Battery health estimation method and system based on Gaussian process regression

Abstract

The invention discloses a battery health estimation method and system based on Gaussian process regression. The method comprises the following steps: firstly, collecting cyclic charging and discharging data in the cyclic charging and discharging process of the lithium battery; features related to electrode degradation loss in the degradation process of the lithium ion battery are extracted in the feature extraction part, mechanism features in the degradation process of the battery are substituted into the estimation process, and the prediction accuracy is improved; in the aspect of model construction, Gaussian process regression and lithium battery mechanism characteristics are combined to construct a series of Gaussian kernel functions integrated with a real degradation mechanism, and the Gaussian kernel functions mainly comprise consideration of discharge current, external temperature, discharge depth and discharge time. According to the method, the disadvantages that a traditional battery health estimation algorithm is low in estimation precision under small sample data, weak in generalization performance, incapable of giving uncertainty estimation and the like are overcome, and the accuracy of battery health estimation is effectively improved.

Description

Technical Field

[0001] The present invention relates to a battery health estimation method in the technical field of battery management, and in particular to a battery health estimation method and system based on Gaussian process regression. Background Art

[0002] Lithium-ion batteries have excellent characteristics such as high energy density, long service life, and green environmental protection. They are widely used in electric vehicles and various electronic devices. However, there are still many challenges in the field of battery management to ensure the safe and stable operation of batteries. The most critical issue is how to accurately estimate the health status of the battery. The health status of the battery is affected by many factors, such as the external operating temperature, the size of the charge and discharge current, the depth of discharge, and other factors, as well as the active lithium content inside the battery, the degradation of the electrolyte interface membrane, and other factors. Therefore, aging batteries should be replaced when the battery health reaches a certain level to ensure the normal operation of the equipment.

[0003] At present, the estimation methods for battery health are mainly divided into two categories, including model-based methods and data-driven methods. Model-based methods require a large number of test experiments, and different types of batteries need to be modeled separately, which requires high experimental equipment and professional knowledge; while data-driven methods can effectively avoid these problems. Data-driven models approach target values ​​by training data, and their generalization and practicality are better than model-based prediction methods. However, there is still a lot of room for improvement in traditional data-driven methods. First, it ignores the importance of effective feature extraction of battery data, and often only uses raw data, but a series of chemical reactions that occur during the charging and discharging process of the battery can provide more available features; secondly, traditional methods often need to establish separate models for the same type of batteries at different temperatures, discharge currents, and discharge depths, which requires high computing power and time; on the other hand, such methods usually only give estimates of battery health, but cannot give the uncertainty interval of the estimate, and the prediction effect for a longer range is poor, and the accuracy will be greatly reduced. Summary of the invention

[0004] In view of this, in order to solve the problems existing in the technical background, the purpose of the present invention is to propose a battery health estimation method and system based on Gaussian process regression. The present invention can accurately predict the health status of the battery under degradation conditions such as different temperatures, discharge currents, and discharge depths, while optimizing the feature extraction of the data, considering the features related to electrode loss, and introducing the mechanism characteristics of the battery into the design of the Gaussian process kernel function, which has long-term battery health prediction capabilities and improves the accuracy, generalization, and robustness of battery health estimation.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is as follows:

[0006] 1. A battery health estimation method based on improved Gaussian process regression

[0007] Step 1: Collect the cycle charge and discharge data of multiple groups of lithium batteries under different conditions;

[0008] Step 2: After preprocessing and feature extraction of multiple groups of lithium battery cycle charge and discharge data under different conditions, the corresponding cycle charge and discharge feature information is obtained, and the training data set is composed of all the cycle charge and discharge feature information and the corresponding battery health data;

[0009] Step 3: After training the improved Gaussian process regression model using the training data set, a lithium battery health estimation model is obtained;

[0010] Step 4: Preprocess the cycle charge and discharge data to be predicted and then input it into the lithium battery health estimation model. The model predicts and outputs the battery health prediction estimate.

[0011] The step 2 is specifically as follows:

[0012] Step 2.1: Among multiple groups of cyclic charge and discharge data of lithium batteries under different conditions, if the charge and discharge data in each charge and discharge cycle is incomplete, the charge and discharge data in the charge and discharge cycle is removed to obtain multiple groups of charge and discharge data in the screened charge and discharge cycles;

[0013] Step 2.2: Normalizing the charge and discharge data of the multiple groups of screened charge and discharge cycles respectively to obtain corresponding normalized charge and discharge data;

[0014] Step 2.3: The normalized discharge current, discharge depth, discharge time, and external temperature in all normalized charge and discharge data constitute the time series data D1, and the battery electrode degradation-related characteristic data D2 is calculated according to the capacity-voltage curve. All the cyclic charge and discharge characteristic information is composed of the time series data D1 and the battery electrode degradation-related characteristic data D2. Finally, all the cyclic charge and discharge characteristic information and the corresponding battery health data constitute the training data set.

[0015] In step 1, the charge and discharge data include discharge current, external temperature, voltage range, discharge depth, discharge time and current remaining capacity of the battery.

[0016] The battery electrode degradation related characteristic data D2 includes lithium inventory loss LLI, positive electrode active material loss LAM PE and negative electrode active material loss LAM NE .

[0017] In step 3, the kernel function of the improved Gaussian process regression model is formed by fusing five Gaussian process regression kernel functions.

[0018] The five Gaussian process regression kernel functions are respectively a kernel function based on electrode degradation characteristic data, a kernel function based on temperature, a kernel function based on discharge time, a kernel function based on discharge current, and a kernel function based on discharge depth.

[0019] In step 3, the fusion of the five Gaussian process regression kernel functions is specifically to multiply the five Gaussian process regression kernel functions.

[0020] 2. A Battery Health Estimation System Based on Gaussian Process Regression

[0021] Cyclic charge and discharge data acquisition module, used to obtain the cyclic charge and discharge data of lithium batteries under different conditions;

[0022] A training data set construction module is used to preprocess and extract features of the cycle charge and discharge data of lithium batteries under different conditions, thereby generating a training data set;

[0023] A lithium battery health estimation module is used to train the improved Gaussian process regression model using a training data set, generate a lithium battery health estimation model and store it;

[0024] The module for estimating the data to be predicted is used to preprocess and extract features of the cyclic charge and discharge data to be predicted, and then use the lithium battery health estimation model to perform health estimation to obtain the battery health prediction estimate.

[0025] 3. A computer device

[0026] The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the steps of the battery health estimation method based on Gaussian process regression when executing the computer program.

[0027] 4. A computer-readable storage medium

[0028] The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the battery health estimation method based on Gaussian process regression are implemented.

[0029] 5. A computer program product

[0030] The computer program product comprises a computer program / instruction, which, when executed by a processor, implements the steps of the battery health estimation method based on Gaussian process regression.

[0031] The beneficial effects of the present invention are:

[0032] (1) The present invention enriches feature information by extracting features related to battery electrode degradation.

[0033] (2) The present invention uses an improved Gaussian process regression model to not only provide an estimate of battery health, but also provide an uncertainty range for the estimate.

[0034] (3) The present invention incorporates the mechanism characteristics of battery degradation into the model by redesigning the kernel function of the Gaussian process, greatly improving the accuracy and robustness of battery health estimation under different operating temperatures, discharge currents, and discharge depths. At the same time, it can provide long-term battery health estimation. Long-term means that the battery health status can be predicted after 2000 cycles. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a system flow chart of the present invention.

[0036] Figure 2 It is a structural diagram of the improved Gaussian process regression model of the present invention.

[0037] Figure 3 This is a comparison chart of the experimental results of the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0039] The present invention also proposes a battery health estimation method based on Gaussian process regression, such as Figure 1 As shown, the following steps are included:

[0040] Step 1: Collect cyclic charge and discharge data of multiple groups of lithium batteries under different conditions and obtain the capacity-voltage curve of the lithium battery; wherein the charge and discharge data include discharge current, external temperature, voltage range, discharge depth, discharge time and current remaining battery capacity.

[0041] Step 2: After preprocessing and feature extraction of multiple groups of lithium battery cycle charge and discharge data under different conditions, the corresponding cycle charge and discharge feature information is obtained. The training data set is composed of all cycle charge and discharge feature information and corresponding battery health data. Among them, the battery health of lithium batteries is generally represented by SOH, and the calculation formula of SOH is as follows:

[0042]

[0043] Among them, C nominalis the nominal capacity value of the battery, that is, the maximum capacity of the battery when it leaves the factory, and C is the maximum capacity of the battery at the current moment; when SOH is lower than 80%, the lithium battery can be considered to be failed, that is, scrapped.

[0044] Step 2 is as follows:

[0045] Step 2.1: Among the cyclic charge and discharge data of multiple groups of lithium batteries under different conditions, if the charge and discharge data in each charge and discharge cycle is incomplete, the charge and discharge data in the charge and discharge cycle are removed, that is, the discharge current, external temperature, voltage range, discharge depth, discharge time and current battery remaining capacity in each charge and discharge cycle need to be present, and multiple groups of screened charge and discharge data in the charge and discharge cycle are obtained;

[0046] Step 2.2: Normalize the charge and discharge data in the multiple groups of screened charge and discharge cycles respectively. Specifically, normalize the discharge current, external temperature, voltage range, discharge depth, discharge time and current battery remaining capacity in the charge and discharge data respectively to obtain the corresponding normalized charge and discharge data; wherein, the normalization formula is:

[0047]

[0048] Among them, x is the original data, x max 、x min are the maximum and minimum values ​​in the original data, respectively, norm is the data after normalization;

[0049] Step 2.3: The normalized discharge current I from all normalized charge and discharge data d , Depth of Discharge DOD d , discharge time Temp d , external temperature t d Construct each time series data D1 that does not need to be calculated, specifically D 14×n = {I d ,DOD d ,Temp d ,t d}, the feature dimension is 4×n, and the battery electrode degradation related feature data D2 is calculated based on the capacity-voltage curve of the lithium battery, specifically D 23×n ={LLI,LAM PE ,LAM NE}, the feature dimension is 3×n; among them, LLI, LAM PE ,LAM NEThey represent lithium inventory loss, positive electrode active material loss, and negative electrode active material loss respectively. All the cycle charge and discharge feature information is composed of time series data D1 and battery electrode degradation related feature data D2. Finally, all the cycle charge and discharge feature information and the corresponding battery health data N constitute the training data set.

[0050] Among them, LLI, LAM PE ,LAM NE The calculation formula is as follows:

[0051]

[0052] in, Indicates the lithium state of the positive and negative electrodes in the fully charged state. Indicates the maximum capacity of the positive and negative electrodes during the degradation process; Indicates the positive and negative electrode capacities of the battery before degradation. And the corresponding battery half-electrode potential curve can be solved by solving the following optimization problem Thus, we can obtain LLI, LAM PE ,LAM NE :

[0053]

[0054] Among them, U p ,U n represents the half-cell potential function of the positive and negative electrodes, Q i Indicates the current capacity of the battery, I indicates the current current, and R ohm Represents the internal resistance of the battery, V(Q i ) represents the battery voltage corresponding to the current capacity; n represents the number of sampling points on the voltage-capacity curve when solving the optimization problem, ‖‖ 2 represents the square of the matrix 1-norm, It means that the objective function value is minimized by adjusting the vector θ.

[0055] Step 3: After training the improved Gaussian process regression model using the training data set, a lithium battery health estimation model, i.e., a lithium battery health prediction model, is obtained;

[0056] like Figure 2 As shown, the kernel function of the improved Gaussian process regression model is formed by fusing five Gaussian process regression kernel functions, specifically, the five Gaussian process regression kernel functions are multiplied together; the time series data D1 and the electrode degradation related feature data D2 are used as the input of the improved Gaussian process regression model.

[0057] Gaussian process regression can be expressed as follows:

[0058] For all x=[x1,x2,…,x3], f(x)=[f(x1),f(x2),…,f(x3)] obeys the multivariate Gaussian distribution, then f is called a Gaussian process, which can be expressed as:

[0059] f(x)~N(m(x),k(x,x′))

[0060] Here m(x) represents the mean function, which returns the mean of each dimension; k(x,x′) is the covariance function (also called the kernel function) which returns the covariance matrix between the dimensions of two vectors. A Gaussian process is uniquely defined by a mean function and a covariance function, and a finite-dimensional subset of a Gaussian process obeys a multivariate Gaussian distribution.

[0061] In the regression problem, the prior distribution of the output y can be expressed as:

[0062]

[0063] Among them, N() represents normal distribution, σ n is the noise term, I n is an n×n identity matrix. If the same Gaussian distribution exists between the test set x′ and the training set x, the predicted output y will follow the joint prior distribution, and the training output y is:

[0064]

[0065] where k(x,x), k(x′,x′), and k(x,x′) represent the covariance matrices of the training set, the test set, and the inputs of the training set and the test set, respectively.

[0066] In order to ensure the performance of the Gaussian process regression model, some hyperparameters θ in the covariance function need to be optimized through n points during the training process. An effective optimization solution is to minimize the negative log marginal likelihood L(θ) as:

[0067]

[0068] Among them, λ(θ) represents the covariance matrix, k(θ) represents the kernel function matrix, det represents the determinant of the matrix, and T is the matrix transpose symbol.

[0069] After optimizing the hyperparameters of the Gaussian process regression model, the predicted output y' can be obtained on the dataset x by calculating the corresponding conditional distribution p(y'|x',x,y):

[0070]

[0071] in, represents the corresponding predicted means; cov(y′) represents the variance matrix reflecting the uncertainty range of these predictions.

[0072] Among them, the customized Gaussian process regression model kernel function is as follows:

[0073] 1) Kernel function k based on electrode degradation feature data loss Using the radial basis kernel function, it is expressed as follows:

[0074]

[0075] x loss ={LLI,LAM PE ,LAM NE}

[0076] x′ loss ={LLI′,LAM PE ′,LAM NE ′}

[0077] Among them, σ loss represents a hyperparameter of the electrode degradation loss, x loss Represents the input kernel function k loss The training data is obtained from D2; x′ loss represents the input kernel function k loss The test data is obtained from the test dataset. Represents the square of the matrix two-norm.

[0078] 2) Temperature-based kernel function k T , considering that the side reactions in the battery will decrease exponentially with the decrease of temperature, that is, considering the Arrhenius equation in the battery degradation process, it is expressed as follows:

[0079]

[0080] Among them, σ T is the first hyperparameter associated with temperature; Temp d Represents the input kernel function k T Training data, Temp d′ represents the input kernel function k T of test data.

[0081] 3) Kernel function k based on discharge time t is the improved radial basis kernel function (ARD-RBF), which is expressed as follows:

[0082]

[0083] Among them, M represents the length of the time series, y drepresents the input kernel function k t The training data, t d′ represents the input kernel function k t of test data.

[0084] 4) Kernel function k based on discharge current I , considering that the effect of discharge rate on battery degradation generally presents a polynomial trend, it is expressed as follows:

[0085]

[0086] Among them, c I ,d I are the first and second hyperparameters associated with the discharge current; I d represents the input kernel function k I The training data, I d′ represents the input kernel function k I of test data.

[0087] 5) Kernel function k based on discharge depth dod , considering the exponential effect of discharge depth on battery health degradation, it is expressed as follows:

[0088]

[0089] Among them, c dod ,d dod The third and fourth hyperparameters are related to the discharge current; DOD d represents the input kernel function k dod training data, DOD d′ represents the input kernel function k dod of test data.

[0090] Finally, the improved Gaussian process regression model kernel function is obtained by multiplication:

[0091] K=σ f ·k loss ·k T ·k t ·k I ·k dod

[0092] In order to verify the prediction effect of the present invention on the missing data set, the collected data set is divided into a training set of 30% and a test set of 70%; after the training data is input into the model, the optimal hyperparameter value is obtained by maximizing the logarithmic marginal likelihood; the evaluation index of the model is the mean maximum absolute error MAE, which can be expressed as:

[0093]

[0094] Where m represents the number of samples, y i represents the i-th true value, Represents the i-th prediction value. The smaller the MAE, the higher the prediction accuracy. || represents the absolute value of the constant.

[0095] Step 4: After preprocessing and feature extraction of the cycle charge and discharge data to be predicted, it is input into the lithium battery health estimation model, and the model predicts and outputs the battery health prediction estimate.

[0096] After preprocessing the charge and discharge data to be predicted, the corresponding time series data D1 that does not require calculation and the battery electrode degradation-related characteristic data D2 are obtained.

[0097] In the embodiment, multiple A123 LFP / Graphite lithium battery data sets are collected, the cycle test temperature distribution is between 25 and 45°C, the discharge rate is 2C to 0.5C, and the discharge depth is 60% to 100%; the data sets are subjected to feature extraction, model building and battery health estimation according to steps 1-4, wherein the first 30% of the data is used as a training set and the last 70% of the data is used as a test set. Comparative experiments are conducted using traditional long short-term memory neural networks, Bayesian neural networks, and Gaussian process regression models without custom kernel functions. The final estimated value of the lithium battery battery health and the actual battery health value are shown in Figure 2. Figure 3 (a) Figure 3 (b) Figure 3 (c) and Figure 3 As shown in (d), the average maximum absolute error is shown in Table 1:

[0098] Table 1 is a comparison table of the mean maximum absolute error of different methods

[0099] MAE Improved Gaussian mixture process regression model ( Figure 3 (a) 0.0043 Unmodified Gaussian process regression model ( Figure 3 (b) 0.0133 Bayesian Neural Networks ( Figure 3 (c) 0.2852 Long Short-Term Memory Networks (LSTM) Figure 3 (d) 0.1034

[0100] By comparing the method proposed in this invention with the other two traditional prediction methods, it can be seen that the Gaussian process regression model can significantly improve the prediction accuracy; at the same time, after using a variety of customized kernel functions that combine the physical properties of lithium batteries, the prediction accuracy is further improved, and its average maximum absolute error is much smaller than the results obtained by the traditional data-driven method. Figure 3 It can be seen that in the case of long-term prediction, the method proposed by the present invention is closer to the actual value.

[0101] The present invention also proposes a battery health estimation system based on Gaussian process regression, comprising:

[0102] Cyclic charge and discharge data acquisition module, used to obtain the cyclic charge and discharge data of lithium batteries under different conditions;

[0103] A training data set construction module is used to preprocess and extract features of the cycle charge and discharge data of lithium batteries under different conditions, thereby generating a training data set;

[0104] A lithium battery health estimation module is used to train the improved Gaussian process regression model using a training data set, generate a lithium battery health estimation model and store it;

[0105] The module for estimating the data to be predicted is used to preprocess and extract features of the cyclic charge and discharge data to be predicted, and then use the lithium battery health estimation model to perform health estimation to obtain the battery health prediction estimate.

[0106] The present invention also proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a battery health estimation method based on Gaussian process regression are implemented.

[0107] The present invention also proposes a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of a battery health estimation method based on Gaussian process regression are implemented.

[0108] Finally, it should be noted that the above embodiments and explanations are only used to illustrate the technical solution of the present invention rather than to limit it. Those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope disclosed in the technical solution of the present invention, which should be included in the scope of protection of the claims of the present invention.

Claims

1. A battery health estimation method based on Gaussian process regression, characterized in that: The following steps are involved: Step 1: Collect the cycle charge and discharge data of multiple groups of lithium batteries under different conditions; Step 2: After preprocessing and feature extraction of multiple groups of lithium battery cycle charge and discharge data under different conditions, the corresponding cycle charge and discharge feature information is obtained, and the training data set is composed of all the cycle charge and discharge feature information and the corresponding battery health data; Step 3: After training the improved Gaussian process regression model using the training data set, a lithium battery health estimation model is obtained; Step 4: Preprocess the cycle charge and discharge data to be predicted and then input it into the lithium battery health estimation model. The model predicts and outputs the battery health prediction estimate.

2. The method for estimating battery health based on Gaussian process regression according to claim 1, characterized in that: The step 2 is specifically as follows: Step 2.1: Among multiple groups of cyclic charge and discharge data of lithium batteries under different conditions, if the charge and discharge data in each charge and discharge cycle is incomplete, the charge and discharge data in the charge and discharge cycle is removed to obtain multiple groups of charge and discharge data in the screened charge and discharge cycles; Step 2.2: Normalizing the charge and discharge data of the multiple groups of screened charge and discharge cycles respectively to obtain corresponding normalized charge and discharge data; Step 2.3: The normalized discharge current, discharge depth, discharge time, and external temperature in all normalized charge and discharge data constitute the time series data D1, and the battery electrode degradation-related characteristic data D2 is calculated according to the capacity-voltage curve. All the cyclic charge and discharge characteristic information is composed of the time series data D1 and the battery electrode degradation-related characteristic data D2. Finally, all the cyclic charge and discharge characteristic information and the corresponding battery health data constitute the training data set.

3. The battery health estimation method based on Gaussian process regression according to claim 1, characterized in that: In step 1, the charge and discharge data include discharge current, external temperature, voltage range, discharge depth, discharge time and current remaining capacity of the battery.

4. The battery health estimation method based on Gaussian process regression according to claim 2, characterized in that: The battery electrode degradation related characteristic data D2 includes lithium inventory loss LLI, positive electrode active material loss LAM PE and negative electrode active material loss LAM NE .

5. The battery health estimation method based on Gaussian process regression according to claim 1, characterized in that: In step 3, the kernel function of the improved Gaussian process regression model is formed by fusing five Gaussian process regression kernel functions.

6. The battery health estimation method based on Gaussian process regression according to claim 5, characterized in that: The five Gaussian process regression kernel functions are respectively a kernel function based on electrode degradation characteristic data, a kernel function based on temperature, a kernel function based on discharge time, a kernel function based on discharge current, and a kernel function based on discharge depth.

7. The method for battery health estimation based on Gaussian process regression according to claim 5, characterized in that: In step 3, the fusion of the five Gaussian process regression kernel functions is specifically to multiply the five Gaussian process regression kernel functions.

8. A battery health estimation system based on Gaussian process regression, characterized in that: include: Cyclic charge and discharge data acquisition module, used to obtain the cyclic charge and discharge data of lithium batteries under different conditions; A training data set construction module is used to preprocess and extract features of the cycle charge and discharge data of lithium batteries under different conditions, thereby generating a training data set; A lithium battery health estimation module is used to train the improved Gaussian process regression model using a training data set, generate a lithium battery health estimation model and store it; The module for estimating the data to be predicted is used to preprocess and extract features of the cyclic charge and discharge data to be predicted, and then use the lithium battery health estimation model to perform health estimation to obtain the battery health prediction estimate.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images