of handling multi-scale and multi-physics modeling for comprehensive system optimization.

11.4. Final Remarks

As collaborative research and open-data-sharing initiatives expand, AI’s transformative potential in driving more sustainable, efficient, and safer energy storage solutions will continue to grow, shaping the future of lithium-ion batteries and their applications in a greener, more energy-efficient world. The continued advancement of AI applications in energy storage systems represents a critical component of the global energy transition.
The convergence of AI technologies with energy storage systems is creating unprecedented opportunities for improving system performance, reducing costs, and enhancing sustainability. While challenges remain, the progress demonstrated in recent research provides a strong foundation for continued advancement. The successful implementation of AI in energy storage will require continued collaboration between researchers, industry, and policymakers to address technical, economic, and regulatory challenges.
In conclusion, this study emphasizes the need for continued research and the development of new algorithms to address existing limitations in the field [58]. The future success of AI applications in energy storage will depend on sustained research efforts, industry investment, and the development of supportive regulatory frameworks that enable innovation while ensuring safety and reliability [135,136,137].
The potential for AI to transform energy storage systems is significant, with implications extending far beyond individual applications to encompass entire energy ecosystems. As these technologies continue to mature and integrate, they will play an increasingly important role in enabling the transition to a sustainable, efficient, and resilient energy future [138,139,140,141,142,143,144].

Funding

This research received no external funding.

Data Availability Statement

This article is a comprehensive literature review and does not involve the generation or analysis of new experimental datasets. All reviewed publications are cited in the reference list, allowing readers to access the original sources. The review encompasses peer-reviewed publications obtained through institutional subscriptions to the IEEE Xplore, ScienceDirect, SpringerLink, and MDPI databases. While most reviewed articles are available through standard academic database subscriptions, some may require institutional access or individual purchase. No proprietary datasets were used in this review. Readers seeking specific papers should consult their institutional libraries or contact the corresponding authors of the original publications. A complete bibliography is provided in the reference section, and citation information for all reviewed works is available within the manuscript text and tables.

Acknowledgments

The authors acknowledge the valuable contributions of researchers worldwide, whose work has advanced the field of AI applications in energy storage systems. The comprehensive nature of this review was made possible by the extensive body of research conducted by the global scientific community.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Jayendra, J.; Mehare, A.; Wandhekar, A.; Pawar, S.; Chavan, P.; Nigade, V. Artificial Intelligence in Energy Management: A Comprehensive Literature Review on Methods, Applications, and Challenges. Int. J. Sci. Res. Eng. Trends 2025, 11, 2683–2688. [Google Scholar] [CrossRef]
  2. Debs, A.S. Modern Power Systems Control and Operation; Springer: New York, NY, USA, 1988. [Google Scholar] [CrossRef]
  3. Huan, Z.; Sun, C.; Ge, M. Progress in Profitable Fe-Based Flow Batteries for Broad-Scale Energy Storage. WIREs Energy Environ. 2024, 13, e541. [Google Scholar] [CrossRef]
  4. Omojola, A.F.; Ilabija, C.O.; Onyeka, C.I.; Ishiwu, J.I.; Olaleye, T.G.; Ozoemena, I.J.; Nzereogu, P.U. Artificial Intelligence-Driven Strategies for Advancing Lithium-Ion Battery Performance and Safety. Int. J. Adv. Eng. Manag. 2024, 6, 452–484. [Google Scholar] [CrossRef]
  5. Miraftabzadeh, S.M.; Longo, M.; Di Martino, A.; Saldarini, A.; Faranda, R.S. Exploring the Synergy of Artificial Intelligence in Energy Storage Systems for Electric Vehicles. Electronics 2024, 13, 1973. [Google Scholar] [CrossRef]
  6. Jabakumar, A.K. Optimizing Energy Storage Systems with AI-Based Control Strategies. Res. J. Comput. Syst. Eng. 2024, 5, 95–105. Available online: https://www.technicaljournals.org/RJCSE/index.php/journal/article/view/98 (accessed on 12 August 2025).
  7. Ma, L.; Hu, C.; Cheng, F. State of charge and state of energy estimation for lithium-ion batteries based on a long short-term memory neural network. J. Energy Storage 2021, 37, 102440. [Google Scholar] [CrossRef]
  8. Ghalkhani, M.; Habibi, S. Review of the Li-Ion Battery, Thermal Management, and AI-Based Battery Management System for EV Application. Energies 2022, 16, 185. [Google Scholar] [CrossRef]
  9. Xiong, R.; Shen, W. Battery State of Charge and State of Energy Estimation. In Advanced Battery Management Technologies for Electric Vehicles; Wiley: Hoboken, NJ, USA, 2019; pp. 67–93. [Google Scholar] [CrossRef]
  10. Vashist, D.; Raj, R.; Deepanshu, S. Artificial Intelligence in Electric Vehicle Battery Management System: A Technique for Better Energy Storage. SAE Technical Paper Series. In Proceedings of the SAENIS TTTMS Thermal Management Systems Conference, Delhi, India, 19–21 September 2024. [Google Scholar] [CrossRef]
  11. Lam, D.H.C.; Lim, Y.S.; Wong, J.; Hau, L.C. Machine Learning for Estimation of State-of-Charge of Energy Storage System. In Proceedings of the 2021 International Conference on Smart City and Green Energy (ICSCGE), Hangzhou, China, 20–22 November 2021; pp. 1–5. [Google Scholar] [CrossRef]
  12. Sorouri, H.; Oshnoei, A.; Che, Y.; Teodorescu, R. A comprehensive review of hybrid battery state of charge estimation: Exploring physics-aware AI-based approaches. J. Energy Storage 2024, 100, 113604. [Google Scholar] [CrossRef]
  13. Nallamala, S.K.; Punukollu, P.; Kodali, S.; Punukollu, M.; Burugu, S.; Yerneni, R.P. AI-Powered Battery Management Systems for Electric Vehicles: Leveraging Machine Learning for State-of-Charge Estimation, Battery Health Monitoring, and Beyond. J. Artif. Intell. Mach. Learn. Syst. 2022, 6, 66–103. [Google Scholar]
  14. Vakharia, V.; Shah, M.; Nair, P.; Borade, H.; Sahlot, P.; Wankhede, V. Estimation of Lithium-ion Battery Discharge Capacity by Integrating Optimized Explainable-AI and Stacked LSTM Model. Batteries 2023, 9, 125. [Google Scholar] [CrossRef]
  15. Vidal, C.; Malysz, P.; Kollmeyer, P.; Emadi, A. Machine Learning Applied to Electrified Vehicle Battery State of Charge and State of Health Estimation: State-of-the-Art. IEEE Access 2020, 8, 52796–52814. [Google Scholar] [CrossRef]
  16. Rauf, H.; Khalid, M.; Arshad, N. Machine learning in state of health and remaining useful life estimation: Theoretical and technological development in battery degradation modelling. Renew. Sustain. Energy Rev. 2022, 156, 111903. [Google Scholar] [CrossRef]
  17. Lipu, M.S.H.; Ansari, S.; Miah, M.S.; Meraj, S.T.; Hasan, K. Deep learning enabled state of charge, state of health and remaining useful life estimation for smart battery management system: Methods, implementations, issues. J. Energy Storage 2022, 55, 105752. [Google Scholar] [CrossRef]
  18. Ali, S.R.M.; Sharma, G.; Sheker, P.; Avns, S.; Rajarajeswari, B. Leveraging Artificial Intelligence and Machine Learning Techniques Improve Performance of Electrical Systems and Smart Grids. Int. J. Integr. Stud. (IJIS) 2025, 1, 1–14. [Google Scholar] [CrossRef]
  19. Hamdan, A.; Ibekwe, K.I.; Ilojianya, V.I. AI in renewable energy: A review of predictive maintenance and energy optimization. Int. J. Sci. Res. Arch. 2024, 11, 718–729. [Google Scholar] [CrossRef]
  20. Cavus, M.; Dissanayake, D.; Bell, M. Next Generation of Electric Vehicles: AI-Driven Approaches for Predictive Maintenance and Battery Management. Energies 2025, 18, 1041. [Google Scholar] [CrossRef]
  21. Rojek, I.; Mikołajewski, D.; Mroziński, A.; Macko, M. Machine Learning- and Artificial Intelligence-Derived Prediction for Home Smart Energy Systems with PV Installation and Battery Energy Storage. Energies 2023, 16, 6613. [Google Scholar] [CrossRef]
  22. Li, G.; Liu, S. Research on fault prediction and diagnosis methods of energy storage system based on big data and artificial intelligence. Energy Storage Sci. Technol. 2024, 13, 3653. [Google Scholar]
  23. Han, S. Research on fault prediction and diagnosis methods for energy storage systems based on big data and artificial intelligence. Energy Storage Sci. Technol. 2025, 14, 2114. [Google Scholar]
  24. Samanta, A.; Chowdhuri, S.; Williamson, S.S. Machine learning-based data-driven fault detection/diagnosis of lithium-ion battery: A critical review. Electronics 2021, 10, 1309. [Google Scholar] [CrossRef]
  25. Bhattacharya, A.; Majumdar, P. Artificial Intelligence-Machine Learning Algorithms for the Simulation of Combustion Thermal Analysis. Heat Transf. Eng. 2023, 45, 176–193. [Google Scholar] [CrossRef]
  26. Chekifi, T.; Boukraa, M.; Benmoussa, A. Artificial Intelligence for Thermal Energy Storage Enhancement: A Comprehensive Review. ASME J. Energy Resour. Technol. 2024, 146, 060802. [Google Scholar] [CrossRef]
  27. Mehraj, N.; Mateu, C.; Cabeza, L.F. Use of artificial intelligence methods in designing thermal energy storage tanks: A bibliometric analysis. J. Energy Storage 2024, 97, 112794. [Google Scholar] [CrossRef]
  28. Wiering, M.A.; Van Otterlo, M. Reinforcement Learning: State-of-the-Art; Springer: Berlin/Heidelberg, Germany, 2012. [Google Scholar]
  29. Song, H.; Liu, C.; Amani, A.M.; Gu, M.; Jalili, M. Smart optimization in battery energy storage systems: An overview. J. Energy Storage 2024, 17, 100378. [Google Scholar] [CrossRef]
  30. Alam, M.M.; Rahman, M.H.; Ahmed, M.F.; Chowdhury, M.Z. Deep learning based optimal energy management for photovoltaic and battery energy storage integrated home micro-grid system. Sci. Rep. 2022, 12, 15133. [Google Scholar] [CrossRef]
  31. Promasa, N.; Songkoh, E.; Phonkaphon, S.; Sirichunchuen, K.; Ketkaew, C.; Unahalekhaka, P. Optimization of Sizing of Battery Energy Storage System for Residential Households by Load Forecasting with Artificial Intelligence (AI): Case of EV Charging Installation. Energies 2025, 18, 1245. [Google Scholar] [CrossRef]
  32. Bertsekas, D.P. A Course in Reinforcement Learning; Athena Scientific: Nashua, NH, USA, 2024. [Google Scholar]
  33. Giglio, E.; Luzzani, G.; Terranova, V.; Trivigno, G.; Niccolai, A.; Grimaccia, F. An Efficient Artificial Intelligence Energy Management System for Urban Building Integrating Photovoltaic and Storage. IEEE Access 2023, 11, 18673–18688. [Google Scholar] [CrossRef]
  34. Ukoba, K.; Olatunji, K.O.; Adeoye, E.; Jen, T.C. Optimizing renewable energy systems through artificial intelligence: Review and future prospects. Energy Environ. 2024, 35, 3833–3879. [Google Scholar] [CrossRef]
  35. Talaat, M.; Elkholy, M.H.; Alblawi, A.; Said, T. Artificial intelligence applications for microgrids integration and management of hybrid renewable energy sources. Artif. Intell. Rev. 2023, 56, 10557–10611. [Google Scholar] [CrossRef]
  36. Cao, Z.; Han, X.; Lyons, W.; O’Rourke, F. Energy management optimisation using a combined Long Short-Term Memory recurrent neural network—Particle Swarm Optimisation model. J. Clean. Prod. 2021, 326, 129246. [Google Scholar] [CrossRef]
  37. Patty, S.; Malakar, T. Assessment of battery energy storage uses in microgrid operation under varied demand uncertainties through a machine learning based stochastic optimization approach. Phys. Scr. 2025, 100, 036013. [Google Scholar] [CrossRef]
  38. Sharma, V.; Cortes, A.; Cali, U. Use of Forecasting in Energy Storage Applications: A Review. IEEE Access 2021, 9, 114690–114704. [Google Scholar] [CrossRef]
  39. Khan, Z.A.; Ullah, A.; Haq, I.U.; Hamdy, M.; Mauro, G.M. Efficient short-term electricity load forecasting for effective energy management. Sustain. Energy Technol. Assess. 2022, 53, 102337. [Google Scholar] [CrossRef]
  40. Fallah, S.N.; Deo, R.C.; Shojafar, M.; Conti, M. Computational intelligence approaches for energy load forecasting in smart energy management grids: State of the art, future challenges, and research opportunities. Energies 2018, 11, 596. [Google Scholar] [CrossRef]
  41. Wang, L.; Qu, Y.; Wang, S.; Liu, Y.; Yang, R.; Ren, X.; Ma, Y. Optimal allocation of customer energy storage based on power big data and improved LSTM load forecasting. Energy Rep. 2024, 11, 3902–3913. [Google Scholar] [CrossRef]
  42. Duan, F.; Eslami, M.; Khajehzadeh, M.; Basem, A.; Jasim, D.J.; Palani, S. Optimization of a photovoltaic/wind/battery energy-based microgrid in distribution network using machine learning and fuzzy multi-objective improved Kepler optimizer algorithms. Sci. Rep. 2024, 14, 13354. [Google Scholar] [CrossRef]
  43. Xuan, J.; Chen, Z.; Zheng, J.; Zhang, Z.; Shi, Y. Optimal planning of hybrid electric-hydrogen energy storage systems via multi-objective particle swarm optimization. Front. Energy Res. 2023, 10, 1034985. [Google Scholar] [CrossRef]
  44. Wong, L.A.; Ramachandaramurthy, V.K.; Walker, S.L. Optimal placement and sizing of battery energy storage system for losses reduction using whale optimization algorithm. J. Energy Storage 2019, 26, 100892. [Google Scholar] [CrossRef]
  45. Fossati, J.P.; Galarza, A.; Martín-Villate, A.; Fontan, L. A method for optimal sizing energy storage systems for microgrids. Renew. Energy 2015, 77, 539–549. [Google Scholar] [CrossRef]
  46. Grisales-Noreña, L.F.; Restrepo-Cuestas, B.J. Optimal location and sizing of distributed generators and energy storage systems in microgrids: A review. Energies 2022, 16, 106. [Google Scholar] [CrossRef]
  47. Nuvvula, R.S.S.; Kumar, P.P.; Ahammed, S.R.; Hushein, R.H.; Reddy, R.S.S.; Babu, B.H.; Ali, A. Distributed Reinforcement Learning Framework for Autonomous Optimization of Grid-Scale Energy Storage Systems in Renewable Energy Integration. In Proceedings of the 2024 12th International Conference on Smart Grid (icSmartGrid), Setubal, Portugal, 27–29 May 2024. [Google Scholar] [CrossRef]
  48. Handel, K.; Anitha, A.E.; Rigo-Mariani, R.; Schulte, K.; Haubrock, J. Optimal Decision Tree Design for Near Real-Time Control of Battery Energy Storage Systems. Symposium on Symbolic and Numeric Algorithms for Scientific Computing. In Proceedings of the 2024 26th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing (SYNASC), Timisoara, Romania, 16–19 September 2024. [Google Scholar] [CrossRef]
  49. Miao, Y.; He, X.; Gu, C. Modeling lithium-ion Battery in Grid Energy Storage Systems: A Big Data and Artificial Intelligence Approach, Industrial Cyber-Physical Systems. In Proceedings of the 2023 IEEE 6th International Conference on Industrial Cyber-Physical Systems (ICPS), Wuhan, China, 8–11 May 2023. [Google Scholar] [CrossRef]
  50. Kumar, N.; Prabhansu. Artificial intelligence and machine learning applications in energy storage system: Technology overview and perspectives. In Emerging Trends in Energy Storage Systems and Industrial Applications; Kumar, N., Prabhansu, Eds.; Academic Press: Cambridge, MA, USA, 2023; pp. 1–26. [Google Scholar] [CrossRef]
  51. Yun, P.; Wu, H.; Alsenani, T.R.; Bouzgarrou, S.M. On the utilization of artificial intelligence for studying and multi-objective optimizing a compressed air energy storage integrated energy system. Energy Rep. 2024, 84, 110839. [Google Scholar] [CrossRef]
  52. Ahmed, I.; Pedersen, A.; Mihet-Popa, L. Smart Microgrid Optimization using Deep Reinforcement Learning by utilizing the Energy Storage Systems. In Proceedings of the 2024 4th International Conference on Smart Grid and Renewable Energy (SGRE), Doha, Qatar, 8–10 January 2024. [Google Scholar] [CrossRef]
  53. Quizhpe, K.; Arévalo, P.; Ochoa-Correa, D.; Villa-Ávila, E. Optimizing Microgrid Planning for Renewable Integration in Power Systems: A Comprehensive Review. Electronics 2024, 13, 3620. [Google Scholar] [CrossRef]
  54. Yan, Z.; Li, X.; Zhou, H.; Xie, Z.; Wang, Y.; Hong, L. Learning Strategies for Volt-VAR Optimization with PV Inverters and Battery Energy Storage Systems in Distribution Network. In Proceedings of the 2023 6th International Conference on Energy, Electrical and Power Engineering (CEEPE), Guangzhou, China, 12–14 May 2023. [Google Scholar] [CrossRef]
  55. Mohammadi, E.; Alizadeh, M.; Asgarimoghaddam, M.; Wang, X.; Simões, M.G. A review on application of artificial intelligence techniques in microgrids. IEEE J. Emerg. Sel. Top. Ind. Electron. 2022, 3, 878–890. [Google Scholar] [CrossRef]
  56. Wu, T.; Wang, J. Artificial intelligence for operation and control: The case of microgrids. Electr. J. 2021, 34, 106890. [Google Scholar] [CrossRef]
  57. Joshi, A.; Capezza, S.; Alhaji, A.; Chow, M.Y. Survey on AI and Machine Learning Techniques for Microgrid Energy Management Systems. IEEE/CAA J. Autom. Sin. 2023, 10, 1513–1529. [Google Scholar] [CrossRef]
  58. Filho, E.M.L.; Silveira, A.B.; Ferreira, A.M.; Marques, J.A.L.; Batista, J.G.; Guimares, G.D.F.; Alexandria, A.R.D.; Rodrigues, J.J.P.C. Optimization of Energy Storage Systems with Renewable Energy Generation and Consumption Data. In Proceedings of the 2024 IEEE 7th Student Conference on Electric Machines and Systems (SCEMS), Macao, China, 6–8 November 2024. [Google Scholar] [CrossRef]
  59. Amini, M.; Baradaran Rohani, M. The role of machine learning and artificial intelligence in enhancing renewable energy through data science. World J. Technol. Sci. Res. 2024, 12, 2341–2365. [Google Scholar] [CrossRef]
  60. Kiasari, M.; Ghaffari, M.; Aly, H.H. A Comprehensive Review of the Current Status of Smart Grid Technologies for Renewable Energies Integration and Future Trends: The Role of Machine Learning and Energy Storage Systems. Energies 2024, 17, 4128. [Google Scholar] [CrossRef]
  61. Ajibade, S.-S.; Zaidi, A.; Al Luhayb, A.S.M.; Adediran, A.O.; Voumik, L.C.; Rabbi, F. New Insights into the Emerging Trends Research of Machine and Deep Learning Applications in Energy Storage: A Bibliometric Analysis and Publication Trends. Int. J. Energy Econ. Policy 2023, 13, 303–314. [Google Scholar] [CrossRef]
  62. Sutton, R.S.; Barto, A.G. Reinforcement Learning: An Introduction; MIT Press: London, UK, 1998. [Google Scholar]
  63. Zhou, Y. Advances of machine learning in multi-energy district communities mechanisms, applications and perspectives. Energy AI 2022, 10, 100187. [Google Scholar] [CrossRef]
  64. Kumar, S.; Kassamkuttiyil, H.S.; Bhadauria, K.S.; Shah, J.V.; Khanna, V.; Jena, P. Machine Learning Technique for Hydrogen Fuel Cell Based Microgrid Energy Management. In Proceedings of the 2024 23rd National Power Systems Conference (NPSC), Indore, India, 14–16 December 2024; pp. 1–6. [Google Scholar] [CrossRef]
  65. Hastie, T.; Tibshirani, R.; Friedman, J. Overview of supervised learning. In The Elements of Statistical Learning; Springer: New York, NY, USA, 2009; pp. 9–42. [Google Scholar]
  66. Radhoush, S.; Whitaker, B.M.; Nehrir, H. An Overview of Supervised Machine Learning Approaches for Applications in Active Distribution Networks. Energies 2023, 16, 5972. [Google Scholar] [CrossRef]
  67. Celebi, M.E.; Aydin, K. Unsupervised learning algorithms. In Unsupervised Learning Algorithms; Springer: Cham, Switzerland, 2016; pp. 3–24. Available online: https://link.springer.com/content/pdf/10.1007/978-3-319-24211-8.pdf (accessed on 12 August 2025).
  68. Dike, H.U.; Zhou, Y.; Deveerasetty, K.K. Unsupervised learning based on artificial neural network: A review. In Proceedings of the 2018 IEEE International Conference on Cyborg and Bionic Systems (CBS), Shenzhen, China, 25–27 October 2018; Available online: https://ieeexplore.ieee.org/abstract/document/8612259/ (accessed on 12 August 2025).
  69. Gao, T.; Lu, W. Machine learning toward advanced energy storage devices and systems. iScience 2021, 24, 101936. [Google Scholar] [CrossRef] [PubMed]
  70. Memarzadeh, G.; Keynia, F. A new optimal energy storage system model for wind power producers based on long short term memory and Coot Bird Search Algorithm. J. Energy Storage 2021, 44 Pt A, 103401. [Google Scholar] [CrossRef]
  71. He, J. Enhancing Energy Management in New Energy Vehicles and Energy Storage Systems Through Advanced Data Analysis and Machine Learning. Adv. Econ. Manag. Political Sci. 2024, 89, 61–67. [Google Scholar] [CrossRef]
  72. Zhang, R.; Li, X.; Sun, C.; Yang, S.; Tian, Y.; Tian, J. State of Charge and Temperature Joint Estimation Based on Ultrasonic Reflection Waves for Lithium-Ion Battery Applications. Batteries 2023, 9, 335. [Google Scholar] [CrossRef]
  73. Meng, X.; Sun, C.; Mei, J.; Tang, X.; Hasanien, H.M.; Jiang, J.; Fan, F.; Song, K. Fuel cell life prediction considering the recovery phenomenon of reversible voltage loss. J. Power Sources 2025, 625, 235634. [Google Scholar] [CrossRef]
  74. Fathollahi, A. Machine Learning and Artificial Intelligence Techniques in Smart Grids Stability Analysis: A Review. Energies 2025, 18, 3431. [Google Scholar] [CrossRef]
  75. Memarzadeh, G.; Noori, H.; Memarzadeh, R.; Amirteimoury, F.; Keynia, F.; Gholizadeh, M.H. Optimal energy storage system management in an uncertain smartgrid. J. Energy Storage 2025, 117, 116125. [Google Scholar] [CrossRef]
  76. Giannelos, S.; Pudjianto, D.; Bellizio, F.; Qiu, D.; Strbac, G. A comprehensive review of option value in energy economics. IET Renew. Power Gener. 2025, in press. [CrossRef]
  77. Lee, J.H.; Kadero, M.; Ramin, K.; Kamal, R.; Khan, A. AI-Driven Fire Detection in Gas-to-Liquids (GTL) Facilities: A Paradigm Shift in Industrial Safety. Future Eng. J. 2025, 2, 23–27. [Google Scholar] [CrossRef]
  78. Ekunke, O.V.; Kehinde, T.O.; Owunna, I.B. Innovations in Fire Detection and Suppression Systems for Oil Refinery Operations. Path Sci. 2024, 10, 4001–4015. [Google Scholar] [CrossRef]
  79. Zhang, T.; Strbac, G. Novel Artificial Intelligence Applications in Energy: A Systematic Review. Energies 2025, 18, 3747. [Google Scholar] [CrossRef]
  80. Krishna, G.; Singh, R.; Gehlot, A.; Akram, S.V.; Priyadarshi, N.; Twala, B. Digital Technology Implementation in Battery-Management Systems for Sustainable Energy Storage: Review, Challenges, and Recommendations. Multidisciplinary Digital Publishing Institute. Electronics 2022, 11, 2695. [Google Scholar] [CrossRef]
  81. Kwon, K.; Lee, H.-B.; Kim, N.; Park, S.; Joshua, S.R. Integrated Battery and Hydrogen Energy Storage for Enhanced Grid Power Savings and Green Hydrogen Utilization. Multidisciplinary Digital Publishing Institute. Appl. Sci. 2024, 14, 7631. [Google Scholar] [CrossRef]
  82. Greenwood, D.M.; Djapic, P.; Sarantakos, I.; Giannelos, S.; Strbac, G.; Creighton, A. Pragmatic method for assessing the security of supply in future smart distribution networks. CIRED 2020, 2020, 221–224. [Google Scholar] [CrossRef]
  83. Ilo, A.; Prata, R.; Strbac, G.; Giannelos, S.; Bissell, G.R.; Kulmala, A.; Constantinescu, N.; Samovich, N.; Iliceto, A. White Paper ETIP SNET—Holistic Architectures For Power Systems. 2019. Available online: http://hdl.handle.net/20.500.12708/39729 (accessed on 12 August 2025).
  84. Giannelos, S.; Konstantelos, I.; Zhang, X.; Strbac, G. A stochastic optimization model for network expansion planning under exogenous and endogenous uncertainty. Electr. Power Syst. Res. 2025, 248, 111894. [Google Scholar] [CrossRef]
  85. Dong, Z.; Zhang, X.; Zhang, L.; Giannelos, S.; Strbac, G. Flexibility enhancement of urban energy systems through coordinated space heating aggregation of numerous buildings. Appl. Energy 2024, 374, 123971. [Google Scholar] [CrossRef]
  86. Konstantelos, I.; Giannelos, S.; Strbac, G. Strategic valuation of smart grid technology options in distribution networks. IEEE Trans. Power Syst. 2017, 32, 1293–1303. [Google Scholar] [CrossRef]
  87. Meng, Z.; Wang, D.; Li, N.; Zhao, F.; Huang, B.; Hu, J. Study on Evaluation Method of Substitution Effect for Grid-side Functional Alternative Energy Storage. In Proceedings of the 2023 International Conference on Power Energy Systems and Applications (ICoPESA), Nanjing, China, 24–26 February 2023; pp. 197–201. [Google Scholar] [CrossRef]
  88. Giannelos, S.; Borozan, S.; Aunedi, M.; Zhang, X.; Ameli, H.; Pudjianto, D.; Konstantelos, I.; Strbac, G. Modelling Smart Grid Technologies in Optimisation Problems for Electricity Grids. Energies 2023, 16, 5088. [Google Scholar] [CrossRef]
  89. Giannelos, S.; Zhang, T.; Pudjianto, D.; Konstantelos, I.; Strbac, G. Investments in Electricity Distribution Grids: Strategic versus Incremental Planning. Energies 2024, 17, 2724. [Google Scholar] [CrossRef]
  90. Giannelos, S.; Konstantelos, I.; Strbac, G. Investment model for cost-effective integration of solar PV capacity under uncertainty using a portfolio of energy storage and soft open points. In Proceedings of the 2019 IEEE Milan PowerTech, Milan, Italy, 23–27 June 2019; pp. 1–6. [Google Scholar] [CrossRef]
  91. Giannelos, S.; Konstantelos, I.; Strbac, G. Option value of Soft Open Points in distribution networks. In Proceedings of the 2015 IEEE Eindhoven PowerTech, Eindhoven, The Netherlands, 29 June–2 July 2015; pp. 1–6. [Google Scholar] [CrossRef]
  92. Giannelos, S.; Borozan, S.; Strbac, G. Option value of EV smart charging concepts in transmission expansion planning under uncertainty. In Proceedings of the 2022 IEEE 21st Mediterranean Electrotechnical Conference (MELECON), Palermo, Italy, 14–16 June 2022; pp. 63–68. [Google Scholar] [CrossRef]
  93. Giannelos, S.; Konstantelos, I.; Strbac, G. Option value of dynamic line rating and storage. In Proceedings of the 2018 IEEE International Energy Conference (ENERGYCON), Limassol, Cyprus, 3–7 June 2018; IEEE: Piscataway, NJ, USA, 2018; pp. 1–6. [Google Scholar] [CrossRef]
  94. Hwang, S.; Mun-Kyeom, K. Optimal investment strategy based on a real options approach for energy storage systems in the Korean power market. Energy Rep. 2024, 11, 5859–5873. [Google Scholar] [CrossRef]
  95. Hammann, E.; Madlener, R.; Hilgers, C. Economic Feasibility of a Compressed Air Energy Storage System Under Market Uncertainty: A Real Options Approach. Energy Procedia 2017, 105, 3798–3805. [Google Scholar] [CrossRef]
  96. Moriarty, J.; Palczewski, J. Imbalance Market Real Options and the Valuation of Storage in Future Energy Systems. Risks 2019, 7, 39. [Google Scholar] [CrossRef]
  97. Xiu, X.; Li, B. NaN. Study on energy storage system investment decision based on real option theory. In Proceedings of the International Conference on Sustainable Power Generation and Supply (SUPERGEN 2012), Hangzhou, China, 8–9 September 2012. [Google Scholar] [CrossRef]
  98. Qu, J.; Jeon, W. Price and subsidy under uncertainty: Real-option approach to optimal investment decisions on energy storage with solar PV. Energy Environ. 2021, 33, 263–282. [Google Scholar] [CrossRef]
  99. Giannelos, S.; Djapic, P.; Papadaskalopoulos, D.; Strbac, G. Report on Cost-Benefit Analysis and Its Components for the UPGRID Project (Technical Report); UPGRID Project; Schneider Electric: Rueil-Malmaison, France, 2018. [Google Scholar]
  100. Münster, M.; Sneum, D.M.; Pedersen, R.B.; Bühler, F.; Elmegaard, B.; Giannelos, S.; Zhang, X.; Strbac, G.; Berger, M.; Radu, D.; et al. Sector Coupling: Concepts, State-of-the-Art and Perspectives; European Technology and Innovation Platform; Imperial College London: London, UK, 2020. [Google Scholar]
  101. Holttinen, H.; Kiviluoma, J.; Helistö, N.; Levy, T.; Menemenlis, N.; Jun, L.; Cutululis, N.; Koivisto, M.; Das, K.; Orths, A.; et al. Design and Operation of Energy Systems with Large Amounts of Variable Generation: Final Summary Report, IEA Wind TCP Task 25; VTT Technology; VTT Technical Research Centre of Finland: Espoo, Finland, 2021. [Google Scholar] [CrossRef]
  102. Beulertz, D.; Charousset, S.; Most, D.; Giannelos, S.; Yueksel-Erguen, I. Development of a modular framework for future energy system analysis. In Proceedings of the 2019 54th International Universities Power Engineering Conference (UPEC), Bucharest, Romania, 3–6 September 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 1–6. [Google Scholar] [CrossRef]
  103. Most, D.; Giannelos, S.; Yueksel-Erguen, I.; Beulertz, D.; Haus, U.U.; Charousset-Brignol, S.; Frangioni, A. A Novel Modular Optimization Framework for Modelling Investment and Operation of Energy Systems at European Level; Zuse Institute Berlin: Berlin, Germany, 2020. [Google Scholar]
  104. Charousset-Brignol, S.; van Ackooij, W.; Oudjane, N.; Daniel, D.; Noceir, S.; Haus, U.U.; Lazzaro, A.; Frangioni, A.; Lobato, R.; Ghezelsoflu, A.; et al. Synergistic Approach of Multi-Energy Models for a European Optimal Energy System Management Tool; European Dissemination Media Agency: Staffordshire, UK, 2021. [Google Scholar]
  105. Giannelos, S.; Borozan, S.; Konstantelos, I.; Strbac, G. Option value, investment costs and deployment levels of smart grid technologies. Sustain. Energy Res. 2024, 11, 47. [Google Scholar] [CrossRef]
  106. Giannelos, S.; Konstantelos, I.; Strbac, G. Option value of demand-side response schemes under decision-dependent uncertainty. IEEE Trans. Power Syst. 2018, 33, 5103–5113. [Google Scholar] [CrossRef]
  107. Landry, M.; Gagnon, Y. Energy storage: Technology applications and policy options. Energy Procedia 2015, 79, 315–320. [Google Scholar] [CrossRef]
  108. Murrant, D.; Radcliffe, J. Assessing energy storage technology options using a multi-criteria decision analysis-based framework. Appl. Energy 2018, 231, 788–802. [Google Scholar] [CrossRef]
  109. Rahman, M.M.; Oni, A.O.; Gemechu, E.; Kumar, A. Assessment of energy storage technologies: A review. Energy Convers. Manag. 2020, 223, 113295. [Google Scholar] [CrossRef]
  110. Acar, C. A comprehensive evaluation of energy storage options for better sustainability. Int. J. Energy Res. 2018, 42, 3732–3746. [Google Scholar] [CrossRef]
  111. Giannelos, S.; Jain, A.; Borozan, S.; Falugi, P.; Moreira, A.; Bhakar, R.; Mathur, J.; Strbac, G. Long-Term Expansion Planning of the Transmission Network in India under Multi-Dimensional Uncertainty. Energies 2021, 14, 7813. [Google Scholar] [CrossRef]
  112. Borozan, S.; Giannelos, S.; Strbac, G. Strategic network expansion planning with electric vehicle smart charging concepts as investment options. Adv. Appl. Energy 2022, 5, 100077. [Google Scholar] [CrossRef]
  113. Giannelos, S.; Borozan, S.; Moreira, A.; Strbac, G. Techno-economic analysis of smart EV charging for expansion planning under uncertainty. In Proceedings of the 2023 IEEE Belgrade PowerTech, Belgrade, Serbia, 25–29 June 2023; IEEE: Piscataway, NJ, USA, 2023; pp. 1–7. [Google Scholar] [CrossRef]
  114. Giannelos, S.; Konstantelos, I.; Strbac, G. Stochastic optimisation-based valuation of smart grid options under firm DG contracts. In Proceedings of the 2016 IEEE International Energy Conference (ENERGYCON), Leuven, Belgium, 4–8 April 2016; pp. 1–7. [Google Scholar] [CrossRef]
  115. Lin, W.; Hu, X. Effect of option’s time value on decision of exercising or selling option. In Proceedings of the 2011 International Conference on Management Science & Engineering 18th Annual Conference Proceedings, Rome, Italy, 13–15 September 2011; pp. 835–840. [Google Scholar] [CrossRef]
  116. Song, E.Y.; FitzPatrick, G.J.; Lee, K.B.; Griffor, E. A Methodology for Modeling Interoperability of Smart Sensors in Smart Grids. IEEE Trans. Smart Grid 2022, 13, 555–563. [Google Scholar] [CrossRef]
  117. Giannelos, S.; Konstantelos, I.; Strbac, G. Endogenously stochastic demand side response participation on transmission system level. In Proceedings of the 2018 IEEE International Energy Conference (ENERGYCON), Limassol, Cyprus, 3–7 June 2018; pp. 1–6. [Google Scholar] [CrossRef]
  118. Giannelos, S.; Konstantelos, I.; Strbac, G. A new class of planning models for option valuation of storage technologies under decision-dependent innovation uncertainty. In Proceedings of the 2017 IEEE Manchester PowerTech, Manchester, UK, 18–22 June 2017; pp. 1–6. [Google Scholar] [CrossRef]
  119. Carmona, R.; Ludkovski, M. Valuation of energy storage: An optimal switching approach. Quant. Financ. 2010, 10, 359–374. [Google Scholar] [CrossRef]
  120. De Sisternes, F.J.; Jenkins, J.D.; Botterud, A. The value of energy storage in decarbonizing the electricity sector. Appl. Energy 2016, 175, 368–379. [Google Scholar] [CrossRef]
  121. Denholm, P.; Jorgenson, J.; Hummon, M.; Jenkin, T. Value of Energy Storage for Grid Applications; National Renewable Energy Laboratory (NREL): Golden, CO, USA, 2013. [Google Scholar]
  122. Locatelli, G.; Invernizzi, D.C.; Mancini, M. Investment and risk appraisal in energy storage systems: A real options approach. Energy 2016, 104, 114–131. [Google Scholar] [CrossRef]
  123. Bakke, I.; Fleten, S.E.; Hagfors, L.I.; Hagspiel, V. Investment in electric energy storage under uncertainty: A real options approach. Comput. Manag. Sci. 2016, 13, 483–500. [Google Scholar] [CrossRef]
  124. Xiao, B.; Wang, J.; Xiao, Z.; Yan, G.; Dong, L.; Wang, M.; Yang, H. Power Source Flexibility Margin Quantification Method for Multi-Energy Power Systems Based on Blind Number Theory. CSEE J. Power Energy Syst. 2023, 9, 2321–2331. [Google Scholar] [CrossRef]
  125. Giannelos, S. Option Valuation of Smart Grid Technology Projects Under Endogenous and Exogenous Uncertainty. Doctoral Dissertation, Imperial College London, London, UK, 2016. [Google Scholar]
  126. Giannelos, S.; Papadaskalopoulos, D.; Konstantinides, C.V.; Strbac, G. Report on Market Barriers for Deployment of UPGRID Innovative Concepts (Technical Report); UPGRID Project; Schneider Electric: Rueil-Malmaison, France, 2017. [Google Scholar]
  127. Pan, C.; Shao, C.; Hu, B.; Xie, K.; Li, C.; Ding, J. Modeling the Reserve Capacity of Wind Power and the Inherent Decision-Dependent Uncertainty in the Power System Economic Dispatch. IEEE Trans. Power Syst. 2023, 38, 4404–4417. [Google Scholar] [CrossRef]
  128. Giannelos, S.; Bellizio, F.; Strbac, G.; Zhang, T. Machine learning approaches for predictions of CO2 emissions in the building sector. Electr. Power Syst. Res. 2024, 235, 110735. [Google Scholar] [CrossRef]
  129. Borozan, S.; Giannelos, S.; Falugi, P.; Moreira, A.; Strbac, G. Machine learning-enhanced Benders decomposition approach for the multi-stage stochastic transmission expansion planning problem. Electr. Power Syst. Res. 2024, 237, 110985. [Google Scholar] [CrossRef]
  130. Giannelos, S.; Moreira, A.; Papadaskalopoulos, D.; Borozan, S.; Pudjianto, D.; Konstantelos, I.; Sun, M.; Strbac, G. A Machine Learning Approach for Generating and Evaluating Forecasts on the Environmental Impact of the Buildings Sector. Energies 2023, 16, 2915. [Google Scholar] [CrossRef]
  131. Giannelos, S. Reinforcement Learning in Energy Finance: A Comprehensive Review. Energies 2025, 18, 2712. [Google Scholar] [CrossRef]
  132. Giannelos, S.; Zhang, X.; Zhang, T.; Strbac, G. Multi-Objective Optimization for Pareto Frontier Sensitivity Analysis in Power Systems. Sustainability 2024, 16, 5854. [Google Scholar] [CrossRef]
  133. Giannelos, S.; Borozan, S.; Strbac, G. A Backwards Induction Framework for Quantifying the Option Value of Smart Charging of Electric Vehicles and the Risk of Stranded Assets under Uncertainty. Energies 2022, 15, 3334. [Google Scholar] [CrossRef]
  134. Giannelos, S.; Pudjianto, D.; Zhang, T.; Strbac, G. Energy Hub Operation Under Uncertainty: Monte Carlo Risk Assessment Using Gaussian and KDE-Based Data. Energies 2025, 18, 1712. [Google Scholar] [CrossRef]
  135. Amann, G.; Escobedo Bermúdez, V.R.; Boskov-Kovacs, E.; Gallego Amores, S.; Giannelos, S.; Iliceto, A.; Ilo, A.; Chavarro, J.R.; Samovich, N.; Schmitt, L.; et al. E-mobility Deployment and Impact on Grids: Impact of EV and Charging Infrastructure on European T&D Grids: Innovation Needs; Amores, S.G., Ed.; MJ-09-22-246-EN-N; Publications Office of the European Union: Luxembourg, 2022. [Google Scholar] [CrossRef]
  136. Giannelos, S.; Borozan, S.; Strbac, G.; Zhang, T.; Kong, W. Vehicle-to-Grid: Quantification of its contribution to security of supply through the F-Factor methodology. Sustain. Energy Res. 2024, 11, 32. [Google Scholar] [CrossRef]
  137. Giannelos, S.; Djapic, P.; Pudjianto, D.; Strbac, G. Quantification of the Energy Storage Contribution to Security of Supply through the F-Factor Methodology. Energies 2020, 13, 826. [Google Scholar] [CrossRef]
  138. Khalil, M.; Sheikh, S.A. Advancing Green Energy Integration in Power Systems for Enhanced Sustainability: A Review. IEEE Access 2024, 12, 151669–151692. [Google Scholar] [CrossRef]
  139. Giannelos, S.; Papadaskalopoulos, D.; Konstantinides, C.V.; Strbac, G. Report on the Incentive Framework for Facilitating Investments in UPGRID Innovative Concepts (Technical Report); UPGRID Project; Schneider Electric: Rueil-Malmaison, France, 2017. [Google Scholar]
  140. Khan, A.A.; Minai, A.F.; Godi, R.K.; Sharma, V.S.; Malik, H.; Afthanorhan, A. Optimal Sizing, Techno-Economic Feasibility and Reliability Analysis of Hybrid Renewable Energy System: A Systematic Review of Energy Storage Systems’ Integration. IEEE Access 2025, 13, 59198–59226. [Google Scholar] [CrossRef]
  141. Long, X.; Li, Y.; Li, Y.; Ge, L.; Gooi, H.B.; Chung, C.; Zeng, Z. Collaborative Response of Data Center Coupled With Hydrogen Storage System for Renewable Energy Absorption. IEEE Trans. Sustain. Energy 2024, 15, 986–1000. [Google Scholar] [CrossRef]
  142. O’Malley, C.; de Mars, P.; Badesa, L.; Strbac, G. Reinforcement learning and mixed-integer programming for power plant scheduling in low carbon systems: Comparison and hybridisation. Appl. Energy 2023, 349, 121659. [Google Scholar] [CrossRef]
  143. Giannelos, S.; Pudjianto, D.; Strbac, G. Smart home economic operation under uncertainty: Comparing Monte Carlo and stochastic optimization using Gaussian and KDE-based data. Oper. Res. Perspect. 2025, 15, 100348. [Google Scholar] [CrossRef]
  144. Giannelos, S.; Konstantelos, I.; Strbac, G. Optimal supply chain design using machine learning, risk assessment and optimisation applied to coal distribution. EURO J. Decis. Process. 2025, 13, 100062. [Google Scholar] [CrossRef]
Figure 1. Classification framework of AI applications in energy storage systems. The taxonomy organizes the reviewed literature into five main categories that represent the major application domains of AI in Energy Storage Systems. (1) AI Techniques encompassing machine learning fundamentals, deep learning applications, reinforcement learning, and optimization algorithms; (2) Battery Management Systems covering state estimation, health monitoring, predictive maintenance, and fault detection; (3) Energy Storage Optimization including charging strategies, energy management, load forecasting, and system sizing; (4) Grid-Scale Applications addressing grid integration, microgrid management, and renewable energy integration; (5) Thermal Management and Safety covering thermal control and fire prevention. Numbers in parentheses indicate the quantity of papers reviewed in each subcategory.
Figure 1. Classification framework of AI applications in energy storage systems. The taxonomy organizes the reviewed literature into five main categories that represent the major application domains of AI in Energy Storage Systems. (1) AI Techniques encompassing machine learning fundamentals, deep learning applications, reinforcement learning, and optimization algorithms; (2) Battery Management Systems covering state estimation, health monitoring, predictive maintenance, and fault detection; (3) Energy Storage Optimization including charging strategies, energy management, load forecasting, and system sizing; (4) Grid-Scale Applications addressing grid integration, microgrid management, and renewable energy integration; (5) Thermal Management and Safety covering thermal control and fire prevention. Numbers in parentheses indicate the quantity of papers reviewed in each subcategory.
Energies 18 04718 g001
Table 1. Overview of AI Applications in Energy Storage Systems Reviewed.
Table 1. Overview of AI Applications in Energy Storage Systems Reviewed.
Application DomainSpecific ApplicationsAI Techniques UsedReferences
Battery Cell-LevelSoC estimation, SoE estimationLSTM, SVM, neural networks, physics-informed models[7,8,9,10,11,12,13]
SoH estimation, RUL prediction, degradation modelingExplainable AI, deep learning, ensemble methods[14,15,16,17]
Battery System-LevelFailure prediction, anomaly detectionML ensemble methods, neural networks[6,18,19,20,21]
Thermal runaway detection, cell imbalancePattern recognition, big data analytics[22,23,24]
Temperature prediction, thermal controlML/DL models, ANN, SVM[25,26,27]
Energy Storage OperationOptimal charging strategies, cycle optimizationReinforcement learning, deep RL, dynamic programming[5,28,29,30,31]
Real-time control, multi-objective optimizationModel predictive control, deep learning[30,32,33,34,35]
Grid IntegrationDemand prediction, consumption patternsLSTM, GPR, CNN–LSTM hybrid[31,36,37,38,39,40,41]
Optimal capacity, location optimizationGenetic algorithms, PSO, whale optimization[31,42,43,44,45,46]
System-Level applicationsGrid stability, frequency regulationDLR, real-time optimization[47,48,49,50,51]
Distributed control, renewable integrationMulti-agent RL, federated learning[52,53,54,55,56,57]
Wind/solar forecasting, variability managementML, neural networks, time series[58,59,60]
Table 2. Mapping of AI Techniques to Specific Objectives in Energy Storage Systems.
Table 2. Mapping of AI Techniques to Specific Objectives in Energy Storage Systems.
AI TechniquePrimary ObjectivesSystem Application PointKey AdvantagesReferences
Long Short-Term Memory (LSTM)SoC estimation, load forecasting, time-series predictionCell-level BMS, grid-level BMSCaptures long-term dependencies, handles sequential data[7,11,31,36,37,38,39]
Convolutional Neural Networks (CNN)Pattern recognition in battery data, fault detection Battery pack monitoring Spatial feature extraction, noise robustness[39,61]
Support Vector Machines (SVM)SoC/SoH estimation, classification tasksCell-level BMSHigh accuracy with limited data, nonlinear mapping[6,8,9,10]
Random Forest/Ensemble Methods Predictive maintenance, SoH estimationSystem-level diagnosticsHandles heterogeneous data, feature importance ranking[6,18,19]
Reinforcement Learning (RL)Charging optimization, energy arbitrageCharging controllers, grid interfaceAdaptive to changing conditions, multi-objective optimization [5,28,29,32,47,62]
Deep Reinforcement Learning (DRL)Grid management, microgrid controlGrid-scale EMSComplex decision making, real-time adaptation [47,52]
Genetic Algorithms (GA)System sizing, placement optimizationPlanning and design phaseGlobal optimization, discrete variables[45,53]
Particle Swarm Optimization (PSO)Multi-objective optimization, parameter tuningSystem design, control optimizationFast convergence, parallel search[36,43]
Artificial Neural Networks (ANN)General prediction tasks, thermal modeling Multiple levelsUniversal approximation, flexibility[25,26,31]
Explainable AI (Ex-AI)SoH monitoring, safety-critical decisionsBMS, safety systemsInterpretability, trust building[14]
Federated LearningDistributed system optimization, privacy-preserving learningFleet management, distributed storageData privacy, collaborative learning[28,57]
Physics-Informed Neural NetworksModel-based prediction, hybrid modellingAdvanced BMSIncorporates domain knowledge, better generalization[12]
Table 3. Comparison of Deep Learning Architectures for Energy Storage Applications.
Table 3. Comparison of Deep Learning Architectures for Energy Storage Applications.
ArchitectureComputational EfficiencyMemory
Requirements
Training SpeedLong-Term DependenciesReal-Time PerformanceTypical Applications in Energy Storage
LSTMModerateHighSlowExcellentModerateSOC/SOH estimation, long-term forecasting
GRUHighModerateModerateGoodGoodReal-time battery monitoring, short-term prediction
Stacked RNNLowModerateSlowPoorPoorBasic time series analysis
TCNVery HighLowFastVery GoodExcellentHigh-frequency data analysis, edge deployment
Table 4. Method Rankings by Application Domain Based on Reviewed Evidence.
Table 4. Method Rankings by Application Domain Based on Reviewed Evidence.
Application DomainRankMethodPerformance EvidenceReferences
SoC/SoH Estimation1LSTMMAE 0.1, RMSE 0.12[11]
2SVMHigh accuracy with limited data[8,9,10,13]
3Physics-Informed Neural NetworksImproved generalization[12]
Grid-Scale Optimization1Distributed Reinforcement Learning40% disruption reduction, 12.2% cost reduction[47]
2Multi-agent Reinforcement Learning, Federated LearningDistributed control capabilities[52,53,54,55,56,57]
3Model Predictive ControlEstablished baseline[30,32,33,34,35]
System Sizing and Placement1MOPSO22.8% power loss reduction, 71% voltage fluctuation reduction[43]
2Whale OptimizationSuperior to PSO and Firefly[44]
3Genetic AlgorithmsProven effectiveness[45]
Load Forecasting1LSTMDominant in the literature [31,37,38,39]
2CNN-LSTM Hybrid3.4% RMSE reduction[39]
3Gaussian Process RegressionHandles uncertainty well[37]
Predictive Maintenance1Random Forest/Ensemble Feature importance capability[6,18,19]
2Neural NetworksGeneral effectiveness[6,20,21]
3Explainable AITrust and interpretability[14]
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhang, T.; Strbac, G. Artificial Intelligence Applications for Energy Storage: A Comprehensive Review. Energies 2025, 18, 4718. https://doi.org/10.3390/en18174718

AMA Style

Zhang T, Strbac G. Artificial Intelligence Applications for Energy Storage: A Comprehensive Review. Energies. 2025; 18(17):4718. https://doi.org/10.3390/en18174718

Chicago/Turabian Style

Zhang, Tai, and Goran Strbac. 2025. "Artificial Intelligence Applications for Energy Storage: A Comprehensive Review" Energies 18, no. 17: 4718. https://doi.org/10.3390/en18174718

APA Style

Zhang, T., & Strbac, G. (2025). Artificial Intelligence Applications for Energy Storage: A Comprehensive Review. Energies, 18(17), 4718. https://doi.org/10.3390/en18174718

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Article metric data becomes available approximately 24 hours after publication online.
Back to TopTop