A Systematic Mapping Study on State Estimation Techniques for Lithium-Ion Batteries in Electric Vehicles

Abstract

The effective administration of lithium-ion batteries is key to the performance and durability of electric vehicles (EVs). This systematic mapping study (SMS) thoroughly examines optimization methodologies for battery management, concentrating on the estimation of state of health (SoH), remaining useful life (RUL), and state of charge (SoC). The findings disclose various methods that boost the accuracy and reliability of SoC, including enhanced variants of the Kalman filter, machine learning models like long short-term memory (LSTM) and convolutional neural networks (CNNs), as well as hybrid optimization frameworks that combine Grey Wolf Optimization (GWO) and Particle Swarm Optimization (PSO). For estimating SoH, prevalent data-driven techniques include support vector regression (SVR) and Gaussian process regression (GPR), alongside hybrid models merging machine learning with conventional estimation techniques to heighten predictive accuracy. RUL prediction sees advancements through deep learning techniques, especially LSTM and gated recurrent units (GRUs), improved using algorithms such as Harris Hawks Optimization (HHO) and Adaptive Levy Flight (ALF). This study underscores the critical role of integrating advanced filtering techniques, machine learning, and optimization algorithms in developing battery management systems (BMSs) that enhance battery reliability, extend lifespan, and optimize energy management for EVs. Moreover, innovations like hybrid models and synthetic data generation using generative adversarial networks (GANs) further augment the robustness and precision of battery management strategies. This review lays out a thorough framework for future exploration and development in the optimization of EV batteries.

1. Introduction

Over the past few years, the acceptance of electric vehicles (EVs) has surged, largely due to factors such as the escalation of carbon dioxide (CO₂) emissions, dwindling fossil fuel resources, and the worldwide drive towards the renewable energy transition [1]. The swift growth of EVs is seen as a key strategy for addressing environmental problems, especially by cutting down greenhouse gas emissions and reducing reliance on fossil fuels.

The global shift towards sustainable energy and the rapid increase in EV adoption have made lithium-ion batteries (LIBs) a cornerstone of energy storage systems [2]. Renowned for their high energy density, long cycle life, and low self-discharge rates, LIBs are the preferred power choice for EVs [3]. Nonetheless, effectively managing these batteries presents significant obstacles. Improving their performance is essential for prolonging battery life, maintaining safety, and enhancing EV efficiency.

The management of LIBs encompasses several key aspects, including state of charge (SoC) estimation, thermal management, state of health (SoH) monitoring, and charge–discharge optimization. These factors directly influence the performance and reliability of EVs [4]. As demand for electric vehicles continues to increase, the need for advanced battery management strategies becomes increasingly urgent to address issues such as thermal runaway, degradation, and energy inefficiencies that can affect both the driving range and the longevity of the batteries [5].

In recent years, a multitude of optimization strategies have emerged to enhance the performance of LIBs in electric vehicles, reflecting the critical role these technologies play in advancing sustainable transportation. These strategies span a wide spectrum, encompassing algorithmic approaches such as predictive modeling, machine learning-based battery management systems, and real-time monitoring algorithms.

Simultaneously, advancements in hardware design, such as enhanced thermal management systems and improved cell architecture, have significantly contributed to prolonging battery life and improving energy density [4]. Similarly, system-level software innovations, including adaptive control algorithms and integrated energy management frameworks, have aimed at optimizing energy usage while maintaining vehicle safety and reliability. Regrettably, despite these efforts, the diversity of techniques, their overlap, and their varying effectiveness in different operational scenarios create a fragmented knowledge base, making it challenging for researchers and industry professionals to identify the most suitable solutions. Moreover, many approaches focus on isolated aspects of optimization, potentially overlooking synergies between hardware, software, and system-level innovations. This complexity underscores the need for a comprehensive review that not only evaluates the strengths and limitations of existing strategies, but also explores their integration potential. A comprehensive review is important to guide future research and enable the integration of these innovations into commercial electric vehicles, accelerating the shift to greener and more efficient transportation with multiple environmental benefits.

This article presents a systematic mapping study (SMS) regarding optimization techniques for managing lithium-ion batteries in electric vehicles. An SMS is a methodologically rigorous approach designed to provide a comprehensive overview of a wide research domain. It systematically identifies and categorizes existing studies to uncover key themes, prevailing trends, and knowledge gaps within the field. Distinct from systematic literature reviews, which are tailored to address specific research questions, systematic mapping studies focus on organizing and structuring the body of knowledge based on predefined dimensions such as research objectives, methodologies, or study outcomes. This method offers valuable insights into the research landscape, facilitates the identification of underexplored areas, and aids in the strategic prioritization of future research endeavors [6]. By integrating recent progress in this area, our goal is to offer an extensive summary of present methodologies, pinpoint emerging trends, and highlight existing research gaps. This work aims to aid the continued advancement of more efficient, reliable, and sustainable battery management systems, paving the way for the future of electric transportation.

By mapping findings from various research studies and industrial advancements, this SMS seeks to provide information on the progression of charging schemes and identify gaps and opportunities for further development in this rapidly evolving field.

2. Background

This section presents the foundational knowledge necessary to understand the research problem and its context. It introduces key concepts central to the study, offering the reader the necessary framework to comprehend the research objectives and methodology.

2.1. Electric Vehicle

An EV serves as a means of transport powered by electricity. Electric motors are widely used in many vehicle types, where the power source may be external, drawing electricity from off-board systems through a collection system, or housed within the vehicle itself, using components such as batteries [7].

EVs represent a pivotal technology for mitigating petroleum consumption and reducing greenhouse gas emissions [8]. Numerous countries have recognized their environmental benefits and implemented financial incentives to encourage EV adoption and expand ownership globally. EVs are also increasingly viewed as viable alternatives to internal combustion engine (ICE) vehicles [9]. EV technologies offer significant advantages, including lower greenhouse gas emissions, reduced air pollution, and improved energy efficiency, making them integral to the transition to sustainable transportation.

All electric vehicles are equipped with batteries, and their performance is related to that, particularly for electric vehicles with an internal power supply. The widespread adoption of electric vehicles depends on ongoing progress in battery technology, particularly improvements in energy storage capacity, decreases in charging time, and cost-effectiveness. Hence, a fascinating research field is the state estimation of batteries [4].

2.2. Lithium-Ion Batteries

Lithium-ion batteries (LIBs) have solidified their role as crucial elements in energy storage advancements [9]. In recent times, lithium-ion batteries have emerged as the preferred power source for electric vehicles due to their high energy and power densities, extended service life, and eco-friendliness [10]. These batteries require a battery management system (BMS) to control charging and discharging, and to monitor battery status and optimize its state for safety, reliability, and optimal performance. Factors such as the number of charge and discharge cycles, environmental temperature, voltage levels, and current flows significantly influence the state of health of the battery [11].

2.3. State of Health

Assessing the SoH of lithium-ion batteries is essential yet difficult, especially in the context of electric vehicles [12]. State of health acts as a critical performance metric, evaluating the battery’s condition and its ability to meet fundamental operational requirements, such as available capacity or permissible instantaneous charge/discharge power. It is commonly characterized by relative variations in capacity or internal resistance throughout the battery’s charge–discharge cycles, offering important insights into the battery’s degradation over time. This is commonly represented by the ratio of the remaining battery capacity to its initial nominal capacity, depicting its current performance compared to its original design specifications [13].

2.4. State of Charge

Estimating the SoC is crucial in a BMS; making it accurate is vital for effective cell balancing, which helps prolong battery pack life and reduce risks such as overheating, explosion, and swelling [4]. In Li-ion batteries, SoC estimation depends mainly on inputs like voltage, current, and temperature to evaluate the battery’s status accurately [4]. Traditional methods for estimating SoC involve open-circuit voltage measurement, coulomb counting, and techniques based on electrical or electrochemical models. SoC estimation in an EV is used to estimate the remaining range of the EV for the driver, who can schedule the charging of the battery [14].

3. Related Work

This section provides an overview of prior studies and advancements closely related to the research presented. By contextualizing the current work within the existing body of knowledge, the foundation for the research problem is established. It also differentiates this study’s contribution by comparing and contrasting it with relevant approaches, methodologies, and findings from previous works. This discussion ensures the research is framed within a broader academic context, demonstrating its relevance and significance. It shows existing research in the literature that concentrates on different topics related to battery management.

In [15], future developments in state of charge (SoC) estimation methodologies are explored. The study presents a comprehensive overview of global research, categorizing publications by country and type of document. It examines several SoC estimation techniques, such as open-circuit voltage measurement, coulomb counting, model-based approaches, and data-driven strategies. In particular, the research emphasizes the advantages of data-driven models in improving SoC estimation accuracy. However, the authors suggest that although optimizing current limits may enhance battery life and decrease safety risks, further research is required to determine optimal values and their relationship with other system variables.

The study in [16] examines how the current limits of charge and discharge affect the degradation and safety of lithium-ion batteries in electric vehicles. It explores the electrochemical processes, mechanical stresses, and thermal effects that drive battery deterioration. The research highlights that higher current densities accelerate capacity loss, increase impedance, and increase the risk of thermal runaway. It emphasizes balancing system performance with safety by proposing strategies adapted to battery chemistry, operating temperature, and state of charge. A key focus is integrating current limit optimization into battery design and operational strategies to extend battery life and enhance safety. The study provides valuable information on the role of current limits in mitigating degradation and improving the reliability of electric vehicle batteries, advancing the development of more efficient and reliable transportation and energy storage systems.

In [17], the researchers evaluate the efficacy of a neural network (NN) algorithm against the coulomb counting technique for estimating the SoC in batteries, specifically within the framework of electric vehicle BMSs. Ensuring accurate SoC estimation is essential for preventing both overcharging and overdischarging, contributing to extending battery life. Moreover, reliable SoC information allows the control system to make better decisions, thereby optimizing energy use in electric vehicles. A significant advantage of the NN approach compared to the coulomb counting method is its capability to be incorporated into BMS hardware, utilizing real-time measurements of current, voltage, and temperature to achieve more precise and adaptable SoC estimation.

In [11], the authors review battery modeling approaches, focusing on electrochemical, equivalent circuit, and data-driven models. It presents the basic principles, applications, and parameters of each. The electrochemical model is the most detailed and accurate, describing internal physical processes using partial differential equations (PDEs). Predicting battery performance and aging requires a deep understanding of these model parameters. The study critically examines parameter estimation methods for electrochemical models, categorizing them into online, offline, and analytical approaches. It emphasizes the importance of online and offline methods for real-time applications and provides a comprehensive state-of-the-art review of each. The analysis reveals that combining online and offline methods delivers superior performance compared to using either method alone.

The research detailed in [9] explores the factors leading to degradation in lithium-ion batteries, highlighting that although battery aging is unavoidable, it can have reduced impact by considering the operational conditions of the vehicle. Different conditions influence the aging process in various ways, making it essential to understand the factors that impact battery capacity to reduce degradation. The paper first provides a detailed explanation of the internal degradation mechanisms, followed by an analysis of the key factors that contribute to battery deterioration during the operation of electric vehicles. In addition, it discusses various techniques used to model battery degradation and predict remaining battery life, as well as methods to slow the aging process. The study also highlights existing research gaps, challenges in accurately predicting battery lifespan, and strategies to mitigate battery degradation in electric vehicles.

The review paper in [4] examines the primary challenges, concerns, and solutions of battery technology, focusing on battery management systems. It provides an in-depth analysis of fundamental BMS technologies, including battery modeling, state estimation, and charging strategies. The study evaluates various battery models, encompassing electrical, thermal, and electrothermal frameworks, and explores techniques for assessing battery charge and health. In addition, it discusses several charging methodologies and optimization strategies. The article emphasizes the critical roles of a BMS, including monitoring voltage and current, estimating charge and discharge states, balancing and safeguarding the battery, regulating temperature, and managing battery data.

In [18], a comprehensive overview of recent advances in intelligent BMSs for electric and hybrid vehicles is presented, focusing on fundamental mathematical principles, methodologies, and practical applications. The research analyzed focuses on three core research domains: Estimation of state of charge, evaluation of the state of health of battery packs, and prediction of the remaining driving range.

The study presented in [19] offers a detailed examination of advanced techniques for managing critical parameters in battery systems, including SoC, SoH, thermal management, and cell balancing. These techniques are essential for safeguarding batteries from potentially harmful conditions such as overcharging, excessive discharging, and overcurrent scenarios, which can compromise battery performance and lifespan. The authors provide a comprehensive literature survey on BMSs, emphasizing their pivotal role in ensuring the safe, reliable, and efficient operation of batteries in EVs. This review not only highlights state-of-the-art methodologies but also identifies challenges and opportunities in BMS design and implementation, contributing to the development of robust solutions for EV applications. By addressing both operational reliability and safety concerns, the study underscores the importance of BMS advancements in the broader context of sustainable energy storage and electric mobility.

The review presented in [20] provides an in-depth analysis of various methodologies for estimating the SoC in EV batteries, offering a comprehensive discussion on both traditional and advanced computational techniques. The study categorizes SoC estimation methods into conventional approaches, which often rely on empirical models and physical principles, and modern computer-based techniques, which leverage computational algorithms, machine learning, and data-driven models to enhance accuracy and adaptability. By systematically classifying these methods, the review sheds light on their underlying principles, strengths, limitations, and applicability in different EV battery systems. This work serves as a valuable resource for understanding the evolution of SoC estimation techniques, highlighting the trade-offs between computational complexity, precision, and real-world implementation challenges. Furthermore, it underscores the critical role of accurate SoC estimation in optimizing battery performance, extending lifespan, and ensuring the safe and efficient operation of EVs.

The study presented in [21] offers a comprehensive review of physics-based models (PBMs) developed to analyze and understand the degradation mechanisms in lithium-ion batteries. These models play a crucial role in elucidating the underlying physical and chemical processes that contribute to battery aging and performance deterioration. The review emphasizes the importance of leveraging PBMs to ensure the reliable and safe operation of lithium-ion batteries, with a particular focus on strategies to extend their operational lifespan. By providing detailed insights into the degradation pathways and their interactions, the study highlights the potential of PBMs to enhance predictive capabilities, guide the design of advanced battery management systems, and inform the development of next-generation battery technologies.

4. Methodology

This research follows established guidelines for conducting a systematic mapping study (SMS) [6,22,23]. As in similar studies, a structured approach is applied to provide a comprehensive overview of the research area. The study was organized into five main stages: study planning, searching for primary studies, study selection, quality assessment, and reporting. Figure 1 illustrates the process flow for the presented SMS.

In the planning stage, the scope of the study was defined, including the primary goal and research questions (RQs). Based on these RQs, the SMS protocol was developed, which included selecting a search strategy, formulating search strings, setting inclusion/exclusion criteria for research work, creating a classification scheme and extraction process, determining visualization methods for the results, and identifying potential threats to validity. The protocol ensured quality through peer-review validation and adherence to established standards, vocabularies, and taxonomies relevant to the field.

During the conduct stage, the SMS protocol was implemented. The selection of the initial research work and the application of inclusion/exclusion criteria were performed iteratively to refine the search strings and criteria. Once validated through testing and peer review, a final set of primary studies was identified based on the criteria. Then, these were categorized using the classification scheme.

The reporting stage involved analyzing the SMS results, answering the RQs, identifying research gaps and trends, and providing future research directions. Section 4.1 gives an overview of the study scope, while Section 4.2 details the paper selection strategy. Section 4.3 explains the inclusion/exclusion procedure, and Section 4.4 and Section 4.5 describe the classification scheme and procedure, respectively.

4.1. Study Planning

This research concentrates on pinpointing strategies and solutions for managing lithium-ion batteries in electric vehicles. This SMS is designed to evaluate the existing battery management strategies and to detect trends and challenges within the field. The research aims to address the following questions:

RQ1. Which research topics in battery management are currently being addressed in the domains of RUL, SoC, and SoH? Knowing the strategies used provides a specific goal and guides decision making during the use and development of EVs. Identifying and understanding the methods, techniques, and technologies included in the strategies provides insight into how closely they resemble real-world scenarios and identifies potential factors that can be considered to improve them.

RQ2. What strategies are the most commonly used? Analyzing the proposals enables us to examine adaptability, accuracy, and robustness to address battery management challenges.

RQ3. What challenges do batteries face in electric vehicles that use lithium-ion batteries? This is essential because acknowledging these challenges paves the way for advancements in this area and facilitates the creation of new proposals.

4.2. Searching for Primary Studies

A database-driven approach was employed to retrieve high-quality peer-reviewed literature, focusing on journals, conference proceedings, and book chapters. Reputable digital libraries in EV research, such as IEEE Xplore, Springer Link, and ScienceDirect, were selected.

The search string was constructed based on our SMS scope and utilized established domain-specific vocabularies and taxonomies. This strategy aimed to capture all relevant literature by including key terms such as ‘electric vehicle’, ‘batteries management technique’, ‘optimization’, and ‘Lithium ion’ to cover a wide range of related research.

Our search string combined these terms with ’AND’ conjunctions, for example, ‘electric vehicle’ AND ‘batteries management technique’ AND ‘optimization’ AND ‘lithium ion’, to ensure relevance and avoid unrelated articles containing isolated keywords. Once refined, the final query was executed in October 2024, targeting titles, abstracts, and keywords, yielding a total of 2291 articles.

4.3. Study Selection

The inclusion and exclusion criteria were defined in accordance with the research questions to ensure that the findings could be accurately interpreted and classified appropriately within the scope of the study.

The following inclusion criteria were established to identify relevant publications that effectively addressed the research questions:

• Publication date of 1 January 2019 to 31 December 2024;
• A new battery management proposal for EVs is presented;
• Only proposals that use lithium-ion batteries are considered.

For the exclusion criteria, we used the following:

• Articles that were not related to EVs;
• Non-peer-reviewed articles;
• Scientific articles in which the written language was not English;
• Research in which a method was manifestly missing.

4.4. Quality Assessment

As shown in

Table 1. Primary studies obtained from each source.

Source	First Search	Inclusion/Exclusion	Quality Assessment
IEEE	77	64	30
Springer	943	433	11
ACM	238	16	8
ScienceDirect	1033	225	14
Total	2291	738	63

, potential primary studies were filtered through a structured inclusion and exclusion process. Some papers retrieved through the search query fell outside the study’s scope, so manual processing steps were applied. The inclusion criteria considering filtering for English-language papers, removal of duplicates, specific document types (conference papers and journals), and publication period constraints.

Following the first phase, we manually selected titles, abstracts, and full texts. This manual process focused on identifying primary contributions in articles relevant to our study and excluding those that did not meet the inclusion criteria.

From the initial assessment of 2291 paper titles, 738 candidates studies were shortlisted as either “included” or “unclear”, moving to the subsequent phase. Throughout the full-text screening, the screener meticulously examined the title, abstract, introduction, and conclusions, ultimately selecting 63 articles for inclusion in the study, as outlined in Table 1.

The sections that follow present and examine the proposals that were identified as pertinent to this study.

4.5. Classification Scheme

The studies were categorized based on the emphasis of their proposals to examine the trends in research papers on state of health (SoH), remaining useful life (RUL), and state of charge (SoC). The classification scheme went through several iterations. Each research paper was attributed to one of these three areas according to the proposal put forward. Subsequently, the specific method proposed in each principal study was identified to aid in understanding how research on the relevant subject is progressing. Each paper was individually classified and the process refined throughout the research endeavor. Figure 2, Figure 3 and Figure 4 illustrate the finalized classification scheme applied to the 63 papers, thereby enabling us to answer RQ1, RQ2, and RQ3.

5. Mapping Results

This section presents the results of the SMS addressing our research questions. The findings are based on collective observations and identified trends, providing a comprehensive, research-driven perspective rather than individual interpretations.

5.1. RQ1. Which Research Topics in Battery Management Are Currently Being Addressed in the Domains of RUL, SoC, and SoH?

Figure 2 shows a gradual and notable increase in the total number of publications. This increase is particularly significant in the last two years, 2023 and 2024, when the number of publications reaches its maximum values. SoC research has been a predominant area during the years studied; however, in the last year, RUL research has increased noticeably, representing a significant portion of the total publications. The graph reflects sustained growth in research related to SoC, SoH, and RUL. The predominance of SoC indicates a greater interest in this category, while the increase in RUL in recent years suggests a diversification of study areas. This pattern suggests an evolution in the research topic and a growing interest in analyzing systems’ useful life and state. The techniques identified in the primary studies are summarized in

Table 2. Techniques identified in primary studies.

Acronym	Technique
APID	Adaptive proportional–integral–derivative
Bi-LSTM	Bidirectional long short-term memory
BNs	Bayesian networks
BNNs	Boltzmann neural networks
CNNs	Convolutional neural networks
DDTS	Data-driven time series
DL	Deep learning
DNNs	Deep neural networks
EKF	Extended Kalman filter
FL	Fuzzy logic
FOMs	Fractional-order models
GAC	Genetic Algorithm Clustering
GANs	Deep convolutional generative adversarial networks
GCNs	Graph convolutional networks
GPR	Gaussian process regression
GRUs	Gated recurrent units
GWO	Grey Wolf Optimization
HHO	Harris Hawks Optimization
LSM	Least squares method
ML	Machine learning
NNn	Neural networks
PCB	Passive cell balancing
PPO	Proximal Policy Optimization
RFR	Random Forest Regressor
RL	Reinforcement learning
RLS	Recursive least squares
LSTM	Long short-term memory
SVR	Support vector regression
SCA	Sine Cosine Algorithm

.

As illustrated in Figure 3, machine learning (ML) techniques are the predominant focus of current research addressing RUL, SoC, and SoH (50%). In particular, primary studies highlight the extensive use of methods such as CNN, DL, Bayesian networks, NN, DNN, Boltzmann neural networks, GCN, GAN, Genetic Algorithm Clustering, RL, RLS, GPR, LST, and LSM (13 primary studies applied this one). These approaches are often combined with complementary ML techniques to improve the efficiency and optimization of battery management systems. The second most used technique is EKF, which is not traditionally categorized as a core machine learning algorithm, but it is a significant algorithmic tool used in conjunction with machine learning in specific contexts, particularly for tasks involving dynamic systems, state estimation, and real-time learning. The third place is for PCB, which is a method to ensure that all cells within a battery pack remain at the same voltage level during charging and discharging cycles.

5.2. RQ2. What Strategies Are the Most Commonly Used?

Based on the summarized results, the reviewed studies predominantly employ the following strategies. When estimating the SoC, the primary methodologies include the following:

• Advanced filtering techniques such as Kalman filters and related variations coupled with Particle Swarm Optimization for tuning purposes.
• Machine learning methods, encompassing convolutional neural networks, long short-term memory, and hybrid frameworks.
• Optimization algorithms, like Particle Swarm Optimization and Bayesian optimization, used to fine-tune hyperparameters. Some innovative approaches involve Grey Wolf Optimization and Genetic Algorithms for optimizing energy allocation and parameter forecasting.
• Hybrid models.

For SoH estimation, the most common approaches include the following:

• Machine learning models.
• Optimization and feature engineering.
• Hybrid frameworks (a combination of data-driven and conventional models, such as dual GPRs and autoregressive models, that thoroughly capture degradation characteristics).

Meanwhile, RUL estimation primarily employs the following:

• Deep learning models.
• Optimization techniques like Harris Hawks Optimization and Adaptive Levy Flight, used with particle filters to enhance model performance.
• Feature extraction and data fusion.
• Real-time applications; for example, integration with interfaces like Ganesan’s UI system supports real-time monitoring, vital for smart city and electric vehicle applications.

Together, these strategies emphasize the critical role of integrating advanced algorithms, effective optimization, and real-time adaptability in managing the complexities involved in battery performance estimation.

5.3. RQ3. What Challenges Do Batteries Face in Electric Vehicles That Use Lithium-Ion Batteries?

The reviewed studies reveal a robust and diverse methodology to improve the accuracy and reliability of state of charge estimation in lithium-ion batteries. Primary approaches include advanced filtering algorithms, machine learning techniques, and hybrid optimization models, emphasizing real-time adaptability and robustness under varying conditions. The convergence of advanced filtering methods, machine learning, and optimization frameworks marks a significant leap in the estimation of SoC, ensuring improved battery reliability, extended useful life, and efficient energy management. Together, these methods provide a comprehensive toolbox for addressing the challenges of SoC prediction in modern electric vehicle applications.

Recent advances in estimating lithium-ion batteries’ SoH highlight the growing significance of data-driven and hybrid approaches for effective battery health monitoring and prognostics. The reviewed studies reveal similarities in the use of advanced machine learning, optimization, and data processing techniques to address challenges in SoH estimation. These studies combine advanced algorithms, robust optimization, and tailored feature engineering to improve the accuracy of SoH estimation. The collective findings underscore the importance of leveraging diverse data sources and innovative methodologies to meet the demands of evolving battery technologies.

The reviewed papers explore various machine learning (ML) and deep learning (DL) techniques to estimate the RUL of lithium-ion batteries, a critical task for enhancing battery longevity and reliability in EVs and other applications. All papers aim to improve battery RUL prediction accuracy by leveraging different algorithms or frameworks, particularly for lithium-ion batteries. This shared objective reflects the importance of accurate RUL estimation in battery management systems, which is essential for EV infrastructure. Techniques like GRUs and optimized LSTM models (e.g., attention-based, hyperparameter-tuned) promise efficiency and accuracy in time-series data applications. Efficiency here refers to the information’s capacity to promptly aid operators and decision-makers in planning battery replacements, budgets, and resources [58]. Meanwhile, fusion models and AutoML frameworks provide robust predictions, albeit with high computational costs.

Future research should focus on integrating synthetic data generation, optimizing computational resources, and exploring new hybrid frameworks to ensure greater adoption and scalability in practical applications. These efforts will be instrumental in the advancement of electric vehicle technology and the transition to sustainable energy solutions.

6. Synthesis of Primary Studies

This section provides a comprehensive synthesis of the key aspects of each primary study analyzed, with a particular emphasis on elements that, while not directly addressing the predefined research questions of this SMS, hold substantial relevance to the broader context of battery management. These elements are instrumental in providing deeper insight into the field, shedding light on secondary findings, emerging trends, and critical challenges that extend beyond the immediate scope of research questions. By identifying and highlighting these aspects, this synthesis not only enriches the content of the current investigation, but also lays a solid foundation for the development of future research trajectories. These findings can inform the prioritization of unresolved issues, guide methodological advances, and inspire innovative approaches in battery management systems. Furthermore, this extended analysis underscores the interconnected nature of the field, demonstrating how various studies contribute to a holistic understanding of key themes, such as efficiency optimization, mitigation of degradation, and integration of advanced technologies in energy storage solutions.

This synthesis serves as a valuable resource for researchers and practitioners alike, fostering a deeper understanding of the multifaceted challenges and opportunities within the field of battery management. Using these insights, future studies can build upon this foundational knowledge to address critical gaps, refine existing strategies, and advance the state of the art in sustainable energy storage technologies.

6.1. Focus on State of Charge

The SoC is a critical parameter in managing battery systems for EVs. Accurate monitoring and estimation of SoC are essential for ensuring the battery’s efficiency, safety, and longevity. This section shows the primary studies selected in the SMS.

The research presented in [26] proposes the use of an extended Kalman filter (EKF) model for SoC estimation, demonstrating superior accuracy compared to more straightforward methods, such as coulomb counting. However, despite its improved precision, potential challenges in the real-time implementation of EKF should be considered.

In the research published in [27], the authors present a Kalman filter (KF)-based algorithm designed to improve the accuracy of SoC estimation in EVs. The algorithm is trained using a machine learning approach grounded in Proximal Policy Optimization (PPO). Although this method introduces a higher computational burden than conventional SoC estimation techniques, it significantly improves accuracy, particularly in initial error scenarios. Furthermore, the model demonstrates robust performance under uncertainty in assumptions or initial information, provided accurate data become available over time. The approach was validated using experimental data from a standard lithium-ion battery cell, with results confirming its effectiveness in delivering precise SoC estimations.

The study presented in [28] applies the extended Kalman filter (EKF) method to estimate the SoC in lithium-ion batteries. Integrating the Multi-Objective Cuckoo Search (MOCS) with EKF, the authors address SoC estimation during dynamic stress testing (DST) conditions. A comparative assessment against the Particle Swarm Optimization (PSO) algorithm reveals the enhanced performance of the combined MOCS+EKF approach, emphasizing its reliability and precision in SoC estimation.

The study presented in [29] introduces an enhanced battery model that leverages an optimized EKF for accurate SoC estimation. This model accounts for key factors like temperature, aging, and self-discharge, which ensure precise estimations. Before estimating the SoC, the model evaluates capacity degradation using a simplified methodology. Additionally, a Particle Swarm Optimization algorithm is employed to fine-tune the process noise covariance, improving state estimation performance. Comparative analyses with existing methods demonstrate the proposed approach’s ability to balance high accuracy and computational efficiency.

In [30], a method is introduced to estimate the charging state of EV lithium-ionn batteries by combining a second-order circuit model with an adaptive unscented Kalman filter (AUKF). An adaptive factor dynamically adjusts the noise covariance in the estimation process using the innovation vector. Comparative results indicate that the AUKF significantly enhances estimation accuracy over the standard UKF, demonstrating strong performance in accurately predicting battery charge state.

The research in [31] introduces a cloud-based SoC estimation algorithm, leveraging the extensive computational and storage capabilities of the cloud platform. The research investigates the impact of time-varying model parameters on the accuracy of SoC estimation. It enhances adaptability through the Noise Matrix Self-Adjustment Extended Kalman Filter (NMSA-EKF), which is designed to handle long data transmission intervals and low transmission precision. A hybrid parameter identification method combining a direct approach with the variable forgetting factor recursive least squares (VFFRLS) algorithm is also evaluated. Using cloud-sourced discharge data, NMSA-EKF achieves SoC estimation with a relative error under 3%, demonstrating the robustness and precision of the proposed methodology.

In [32], an advanced second-order adaptive extended Kalman filter (AEKF) is proposed to minimize the truncation error and improve the accuracy of SoC estimation. Recognizing the influence of the sliding window length (SWL) on the precision of the estimation, a correspondence analysis is performed to identify an optimal SWL, ensuring stable accuracy across varying conditions without the need for SWL adjustments. The effectiveness of the algorithm is validated using diverse datasets and temperature ranges. The experimental results confirm that the proposed second-order AEKF achieves superior accuracy and robustness in the estimation of SoC.

In [33], a CNN-based method is introduced for estimating lithium-ion battery SoC. The model’s effectiveness was assessed using actual data from the New York City Cycle (NYCC) at temperatures of 15 °C, 25 °C, and 45 °C. To improve precision, the model’s validation involved modifying hyperparameters, particularly the filter count in convolutional layers. The findings reveal that the CNN proposed is on par with conventional model-based techniques, highlighting its ability to process and analyze battery data for SoC estimation.

The research detailed in [34] estimated SoC using a convolutional neural network (CNN) model. To enhance the CNN architecture, three optimization algorithms were applied: Particle Swarm Optimization (PSO), Elephant Search Algorithm (ESA), and Equilibrium Optimization (EO). The sensor data from lithium-ion batteries were processed and fed into the CNN and the three optimized CNN models. These models were evaluated using error, R², and time metrics to determine the best approach. The CNN-ESA model, a novel combination, outperformed the others, achieving the lowest error rates and the highest R² value of 0.9987. The results highlight the unique effectiveness of ESA in improving CNN architectures for more accurate SoC estimates, contributing to greater efficiency and longer lifespan of electric vehicles.

The study in [35] proposes a hybrid method combining deep learning with a particle filter (PF) to estimate the SoC in lithium-ion batteries, addressing the PF’s dependence on mechanism models. The approach involves constructing a battery degradation model using convolutional neural networks (CNNs) to extract key health features and predict degradation trends. The CNN output is then refined using the PF algorithm for precise SoC estimation. Validation on two lithium-ion battery datasets demonstrates the method’s effectiveness in tracking degradation, delivering accurate and stable predictions, and achieving model independence, making it well suited for practical applications.

A framework based on convolutional neural networks is presented in [36] to directly estimate lithium-ion batteries’ SoC using voltage, current, and temperature data collected during charging. The CNN model was trained on randomized data, with added noise and error to enhance its robustness. The framework incorporates multiple CNN layers and neurons to improve accuracy, and the model’s performance was tested across a range of temperature distributions to reflect real-world conditions.

The primary study detailed in [14] employs two machine learning algorithms, support vector regression (SVR) and XGBoost, to estimate the SoC of Li-Iron-Phosphate battery cells based on experimental test data. SVR, derived from the support vector machine (SVM) methodology, is widely utilized in data science for regression tasks. At the same time, XGBoost introduces an advanced gradient-boosting framework featuring parallel computation and accelerated training times. The research compares the two algorithms regarding implementation complexity, performance, accuracy, and processing speed. Through parameter tuning, the SoC estimations achieved a coefficient of determination ranging from 97% to 99%, demonstrating the effectiveness of both approaches.

The research in [37] investigates the use of deep neural networks to accurately estimate the SoC in electric vehicles. By combining historical data with real-time parameters, DNNs improve SoC prediction, enabling more efficient charging strategies and improving overall EV performance and reliability. The results of the study support advances in the management of charging networks, highlighting improved accessibility, cost-effectiveness, user satisfaction, and the promotion of sustainable transportation solutions.

The study published in [38] introduces a novel framework employing the least squares support vector machine (LSSVM) approach to improve the accuracy of battery state of charge estimation. A comparative evaluation against the conventional ampere-hour (AH) method demonstrates the enhanced precision of the LSSVM model in predicting SoC.

The study in [39] introduces a graph convolutional network (GCN)-based model for improved performance and interpretability in the analysis of EV data, surpassing traditional recurrent neural network (RNN) models. Unlike RNNs, the GCN model uses a learnable adjacency matrix to dynamically represent variable relationships during training. Two adjacency matrices are utilized: one treating variables as nodes and the other considering timestamps as nodes. This dual approach enables the model to interpret the data from different perspectives, enhancing clarity in the relationships between variables. Experimental results with real-world EV data show that the GCN model outperforms RNN-based methods and improves data interpretability.

The authors of [40] present an innovative adaptive online gated recurrent unit (GRU) approach for state of charge estimation. The GRU, a variant of deep recurrent neural networks (RNNs), effectively mitigates some issues commonly encountered in RNNs. Optimization is achieved through a robust adaptive online gradient learning algorithm, which dynamically adjusts the learning rate during the training process. Unlike conventional methods, the adaptive GRU eliminates the need for a nonlinear battery model, simplifying mathematical computations. This approach was validated using real-world data from LifePO4 lithium-ion batteries. The experimental results demonstrate that the adaptive GRU outperforms standard RNNs in terms of SoC estimation accuracy.

The Particle Swarm Optimization (PSO) algorithm is employed to fine-tune the error covariances of the extended Kalman filter (EKF) using initial segmental data derived from an LIB application as described in [41]. This tuning method, utilizing PSO, establishes the search constraints by examining the system’s error transition characteristics. Experiments were carried out to confirm the effectiveness of the proposed two-step, PSO-optimized SoC estimation technique. The findings revealed that this method achieved SoC accuracy akin to that of more sophisticated high-order models while using merely a simple first-order model. This technique optimizes the utility of model-based estimators, avoiding the necessity for expensive model enhancements.

The work described in [42] focuses on improving equivalent circuit models (ECMs) for EV applications by employing parameter identification methods based on recursive least squares (RLS) filters. Among these, the variable forgetting factor (VFFRLS) algorithms are highlighted for their effectiveness in parameter estimation for lithium-ion batteries. Traditional RLS methods use a fixed forgetting factor to improve accuracy, but struggle to adapt to real-time dynamic environmental changes. To address this, the study proposes optimizing the VFFRLS parameters using Particle Swarm Optimization (PSO), ensuring a balance between precision and stability in the estimation process.

The study presented in [43] on the prediction of the lifespan of lithium-ionn batteries using optimization techniques demonstrates the determination of optimal battery parameters and evaluates the battery performance under various conditions. The Grey Wolf Optimization (GWO) algorithm is employed to predict the battery’s lifecycle. The GWO method is notable for its simplicity and efficiency, requiring less computational time. The results obtained using the GWO algorithm are highly efficient, and the findings of multimodal functions confirm its effectiveness.

The researchers in [59] employ a variety of least squares methods to determine the capacity of lithium-ion batteries, with the goal of reducing the sum of squared errors using the teaching–learning-based optimization technique. Proportional total least squares (PTLS) and adaptive weighted total least squares (AWTLS) achieve superior precision compared to other methods examined. Starting the recursive parameters with plausible values significantly refines the estimation process, enhancing accuracy and decreasing error margins.

The study presented in [44] explores the enhancement of the noise covariance matrix in the extended Kalman filter (EKF) to refine the state of charge estimation for lithium-ion batteries used in electric vehicles. This methodology employs an adaptive Sine Cosine–Levy Flight–Quantum Particle Swarm Optimization (ASL-QPSO) algorithm. Initially, parameters of the battery’s equivalent circuit model are derived using a variable forgetting factor recursive least squares (VFFRLS) algorithm. These parameters are subsequently integrated into the EKF, enabling the ASL-QPSO algorithm to dynamically modify the local attraction factor for noise covariance matrix optimization. This improvement results in considerably more precise SoC estimations.

The work in [45] introduces an adaptive proportional–integral–derivative (PID) control-based observer for improved state estimation in lithium-ion batteries. It addresses the limitations of conventional observer methods, such as slow convergence and sensitivity to nonlinearities. The observer uses an equivalent circuit model of LIBs, offering a good balance between complexity and accuracy. The proposed algorithm is validated through analysis and error convergence testing using multiple experimental driving datasets replicating real-world conditions. The experimental results show that the proposed method achieves better accuracy, greater resilience to nonlinearities, and five times faster convergence than conventional approaches.

In [46], the AFFRLS (Adaptive Forgetting Factor Recursive Least Squares) algorithm is introduced to determine battery model parameters and project output voltage, showing its effectiveness relative to the traditional RLS (recursive least squares) technique. The tests were conducted using the dynamic profile from the Urban Dynamometer Driving Schedule (UDDS), and the projected output voltages underwent comparison.

The study presented in [47] proposes an SoC estimation technique for lithium-ion batteries utilizing an extended Kalman quantum particle filter. This approach determines its parameters through Quantum Particle Swarm Optimization (QPSO) within a second-order equivalent circuit model. Theoretical evaluations and practical experiments confirm the QPSO-enhanced Kalman quantum particle filter’s efficacy in precisely estimating SoC.

The work presented in [48] introduces an IoT-optimized battery management system (IoT-OBMS) for practical energy storage management in EVs. The system is divided into two primary modules: IoT and charging. Particle filtering estimates the state of charge and internal battery temperature, with direct cell parameter estimations integrated into SoC calculations. Data management is handled via a software defined network (SDN), while the Spider Swarm Monkey Optimization (SSMO) algorithm optimizes data routing paths. In the charging module, a Mamdani fuzzy rule-based system is combined with a Boltzmann neural network to improve battery control and management. This integration applies fuzzy logic within a deep learning framework to ensure precise and efficient battery regulation for EVs.

The authors of [49] present a time-series augmentation model (TS-DCGAN) using a deep convolutional generative adversarial network (GAN) to generate high-quality synthetic battery data. By reframing data generation as an image generation problem, TS-DCGAN leverages advances in GAN-based image synthesis. The similarity and diversity of synthetic data to real data are evaluated quantitatively and qualitatively through multiple metrics. SoC estimators trained on the generated data achieved performance comparable to those trained on real datasets. In addition, they combine synthetic data with limited real data to improve the performance of the SoC estimator. Extensive experiments confirm that TS-DCGAN effectively captures original feature distributions and temporal patterns, providing valuable support for developing accurate battery SoC estimators.

The research in [50] introduces an advanced deep learning framework integrated with a dimensionality reduction technique for estimating the state of charge in lithium-ion batteries used in electric vehicles. Using current, voltage, and temperature data from a public dataset, the input data undergo normalization for standardization. Dimensionality reduction is performed through the analysis of principal components based on the Moore–Penrose pseudoinverse (MPPI). The refined data are then processed by an optimized long short-term memory (LSTM) model to achieve highly accurate SoC predictions. The simulation results reveal that the proposed method provides fast and precise SoC estimates with minimal error, surpassing traditional approaches.

In [51], the authors present a deep long short-term memory (LSTM) neural network model designed to estimate Li-ion batteries’ state of charge across two distinct datasets. The model is trained and tested on discharge cycles involving constant and variable currents under ambient temperatures. The results demonstrate that the LSTM model successfully captures the batteries’ dynamic discharge behavior, improving SoC estimation accuracy.

In [3], the authors present an automated approach for hyperparameter optimization using a Bayesian algorithm to improve the accuracy of state of charge predictions. Beyond standard battery parameters such as voltage, current, and temperature, the model incorporates additional inputs, including vehicle velocity, road conditions, motor characteristics, and environmental factors. Feature selection is performed using the Minimum Redundancy Maximum Relevance (MRMR) algorithm. The experimental results demonstrate that the optimized model achieves significantly higher SoC estimation accuracy than conventional models.

In the study by [52], an advanced deep-learning framework is introduced for predicting the state of charge in batteries. The process starts with empirical mode decomposition (EMD) to preprocess data by reducing noise and enhancing dimensionality for better feature extraction. To strengthen the relationship between features and SoC, the Box–Cox transformation is used, thereby unlocking the predictive potential of the data. SoC predictions are modeled with a bidirectional long short-term memory (Bi-LSTM) network, which is improved with a dual-stage attention mechanism. The model’s parameters undergo fine-tuning through Bayesian optimization (BO). The experimental results indicate that this method outperforms existing prediction techniques, achieving an average absolute error of less than 0.50% and a prediction accuracy of 99.91%.

The research in [53] aims to enhance the effectiveness and performance of an electric vehicle’s energy storage system by investigating the SoC of three batteries. A battery management system with passive cell balancing was developed on Matlab Simulink. In addition, a novel protective system was proposed to avoid excessive discharges and overcharges. The results obtained demonstrated the use of the developed model for future improvements.

The authors of [54] introduce a novel battery management approach for lithium-ionn battery packs using a passive cell balancing system. The system utilizes a proportional–integral (PI) controller to correct voltage imbalances between individual cells, enhancing battery life and longevity without needing a complex active control circuit. Performance is evaluated under various conditions, focusing on capacity retention, energy efficiency, and reliability.

The primary objective in [55] is to demonstrate how to balance the voltages within each battery pack cell using various cell balancing techniques. This research focuses on the implementation of passive and active cell balancing strategies. These techniques enhance the battery’s total capacity by preventing overcharging in individual cells and ensuring uniform voltage across all cells. By preventing cell failure, these methods ensure the optimal performance of the battery pack. The emphasis on reliability should provide confidence in the efficacy of these techniques. Active cell balancing was found to be 23% faster than passive cell balancing to achieve balanced voltage levels.

Article [56] explores the use of a Proximal Policy Optimization (PPO) agent, trained through reinforcement learning (RL), to optimize the dynamic balancing of the SoC and temperature in lithium-ion battery cells integrated with an active BMS. The implementation, which includes active BMS models, battery cell simulations, and the training environment, was developed in Python. The observation space consists of the SoC and temperature of the battery cells, while the reward function encourages efficient balancing. The simulation results of the most effective models, fine-tuned using hyperparameter optimization (HPO), revealed a reduction in the variability of the balanced parameters by at least 28%, with some cases achieving reductions of up to 72%.

In [57], the research presents an innovative active equalization circuit coupled with a fresh energy redistribution technique. This bidirectional equalization configuration employs a forward transformer and a switch matrix. The new equalization strategy combines clustering analysis with a Genetic Algorithm (GA). Through simulations and experiments, the findings confirm that this method improves cell uniformity and markedly speeds up the equalization process.

In [60], a novel reduced decoupling algorithm is proposed for co-estimating the SoC and SoH of batteries, based on convex optimization. Unlike traditional methods, this approach estimates SoC directly from the battery model without relying on the coulomb counting method, allowing it to decouple capacity estimation from SoC estimation. This reduces the strong interactions typically found in conventional co-estimation techniques. The algorithm simplifies the process by solving all state variables using a single estimator, avoiding the complexity of observer networks. The result is a robust weakly interacting co-estimation algorithm for SoC and SoH.

The research in [61] introduces a square root cubature particle filter (SR-CPF) method to estimate lithium-ion batteries’ SoC, which enhances numerical stability and ensures positive definiteness of state covariance. The method leverages a fractional-order model of the battery, which better captures its capacitance characteristics, and uses Particle Swarm Optimization for parameter identification, balancing accuracy with computational efficiency. To validate the proposed SR-CPF method, dynamic cycle tests are conducted at various temperatures, comparing them with the unscented Kalman filter and the cubature Kalman filter.

The primary studies analyzed in this section propose a diverse yet complementary array of approaches designed to improve the accuracy and efficiency of state of charge estimation (see Figure 4) in lithium-ion batteries. These methodologies include model-based algorithms, machine learning techniques, and hybrid optimization strategies, each addressing the inherent complexities and challenges of SoC estimation with innovative solutions. Using advanced modeling and computational techniques, these approaches contribute to improving the precision and reliability of battery monitoring systems, which are essential to ensure optimal performance and safety in various applications, particularly electric vehicles.

The authors use cell balancing in SoC management to address battery imbalances that affect electric vehicle performance, lifespan, and safety [53,54,55,56,57]. Maintaining similar SoC levels among cells prevents overcharging, overdischarging, and reduced capacity due to inconsistency. Cell balancing maximizes battery usage, improving reliability and efficiency. They also reduce equalization time through clustering and use passive balancing and PI control for energy transfer based on SoC variations. These actions ensure cell uniformity, enhance battery performance, and ensure the safety and longevity of the system, crucial for electric vehicles.

6.2. Focus on State of Health

The SoH is a fundamental metric in the management of battery systems for EVs, representing the battery’s overall condition and performance capability relative to its original state. This section examines the primary research that explores the methods and algorithms used to estimate SoH, addressing challenges such as aging effects, environmental influences, and usage patterns. By focusing on the SoH, this study emphasizes its critical role in predicting battery lifespan, enabling proactive maintenance, and optimizing the performance of EV battery systems.

The study in [62] presents a robust framework to estimate battery health using machine learning techniques, specifically support vector regression (SVR). Recognizing the critical role of batteries in various applications, the research focuses on assessing health status to enhance performance and longevity. A diverse dataset comprising key indicators of battery health is used to train and evaluate the SVR model. The model is trained and tested for 168 cycles, demonstrating its effectiveness in accurately determining battery health status and providing valuable insights into performance over time.

The research in [63] employs data-driven methodologies to estimate battery degradation using the state of health metric. Leveraging extensive voltage, current, and temperature datasets provided by NASA’s Prognostics Center of Excellence, the data are refined using the Fourier resampling method before being processed by machine learning algorithms. The research utilizes three key techniques—long short-term memory (LSTM), DNN, and gated recurrent units (GRUs)—for SoH estimation. The performance and accuracy of these models are significantly influenced by precise hyperparameter tuning aimed at reducing computational complexity and estimation time.

The authors of [64] explore the charging and discharging processes of lithium batteries and propose a health estimation algorithm based on multiple health indicators and an enhanced radial basis function neural network (RBFNN). The NASA Li-ion battery dataset identifies key correlated parameters as indirect health factors. In addition, simulation experiments optimize the RBFNN with the whale optimization algorithm. The results indicate that the optimized RBFNN predicts the SoH better, with a 6-fold improvement in performance and better results in the degree of fit, and can anticipate the battery SoH in real time.

The authors of [65] enhance SoH estimation accuracy using hyperparameter-optimized Gaussian process regression (GPR). GPR offers flexibility as a nonparametric, stochastic model capable of providing uncertainty estimates, which account for potential variations in evaluation and forecasting. This study utilizes a NASA lithium-ion battery dataset, demonstrating the algorithm’s effectiveness in battery monitoring and prognostics. Additionally, the accuracy of SoH predictions improves with increased training data, further validating the method’s applicability for battery health forecasting.

The publication [66] presents a novel approach to modeling the SoH of cyclically operating lithium-ion batteries using Gaussian process regression. This method enables estimation of the degradation of the LIB over equivalent duty cycles under varying load patterns. Drawing on extensive research data, the model considers current SoH as a key degradation factor, improving existing modeling techniques.

The authors of [67] propose a novel data fusion model to estimate the state of health of lithium-ion batteries. The approach extracts two aging features (AFs) from partial charging curves to capture battery degradation characteristics effectively. A flexible, data-driven aging model is developed using dual Gaussian process regressions (GPRs) to address the nonlinear and non-Gaussian nature of battery aging. To improve tracking accuracy, a rejection sampling particle filter (RSPF) enhances the traditional particle filter (PF) by reducing measurement uncertainty and refining the posterior distribution. The proposed framework integrates these methods into a cohesive data fusion model, demonstrating improved precision in the estimation of SoH.

The research presented in [68] seeks to create multiple hybrid artificial intelligence (MHAI) models aimed at optimizing lithium-ion battery charging current profiles. The primary focus is on reducing both charging time and temperature increase while maintaining the battery’s SoH. Utilizing NASA’s public dataset for Li-ion 18,650 batteries, the models employ long short-term memory (LSTM) networks for predicting temperature, Random Forest (RF) algorithms for assessing maximum chargeable capacity, and coulomb counting for estimating charging duration. Optimization is realized through Particle Swarm Optimization (PSO) to find optimal current settings, and Multi-Objective Particle Swarm Optimization (MOPSO) to weigh competing objectives through a weighted-sum technique. The results validate the success of the proposed framework.

In [69], a hybrid method for optimal parameter estimation is introduced based on a combination factor-driven semi-empirical model to assess Li-ion batteries’ state of health. The model parameters are optimized using the Sine Cosine Algorithm (SCA) and benchmarked against other heuristic algorithms, including the Grey Wolf Optimizer (GWO), Differential Evolution (DE), Harris Hawks Optimization (HHO), and Genetic Algorithm (GA). The proposed approach outperforms traditional models, achieving an SoH of 0.9591, compared to 0.9467 in the conventional framework.

The study in [13] utilizes complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) to analyze the charge–discharge capacity, producing a nonlinear representation of the battery capacity. Health indicators (HIs) are derived from user behavior patterns and hourly meteorological data, which are resampled to integrate multivariate data across different time scales. The Informer network captures the global dependency between HIs and the state of health. An SoH estimation model based on the Informer network is then established. Its structure is optimized through Bayesian optimization.

An innovative approach to estimate the state of health and diagnose aging mechanisms in lithium-ion batteries using a fractional-order model (FOM) incorporating the dispersion effect is presented in [12]. A parameter identification technique is introduced, and the FOM’s predictive accuracy is benchmarked against traditional integer-order models. The robustness of the SoH estimation method is confirmed through cross-validation in batteries subjected to varied aging conditions, achieving a remaining capacity estimation error below 3.1%.

In research published in [70], a fusion prognostic framework is introduced to predict lithium-ionn battery capacity by integrating a data-driven time-series model with extracted characteristics. The method employs an autoregressive model with exogenous variables that self-adaptively update at each cycle to predict the state of health in the upcoming cycles. Given historical capacity data, the experimental results demonstrate high prediction accuracy, validating the effectiveness of the framework in forecasting battery health.

In [71], an innovative SoH estimation framework is presented that combines Mixers with a bidirectional temporal convolutional neural network (BTCN) to predict lithium-ion battery (LIB) health. This approach effectively leverages local and global feature properties for accurate SoH estimation while reducing redundancy in temporal and channel information. During data preprocessing, voltage changes are extracted at equal intervals from the constant current (CC) charging phase, simulating realistic charging scenarios. Features strongly correlated with SoH are selected using the Pearson correlation coefficient (PCC), and all data are normalized via min–max scaling to expedite convergence and reduce initial learning rate requirements.

Article [72] proposes a state of health estimation method for lithium-ion batteries that combines incremental capacity (IC) analysis with a long short-term memory (LSTM) network. An enhanced IC curve acquisition technique is introduced based on reference voltage, which retains essential IC curve features while lowering computational complexity. The variables of key health characteristics are extracted from the IC curves based on their correlation with SoH. To address the time-series characteristics and long-term dependencies of battery degradation, an LSTM network is employed to construct the SoH estimation model, ensuring accurate and efficient predictions.

In addition to SoC estimation, the analysis of state of health estimation is presented in Figure 5, providing a detailed overview of advancements in this crucial domain. The reviewed studies highlight a variety of strategies for SoH estimation, including data-driven methods, empirical models, and integration of real-time diagnostic tools. These advancements aim to improve the ability to monitor and predict battery degradation, thereby supporting the development of more efficient and sustainable battery management systems. Collectively, the insights gained from these studies lay the groundwork for further exploration and refinement of battery diagnostics and prognostics, emphasizing the critical role of accurate SoC and SoH estimation in advancing the field of battery management.

6.3. Focus on Remaining Useful Life

The RUL of a battery is a crucial indicator in battery management systems for EVs, as it estimates the remaining time or cycles for which a battery can continue to operate effectively before reaching the end of its usable life. This section explores the significance of RUL estimation in enhancing the reliability, efficiency, and cost-effectiveness of EVs. It reviews various approaches focusing on RUL, and highlights its role in enabling predictive maintenance strategies and optimizing battery utilization throughout its lifecycle.

In [73], an innovative machine learning-based method is proposed to accurately estimate the remaining useful life of lithium-ion batteries. The approach utilizes critical performance variables such as voltage, current, and temperature to develop a predictive RUL model. Harris Hawks Optimization (HHO) is employed for hyperparameter tuning to improve model accuracy. The method is validated on a lithium-ion battery dataset and compared with existing RUL prediction models, demonstrating its effectiveness.

In [74], the authors present a machine learning approach utilizing the Random Forest Regressor for predicting the RUL of batteries. Moreover, they introduce an innovative neural network architecture based on long short-term memory (LSTM) networks, appreciated for their capability to handle temporal dependencies and nonlinear behaviors in sequential datasets. These LSTM networks play a crucial role in battery capacity degradation models, supporting performance enhancement and anticipatory maintenance. A proposed user interface application would offer real-time battery capacity data.

In [75], an innovative methodology is presented for estimating the remaining useful life of batteries using measurable parameters such as discharge time and temperature. A novel feature extraction strategy is introduced to better capture battery degradation dynamics, improving the precision of RUL predictions. The study evaluates multiple machine learning algorithms, including support vector regression, Random Forest, artificial neural networks, and boosting techniques. The experimental results demonstrate that the proposed approach delivers accurate capacity estimates with minimal hyperparameter tuning, highlighting its efficiency and reliability.

The study presented in [76] introduces an AutoML-driven model designed to precisely predict the lifespan of lithium-ion batteries used in electric vehicles, surpassing conventional constant-current (CC) and constant-voltage (CV) charging methods. This method incorporates incremental capacity analysis (ICA) alongside direct analysis to formulate eight innovative multi-dimensional health indicators. K-means clustering is utilized to group analogous data points, and a change-point detection algorithm, executed via the ruptures library, pinpoints vital transitions in battery datasets. This thorough framework markedly improves the precision of battery lifespan predictions.

The research in [77] employs hyperparameter-tuned long short-term memory (LSTM) techniques to automate the analysis of incremental capacity curves (ICCs) and extract valuable battery degradation information. These techniques use statistical models to process large ICC datasets and identify degradation patterns. By training on historical data, the models can accurately predict battery degradation and estimate a battery’s remaining useful life. To further enhance RUL estimation, a hyperparameter-tuned LSTM model is proposed and compared with other well-known techniques, such as fully connected neural networks (FNNs), artificial neural networks (ANNs), and convolutional neural networks (CNNs). The results show that the robust LSTM model outperforms others regarding computational efficiency and speed.

The study in [78] focuses on battery remaining useful life prediction using long short-term memory (LSTM) and gated recurrent unit (GRU) models. A comparative analysis is conducted to evaluate both models in terms of performance and computational complexity. The results show that GRU outperforms LSTM, requiring about 20% fewer parameters. The models are tested using the NASA battery dataset, confirming GRU’s efficiency and accuracy in RUL prediction.

The study in [58] addresses the challenge of forecasting the remaining useful life of lithium-ion batteries, which is essential to improve battery reliability and longevity. The proposed approach uses an attention-based LSTM model optimized for time-series data, integrating 1D dilated convolution layers to extract feature vectors that boost prediction accuracy. A new pre-training method, Mutual Learning-based Artificial Bee Colony (ML-ABC), is introduced to initialize the model’s weights effectively. Extensive testing on NASA lithium-ion battery datasets demonstrates the high accuracy and effectiveness of this framework.

The research published in [79] explores a fusion approach for enhancing the reliability of RUL predictions by reducing uncertainty. Based on feature data, it develops separate models using the extended Kalman filter (EKF) and particle filter (PF). These results are combined using Dempster–Shafer Theory (DST) to integrate multimodel prognostics and optimize diagnostic and RUL predictive performance. The proposed method is validated through lithium-ion battery prognosis, demonstrating its effectiveness in producing more accurate and robust RUL predictions.

The study in [80] introduces a hybrid system designed to enhance prediction accuracy for lithium-ion batteries. This framework integrates an Adaptive Levy Flight (ALF)-enhanced particle filter (PF) with a long short-term memory (LSTM) network. The ALF method improves the traditional PF by addressing issues related to weight degeneracy and particle impoverishment, while the LSTM network models LIB degradation. This unified ALF-PF-LSTM method yields accurate RUL forecasts. Experiments utilizing NASA and HUST battery datasets reveal that the proposed framework notably surpasses existing PF-based algorithms in terms of prediction accuracy and robustness.

The study presented in [81] outlines a comprehensive method for forecasting the remaining useful life of lithium-ion batteries, utilizing an optimal combination strategy (OCS) in conjunction with an unscented particle filter (UPF). The process begins with the unscented Kalman filter, which generates the proposal distribution necessary for evaluating particle weights in the particle filter (PF). OCS is subsequently implemented during the resampling phase to improve particle distribution and preserve diversity. Two tests conducted on the RUL prediction of lithium-ion batteries confirm the technique’s efficacy. The findings indicate enhanced prediction accuracy and robustness over conventional methods.

Figure 6 summarizes the proposals that focus on improving the RUL, where the main technologies were machine learning, deep learning, and filter-based approaches.

7. Discussion

The study of lithium-ion battery management has seen significant advancements, particularly in estimating the SoC, SoH, and RUL. These efforts focus on enhancing predictive models’ accuracy, reliability, and adaptability under dynamic and challenging conditions.

Various methodologies have emerged for SoC estimation, ranging from advanced filtering algorithms to machine learning and optimization techniques. Kalman filter variants, such as the extended Kalman filter and adaptive unscented Kalman filter, have become foundational tools, often coupled with optimization algorithms like Particle Swarm Optimization to refine error covariance matrices. Machine learning models, including convolutional neural networks and long short-term memory networks, have improved SoC estimation by leveraging data-driven insights. Hybrid approaches that integrate traditional estimation methods with machine learning models have proven especially effective, as they combine the strengths of both paradigms.

In SoH estimation, researchers have similarly adopted data-driven and hybrid approaches to monitor and predict battery health. Supervised learning models, such as support vector regression and optimized radial basis function neural networks, have shown great potential in accurately estimating SoH. LSTM frameworks and feature extraction techniques like incremental capacity analysis capture the complexities of battery degradation dynamics. Nonparametric tools, such as Gaussian process regression, have been employed to address nonlinear and non-Gaussian aging patterns, with enhancements like rejection sampling particle filters providing additional robustness. Optimization techniques, such as the Sine Cosine Algorithm and hybrid artificial intelligence approaches, further refine model performance, ensuring high accuracy. Preprocessing and feature engineering are crucial in this context, with frameworks leveraging statistical tools like the Pearson correlation coefficient to identify relevant features. Hybrid models that fuse multiple methodologies, such as dual GPRs or autoregressive models with exogenous variables, offer comprehensive solutions for dynamic SoH updates.

Regarding RUL estimation, the temporal nature of battery aging makes time-series models indispensable. Sequential architectures like LSTMs and gated recurrent units effectively capture dependencies in performance indicators such as voltage, current, and temperature. Optimization techniques, including Harris Hawks Optimization and Adaptive Levy Flight algorithms, have been integrated into these models to enhance hyperparameter tuning and computational efficiency. Feature extraction is pivotal in improving prediction accuracy, with innovative methods identifying critical degradation indicators. While real-time applications and user interfaces have been developed to monitor battery health effectively, computational costs remain challenging, particularly for multimodel fusion approaches. Benchmarking studies provide valuable insights into the trade-offs between computational complexity and prediction accuracy, enabling researchers to tailor solutions for specific applications.

Advancements in SoC, SoH, and RUL estimation methodologies have contributed to improving battery reliability, extending lifespan, and optimizing energy management. By integrating advanced algorithms, robust optimization, and tailored feature engineering, these approaches address the growing demands of modern applications, such as electric vehicles. However, scalability and real-time applicability remain areas for further exploration, particularly in resource-constrained environments, highlighting the need for continued innovation in battery management systems.

8. Conclusions

This SMS has provided an in-depth analysis of optimization strategies for lithium-ion battery management in electric vehicles, focusing on key metrics such as state of charge, state of health, and remaining useful life. The findings underscore the pivotal role of advanced filtering algorithms, machine learning techniques, and hybrid optimization models in enhancing battery performance, reliability, and longevity.

Integrating Kalman filter variants, machine learning models (e.g., convolutional neural networks and long short-term memory networks), and hybrid frameworks has proven highly effective for SoC estimation. These approaches demonstrate significant advances in real-time adaptability, accuracy, and robustness under dynamic conditions, positioning them as foundational tools for modern battery management systems.

The reviewed studies highlight the growing reliance on data-driven techniques and hybrid models in SoH estimation. Supervised learning algorithms such as support vector regression and Gaussian process regression, coupled with advanced feature extraction and preprocessing methods, have substantially improved accuracy and computational efficiency. Innovative optimization algorithms further refine model performance and support dynamic health monitoring.

RUL prediction strategies emphasize time-series modeling through deep learning architectures such as LSTMs and GRUs, often optimized with metaheuristic algorithms. These techniques enhance prediction accuracy and computational efficiency, making them highly suitable for real-time applications in electric vehicle ecosystems. However, challenges remain in balancing computational costs and model complexity, especially for multimodel fusion frameworks.

This study highlights the transformative impact of the convergence and integration of machine learning, optimization algorithms, and advanced filtering methods on the reshaping of the battery management landscape (see Figure 3). The increasing adoption of hybrid approaches, which combine data-driven models with traditional methods, shows significant promise in addressing the multifaceted challenges associated with lithium-ion battery degradation and energy management.

Machine learning techniques, including deep learning and reinforcement learning, are playing a pivotal role in extracting complex patterns from battery performance data, enabling precise SoC and SoH estimations. Optimization algorithms, such as evolutionary strategies and metaheuristics, further complement these efforts by fine-tuning predictive models and enhancing system reliability. Meanwhile, advanced filtering methods, such as Kalman filters and particle filters, are integral in improving the robustness of real-time data processing and anomaly detection.

Hybrid methodologies that synergize these cutting-edge techniques with traditional physics-based and empirical models offer a scalable and versatile framework for tackling the inherent complexities of lithium-ion battery systems. By effectively balancing computational efficiency with predictive accuracy, these approaches address critical issues such as thermal management, capacity fading, and safety concerns. The study underscores the potential of such integrated solutions to revolutionize battery management systems, paving the way for more sustainable and reliable applications in electric vehicles, renewable energy storage, and beyond.

The primary real-world challenges identified that algorithms for BMSs face are difficulties like dynamic environmental conditions, variable battery aging patterns, and nonlinear behavior of lithium-ion batteries. For instance, traditional and machine learning-based methods must adapt to temperature and voltage fluctuations, and self-discharge during SoC estimation. Integrating algorithms such as the extended Kalman filter with advanced optimization techniques like Particle Swarm Optimization or Adaptive Levy Flight addresses uncertainties in battery performance but adds computational overhead.

Machine learning and deep learning approaches, such as CNNs, LSTMs, and GANs, require significant computational resources for training and inference. This includes powerful processors, GPUs, and sufficient memory, particularly for real-time applications. Cloud-based approaches are proposed to offload computational tasks, leveraging extensive storage and processing capabilities for adaptive algorithms like Noise Matrix Self-Adjustment Extended Kalman Filter.

Finally, the main problems in implementation are (a) computational complexity: advanced optimization methods can increase computational costs, making real-time implementation challenging. For instance, using LSTM models for RUL estimation demands hyperparameter tuning and large datasets for accuracy. (b) Real-time data handling: synchronizing real-time inputs like temperature, voltage, and current with computational models is challenging due to delays in data transmission or processing. (c) Robustness: many algorithms, such as those based on Gaussian process regression, perform well in controlled conditions but require enhancements to handle real-world uncertainties. (d) Efforts to address these challenges include hybrid frameworks that combine traditional methods with AI-driven techniques, synthetic data generation for better model training, and innovative optimization strategies.

The findings presented in this work lay a solid foundation for further exploration of hybrid strategies, advocating for a multi-disciplinary approach to advancing the state of the art in battery diagnostics, optimization, and lifecycle management.