Battery health state estimation method based on battery local fragment data extraction

Abstract

The invention provides a battery health state estimation method based on battery local fragment data extraction, and belongs to the technical field of lithium ion batteries, and the method comprises the steps: S1, constructing an estimation model; s2, in m charge-discharge cycles, determining an interval section for feature extraction on a battery charging curve corresponding to each charge-discharge cycle, selecting n voltage sections at the same interval V, and obtaining an input feature matrix; s3, obtaining an optimal interval Vp, and obtaining an optimal input feature matrix; s4, repeating the step S2 and the step S3 for a set number of times, selecting voltage segments at different intervals V in each time of repetition, and obtaining an optimal input characteristic matrix corresponding to the current time of repetition; and S5, inputting each optimal input feature matrix into the estimation model to determine a final optimal interval, and repeating the step S2 at the final optimal interval to obtain a final input feature matrix. The complexity and accuracy of the estimation model can be balanced.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion batteries, and in particular relates to a battery health state estimation method based on battery local segment data extraction. Background Art

[0002] Lithium-ion batteries (LIBs) have the advantages of high energy and power density, low self-discharge rate, and long life, and have been widely used in consumer electronics, energy storage systems, electric vehicles, and other fields. In order to improve the reliability of battery-powered systems, it is crucial for the battery management system (BMS) to estimate and monitor the battery status in real time. Throughout the life cycle of LIBs, due to the interaction of complex physical and chemical reactions, the battery capacity gradually degrades with use and aging, and the battery capacity decays to 80% is considered the end of life. As one of the key parts of BMS, accurate battery capacity estimation is a critical and challenging issue.

[0003] In recent years, research on battery capacity estimation has been mainly divided into physical model-based and data-driven methods. Physical model-based methods link the battery's electrochemical model (EM), equivalent circuit model (ECM)) with the battery capacity and use some adaptive techniques known in control theory, such as Kalman filtering (KF) or particle filtering (PF) to predict battery capacity degradation. However, in order to achieve accurate capacity estimation, a suitable and universal aging model needs to be established based on physical model methods, which is not an easy task in some cases.

[0004] The data-driven method refers to learning the mapping relationship between features and SOH through neural networks and support vector machines. This method has wide applicability. With the rapid development of artificial intelligence, data-driven methods have become a research hotspot for SOH estimation. However, feature selection is the key to using this method. Therefore, in the extraction process, we need to select features that are easy to obtain, simple to process, and highly relevant to extracting the health characteristics of the battery. However, since the battery discharge process is dynamic, the extraction of its stable features is challenging. For example, under different discharge conditions (such as operating temperature, discharge current, and cut-off voltage), there are differences in discharge data, so it is difficult to obtain regular and reliable discharge data. In order to improve the accuracy of SOH estimation, more complex models may be required for feature extraction and SOH estimation, but the complexity of the model will bring other problems. How to balance complexity and accuracy still needs further research. Summary of the invention

[0005] The purpose of the present invention is to propose a battery health state estimation method based on battery local fragment data extraction, which can balance the complexity and accuracy of the estimation model.

[0006] The present invention is achieved through the following technical solutions: A method for estimating a battery health state based on battery local segment data extraction comprises the following steps: Step S1, constructing an estimation model, the estimation model includes an attention mapping layer for generating feature nodes, an enhancement node layer for enhancing feature nodes into enhancement nodes, a discard layer connected to the enhancement node layer, and an output layer connected to the attention mapping layer and the discard layer, the output weight of the output layer is calculated by elastic network regression, the attention mapping layer includes a feature attention module, a time attention module, and a fusion module for fusing feature attention and time attention; Step S2: m Based on statistical data, the interval segment for extracting features is determined on the battery charging curve corresponding to each charge and discharge cycle. , with the same interval △ V Select n Voltage segment ,in accordance with n The battery charging curve is intercepted by voltage segments, and the n Battery charging curve segment, m The battery charging curve segments obtained from the charge-discharge cycle are fitted, and the fitted parameters are used as input features to form an input feature matrix, where: , ; Step S3: Input the input feature matrix into the estimation model, and use the multi-objective particle swarm optimization algorithm to obtain the optimal interval △ with the calculation cost and calculation accuracy as the target V p , and at the optimal interval △ V p Repeat step S2 to obtain the best input feature matrix; Step S4: Repeat steps S2 and S3 for a set number of times, with each repetition taking place at different intervals. V Select the voltage segment and obtain the optimal input feature matrix corresponding to this repetition; Step S5: Input each optimal input feature matrix into the estimation model, and evaluate the estimation results by MAPE and RMSE to determine the final optimal interval, and repeat step S2 with the final optimal interval to obtain the final input feature matrix, and input the final input feature matrix into the estimation model for estimation.

[0007] Furthermore, in the step S2, in each of the charge and discharge cycles, based on statistical data, the range of the interval segment for extracting features is [3.8V, 4.2V].

[0008] Furthermore, in step S2, in each charge and discharge cycle, a quadratic equation is used to calculate the Each of the battery charging curve segments is fitted, and the parameters obtained by fitting include quadratic parameters, primary parameters and constant parameters. The correlations between the quadratic parameters, primary parameters and constant parameters and the average value of the battery capacity under the corresponding battery charging curve segment are compared respectively, and the parameters with the largest correlation are taken to form the input feature matrix.

[0009] Furthermore, in step S2, the process of comparing the correlation between the secondary parameter, the primary parameter and the constant parameter and the average value of the battery capacity under the corresponding battery charging curve segment includes: Step S21, m In the first charge and discharge cycle, n The parameters obtained by fitting the battery charging curve segment corresponding to the voltage segment are: , , , respectively m The parameters obtained from the charge and discharge cycles are combined into three matrices , , ; Step S22: According to the formula Calculate the n The correlation parameter sequence of the voltage segments is calculated according to the formula , , Calculate the average value of the battery capacity under the corresponding battery charging curve segment with respect to the secondary parameter, primary parameter and constant parameter respectively, and take , , The largest number The corresponding matrix is ​​used as the input feature matrix ,in, Indicates n The secondary parameter time series under the battery charging curve segment, Indicates n A time series of parameters under a battery charging curve segment, Indicates n The constant parameter time series under the battery charging curve segment, Indicates n The average value sequence of battery capacity under the battery charging curve segment, E () indicates taking the mean value, D () indicates taking the variance.

[0010] Furthermore, in step S1, the loss function of the elastic network regression calculation is expressed as ,in, x i Indicates i The features of the estimated observations, y i Indicates i The actual observed values, represents the Lasso regression regularization coefficient, represents the ridge regression regularization coefficient, p represents the number of actual observations, represents the regression coefficient, represents the Lasso regression regularization term, represents the ridge regression regularization term.

[0011] Furthermore, in step S1, in the feature attention module, the feature attention weight matrix is ​​expressed as ,in, Indicates i The first charge and discharge cycle k The weight of the input feature fitted by the battery charging curve segment corresponding to the voltage segment, i =1,2,…, m , k =1,2,…, n .

[0012] Furthermore, in step S1, in the temporal attention module, the temporal attention weight matrix is ​​expressed as ,in, .

[0013] Furthermore, in step S1, in the fusion module, the fusion matrix is ​​expressed as .

[0014] The present invention has the following beneficial effects: The present invention first constructs an estimation model, which includes an attention mapping layer for generating feature nodes, an enhancement node layer for enhancing feature nodes into enhancement nodes, a discard layer connected to the enhancement node layer, and an output layer connected to the attention mapping layer and the discard layer. The output weight of the output layer is calculated by elastic network regression. The attention mapping layer includes a feature attention module, a time attention module, and a fusion module for fusing feature attention and time attention. Then, m In the charge and discharge cycle, the interval segment for extracting features is determined on the corresponding battery charging curve, with the same interval △ V Select n voltage segments, and based on nThe battery charging curve is intercepted by voltage segments, and the n The battery charging curve segments are fitted, and the fitted parameters are used as input features to form an input feature matrix. Then the input feature matrix is ​​input into the estimation model, and the multi-objective particle swarm optimization algorithm is used to obtain the optimal interval △ with the calculation cost and calculation accuracy as the target. V p , and at the optimal interval △ V p Get the best input feature matrix. In order to further improve the accuracy, set multiple intervals △ V , for each interval △ V , the above process is repeated to obtain the corresponding optimal input feature matrix, each optimal input feature matrix is ​​input into the estimation model, and the estimation model is evaluated by MAPE and RMSE to determine the final optimal interval. In the actual estimation, the final input feature matrix is ​​obtained with the final optimal interval, and the final input feature matrix is ​​input into the estimation model for estimation. In this process, features are extracted based on the battery charging curve. Compared with the discharge process, the charging process is relatively standardized, and more stable and representative features can be collected. The interval segments on the battery charging curve are selected to extract input features, which is more in line with the daily use of users of new energy vehicles. The multi-objective particle swarm optimization algorithm is used to determine the optimal interval with computational cost and computational accuracy as the goals. In this way, while ensuring the computational accuracy, the situation in which the amount of calculation is greatly increased due to too many segmented voltage segments can be avoided, and the risk of overfitting of the estimation model can also be reduced. The elastic network regression calculation of the estimation model can stabilize the estimation of the fitting coefficients of the battery charging curve segments, reduce the variance of the coefficients, and is more suitable for estimating the battery health status. The addition of the discard layer can reduce the risk of overfitting of the estimation model as a whole. The attention mapping layer in the estimation model calculates the attention weights for the time series and the feature series respectively, so that the estimation model can better focus on key information instead of uniformly processing all inputs, thereby improving the performance of the estimation model and shortening the training time. In summary, the present invention can effectively balance the complexity and accuracy of the estimation model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention is further described in detail below with reference to the accompanying drawings.

[0016] Figure 1 It is a flow chart of the present invention.

[0017] Figure 2 The SOH comparison simulation diagram for different selections of K (K=1, 2, 3) for a univariate K-order equation.

[0018] Figure 3 This is a comparison simulation diagram of SOH fitting of a linear equation and a quadratic equation.

[0019] Figure 4 It is a framework diagram of the estimation model of the present invention.

[0020] Figure 5 It is a simulation diagram of battery health state estimation of the present invention and the existing SOH estimation method.

[0021] Figure 6 It is a simulation diagram of the battery health state estimation error of the present invention and the existing SOH estimation method. DETAILED DESCRIPTION

[0022] The data set used in this example comes from NASA Ames Prediction Center of Excellence, where three lithium-ion batteries (labeled 5, 6, and 7) are second-generation 18650 batteries produced by the Idaho National Laboratory. All batteries were repeatedly operated at room temperature 24°C in three different operating modes, namely constant current and constant voltage (CC-CV) charging mode, CC discharge mode, and impedance measurement mode. The CC charging current, upper limit voltage, and CV charging voltage were 1.5A, 4.2V, and 4.2V, respectively, and CV ended when the current dropped to 50mA. CC discharge and CV charging were repeated continuously. The experiment was terminated when the LIB reached the end of life (EOL) standard.

[0023] In this embodiment, SOH is defined as capacity decay, that is, ,in, C i Indicates the maximum capacity of the current cycle, C O Indicates the initial capacity.

[0024] like Figure 1 As shown, the battery health status estimation method based on battery local segment data extraction includes the following steps: Step S1, constructing an estimation model, the estimation model includes an attention mapping layer for generating feature nodes, an enhancement node layer for enhancing feature nodes into enhancement nodes, a discard layer connected to the enhancement node layer, and an output layer connected to the attention mapping layer and the discard layer, the output weight of the output layer is calculated by elastic network regression, the attention mapping layer includes a feature attention module, a time attention module, and a fusion module for fusing feature attention and time attention; Step S2: m Based on statistical data, the interval segment for extracting features is determined on the battery charging curve corresponding to each charge and discharge cycle. , with the same interval △ V Select n Voltage segment ,in accordance withn The battery charging curve is intercepted by voltage segments, and the n Battery charging curve segment, m The battery charging curve segments obtained from the charge-discharge cycle are fitted, and the fitted parameters are used as input features to form an input feature matrix, where: , ; As the battery ages, the battery charging curve has a clear upward trend, which can well explain the battery aging process. Statistics of 11,000 user charging behaviors in 100 electric vehicles show that most users choose to start charging at a voltage of 3.8V and end charging at a voltage of 4.18V. Therefore, the voltage range of the interval segment for extracting features in this embodiment is determined to be [3.8V, 4.2V].

[0025] like Figure 2 The following is a comparison of the SOH simulation diagrams for different choices of K (K = 1, 2, 3) for a univariate K-order equation. Figure 2 It can be seen that the one-dimensional linear function and the one-dimensional quadratic function can still present accurate predictions when fitting and predicting, but when the function index reaches 3 times, the overall predicted value fluctuates violently and the result is very inaccurate, proving that the robustness is poor and is not suitable for battery health status prediction. Since other functions or other functions with better exponents take a lot of time in the fitting process, it deviates from the original intention of our method, which is to achieve fast real-time measurement on board, so other fitting methods are not considered. Next, we will further discuss the one-dimensional linear fitting function and the one-dimensional quadratic fitting function on the basis of the above to finally determine the fitting function.

[0026] like Figure 3 The figure shows the SOH comparison simulation diagram of the linear equation and the quadratic equation. It can be seen that in the second half of the curve, the linear equation and the quadratic equation are comparable in results, and the linear equation has an advantage in fitting speed. However, in the initial stage of the battery cycle, the linear equation cannot accurately estimate the SOH of the battery, and there are still large errors and fluctuations. Therefore, after weighing the pros and cons, it is finally chosen to use the quadratic equation as the fitting function to extract features from the battery health curve and make estimates.

[0027] That is, in this embodiment, in each charge and discharge cycle, a quadratic equation is used Each battery charging curve segment is fitted, and the fitting parameters include quadratic parameters a , primary parameter b and constant parameters d, respectively compare the correlations between the secondary parameters, primary parameters and constant parameters and the average value of the battery capacity under the corresponding battery charging curve segment, and take the parameters with the largest correlation to form the input feature matrix.

[0028] The process of comparing the correlations of the secondary parameter, the primary parameter and the constant parameter with the average value of the battery capacity under the corresponding battery charging curve segment includes: Step S21, m In the first charge and discharge cycle, n The parameters obtained by fitting the battery charging curve segment corresponding to the voltage segment are: , , , respectively m The parameters obtained from the charge and discharge cycles are combined into three matrices , , ; Step S22: According to the formula Calculate the n The correlation parameter sequence of the voltage segments is calculated according to the formula , , Calculate the average value of the battery capacity under the corresponding battery charging curve segment with respect to the secondary parameter, primary parameter and constant parameter respectively, and take , , The largest number The corresponding matrix is ​​used as the input feature matrix ,in, Indicates n The secondary parameter time series under the battery charging curve segment, Indicates n A time series of parameters under a battery charging curve segment, Indicates n The constant parameter time series under the battery charging curve segment, Indicates n The average value sequence of battery capacity under the battery charging curve segment, E () indicates taking the mean value, D () indicates taking the variance.

[0029] Step S3: Input the input feature matrix into the estimation model, and use the multi-objective particle swarm optimization algorithm to obtain the optimal interval △ with the calculation cost and calculation accuracy as the target V p , and at the optimal interval △ V p Repeat step S2 to obtain the best input feature matrix; Using multi-objective particle swarm optimization algorithm to obtain the optimal interval △V p , specifically including the following steps: Step S11: Initialize the particle position x n0 ,speed v n0 , the global optimal solution set G n0 , and the individual optimal solution for each particle P n0 ; Step S12: In the kth iteration, according to the formula Update speed v nk , according to the formula Update particle position x nk ,in, w Represents the inertia weight, which describes the degree to which the current velocity of the particle affects its next velocity. r 1 represents the individual learning factor, which is a random number, usually varying between [0,1], and is used to control the degree to which the particle approaches its historical optimal position. r 2 represents the social learning factor, which is a random number, usually varying between [0,1], and is used to control the degree to which particles approach the global optimal position in the group. e represents random disturbance; Step S13, after the set number of iterations G, according to the formula The particle position obtained x nG Constraints are imposed to ensure that the particle position is within a reasonable range, according to the formula y n = f ( x n ) calculates the fitness value to evaluate the quality of the current particle position. For the non-dominated solution set, the new solution is added and the dominated solution is removed. If the new solution is better than the current optimal solution, the new solution is used as the optimal solution. enforceConstrans () is the constraint function, f ( x n )satisfy min (△ V p )=[ f 1( x n ), f 2( x n )], min (△ V p ) means in satisfying f 1(x n )and f 2( x n ) Under the premise of two objective functions △ V p Take the minimum value, x n Represents the optimal interval △ V p , f 1( x n )express x n As the computational cost at the optimal interval, f 2( x n )express x n The Pearson correlation coefficient of the input features obtained under the interval is expressed as G is an integer. k =1,2,…,G. The optimal interval △ determined in this embodiment V p is 0.05V.

[0030] Step S4: Repeat steps S2 and S3 for a set number of times, with each repetition taking place at different intervals. V Select the voltage segment and obtain the optimal input feature matrix corresponding to this repetition; In this embodiment, the number of times is set to 10 times, with a △ V The values ​​are 0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, and 0.1 respectively.

[0031] Step S5: Input each optimal input feature matrix into the estimation model, and evaluate the estimation results by MAPE and RMSE to determine the final optimal interval, and repeat step S2 with the final optimal interval to obtain the final input feature matrix, and input the final input feature matrix into the estimation model for estimation.

[0032] In the estimation model, the loss function of elastic net regression calculation is expressed as ,in, x i Indicates i The features of the estimated observations, y i Indicates i The actual observed values, represents the Lasso regression regularization coefficient, represents the ridge regression regularization coefficient, p represents the number of actual observations, represents the regression coefficient, represents the Lasso regression regularization term, Represents the ridge regression regularization term. Elastic net regression uses both the Lasso regression regularization coefficient and the ridge regression regularization coefficient in the calculation, thereby achieving a trade-off between feature selection and parameter reduction, and can accurately reduce some coefficients to zero, that is, it can automatically identify features that are not important to the target variable, thereby simplifying the estimation model and improving the interpretability. Elastic net regression is suitable for dealing with multicollinearity problems, thereby stabilizing coefficient estimation and reducing coefficient variance.

[0033] In the feature attention module, the feature attention weight matrix is ​​expressed as ,in, Indicates i The first charge and discharge cycle k The weight of the input feature fitted by the battery charging curve segment corresponding to the voltage segment, i =1,2,…, m , k =1,2,…, n .

[0034] In the temporal attention module, the temporal attention weight matrix is ​​expressed as ,in, .

[0035] In the fusion module, the feature attention weight matrix and the time attention matrix obtained above are multiplied by the corresponding elements of the matrices to obtain a matrix of the same size as the input two-dimensional matrix, namely the fusion matrix C , expressed as .

[0036] The feature enhancement layer can be implemented by existing technologies, such as the feature enhancement layer technology using a width learning network. The discard layer can also be implemented by existing technologies.

[0037] For multi-features like battery data, time series input cannot be effectively processed. The introduction of the attention mapping layer enables the estimation model to effectively focus on key information. In order to avoid the overfitting problem of the estimation model, the dropout layer is added to the width learning network after adding it to the enhancement node. The dropout layer randomly ignores the node to prevent the prediction model from over-relying on the relevant feature values ​​during training. Figure 4 The framework structure diagram of the estimation model.

[0038] After inputting each optimal input feature matrix into the estimation model, the estimation results are evaluated by MAPE and RMSE, and the optimal interval corresponding to the optimal input feature matrix with the best evaluation effect is selected as the final optimal interval. In actual estimation, step S2 is repeated with the final optimal interval to obtain the final input feature matrix. Among them, the MAPE formula is , the RMSE formula is ,in, represents the predicted value, represents the true value, n Indicates the sample size of the test dataset.

[0039] like Figure 5 The above is a battery health state estimation simulation diagram of the present invention and the existing SOH estimation methods (convolutional neural network CNN model, long short neural network LSTM model, width learning network BLS model, width learning A-BLS estimation with added attention mechanism), as shown in Figure 6 The above is a battery estimation error Error simulation diagram of the present invention and the existing SOH estimation method. It can be seen that the RMSE of the present invention can be reduced to 0.57%, and the MAPE can be reduced to 0.46%, both lower than other estimation methods, which effectively improves the accuracy of the present invention in estimating the battery health state.

[0040] The above description is only a preferred embodiment of the present invention, and therefore cannot be used to limit the scope of implementation of the present invention. That is, equivalent changes and modifications made according to the scope of the patent application of the present invention and the contents of the specification should still fall within the scope covered by the patent of the present invention.

Claims

1. A battery health status estimation method based on battery local segment data extraction, characterized in that: The steps include: Step S1, constructing an estimation model, the estimation model includes an attention mapping layer for generating feature nodes, an enhancement node layer for enhancing feature nodes into enhancement nodes, a discard layer connected to the enhancement node layer, and an output layer connected to the attention mapping layer and the discard layer, the output weight of the output layer is calculated by elastic network regression, the attention mapping layer includes a feature attention module, a time attention module, and a fusion module for fusing feature attention and time attention; Step S2: m Based on statistical data, the interval segment for extracting features is determined on the battery charging curve corresponding to each charge and discharge cycle. , with the same interval △ V Select n Voltage segment ,in accordance with n The battery charging curve is intercepted by voltage segments, and the n Battery charging curve segment, m The battery charging curve segments obtained from the charge-discharge cycle are fitted, and the fitted parameters are used as input features to form an input feature matrix, where: , ; Step S3: Input the input feature matrix into the estimation model, and use the multi-objective particle swarm optimization algorithm to obtain the optimal interval △ with the calculation cost and calculation accuracy as the target V p , and at the optimal interval △ V p Repeat step S2 to obtain the best input feature matrix; Step S4: Repeat steps S2 and S3 for a set number of times, with each repetition taking place at different intervals. V Select the voltage segment and obtain the optimal input feature matrix corresponding to this repetition; Step S5: Input each optimal input feature matrix into the estimation model, and evaluate the estimation results by MAPE and RMSE to determine the final optimal interval, and repeat step S2 with the final optimal interval to obtain the final input feature matrix, and input the final input feature matrix into the estimation model for estimation.

2. The method for estimating the battery health status based on battery local segment data extraction according to claim 1, characterized in that: In the step S2, in each of the charge and discharge cycles, based on statistical data, the range of the interval segment for extracting features is [3.8V, 4.2V].

3. The method for estimating the battery health status based on battery local segment data extraction according to claim 2, characterized in that: In step S2, in each charge and discharge cycle, a quadratic equation is used to Each of the battery charging curve segments is fitted, and the parameters obtained by fitting include quadratic parameters, primary parameters and constant parameters. The correlations between the quadratic parameters, primary parameters and constant parameters and the average value of the battery capacity under the corresponding battery charging curve segment are compared respectively, and the parameters with the largest correlation are taken to form the input feature matrix.

4. The method for estimating the battery health status based on battery local segment data extraction according to claim 3, characterized in that: In step S2, the process of comparing the correlation between the secondary parameter, the primary parameter and the constant parameter and the average value of the battery capacity under the corresponding battery charging curve segment includes: Step S21, m In the first charge and discharge cycle, n The parameters obtained by fitting the battery charging curve segment corresponding to the voltage segment are: , , , respectively m The parameters obtained from the charge and discharge cycles are combined into three matrices , , ; Step S22: According to the formula Calculate the n The correlation parameter sequence of the voltage segments is calculated according to the formula , , Calculate the average value of the battery capacity under the corresponding battery charging curve segment with respect to the secondary parameter, primary parameter and constant parameter respectively, and take , , The largest number The corresponding matrix is ​​used as the input feature matrix ,in, Indicates n The secondary parameter time series under the battery charging curve segment, Indicates n A time series of parameters under a battery charging curve segment, Indicates n The constant parameter time series under the battery charging curve segment, Indicates n The average value sequence of battery capacity under the battery charging curve segment, E () indicates taking the mean value, D () indicates taking the variance.

5. The method for estimating the battery health status based on battery local segment data extraction according to claim 4, characterized in that: In step S1, the loss function of the elastic network regression calculation is expressed as ,in, x i Indicates i The features of the estimated observations, y i Indicates i The actual observed values, represents the Lasso regression regularization coefficient, represents the ridge regression regularization coefficient, p represents the number of actual observations, represents the regression coefficient, represents the Lasso regression regularization term, represents the ridge regression regularization term.

6. The method for estimating the battery health status based on battery local segment data extraction according to claim 5, characterized in that: In step S1, in the feature attention module, the feature attention weight matrix is ​​expressed as ,in, Indicates i The first charge and discharge cycle k The weight of the input feature fitted by the battery charging curve segment corresponding to the voltage segment, i =1,2,…, m , k =1,2,…, n .

7. The method for estimating the battery health status based on battery local segment data extraction according to claim 6, characterized in that: In step S1, in the temporal attention module, the temporal attention weight matrix is ​​expressed as ,in, .

8. The method for estimating the battery health status based on battery local segment data extraction according to claim 7, characterized in that: In the step S1, in the fusion module, the fusion matrix is ​​expressed as .

Images