With the rapid development of commercial lithium-ion batteries (LIBs), traditional experimental methods face challenges in handling complex data and optimizing designs. Recently, artificial intelligence (AI) technology has shown great potential in data processing, pattern recognition, and predictive analysis, providing new solutions for the research and development of LIBs. This paper reviews the application of AI in the characterization of LIB materials, including spectroscopic and imaging techniques. AI improves the accuracy and efficiency of spectroscopic analysis through feature extraction and data analysis. Combined with advanced imaging techniques, researchers can now explore the microstructure of materials with unprecedented precision and speed using AI. AI applications in image recognition, classification, and segmentation further enhance data processing efficiency and accuracy. In the future, AI will play a crucial role in the battery community through technological innovation and interdisciplinary collaboration, driving the development and application of high-performance batteries.

During the repeated charging and discharging processes of lithium-ion batteries, the uneven Li⁺ deposition leads to uncontrollable dendrite growth on the electrode surface, severely affecting the battery's safety use. A phase-field simulation is a powerful tool for describing and predicting dendrite growth, while solving partial differential equations that describe the evolution of field variables requires high computational resources. Machine learning approaches that can quickly fit the underlying laws in historical data to predict material performance have been widely used in battery materials. This paper focuses on the effects of lithium-ion transport parameters on dendrite morphology. The corresponding dendrite images were first obtained from the phase-field simulation. The machine learning models were then trained to predict the dendrite metrics (maximum dendrite height and space utilization rate). The results show that the K-nearest neighbors model can accurately capture the relationship between lithium-ion transport parameters and dendrite metrics (R² = 0.995 and 0.992). Meanwhile, the choice of machine learning models and the intervals of dendrite metrics all affect the accuracy of the results. This study can effectively save computational costs and contribute to the effective design of battery materials with dendrite suppression performance.

In recent years, artificial intelligence (AI) has made significant advancements in battery design and optimization, showing particular promise in the study of redox flow batteries (RFBs). RFBs are attractive for their low cost, scalability, long cycle life, and high safety, positioning them as critical in advancing new energy storage systems. However, traditional experimental and simulation methods have been inefficient in navigating the design space of RFBs and uncovering their complex physicochemical processes. Our research team presents a novel approach that synergizes computational simulation with data-driven AI technologies to develop a highly interpretable, multiphysics-driven model. This model is further refined through machine learning enhancements, improving the analysis and optimization of RFB designs. Our findings indicate that machine learning models, especially the Gradient Boosting model, are highly effective in predicting voltage efficiency, coulombic efficiency, and capacity. Key factors influencing these metrics were identified using SHAP analysis and interpreted through electrochemical reaction mechanisms, providing a scientific foundation for optimizing battery performance. Additionally, we developed a large language model tailored specifically for the RFB field. By employing refined prompt engineering and text analysis techniques, this model reduces errors typically known as "hallucinations", thus significantly improving the accuracy of information processing. This research underscores the transformative potential of AI-driven simulation and optimization in enhancing the design and performance of RFBs, with the ongoing evolution of computational capabilities and algorithms likely to broaden AI applications in RFBs and other energy storage technologies significantly.

The deep potential (DP) model constructs high-precision potential energy surfaces by leveraging advanced machine learning techniques to extract knowledge from vast amounts of atomic structure and energy data. This innovative approach overcomes the limitations of traditional force field methods and offers new insights into materials science. This study outlines the basic principles, development process, and application flow of the proposed DP model and its software. This study reviews the application of the DP model in electrochemical energy storage materials and highlights its advantages in revealing the microstructure and kinetic behavior of battery materials. The model accurately describes the structural evolution and free energy changes during lithium deintercalation in the cathode and anode materials. The material structure and ion transport behavior of solid electrolytes are precisely captured for solid electrolytes. For electrolytes, the model not only enhances the understanding of their dynamic structures and properties but also offers a new strategy for accurately calculating their physicochemical properties, such as their redox potential and acidity. For interfaces, the model resolves the structural evolution and properties during interface formation. These accurate material descriptions facilitate the accelerated development of energy materials. In addition, the study identifies areas for improvement in simulating battery materials using the DP model and envisions its potential applications in battery material design and optimization. The results demonstrate that the proposed DP model, as a powerful computational tool, has great potential for studying electrochemical energy storage materials. With ongoing model optimization and algorithmic innovation, the DP model is expected to play an increasingly vital role in future material design and battery technology development.

In the pursuit of carbon neutrality, energy storage technology plays an increasingly crucial role in modern society. Addressing future challenges requires innovative methods in energy storage research, given its interdisciplinary and information-intensive nature. With the rapid development of artificial intelligence technology, large language models (LLMs) have achieved significant success in various domains, including text processing, information collection and integration, and picture and video generation. Moreover, the application of LLMs has extended to natural science research, demonstrating promising potential for improving research efficiency. Thus, LLMs are expected to assist in addressing future challenges in energy storage science and technology. This paper first focuses on ChatGPT and reviews AI advancements and LLMs, analyzing their impact on civil use and scientific research, particularly focusing on domestic LLMs. Subsequently, it discusses the basic concepts and fundamentals of LLMs and their application in energy storage, covering information processing, information generation, and system integration. Detailed examples are provided to illustrate the effectiveness of these new methods. Finally, this study outlines the remaining challenges and future development directions of the interdisciplinary nature of LLMs and energy storage.

Solid-state batteries are one of the most promising next-generation energy storage technologies. Designing new battery materials with excellent comprehensive performance is expected to improve the performance of batteries. This article focuses on key scientific issues in solid-state batteries, such as migrated ion transport behavior, complex surface/interface phenomena, and microstructural dynamic evolution. Furthermore, it proposes a high-throughput material screening strategy based on multiprecision transmission. It also discusses the outstanding role of machine learning techniques in accelerating the simulation of complex physical and chemical processes and analyzing the structure-properties relationships of batteries. Benefiting from the application of multiprecision transmission concepts and machine learning techniques, the efficient acquisition of fast ion conductor materials can be achieved from various perspectives, such as direct screening, element substitution, crystal structure construction, and the design of amorphous structures. This allows for a multidimensional analysis of ionic transport performance and microscopic mechanisms, greatly enriching the range of candidate materials for electrode and solid-state electrolyte materials. In addition, the cloud toolkit designed for the development process of battery materials provides multiple features, such as data archiving, analysis, and reutilization, aimed at automating the material development process. The research paradigm and models based on big data introduced in this article, together with emerging machine learning technologies, can effectively accelerate the design and development process of new battery materials, deepen the understanding of complex physical and chemical phenomena inside batteries, and promote the design of practical new solid-state battery materials.

The rapid proliferation of electrification has driven a global surge in the demand for power and energy storage batteries. This rise has intensified concerns regarding battery safety and reliability, emphasizing the need for accurate methods for diagnosing and predicting battery aging, making this a notable area of research in the battery domain. Electrochemical impedance spectroscopy (EIS) is widely used to analyze the complex aging processes of batteries because it can effectively decouple various frequency-domain processes. The integration of machine learning methods not only facilitates the acquisition and analysis of EIS data but also offers deeper insights into battery aging and failure mechanisms. This paper reviews the latest applications of machine learning methods in EIS technique, focusing on machine learning-based acquisition and analysis of EIS data for battery life assessment and prediction. In addition, this paper explores the potential of data fusion methods for analyzing the aging behavior of batteries and predicting their lifespan, discusses the current limitations of applying machine learning to EIS research, and describes the future prospects of EIS-based battery life prediction.

Fast and accurate identification of electrochemical parameters is crucial for mechanistic modeling of lithium-ion batteries. Traditional parameter identification methods mostly use direct fitting, which makes it difficult to accurately reflect the internal state of a battery. To solve this problem, in this study, a modified simplified impedance model mapped with an electrochemical model was constructed based on the Faradaic process of electrochemical reactions, the non-Faradaic process of the double-layer capacitance dispersion effect, and the conduction process in the solid and liquid phases. The model was applied to a 37 Ah ternary battery. The model's inputs are the three-electrode electrochemical impedance spectra (EIS) under different states of charge (SOC), unlike the P2D model, which are used as inputs to the three-electrode EIS under the different SOCs. The corresponding working conditions of the electrochemical parameters were obtained by fitting the EIS to achieve accurate parameter identification of the battery model. By fitting the impedance spectra, 16 highly sensitive electrochemical parameters were identified: 7 for the positive and 9 for the negative electrodes. Further, we compared the performance of 66 metaheuristic algorithms in lithium-ion battery electrochemical parameter identification and analyzed them multidimensionally in terms of identification accuracy, computational efficiency, and robustness. The results showed that the adaptive differential evolutionary algorithm has the best overall effect in parameter identification, with its average absolute percentage error of less than 3% and the number of non-repeating function calculations of less than 35000, indicating that it achieves maximum accuracy with low arithmetic and that the proposed identification method not only better reflects the physical significance of the parameters, but it also provides strong support for simplified computation and on-line identification of the electrochemical model.

Lithium-ion batteries, a crucial component of electric vehicles, directly influence vehicles' range, safety performance, and overall operational efficiency. Capacity, a key indicator of battery health, presents challenges in estimation and acquisition under real vehicle conditions. In response, this study introduced a method that integrates ampere-hour integration with the equivalent circuit model, treating capacity as a parameter to be identified through the particle swarm optimization algorithm. Building on this, the study focused on the peculiarities of lithium iron phosphate batteries and proposed a specific optimization approach for slow charging conditions to address issues of poor model voltage fitting during the capacity identification process, which is primarily implemented by deleting the voltage segment at the end of charging and employing a two-dimensional loss function. The method was precision-validated across two electric vehicle models equipped with lithium iron phosphate batteries. Given the absence of direct capacity labels in real vehicle data, the study first calculated capacity based on static charging segments as label values. Due to insufficient label quantities, the nominal capacity under small mileage was also used as a label for accuracy validation. The results show that the mean absolute percentage error for the two vehicle models was 2.33% and 3.38%, respectively. These results demonstrated the method's high accuracy and applicability, offering a new perspective and approach for estimating real-vehicle battery capacity.

An accurate assessment of the state-of-health (SOH) of lithium-ion batteries is critical to ensure the safe and stable operation of electric vehicles. However, traditional SOH estimation methods face challenges in effectively extracting health features (HFs) and relying on large amounts of HF test data. This paper proposes an SOH estimation method for lithium-ion batteries based on multifeature analysis and the long short-term memory (LSTM)-eXtreme gradient boosting (XGBoost) model. First, to accurately describe the aging mechanism of a battery, six HFs were extracted from the battery charging data in three categories: time, energy, and IC. Considering a lot of redundant information exists among the same types of HFs, a feature-processing method based on double correlation was presented to screen out the combined HFs that can accurately characterize the trend of battery degradation. Second, to solve the problem that the traditional SOH estimation model requires a large amount of HF test data, an SOH estimation model based on the LSTM-XGBoost was proposed. In this model, the LSTM algorithm was used to predict the HF data of the number of battery remaining cycles. At the same time, to solve the problem of low computational efficiency in HF prediction using the LSTM model, the LSTM-XGBoost model was developed to estimate the SOH of batteries. The results show that the proposed method can accurately estimate the SOH of lithium batteries under different test data amounts, and the root-mean-square error is kept within 1%, which has high estimation accuracy and robustness.

Accurate state of charge (SOC) estimation is the key to ensure the safe and stable operation of power batteries. However, in practical applications, the environment factors such as temperature change and noise interference make the accurate estimation of SOC difficult. In order to solve this problem, this paper proposes a joint estimation method of multi-timescales of the SOC of lithium ion batteries in wide temperature range based on the multi-new interest adaptive robust untrace Kalman filter (MIARUKF) algorithm, which integrates multi-new interest theory, adaptive filtering and robust algorithm based on the UKF algorithm. The proposed algorithm uses the multi-interest vector to correct the state estimates and timely update the noise covariance, so as to improve the estimation accuracy of SOC and improve the robustness of the algorithm by introducing the H filtering algorithm. Meanwhile, in order to reduce the computational burden of BMS, the UKF algorithm was used to estimate the model parameters online on the macroscopic time scale, and the MIARUKF algorithm was used to estimate the battery SOC on the microscopic time scale. Finally, the estimation results of battery SOC were compared and analyzed under different initial SOC initial values and temperature conditions, and the maximum and average absolute errors of the proposed method were 1.05% and 0.42%, respectively, indicating the high accuracy and good robustness.

To enhance the accuracy of state-of-health (SOH) estimation for lithium-ion batteries, this study proposes a convolutional neural network (CNN)-transformer fusion model based on polynomial feature expansion. The model leverages the powerful local feature extraction capability of CNNs and the sequence processing ability of transformers. Key health factors, highly correlated with battery capacity, such as peak values of incremental capacity curves, corresponding voltages, areas, and charging time, were extracted and expanded using polynomial features. This expansion enhances the model's ability to handle nonlinearities in the input features. Principal component analysis was employed to reduce the dimensionality of the feature space, which aided in capturing adequate data information and reduced training time. The effectiveness and accuracy of the proposed fusion algorithm were validated using open-source datasets from the National Aeronautics and Space Administration (NASA) and the University of Maryland. Comparative analyses of SOH estimation were conducted for the CNN-transformer model with and without polynomial features and for single-model algorithms. The results indicate that the SOH estimation accuracy of the proposed model, compared to the CNN-transformer model without polynomial features, improved by 38.71%, 50.28%, 4.71%, and 17.58% for datasets B0005, B0006, B0007, and B0018, respectively.

Predicting the temperature of the thermal process in lithium batteries has significant implications for the lifespan management and safety of these batteries. Thermal management in battery management systems typically depends on accurate thermal process models. However, the thermal process in lithium batteries is complex, constituting a strongly nonlinear distributed parameter system with the characteristics of parameter spatiotemporal coupling, time variation, and strong nonlinearity. Conventional methods struggle to accurately model this thermal process.To address these issues, this study proposes a three-stage lithium battery thermal process modeling method based on weighted uniform manifold approximation and projection (WUMAP) and an improved broad learning system (BLS). First, a WUMAP dimensionality reduction algorithm was introduced to solve the nonlinear dimensionality reduction problem while preserving global and local data information. Then, a BLS model was introduced to predict the temporal data obtained from the dimensionality reduction. Finally, a mixed multikernel BLS (MKBLS) model optimized by particle swarm optimization was used to reconstruct the spatiotemporal temperature data.To validate the effectiveness of the model, modeling experiments were conducted on the thermal process of a flat plate 32 Ah Li(Ni_0.5Co_0.2Mn_0.3)O₂ ternary lithium battery. The experimental results show that the final model, compared to its previous version, increased R² by 0.0546 and decreased MAE and RMSE by 0.0082 and 0.0092, respectively. When compared with several other models, the final model demonstrated a lower relative error (ARE) of less than 0.035 and better performance on all error indicators, confirming its high prediction accuracy.

The capacity of lithium-ion batteries degrades after numerous charge-discharge cycles, posing a risk to energy storage systems. This study proposes a hybrid model for the early lifetime prediction of lithium-ion batteries considering their energy and temperature features. The proposed model addresses the insufficient analysis of temperature and energy features and the lack of research inon the significance of features extracted via deep learning. First, to fully mine the effective information from the temperature data, the voltage, current, and temperature data were employed to indirectly compute and extract the capacity, energy, and temperature signal energy curves of the battery. The first 100 cycles were selected to construct the corresponding two-dimensional features. Second, to address the inability of convolutional neural networks (CNNs) to filter extracted feature maps, a feature extraction architecture based on CNNs and a convolutional block attention mechanism was proposed, The attention mechanism identifies the importance of each feature map, facilitating mapping from features to early lifetime predictions. Experiments conducted on the MIT lithium-ion battery degradation dataset validated the effectiveness of the proposed features and methods. The results indicated that the proposed hybrid model outperformed the basic CNN, achieving superior prediction performance with an average root mean square error of 97.43. Furthermore, a series of experiments using different features as inputs revealed that the proposed temperature signal energy features provide superior prediction performance, whereas the multi-feature fusion technology can achieve better prediction performance. Finally, in scenarios with limited period data application, the model requires at least 70 cycles to maintain good prediction performance and high stability.

Accurate state estimation is crucial for ensuring the safe and reliable operation of lithium-ion batteries. However, achieving simultaneous online estimation of multiple parameters across a broad temperature range is challenging due to strong nonlinearity and multi-parameter coupling. To address this, an electro-thermal coupling model was developed, and battery parameters were identified online using the extended Kalman filter algorithm, the model's accuracy was verified by voltage and temperature simulations. To enhance the utilization of historical data and address the limitations of unscented Kalman filter algorithm (UKF), the multi-innovation theory (MI) was introduced to improve the UKF. The root mean square error of state of charge (SOC) estimation with the improved algorithm in the non-voltage platform areas is reduced to under 1.2%, representing more than 30% improvement. A switching algorithm was also designed, integrating the ampere-hour method to overcome the MIUKF algorithm's limitation of not correcting SOC estimation errors through voltage feedback in the voltage platform areas of lithium iron phosphate batteries. This approach enabled accurate full-range SOC estimation under complex working conditions and at various temperatures. The combined algorithm's accuracy was validated across different initial SOC values, with a root mean square error of less than 3%, providing a reliable SOC value for state of power (SOP) estimation. Finally, under multiple constraint method, a temperature constraint was introduced in SOP estimation. The results show that at high temperature,temperature limitation plays a critical role in preventing excessive temperature rise, thereby reducing potential safety hazards.

The evaluation of the state of health (SOH) for real-vehicle batteries is challenging owing to poor data quality, inconsistent operating conditions, and limited data utilization. This paper presents a multisource feature extraction and SOH estimation model specifically designed for step-rate charging conditions. First, charging segments are obtained through data cleaning, segmenting, and filling processes. Next, capacity is calculated using data from various current stages, achieving a raw data utilization rate of 96.9%. Compared to methods that calculate capacity within a restricted state of charge (SOC) range, this approach reduces error by over 48.1%. Finally, health factors are extracted based on current operating conditions and historical data accumulation. For current operating condition feature values, dual screening is performed using grey correlation analysis and random forest importance analysis to manage interference. For historical cumulative feature values, Spearman correlation analysis and Kernel Principal Component Analysis (KPCA) are employed to reduce information redundancy. Finally, an attention mechanism and Runge-Kutta optimizer (RUN) are integrated into the Gated Recurrent Unit (GRU) network model. The performance of this optimized model is then compared with five existing models using an actual vehicle operation dataset. The experimental results demonstrate that the optimized model achieves superior estimation accuracy, with an error margin of no more than 0.0086, regardless of whether the test samples include single-stage or multi-stage currents. Additionally, the model shows excellent error convergence as the number of charging cycles increases and effectively predicts the trend of SOH fluctuations.

Accurate estimation of the state of health (SOH) in lithium-ion batteries (LIB) is crucial for the safe and stable operation of energy storage systems. Current data-driven approaches often rely on manual feature extraction or fall short in single-scale feature representation. To address these issues, this paper introduces a novel SOH estimation model that leverages automatic feature extraction and channel attention mechanisms for multi-timescale modeling. The approach begins with inputting charging process data into multiple parallel dilation convolution modules (DCM), which automatically extract features across various time scales, creating a rich and comprehensive feature representation. These multi-scale features are then integrated and processed by a gated recurrent unit (GRU) to capture long-term dependencies in the time series data. Furthermore, the model incorporates the efficient channel attention (ECA) mechanism, which dynamically adjusts the importance of historical information and emphasizes critical features. The proposed method's effectiveness is validated through experiments two public datasets, showcasing a significant improvement over common deep learning models. Results demonstrate that the model proposed in this study exhibits high precision in SOH estimation and robust transferability. The model achieves low Root Mean Square Errors (RMSE) of 0.0110 and 0.0095 on these datasets, respectively, and maintains an RMSE of only 0.0092 in cross-dataset transfer experiments. These findings underscore the efficacy and adaptability of the proposed model in executing SOH predictions across different datasets.

The accurate and efficient assessment of the state of charge (SOC) of lithium-ion batteries is critical to ensuring the satisfactory performance and safety of electric vehicles and energy storage devices. The equivalent circuit model is considered to be effective for describing complex reaction processes inside Li-ion batteries. To use the equivalent circuit model to address the difficult trade-off between accuracy and complexity in SOC estimation, we use the first-order RC model as the foundation of this study. In order to improve the performance of the model over the SOC interval, the RC model is optimized according to electrochemical principles. By adding an improved error term for the solid-phase diffusion process inside the reactive battery to the open-circuit voltage (OCV) module of the first-order RC model, we reduce the computational complexity. By adding a modified error term that reflects the solid-phase diffusion process inside the cell to the first-order RC model of the OCV module, we also reduce the error between the equivalent circuit model and the more accurate mechanism model while ensuring that the computational complexity remains low. Then, based on the multiplicity test and pulse test data, a particle swarm algorithm is used to reduce the complexity and improve the accuracy of parameter identification through parameter decoupling. At the same time, a polynomial method is used to fit the OCV-SOC curve based on the OCV data from a small-multiplication test. Subsequently, based on the model parameter identification results, SOC estimation research is carried out. To address the insufficient accuracy of conventional Kalman filtering, a weighted sliding window is used with traceless Kalman filtering to improve the accuracy and robustness of the SOC estimation, and the Kalman filtering algorithm is verified based on the UDDS and DST dynamic test data. The final estimation results show excellent accuracy and robustness, unlike the traditional method. The results quickly converge to the accurate value when the initial SOC has a large deviation.

In the use of lithium-ion batteries, real-world operating conditions often restrict the availability of extensive, fully labeled data. This limitation presents a significant challenge in accurately predicting battery capacity degradation. To address this issue, this study introduces an approach that combines a capacity degradation curve augmentation algorithm with traditional neural network techniques to predict the trajectory of battery capacity degradation. First, a small set of fully labeled battery capacity degradation data, polynomial functions, and the Monte Carlo method are used to generate virtual capacity degradation curves. These augmented curves are subsequently filtered using KL divergence and Euclidean distance metrics. Next, four widely used neural network models—Multilayer Perceptron, Convolutional Neural Network, Gated Recurrent Unit, and Long Short-Term Memory—are developed to map the virtual degradation curve data to the actual battery capacity. Finally, the models are pretrained with a small amount of fully labeled battery data using the virtual capacity degradation curve data as input and the actual capacity as output. They are then fine-tuned with early degradation data from the target battery to predict its capacity degradation trajectory. The effectiveness of the proposed method is validated using data from 77 lithium-ion batteries subjected to various discharge schemes. The results demonstrate that, even with just three fully labeled battery capacity degradation datasets, the prediction performance of the proposed method remains robust regardless of the type of neural network used. All four neural networks effectively predict the capacity degradation trajectories of the remaining batteries, achieving a mean MAPE and RMSE of less than 2.3% and 31 mAh, respectively.

In the context of actively pursuing the "dual carbon" goals, the advancement of hydrogen energy has experienced unprecedented opportunities. As a vital component of the green transportation revolution, fuel cell vehicles play a key role in carbon reduction and achieving carbon neutrality. Such vehicles have also become a focus of research on new energy vehicles. Improving the intelligence of fuel cell vehicles and continuously optimizing their performance through machine learning algorithms have become important ways to enhance the efficiency of such vehicles. Proton exchange membrane fuel cells, one of the core technologies of fuel cells, continue to encounter significant challenges in commercialization and market adoption due to durability issues. Accurately predicting the aging performance of fuel cells is highly challenging due to their nonlinearity and dynamic characteristics, coupled with their ever-changing operating conditions. This paper proposes a novel fuel cell performance prediction model based on conditional convolutional neural networks. The proposed model combines linear trends and nonlinear dynamic feature predictions to iteratively forecast the aging performance of fuel cells using a recursive method. The experimental results confirmed the high accuracy of this model in long-term performance prediction, instilling confidence in its practical implications for enhancing the reliability and efficiency of fuel cell systems.

Predicting the remaining useful life (RUL) of lithium-ion batteries is crucial for maintaining their safe and reliable performance. This paper addresses several challenges such as nonlinear capacity changes due to recovery and external disturbances, sparse degradation data, and the difficulty in acquiring complete lifecycle data. The study introduces a novel approach that leverages variational mode decomposition and permutation entropy to denoise and reconstruct degradation data for similar batteries, normalizing it for effective model training. Additionally, a rolling prediction strategy is employed, using sliding windows to partition and concatenate training data. It then trains a Transformer network that is proficient at capturing global dependency relationships. For predictions, the initial 10% of the target battery data is iteratively used for rolling predictions. This approach's effectiveness is initially validated using battery capacity datasets from the Center for Advanced Life Cycle Engineering (CALCE) at the University of Maryland, using a leave-one-out evaluation. Experimental results demonstrate that the prediction method yields strong performance metrics, with an average relative error of 2.21% for RUL across four batteries. Additional validation is conducted with battery data from the National Aeronautics and Space Administration (NASA), specifically battery B0005, to assess model generalization. Battery B0005 achieves an RUL relative error of 2.34%, further confirming the method's effectiveness.

Energy batteries with high energy density have attracted much attention as an important way to achieve China's carbon peak and carbon neutrality goals; however, the existing technologies can no longer meet the urgent need for efficient, safe, and stable operation of such energy batteries. Digital twin technology, with its characteristics of real-time sensing, efficient simulation, accurate prediction, and rapid optimization of complex systems, is expected to be an effective means of addressing these challenges. This paper analyzed the constituent elements of digital twin technology for energy batteries at the cell level. Furthermore, it describes the roles of three key technologies in the battery digital twin: implanted sensing technology, highly efficient and fidelity physical models, and machine learning algorithms. The current status of implanted sensing technology in battery temperature, strain, pressure, and gas sensing was introduced. It reviews related research on coupled models that describe the behavior of different physical fields of batteries. In addition, it discusses the application of machine learning algorithms in battery digital twin technology and recent advances in physics-based machine learning algorithms. Finally, the main challenges and development trends of battery digital twin technology are summarized, and suggestions for overcoming these challenges in future research are proposed. This research work can provide deep insights into battery digital twin technology and contribute to its further popularization and application in academic research and industrial applications.

The performance degradation of lithium-ion batteries encompasses detailed processes at multiple scales, ranging from materials, interfaces, and porous electrodes to devices and involving complex chemical and electrochemical reactions. In recent years, informatics has emerged as a vibrant new field for the study of the degradation of lithium-ion batteries, and it intersects data science with battery materials science. These developments promise to accelerate the resolution of complex issues such as battery state modeling, performance management, and lifetime prediction. Various machine learning (ML) methods serve as crucial tools for modeling complex data, discovering patterns, and informing applications. This study focuses on modeling the remaining useful life (RUL) of lithium-ion batteries by reviewing the latest advancements for predicting the RUL of lithium-ion batteries based on ML. It presents breakthroughs by ML methods for data-driven battery management, predictive modeling, and enhanced battery performance and lifespan. Despite numerous achievements in this field, several key challenges hinder its further development. Finally, this study summarizes the primary problem within the field at present, with the intent to provide a comprehensive perspective on ML-based battery RUL prediction and an outlook for future trends and directions.

Lithium metal anodes in lithium-ion batteries (LIBs) have garnered significant research interest due to their exceptionally high theoretical specific capacity. However, their high reactivity can trigger a series of complex degradation reactions of electrolyte components, ultimately forming a solid electrolyte interface (SEI) on the electrode surfaces. This SEI passivation layer suppresses the continuous loss of electrolytes but can significantly affect the cycling performance of the LIBs. Consequently, understanding the formation and growth mechanisms of SEI at the atomic/molecular level has become a major research focus in recent years. This study summarizes the latest research progress on the structure, composition, and growth process of SEI using various theoretical approaches. Particularly, it introduces some successful examples of classical molecular dynamics, reactive force field molecular dynamics, first-principles molecular dynamics, machine learning force field molecular dynamics, and kinetic Monte Carlo methods in SEI modeling. In addition, this study discusses the limitations of current theoretical approaches in simulating the formation and growth mechanisms of SEI. Finally, this study proposes that combining machine learning methods and kinetic Monte Carlo approaches to enhance the simulations of the formation and growth processes of SEI over long-time domains.

With the continuous adaptation of large-scale energy storage systems and electrical equipment, the energy storage capacity of batteries and supercapacitors is facing increasing demands and challenges. The long research and development cycle and inefficient material screening are two major problems in the development of energy storage materials (ESM). The application of artificial intelligence (AI) to ESM is a new solution to this problem. Furthermore, machine learning (ML), an aspect of AI, has proven to be a powerful tool to for gaining insights from data. ML can mine data behind valuable information and implicit correlation, help to reveal the key structure of ESM or properties and performance relationships, and significantly accelerate ESM development and screening. At the same time, AI for energy storage system design and operation provides advanced prediction tools. Therefore, future integration research on AI and energy storage technology will be an emerging field worthy of attention. This study first provides an overview of the key technical framework for AI, including the ML process, supervised and unsupervised learning, and explainable AI. Then, the latest research progress of AI in ESM design, identification, screening, and performance prediction is summarized. A list of databases commonly used in ML in energy storage materials research is also provided. The contribution of this fusion technology to smart grid optimization and renewable energy integration and management is briefly analyzed. Finally, this study looks at the opportunities and challenges facing the integration of AI and energy storage technology, as well as the research directions to focus on in the future.

In the era of artificial intelligence (AI) in science, the battery design automation (BDA) intelligent R&D platform has revolutionized battery R&D by integrating advanced AI technologies. The BDA platform covers five key aspects of battery R&D: Read, Design, Make, Test, and Analysis. It uses advanced algorithms, such as machine learning, multi-scale modeling, and pre-training models, combined with software engineering to develop user-friendly tools for accelerating the complete battery R&D cycle from theoretical design to experimental validation. Through synthesis and preparation, characterization testing, performance optimization, and automated experimental design, the BDA platform enhances R&D efficiency and improves the accuracy and reliability of battery design, which results in battery technology with higher energy density, longer cycle life, and lower costs.

As the demand for electric vehicles and energy storage continues to grow, the production of high-performance batteries is increasing rapidly, leading to more stringent requirements for advanced manufacturing processes. Smart manufacturing is vital in this context, leveraging automation technology, information systems, computational simulation, and artificial intelligence to enhance production efficiency and flexibility, minimize human errors, elucidate material mechanisms, and improve product performance. Battery data analysis platforms are critical for smart manufacturing, helping researchers predict and optimize various battery properties through advanced data analysis techniques. However, current battery big data analysis platforms encounter several challenges, including issues with data integration, limited analysis tools, poor user-friendliness, and insufficient scalability. To address these challenges, we have developed efficient algorithms using machine learning techniques for common tasks in battery data analysis, including feature analysis, battery consistency evaluation, state of health estimation, and remaining useful life prediction. Furthermore, we have created a specialized big data analysis platform for batteries named BatAi Craft. This platform uses comprehensive data analysis and performance prediction to help researchers intuitively understand complex battery datasets and uncover underlying patterns and relationships. By deploying these advanced algorithms, BatAi Craft improves the efficiency of battery analysis and intelligent management, driving the digital and smart advancement of the battery industry.

In recent years, the surge in research investment within the battery field has presented researchers with challenges of information overload and knowledge gaps. This study examines the Retrieval-Augmented Generation (RAG) architecture of large language models in the battery domain, offering a review of contemporary research and future prospects. We describe the working principles of the RAG architecture, affirm its reliability in specialized domains, and discuss its applications across three key areas as follows: battery material design, battery cell design and manufacturing, and battery management systems for e-mobility and electric grids. In the section on battery material design, the study highlights the hallucination-free generation capabilities of RAG in data extraction, research protocol design, and multimodal data querying. The section on battery cell design and manufacturing elucidates RAG's role in enhancing research-driven battery cell design and bridging the gap between industry and academia, thereby aiding industrial control processes. The discussion on battery management systems for e-mobility and electric grids underscores RAG's contribution to cross-domain knowledge integration and system-level operation and maintenance support. The study concludes by considering the application of multimodal RAG technology in battery research and anticipates further expansion of RAG applications in this field.

This bimonthly review provides a comprehensive overview of the recent research on lithium batteries. A total of 6113 online papers, published between June 1, 2024 and July 31, 2024, were examined using the Web of Science database. The BERTopic topic model was employed to analyze abstracts and map research topics related to lithium batteries. From these, 100 papers were selected for detailed review to cover various aspects of lithium battery development. Cathode materials such as LiNi_0.5Mn_1.5O₄ and lithium-rich oxides have been enhanced by doping, surface coating, and microstructural modification. The cycling performance of silicon-based anodes has improved through structural design innovations. Great efforts have focused on the interface design of lithium metal anodes. Studies on solid-state electrolytes have focused on the structural design and performance of polymer- and halide-based systems. Conversely, liquid electrolytes have seen improvements through the optimization of solvents and lithium salts for different battery applications, along with the incorporation of novel functional additives. In the context of solid-state batteries, extensive investigations have been conducted on the modification, surface coating, cathode synthesis methods, interface construction, three-dimensional structural design of lithium metal anodes, and the use of multilayer electrolytes. The structural design of cathodes and liquid electrolytes for lithium-sulfur batteries has proven beneficial in extending their cycling life. In addition, there are a few studies related to dry electrode technology, binders, and separators for cells. New current collectors and electrolytes have been explored for lithium-oxygen batteries. Moreover, numerous studies address ion transport and degradation mechanisms in electrodes, lithium deposition morphology, and solid-electrolyte interphase structural evolution in electrolytes. Other topics include thermal runaway analysis in full batteries, theoretical simulations of solvent effects on cathode-electrolyte interphase components, and efforts to reduce battery costs and optimize manufacturing processes.

In recent years, LiFePO₄ (LFP) has garnered significant attention because of its low cost, high safety, and long cycle life. However, conventional LFP energy storage devices typically have a cycle life of approximately 2000 cycles at a rate of 0.1—2 C. To further develop high-power and long-life LFP energy storage devices, pouch-type energy storage devices based on different anode materials (hard carbon/soft carbon/graphite) with a capacity of 9 Ah were designed. Examining the cycling performance at a high rate of 4 C revealed that after 4000 cycles, the capacity retentions of energy storage devices using hard and soft carbons as anodes were 83.0% and 78.9%, respectively, outperforming those using graphite, which had a retention of 51.6%. Analytical techniques such as XRD and EIS analyses, incremental capacity analysis, and different voltage analyses revealed that the primary causes of capacity fading include increased contact resistance, SEI film resistance, and lithium-ion consumption in the anode due to structural changes. Further investigation into the poor cycle life of graphite as an anode material revealed that graphite exhibits a lower Coulomb efficiency, higher operating temperature, greater displacement, and a decrease in peak intensity in the dQ/dV peak during cycling. The interlayer spacing of graphite was 0.335 nm, which is smaller than that of soft carbon (0.360 nm) and hard carbon (0.395 nm). Graphite is more susceptible to structural changes and volume expansion during repeated lithium removal and insertion processes. Thus, graphite exhibits higher contact resistance and SEI film resistance for lithium ions and poorer diffusion kinetics. Compared with graphite, hard and soft carbons exhibit longer cycle lives as anode-active materials in high-power LFP energy storage devices.

Degradation of battery packs in electric vehicles is inevitable during their operational lifetime, making the estimated remaining useful life (RUL) a critical indicator of battery performance. This study proposes an optimized long short-term memory (LSTM) network-based RUL prediction model for EV lithium-ion batteries using the coyote optimization algorithm (COA). First, this study examines the capacity degradation characteristics of lithium-ion batteries. Indirect health indicators were extracted from the charge and discharge curves of the batteries, including constant current charging and discharging intervals and constant voltage holding time intervals. The correlations of these indicators were examined using the Pearson approach. Then, variational mode decomposition (VMD) was applied to decompose the health indicators into modal components. The LSTM model was used to predict the RUL of the battery pack. To address the issue of inaccurate LSTM model parameters affecting RUL prediction accuracy, COA was used to optimize these parameters and enhance the predictive capabilities of the model. The proposed method was validated using publicly available datasets from the NASA research center and compared with LSTM, VMD-LSTM, Gaussian process regression, and backpropagation neural network models. The experimental results indicate that the proposed approach can achieve RUL prediction errors of within 2%, demonstrating its ability to accurately predict RUL.

Gravity energy storage offers numerous advantages, including high safety, low cost, long lifespan, no attenuation of stored energy, short construction period, and environmental friendliness. In particular, slope gravity energy storage leverages the natural incline of mountains to reduce construction costs and minimize the use of flat land resources. The proposed technology is a promising approach for large-scale, long-term energy storage. However, slope gravity energy storage systems exhibit high coupling between mechanical and electrical dynamics, and the existing simulation model makes it challenging to fully describe their dynamic characteristics. To address this issue, this study introduces a multi-software collaborative modeling approach for the mechanical and electrical co-simulation of slope gravity energy storage systems. First, a mechanical system frame model, including the slope, track, load car, and mass block, was constructed using Solidworks, a three-dimensional mechanical design software. The model from Solidworks is then imported into the multi-body dynamics simulation software Adams, where the connection between the load car and the chain transmission mechanism is established, resulting in a complete mechanical system model. Finally, the mechanical model from Adams was imported into Simulink, where it was integrated with the electrical model to obtain a mechanical-electrical joint simulation model. The proposed multi-software co-simulation model was compared with an independent Simulink simulation model under normal and abnormal power grid conditions. The results indicate that the multi-software co-simulation model provides a more comprehensive description of the dynamic characteristics of slope gravity energy storage systems. Consequently, this model can better support the power characteristic analysis, safety evaluation, and mechanical parameter optimization design of slope gravity energy storage.

Transient overvoltages in power systems can cause voltage fluctuations and affect the safe and stable operation of electrochemical energy storage stations during grid integration. Research on fire incidents involving electrochemical energy storage systems has revealed inadequate transient overvoltage protection capabilities for lithium-ion battery modules. This study proposes a transient overvoltage protection circuit design for energy storage lithium-ion battery modules by examining the performance of passive overvoltage surge protection devices. Furthermore, this study involves device selection, coordination design, and experimental testing to develop a multi-level step-down and cascaded voltage regulation self-breaking function for a battery module overvoltage protection circuit. Results indicate that for a 12 V lithium-ion battery module, the designed protection circuit gradually limits the voltage and dissipates energy under common- and differential-mode inputs with peak values of 500 V, 1 kV, 2 kV, and 4 kV and a waveform of 1.2/50 μs impulse overvoltage. The circuit suppressed the voltage amplitude at the input battery module ports up to 15.2—26.4 V. Steady-state DC overvoltage testing showed that the module voltage stabilized to 12 V when the port voltage was below 26.5 V. When the port voltage exceeded 26.5 V, the cut-off function isolated the lithium-ion battery module. This study contributes to the realization of effective transient overvoltage suppression and steady-state DC overvoltage breaking and stabilization functions for lithium-ion battery modules.

This study examines the physical model of a turbine generator set, focusing on a detailed explanation of the working principles of the turbine expander speed control system, synchronous generator, and excitation system. A mathematical model of the turbine generator set was developed using MATLAB/Simulink. This model incorporates a turbine expander and its speed control system, a synchronous generator and excitation system, and a power-load model. Conventional PID simulations were performed under various operating conditions for turbine generator sets using the established model. Subsequently, the control approach of the system was optimized, and fuzzy PID control was used to simulate the no-load start and load switching. Based on the simulation results, the control performances of conventional and fuzzy PID were compared. The simulation results indicated that the fuzzy PID controller exhibited a shorter starting time and a smaller overshoot than the conventional PID controller, resulting in a faster and more stable response process for the generator set.

Accurate estimation of the state of charge (SOC) of Lithium-ion (Li-ion) batteries is crucial for the reliable and safe operation of new energy electric vehicles, and a high-precision battery model is fundamental for SOC estimation. Traditional resistance-capacitance (RC) models of Li-ion batteries face several challenges, including the complex structure of the high-order models, the low approximation of the low-order models, and difficulties in accurately estimating the SOC due to abrupt changes in the Li-ion states. This study proposes a fractional variable resistance-capacitance model of Li-ion based on fractional calculus theory. Using the Akachi Information criterion, the optimal order of the fractional RC model was determined for various SOC levels, and a time-varying RC battery model was developed, which was adapted to different SOCs. A strong tracking fractional extended Kalman filter algorithm was constructed by incorporating an attenuation factor, and the SOC of Li-ion was estimated to address the influence of historical data on the current estimated value. The model's performance was validated using the fractional extended Kalman filter algorithm under three different working conditions, including urban road cycle conditions. The findings indicate that the average absolute error (AAE) of the model decreased from 0.0197 to 0.0160 V under pulse discharge conditions, the voltage errors are all less than 50 mV, and the prediction accuracy is relatively improved by 18.8%. The AAE and root mean square error are reduced by the improved approach, which not only verifies the effectiveness of the proposed approach but also provides a new insight for improving the accuracy of SOC estimation and the computation efficiency of Li-ion batteries.