Method for estimating health state of lithium ion battery

Abstract

The invention relates to a method for estimating the health state of a lithium ion battery. The method comprises the following operation steps: collecting fragment information and health state information of a random state-of-charge interval of a vehicle-mounted lithium ion battery in each cycle of constant-current charging stage; and training a variational auto-encoder by adopting fragment information of each cycle of the lithium ion battery, and reconstructing and restoring complete information of the cycle. And extracting a curve peak value of complete information of each cycle as an aging characteristic by using incremental capacity analysis and incremental energy analysis. And optimizing hyper-parameters of the bidirectional long-short-term memory neural network by using a grey wolf optimization algorithm to construct a mapping relation between the optimal feature and the lithium ion battery health state, and obtaining a Bi-LSTM health estimation model. And sending the aging information of the lithium ion battery subjected to feature extraction into the Bi-LSTM health estimation model to obtain the health state of the cycle number to be estimated. According to the scheme provided by the invention, the lithium ion battery health state can be accurately, stably and efficiently estimated.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium-ion batteries, and particularly relates to a method for estimating the health state of a lithium-ion battery. Background Art

[0002] Accurate estimation of the health state of lithium-ion batteries is of great significance for the safe and stable operation of electric vehicles. Utilizing fragmentary data is a common solution for estimating the health state of lithium-ion batteries in the actual scenarios of current electric vehicles. For example, the technical solutions described in the already published patent documents "A Method for Estimating the Health State of a Lithium-Ion Battery Based on Fragmentary Charging Data" (Application No.: 202311860052.1), "A Method for Estimating the Available Capacity of a Lithium-Ion Battery Based on Optimized Support Vector Regression" (Application No.: 202311054327.2), etc. Although these solutions can achieve the estimation of the health state of lithium-ion batteries, due to the lack of information, fragmentary data usually cannot accurately characterize the problems of the aging process. The existing fragmentary data estimation solutions weaken the expression of electrical aging information due to their fragmentary acquisition methods and have no means to suppress noise, reducing the estimation accuracy of the existing models. Therefore, it is necessary to develop a method for accurately, stably, and efficiently estimating the health state of lithium-ion batteries using fragmentary data. Summary of the Invention

[0003] The object of the present invention is to provide a method for estimating the health state of a lithium-ion battery, which can improve the accuracy and stability of the health state estimation of the lithium-ion battery.

[0004] The present invention is specifically implemented according to the technical solution described below.

[0005] A method for estimating the health state of a lithium-ion battery, characterized by comprising the following operating steps:

[0006] S10: Collect fragmentary information and health state information in a random state of charge interval during the constant current charging stage of each cycle of the in-vehicle lithium-ion battery.

[0007] S20: Train a variational autoencoder using the fragmentary information of each cycle of the lithium-ion battery and reconstruct and restore the complete information of this cycle.

[0008] S30: Use incremental capacity analysis and incremental energy analysis to extract the curve peaks of the complete information of each cycle as aging characteristics.

[0009] S40: Construct the mapping relationship between the optimal features and the health state of the lithium-ion battery to obtain the Bi-LSTM health estimation model.

[0010] S50: Send the aging information obtained by feature extraction of the lithium-ion battery into the Bi-LSTM health estimation model to obtain the health state of the cycle number to be estimated.

[0011] Specifically, the fragment information includes voltage, capacity, and energy.

[0012] Use the Grey Wolf Optimization algorithm to optimize the hyperparameters of the Bidirectional Long Short-Term Memory neural network to construct the mapping relationship between the optimal features and the health state of the lithium-ion battery.

[0013] In step S30:

[0014] The calculation method of taking the peak value of the curve of the complete information per cycle as the aging feature in incremental capacity analysis and incremental energy analysis is:

[0015]

[0016] where m is the serial number of the voltage, capacity, and energy sequences, and 1 ≤ m ≤ v - 1;

[0017] The specific calculation method for extracting the peak values of the IC and IE curves is as follows:

[0018] IC i,Max = Max(IC i,m )

[0019] IE i,Max = Max(IE i,m )

[0020] where Max(IC i,m ) and Max(IE i,m ) represent taking the maximum value of the sequences IC i and IE i at the i-th cycle.

[0021] In step S40:

[0022] The specific calculation method of the Bi-LSTM health estimation model is as follows:

[0023] f t = σ(w f [h t-1 ,x t +b f )

[0024] e t = σ(w e [h t-1 ,x t +b e )

[0025] o t = σ(w o[h t-1 , x t + b o )

[0026]

[0027]

[0028] h t = o t ⊙ tanh(C t )

[0029] Among them, w and b are the weights and biases corresponding to the gate respectively; x t , h t-1 are the input at time t and the output of the previous LSTM at the previous moment respectively; C is the cell state at different moments; e t is the output of the input gate at time t, f t is the output of the forget gate at time t, o t is the output of the output gate at time t, C t is the cell state at time t;

[0030] The Bi-LSTM health estimation model is composed of forward and backward LSTM networks connected to the same input layer:

[0031]

[0032] Among them, is the weight of the forward LSTM hidden layer; is the weight of the backward LSTM hidden layer; h t is the result after the linear superposition of the forward and backward hidden layer states; g t is the input of the input LSTM, and f(·) is the calculation process of the LSTM.

[0033] The operations of optimizing the hyperparameters of the bidirectional long short-term memory neural network using the grey wolf optimization algorithm in step S40 include the following:

[0034] S41: Initialize the grey wolf population, and the convergence factor Track, chase, and approach the prey:

[0035]

[0036] Among them, represents the distance between the individual and the prey, and represent the position vectors of the grey wolf and the prey respectively, is the convergence factor, is a random number with a modulus length between 0 and 1;

[0037] S42: Calculate the fitness of the gray wolf individuals, and save the top three wolves α, β, and δ with the best fitness. Wolves α, β, and δ hunt, surround, and harass the prey until it stops moving:

[0038]

[0039] Among them, and represent the distances between α, β, δ and other individuals and the current positions respectively; is a random vector; is the position of the current gray wolf;

[0040] S43: Update the current position of the gray wolf:

[0041]

[0042] Among them, represents the position after other wolves in the wolf pack move towards α, β, and δ, represents the final position of the wolf pack;

[0043] S44: If the maximum number of iterations is reached, stop. If the maximum number of loops is not reached, jump to step S41 for execution.

[0044] In step S40, the first u of the obtained battery aging data are used as training data, and the parameter optimization process for constructing the best mapping model is as follows:

[0045] Take the first u sample values of the input sequence and the health status sequence and as the training set;

[0046] Input the sample values into the Bi-LSTM health estimation model to obtain

[0047] Optimize the mean square error loss function through backpropagation: Make the network parameters parm reach the best combination.

[0048] The specific operation in step S50 is as follows:

[0049] Input the test set into the HAMN network with trained parameters to obtain the predicted value

[0050] In step S10, record the lithium-ion battery health status data of each charge-discharge cycle as sequence H 1 , H 2 ,..., H total ;

[0051]

[0052] Among them, H i is the battery health state of the i-th (i = 1, 2, …, n) charge-discharge cycle, total is the number of charge-discharge cycles, C i is the maximum discharge capacity of the lithium-ion battery in the i-th charge-discharge cycle, and C is the rated capacity of the lithium-ion battery;

[0053] The charging voltage data in the random state of charge interval of the i-th charge-discharge cycle is the sequence U i,k×d , U i,(k+1)×d , …, U i,q×d , and the charging capacity data is the sequence C i,k×d , C i,(k+1)×d , …, C i,q×d , and the charging energy data is the sequence E i,k×d , E i,(k+1)×d , …, E i,q×d ;

[0054] Among them, U i,k×d is the voltage at the initial sampling time of the i-th charge-discharge cycle, U i,q×d is the voltage at the termination sampling time of the i-th charge-discharge cycle, C i,k×d is the capacity at the initial sampling time of the i-th charge-discharge cycle, C i,q×d is the capacity at the termination sampling time of the i-th charge-discharge cycle, E i,k×d is the energy at the initial sampling time of the i-th charge-discharge cycle, E i,q×d is the energy at the termination sampling time of the i-th charge-discharge cycle, d is the data sampling interval time during charging, k and q are the initial and termination sampling times respectively, and (q - k) × d is the total sampling time.

[0055] In step S20: The variational autoencoder maps the true distribution p(x) of the data D in the source domain x = {U, C, E} to the prior latent distribution p (z) of the latent domain z through the encoder, and maps the samples back from the low-dimensional latent distribution points to the reconstructed high-dimensional data distribution using a set of shared parameters θ through the decoder p θ (x|z), and finally obtains the likelihood-maximized joint probability distribution through backpropagation: θ (x|z) uses a set of shared parameters θ to map the samples back from the low-dimensional latent distribution points to the reconstructed high-dimensional data distribution, and finally obtains the likelihood-maximized joint probability distribution through backpropagation:

[0056]

[0057] Among them, p θ (z) is a multivariate standard normal distribution;

[0058] After processing the fragment voltage, capacity, and energy information of each cycle by the variational autoencoder, the complete voltage, capacity, and energy sequences U i,1 , Ui,2 ,...,U i,v ,C i,1 ,C i,2 ,...,C i,v and E i,1 ,E i,2 ,...,E i,v , where v is the length of the reconstructed information sequence, and 1 ≤ k ≤ q ≤ v.

[0059] The above solution provided by the present invention uses a variational autoencoder to learn the data distribution of fragment charging aging information and reconstructs and generates the voltage, capacity, and energy information of a complete lithium battery charge. While suppressing the noise of the original data, new data that conforms to the aging characteristic distribution is generated, enabling the effective implementation of traditional feature engineering; a Bi-LSTM health estimation model is adopted to improve the overall attention to data while alleviating overfitting of traditional methods, and finally a high-performance SOH estimation learning machine is iteratively constructed; and the best Bi-LSTM hyperparameters are optimized through GWO, avoiding cumbersome manual adjustment of hyperparameters, thereby achieving accurate, stable, and efficient estimation of the health state of lithium-ion batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a schematic flow chart of the present invention.

[0061] Figure 2 is the SOC interval range of each cycle of the aging test of the No. 1 experimental battery.

[0062] Figure 3 is the data graph of the No. 1 experimental battery before reconstruction by the variational autoencoder.

[0063] Figure 4 is the data graph of the No. 1 experimental battery after reconstruction by the variational autoencoder.

[0064] Figure 5 is the distribution graph of the IC curves of each cycle of the No. 1 experimental battery.

[0065] Figure 6 is the distribution graph of the IE curves of each cycle of the No. 1 experimental battery.

[0066] Figure 7 is the peak feature graph of the IC / IE curves of the No. 1 experimental battery.

[0067] Figure 8 is the Bi-LSTM network structure diagram.

[0068] Figure 9 is the flow chart of the GWO algorithm.

[0069] Figure 10 is the stability verification of the method proposed by the present invention for health state estimation of the No. 1 battery.

[0070] Figure 11 It is a comparison chart of the health state estimation effects of each method and the method proposed in the present invention.

[0071] Figure 12 It is a partial enlarged view of the last 50% cycles of the comparison chart of the health state estimation effects of each method and the method proposed in the present invention.

[0072] Figure 13 It is a comparison chart of the health state estimation effects on the Bi-LSTM health estimation model using the complete data set and the fragmented data set reconstructed by the present invention.

[0073] Figure 14 It is a partial enlarged view of the last 50% cycles of the comparison chart of the health state estimation effects on the Bi-LSTM health estimation model using the complete data set and the fragmented data set reconstructed by the present invention. Detailed implementation manners

[0074] In order to make the purpose and advantages of the present invention clearer, the present invention will be specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific implementation manners of the present invention, and does not strictly limit the scope of protection specifically claimed by the present invention.

[0075] As used herein, terms such as "parallel", "perpendicular", etc. are not limited to their strict geometric definitions, but include tolerances for machining or human errors that are reasonable and inconsistent.

[0076] The following specifically describes the operation with Battery No. 1 as an example. Taking 50% of the data as the training set and 50% of the data as the test set, a health state estimation experiment is carried out.

[0077] A method for estimating the health state of a lithium-ion battery is specifically as Figure 1 shown, and includes the following operation steps.

[0078] First, collect the fragment information and health state information of the random state of charge (SOC) interval in the constant current charging stage of the on-vehicle lithium-ion battery for each cycle, as Figure 2 shown, and the fragment information includes voltage, capacity, and energy.

[0079] Specifically, record the health state data of the lithium-ion battery for each charge-discharge cycle as sequence H 1 ,H 2 ,...,H total ;

[0080]

[0081] Among them, H iis the state of health of the battery for the i-th (i = 1, 2, K, n) charge and discharge cycle, total is the number of charge and discharge cycles, C i is the maximum discharge capacity of the lithium-ion battery for the i-th charge and discharge cycle, C is the rated capacity of the lithium-ion battery;

[0082] The charging voltage data of the random state of charge interval for the i-th charge and discharge cycle is the sequence U i,k×d , U i,(k+1)×d ,..., U i,q×d , and the charging capacity data is the sequence C i,k×d , C i,(k+1)×d ,..., C i,q×d , and the charging energy data is the sequence E i,k×d , E i,(k+1)×d ,..., E i,q×d ;

[0083] Among them, U i,k×d is the voltage at the initial sampling time of the i-th charge and discharge cycle, U i,q×d is the voltage at the termination sampling time of the i-th charge and discharge cycle, C i,k×d is the capacity at the initial sampling time of the i-th charge and discharge cycle, C i,q×d is the capacity at the termination sampling time of the i-th charge and discharge cycle, E i,k×d is the energy at the initial sampling time of the i-th charge and discharge cycle, E i,q×d is the energy at the termination sampling time of the i-th charge and discharge cycle, d is the data sampling interval time during charging, k and q are the initial and termination sampling times respectively, and (q - k) × d is the total sampling time.

[0084] Second, use the segment information of each cycle of the lithium-ion battery to train the variational autoencoder and reconstruct and restore the complete information of this cycle.

[0085] The variational autoencoder passes the encoder to map the true distribution p(x) of the data D in the source domain x = {U, C, E} to the prior latent distribution p θ (z) of the latent domain z, and uses a set of shared parameters θ through the decoder p θ (x|z) to map the samples from the low-dimensional latent distribution points back to the reconstructed high-dimensional data distribution, and finally obtains the likelihood-maximized joint probability distribution through backpropagation:

[0086] argmax p θ (x, z) = argmin p θ (x|z) p θ (z)

[0087] Among them, p θ (z) is a multivariate standard normal distribution; since the latent posterior probability pθ (x|z) is a problem without an analytical solution, so variational distributions are usually used to approximate the posterior, that is, the problem is transformed into minimizing the KL divergence between the true distribution p(x) and the latent distribution p θ (z):

[0088]

[0089] After processing the fragment voltage, capacity, and energy information of each cycle through a variational autoencoder, the complete voltage, capacity, and energy sequences U i,1 , U i,2 ,..., U i,v , C i,1 , C i,2 ,..., C i,v and E i,1 , E i,2 ,..., E i,v are obtained, where v is the length of the reconstructed information sequence, and 1 ≤ k ≤ q ≤ v. The data results before and after reconstruction are as Figure 3 、 Figure 4 shown.

[0090] Third, use incremental capacity analysis (ICA) and incremental energy analysis (IEA) to extract the curve peaks of the complete information of each cycle as aging characteristics.

[0091] Using the difference method to approximately simulate the differential operation, processing the discrete voltage, capacity, and energy sequences, the specific calculation method for using ICA and IEA to extract the curve peaks of the complete information of each cycle as aging characteristics is as follows:[[]]

[0092]

[0093] where m is the number of the voltage, capacity, and energy sequences, and 1 ≤ m ≤ v - 1; the IC and IE curves of the No. 1 battery obtained are respectively as Figure 5 , Figure 6 shown.

[0094] The specific calculation method for extracting the peaks of the IC and IE curves is as follows:[[]]

[0095] IC i,Max = Max(IC i,m )

[0096] IE i,Max = Max(IE i,m )

[0097] where Max(IC i,m ) and Max(IE i,m ) represent taking the maximum value of the sequence IC i and IE at the i-th cyclei The maximum value. The battery feature extraction results are as Figure 7 shown.

[0098] Bi-LSTM is composed of LSTM. Compared with the traditional recurrent neural network (RNN), LSTM introduces the input gate $e_t$, forget gate $f_t$, output gate $o_t$ and cell state $C_t$. These mechanisms enable LSTM to better handle the long-term dependencies in the sequence. The specific calculation method of the Bi-LSTM health estimation model is as follows:

[0099] f t = σ(w f [h t-1 , x t + b f )

[0100] e t = σ(w e [h t-1 , x t + b e )

[0101] o t = σ(w o [h t-1 , x t + b o )

[0102]

[0103]

[0104] h t = o t ⊙ tanh(C t )

[0105] Among them, w and b are the weights and biases of the corresponding gates respectively; x t , h t-1 are the input at time t and the output of the previous LSTM respectively; C is the cell state at different times; e t is the output of the input gate at time t, f t is the output of the forget gate at time t, o t is the output of the output gate at time t, C t is the cell state at time t;

[0106] The Bi-LSTM health estimation model is composed of forward and backward LSTM networks connected to the same input layer:

[0107]

[0108] Among them, is the weight of the forward LSTM hidden layer; is the weight of the backward LSTM hidden layer; h t is the result after the linear superposition output of the forward and backward hidden layer states; g t is the input to the input LSTM, and f(·) is the calculation process of the LSTM. The structure diagram of the Bi-LSTM health estimation model is as Figure 8 shown.

[0109] Fourth, use the Grey Wolf Optimization Algorithm (GWO) to optimize the hyperparameters of the Bidirectional Long Short-Term Memory Neural Network (Bi-LSTM) to construct the best mapping relationship between features and the health state of lithium-ion batteries, and obtain the Bi-LSTM health estimation model. As Figure 9 shown, it includes the following operations:

[0110] (1): Initialize the grey wolf population (hyperparameters of the Bi-LSTM health estimation model), and the convergence factor Track, chase, and approach the prey:

[0111]

[0112] Among them, represents the distance between an individual and the prey, and represent the position vectors of the grey wolf and the prey respectively, is the convergence factor, is a random number with a modulus length between 0 and 1.

[0113] (2): Calculate the fitness of the grey wolf individuals, and save the top three wolves α, β, δ with the best fitness. Wolves α, β, δ chase, surround, and harass the prey until it stops moving:

[0114]

[0115] Among them, and represent the distances between α, β, δ and other individuals and the current positions respectively; is a random vector; is the position of the current grey wolf;

[0116] (3): Update the current position of the grey wolf:

[0117]

[0118] Among them, represents the position after other wolves in the wolf pack move towards α, β, δ, represents the final position of the wolf pack;

[0119] (4): Stop if the maximum number of iterations is reached; otherwise, jump to step (1) and execute.

[0120] Take the first u of the obtained battery aging data as training data. The parameter optimization process for constructing the best mapping model is as follows:

[0121] Take the first u sample values of the input sequence and the health status sequence and as the training set;

[0122] Input the sample values into the Bi-LSTM health estimation model to obtain

[0123] Optimize the mean square error loss function through backpropagation: Make the network parameters parm reach the best combination.

[0124] Fifth, send the aging information obtained by feature extraction of the lithium-ion battery into the Bi-LSTM health estimation model to obtain the health status of the number of cycles to be estimated. The specific operation is as follows:

[0125] Input the test set into the HAMN network with trained parameters to obtain the predicted value

[0126] As Figure 10 shown, it is the model test result of Battery No. 1 and its stable confidence interval for 100 tests at the 99.7% confidence level. It can be seen that the proposed SOH estimation algorithm has excellent lithium-ion battery health status estimation ability.

[0127] To evaluate the performance of the proposed algorithm for battery health status estimation, the present invention compares various methods applied to the regression model: the machine learning model support vector regression (SVR), the sequence prediction model long short-term memory neural network (LSTM), and the gated recurrent unit (GRU). Figure 11 For the prediction effects of various models on Battery No. 1, Figure 12 For Figure 11 the local enlarged view of. The regression evaluation indexes of each model are shown in Table 1.

[0128] Table 1 Regression evaluation indexes of each model for Battery No. 1

[0129]

[0130] It can be concluded from Table 1 that the R of the present invention 2It is 0.9392, the highest among the four models, indicating that the Bi-LSTM model fits the shape best; the MAE and RMSE of the model of the present invention are 0.2927% and 0.0034% respectively, both of which are the lowest among all models, indicating that the model proposed by the present invention has better stability. To verify the performance loss after reconstruction, the present invention compares the complete data with the data reconstructed by the present invention, and the results are as Figure 13 , Figure 14 shown. The regression evaluation indexes are shown in Table 2.

[0131] Table 2 Regression evaluation indexes of the segment / complete data set under the condition of Bi-LSTM

[0132]

[0133] It can be seen from Table 2 that the SOH estimation effect by the segment data reduction method of the present invention does not decrease significantly compared with the estimation effect using the complete data. Among them, R2 only decreases by 2.81%, and the values of MAE and RMSE only increase by 0.0802% and 0.1052% respectively. It shows that the reconstruction effect of the present invention is good.

[0134] The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention. The structures, devices and operation methods not specifically described and explained in the present invention, unless otherwise specified and limited, are implemented according to the conventional means in the art.

Claims

1. A method for estimating the health status of a lithium-ion battery, characterized in that: The steps are as follows: S10: Collecting fragment information and health status information of random state of charge intervals of the on-board lithium-ion battery during each cycle of constant current charging; S20: Using the fragment information of each cycle of the lithium-ion battery to train the variational autoencoder, and reconstructing and restoring the complete information of this cycle; S30: Using incremental capacity analysis and incremental energy analysis to extract the curve peak of each cycle complete information as the aging feature; S40: constructing a mapping relationship between the best features and the health status of the lithium-ion battery, and obtaining a Bi-LSTM health estimation model; S50: The aging information of the lithium-ion battery obtained by feature extraction is sent to the Bi-LSTM health estimation model to obtain the health status of the cycle number to be estimated.

2. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: The fragment information includes voltage, capacity, and energy.

3. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: The Grey Wolf Optimization Algorithm is used to optimize the hyperparameters of the bidirectional long short-term memory neural network to construct the best mapping relationship between features and the health status of lithium-ion batteries.

4. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S30: The calculation method for extracting the curve peak of each cycle complete information in incremental capacity analysis and incremental energy analysis as the aging characteristic is: Wherein, m is the number of the voltage, capacity, and energy sequence, 1≤m≤v-1; The specific calculation method for extracting the peak values ​​of IC and IE curves is as follows: IC i,Max =Max(IC i,m ) IE i,Max =Max(IE i,m ) Where Max(IC i,m ) and Max(IE i,m ) means taking the sequence IC in the i-th cycle i and IE i The maximum value of .

5. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S40: The specific calculation method of the Bi-LSTM health estimation model is as follows: f t =σ(w f [h t-1 ,x t ]+b f ) e t =σ(w e [h t-1 ,x t ]+b e ) the t =σ(w o [h t-1 ,x t ]+b o ) h t =o t ⊙tanh(C t ) Among them, w, b are the weight and bias of the corresponding gate respectively; x t ,h t-1 are the input at time t and the output of LSTM at the previous time; C is the cell state at different times; e t is the output of the input gate at time t, f t is the output of the forget gate at time t, o t is the output of the output gate at time t, C t is the cell state at time t; The Bi-LSTM health estimation model consists of two LSTM networks connected to the same input layer: in, is the weight of the forward LSTM hidden layer; is the weight of the backward LSTM hidden layer; h t is the result of linearly superimposing the output to the hidden layer states; g t is the input of LSTM, and f(·) is the calculation process of LSTM.

6. The method for estimating the health status of a lithium-ion battery according to claim 5, characterized in that: In step S40, the hyperparameters of the bidirectional long short-term memory neural network are optimized by using the gray wolf optimization algorithm, including the following operations: S41: Initialize the gray wolf population, convergence factor Tracking, chasing and approaching prey: in, represents the distance between the individual and the prey, and Represent the position vectors of the gray wolf and prey respectively, is the convergence factor, is a random number with a modulus between 0 and 1; S42: Calculate the fitness of individual gray wolves and save the top three wolves with the best fitness, α, β, δ. α, β, δ wolves hunt, surround and harass the prey until it stops moving: in, and Represent the distance and current position between α, β, δ and other individuals respectively; is a random vector; is the current position of the gray wolf; S43: Update the current position of the Gray Wolf: in, Represents the positions of other wolves in the wolf pack after they move towards α, β, and δ. represents the final position of the wolf pack; S44: If the maximum number of iterations is reached, stop; if the maximum number of cycles is not reached, jump to step S41 for execution.

7. The method for estimating the health status of a lithium-ion battery according to claim 6, characterized in that: In step S40, the first u pieces of battery aging data obtained are used as training data, and the parameter optimization process for constructing the best mapping model is as follows: Take the first u sample values ​​of the input sequence and health status sequence and As a training set; Input sample values ​​into the Bi-LSTM health estimation model to obtain Optimize the mean square error loss function through back propagation: Make the network parameter parm reach the best combination.

8. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: The specific operations in step S50 are: The test set Input the HAMN network with trained parameters to get the predicted value 9. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S10: The health status data of the lithium-ion battery in each charge and discharge cycle is recorded as a sequence H1, H2, ..., H total ; Among them, H i is the battery health status of the i-th (i=1,2,K,n) charge-discharge cycle, total is the number of charge-discharge cycles, C i is the maximum discharge capacity of the lithium-ion battery in the i-th charge and discharge cycle, and C is the rated capacity of the lithium-ion battery; The charging voltage data of the random state of charge interval of the i-th charge and discharge cycle is the sequence U i,k×d ,U i,(k+1)×d ,...,U i,q×d , the charging capacity data is sequence C i,k×d ,C i,(k+1)×d ,...,C i,q×d , the charging energy data is sequence E i,k×d ,E i,(k+1)×d ,...,E i,q×d ; Among them, U i,k×d is the initial sampling time voltage of the ith charge and discharge cycle, U i,q×d is the voltage at the end sampling time of the i-th charge and discharge cycle, C i,k×d is the initial sampling time capacity of the ith charge and discharge cycle, C i,q×d is the end sampling time capacity of the ith charge and discharge cycle, E i,k×d is the initial sampling time energy of the ith charge and discharge cycle, E i,q×d is the termination sampling time energy of the ith charge-discharge cycle, d is the data sampling interval during charging, k and q are the initial and termination sampling moments respectively, and (qk)×d is the total sampling time.

10. The method for estimating the health status of a lithium-ion battery according to claim 1, characterized in that: In step S20: the variational autoencoder passes through the encoder Map the true distribution p(x) of the data D in the source domain x = {U, C, E} to the prior potential distribution p of the hidden domain z θ (z), through the decoder p θ (x|z) uses a set of shared parameters θ to map samples from low-dimensional potential distribution points back to the reconstructed high-dimensional data distribution, and finally obtains the likelihood-maximized joint probability distribution through back propagation: argmaxp θ (x,z)=argminp θ (x|z)p θ (z) Among them, p θ (z) is a multivariate standard normal distribution; After the variational autoencoder processes the segment voltage, capacity, and energy information of each cycle, the complete voltage, capacity, and energy sequence U is finally obtained. i,1 ,U i,2 ,...,U i,v , C i,1 ,C i,2 ,...,C i,v and E i,1 ,E i,2 ,...,E i,v , where v is the length of the reconstructed information sequence, 1≤k≤q≤v.

Images