Lithium ion battery health state estimation method based on feature optimization

Abstract

The invention discloses a lithium ion battery state of health (SOH) estimation method based on feature optimization, and the method comprises the following specific steps: obtaining the experimental data of a lithium ion battery, extracting initial health features reflecting the degradation of the battery, and carrying out the normalization processing of the initial health features; performing screening and dimension reduction on the preliminary health features by adopting a Spearman correlation coefficient and a grey correlation analysis method, and selecting six features with the highest SOH correlation degree; using an improved fully adaptive noise set empirical mode decomposition and sample entropy method to decompose and reconstruct health features, adopting t-SNE to perform transformation and dimension reduction on sample entropy features, using k-means + + clustering analysis, optimizing feature data and dividing the feature data into a training set and a test set; establishing a mapping relation between the optimized health features and the SOH through a bidirectional long short-term memory network (BiLSTM); a sparrow search algorithm (SCSSA) fusing sine, cosine and Cauchy variation is adopted to optimize BiLSTM parameters; and constructing an SOH estimation model based on SCSSA-BiLSTM, and obtaining a battery health state estimation value.

Description

Technical Field

[0001] The present invention belongs to the technical field of lithium-ion battery health state estimation, and relates to a lithium-ion battery health state estimation method based on feature optimization. Background Art

[0002] With the continuous development of battery technology, lithium-ion batteries have the advantages of high efficiency, light weight, high energy density and green environmental protection. They have been widely used in many fields such as electric vehicles and new energy storage. Accurate estimation of the state of health (SOH) of lithium-ion batteries is crucial to improving the reliability and safety of battery systems.

[0003] At present, there are four main methods for estimating SOH of lithium-ion batteries: model-based method, data-driven method, method based on the fusion of model and data-driven method, and method based on the fusion of multiple data-driven methods. Among them, the data-driven machine learning method has the advantages of not relying on expert knowledge, low computational load, and strong adaptability compared to the model-based method, so it has become the mainstream method for SOH estimation.

[0004] However, the current data-driven SOH estimation method still has some shortcomings. The current data-driven SOH estimation method has problems such as the extracted health features (HFs) are difficult to effectively and accurately reflect the capacity decay of the battery and the inaccurate selection of model parameters. In addition, the model is more susceptible to the redundancy of health features. In most studies, the reconstruction of health features is usually achieved by directly adding important IMFs, ignoring the weight differences and frequency and time information of different IMFs, resulting in the inability to distinguish changes on different time scales. Directly adding multiple IMFs will also cause the reconstructed signal to contain multiple frequency band information, increasing the complexity and being unfavorable for subsequent feature extraction and clustering. The selection of hyperparameters for most models is also a difficult problem. Patent (CN117471348A) discloses a method for estimating the health state of lithium-ion batteries based on the IPSO-GPR algorithm. The method is based on feature extraction based on discrete wavelet packet transform and Box-Cox transform, uses GPR to construct a mapping relationship between features and SOH, uses an improved particle swarm algorithm to optimize the kernel parameters of GPR, and establishes a SOH estimation model based on GPR to accurately and quickly realize the SOH estimation of lithium-ion batteries. However, in terms of feature extraction optimization, this method is more dependent on predefined wavelet basis functions, so the generalization ability is poor, and when the frequency components of modal aliasing are very close or change rapidly, it cannot separate the aliased modes well. The present invention uses the ICEEMDAN-SE decomposition and reconstruction method, which not only effectively solves the problem of modal aliasing, but also can adaptively perform modal decomposition according to the dynamic characteristics of the signal itself, does not rely on fixed frequency segments, and avoids the problems caused by direct addition of IMFs. Combined with k-means++, the features are clustered, and the health state of the battery is divided into different categories to achieve unsupervised learning. Patent (CN117849628B) discloses a lithium-ion battery health state estimation method based on a time-series transformation memory network. The method divides the voltage curve of the charging process in the data set into equal voltage intervals, and extracts the time difference between the equal voltage intervals to construct features. A new model structure is used to map and transform the decoded data obtained from the time-series transformation memory network through two fully connected layers to achieve the final SOH estimation. It can solve the short-term pattern and long-term dependency problems in the sequence data, accelerate the parallel processing of information at different levels, and significantly improve the estimation performance and generalization ability of the deep learning estimation method. However, the hyperparameters in the model need to be selected and tuned, which reduces the training efficiency of the model. The present invention uses the sparrow optimization algorithm that integrates sine, cosine and Cauchy mutations to optimize the hyperparameters of the model, thereby improving the training efficiency of the model. Patent (CN116540134A) discloses a GWO-LSTM lithium-ion battery health state estimation method. First, an LSTM long short-term memory network model for estimating the SOH of lithium-ion batteries is established.Then, the hyperparameters of the LSTM long short-term memory network are optimized by the GWO algorithm to obtain the best initial parameter configuration, and finally the pre-processed discharge voltage data MDV-SOH data of the lithium-ion battery cycle discharge is input into the trained model to obtain the lithium-ion battery SOH estimation result. However, the LSTM long short-term memory network can only capture the long-term dependencies in the sequence data and solve the gradient disappearance problem. Therefore, the present invention uses BiLSTM instead of LSTM, so that the model estimation result takes into account the past and future feature information, and improves the estimation ability of the model. Summary of the invention

[0005] The problem solved by the present invention is to provide a method for estimating the health status of a lithium-ion battery based on feature optimization, the main contents of which are as follows:

[0006] (1) The ICEEMDAN-SE decomposition and reconstruction method not only effectively solves the problem of modal aliasing, but also can adaptively perform modal decomposition based on the dynamic characteristics of the signal itself. It does not rely on fixed frequency segments and avoids the problems caused by direct addition of IMFs. It is combined with k-means++ to cluster features and classify the health status of the battery into different categories to achieve unsupervised learning.

[0007] (2) The sparrow optimization algorithm combining sine-cosine and Cauchy mutation (SCSSA) is used to optimize the initial learning rate and number of neurons of BiLSTM, which solves the problem of the influence of artificial hyperparameter tuning on the SOH estimation effect and accuracy.

[0008] To achieve the above object, the present invention provides the following technical solution: a method for estimating the health status of a lithium-ion battery based on feature optimization, the specific steps of which are as follows:

[0009] S1: Obtain the battery voltage, battery current, and charge and discharge time data during the battery experimental data set aging cycle charge and discharge process, and extract the preliminary health features that reflect battery degradation in intervals. The preliminary health features include the charging time, charging energy, and the peak and corresponding position voltage of the capacity increment curve within the constant current stage charging voltage interval and the constant voltage stage charging current interval.

[0010] S2: The extracted preliminary health features are normalized using the min-max normalization method;

[0011] S3: Use Spearman correlation analysis (SCA) and grey relational analysis (GRA) to jointly screen and reduce the input preliminary health characteristics;

[0012] S4: The health features after dimensionality reduction are optimized by using the reconstruction clustering module. The specific steps are as follows: First, the health features after dimensionality reduction are used as input, and the improved Intrinsic Computing Expressive Empirical Mode Decomposition With AdaptiveNoise-Sample Entrop (ICEEMDAN-SE) method is used to decompose and reconstruct the sample entropy features to obtain the sample entropy features; then the sample entropy features are transformed and reduced in dimension by the t-SNE algorithm; finally, k-means++ clustering analysis is used in the feature space after dimensionality reduction to obtain the optimized health features, and the optimized health features are divided into training set data and test set data according to a certain ratio;

[0013] S5: Input the optimized training set data to train BiLSTM, use the sparrow search algorithm (SCSSA) that combines sine and cosine and Cauchy mutations to optimize the hyperparameters in BiLSTM, including the initial learning rate and the number of neurons, and build a SCSSA-BiLSTM estimation model;

[0014] S6: Input the test set data of the optimized health features into the SCSSA-BiLSTM model to obtain the battery SOH estimation value, and verify whether the SOH estimation accuracy meets the design requirements. If it does not meet the requirements, reconstruct and cluster the health features after dimensionality reduction. If it meets the requirements, output the battery SOH estimation value.

[0015] Preferably, the preliminary health characteristics described in step S1 include the charging time and charging energy within the charging voltage interval in the constant current stage and the charging current interval in the constant voltage stage; in the constant current stage, the charging voltage interval is divided into different voltage partitions with a variation interval of 0.03V, and the charging time and charging energy within each voltage partition are extracted as preliminary health characteristics; in the constant voltage stage, the charging current interval is divided into different current partitions with a variation interval of 0.01A, and the charging time and charging energy within each current partition are extracted as preliminary health characteristics; in the constant current discharge stage, the peak value of the capacity increment (IC) curve and the voltage at its corresponding position are extracted as preliminary health characteristics.

[0016] Preferably, in step S3, the preliminary health characteristics are jointly screened and reduced using the Spearman correlation coefficient analysis (SCA) and the grey relational analysis (GRA), and the specific steps are as follows: (1) taking the preliminary health characteristics as input, calculating the Spearman correlation coefficient In the formula, n is the total number of observed sample features, d i represents the difference in the rank value of the i-th data pair; (2) according to the formula Get the grey correlation coefficient ξ between the i-th HF and SOH i (k), where k = 1, 2, ..., n, n is the number of battery cycles, y(k) is the SOH sequence, x i (k) is the HF sequence, ρ is the resolution coefficient, and then the formula Get the grey relational degree γ i , whose value range is [0,1]; (3) Spearman correlation coefficient ρ and grey relational degree γ i The eigenvalues ​​that all reached above 0.95 were used as healthy features after dimensionality reduction.

[0017] Preferably, in step S4, (1) the extracted health features after dimensionality reduction are decomposed and reconstructed using the ICEEMDAN-SE method, and the specific steps are as follows: ① The original battery health feature sequence L (i) =L+α0E k (ξ i ) is decomposed, where: L (i) is the ith health characteristic value of the battery, L is the baseline health characteristic value, α0 is the disturbance amplitude parameter, ξ i is the ith random perturbation input, E k (ξ i (t)) is the disturbance function E k , applied to the random perturbation input ξ i , and calculate the sequence signal M(L (i) ) defines M(·) as an operator that produces the local mean of the applied signal, and obtains the first residual, which is: Where N is the number of white noises added, the first-order (k=1) modal components can be derived, namely: According to Calculate the subsequent modal components r k It can be expressed as Among them, r k represents the eigenvalue of the kth iteration, r k-1 represents the eigenvalue of the previous iteration step, α k-1 is the perturbation coefficient of the previous iteration step. ICEEMDAN decomposes and obtains multiple intrinsic mode functions IMFs. According to the size of the sample entropy value, the IMFs related to SOH are initially obtained for reconstruction to obtain the sample entropy characteristics. ② The sample entropy is defined as: First prepare the time series data: x=[x1,x2,x3,…,x N ], where N is the length of the time series, then determine the embedding dimension m and tolerance r, and for a given time series x, construct all subsequences with an embedding dimension of m: (i=1,2,...,N-m+1), these subsequences are vectors consisting of m consecutive data points. Finally, the number of matching subsequence pairs B(m,r) is calculated. For each pair of subsequences and Calculate the distance between them: If the distance is less than or equal to the tolerance r, then it is considered and is a match. Define B(m,r) as the number of subsequence pairs that meet this condition divided by the total number of subsequence pairs N-m+1. Repeat the above process, but increase the embedding dimension to m+1, and you can calculate B(m+1,r);

[0018] (2) Transformation and dimension reduction of the obtained sample entropy features through the t-SNE algorithm;

[0019] (3) Use k-means++ clustering analysis in the feature space after dimensionality reduction. ① Randomly select a data point from the sample entropy feature after dimensionality reduction as the first cluster center. For each data point x, calculate its shortest distance D(x) to all currently selected cluster centers, and then use probability distribution to select the next cluster center, that is: ① is the sum of the squares of the distances from all data points not selected as centroids to the nearest centroid. ② select new cluster centers from the remaining data points until k cluster centers are selected (where k is the predetermined number of clusters). Once k initial cluster centers are selected, the conventional k-means algorithm is used for iteration: the distance from each data point to all cluster centers is calculated, and the data point is assigned to the cluster corresponding to the nearest center. The center of each cluster is recalculated until the cluster center no longer changes or the convergence condition is reached. The optimized health feature data is divided into training set data and test set data according to a certain ratio.

[0020] Preferably, in step S5, the specific working steps of BiLSTM in the SCSSA-BiLSTM estimation model are as follows:

[0021] (1) Start processing from the beginning of the sequence, update the hidden state through the forget gate, input gate, and output gate to capture positive information;

[0022] (2) Start processing from the end of the sequence, calculate the hidden state in reverse order, and capture reverse information;

[0023] (3) Merge the forward and reverse hidden states to form a bidirectional hidden state;

[0024] (4) The hidden state sequence of BiLSTM is input into the fully connected layer, the hidden state is linearly transformed through the weight matrix W and the bias term b, and the activation function ReLU is added to introduce nonlinear transformation;

[0025] (5) Map the output of the fully connected layer to the final output category or regression result;

[0026] The sparrow search algorithm (SCSSA) that combines sine and cosine and Cauchy mutations optimizes the BiLSTM hyperparameters. Its design is as follows:

[0027] (1) Using the sine-cosine strategy, the step search factor r1′ is calculated as follows: Among them, t is the number of iterations, Iter max is the maximum number of iterations, η is the adjustment coefficient, η≥1, a=1. Considering that the position update of individuals in the population is often affected by the current position during the entire search process of the SSA algorithm, the formula is introduced: The nonlinear weight factor ω is used to adjust the position of individuals in the population and update the dependence on the individual information at this time;

[0028] (2) The new discoverer position update formula is: in, is the value of the j-th variable of individual i at the t+1th iteration, is the value of the jth variable of individual i at the tth iteration, r1′∈[0,1] is a random number, r2∈[0,2π] is a random number, r3∈[0,2π] is a random number, X best is the position of the global or current best solution, ST is the threshold, and R2 is the variable;

[0029] (3) To avoid the algorithm from falling into local optimality, the Cauchy mutation strategy is introduced into the follower update formula. The new follower position update formula is: Among them, cauchy(0,1) is the standard Cauchy distribution function, It means multiplication.

[0030] Compared with the prior art, the present invention has the following technical effects: First, the ICEEMDAN-SE decomposition and reconstruction method not only effectively solves the problem of modal aliasing, but also can adaptively perform modal decomposition according to the dynamic characteristics of the signal itself, without relying on fixed frequency segments, and also avoids the problems caused by direct addition of IMF. Combined with k-means++, the features are clustered and the battery health status is divided into different categories to achieve unsupervised learning. Second, the SCSSA algorithm is used to optimize the hyperparameters of BiLSTM, which reduces the impact of randomly selected parameters on the model estimation results and improves the accuracy of SOH estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention.

[0032] In the attached picture:

[0033] Figure 1 is a flow chart of a method for estimating the health status of a lithium-ion battery according to the present invention;

[0034] Figure 2 It is a schematic diagram of the LSTM structure in the present invention.

[0035] Figure 3 It is a schematic diagram of the SCSSA-BiLSTM structure in the present invention. DETAILED DESCRIPTION

[0036] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0037] Example: Figure 1 As shown, a lithium-ion battery health status estimation method based on feature optimization, the specific steps are as follows:

[0038] S1: Obtain the voltage, current, and time data of battery B0005 in the NASA battery experiment data set during the aging cycle charge and discharge process, and extract the preliminary health characteristics reflecting battery degradation in intervals. The battery charge and discharge experiment process is as follows: first, charge with a constant current (CC) of 1.5A until the voltage rises to 4.3V, then charge with a constant voltage (CV) until the current drops to 15mA, and keep the current 2A unchanged during the discharge process. When the battery voltage drops from 4.3V to the cut-off voltage, the discharge ends. The extracted health characteristic parameters are the charging time and charging energy within the charging voltage range of 3.7V to 4.3V in the CC stage, which are recorded as IHF1 and IHF2, and the charging time and charging energy within the charging current range of 1.5A to 0.5A in the CV stage; the charging voltage interval of the CC stage is divided into different voltage intervals with a change interval of 0.03V, and the charging current interval of the CV stage is divided into different current intervals with a change interval of 0.01A. The charging time and charging energy in each voltage interval and current interval are extracted as health characteristic parameters, which are recorded as IHF3 and IHF4. In the CC stage, the charging interval is from 3.5V to 4.3V, with 0.01V as the change interval, and the peak value of the capacity increment (IC) curve and the voltage at the corresponding position are extracted as the preliminary health characteristics, which are recorded as IHF5 and IHF6;

[0039] S2: The min-max standardization method is used, and the calculation formula is Where, X i is the health characteristic data, y i is the standardized data, X max and X min are the maximum and minimum values ​​corresponding to the preliminary health characteristics;

[0040] S3: Use Spearman correlation coefficient analysis (SCA) and grey relational analysis (GRA) to jointly screen and reduce the dimension of preliminary health characteristics. The specific steps are as follows: (1) Take preliminary health characteristics as input and calculate the Spearman correlation coefficient In the formula, n is the total number of observed sample features, d i represents the difference in the rank value of the i-th data pair; (2) according to the formula Get the grey correlation coefficient ξ between the i-th HF and SOH i (k), where k = 1, 2, ..., n, n is the number of battery cycles, y(k) is the SOH sequence, x i (k) is the HF sequence, ρ is the resolution factor, which is set to 0.5, and then the formula Get the grey relational degree γ i , whose value range is [0,1]; (3) Spearman correlation coefficient ρ and grey relational degree γ i The eigenvalues ​​that all reached above 0.95 were used as health features after dimensionality reduction;

[0041] S4: (1) Using the ICEEMDAN-SE method to decompose and reconstruct the extracted health features after dimensionality reduction, the specific steps are as follows: ① The original battery health feature sequence L (i) =L+α0E k (ξ i ) is decomposed, where: L (i) is the ith health characteristic value of the battery, L is the baseline health characteristic value, α0 is the disturbance amplitude parameter, ξ i is the ith random perturbation input, E k (ξ i (t)) is the disturbance function E k , applied to the random perturbation input ξ i , the noise standard deviation Nstd is set to 0.2, the number of iterations NR is set to 50, the maximum number of decomposition iterations MaxIter is set to 500, the signal-to-noise ratio flag SNRFlag is set to 1, and the sequence signal M(L (i) ) defines M(·) as an operator that produces the local mean of the applied signal, and obtains the first residual, which is: Where N is the number of white noises added, the first-order (k=1) modal components can be derived, namely: According to Calculate the subsequent modal components r k It can be expressed as Among them, r k represents the eigenvalue of the kth iteration, r k-1represents the eigenvalue of the previous iteration step, α k-1 is the perturbation coefficient of the previous iteration step. ICEEMDAN decomposes and obtains multiple intrinsic mode functions IMFs. According to the size of the sample entropy value, the IMFs related to SOH are initially obtained for reconstruction to obtain the sample entropy characteristics. ② The sample entropy is defined as: First prepare the time series data: x=[x1,x2,x3,…,x N ], where N is the length of the time series, then determine the embedding dimension m and tolerance r, and for a given time series x, construct all subsequences with an embedding dimension of m: (i=1,2,...,N-m+1), these subsequences are vectors consisting of m consecutive data points. Finally, the number of matching subsequence pairs B(m,r) is calculated. For each pair of subsequences and Calculate the distance between them: If the distance is less than or equal to the tolerance r, then it is considered and is a match. Define B(m,r) as the number of subsequence pairs that meet this condition divided by the total number of subsequence pairs N-m+1. Repeat the above process, but increase the embedding dimension to m+1, and you can calculate B(m+1,r);

[0042] (2) Transformation and dimension reduction of the obtained sample entropy features through the t-SNE algorithm;

[0043] (3) Use k-means++ clustering analysis in the feature space after dimensionality reduction. ① Randomly select a data point from the sample entropy feature after dimensionality reduction as the first cluster center. For each data point x, calculate its shortest distance D(x) to all currently selected cluster centers, and then use probability distribution to select the next cluster center, that is: The sum of the squares of the distances from all data points not selected as centroids to the nearest centroid. ② Select new cluster centers from the remaining data points until k cluster centers are selected (where k is the predetermined number of clusters). k is selected as 3, indicating that they are divided into three categories: high, medium and low frequency. Once k initial cluster centers are selected, the conventional k-means algorithm is used for iteration: the distance from each data point to all cluster centers is calculated, and the data point is assigned to the cluster corresponding to the nearest center. The center of each cluster is recalculated until the cluster center no longer changes or reaches the convergence condition, and the optimized health feature data is divided into training set data and test set data according to a certain ratio.

[0044] S5: BiLSTM is a recurrent neural network (RNN). The hyperparameter settings are: the number of hidden neurons is 15, the maximum number of training iterations is 150, the initial learning rate is set to 0.001, and the L2 regularization parameter is 0.00001. The specific working steps of BiLSTM in the SCSSA-BiLSTM estimation model are as follows:

[0045] (1) Start processing from the beginning of the sequence, update the hidden state through the forget gate, input gate, and output gate to capture positive information;

[0046] (2) Start processing from the end of the sequence, calculate the hidden state in reverse order, and capture reverse information;

[0047] (3) Merge the forward and reverse hidden states to form a bidirectional hidden state;

[0048] (4) The hidden state sequence of BiLSTM is input into the fully connected layer, the hidden state is linearly transformed through the weight matrix W and the bias term b, and the activation function ReLU is added to introduce nonlinear transformation;

[0049] (5) Map the output of the fully connected layer to the final output category or regression result;

[0050] The sparrow search algorithm (SCSSA) that combines sine and cosine and Cauchy mutations optimizes the BiLSTM hyperparameters. The population size is set to 30 and the number of iterations is set to 10. The design is as follows:

[0051] (1) Using the sine-cosine strategy, the step search factor r1′ is calculated as follows: Among them, t is the number of iterations, Iter max is the maximum number of iterations, η is the adjustment coefficient, η≥1, a=1. Considering that the position update of individuals in the population is often affected by the current position during the entire search process of the SSA algorithm, the formula is introduced: The nonlinear weight factor ω is used to adjust the position of individuals in the population and update the dependence on the individual information at this time;

[0052] (2) The new discoverer position update formula is: in, is the value of the j-th variable of individual i at the t+1th iteration, is the value of the jth variable of individual i at the tth iteration, r1′∈[0,1] is a random number, r2∈[0,2π] is a random number, r3∈[0,2π] is a random number, X best is the position of the global or current best solution, ST is the threshold, and R2 is the variable;

[0053] (3) To avoid the algorithm from falling into local optimality, the Cauchy mutation strategy is introduced into the follower update formula. The new follower position update formula is: Among them, cauchy(0,1) is the standard Cauchy distribution function, It means multiplication.

[0054] S6: Input the test set data into the SCSSA-BiLSTM estimation model to verify whether the model accuracy meets the expected requirements. The mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are selected as evaluation indicators to evaluate the estimation accuracy of the SCSSA-BiLSTM model to determine whether the expected results are met. The specific formula is: Where N is the number of experimental estimates, y n is the actual value of SOH, It is the SOH estimation value. The test results show that MAE is 0.0057-0.013, RMSE is 0.0051-0.011, both within 0.015, and MAPE is 0.83%-1.13%, within 1.5%. If the requirements are not met, the reconstruction and clustering of the health features after dimensionality reduction are performed again.

[0055] Finally, it should be noted that the above description is only an example of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for estimating the health status of a lithium-ion battery based on feature optimization, characterized in that: The method for estimating the health status of a lithium-ion battery based on feature optimization and data-driven includes the following steps: S1 obtains the aging test data of the battery during charging and discharging, including the battery voltage, battery current, and charging and discharging time data, and extracts the initial health features (IHFs) reflecting the battery degradation in different intervals. The initial health features include the charging time, charging energy, and the peak value and corresponding voltage of the capacity increment curve in the charging voltage interval of the constant current stage and the charging current interval of the constant voltage stage; S2 normalizes the extracted preliminary health features using the min-max normalization method; S3 uses Spearman correlation coefficient analysis (SCA) and grey relational analysis (GRA) to jointly screen and reduce the dimensions of the proposed preliminary health features to obtain the reduced-dimensional health features; S4 uses the reconstruction clustering module to optimize the health features after dimensionality reduction. The specific steps are as follows: first, the health features after dimensionality reduction are used as input, and the improved completely adaptive noise ensemble empirical mode decomposition and sample entropy (ICEEMDAN-SE) method is used to decompose and reconstruct to obtain sample entropy features; then the t-SNE algorithm is used to transform and reduce the dimensionality of the sample entropy features; finally, k-means++ clustering analysis is used in the feature space after dimensionality reduction to obtain the optimized health features, and the optimized health features are divided into training set data and test set data according to a certain ratio; S5 inputs the training set data of the optimized health features into the BiLSTM model, and uses the sparrow search algorithm (SCSSA) fused with sine-cosine and Cauchy mutation to optimize the hyperparameters in BiLSTM, including the initial learning rate and the number of neurons, and constructs the SCSSA-BiLSTM estimation model; S6 inputs the test set data of the optimized health features into the SCSSA-BiLSTM model to obtain the battery SOH estimation value and verify whether the SOH estimation accuracy meets the design requirements. If it does not meet the requirements, the health features are reconstructed and clustered after dimensionality reduction. If it meets the requirements, the battery SOH estimation value is output.

2. A lithium-ion battery health status estimation method based on feature optimization according to claim 1, characterized in that: The preliminary health characteristics include the charging time and charging energy within the charging voltage interval in the constant current stage and the charging current interval in the constant voltage stage; in the constant current stage, the charging voltage interval is divided into different voltage partitions with a variation interval of 0.03V, and the charging time and charging energy within each voltage partition are extracted as preliminary health characteristics; in the constant voltage stage, the charging current interval is divided into different current partitions with a variation interval of 0.01A, and the charging time and charging energy within each current partition are extracted as preliminary health characteristics; in the constant current discharge stage, the peak value of the capacity increment (IC) curve and the corresponding position voltage are extracted as preliminary health characteristics.

3. The method for estimating the health status of a lithium-ion battery based on feature optimization according to claim 1, characterized in that: The Spearman correlation coefficient analysis (SCA) and grey relational analysis (GRA) were used to jointly screen and reduce the dimension of the preliminary health characteristics. The specific steps are as follows: (1) The preliminary health characteristics were used as input and the Spearman correlation coefficient was calculated. In the formula, n is the total number of observed sample features, d i represents the difference in the rank value of the i-th data pair; (2) according to the formula Get the grey correlation coefficient ξ between the i-th HF and SOH i (k), where k = 1, 2, ..., n, n is the number of battery cycles, y(k) is the SOH sequence, x i (k) is the HF sequence, ρ is the resolution coefficient, and then the formula Get the grey relational degree γ i , whose value range is [0,1]; (3) Spearman correlation coefficient ρ and grey relational degree γ i The eigenvalues ​​that all reached above 0.95 were used as healthy features after dimensionality reduction.

4. The method for estimating the health status of a lithium-ion battery based on feature optimization according to claim 1, characterized in that: In the reconstruction clustering module, the specific steps are as follows: (1) The ICEEMDAN-SE method is used to decompose and reconstruct the extracted health features after dimensionality reduction. The specific steps are as follows: ① The original battery health feature sequence L (i) =L+α0E k (ξ i ) is decomposed, where: L (i) is the ith health characteristic value of the battery, L is the baseline health characteristic value, α0 is the disturbance amplitude parameter, ξ i is the ith random perturbation input, E k (ξ i (t)) is the disturbance function E k , applied to the random perturbation input ξ i , and calculate the sequence signal M(L (i) ) defines M(·) as an operator that produces the local mean of the applied signal, and obtains the first residual, which is: Where N is the number of white noises added, the first-order (k=1) modal components can be derived, namely: According to Calculate the subsequent modal components r k It can be expressed as Among them, r k represents the eigenvalue of the kth iteration, r k-1 represents the eigenvalue of the previous iteration step, α k-1 is the disturbance coefficient of the previous iteration step, ICEEMDA N decomposes to obtain multiple intrinsic mode functions IMFs, and according to the size of the sample entropy value, the IMFs related to SOH are initially obtained for reconstruction to obtain the sample entropy characteristics; ② The sample entropy is defined as: First prepare the time series data: x=[x1,x2,x3,...,x N ], where N is the length of the time series, then determine the embedding dimension m and tolerance r, and for a given time series x, construct all subsequences with an embedding dimension of m: These subsequences are vectors consisting of m consecutive data points. Finally, the number of matching subsequence pairs B(m,r) is calculated. For each pair of subsequences and Calculate the distance between them: If the distance is less than or equal to the tolerance r, then it is considered and is a match; define B(m,r) as the number of subsequence pairs that meet this condition divided by the total number of subsequence pairs N-m+1. Repeat the above process, but increase the embedding dimension to m+1, and you can calculate B(m+1,r); (2) Transformation and dimension reduction of the obtained sample entropy features through the t-SNE algorithm; (3) Use k-means++ clustering analysis in the feature space after dimensionality reduction. ① Randomly select a data point from the sample entropy feature after dimensionality reduction as the first cluster center. For each data point x, calculate its shortest distance D(x) to all currently selected cluster centers, and then use probability distribution to select the next cluster center, that is: The sum of the squares of the distances from all data points not selected as centroids to the nearest centroid; ② Select new cluster centers from the remaining data points until k cluster centers are selected (where k is the predetermined number of clusters). Once k initial cluster centers are selected, the conventional k-means algorithm is used for iteration: the distance from each data point to all cluster centers is calculated, and the data point is assigned to the cluster corresponding to the nearest center. The center of each cluster is recalculated until the cluster center no longer changes or the convergence condition is reached, and the optimized health feature data is divided into training set data and test set data according to a certain ratio.

5. The method for estimating the health status of a lithium-ion battery based on feature optimization according to claim 1, characterized in that: The specific working steps of BiLSTM in the SCSSA-BiLSTM estimation model are as follows: (1) Start processing from the beginning of the sequence, update the hidden state through the forget gate, input gate, and output gate to capture positive information; (2) Start processing from the end of the sequence, calculate the hidden state in reverse order, and capture reverse information; (3) Merge the forward and reverse hidden states to form a bidirectional hidden state; (4) The hidden state sequence of BiLSTM is input into the fully connected layer, the hidden state is linearly transformed through the weight matrix W and the bias term b, and the activation function ReLU is added to introduce nonlinear transformation; (5) Map the output of the fully connected layer to the final output category or regression result; The sparrow search algorithm (SCSSA) that combines sine and cosine and Cauchy mutations optimizes the BiLSTM hyperparameters. Its design is as follows: (1) Using the sine-cosine strategy, the step search factor r1′ is calculated as follows: Among them, t is the number of iterations, Iter max is the maximum number of iterations, η is the adjustment coefficient, η≥1, a=1. Considering that the position update of the individual population is often affected by the current position during the entire search process of the SSA algorithm, the formula is introduced: The nonlinear weight factor ω is used to adjust the position of individuals in the population and update the dependence on the individual information at this time; (2) The new discoverer position update formula is: in, is the value of the j-th variable of individual i at the t+1th iteration, is the value of the jth variable of individual i at the tth iteration, r1′∈[0,1] is a random number, r2∈[0,2π] is a random number, r3∈[0,2π] is a random number, X best is the position of the global or current best solution, ST is the threshold, and R2 is the variable; (3) To avoid the algorithm from falling into local optimality, the Cauchy mutation strategy is introduced into the follower update formula. The new follower position update formula is: Among them, cauchy(0,1) is the standard Cauchy distribution function, It means multiplication.

Images