Lithium battery SOH estimation method based on pre-training fused multi-source data

Abstract

The invention discloses a lithium battery SOH (state of health) estimation method based on pre-training fused multi-source data. The method comprises the following steps: acquiring constant-voltage charging current data and electrochemical impedance spectroscopy data; preprocessing the data and forming a data set; constructing a pre-training task 1, outputting two features through an encoder 1 and an encoder 2, and constructing a loss function loss1 of the task 1 according to cosine similarity between feature fragments; constructing a pre-training task 2, randomly masking segments of two features, outputting fusion features through an encoder 3, constructing two decoder reconstruction mask segments to obtain reconstruction loss loss2 and loss3, and weighting based on correlation to obtain a loss function loss4 of the task 2; the encoder is frozen to form an SOH estimation model with an estimator; and estimating the SOH on line. According to the invention, the encoder can fully fuse two kinds of data aging information through the two pre-training tasks, and the SOH of the lithium battery is accurately estimated.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a lithium battery SOH estimation method based on pre-training fusion of multi-source data. Background Art

[0002] Lithium batteries have become the main energy storage system for smart grids, microgrids, smart electronic devices and electric vehicles due to their high energy density, low self-discharge rate and long operating life. The state of health (SOH) of lithium batteries intuitively reflects the current aging of the battery. The capacity of lithium-ion batteries will inevitably age during their life cycle. These complex electrochemical reactions and changes in external performance make it difficult to accurately estimate SOH. Therefore, accurate estimation of SOH is very important for the safety performance evaluation of lithium batteries.

[0003] Today, data-driven lithium battery SOH estimation methods are a hot topic. This method generally constructs manual features from different charging operation curves, and then inputs them into the machine learning model to estimate SOH, or directly inputs the charging operation curve into the machine learning model to estimate SOH. Existing data-driven lithium battery SOH estimation methods can be roughly divided into SOH estimation methods based on battery operation data such as current, current and temperature, and SOH estimation methods based on multi-source data. Among them, the SOH estimation method based on multi-source data comprehensively utilizes the battery aging information contained in the operation data and electrochemical impedance spectroscopy data, avoiding the problem that SOH estimation is easily interfered by single sensor noise. Existing SOH estimation methods based on multi-source data generally combine or splice the health features of the operation data and the health features of the electrochemical impedance spectrum to obtain new health features and use them to estimate SOH. However, there are usually heterogeneity and semantic differences between multi-source data. Simple fusion methods are difficult to capture the interactive relationship between different data. Therefore, the aging information of multi-source data is not fully integrated, resulting in limitations in the application of existing methods, which is not conducive to improving the estimation accuracy of SOH. Summary of the invention

[0004] In view of the above-mentioned deficiencies in the prior art, the present invention proposes a lithium battery SOH estimation method based on pre-training fusion of multi-source data, aiming to fully fuse multi-source data and improve the SOH estimation accuracy.

[0005] The technical solution of the present invention is as follows:

[0006] S1: Data collection of lithium battery aging experiment, i.e., a cyclic charge and discharge experiment is performed on multiple lithium batteries in a constant current-constant voltage charging and constant current discharging mode. In each charge and discharge cycle, constant voltage charging current data is continuously collected in the constant voltage charging stage. After charging, the lithium battery is left to stand for 15 minutes and then the electrochemical impedance spectrum data in the frequency range of 0.02Hz-20kHz is collected. After discharging, the SOH value of the lithium battery in this cycle is calculated.

[0007] S2: Data preprocessing and data set construction, that is, data cleaning of the constant voltage charging current data and electrochemical impedance spectrum data described in step S1, and calculation of differential current data based on the constant voltage charging current data. The specific calculation method is as follows:

[0008] dI n =I n+1 -I n (1)

[0009] Among them, dI n Represents the differential current data at the nth time sampling point, I n Represents the constant voltage charging current data at the nth time sampling point;

[0010] According to the constant voltage charging current data and differential current data, the current matrix X is constructed. CV , as follows:

[0011]

[0012] Wherein, p is the total number of constant voltage charging current data collected;

[0013] According to the real and imaginary parts of all impedances in the electrochemical impedance spectroscopy data, the electrochemical impedance spectroscopy matrix X is constructed. EIS , as follows:

[0014]

[0015] Among them, Re m and Im m They represent the real and imaginary parts of the impedance data corresponding to the mth frequency sampling point, respectively, and u is the total number of impedance data collected;

[0016] The current matrix X CV Divide it into N sub-matrices evenly according to the time dimension, then the current matrix X CV It can be expressed as:

[0017]

[0018] Among them, X CV,i represents the i-th current submatrix, whose dimension is L cv ×2, Lcv is the number of sampling points corresponding to the sub-matrix;

[0019] The electrochemical impedance spectroscopy matrix X EIS According to the frequency dimension, it is evenly divided into M sub-matrices, then the electrochemical impedance spectroscopy matrix X EIS It can be expressed as:

[0020]

[0021] Among them, X EIS,j Represents the jth electrochemical impedance spectroscopy submatrix, whose dimension is L eis ×2, L eis is the number of sampling points corresponding to the sub-matrix;

[0022] The above sub-matrix-divided current matrix and electrochemical impedance spectrum matrix are defined as and Will and Constructing an unlabeled dataset Combined with the SOH described in step S1 to form a labeled data set

[0023] S3: Construct pre-training task 1, that is, build encoder 1 and encoder 2, and convert the Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the step S2 described Input to encoder 2, encoder 2 outputs EIS characteristic HF EIS , calculate HF CV Each characteristic fragment is related to HF EIS The cosine similarity between each feature segment in the pre-training task 1 is calculated, and a loss loss1 that takes all cosine similarities into consideration is constructed as the loss function of the pre-training task 1. The unlabeled dataset D1 described in step S2 is used to train encoder 1 and encoder 2. After the training is completed, the weight parameters of encoder 1 and encoder 2 are saved, and the unlabeled dataset D1 is input into the trained encoder 1 and encoder 2 to obtain the feature set.

[0024] S4: Construct pre-training task 2, that is, the feature HF described in step S3 CV and HF EIS In each case, a feature segment is randomly selected for masking, and the two masked features are concatenated and then input into the encoder 3 based on the self-attention mechanism. The encoder 3 outputs and The fusion feature matrix is It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current feature aging information. Decoder 1 and decoder 2 are built to Input to decoder 1, HF CV The masked fragments are reconstructed. Input to decoder 2, HF EIS The masked segments are reconstructed to construct the mask reconstruction loss loss2 of the constant voltage current feature and the mask reconstruction loss loss3 of the EIS feature. According to the correlation score between the constant voltage charging current data, the electrochemical impedance spectroscopy data and the SOH after data cleaning described in step S2, loss2 and loss3 are weighted to obtain loss4. Loss4 is used as the loss function of the pre-training task 2. The feature set F described in step S3 is used to train the encoder 3, the decoder 1 and the decoder 2. After the training is completed, the weight parameters of the encoder 3, the decoder 1 and the decoder 2 are saved.

[0025] S5: Construct and train a SOH estimation model, which includes an encoder module and an estimator, that is, the encoder 1 and encoder 2 described in step S3 and the encoder 3 described in step S4 together constitute an encoder module, and the weight parameters of the encoder 1, encoder 2 and encoder 3 obtained in step S3 and step S4 are frozen, and the estimator to be trained is built behind the encoder module to establish the estimator described in step S4. and The mapping relationship between the fusion feature matrix and the SOH value is formed, the labeled data set D2 described in step S2 is divided into a training set and a test set, and the SOH estimation model including the encoder module and the estimator is trained and tested. If the test result meets the accuracy requirement, the SOH estimation model is saved;

[0026] S6: Online estimation of the SOH of the lithium battery, that is, in the actual charge and discharge cycle of the lithium battery, constant voltage charging current data and electrochemical impedance spectrum data are collected, and after data preprocessing according to the method of step S2, the constant voltage charging current data and electrochemical impedance spectrum data are input into the SOH estimation model described in step S5 to output the SOH estimation value.

[0027] In this solution, the specific steps of constructing the pre-training task 1 described in step S3 are as follows:

[0028] S301: Build encoder 1, and convert the encoder described in step S2 Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the encoder 1 will Linearly project the N sub-matrices in to obtain the corresponding N constant voltage and current feature fragments with a dimension of 1×c, as follows:

[0029]

[0030] Among them, Encoder1(·) represents encoder 1, HF CV,i represents the i-th constant voltage and current feature segment, whose dimension is 1×c, HF CV The dimension is N×c;

[0031] S302: Build encoder 2, and convert the encoder described in step S2 Input to encoder 2, encoder 2 outputs EIS characteristic HF EIS , the encoder 2 will Linear projection is performed on the M sub-matrices in to obtain the corresponding M EIS characteristic fragments with a dimension of 1×c, as follows:

[0032]

[0033] Among them, Encoder2(·) represents encoder 2, HF EIS,j represents the jth EIS characteristic fragment, whose dimension is 1×c, HF EIS The dimension is M×c;

[0034] S303: Calculate the HF in step S301 CV Each characteristic segment in step S302 is consistent with the HF EIS The cosine similarity between each feature fragment in is as follows:

[0035]

[0036] Among them, CS ij Indicates HF CV,i and HF EIS,j The cosine similarity between HF CV,i ·HF EIS,j represents the dot product of two feature fragments, ‖·‖ represents the magnitude of the vector;

[0037] S304: Construct a loss1 that takes into account all cosine similarities, as follows:

[0038] CL ij =1-CS ij (9)

[0039] Q=Var({CL ij |i=1,2,…,N,j=1,2,…,M}) (10)

[0040]

[0041] Among them, CLij Indicates HF CV,i and HF EIS,j The cosine similarity loss between them, Var(·) represents the variance calculation function, Q represents the variance of all cosine similarity losses, and λ1 is the regularization coefficient of loss1;

[0042] S305: Use the unlabeled data set D1 described in step S2 to train encoder 1 and encoder 2, use loss1 as the loss function of the training, iteratively update the weight parameters of encoder 1 and encoder 2 through the back propagation algorithm, so as to complete the training of encoder 1 and encoder 2, save the weight parameters after the training is completed, input the unlabeled data set D1 into the encoder 1 and encoder 2 after the training is completed, and obtain the feature set

[0043] In this solution, the specific steps of constructing the pre-training task 2 described in step S4 are as follows:

[0044] S401: Constant voltage current characteristic HF described in step S3 CV In the process, a feature segment is randomly selected for masking operation, that is, for each charging cycle sample, the kth constant voltage current feature segment HF is randomly masked. CV,k All c elements in are set to 0, and the masked constant voltage and current characteristics are obtained. The dimension is N×c;

[0045] S402: EIS characteristic HF described in step S3 EIS In the process, a characteristic segment is randomly selected for masking operation, that is, for each charging cycle sample, the hth EIS characteristic segment HF is randomly masked. EIS,h All c elements in are set to 0 to obtain the masked EIS feature The dimension is M×c;

[0046] S403: Build an encoder 3 based on the self-attention mechanism. and After being spliced ​​into a matrix of dimension (N+M)×c, it is input into encoder 3, and the connection between all feature fragments is learned through the self-attention mechanism, thereby outputting a fused feature matrix. The fused feature matrix is ​​composed of and Composition, of which It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current characteristic aging information, which is specifically expressed as follows:

[0047]

[0048] Among them, concat(·) represents the concatenation function, Encoder3(·) represents the encoder 3 based on the self-attention mechanism, and its output and The dimensions are N×c and M×c respectively;

[0049] S404: Build decoder 1, and convert the decoder described in step S403 Input decoder 1 to reconstruct the feature segment HF described in step S401 CV,k , the reconstruction result is obtained as The specific expressions are as follows:

[0050]

[0051] Wherein, Decoder1(·) represents decoder 1;

[0052] According to HF CV,k The original value y of the ath element in cv,a and The ath reconstruction value in Calculate HF CV,k The mask reconstruction error l of the a-th element in cv,a , specifically expressed as follows:

[0053]

[0054] Construct the mask reconstruction loss loss2 of the constant voltage and current feature to describe the feature segment HF CV,k The mask reconstruction error and error stability of all elements in are specifically expressed as follows:

[0055]

[0056] Among them, Var(l cv,1 ,l cv,2 ,…,l cv,c ) indicates the calculation of HF CV,k The variance of the mask reconstruction error of all elements in , λ2 is the regularization coefficient of loss2;

[0057] S405: Build decoder 2, and convert the decoder described in step S403 Input to decoder 2 to reconstruct the feature segment HF described in step S402 EIS,h , the reconstruction result is obtained as The specific expressions are as follows:

[0058]

[0059] Wherein, Decoder2(·) represents decoder 2;

[0060] According to HF EIS,h The original value y of the bth element in eis,b and The bth reconstruction value in Calculate HF EIS,h The mask reconstruction error l of the bth element in eis,b , specifically expressed as follows:

[0061]

[0062] Construct the mask reconstruction loss loss3 of EIS feature to describe the feature fragment HF EIS,h The mask reconstruction error and error stability of all elements in are specifically expressed as follows:

[0063]

[0064] Among them, Var(l eis,1 ,l eis,2 ,…,l eis,c ) indicates the calculation of HF EIS,h The variance of the mask reconstruction error of all elements in , λ3 is the regularization coefficient of loss3;

[0065] S406: for the constant voltage charging current data after data cleaning in step S2, calculate the mean, standard deviation, kurtosis and skewness of the constant voltage charging current data in each cycle, and for the electrochemical impedance spectroscopy data after data cleaning in step S2, calculate the real part mean, imaginary part mean, covariance between the real part and the imaginary part, and dot product of the real part and the imaginary part of the electrochemical impedance spectroscopy data corresponding to each cycle, to form 8 artificial features;

[0066] S407: Calculate the Pearson correlation coefficient and the Spearman correlation coefficient between the eight artificial features described in step S406 and the battery SOH, and calculate the aging correlation score Z of the vth artificial feature by combining the Softmax function. v , the specific calculation process is as follows:

[0067]

[0068] Among them, Softmax(·) is the Softmax function, γ v represents the Pearson correlation coefficient between the vth artificial feature and SOH, ρ v represents the Spearman correlation coefficient between the vth artificial feature and SOH;

[0069] S408: Add Z1 to Z4 calculated in step S407 to obtain the weight α of loss2 described in S404, add Z5 to Z8 calculated in step S407 to obtain the weight β of loss3 described in S405, linearly weight loss2 and loss3 according to their respective weights to obtain the joint loss loss4, as follows:

[0070] loss4=αloss2+βloss3 (20)

[0071] S409: Use the feature set F described in step S3 to train encoder 3, decoder 1 and decoder 2, use loss4 as the loss function of training, iteratively update the weight parameters of encoder 3, decoder 1 and decoder 2 through the back propagation algorithm, thereby completing the training of encoder 3, decoder 1 and decoder 2, and save the weight parameters after the training is completed.

[0072] Compared with the prior art, the technical solution of the present invention has the following beneficial effects:

[0073] The present invention designs two pre-training tasks to fuse multi-source data. Through pre-training task 1, the cosine similarity loss between the constant-voltage current feature fragment and the EIS feature fragment is minimized, which can effectively improve the similarity of the two data in the feature space, align the local aging information between the two features in a finer granularity, and obtain the aging features that correspond and match each other between the constant-voltage charging current and the EIS; through pre-training task 2, the two features are fused by the self-attention mechanism, and the masked EIS feature fragment is reconstructed using the fused constant-voltage current feature, and the masked constant-voltage current feature fragment is reconstructed using the fused EIS feature. This design helps the model learn the complementarity of the two features and strengthen the collaborative information between the two features; the above two pre-training tasks enable the encoder module to fully fuse the aging information from the constant-voltage charging current data and the EIS data, thereby improving the estimation accuracy of SOH. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0075] Figure 1 A schematic diagram of a lithium battery SOH estimation method based on pre-training fusion of multi-source data;

[0076] Figure 2 Schematic diagram of matrix construction and partitioning of electrochemical impedance spectroscopy data (taking the partitioning into 5 sub-matrices as an example);

[0077] Figure 3 The diagram of matrix construction and partitioning of constant voltage charging current data (taking the partitioning into 5 sub-matrices as an example); DETAILED DESCRIPTION

[0078] In order to more clearly understand the above-mentioned objectives, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0079] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0080] In a specific embodiment, Figure 1 As shown, a lithium battery SOH estimation method based on pre-training fusion of multi-source data includes the following steps:

[0081] S1: Data collection of lithium battery aging experiment, i.e., a cyclic charge and discharge experiment is performed on multiple lithium batteries in a constant current-constant voltage charging and constant current discharging mode. In each charge and discharge cycle, constant voltage charging current data is continuously collected in the constant voltage charging stage. After charging, the lithium battery is left to stand for 15 minutes and then the electrochemical impedance spectrum data in the frequency range of 0.02Hz-20kHz is collected. After discharging, the SOH value of the lithium battery in this cycle is calculated.

[0082] S2: Data preprocessing and data set construction, that is, data cleaning of the constant voltage charging current data and electrochemical impedance spectrum data described in step S1, and calculation of differential current data based on the constant voltage charging current data. The specific calculation method is as follows:

[0083] dI n =I n+1 -I n (1)

[0084] Among them, dI n Represents the differential current data at the nth time sampling point, I n Represents the constant voltage charging current data at the nth time sampling point;

[0085] According to the constant voltage charging current data and differential current data, the current matrix X is constructed. CV , as follows:

[0086]

[0087] Wherein, p is the total number of constant voltage charging current data collected;

[0088] According to the real and imaginary parts of all impedances in the electrochemical impedance spectroscopy data, the electrochemical impedance spectroscopy matrix X is constructed. EIS , as follows:

[0089]

[0090] Among them, Re m and Im m They represent the real and imaginary parts of the impedance data corresponding to the mth frequency sampling point, respectively, and u is the total number of impedance data collected;

[0091] The current matrix X CV Divide it into N sub-matrices evenly according to the time dimension, then the current matrix X CV It can be expressed as:

[0092]

[0093] Among them, X CV,i represents the i-th current submatrix, whose dimension is L cv ×2, L cv is the number of sampling points corresponding to the sub-matrix;

[0094] The electrochemical impedance spectroscopy matrix X EIS According to the frequency dimension, it is evenly divided into M sub-matrices, then the electrochemical impedance spectroscopy matrix X EIS It can be expressed as:

[0095]

[0096] Among them, X EIS,j Represents the jth electrochemical impedance spectroscopy submatrix, whose dimension is L eis ×2, L eis is the number of sampling points corresponding to the sub-matrix;

[0097] The above sub-matrix-divided current matrix and electrochemical impedance spectrum matrix are defined as and Will and Constructing an unlabeled dataset Combined with the SOH described in step S1 to form a labeled data set

[0098] S3: Construct pre-training task 1, that is, build encoder 1 and encoder 2, and convert the Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the step S2 described Input to encoder 2, encoder 2 outputs EIS characteristic HFEIS , calculate HF CV Each characteristic fragment is related to HF EIS The cosine similarity between each feature segment in the pre-training task 1 is calculated, and a loss loss1 that takes all cosine similarities into consideration is constructed as the loss function of the pre-training task 1. The unlabeled dataset D1 described in step S2 is used to train encoder 1 and encoder 2. After the training is completed, the weight parameters of encoder 1 and encoder 2 are saved, and the unlabeled dataset D1 is input into the trained encoder 1 and encoder 2 to obtain the feature set.

[0099] S4: Construct pre-training task 2, that is, the feature HF described in step S3 CV and HF ESI In each case, a feature segment is randomly selected for masking, and the two masked features are concatenated and then input into the encoder 3 based on the self-attention mechanism. The encoder 3 outputs and The fusion feature matrix is It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current feature aging information. Decoder 1 and decoder 2 are built to Input to decoder 1, HG CV The masked fragments are reconstructed. Input to decoder 2, HF EIS The masked segments are reconstructed to construct the mask reconstruction loss loss2 of the constant voltage current feature and the mask reconstruction loss loss3 of the EIS feature. According to the correlation score between the constant voltage charging current data, the electrochemical impedance spectroscopy data and the SOH after data cleaning described in step S2, loss2 and loss3 are weighted to obtain loss4. Loss4 is used as the loss function of the pre-training task 2. The feature set F described in step S3 is used to train the encoder 3, the decoder 1 and the decoder 2. After the training is completed, the weight parameters of the encoder 3, the decoder 1 and the decoder 2 are saved.

[0100] S5: Construct and train a SOH estimation model, which includes an encoder module and an estimator, that is, the encoder 1 and encoder 2 described in step S3 and the encoder 3 described in step S4 together constitute an encoder module, and the weight parameters of the encoder 1, encoder 2 and encoder 3 obtained in step S3 and step S4 are frozen, and the estimator to be trained is built behind the encoder module to establish the estimator described in step S4. and The mapping relationship between the fusion feature matrix and the SOH value is formed, the labeled data set D2 described in step S2 is divided into a training set and a test set, and the SOH estimation model including the encoder module and the estimator is trained and tested. If the test result meets the accuracy requirement, the SOH estimation model is saved;

[0101] S6: Online estimation of the SOH of the lithium battery, that is, in the actual charge and discharge cycle of the lithium battery, constant voltage charging current data and electrochemical impedance spectrum data are collected, and after data preprocessing according to the method of step S2, the constant voltage charging current data and electrochemical impedance spectrum data are input into the SOH estimation model described in step S5 to output the SOH estimation value.

[0102] The specific steps of constructing the pre-training task 1 described in step S3 are as follows:

[0103] S301: Build encoder 1, and convert the encoder described in step S2 Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the encoder 1 will Linearly project the N sub-matrices in to obtain the corresponding N constant voltage and current feature fragments with a dimension of 1×c, as follows:

[0104]

[0105] Among them, Encoder1(·) represents encoder 1, HF CV ,i represents the i-th constant voltage current feature segment, whose dimension is 1×c, HF CV The dimension is N×c;

[0106] S302: Build encoder 2, and convert the encoder described in step S2 Input to encoder 2, encoder 2 outputs EIS characteristic HF EIS , the encoder 2 will Linear projection is performed on the M sub-matrices in to obtain the corresponding M EIS characteristic fragments with a dimension of 1×c, as follows:

[0107]

[0108] Among them, Encoder2(·) represents encoder 2, HF EIS,j represents the jth EIS characteristic fragment, whose dimension is 1×c, HF EIS The dimension is M×c;

[0109] S303: Calculate the HF in step S301 CV Each characteristic segment in step S302 is consistent with the HF EIS The cosine similarity between each feature fragment in is as follows:

[0110]

[0111] Among them, CS ij Indicates HF CV,i and HF EIS,j The cosine similarity between HF CV,i ·HF EIS,j represents the dot product of two feature fragments, ‖·‖ represents the magnitude of the vector;

[0112] S304: Construct a loss1 that takes into account all cosine similarities, as follows:

[0113] CL ij =1-CS ij (9)

[0114] Q=Var({CL ij |i=1,2,…,N,j=1,2,…,M}) (10)

[0115]

[0116] Among them, CL ij Indicates HF CV,i and HF EIS,j The cosine similarity loss between them, Var(·) represents the variance calculation function, Q represents the variance of all cosine similarity losses, and λ1 is the regularization coefficient of loss1;

[0117] S305: Use the unlabeled data set D1 described in step S2 to train encoder 1 and encoder 2, use loss1 as the loss function of the training, iteratively update the weight parameters of encoder 1 and encoder 2 through the back propagation algorithm, so as to complete the training of encoder 1 and encoder 2, save the weight parameters after the training is completed, input the unlabeled data set D1 into the encoder 1 and encoder 2 after the training is completed, and obtain the feature set

[0118] The specific steps of constructing the pre-training task 2 described in step S4 are as follows:

[0119] S401: The constant voltage current characteristic GF described in step S3 CV In the process, a feature segment is randomly selected for masking operation, that is, for each charging cycle sample, the kth constant voltage current feature segment HF is randomly masked. CV,k All c elements in are set to 0, and the masked constant voltage and current characteristics are obtained. The dimension is N×c;

[0120] S402: EIS characteristic HF described in step S3 EIS In the process, a characteristic segment is randomly selected for masking operation, that is, for each charging cycle sample, the hth EIS characteristic segment HF is randomly masked. EIS,h All c elements in are set to 0 to obtain the masked EIS feature The dimension is M×c;

[0121] S403: Build an encoder 3 based on the self-attention mechanism. and After being spliced ​​into a matrix of dimension (N+M)×c, it is input into encoder 3, and the connection between all feature fragments is learned through the self-attention mechanism, thereby outputting a fused feature matrix. The fused feature matrix is ​​composed of and Composition, of which It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current characteristic aging information, which is specifically expressed as follows:

[0122]

[0123] Among them, concat(·) represents the concatenation function, Encoder3(·) represents the encoder 3 based on the self-attention mechanism, and its output and The dimensions are N×c and M×c respectively;

[0124] S404: Build decoder 1, and convert the decoder described in step S403 Input decoder 1 to reconstruct the feature segment HF described in step S401 CV,k , the reconstruction result is obtained as The specific expressions are as follows:

[0125]

[0126] Wherein, Decoder1(·) represents decoder 1;

[0127] According to HF CV,k The original value y of the ath element in cv,a and The ath reconstruction value in Calculate HF CV,k The mask reconstruction error l of the a-th element in cv,a , specifically expressed as follows:

[0128]

[0129] Construct the mask reconstruction loss loss2 of the constant voltage and current feature to describe the feature segment HF CV,k The mask reconstruction error and error stability of all elements in are specifically expressed as follows:

[0130]

[0131] Among them, Var(l cv,1 ,l cv,2 ,…,l cv,c ) indicates the calculation of HF CV,k The variance of the mask reconstruction error of all elements in , λ2 is the regularization coefficient of loss2;

[0132] S405: Build decoder 2, and convert the decoder described in step S403 Input to decoder 2 to reconstruct the feature segment HF described in step S402 EIS,h , the reconstruction result is obtained as The specific expressions are as follows:

[0133]

[0134] Wherein, Decoder2(·) represents decoder 2;

[0135] According to HF EIS,h The original value y of the bth element in eis,b and The bth reconstruction value in Calculate HF EIS,h The mask reconstruction error l of the bth element in eis,b , specifically expressed as follows:

[0136]

[0137] Construct the mask reconstruction loss loss3 of EIS feature to describe the feature fragment HF EIS,h The mask reconstruction error and error stability of all elements in are specifically expressed as follows:

[0138]

[0139] Among them, Var(l eis,1 ,l eis,2 ,…,l eis,c ) indicates the calculation of HF EIS,h The variance of the mask reconstruction error of all elements in , λ3 is the regularization coefficient of loss3;

[0140] S406: for the constant voltage charging current data after data cleaning in step S2, calculate the mean, standard deviation, kurtosis and skewness of the constant voltage charging current data in each cycle, and for the electrochemical impedance spectroscopy data after data cleaning in step S2, calculate the real part mean, imaginary part mean, covariance between the real part and the imaginary part, and dot product of the real part and the imaginary part of the electrochemical impedance spectroscopy data corresponding to each cycle, to form 8 artificial features;

[0141] S407: Calculate the Pearson correlation coefficient and the Spearman correlation coefficient between the eight artificial features described in step S406 and the battery SOH, and calculate the aging correlation score Z of the vth artificial feature by combining the Softmax function. v , the specific calculation process is as follows:

[0142]

[0143] Among them, Softmax(·) is the Softmax function, γ v represents the Pearson correlation coefficient between the vth artificial feature and SOH, ρ v represents the Spearman correlation coefficient between the vth artificial feature and SOH;

[0144] S408: Add Z1 to Z4 calculated in step S407 to obtain the weight α of loss2 described in S404, add Z5 to Z8 calculated in step S407 to obtain the weight β of loss3 described in S405, linearly weight loss2 and loss3 according to their respective weights to obtain the joint loss loss4, as follows:

[0145] loss4=αloss2+βloss3 (20)

[0146] S409: Use the feature set F described in step S3 to train encoder 3, decoder 1 and decoder 2, use loss4 as the loss function of training, iteratively update the weight parameters of encoder 3, decoder 1 and decoder 2 through the back propagation algorithm, thereby completing the training of encoder 3, decoder 1 and decoder 2, and save the weight parameters after the training is completed.

Claims

1. A lithium battery SOH estimation method based on pre-training fusion of multi-source data, characterized in that: The specific steps are as follows: S1: Data collection of lithium battery aging experiment, i.e., a cyclic charge and discharge experiment is performed on multiple lithium batteries in a constant current-constant voltage charging and constant current discharging mode. In each charge and discharge cycle, constant voltage charging current data is continuously collected in the constant voltage charging stage. After charging, the lithium battery is left to stand for 15 minutes and then the electrochemical impedance spectrum data in the frequency range of 0.02Hz-20kHz is collected. After discharging, the SOH value of the lithium battery in this cycle is calculated. S2: Data preprocessing and data set construction, that is, data cleaning of the constant voltage charging current data and electrochemical impedance spectrum data described in step S1, and calculation of differential current data based on the constant voltage charging current data. The specific calculation method is as follows: of n =I n+1 -I n Among them, dI n Represents the differential current data at the nth time sampling point, I n Represents the constant voltage charging current data at the nth time sampling point; According to the constant voltage charging current data and differential current data, the current matrix X is constructed. CV , as follows: Wherein, p is the total number of constant voltage charging current data collected; According to the real and imaginary parts of all impedances in the electrochemical impedance spectroscopy data, the electrochemical impedance spectroscopy matrix X is constructed. EIS , as follows: Among them, Re m and Im m They represent the real and imaginary parts of the impedance data corresponding to the mth frequency sampling point, respectively, and u is the total number of impedance data collected; The current matrix X CV Divide it into N sub-matrices evenly according to the time dimension, then the current matrix X CV It can be expressed as: Among them, X CV,i represents the i-th current submatrix, whose dimension is L cv ×2, L cv is the number of sampling points corresponding to the sub-matrix; The electrochemical impedance spectroscopy matrix X EIS According to the frequency dimension, it is evenly divided into M sub-matrices, then the electrochemical impedance spectroscopy matrix X EIS It can be expressed as: Among them, X EIS,j Represents the jth electrochemical impedance spectroscopy submatrix, whose dimension is L eis ×2, L eis is the number of sampling points corresponding to the sub-matrix; The above sub-matrix-divided current matrix and electrochemical impedance spectrum matrix are defined as and Will and Constructing an unlabeled dataset Combined with the SOH described in step S1 to form a labeled data set S3: Construct pre-training task 1, that is, build encoder 1 and encoder 2, and convert the Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the step S2 described Input to encoder 2, encoder 2 outputs EIS characteristic HF EIS , calculate HF CV Each characteristic fragment is related to HF EIS The cosine similarity between each feature segment in the pre-training task 1 is calculated, and a loss loss1 that takes all cosine similarities into consideration is constructed as the loss function of the pre-training task 1. The unlabeled dataset D1 described in step S2 is used to train encoder 1 and encoder 2. After the training is completed, the weight parameters of encoder 1 and encoder 2 are saved, and the unlabeled dataset D1 is input into the trained encoder 1 and encoder 2 to obtain the feature set. S4: Construct pre-training task 2, that is, the feature HF described in step S3 CV and HF EIS In each case, a feature segment is randomly selected for masking, and the two masked features are concatenated and then input into the encoder 3 based on the self-attention mechanism. The encoder 3 outputs and The fusion feature matrix is It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current feature aging information. Decoder 1 and decoder 2 are built to Input to decoder 1, HF CV The masked fragments are reconstructed. Input to decoder 2, HF EIS The masked segments are reconstructed to construct the mask reconstruction loss loss2 of the constant voltage current feature and the mask reconstruction loss loss3 of the EIS feature. According to the correlation score between the constant voltage charging current data, the electrochemical impedance spectroscopy data and the SOH after data cleaning described in step S2, loss2 and loss3 are weighted to obtain loss4. Loss4 is used as the loss function of the pre-training task 2. The feature set F described in step S3 is used to train the encoder 3, the decoder 1 and the decoder 2. After the training is completed, the weight parameters of the encoder 3, the decoder 1 and the decoder 2 are saved. S5: Construct and train a SOH estimation model, which includes an encoder module and an estimator, that is, the encoder 1 and encoder 2 described in step S3 and the encoder 3 described in step S4 together constitute an encoder module, and the weight parameters of the encoder 1, encoder 2 and encoder 3 obtained in step S3 and step S4 are frozen, and the estimator to be trained is built behind the encoder module to establish the estimator described in step S4. and The mapping relationship between the fusion feature matrix and the SOH value is formed, the labeled data set D2 described in step S2 is divided into a training set and a test set, and the SOH estimation model including the encoder module and the estimator is trained and tested. If the test result meets the accuracy requirement, the SOH estimation model is saved; S6: Online estimation of the SOH of the lithium battery, that is, in the actual charge and discharge cycle of the lithium battery, constant voltage charging current data and electrochemical impedance spectrum data are collected, and after data preprocessing according to the method of step S2, the constant voltage charging current data and electrochemical impedance spectrum data are input into the SOH estimation model described in step S5 to output the SOH estimation value.

2. According to claim 1, a lithium battery SOH estimation method based on pre-training fusion of multi-source data is characterized in that: The specific steps of constructing the pre-training task 1 described in step S3 are as follows: S301: Build encoder 1, and convert the encoder described in step S2 Input to encoder 1, encoder 1 outputs constant voltage current characteristic HF CV , the encoder 1 will Linearly project the N sub-matrices in to obtain the corresponding N constant voltage and current feature fragments with a dimension of 1×c, as follows: Among them, Encoder1(·) represents encoder 1, HF CV,i represents the i-th constant voltage and current feature segment, whose dimension is 1×c, HF CV The dimension is N×c; S302: Build encoder 2, and convert the encoder described in step S2 Input to encoder 2, encoder 2 outputs EIS characteristic HF EIS , the encoder 2 will Linear projection is performed on the M sub-matrices in to obtain the corresponding M EIS characteristic fragments with a dimension of 1×c, as follows: Among them, Encoder2(·) represents encoder 2, HF EIS,j represents the jth EIS characteristic fragment, whose dimension is 1×c, HF EIS The dimension is M×c; S303: Calculate the HF in step S301 CV Each characteristic segment in step S302 is consistent with the HF EIS The cosine similarity between each feature fragment in is as follows: Among them, CS ij Indicates HF CV,i and HF EIS,j The cosine similarity between CV,i ·HF EIS,j represents the dot product of two feature fragments, ‖·‖ represents the magnitude of the vector; S304: Construct a loss1 that takes into account all cosine similarities, as follows: CL ij =1-CS ij Q=Var({CL ij |i=1,2,…,N,j=1,2,…,M}) Among them, CL ij Indicates HF CV,i and HF EIS,j The cosine similarity loss between them, Var(·) represents the variance calculation function, Q represents the variance of all cosine similarity losses, and λ1 is the regularization coefficient of loss1; S305: Use the unlabeled data set D1 described in step S2 to train encoder 1 and encoder 2, use loss1 as the loss function of the training, iteratively update the weight parameters of encoder 1 and encoder 2 through the back propagation algorithm, so as to complete the training of encoder 1 and encoder 2, save the weight parameters after the training is completed, input the unlabeled data set D1 into the encoder 1 and encoder 2 after the training is completed, and obtain the feature set 3. The lithium battery SOH estimation method based on pre-training fusion of multi-source data according to claim 1 is characterized in that: The specific steps of constructing the pre-training task 2 described in step S4 are as follows: S401: Constant voltage current characteristic HF described in step S3 CV In the process, a feature segment is randomly selected for masking operation, that is, for each charging cycle sample, the kth constant voltage current feature segment HF is randomly masked. CV,k All c elements in are set to 0, and the masked constant voltage and current characteristics are obtained. The dimension is N×c; S402: EIS characteristic HF described in step S3 EIS In the process, a characteristic segment is randomly selected for masking operation, that is, for each charging cycle sample, the hth EIS characteristic segment HF is randomly masked. EIS,h All c elements in are set to 0 to obtain the masked EIS feature The dimension is M×c; S403: Build an encoder 3 based on the self-attention mechanism. and After being spliced ​​into a matrix of dimension (N+M)×c, it is input into encoder 3, and the connection between all feature fragments is learned through the self-attention mechanism, thereby outputting a fused feature matrix. The fused feature matrix is ​​composed of and Composition, of which It is a constant voltage current feature that integrates the EIS characteristic aging information. It is an EIS feature that combines the constant voltage and current characteristic aging information, which is specifically expressed as follows: Among them, concat(·) represents the concatenation function, Encoder3(·) represents the encoder 3 based on the self-attention mechanism, and its output and The dimensions are N×c and M×c respectively; S404: Build decoder 1, and convert the decoder described in step S403 Input decoder 1 to reconstruct the feature segment HF described in step S401 CV,k , the reconstruction result is obtained as The specific expressions are as follows: Wherein, Decoder1(·) represents decoder 1; According to HF CV,k The original value y of the ath element in cv,a and The ath reconstruction value in Calculate HF CV,k The mask reconstruction error l of the a-th element in cv,a , specifically expressed as follows: Construct the mask reconstruction loss loss2 of the constant voltage and current feature to describe the feature segment HF CV,k The mask reconstruction error and error stability of all elements in are specifically expressed as follows: Among them, Var(l cv,1 ,l cv,2 ,…,l cv,c ) indicates the calculation of HF CV,k The variance of the mask reconstruction error of all elements in , λ2 is the regularization coefficient of loss2; S405: Build decoder 2, and convert the decoder described in step S403 Input to decoder 2 to reconstruct the feature segment HF described in step S402 EIS,h , the reconstruction result is obtained as The specific expressions are as follows: Wherein, Decoder2(·) represents decoder 2; According to HF EIS,h The original value y of the bth element in eis,b and The bth reconstruction value in Calculate HF EIS,h The mask reconstruction error l of the bth element in eis,b , specifically expressed as follows: Construct the mask reconstruction loss loss3 of EIS feature to describe the feature fragment HF EIS,h The mask reconstruction error and error stability of all elements in are specifically expressed as follows: Among them, Var(l eis,1 ,l eis,2 ,…,l eis,c ) indicates the calculation of HF EIS,h The variance of the mask reconstruction error of all elements in , λ3 is the regularization coefficient of loss3; S406: for the constant voltage charging current data after data cleaning in step S2, calculate the mean, standard deviation, kurtosis and skewness of the constant voltage charging current data in each cycle, and for the electrochemical impedance spectroscopy data after data cleaning in step S2, calculate the real part mean, imaginary part mean, covariance between the real part and the imaginary part, and dot product of the real part and the imaginary part of the electrochemical impedance spectroscopy data corresponding to each cycle, to form 8 artificial features; S407: Calculate the Pearson correlation coefficient and the Spearman correlation coefficient between the eight artificial features described in step S406 and the battery SOH, and calculate the aging correlation score Z of the vth artificial feature in combination with the Softmax function. v , the specific calculation process is as follows: Among them, Softmax(·) is the Softmax function, γ v represents the Pearson correlation coefficient between the vth artificial feature and SOH, ρ v represents the Spearman correlation coefficient between the vth artificial feature and SOH; S408: Add Z1 to Z4 calculated in step S407 to obtain the weight α of loss2 described in S404, add Z5 to Z8 calculated in step S407 to obtain the weight β of loss3 described in S405, linearly weight loss2 and loss3 according to their respective weights to obtain the joint loss loss4, as follows: loss4 = αloss2 + βloss3 S409: Use the feature set F described in step S3 to train encoder 3, decoder 1 and decoder 2, use loss4 as the loss function of training, iteratively update the weight parameters of encoder 3, decoder 1 and decoder 2 through the back propagation algorithm, thereby completing the training of encoder 3, decoder 1 and decoder 2, and save the weight parameters after the training is completed.

Images