Machine learning and neural network supported state of health simulation and forecasting model for lithium-ion battery

Abstract

As the intersection of disciplines deepens, the field of battery modeling is increasingly employing various artificial intelligence (AI) approaches to improve the efficiency of battery management and enhance the stability and reliability of battery operation. This paper reviews the value of AI methods in lithium-ion battery health management and in particular analyses the application of machine learning (ML), one of the many branches of AI, to lithium-ion battery state of health (SOH), focusing on the advantages and strengths of neural network (NN) methods in ML for lithium-ion battery SOH simulation and prediction. NN is one of the important branches of ML, in which the application of NNs such as backpropagation NN, convolutional NN, and long short-term memory NN in SOH estimation of lithium-ion batteries has received wide attention. Reports so far have shown that the utilization of NN to model the SOH of lithium-ion batteries has the advantages of high efficiency, low energy consumption, high robustness, and scalable models. In the future, NN can make a greater contribution to lithium-ion battery management by, first, utilizing more field data to play a more practical role in health feature screening and model building, and second, by enhancing the intelligent screening and combination of battery parameters to characterize the actual lithium-ion battery SOH to a greater extent. The in-depth application of NN in lithium-ion battery SOH will certainly further enhance the science, reliability, stability, and robustness of lithium-ion battery management.

Graphical abstract

Keywords

machine learning / lithium-ion battery / state of health / neural network / artificial intelligence

1 Introduction

As society continues to progress, the energy and environmental crisis is becoming a global issue that countries need to face. Today, the global energy system is changing. Fossil fuels have been depleted and renewable energy sources such as solar, biomass, wind, and hydro energy are rapidly developing [1]. As a very important clean energy storage technology, lithium-ion batteries are widely used in the field of power batteries due to their low pollution characteristics, high energy density, and long cycle life [2]. To ensure efficient and safe battery operation and to improve the lifetime of lithium-ion battery systems, predicting the remaining life and assessing the state of health (SOH) of the battery are crucial. In recent years, artificial intelligence (AI) has been widely used to evaluate SOH of lithium-ion batteries. AI is significantly important for lithium-ion battery in modeling, not only to improve modeling efficiency but also to better adapt to the needs of lithium-ion battery management (including battery load stationary energy storage batteries for vehicles) [3,4].

As a branch of AI, machine learning (ML) has made many advances in the management of lithium-ion battery SOH. There are several ML methods that have been applied to the estimation of battery conditions. These methods fall into two broad categories, those that measure directly and those that measure indirectly. Direct measurement methods have included internal resistance measurements [5], impedance measurements [6], etc. Indirect measurement methods include model-based methods such as empirical models [7], equivalent circuit models [8], and data-driven methods such as support vector machines (SVMs) [9], relevance vector machines (RVMs) [10], and artificial neural network (ANN) [11].

However, existing SOH prediction methods still face several challenges. First, although SOH estimation utilizing ML does not allow for the analysis of the complex chemical processes within the battery, it often involves a large amount of battery data and requires high-quality battery data. Second, the reliability and robustness of the algorithm needs to be improved. Therefore, there is an urgent need to improve the ML approach and enhance its usefulness in lithium-ion battery management [12,13].

Neural network (NN), as a subset of ML methods, offer a number of advantages in estimating the SOH of the battery. NN is capable of handling large amounts of data in nonlinear systems and have effective fitting capabilities. However, due to the “black box” nature of NN modeling, it is not possible to accurately capture the correlation and one-to-one correspondence between battery SOH and various input features such as voltage and temperature. Therefore, there is an urgent need to strengthen research on NN principles, especially to identify input indicators that are representative of battery SOH, and to strengthen the optimization of the “black box” structure inside NN to better simulate and predict the SOH with NN.

To this end, this paper, starting with AI, analyzes the application of ML in battery SOH, the ML methods commonly used for SOH modeling, and finally and most importantly, the applications and advantages existed for NN as an important branch of ML in battery SOH, and explores the enhancement of NN in the science, reliability, stability, and robustness of battery management in the future.

2 Preparation of ML to estimate SOH of lithium-ion battery

2.1 Relationship between AI, ML, NN, and deep learning

AI has many application prospects in the field of battery research and development. In the research and development of battery materials, advanced electrode materials are the key to the fundamental progress of the battery system. For instance, metal organic framework materials can make the battery have excellent electrochemical performance. Nowadays, AI can be used in combination with data mining, high-throughput computing technology, etc., to find suitable, high-performance electrode materials. Through the use of a large amount of data fitting, it can predict the performance of the material, such as identifying the lattice parameters. Compared with the traditional way of battery material, it is more time-saving and efficient. In addition, AI can also be used to optimize the battery system, screen the components in the battery system, and build relevant models to optimize the internal structure of the battery, such as the gap structure and the morphology of the battery materials [11,14–18].

NN is an important branch of ML which is a core technology in AI research. NN has received a lot of attention in the SOH of lithium-ion battery because of its powerful ability to nonlinearly model complex systems. Deep learning is derived from the extension of NN, giving rise to the ability to automatically extract features from large amounts of data, and is increasingly popular for estimating the SOH of battery [19]. The relationship between AI, ML, NN, and deep learning is shown in Fig.1.

2.2 ML and its main operating mechanisms

ML methods are becoming increasingly popular in the field of lithium-ion batteries in the era of Big Data.

ML can be divided into two main categories, supervised learning and unsupervised learning, depending on the learning method. Supervised learning includes decision tree, Naive Bayes, SVM, ANN, etc.; unsupervised learning includes clustering, probability estimation, dimensionality reduction algorithms, etc. [17]. The most obvious feature that distinguishes supervised learning from unsupervised learning is the presence or absence of labels. Fig.2 shows the general working principles of an ML approach for supervised/unsupervised and classification/regression methods.

ML has four main steps, data collection, feature extraction, model training, and model evaluation. To use ML to assess the SOH of lithium-ion batteries, it is necessary to determine the battery type and collect battery data. Generally, battery data contains the basic information, such as operating conditions of the battery. From the data, the features to be studied can be obtained. The data obtained and processed are divided into the training set, the validation set, and the test set. Then, model training is conducted. Finally, the effect of the model is evaluated. Fig.3 demonstrates the main operating mechanisms of ML.

2.3 SOH and data sets for its simulation and forecast validation

SOH is a fractional representation of the current SOH of a battery that is bound to experience aging or deterioration after use. To ensure the performance requirements of the battery, it is stated in IEEE standard 1188—1996 that the battery should be replaced when the capacity of the power battery drops to 80% [22]. There are numerous factors that affect the SOH of lithium-ion batteries, such as temperature, and current multiplier. The actual manifestation of this is in the variation of certain parameters within the battery, such as capacity and internal resistance. Therefore, SOH is usually defined in terms of capacity (SOH_c) and internal resistance (SOH_ir).

(1){\mathrm{S}\mathrm{O}\mathrm{H}}_{\mathrm{c}}=\frac{C_{\mathrm{c}\mathrm{u}\mathrm{r}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{t}}}{C_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}}} ,

where C_{\mathrm{c}\mathrm{u}\mathrm{r}\mathrm{r}\mathrm{e}\mathrm{n}\mathrm{t}} represents the current maximum capacity and C_{\mathrm{i}\mathrm{n}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{a}\mathrm{l}} represents the initial capacity.

(2){\mathrm{S}\mathrm{O}\mathrm{H}}_{\mathrm{i}\mathrm{r}}=\frac{R_{\mathrm{e}\mathrm{o}\mathrm{l}}-R}{R_{\mathrm{e}\mathrm{o}\mathrm{l}}-R_0},

where R_{\mathrm{e}\mathrm{o}\mathrm{l}} represents the internal resistance at the end of battery life, R_0 represents the internal resistance of a new battery at the factory, and R represents the internal resistance measured at the current state.

In ML, the most important part is the data. The quality and quantity of the data available determines how well the final model is evaluated. Once the experimenter has the data, the data needs to be manually extracted to determine the feature values that need to be fed into the model. The data are divided into a training set, a validation set, a test set to train the model, and finally the model is evaluated. Data acquisition plays a key role in the subsequent training of the model [23,24]. Battery data are obtained through tests such as normal operation or accelerated aging. Fig.4 exhibits the SOH data acquisition routes. However, for reasons such as battery performance testing, which requires hundreds or thousands of cycles and can take months, there are currently fewer battery data sets available to the public. The data sets in Tab.1 are publicly available on the Internet today.

2.4 Selection of battery health features

Numerous factors affect the SOH of a lithium-ion battery, such as temperature, charge/discharge current, charge/discharge multiplier, charge/discharge voltage, which are also used as features in estimating the health state of a lithium-ion battery. The selection principle for estimating health features in the SOH of a lithium-ion battery using ML methods is to select features that are highly correlated with the SOH of the lithium-ion battery. Battery health features can be divided into direct and indirect features. Direct features include current, voltage, temperature, etc. Indirect features evolve from direct features, such as incremental capacity (IC) curve peak, and IC curve slope. The input features of the SOH of batteries are listed in Tab.2. Fig.5 displays the framework of the battery health features.

3 Application of AI to lithium-ion battery SOH estimation

There are many ways to classify the SOH assessment methods for lithium-ion batteries, which are described in this paper in terms of experimental approach method, model-based method with adaptive models, and data-driven method based on ML.

3.1 Experimental approach

The experimental approach refers to understanding the behavior of battery aging through battery aging experiments. The analysis of the battery aging mechanism provides the basis for the model. The experimental approach can be divided into direct measurement and indirect analysis.

Direct measurement includes Coulomb counting, open circuit voltage (OCV), etc. The Coulomb counting method is associated with the continuous monitoring of input and output currents. As capacity is the integral of current over time, the change in capacity or capacity degradation of a battery can be easily measured by measuring the input and output currents [51]. The OCV method is a way of exploring the corresponding relationship between OCV and state of charge (SOC). However, accurate measurement of OCV takes a long time and requires the battery to reach a steady-state. The advantage of the experimental measurement method is the small amount of calculation. At present, the acquisition of the value of SOH is still mainly based on the experimental method, which is based on the model-based method and the data-driven method for predicting the SOH, but it has some disadvantages, such as low accuracy, need for special equipment, and unsuitable for on-site estimation [52].

3.2 Model-based method with adaptive models

In response to the shortcomings of direct measurement methods, model-based approaches have been proposed. Model-based SOH estimation is to use the battery signals measured (such as voltage, current, and temperature) to output a battery health indicator, employing a battery input−output model. The accuracy and robustness of the model mostly affect the accuracy of the method. Adaptive methods can automatically adjust their parameters according to the input signal. Adaptive models can describe the dynamic characteristics of the lithium-ion battery, while the battery is in the process of charging and discharging, including electrical and electrochemical models. Commonly used adaptive models are the Kalman filter, extended Kalman filter (EKF), and particle filter (PF). Kalman filter improves the accuracy of the model by comparing the values estimated with the true values measured in a recursive manner [53]. However, the disadvantage of Kalman filter for estimating battery SOH is that the conventional Kalman filter does not apply to nonlinear systems. Therefore, EKF, which can convert linearity to nonlinearity, is applied to estimate battery SOH [54,55]. However, EKF also has some disadvantages, such as system noise and measurement noise that must satisfy a Gaussian distribution, but the degradation of the lithium-ion battery does not meet these limits and the filtering performance suffers as a result [56]. On the contrary, PF does not restrict the system process noise model and is more widely used [57].

3.3 Data-driven method based on ML

Recently, the use of data-driven approaches has become increasingly popular in the field of SOH assessment of lithium-ion batteries. The data-driven approach does not require the researcher to have a deep understanding of the battery operating mechanism and does not require complex electrochemical modeling [58,59]. A sufficient amount of battery operation data is required to extract information from the data that can indicate the degree of battery degradation, and the data-driven approach enables the estimation of the SOH of lithium-ion batteries. Combined with the current state of research, data-driven methods are classified as Gaussian process regression (GPR), SVM, RVM, ANN, etc.

The Gaussian regression process is a statistical method, which builds the relationships between features by observing data, gives predictions using posterior probabilities, and finally expresses the uncertainty estimated in terms of confidence intervals [60]. Yang et al. [61] used SOH in terms of capacity based on GPR, took the parameters of the battery charge curve as the input and SOH as the output and trained the GPR model to estimate SOH. Jia et al. [62] extracted additional health features from lithium-ion battery charging and discharging curves and combined them with Gaussian regression methods for the short-term prediction of SOH. Lin et al. [63] proposed a hierarchical feature construction method to reduce the difficulty of feature extraction to effectively predict the SOH of batteries.

SVM is a statistical theory-based method proposed by Vanpnik in 1995, which constructs a hyperplane or a set of hyperplanes in a high- or infinite-dimensional space and is suitable for solving classification and regression problems. In the field of SOH estimation, SVM are commonly used to determine the relationship between input features and SOH. Xiong et al. [64] proposed a weighted least squares SVM to estimate the SOH of second-use lithium-ion batteries. Li et al. [65] used a modified SVM with a particle swarm algorithm to achieve the joint estimation of the SOC of batteries and SOH. In this study, the kernel function of the SVM was optimized using the particle swarm algorithm, and the rate of change of battery SOC and the rate of change of discharge voltage were used as inputs to the SVM to finally achieve online estimation of the SOH of batteries. However, this method suffers from the problem that the estimated SOH of the battery is not highly stable. To address this problem Li et al. [66] used integrated learning adaptive boosting (AdaBoost) to improve the particle swarm optimization-support vector machine model (PSO-SVM) by using multiple learners, which effectively improved the stability and accuracy of the SOH of batteries.

RVM is a method of sparse supervised ML based on an improvement in SVM, which is suitable for solving classification and regression problems [67]. Widodo et al. [68] used the entropy of the sample based on the discharge voltage as the input to RVM and SOH as the output. The results show that RVM has a good performance in the estimation of the SOH of batteries. Li et al. [10] combined average entropy and RVM for the estimation of the SOH of batteries. In this study, the optimal embedding dimension was determined using an average entropy-based approach to perform the correct time series reconstruction for the final estimation of the SOH of batteries, and the results proved to be effective for battery monitoring. In addition to the most basic RVM, researchers are improving the basic RVM. Yang et al. [69] changed the RVM to a multicore RVM and used a particle swarm optimization algorithm to optimize parameters and improve the accuracy of the estimation of the SOH of batteries. Chen et al. [70] used the bat algorithm, which has a simple structure, fewer input parameters, and better search capabilities than the traditional particle swarm algorithm, to optimize the RVM parameters and achieve an online estimation of the SOH of batteries based on dynamically integrated bat algorithm-relevance vector machine (BA-RVM). Wang et al. [71] proposed a multi-kernel RVM and whale optimization algorithm (WOA) to estimate the SOH of batteries.

ANN, or simply NN is obtained by mimicking neurons in the brain that collaborate to process and transmit information and are a popular research topic that emerged in the field of AI in the 1980s [72]. Similar to SVM, NN is commonly used in the field of estimation of the SOH of batteries to determine the relationship between input features and SOH. Xia & Abu Qahouq [73] took advantage of the ability of ANN to handle complex nonlinear problems to learn the relationship between minimum battery impedance and capacity decay. A hybrid, ANN-based method capable of obtaining the minimum complex impedance amplitude of a battery online was proposed to enable fast estimation of SOH. Kim et al. [74] combined reference-based performance testing with NN to transform the continuous SOH estimation problem of the battery into a classification problem, which not only alleviates the degradation of the battery caused by the test but also saves the cost and time of testing the battery.

Herein, NNs are described in detail in Section 4. Tab.3 focuses on the progress of research on SVM, RVM, and ANN in the assessment of the SOH of lithium-ion batteries.

There is a large number of studies published on the prediction of the SOH of lithium-ion batteries using AI methods. However, from the current literature on the estimation of the SOH of lithium-ion batteries using AI methods, several major limitations can be identified, including the lack of massive battery data, the long experimental period for obtaining battery data, the effective extraction of health features, the selection of models and the implementation of algorithms, and the application in practice. Herein, based on the application process of ML in the SOH of lithium-ion batteries, the estimation of ML in lithium-ion batteries is described, including the current publicly available data set of lithium-ion batteries, the main features extracted during the extraction of battery health features, the application of NN models in the estimation of the SOH of lithium-ion batteries, and the indicators of model evaluation. Finally, based on the limitations of the current ML methods in predicting lithium-ion batteries, an outlook is given for future research on ML-based methods for the estimation of the SOH batteries. Tab.4 exhibits the literature on the estimation methods of SOH for lithium-ion batteries.

4 NN for lithium-ion battery SOH simulation and forecasting

4.1 NN process in SOH modeling

The data-driven NN treats the battery as a “black box” and trains the mapping between input and output. There are four main steps in using NN to estimate the SOH of lithium-ion batteries. The first step is to collect the battery data, and determine the information of battery type and battery condition. The second step is to process the data obtained by filling in missing values, filtering outliers, and normalizing. In the third step, the input battery health features are determined through Spearman correlation, Pearson correlation, grey relation analysis, and other methods to determine the correlation between input and output. In the last step, the NN determined is input for training, adjusting the number of neurons, batch size, epoch, learning rate, and other parameters to optimize the model. Fig.6 shows the approximate flow of the NN to estimate the SOH of a battery, and Fig.7 shows some specific examples of applications.

4.2 Recent reports of ANN research in SOH modeling

There are many types of NNs, and the main ones applied to the health status of lithium-ion batteries include the backpropagation NN (BPNN), the convolutional NN (CNN), the recurrent NN (RNN), the long short-term memory (LSTM) NN, etc.

Wen et al. [88] extracted the voltage corresponding to the peak in the IC curve, the slope to the left of the peak, and the peak as health features, established the relationship between health features and temperature, in combination with the BPNN to predict the SOH of the lithium-ion battery at different temperatures, and the average error of the model prediction result was 1.16%. Chemali et al. [89] extracted the voltage, current, and temperature measured from the charging process as health features and inputted them into a CNN model for estimation of the SOH of batteries with an average mean error as low as 0.8%. Venugopal [90] extracted 18 health features related to battery capacity such as the average of voltage, current and temperature at a particular cycle of battery operation data, the battery capacity measured before the start of the current cycle, the difference in time in each cycle, the time spent in each discharge cycle, and other health features input to the IndRNN network to predict the SOH of batteries with a root mean square error of 1.33%. Park et al. [91] used the wavelet transform technique to pre-process the battery data, extract the nonlinear features associated with the intrinsic transformations of the voltage and temperature of the lithium-ion battery, in combination with CNN and LSTM to estimate the SOH of the lithium-ion battery. The final LSTM model estimation accuracy of 98.92% was obtained. Teng et al. [92] proposed a BPNN and an LSTM to estimate the SOH of retired batteries for more efficient use in the future. In addition to this, researchers also use various variants of NNs for prediction of the SOH of batteries. In addition to the above-mentioned variant of RNN, LSTM, Cui & Joe [93] used another variant of RNN, dynamic spatial-temporal attention-gate recurrent unit (DSTA-GRU), for prediction of the SOH of batteries, which is more effective than SVM and other models under the same conditions. Li et al. [94] used a variant of LSTM, AST-LSTM, for prediction of SOH, which is unique in that it can select both old and new data and performs well on the NASA battery data set. Yayan et al. [42] used a stacked Bi-LSTM network to estimate the SOH of lithium-ion batteries, and found that Bi-LSTM has a higher feature extraction efficiency than LSTM and is a more robust model. Several kinds of NN research for SOH modeling in recent reports are shown in Fig.8.

4.3 Hyperparameter optimization of ANN models

When NNs are computed, determining the NN hyperparameters (number of neurons, Bath size, epoch, learning rate, etc.) is essential and this process allows for more scientific training of efficient models. Methods to optimize the hyperparameters of NNs include nature-based optimization methods such as the particle swarm algorithm [98], the moth-flame optimization algorithm [99], the grasshopper algorithm [100], and the grey wolf optimization [101]. In addition to nature-based optimization methods, there are grid search methods [102], Bayesian optimization [103], etc. Based on this, the problem of NN hyperparameters needs to be considered when using ANN to estimate the SOH of batteries. At the present time, NNs are used in the estimation of the SOH of batteries, for the NN tuning hyperparameter methods such as particle swarm optimization [98], grey wolf optimization [101], and Bayesian optimization [103].

Gong et al. [104] used the particle swarm optimization algorithm to determine the number of neurons in each layer of LSTM and dropout for the LSTM used, to get the optimal hyperparameter values and improve the estimation accuracy. Ma et al. [87] proposed a grey wolf optimizer based on differential evolution to optimize the number of hidden layer neurons and the initial learning rate in the hyperparameters of LSTM networks to improve the estimation accuracy. Kong et al. [105] adopted Bayesian optimization to perform hyperparameter optimization of the proposed network with a mixture of deep CNN (DCNN) and LSTM for dropout rate, regularization factor, learning rate, batch size and number of iterations in DCNN, and for sliding window, dropout rate, learning rate, batch size and number of iterations in LSTM. The SOH of the battery is estimated using the optimized network and a result of SOH estimation root mean square error less than 0.0061 is obtained. The choice of NN optimizer needs to be considered for different problems and requirements.

4.4 Evaluation metrics and uncertainty estimation of ANN in SOH modeling

When using ANN for estimation of the SOH of batteries, the trained NN needs to be evaluated, and currently, the metrics commonly used for ANN model evaluation in estimation of the SOH of batteries are MSE (mean square error), RMSE (root mean square error), MAE (mean absolute error), MAPE (mean absolute percentage error), and R² (coefficient of determination).

(3)\mathrm{M}\mathrm{S}\mathrm{E}=\frac{1}{n} {\sum }_{i=1}^n(y_i- {\hat{y}}_i{)}^2,

(4)\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{\frac{1}{n}{\sum }_{i=1}^n(y_i-{\hat{y}}_i{)}^2},

(5)\mathrm{M}\mathrm{A}\mathrm{E}=\frac{1}{n} {\sum }_{i=1}^n|y_i-{\hat{y}}_i|,

(6)\mathrm{M}\mathrm{A}\mathrm{P}\mathrm{E}=\frac{1}{n}{\sum }_{i=1}^n\frac{|y_i-{\hat{y}}_i|}{y_i}\times 100\mathrm{\%} ,

(7)R^2= \frac{{\sum _{i=1}^n(y_i-{\hat{y}}_i)}^2}{\sum _{i=1}^n(y_i-{\overline{y}}_i)},

where n represents the number of samples, y_i represents the true value, {\hat{y}}_{\mathrm{i}} represents the mod model predicted value, and {\overline{y}}_i represents the mean of the true value.

Among them, the range of MSE, RMSE, MAE, and MAPE are [0, +∞), the smaller the value represents the higher accuracy of the model, and the value of R² is in the range of [0, 1], the closer to 1, indicating that the model effect is better. As shown in Tab.5, NN was used to estimate the SOH of lithium-ion batteries based on the same data set, and the network was evaluated using the aforementioned evaluation metrics.

The presence of incomplete data information, noise, and uncertainty in the NN structure contribute to uncertainty in the NN hyperparameters [109,110]. NN uncertainty assessment can be performed to obtain information on the confidence or uncertainty of the prediction results, and the prediction performance of NN can be fully exploited if the model assessment set uncertainty assessment is considered simultaneously in the estimation of the SOH for lithium-ion batteries.

4.5 Strengths and weaknesses of ANN in SOH modeling

ANN has an outstanding ability to perform in nonlinear mapping, self-organizing adaptability, strong parallelism and all neurons can be computed at the same time. The BPNN is a feed-forward NN with a simple structure and strong learning and generalization capabilities. It is one of the longest and most stable NN models in use today [111]. Researchers often use BPNN and its variants for scientific research [112]. The long short-term memory LSTM is a special kind of RNN, which can effectively solve the problem of RNN gradient disappearance gradient explosion, and the LSTM performs well in the field of time series prediction [113]. The LSTM can learn the long-term dependence of battery aging and reduce the noise sensitivity of the model, Tan & Zhao [114] proposed a migration learning method based on LSTM to achieve the SOH of battery prediction accurately and quickly. Deng et al. [115] used transfer learning in combination with LSTM to retrain or fine-tune the original model using transfer learning to further improve the model estimation accuracy.

However, the disadvantages of using a NN approach to the estimation of battery SOH are also apparent. First, the training of NN requires a large amount of data for support, otherwise, it will have an impact on the robustness of the NN; second, NNs, as a kind of “black-box” model, have poor interpretability, and it is common to explain how the bias and weights in NNs work; third, neural work computation is time consuming; fourth, there are numerous parameters affecting the battery SOH, and if an artificial approach is used to extract the health features, the representativeness of the health features may be affected, resulting in less accurate SOH prediction.

5 Conclusions and perspectives

Despite the problems of using ANN for SOH of batteries, there is no denying that the ANN family of methods is one of the most successful approaches to data and model-based SOH practice in recent years. The large number of ANN publications confirms its widespread use.

The core foundation of ANN is AI, and the depth of development of AI in the world today will facilitate a better improvement of ANN methods. The current public data set is not enough. In the future, if more high-quality and richer data can be obtained, it will be conducive to improve the accuracy, robustness and generalization ability of NN, for example, for battery data, more comprehensive and reliable measurements to obtain battery data of different aging states, reasonable data preprocessing, more accurate estimation of the SOH of the battery; to improve the feature mining ability, to ensure the effectiveness of the features; the hyperparameter tuning of the model, the use of a variety of hyperparameter tuning methods to obtain the optimal hyperparameters of the model, to improve the ability of the model to learn; in reality, the operating conditions of the battery is very complex, in practice, the need for a combination of methods to obtain the best model, to improve the learning ability of the model. In reality, the operating conditions of the battery are very complex, and in practice, it is necessary to combine multiple methods to obtain more accurate results. With the arrival of the big data era, the in-vehicle implementation of the estimation of the SOH of the battery will be the mainstream direction of future development.

References

[1] Tian H, Qin P, Li K. . A review of the state of health for lithium-ion batteries: Research status and suggestions. Journal of Cleaner Production, 2020, 261: 120813

[2] Sui X, He S, Vilsen S B. . A review of non-probabilistic machine learning-based state of health estimation techniques for lithium-ion battery. Applied Energy, 2021, 300: 117346

[3] XiaoHWang YXiaoD, . Distributed computing based on AI algorithms in battery early warning and SOH prediction of the intelligent connected vehicles. Neural Computing & Applications, 2020, 4: 1–12

[4] Ghalkhani M, Habibi S. Review of the Li-ion battery, thermal management, and AI-based battery management system for EV application. Energies, 2022, 16(1): 185

[5] Grandjean T, Groenewald J, McGordon A. . Accelerated internal resistance measurements of lithium-ion cells to support future end-of-life strategies for electric vehicles. Batteries, 2018, 4(4): 49

[6] Ungurean L, Cârstoiu G, Micea M V. . Battery state of health estimation: A structured review of models, methods and commercial devices. International Journal of Energy Research, 2017, 41(2): 151–181

[7] Singh P, Chen C, Tan C M. . Semi-empirical capacity fading model for SoH estimation of Li-ion batteries. Applied Sciences, 2019, 9(15): 3012

[8] Tran M K, Fowler M. A review of lithium-ion battery fault diagnostic algorithms: Current progress and future challenges. Algorithms, 2020, 13(3): 62

[9] MengJCai LLuoG, . Lithium-ion battery state of health estimation with short-term current pulse test and support vector machine. Microelectronics and Reliability, 2018, 88–90: 1216−1220

[10] Li H, Pan D, Chen C L P. Intelligent prognostics for battery health monitoring using the mean entropy and relevance vector machine. IEEE Transactions on Systems, Man, and Cybernetics. Systems, 2014, 44(7): 851–862

[11] Lipu M S H, Hannan M A, Hussain A. . A review of state of health and remaining useful life estimation methods for lithium-ion battery in electric vehicles: Challenges and recommendations. Journal of Cleaner Production, 2018, 205: 115–133

[12] Zhang S, Zhai B, Guo X. . Synchronous estimation of state of health and remaining useful lifetime for lithium-ion battery using the incremental capacity and artificial neural networks. Journal of Energy Storage, 2019, 26: 100951

[13] Ren Z, Du C. A review of machine learning state-of-charge and state-of-health estimation algorithms for lithium-ion batteries. Energy Reports, 2023, 9: 2993–3021

[14] Jiang Y, Zhao H, Yue L. . Recent advances in lithium-based batteries using metal organic frameworks as electrode materials. Electrochemistry Communications, 2021, 122: 106881

[15] Yue L, Liang J, Wu Z. . Progress and perspective of metal phosphide/carbon heterostructure anodes for rechargeable ion batteries. Journal of Materials Chemistry. A, Materials for Energy and Sustainability, 2021, 9(20): 11879–11907

[16] Liu K, Wei Z, Zhang C. . Towards long lifetime battery: AI-based manufacturing and management. IEEE/CAA Journal of Automatica Sinica, 2022, 9(7): 1139–1165

[17] Sendek A D, Ransom B, Cubuk E D. . Machine learning modeling for accelerated battery materials design in the small data regime. Advanced Energy Materials, 2022, 12(31): 2200553

[18] Lv C, Zhou X, Zhong L. . Machine learning: An advanced platform for materials development and state prediction in lithium-ion batteries. Advanced Materials, 2022, 34(25): 2101474

[19] Wang F, Zhao Z, Zhai Z. . Explainability-driven model improvement for SOH estimation of lithium-ion battery. Reliability Engineering & System Safety, 2023, 232: 109046

[20] Lombardo T, Duquesnoy M, El-Bouysidy H. . Artificial intelligence applied to battery research: Hype or reality?. Chemical Reviews, 2022, 122(12): 10899–10969

[21] Zhang Z, Li L, Li X. . State-of-health estimation for the lithium-ion battery based on gradient boosting decision tree with autonomous selection of excellent features. International Journal of Energy Research, 2022, 46(2): 1756–1765

[22] Liu K, Shang Y, Ouyang Q. . A data-driven approach with uncertainty quantification for predicting future capacities and remaining useful life of lithium-ion battery. IEEE Transactions on Industrial Electronics, 2021, 68(4): 3170–3180

[23] dos Reis G, Strange C, Yadav M. . Lithium-ion battery data and where to find it. Energy and AI, 2021, 5: 100081

[24] Nagulapati V M, Lee H, Jung D. . Capacity estimation of batteries: Influence of training dataset size and diversity on data driven prognostic models. Reliability Engineering & System Safety, 2021, 216: 108048

[25] Li Y, Stroe D I, Cheng Y. . On the feature selection for battery state of health estimation based on charging–discharging profiles. Journal of Energy Storage, 2021, 33: 102122

[26] Luo F, Huang H, Ni L. . Rapid prediction of the state of health of retired power batteries based on electrochemical impedance spectroscopy. Journal of Energy Storage, 2021, 41: 102866

[27] Bi J, Lee J C, Liu H. Performance comparison of long short-term memory and a temporal convolutional network for state of health estimation of a lithium-ion battery using its charging characteristics. Energies, 2022, 15(7): 2448

[28] SahaBGoebel K. Battery data set. NASA AMES Prognostics Data Repository. 2023-8-18, available at website of NASA

[29] Preger Y, Barkholtz H M, Fresquez A. . Degradation of commercial lithium-ion cells as a function of chemistry and cycling conditions. Journal of the Electrochemical Society, 2020, 167(12): 120532

[30] Severson K A, Attia P M, Jin N. . Data-driven prediction of battery cycle life before capacity degradation. Nature Energy, 2019, 4(5): 383–391

[31] PechtM. Battery data set. In: CALCE Battery Research Group. 2023-8-18, available at website of University of Maryland

[32] Klaas D C, Khiem T. Cyclic ageing with driving profile of a lithium-ion battery module. 2023-2-5, available at website of ResearchData

[33] SteinbußGRzepkaBBischofS, . Frequent observations from a battery system with subunits. 2023-8-18, available at website of Karlsruhe Institute of Technology

[34] DefneGHector PScottM. Fast charging tests. 2023-8-18, available at website of Datadryad

[35] ZhangS Z. Data for: A data-driven coulomb counting method for state of charge calibration and estimation of lithium-ion battery. 2023-06-07, available at website of Mendeley

[36] DamianBLeszek K. NMC cell 2600 mAh cyclic aging data. 2023-01-06, available at website of Mendeley

[37] PhilipKCarlos VMinaN, . LG 18650HG2 Li-ion battery data and example deep neural network xEV SOC estimator script. 2023-05-06, available at website of Mendeley

[38] RaiT. Path dependent battery degradation dataset part 1. 2023-8-18, available at website of University of Oxford

[39] Maleki S, Mahmoudi A, Yazdani A. Knowledge transfer-oriented deep neural network framework for estimation and forecasting the state of health of the lithium-ion batteries. Journal of Energy Storage, 2022, 53: 105183

[40] Driscoll L, de la Torre S, Gomez-Ruiz J A. Feature-based lithium-ion battery state of health estimation with artificial neural networks. Journal of Energy Storage, 2022, 50: 104584

[41] Tian J, Xiong R, Shen W. State-of-health estimation based on differential temperature for lithium-ion batteries. IEEE Transactions on Power Electronics, 2020, 35(10): 10363–10373

[42] Yayan U, Arslan A T, Yucel H. A novel method for SOH prediction of batteries based on stacked LSTM with quick charge data. Applied Artificial Intelligence, 2021, 35(6): 421–439

[43] Goh H H, Lan Z, Zhang D. . Estimation of the state of health (SOH) of batteries using discrete curvature feature extraction. Journal of Energy Storage, 2022, 50: 104646

[44] Bao Z, Jiang J, Zhu C. . A new hybrid neural network method for state-of-health estimation of lithium-ion battery. Energies, 2022, 15(12): 4399

[45] Yu Z, Zhang Y, Qi L. . SOH estimation method for lithium-ion battery based on discharge characteristics. International Journal of Electrochemical Science, 2022, 17(7): 220725

[46] Fu Y, Xu J, Shi M. . A fast impedance calculation-based battery state-of-health estimation method. IEEE Transactions on Industrial Electronics, 2022, 69(7): 7019–7028

[47] Deng Z, Hu X, Lin X. . General discharge voltage information enabled health evaluation for lithium-ion batteries. IEEE/ASME Transactions on Mechatronics, 2021, 26(3): 1295–1306

[48] Ospina Agudelo B, Zamboni W, Postiglione F. . Battery state-of-health estimation based on multiple charge and discharge features. Energy, 2023, 263: 125637

[49] Gong D, Gao Y, Kou Y. . State of health estimation for lithium-ion battery based on energy features. Energy, 2022, 257: 124812

[50] Ma B, Yu H Q, Wang W T. . State of health and remaining useful life prediction for lithium-ion batteries based on differential thermal voltammetry and a long and short memory neural network. Rare Metals, 2022, 42: 885–901

[51] Hu X, Jiang J, Cao D. . Battery health prognosis for electric vehicles using sample entropy and sparse Bayesian predictive modeling. IEEE Transactions on Industrial Electronics, 2015, 63(4): 2645–2656

[52] Liu H, Deng Z, Yang Y. . Capacity evaluation and degradation analysis of lithium-ion battery packs for on-road electric vehicles. Journal of Energy Storage, 2023, 65: 107270

[53] KimYBangH. Introduction to Kalman filter and its application. In: Govaers F, ed. Introduction and Implementations of the Kalman Filter. London: IntechOpen, 2019

[54] Sepasi S, Ghorbani R, Liaw B Y. Inline state of health estimation of lithium-ion batteries using state of charge calculation. Journal of Power Sources, 2015, 299: 246–254

[55] Park J, Lee M, Kim G. . Integrated approach based on dual extended Kalman filter and multivariate autoregressive model for predicting battery capacity using health indicator and SOC/SOH. Energies, 2020, 13(9): 2138

[56] Liu D, Yin X, Song Y. . An on-line state of health estimation of lithium-ion battery using unscented particle filter. IEEE Access: Practical Innovations, Open Solutions, 2018, 6: 40990–41001

[57] Wu T, Liu S, Wang Z. . SOC and SOH joint estimation of lithium-ion battery based on improved particle filter algorithm. Journal of Electrical Engineering & Technology, 2022, 17(1): 307–317

[58] Zhang S, Guo X, Zhang X. Modeling of back-propagation neural network based state-of-charge estimation for lithium-ion batteries with consideration of capacity attenuation. Advances in Electrical and Computer Engineering, 2019, 19(3): 3–10

[59] Pang B, Chen L, Dong Z. Data-driven degradation modeling and SOH prediction of Li-ion batteries. Energies, 2022, 15(15): 5580

[60] Zhou D, Zheng W, Chen S. . Research on state of health prediction model for lithium batteries based on actual diverse data. Energy, 2021, 230: 120851

[61] Yang D, Zhang X, Pan R. . A novel Gaussian process regression model for state-of-health estimation of lithium-ion battery using charging curve. Journal of Power Sources, 2018, 384: 387–395

[62] Jia J, Liang J, Shi Y. . SOH and RUL prediction of lithium-ion batteries based on Gaussian process regression with indirect health indicators. Energies, 2020, 13(2): 375

[63] Lin M, Wu D, Meng J. . Health prognosis for lithium-ion battery with multi-feature optimization. Energy, 2023, 264: 126307

[64] Xiong W, Mo Y, Yan C. Online state-of-health estimation for second-use lithium-ion batteries based on weighted least squares support vector machine. IEEE Access: Practical Innovations, Open Solutions, 2021, 9: 1870–1881

[65] Li R, Li W, Zhang H. . Online estimation method of lithium-ion battery health status based on PSO-SVM. Frontiers in Energy Research, 2021, 9: 693249

[66] Li R, Li W, Zhang H. State of health and charge estimation based on adaptive boosting integrated with particle swarm optimization/support vector machine (AdaBoost-PSO-SVM) model for lithium-ion batteries. International Journal of Electrochemical Science, 2022, 17(2): 220212

[67] Shah A, Shah K, Shah C. . State of charge, remaining useful life and knee point estimation based on artificial intelligence and Machine Learning in lithium-ion EV batteries: A comprehensive review. Renewable Energy Focus, 2022, 42: 146–164

[68] Widodo A, Shim M C, Caesarendra W. . Intelligent prognostics for battery health monitoring based on sample entropy. Expert Systems with Applications, 2011, 38(9): 11763–11769

[69] Yang Y, Wen J, Shi Y. . State of health prediction of lithium-ion batteries based on the discharge voltage and temperature. Electronics, 2021, 10(12): 1497

[70] Chen Z, Zhang S, Shi N. . Online state-of-health estimation of lithium-ion battery based on relevance vector machine with dynamic integration. Applied Soft Computing, 2022, 129: 109615

[71] WangSZhang XChenW, . State of health prediction based on multi-kernel relevance vector machine and whale optimization algorithm for lithium-ion battery. Transactions of the Institute of Measurement and Control, 2021, online, https://doi.org/10.1177/01423312211042009

[72] Kumar B, Khare N, Chaturvedi P K. FPGA-based design of advanced BMS implementing SoC/SoH estimators. Microelectronics and Reliability, 2018, 84: 66–74

[73] Xia Z, Abu Qahouq J A. State-of-charge balancing of lithium-ion batteries with state-of-health awareness capability. IEEE Transactions on Industry Applications, 2021, 57(1): 673–684

[74] Kim J, Chun H, Kim M. . Data-driven state of health estimation of Li-ion batteries with RPT-reduced experimental data. IEEE Access: Practical Innovations, Open Solutions, 2019, 7: 106987–106997

[75] Wang J, Deng Z, Yu T. . State of health estimation based on modified Gaussian process regression for lithium-ion batteries. Journal of Energy Storage, 2022, 51: 104512

[76] Deng Z, Hu X, Li P. . Data-driven battery state of health estimation based on random partial charging data. IEEE Transactions on Power Electronics, 2022, 37(5): 5021–5031

[77] Feng H, Shi G. SOH and RUL prediction of Li-ion batteries based on improved Gaussian process regression. Journal of Power Electronics, 2021, 21(12): 1845–1854

[78] Cai L, Lin J, Liao X. An estimation model for state of health of lithium-ion batteries using energy-based features. Journal of Energy Storage, 2022, 46: 103846

[79] Sahoo S, Hariharan K S, Agarwal S. . Transfer learning based generalized framework for state of health estimation of Li-ion cells. Scientific Reports, 2022, 12(1): 13173

[80] Ezemobi E, Silvagni M, Mozaffari A. . State of health estimation of lithium-ion batteries in electric vehicles under dynamic load conditions. Energies, 2022, 15(3): 1234

[81] Wang Y, Tian J, Sun Z. . A comprehensive review of battery modeling and state estimation approaches for advanced battery management systems. Renewable & Sustainable Energy Reviews, 2020, 131: 110015

[82] Wang Z, Feng G, Zhen D. . A review on online state of charge and state of health estimation for lithium-ion batteries in electric vehicles. Energy Reports, 2021, 7: 5141–5161

[83] Pradhan S K, Chakraborty B. Battery management strategies: An essential review for battery state of health monitoring techniques. Journal of Energy Storage, 2022, 51: 104427

[84] Wu Y, Xue Q, Shen J. . State of health estimation for lithium-ion batteries based on healthy features and long short-term memory. IEEE Access: Practical Innovations, Open Solutions, 2020, 8: 28533–28547

[85] Cheng G, Wang X, He Y. Remaining useful life and state of health prediction for lithium batteries based on empirical mode decomposition and a long and short memory neural network. Energy, 2021, 232: 121022

[86] Sun H, Sun J, Zhao K. . Data-driven ICA-Bi-LSTM-combined lithium battery SOH estimation. Mathematical Problems in Engineering, 2022, 2022: 1–8

[87] Ma Y, Shan C, Gao J. . A novel method for state of health estimation of lithium-ion batteries based on improved LSTM and health indicators extraction. Energy, 2022, 251: 123973

[88] Wen J, Chen X, Li X. . SOH prediction of lithium battery based on IC curve feature and BP neural network. Energy, 2022, 261: 125234

[89] Chemali E, Kollmeyer P J, Preindl M. . A convolutional neural network approach for estimation of Li-ion battery state of health from charge profiles. Energies, 2022, 15(3): 1185

[90] Venugopal P. State-of-health estimation of Li-ion batteries in electric vehicle using IndRNN under variable load condition. Energies, 2019, 12(22): 4338

[91] Park M S, Lee J, Kim B W. SOH estimation of Li-ion battery using discrete wavelet transform and long short-term memory neural network. Applied Sciences, 2022, 12(8): 3996

[92] Teng J H, Chen R J, Lee P T. . Accurate and efficient SOH estimation for retired batteries. Energies, 2023, 16(3): 1240

[93] Cui S, Joe I. A dynamic spatial-temporal attention-based GRU model with healthy features for state-of-health estimation of lithium-ion batteries. IEEE Access: Practical Innovations, Open Solutions, 2021, 9: 27374–27388

[94] Li P, Zhang Z, Xiong Q. . State-of-health estimation and remaining useful life prediction for the lithium-ion battery based on a variant long short-term memory neural network. Journal of Power Sources, 2020, 459: 228069

[95] Kaur K, Garg A, Cui X. . Deep learning networks for capacity estimation for monitoring SOH of Li-ion batteries for electric vehicles. International Journal of Energy Research, 2021, 45(2): 3113–3128

[96] Wei Z, Han X, Li J. State of health assessment for echelon utilization batteries based on deep neural network learning with error correction. Journal of Energy Storage, 2022, 51: 104428

[97] Bhattacharya S, Kumar Reddy Maddikunta P, Meenakshisundaram I. . Deep neural networks based approach for battery life prediction. Computers, Materials & Continua, 2021, 69(2): 2599–2615

[98] Sibalija T V. Particle swarm optimisation in designing parameters of manufacturing processes: A review (2008–2018). Applied Soft Computing, 2019, 84: 105743

[99] Mirjalili S. Moth-flame optimization algorithm: A novel nature-inspired heuristic paradigm. Knowledge-Based Systems, 2015, 89: 228–249

[100] Saremi S, Mirjalili S, Lewis A. Grasshopper optimisation algorithm: Theory and application. Advances in Engineering Software, 2017, 105: 30–47

[101] Mirjalili S, Mirjalili S M, Lewis A. Grey wolf optimizer. Advances in Engineering Software, 2014, 69: 46–61

[102] Abbasimehr H, Shabani M, Yousefi M. An optimized model using LSTM network for demand forecasting. Computers & Industrial Engineering, 2020, 143: 106435

[103] Jin X B, Zheng W Z, Kong J L. . Deep-learning forecasting method for electric power load via attention-based encoder-decoder with Bayesian optimization. Energies, 2021, 14(6): 1596

[104] Gong Y, Zhang X, Gao D. . State-of-health estimation of lithium-ion batteries based on improved long short-term memory algorithm. Journal of Energy Storage, 2022, 53: 105046

[105] Kong D, Wang S, Ping P. State-of-health estimation and remaining useful life for lithium-ion battery based on deep learning with Bayesian hyperparameter optimization. International Journal of Energy Research, 2022, 46(5): 6081–6098

[106] Guo Y, Yu P, Zhu C. . A state-of-health estimation method considering capacity recovery of lithium batteries. International Journal of Energy Research, 2022, 46(15): 23730–23745

[107] Zhang L, Ji T, Yu S. . Accurate prediction approach of SOH for lithium-ion batteries based on LSTM method. Batteries, 2023, 9(3): 177

[108] Xu H, Wu L, Xiong S. . An improved CNN-LSTM model-based state-of-health estimation approach for lithium-ion batteries. Energy, 2023, 276: 127585

[109] Gawlikowski J, Tassi C R N, Ali M, et al. A survey of uncertainty in deep neural networks. A survey of uncertainty in deep neural networks, 2023, online, https://doi.org/10.1007/s10462-023-10562-9

[110] Sun L, You F. Machine learning and data-driven techniques for the control of smart power generation systems: An uncertainty handling perspective. Engineering, 2021, 7(9): 1239–1247

[111] Zheng Y, Lv X, Qian L. . An optimal BP neural network track prediction method based on a GA–ACO hybrid algorithm. Journal of Marine Science and Engineering, 2022, 10(10): 1399

[112] Lin H, Kang L, Xie D. . Online state-of-health estimation of lithium-ion battery based on incremental capacity curve and BP neural network. Batteries, 2022, 8(4): 29

[113] Qu J, Liu F, Ma Y. . A neural-network-based method for RUL prediction and SOH monitoring of lithium-ion battery. IEEE Access: Practical Innovations, Open Solutions, 2019, 7: 87178–87191

[114] Tan Y, Zhao G. Transfer learning with long short-term memory network for state-of-health prediction of lithium-ion batteries. IEEE Transactions on Industrial Electronics, 2020, 67(10): 8723–8731

[115] Deng Z, Lin X, Cai J. . Battery health estimation with degradation pattern recognition and transfer learning. Journal of Power Sources, 2022, 525: 231027

RIGHTS & PERMISSIONS

Higher Education Press