Fast-charging lithium ion battery health state estimation method based on time-frequency characteristics and TS-KELM-AE

Abstract

The invention discloses a fast-charging lithium ion battery health state estimation method based on time-frequency characteristics and TS-KELM-AE, and the method comprises the steps: 1), obtaining a current and voltage data sequence of a lithium battery in each charging step in different fast-charging circulation processes, and calculating a health state data sequence of the corresponding fast-charging circulation process; 2) calculating capacity data in the charging step; 3) performing health index extraction on the charging step content data from a time domain and a frequency domain, and calculating a Pearson correlation coefficient of the health index and the health state data sequence; 4) preprocessing the extracted health index and health state data sequence to construct a training set; 5) constructing a TS-KELM-AE network model, and training the TS-KELM-AE network model by using the training set to obtain a lithium battery health condition estimation model; and 6) extracting the health index of the lithium battery to be detected, inputting the health index of the lithium battery to be detected into the lithium battery health state estimation model, and estimating the health state of the lithium ion battery. According to the method, the health state of the aged fast-charging lithium battery at different temperatures can be quickly and accurately estimated.

Description

Technical Field

[0001] The present invention relates to the field of battery technology, and in particular to a method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE. Background Art

[0002] As the core driving force of the green energy transition, lithium-ion batteries have attracted much attention due to their high energy density, long cycle life, and environmentally friendly and recyclable characteristics. With the widespread popularity of consumer electronic devices such as electric vehicles (EVs) and smartphones, traditional slow charging technology can no longer meet the requirements for convenience and high efficiency. Therefore, fast charging technology is becoming a key technology to resolve this contradiction with its advantages such as shortening charging time and improving user experience. However, the popularization of fast charging technology has also brought new technical challenges. Under high-rate charging and discharging conditions, the chemical and physical reactions inside the battery are significantly accelerated, resulting in an increase in the aging rate of the battery, which directly affects its service life and safety.

[0003] The application of fast charging technology places higher demands on battery management systems (BMS) to monitor and estimate the state of health (SOH) of lithium batteries in real time. Furthermore, in traditional SOH estimation methods, a single health indicator often only reveals partial information about the battery's health status, resulting in a lack of reliability and accuracy in the SOH estimation results.

[0004] In recent years, data-driven algorithms such as Extreme Learning Machine (ELM), Support Vector Machine (SVM), Gaussian Process Regression (GPR), and Long Short-Term Memory (LSTM) have achieved remarkable success in battery SOH estimation, but they also have various limitations. For example, ELM has efficient learning capabilities, but it is difficult to establish reliable nonlinear mapping when faced with complex time series modeling tasks. Classic machine learning methods SVM and GPR have a complete statistical theoretical foundation, but it is difficult to fully explore deep features. On the other hand, although deep learning methods such as LSTM have powerful representation capabilities, their high dependence on large-scale data and computing resources limits their application in practice.

[0005] Therefore, developing a feature extraction engineering that can effectively process large-scale, multi-dimensional data and designing a fast-charging lithium battery SOH estimation model that adapts to large-scale input data have become the key to improving the accuracy and efficiency of modern BMS. Summary of the Invention

[0006] The purpose of the present invention is to provide a method for estimating the state of health of a fast-charged lithium-ion battery based on time-frequency characteristics and TS-KELM-AE, comprising the following steps:

[0007] 1) Obtain the current and voltage data sequences of each charging step of the lithium battery during different fast charging cycles, and calculate the health status data sequences corresponding to different fast charging cycles.

[0008] 2) Based on the current and voltage data sequence of each charging step, the capacity data of the charging step is calculated.

[0009] 3) Extract health indicators from the charging step capacity data in the time domain and frequency domain, and perform Pearson correlation analysis on the extracted health indicators and health status data sequence to obtain the Pearson correlation between each health indicator and the health status of the lithium battery.

[0010] Based on the numerical value of Pearson correlation, the corresponding relationship between health indicators and health status data is constructed.

[0011] 4) Preprocessing the extracted health indicators and health status data sequences, and constructing a training set with the preprocessed health indicators as input and the corresponding health status data as output.

[0012] 5) Based on the extreme learning machine improved by kernel function and autoencoder, a TS-KELM-AE network model is constructed, and the TS-KELM-AE network model is trained using the training set to obtain a lithium battery health status estimation model.

[0013] 6) Extracting the health indicators of the lithium battery to be tested, and inputting the health indicators of the lithium battery to be tested into the lithium battery health status estimation model to estimate the health status data of the lithium-ion battery.

[0014] Furthermore, the health status data sequences corresponding to the different fast charging cycle processes are as follows:

[0015]

[0016] Where i is the fast charge cycle index, i = 1, 2, ..., n, n is the total number of fast charge cycles, H i is the battery health status of the i-th fast charge cycle, C i is the maximum discharge capacity of the lithium battery in the i-th fast charge cycle, and C is the rated capacity of the lithium battery.

[0017] Furthermore, the charging step capacity data is as follows:

[0018]

[0019] Where i is the fast charge cycle index, n is the total number of cycles, k is the charge step index, K is the total number of charge steps, t is the time, and e is the total time. i,k is the charging step capacity of the kth charging step in the i-th fast charging cycle. nom is the nominal voltage. i,k,t is the current value at time t of the kth charging step in the i-th fast charging cycle. V i,k,t is the voltage value at time t of the kth charging step in the i-th fast charging cycle.

[0020] Furthermore, the method for extracting health indicators from the charging step capacity data in the time domain includes fast Fourier transform.

[0021] The health indicators extracted from the time domain of the charging step content data include: mean, standard deviation, maximum, minimum, root mean square, kurtosis, skewness, median, energy, peak frequency, spectral entropy, root mean square spectrum, average spectrum, bandwidth and average frequency.

[0022] Furthermore, the method for extracting health indicators from the charging step capacity data in the frequency domain includes wavelet analysis.

[0023] The health indicators obtained by extracting the charging step capacity data from the frequency domain include: average energy, maximum energy, energy standard deviation, low-frequency energy, medium-frequency energy, high-frequency energy, wavelet energy, and wavelet correlation coefficient.

[0024] The low-frequency energy refers to energy with a frequency less than or equal to 50 Hz.

[0025] The intermediate frequency energy refers to the energy with a frequency within the range of 50 Hz to 200 Hz.

[0026] The high-frequency energy refers to energy with a frequency greater than 200 Hz.

[0027] Furthermore, the calculation formula of the Pearson correlation analysis is as follows:

[0028]

[0029] Where i is the fast charge cycle index, n is the total number of cycles, p is the health indicator index, and ξ(p) is the Pearson correlation coefficient between the pth health indicator and the health status data sequence. i is the battery health status of the i-th fast charge cycle, F p,i is the pth health indicator of the i-th fast charging cycle.

[0030] Among them, the battery health status H i The mean value E(H i ) and health index F p,i The mean value E(Fp,i ) is as follows:

[0031]

[0032] Furthermore, the preprocessing includes normalization processing.

[0033] The normalization process includes maximum and minimum value normalization process, as shown below:

[0034]

[0035] Where x0(n) is the data to be processed. max is the maximum value in the data x0(n), x0(n) min is the minimum value in the data x0(n). * The data are normalized.

[0036] Furthermore, the steps of constructing the TS-KELM-AE network model are as follows:

[0037] 5.1) Construct the KELM model.

[0038] The output of the KELM model is as follows:

[0039]

[0040] Where, F KELM is the output result of the KELM model. is the output weight of the KELM model. m x(n) , M are health indicator sequences [F p,1 ,F p,2 ,…,F p,n ]The output vector form and spatial matrix form of hidden nodes. is the output weight. T is the Moore-Penrose generalized inverse matrix of the spatial matrix M. U is the identity matrix, λ is the regularization coefficient, and H is the battery health state sequence [H1,H2,…,H n ]. K is the kernel function.

[0041] The radial basis function kernel K(x,x(n)) is as follows:

[0042]

[0043] Where σ is the kernel function width. x represents the health indicator sequence to be evaluated. x(n) represents a health indicator sequence in the training set.

[0044] 5.2) Substitute the autoencoder into the output of the KELM model FKELM , construct the KELM-AE model.

[0045] The output weights of the KELM-AE model As shown below:

[0046]

[0047] 5.3) Set the TS-KELM-AE network model to include L KELM-AE hidden layers, and based on the output weights of the KELM-AE model The output weight of the Lth layer KELM-AE hidden layer of the TS-KELM-AE network model is obtained by recursive calculation and the output M of the Lth KELM-AE hidden layer L , as shown below:

[0048]

[0049] Where λ L-1 M represents the regularization coefficient of the L-1th KELM-AE hidden layer. L-1 g represents the output of the L-1th KELM-AE hidden layer. * is the activation function.

[0050] 5.4) The output weight of the Lth layer of the TS-KELM-AE network model and the output M of the hidden layer L Bring in the output result F of the KELM model KELM , the TS-KELM-AE network model is constructed.

[0051] Furthermore, the autoencoder includes an encoding part and a decoding part.

[0052] The coding part is as follows:

[0053] Z=g * (E·X+B) (13)

[0054] Where X represents the input health indicator sequence [F p,1 ,F p,2 ,…,F p,n ]. Z represents the compressed data of the health indicator sequence X. E represents the encoded weight matrix. B represents the encoded bias matrix. g * is the activation function.

[0055] The decoding part is as follows:

[0056]

[0057] Where, Denotes the reconstructed data of the compressed data Z. D denotes the decoded weight matrix. Represents the bias matrix for decoding.

[0058] Furthermore, the TS-KELM-AE network model is trained using the mean square error between the health status data sequence in the training set and the predicted value as the loss function.

[0059] The technical effect of the present invention is unquestionable. The present invention proposes a method for estimating the health status of fast-charging lithium-ion batteries based on time-frequency features and TS-KELM-AE. This method combines the advantages of time domain and frequency domain feature extraction, and designs a two-stage kernel extreme learning machine autoencoder (TS-KELM-AE) model by combining kernel functions and encoders (AE). It can realize rapid and accurate estimation of the health status of fast-charging lithium batteries aged at different temperatures.

[0060] The present invention provides a new perspective for extracting health indicators from the time domain and frequency domain, identifying more comprehensive health indicator information for fast-charging lithium batteries. This multi-perspective feature extraction method can evaluate the health status of fast-charging batteries from different levels and angles, effectively avoiding the limitations of single time domain or frequency domain analysis.

[0061] This paper designs the TS-KELM-AE model to adapt to the SOH estimation of fast-charge lithium batteries under a large range of health indicator inputs. This two-stage structure optimizes the health estimation process, ensuring high accuracy while maintaining computational speed. It is also adaptable to SOH estimation under different aging temperatures. The first stage focuses on extracting deep features, while the second stage focuses on making accurate SOH estimation decisions.

[0062] The present invention develops a time-frequency feature extraction project to effectively solve the shortcomings of traditional single health indicators; in addition, a two-stage kernel extreme learning machine autoencoder (TS-KELM-AE) model is designed to adapt to the SOH estimation of fast-charging lithium batteries under large-scale health indicator input. The health state estimation process is optimized through a two-stage structure, ensuring high accuracy while guaranteeing calculation speed, and can adapt to the SOH estimation of fast-charging lithium batteries aged at different temperatures. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 1 is a flow chart of a method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE, provided in an embodiment of the present invention;

[0064] Figure 2 1 is a nine-step fast charging protocol diagram for lithium-ion batteries provided by an embodiment of the present invention;

[0065] Figure 3This is a Pearson correlation heat map result diagram of 23 health indicators and SOH at different temperatures provided by an embodiment of the present invention;

[0066] Figure 4 1 is a diagram showing the specific structure and calculation process of the TS-KELM-AE model provided by an embodiment of the present invention;

[0067] Figure 5 This is a comparison chart of SOH estimation results of fast-charge lithium batteries aged at 25°C provided by an embodiment of the present invention;

[0068] Figure 6 This is a comparison chart of SOH estimation results of fast-charge lithium batteries aged at 35°C provided by an embodiment of the present invention;

[0069] Figure 7 This is a comparison chart of SOH estimation results of fast-charge lithium batteries aged at 45°C provided by an embodiment of the present invention;

[0070] Figure 8 This is a comparison chart of SOH estimation results of fast-charge lithium batteries aged at 55° C. provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The present invention will be further described below with reference to the following examples, but it should not be understood that the scope of the present invention is limited to the following examples. Without departing from the above technical ideas of the present invention, various substitutions and modifications can be made according to common technical knowledge and customary means in the art, and all of these should be included in the scope of protection of the present invention.

[0072] Example 1:

[0073] See also Figures 1 to 8 , a fast-charge lithium-ion battery health status estimation method based on time-frequency characteristics and TS-KELM-AE includes the following steps:

[0074] 1) Obtain the current and voltage data sequences of each charging step of the lithium battery during different fast charging cycles, and calculate the health status data sequences corresponding to different fast charging cycles.

[0075] 2) Based on the current and voltage data sequence of each charging step, the capacity data of the charging step is calculated.

[0076] 3) Extract health indicators from the charging step capacity data in the time domain and frequency domain, and perform Pearson correlation analysis on the extracted health indicators and health status data sequence to obtain the Pearson correlation between each health indicator and the health status of the lithium battery.

[0077] Based on the numerical value of Pearson correlation, the corresponding relationship between health indicators and health status data is constructed.

[0078] 4) Preprocessing the extracted health indicators and health status data sequences, and constructing a training set with the preprocessed health indicators as input and the corresponding health status data as output.

[0079] 5) Based on the extreme learning machine improved by kernel function and autoencoder, a TS-KELM-AE network model is constructed, and the TS-KELM-AE network model is trained using the training set to obtain a lithium battery health status estimation model.

[0080] 6) Extracting the health indicators of the lithium battery to be tested, and inputting the health indicators of the lithium battery to be tested into the lithium battery health status estimation model to estimate the health status data of the lithium-ion battery.

[0081] Example 2:

[0082] The health status estimation method of fast-charge lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which is shown in Example 1, further, the health status data sequences corresponding to the different fast-charge cycle processes are as follows:

[0083]

[0084] Where i is the fast charge cycle index, i = 1, 2, ..., n, n is the total number of fast charge cycles, H i is the battery health status of the i-th fast charge cycle, C i is the maximum discharge capacity of the lithium battery in the i-th fast charge cycle, and C is the rated capacity of the lithium battery.

[0085] Example 3:

[0086] A method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 1 to 2. Furthermore, the capacity data within the charging step is shown below:

[0087]

[0088] Where i is the fast charge cycle index, n is the total number of cycles, k is the charge step index, K is the total number of charge steps, t is the time, and e is the total time. i,k is the charging step capacity of the kth charging step in the i-th fast charging cycle. nom is the nominal voltage. i,k,t is the current value at time t of the kth charging step in the i-th fast charging cycle. V i,k,t is the voltage value at time t of the kth charging step in the i-th fast charging cycle.

[0089] Example 4:

[0090] The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 1 to 3. Furthermore, the method for extracting health indicators from the charging step capacity data in the time domain includes fast Fourier transform.

[0091] The health indicators extracted from the time domain of the charging step content data include: mean, standard deviation, maximum, minimum, root mean square, kurtosis, skewness, median, energy, peak frequency, spectral entropy, root mean square spectrum, average spectrum, bandwidth and average frequency.

[0092] Example 5:

[0093] The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 1 to 4. Furthermore, the method for extracting health indicators from the charging step capacity data in the frequency domain includes wavelet analysis.

[0094] The health indicators obtained by extracting the charging step capacity data from the frequency domain include: average energy, maximum energy, energy standard deviation, low-frequency energy, medium-frequency energy, high-frequency energy, wavelet energy, and wavelet correlation coefficient.

[0095] The low-frequency energy refers to energy with a frequency less than or equal to 50 Hz.

[0096] The intermediate frequency energy refers to the energy with a frequency within the range of 50 Hz to 200 Hz.

[0097] The high-frequency energy refers to energy with a frequency greater than 200 Hz.

[0098] Example 6:

[0099] A method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 1 to 5. Furthermore, the calculation formula for the Pearson correlation analysis is as follows:

[0100]

[0101] Where i is the fast charge cycle index, n is the total number of cycles, p is the health indicator index, and ξ(p) is the Pearson correlation coefficient between the pth health indicator and the health status data sequence. i is the battery health status of the i-th fast charge cycle, F p,i is the pth health indicator of the i-th fast charging cycle.

[0102] Among them, the battery health status H i The mean value E(H i ) and health index F p,iThe mean value E(F p,i ) is as follows:

[0103]

[0104] Example 7:

[0105] A method for estimating the health status of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 1 to 6. Furthermore, the preprocessing includes normalization processing.

[0106] The normalization process includes maximum and minimum value normalization process, as shown below:

[0107]

[0108] Where x0(n) is the data to be processed. max is the maximum value in the data x0(n), x0(n) min is the minimum value in the data x0(n). * The data are normalized.

[0109] Example 8:

[0110] The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is described in any one of Examples 1 to 7. Furthermore, the steps for constructing the TS-KELM-AE network model are as follows:

[0111] 5.1) Construct the KELM model.

[0112] The output of the KELM model is as follows:

[0113]

[0114] Where, F KELM is the output result of the KELM model. is the output weight of the KELM model. m x(n) , M are health indicator sequences [F p,1 ,F p,2 ,…,F p,n ]The output vector form and spatial matrix form of hidden nodes. is the output weight. T is the Moore-Penrose generalized inverse matrix of the spatial matrix M. U is the identity matrix, λ is the regularization coefficient, and H is the battery health state sequence [H1,H2,…,H n ]. K is the kernel function.

[0115] The radial basis function kernel K(x,x(n)) is as follows:

[0116]

[0117] Where σ is the kernel function width. x represents the health indicator sequence to be evaluated. x(n) represents a health indicator sequence in the training set.

[0118] 5.2) Substitute the autoencoder into the output of the KELM model F KELM , construct the KELM-AE model.

[0119] The output weights of the KELM-AE model As shown below:

[0120]

[0121] 5.3) Set the TS-KELM-AE network model to include L KELM-AE hidden layers, and based on the output weights of the KELM-AE model The output weight of the Lth layer KELM-AE hidden layer of the TS-KELM-AE network model is obtained by recursive calculation and the output M of the Lth KELM-AE hidden layer L , as shown below:

[0122]

[0123] Where λ L-1 M represents the regularization coefficient of the L-1th KELM-AE hidden layer. L-1 g represents the output of the L-1th KELM-AE hidden layer. * is the activation function.

[0124] 5.4) The output weight of the Lth layer of the TS-KELM-AE network model and the output M of the hidden layer L Bring in the output result F of the KELM model KELM , the TS-KELM-AE network model is constructed.

[0125] Example 9:

[0126] A method for estimating the health status of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 1 to 8. Furthermore, the autoencoder includes an encoding part and a decoding part.

[0127] The coding part is as follows:

[0128] Z=g * (E·X+B) (13)

[0129] Where X represents the input health indicator sequence [F p,1 ,F p,2 ,…,F p,n ]. Z represents the compressed data of the health indicator sequence X. E represents the encoded weight matrix. B represents the encoded bias matrix. g * is the activation function.

[0130] The decoding part is as follows:

[0131]

[0132] Where, Denotes the reconstructed data of the compressed data Z. D denotes the decoded weight matrix. Represents the bias matrix for decoding.

[0133] Example 10:

[0134] A method for estimating the health status of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 1 to 9. Furthermore, the TS-KELM-AE network model is trained using the mean square error between the health status data sequence in the training set and the predicted value as the loss function.

[0135] Example 11:

[0136] See also Figures 1 to 8 , a fast-charge lithium-ion battery health status estimation method based on time-frequency characteristics and TS-KELM-AE includes the following steps:

[0137] 1) Obtain the current and voltage data sequences of each charging step of the lithium battery during different fast charging cycles, and calculate the health status data sequences corresponding to different fast charging cycles.

[0138] 2) Based on the current and voltage data sequence of each charging step, the capacity data of the charging step is calculated.

[0139] 3) Extract health indicators from the charging step capacity data in the time domain and frequency domain, and perform Pearson correlation analysis on the extracted health indicators and health status data sequence to obtain the Pearson correlation between each health indicator and the health status of the lithium battery.

[0140] Based on the numerical value of Pearson correlation, the corresponding relationship between health indicators and health status data is constructed.

[0141] 4) Preprocessing the extracted health indicators and health status data sequences, and constructing a training set with the preprocessed health indicators as input and the corresponding health status data as output.

[0142] 5) Based on the extreme learning machine improved by kernel function and autoencoder, a TS-KELM-AE network model is constructed, and the TS-KELM-AE network model is trained using the training set to obtain a lithium battery health status estimation model.

[0143] The TS-KELM-AE model is designed to adapt to SOH estimation under large-scale feature input. The first stage focuses on deep feature extraction, and the second stage is used to make accurate SOH estimation decisions, thereby effectively improving the accuracy and robustness of the SOH estimation model.

[0144] 6) Extracting the health indicators of the lithium battery to be tested, and inputting the health indicators of the lithium battery to be tested into the lithium battery health status estimation model to estimate the health status data of the lithium-ion battery.

[0145] By extracting the j+1,…,nth charge and discharge cycle data of the lithium battery health indicator sequence aged at different temperatures As input, the established lithium battery health status estimation model is applied for estimation, where T is 25℃, 35℃, 45℃, and 55℃ respectively.

[0146] Example 12:

[0147] A method for estimating the state of health of a fast-charged lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is described in Example 11. Furthermore, the health state data sequences corresponding to different fast-charge cycle processes are shown below:

[0148]

[0149] Where i is the fast charge cycle index, i = 1, 2, ..., n, n is the total number of fast charge cycles, H i is the battery health status of the i-th fast charge cycle, C i is the maximum discharge capacity of the lithium battery in the i-th fast charge cycle, and C is the rated capacity of the lithium battery.

[0150] Example 13:

[0151] A method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 11 to 12. Furthermore, the capacity data within the charging step is as follows:

[0152]

[0153] Where i is the fast charge cycle index, n is the total number of cycles, k is the charge step index, K is the total number of charge steps, t is the time, and e is the total time. i,k is the charging step capacity of the kth charging step in the i-th fast charging cycle.nom is the nominal voltage. i,k,t is the current value at time t of the kth charging step in the i-th fast charging cycle. V i,k,t is the voltage value at time t of the kth charging step in the i-th fast charging cycle.

[0154] Example 14:

[0155] The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 11 to 13. Furthermore, the method for extracting health indicators from the charging step capacity data in the time domain includes fast Fourier transform.

[0156] The health indicators extracted from the time domain of the charging step content data include: mean, standard deviation, maximum, minimum, root mean square, kurtosis, skewness, median, energy, peak frequency, spectral entropy, root mean square spectrum, average spectrum, bandwidth and average frequency.

[0157] Example 15:

[0158] The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 11 to 14. Furthermore, the method for extracting health indicators from the charging step capacity data in the frequency domain includes wavelet analysis.

[0159] The health indicators obtained by extracting the charging step capacity data from the frequency domain include: average energy, maximum energy, energy standard deviation, low-frequency energy, medium-frequency energy, high-frequency energy, wavelet energy, and wavelet correlation coefficient.

[0160] The low-frequency energy refers to energy with a frequency less than or equal to 50 Hz.

[0161] The intermediate frequency energy refers to the energy with a frequency within the range of 50 Hz to 200 Hz.

[0162] The high-frequency energy refers to energy with a frequency greater than 200 Hz.

[0163] Example 16:

[0164] A method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 11 to 15. Furthermore, the calculation formula for the Pearson correlation analysis is as follows:

[0165]

[0166] Where i is the fast charge cycle index, n is the total number of cycles, p is the health indicator index, and ξ(p) is the Pearson correlation coefficient between the pth health indicator and the health status data sequence. i is the battery health status of the i-th fast charge cycle, F p,i is the pth health indicator of the i-th fast charging cycle.

[0167] Among them, the battery health status H i The mean value E(H i ) and health index F p,i The mean value E(F p,i ) is as follows:

[0168]

[0169] Example 17:

[0170] A method for estimating the health status of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 11 to 16. Furthermore, the preprocessing includes normalization processing to normalize the data between the interval [-1, 1].

[0171] The normalization process includes maximum and minimum value normalization process, as shown below:

[0172]

[0173] Where x0(n) is the data to be processed. max is the maximum value in the data x0(n), x0(n) min is the minimum value in the data x0(n). * The data are normalized.

[0174] Example 18:

[0175] A method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 11 to 17. Furthermore, the steps of constructing the TS-KELM-AE network model are as follows:

[0176] 5.1) Construct the KELM model.

[0177] The output of the KELM model is as follows:

[0178]

[0179] Where, F KELM is the output result of the KELM model. is the output weight of the KELM model. m x(n), M are health indicator sequences [F p,1 ,F p,2 ,…,F p,n ]The output vector form and spatial matrix form of hidden nodes. is the output weight. T is the Moore-Penrose generalized inverse matrix of the spatial matrix M. U is the identity matrix, λ is the regularization coefficient, and H is the battery health state sequence [H1,H2,…,H n ]. K is the kernel function.

[0180] The radial basis function kernel K(x,x(n)) is as follows:

[0181]

[0182] Where σ is the kernel function width. x represents the health indicator sequence to be evaluated. x(n) represents a health indicator sequence in the training set.

[0183] 5.2) Substitute the autoencoder into the output of the KELM model F KELM , construct the KELM-AE model.

[0184] The output weights of the KELM-AE model As shown below:

[0185]

[0186] 5.3) Set the TS-KELM-AE network model to include L KELM-AE hidden layers, and based on the output weights of the KELM-AE model The output weight of the Lth layer KELM-AE hidden layer of the TS-KELM-AE network model is obtained by recursive calculation and the output M of the Lth KELM-AE hidden layer L , as shown below:

[0187]

[0188] Where λ L-1 M represents the regularization coefficient of the L-1th KELM-AE hidden layer. L-1 g represents the output of the L-1th KELM-AE hidden layer. * is the activation function.

[0189] 5.4) The output weight of the Lth layer of the TS-KELM-AE network model and the output M of the hidden layer L Bring in the output result F of the KELM model KELM , the TS-KELM-AE network model is constructed.

[0190] Example 19:

[0191] A method for estimating the health status of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE is provided. The main technical content is shown in any one of Examples 11 to 18. Furthermore, the autoencoder includes an encoding part and a decoding part.

[0192] The encoding part maps the input data into a latent low-dimensional space, effectively compressing the input data into a compact representation.

[0193] The decoding part reconstructs the original input data from the encoder as accurately as possible, capturing the essential characteristics of the data.

[0194] The coding part is as follows:

[0195] Z=g * (E·X+B) (13)

[0196] Where X represents the input health indicator sequence [F p,1 ,F p,2 ,…,F p,n ]. Z represents the compressed data of the health indicator sequence X. E represents the encoded weight matrix. B represents the encoded bias matrix. g * is the activation function.

[0197] The decoding part is as follows:

[0198]

[0199] Where, Denotes the reconstructed data of the compressed data Z. D denotes the decoded weight matrix. Represents the bias matrix for decoding.

[0200] Example 20:

[0201] A method for estimating the health status of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE, the main technical content of which can be found in any one of Examples 11 to 19. Furthermore, the TS-KELM-AE network model is trained using the mean square error between the health status data sequence in the training set and the predicted value as the loss function.

[0202] Example 21:

[0203] See also Figures 1 to 8 The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE includes the following main technical contents:

[0204] S1. Obtain the current and voltage data sequence of each charging step during the nine-step fast charging cycle, and extract the health status data sequence corresponding to each charging cycle.

[0205] S2. Calculate the capacity data within the charging step based on the current and voltage sequence within the charging step.

[0206] S3. Extract health indicators from the charging step capacity data in the time domain and frequency domain, and calculate the Pearson correlation analysis between the health indicators and SOH.

[0207] S4. Normalize and preprocess the health indicators and health status sequences and use them as input data.

[0208] S5. A TS-EKLM-AE network model is designed based on the ELM improved by kernel function and autoencoder as the basic structure, and a SOH estimation model is constructed with health indicators as input and the health status of fast-charging lithium-ion batteries as output.

[0209] S6. Use the constructed health status estimation model to estimate the SOH of fast-charged lithium-ion batteries at different temperatures.

[0210] The specific contents of step S1 are as follows:

[0211] The charging current and voltage sequences of each charging step in the nine-step fast charging cycle are [I i,k,1 ,I i,k,2 ,…,I i,k,t …,I i,k,e ] and [V i,k,1 ,V i,k,2 ,…,V i,k,t …,V i,k,e ], where I i,s,t is the current collected at time t in the k-th fast charging stage in the i-th (i=1,2,…,n) cycle, and e is the end time. i,s,t is the voltage collected at time t in the k-th fast charge phase of the i-th cycle, e is the end time, k = 1,…, 9. The health status data sequence of the lithium battery during the cyclic charge and discharge process is H1, H2,…, H n , the measured data Among them H i is the battery health status of the i-th charge and discharge cycle, n is the total number of charge and discharge cycles, C i is the maximum discharge capacity of the lithium battery in the i-th charge and discharge process, and C is the rated capacity of the lithium battery.

[0212] The specific contents of step S2 are as follows:

[0213] According to the current and voltage sequence in the charging step, the capacity of the charging step can be calculated as: Among them, Qi,k is the charging capacity of the kth fast charging stage in the i-th cycle, n is the total number of cycles, V nom is the nominal voltage.

[0214] The specific contents of step S3 are as follows:

[0215] The data of health indicators extracted from time domain and frequency domain along with the cycle charge and discharge process are [F p,1 ,F p,2 ,…,F p,n ], where p is the feature number, p=1,…,23, and n is the number of charge and discharge cycles.

[0216] The health indicators extracted in the time domain and frequency domain include a total of 23 types. Among them, the extracted time domain features include mean, standard deviation, maximum value, minimum value, root mean square (RMS), kurtosis, skewness, median and energy. The time domain features extracted using fast Fourier transform include peak frequency, spectral entropy, root mean square spectrum, average spectrum, bandwidth and average frequency. The energy in the frequency domain features includes average energy, maximum energy and energy standard deviation. The frequency band energy is divided into low-frequency energy (≤50HZ), medium-frequency energy (50~200HZ) and high-frequency energy (>200HZ). Based on wavelet analysis, wavelet energy and wavelet correlation coefficient are also included.

[0217] For each health indicator sequence [F p,1 ,F p,2 ,…,F p,n ] are respectively related to the health state sequence [H1,H2,…,H n ] to calculate the Pearson correlation coefficient [ξ(1),ξ(2),…,ξ(p)], where the Pearson correlation analysis is For each health indicator F p Correlation with lithium battery health status,

[0218] The specific contents of step S4 are as follows:

[0219] The data is preprocessed using the maximum and minimum normalization method, and the data is normalized to the interval [-1, 1]. The normalization formula is: x(n)=[F p,1 ,F p,2 ,…,F p,n ]Preprocessing sequence, x(n) max is the maximum value in the corresponding sequence x(n), x(n) min is the minimum value in the corresponding sequence x(n).

[0220] The specific contents of step S5 are as follows:

[0221] The TS-KELM-AE model is designed to adapt to SOH estimation under large-scale feature input. The first stage focuses on deep feature extraction, and the second stage is used to make accurate SOH estimation decisions, thereby effectively improving the accuracy and robustness of the SOH estimation model. The output result F of KELM is KELM The calculation process is as follows:

[0222]

[0223] Among them, m x(n) and M are the input health indicator sequence [F p,1 ,F p,2 ,…,F p,n ]The output vector form and spatial matrix form of the hidden node, is the output weight, M T is the Moore-Penrose generalized inverse of M, U is the identity matrix, λ is the regularization coefficient, and H is the output health state sequence [H1,H2,…,H n ], kernel function is the radial basis function kernel, and σ is the kernel function width.

[0224] The autoencoder can be divided into two main parts: (1) the encoding part, which maps the input data to a latent low-dimensional space, effectively compressing the input data into a compact representation. (2) the decoding part, which reconstructs the original input data from the encoder as accurately as possible, capturing the essential characteristics of the data. The calculation formula is as follows:

[0225] Z=g * (E·X+B)

[0226]

[0227] Where Z represents the compressed form of the input health indicator sequence X generated by the encoder, Denotes the reconstructed form of the input data Z produced by the decoder. E and D denote the weight matrices for encoding and decoding, respectively. B and Denotes the corresponding bias matrix, denotes the activation function. Substitute the encoding and decoding calculation formulas into F KELM , we can further get the output weight as:

[0228]

[0229] Assuming that TS-KELM-AE contains L+1 hidden layers, according to the formula Then in the first stage KELM-AE, the Lth layer output weight and hidden layer output M L It can be expressed as:

[0230]

[0231] Then the SOH decision result of KELM-AE in the second stage can be obtained through M L and Bring in F KELM Calculated.

[0232] Take the first j samples of the input health index and health status sequence, j = 1, 2, ..., n-1 and As the training set, take the j+1, j+2,…,n sample values and As the test set, the designed TS-KELM-AE network is used to build a lithium battery health status estimation model, and the root mean square error of the true value and the estimated value of the health status is used as the loss function for training.

[0233] The specific contents of step S6 are as follows:

[0234] By extracting the j+1,…,nth charge and discharge cycle data of the lithium battery health indicator sequence aged at different temperatures As input, the established lithium battery health status estimation model is applied for estimation, where T is 25℃, 35℃, 45℃, and 55℃ respectively.

[0235] Example 22:

[0236] See also Figures 1 to 8 The health status estimation method of fast-charging lithium-ion batteries based on time-frequency characteristics and TS-KELM-AE includes the following main technical contents:

[0237] The current and voltage distribution of the 9-step fast charge in one cycle is as follows: Figure 2 As shown. This protocol attempts to simulate the charging behavior under real-world usage scenarios and evaluate the degradation characteristics at different temperatures. It is worth noting that after setting a fixed SOC increment value for each charging step, the battery is charged by applying a constant current charging stage (CC) with different rates, for example, D1 is in the SOC range of 0%-8%, D2 is in the SOC range of 8%-20%, D3 is in the SOC range of 20%-30%, until D8 completes the SOC charge to 297. Table 1 summarizes the more detailed protocol of the 9-step fast charging SOC distribution, where "+" indicates charging.

[0238] In order to verify the reliability of the extracted health indicators, we performed Pearson correlation analysis between the health indicators extracted from the charging step capacity data and the battery SOH in the time domain and frequency domain. For convenience, these health indicators are marked as F1-F23 in this invention. The correlation heat map results are shown as follows: Figure 3 As shown. Figure 3 As can be observed, the color difference fluctuations in the correlation between the F1-F23 health indicators and SOH are relatively small as the temperature increases. This indicates that the feature extraction process of the present invention has a good correlation with SOH under different temperature conditions. Through our innovative method, the extracted features exhibit relatively stable performance and can more accurately reflect the degradation trend of SOH at different temperatures. This feature extraction process effectively improves the accuracy of battery health estimation, especially in the relatively small fluctuations in the correlation results at 25°C and 45°C, where the temperature changes are large.

[0239] In order to verify the superiority of the TS-KELM-AE model of the present invention, we conducted comparative experiments with ELM, SVM, LSTM, KELM and other methods.

[0240] The present invention uses the mean absolute error (MAE), root mean square error (RMSE) and determination coefficient (R 2 )Three evaluation indicators are used to measure the effectiveness and superiority of the method. Figure 5-8 The comparative experiments of SOH estimation results at four different temperatures are shown in Table 2.

[0241] from Figures 5 to 8 It can be clearly seen that in the early stages of SOH estimation, other methods attempt to estimate by tracking the SOH degradation trend observed in the training data. However, after the SOH regeneration point, especially in the later stages, the deviation of the estimation results increases significantly. In contrast, the TS-KELM-AE model proposed in this paper is able to maintain high accuracy throughout the entire estimation cycle, especially during the SOH change process. Its estimation results are always closer to the actual SOH changes, demonstrating excellent SOH tracking capabilities.

[0242] exist Figure 5 The SOH estimation results of all methods show obvious fluctuations. One possible reason is that SOH regeneration points frequently appear in the training data, which increases the variability and uncertainty of the data during feature extraction. Figure 6 、 7 Compared with the results in 8, Figure 5There are significant differences in the SOH degradation trends of the training data and the test data in the data. Specifically, the training data shows a relatively gentle SOH degradation trend, while the test data shows a steeper degradation trend. This phenomenon poses a great challenge to SOH estimation. It is satisfying that the estimation results of the TS-KELM-AE method of the present invention always fluctuate around the true value, while the estimation results of other methods are far away from the true value and show a large deviation. This performance further verifies the significant advantages of the TS-KELM-AE method proposed in the present invention in feature extraction and nonlinear mapping capabilities.

[0243] One of the core innovations of the TS-KELM-AE model designed in this paper lies in its first-stage stacked KELM-AE structure. This structure is a multi-layer feature extraction network based on deep learning, which can effectively capture the high-order nonlinear relationships in the battery degradation process. By introducing kernel techniques, the stacked KELM-AE structure significantly enhances the regression capability of nonlinear data, thereby improving the accurate characterization of battery SOH degradation patterns. This method is not only innovative in theory but also demonstrates broad application potential in practical applications, especially in the fields of battery management, predictive maintenance, and performance optimization, and has important technical value.

[0244] Table 1 Detailed protocol for 9-step fast charging SOC distribution

[0245]

[0246] Table 2 Evaluation index results

[0247]

Claims

1. A fast-charge lithium-ion battery health status estimation method based on time-frequency characteristics and TS-KELM-AE is characterized by: The following steps are involved: 1) Obtain the current and voltage data sequences of each charging step of the lithium battery during different fast charging cycles, and calculate the health status data sequences corresponding to different fast charging cycles. 2) Calculate the charging step capacity data based on the current and voltage data sequence of each charging step; 3) Extract health indicators from the charging step capacity data in the time domain and frequency domain, and perform Pearson correlation analysis on the extracted health indicators and health status data sequence to obtain the Pearson correlation between each health indicator and the health status of the lithium battery. Based on the value of Pearson correlation, the corresponding relationship between health indicators and health status data is constructed; 4) Preprocessing the extracted health indicator and health status data series, and constructing a training set with the preprocessed health indicators as input and the corresponding health status data as output; 5) Based on the extreme learning machine improved by kernel function and autoencoder, a TS-KELM-AE network model is constructed, and the TS-KELM-AE network model is trained using the training set to obtain a lithium battery health status estimation model; 6) Extracting the health indicators of the lithium battery to be tested, and inputting the health indicators of the lithium battery to be tested into the lithium battery health status estimation model to estimate the health status data of the lithium-ion battery.

2. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1 is characterized in that: The health status data sequences corresponding to the different fast charging cycles are as follows: Where i is the fast charge cycle index, i = 1, 2, ..., n, n is the total number of fast charge cycles, H i is the battery health status of the i-th fast charge cycle, C i is the maximum discharge capacity of the lithium battery in the i-th fast charge cycle, and C is the rated capacity of the lithium battery.

3. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The capacity data of the charging step is as follows: Where i is the fast charge cycle index, n is the total number of cycles; k is the charging step index, K is the total number of charging steps; t is the time, e is the total time; Q i,k is the charging step capacity of the kth charging step in the i-th fast charging cycle; V nom is the nominal voltage; I i,k,t is the current value at time t of the kth charging step in the i-th fast charging cycle; V i,k,t is the voltage value at time t of the kth charging step in the i-th fast charging cycle.

4. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The method for extracting health indicators from the charging step capacity data in the time domain includes fast Fourier transform; The health indicators extracted from the time domain of the charging step content data include: mean, standard deviation, maximum, minimum, root mean square, kurtosis, skewness, median, energy, peak frequency, spectral entropy, root mean square spectrum, average spectrum, bandwidth and average frequency.

5. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The method for extracting health indicators from the charging step capacity data in the frequency domain includes wavelet analysis; The health indicators extracted from the frequency domain of the charging step content data include: average energy, maximum energy, energy standard deviation, low-frequency energy, medium-frequency energy, high-frequency energy, wavelet energy, and wavelet correlation coefficient; The low-frequency energy refers to energy with a frequency less than or equal to 50 Hz; The medium frequency energy refers to the energy with a frequency within the range of 50HZ to 200HZ; The high-frequency energy refers to energy with a frequency greater than 200 Hz.

6. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The calculation formula of the Pearson correlation analysis is as follows: Where i is the fast charge cycle index, n is the total number of cycles; p is the health indicator index; ξ(p) is the Pearson correlation coefficient between the pth health indicator and the health status data sequence; H i is the battery health status of the i-th fast charge cycle, F p,i is the pth health indicator of the i-th fast charging cycle; Among them, the battery health status H i The mean value E(H i ) and health index F p,i The mean value E(F p,i ) is as follows:

7. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The preprocessing includes normalization processing; The normalization process includes maximum and minimum value normalization process, as shown below: In the formula, x0(n) is the data to be processed; x0(n) max is the maximum value in the data x0(n), x0(n) min is the minimum value in the data x0(n); x0(n) * The data are normalized.

8. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The steps of constructing the TS-KELM-AE network model are as follows: 5.1) Constructing the KELM model; The output of the KELM model is as follows: Where, F KELM is the output result of the KELM model; is the output weight of the KELM model; m x(n) , M are health indicator sequences [F p,1 ,F p,2 ,…,F p,n ] Output vector form and spatial matrix form of hidden nodes; is the output weight; M T is the Moore-Penrose generalized inverse matrix of the spatial matrix M; U is the identity matrix, λ is the regularization coefficient, and H is the battery health state sequence [H1,H2,…,H n ]; K is the kernel function; The radial basis function kernel K(x,x(n)) is as follows: Where σ is the kernel function width; x represents the health indicator sequence to be evaluated; x(n) represents a health indicator sequence in the training set; 5.2) Substitute the autoencoder into the output of the KELM model F KELM , construct the KELM-AE model; The output weights of the KELM-AE model As shown below: 5.3) Set the TS-KELM-AE network model to include L KELM-AE hidden layers, and based on the output weights of the KELM-AE model The output weight of the Lth layer KELM-AE hidden layer of the TS-KELM-AE network model is obtained by recursive calculation and the output M of the Lth KELM-AE hidden layer L , as shown below: Where λ L-1 represents the regularization coefficient of the L-1th KELM-AE hidden layer; M L-1 surface shows the output of the L-1th KELM-AE hidden layer; g* is the activation function; 5.4) The output weight of the Lth layer of the TS-KELM-AE network model and the output M of the hidden layer L Bring in the output result F of the KELM model KELM , the TS-KELM-AE network model is constructed.

9. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 8, characterized in that: The autoencoder includes an encoding part and a decoding part; The coding part is as follows: Z=g * (E·X+B) (13) Where X represents the input health indicator sequence [F p,1 ,F p,2 ,…,F p,n ]; Z represents the compressed data of the health indicator sequence X; E represents the encoded weight matrix; B represents the encoded bias matrix; g * is the activation function; The decoding part is as follows: Where, represents the reconstructed data of the compressed data Z; D represents the decoded weight matrix; Represents the bias matrix for decoding.

10. The method for estimating the state of health of a fast-charging lithium-ion battery based on time-frequency characteristics and TS-KELM-AE according to claim 1, characterized in that: The TS-KELM-AE network model is trained using the mean square error between the health status data sequence in the training set and the predicted value as the loss function.

Images