SOH evaluation method based on time-frequency feature fusion and LSTM

Abstract

The invention belongs to the technical field of lithium batteries, and relates to an SOH (state of health) evaluation method based on time-frequency feature fusion and LSTM (long short term memory), which comprises the following steps: respectively extracting time-domain health indexes from three kinds of lithium battery charge-discharge cycle working condition data of complete charge-discharge cycle, partial charge cycle and partial charge-discharge cycle; performing wavelet packet decomposition on the lithium battery charge-discharge cycle working condition data to obtain sub-frequency bands; calculating the sum of the energy of each sub-frequency band by adopting the Parseval theorem, and taking the sum as a frequency domain characteristic signal; calculating the Pearson's correlation coefficients of the time domain health indexes and the SOH, and screening the time domain health indexes according to the sizes of the Pearson's correlation coefficients; taking the screened time domain health indexes as the input of an LSTM network model; the LSTM network model is trained, and an Adam algorithm is adopted to optimize the LSTM network model; and the full connection layer of the LSTM network model outputs a lithium battery SOH estimation value. The method has the beneficial effect that the potential reliability problem of certain health indexes under different operation conditions is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of lithium batteries, and particularly relates to an SOH evaluation method based on time-frequency feature fusion and LSTM. BACKGROUND

[0002] As one of the most widely used battery technologies, lithium-ion batteries have experienced rapid development in recent years. As the use time of lithium batteries increases and the number of charge and discharge cycles increases, the active material inside the battery will gradually lose, which may lead to a decline in battery performance, which is manifested as a decrease in charge and discharge efficiency, capacity attenuation, and poor thermal management performance. This inevitable degradation can even cause catastrophic events. IEEE Standard 1188-1996 clearly stipulates that when the capacity of the power battery decreases to 80%, i.e., SOH (State of Health) is less than 80%, the battery should be replaced. As an indispensable indicator of the battery management system, accurate SOH estimation can help predict the remaining life of the battery. This is crucial for the development of battery replacement plans, optimization of battery maintenance strategies, and overall management of battery systems.

[0003] In order to avoid the limitations of model-driven methods that are highly dependent on their accurate description of the real world, data-driven methods for estimating battery SOH have become increasingly popular. "Journal of Energy Storage, 2023; 74 (Part B), 109370" proposes to use random segments of the charge curve as input, and uses a multi-layer long short-term memory network with attention mechanism as the basic learner. Multiple basic learners are trained using different parts of the charge curve segment, and finally a BP neural network is used to integrate the health state estimation value of each basic learner. "Energy, 2024; 311; 133418" proposes a BiLSTM-Transformer model to estimate the SOH of lithium-ion batteries during fast charging, combining the advantages of BiLSTM in capturing long-term dependencies and the advantages of Transformer in parallel processing and global context understanding. LSTM (Long Short-Term Memory) through its unique gating mechanism, through in-depth mining and analysis of a large amount of data, can effectively process data with time series characteristics, such as voltage, current, temperature, and other time series information of the battery. This makes the LSTM network have a natural advantage in predicting the SOH of the battery, and can more accurately predict future trends and results.

[0004] However, when selecting battery health indicators for battery charging and discharging operation, complete time series signals are often needed, such as constant current charging time, constant voltage charging time and the like in a charging and discharging cycle. However, in actual work, the charging and discharging process of the lithium battery often cannot be strictly carried out according to the ideal constant current charging, constant voltage charging and constant current discharging cycle. At present, most of the battery health indicators are time domain characteristics, which enable us to understand the global characteristics of the battery, but cannot determine the local behavior. In the superimposed cycle of the partial charging cycle and the variable charging and discharging mode, these characteristic parameters cannot be completely captured, so more reliance on raw data information is needed. With the evolution of the operation mode of the lithium ion battery, the health characteristics that previously showed strong linear correlation with SOH may fail, because the time domain health indicators are directly susceptible to the influence of the charging and discharging rate, temperature fluctuation and operation mode conversion change. SUMMARY

[0005] Technical problems to be solved In view of the above-mentioned defects and deficiencies of the prior art, the present application provides a SOH evaluation method based on time-frequency feature fusion and LSTM, which solves the technical problem of how to avoid the invalidation of health indicators caused by the change of battery operation mode under actual working conditions on the basis of accurately estimating the SOH of the lithium battery.

[0006] Technical scheme In order to achieve the above-mentioned purpose, the main technical scheme adopted by the present application comprises: In a first aspect, the present application provides a SOH evaluation method based on time-frequency feature fusion and LSTM, comprising: Extracting time domain health indicators from complete charging and discharging cycle, partial charging cycle and partial charging and discharging cycle lithium battery charging and discharging cycle working condition data respectively, for reflecting the aging degree and health condition of the battery; Performing wavelet packet decomposition on the lithium battery charging and discharging cycle working condition data to obtain sub-frequency bands; Calculating the sum of the energy of each sub-frequency band as a frequency domain characteristic signal by using the Parseval theorem; Calculating the Pearson correlation coefficient of the time domain health indicators and the SOH, and screening the time domain health indicators according to the size of the Pearson correlation coefficient; Taking the screened time domain health indicators as the input of the LSTM network model; Training the LSTM network model and optimizing the LSTM network model by using the Adam algorithm; The full connection layer output of the LSTM network model is the SOH estimation value of the lithium battery under the complete charging and discharging cycle, the partial charging cycle and the partial charging and discharging cycle lithium battery working condition.

[0007] Optionally, the time-domain health indicators are extracted from the three lithium battery charge-discharge cycle working condition data, including: The voltage, current and temperature parameters are obtained from the three lithium battery charge-discharge cycle working condition data, to extract the time-domain health indicators, including constant current charging time, constant voltage charging time, maximum charging temperature, average charging temperature, total battery input power, heat accumulation degree and thermal-electricity ratio.

[0008] Optionally, the lithium battery charge-discharge cycle working condition data is decomposed by wavelet packet to obtain sub-frequency bands, including: The decomposition level of wavelet packet decomposition is 3, and the third layer has 8 nodes. The frequency bands of the voltage, current and temperature parameters in the lithium battery charge-discharge cycle working condition data are decomposed by the tree diagram to obtain 8 different sub-frequency bands.

[0009] Optionally, the sum of the energy of each sub-frequency band is calculated as a frequency domain feature signal using the Parseval theorem, including: Let d(j, k) be the wavelet packet decomposition coefficient of the jth layer and the kth sub-frequency band, and N be the length of the lithium battery charge-discharge cycle data. The energy E(j, k) of the jth layer and the kth sub-frequency band is: E(j, k)=∑[d(j,k,n)]², n=1,2,...,N; Where d(j, k, n) represents the nth coefficient in the jth layer and the kth sub-frequency band.

[0010] Optionally, the Pearson correlation coefficient of the time-domain health indicators and SOH is calculated according to the following formula PCC : ; Where X and Y represent the reference values of the time-domain health indicators and SOH, and represent the average values of the time-domain health indicators and SOH, respectively.

[0011] Optionally, the time-domain health indicators are screened according to the size of the Pearson correlation coefficient, including: In the case where the Pearson correlation coefficient is less than a preset value, the corresponding time-domain health indicators are excluded; Where the preset value is 0.9.

[0012] Optionally, the method further comprises: Setting the hyperparameters of the LSTM network model, including the number of hidden layer neurons, the number of iterations and the learning rate; The number of hidden layer neurons is 100, the number of iterations for each training of the LSTM network model is 1000, the initial learning rate is 0.01, and the learning rate reduction factor is 0.1 / 100 times.

[0013] In a second aspect, the present application provides a computer readable storage medium, which stores a computer program, and the program realizes the SOH evaluation method based on time-frequency feature fusion and LSTM in any one of the first aspect when executed.

[0014] In a third aspect, the present application provides a storage device, which comprises a storage medium and a processor, and the storage medium stores a computer program, and the program realizes the SOH evaluation method based on time-frequency feature fusion and LSTM in any one of the first aspect when executed by the processor. Advantages

[0015] The SOH evaluation method based on time-frequency feature fusion and LSTM of the present application fuses the time domain and frequency domain features of the lithium battery charge and discharge operation data, and estimates the SOH of the lithium battery under multiple time scales and partial cycles. The wavelet packet decomposition technology is used to analyze the characteristics of the voltage signal, current signal and temperature signal of the lithium ion battery in the charge and discharge cycle process. The energy distribution between frequency bands is quantified by applying the Parseval theorem, and the dynamic energy characteristics under various operating states are revealed. The network model is provided with interpretable input, and the potential mechanism of energy change is clarified. The sum of energy under different frequency spectrums after wavelet packet decomposition is used as the frequency domain feature signal, and this method captures the overall energy characteristics of the current, voltage and temperature signals in the frequency range represented by each layer after decomposition. The limitation of ideal constant charge and discharge cycle in practical application is solved, which experiences partial or variable cycle mode. In order to better reflect the actual operating conditions, the SOH prediction performance under three different scenarios is evaluated: complete cycle, partial cycle and variable partial cycle, which solves the potential reliability problem of some health indicators under different operating conditions. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 A flowchart of the SOH evaluation method based on time-frequency feature fusion and LSTM provided by the embodiment of the present application is shown in the figure. Figure 2 A SOH estimation process diagram provided by the embodiment of the present application is shown in the figure. Figure 3 A wavelet packet decomposition diagram provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0017] For a better explanation of the present application, in order to facilitate understanding, the following will be described in detail by specific embodiments in combination with the drawings. Although the exemplary embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a clearer, more thorough understanding of the present application and to fully convey the scope of the present application to those skilled in the art.

[0018] In a first aspect, with reference to Figure 1 and Figure 2 The embodiment provides a SOH evaluation method based on time-frequency feature fusion and LSTM, comprising: S1, extracting time-domain health indicators from full charge-discharge cycle, partial charge cycle and partial charge-discharge cycle three lithium battery charge-discharge cycle working condition data respectively, for reflecting the aging degree and health condition of the battery.

[0019] S2, wavelet packet decomposition is performed on the lithium battery charge-discharge cycle working condition data to obtain sub-frequency bands.

[0020] S3, the sum of the energy of each sub-frequency band is calculated as a frequency domain feature signal by using the Parseval theorem.

[0021] S4, the Pearson correlation coefficient of the time-domain health indicators and the SOH is calculated, and the time-domain health indicators are screened according to the size of the Pearson correlation coefficient.

[0022] S5, the screened time-domain health indicators are taken as the input of the LSTM network model.

[0023] S6, the LSTM network model is trained, and the Adam algorithm is used to optimize the LSTM network model.

[0024] S7, the full connection layer output of the LSTM network model is the lithium battery SOH estimation value under the three lithium battery working conditions of full charge-discharge cycle, partial charge cycle and partial charge-discharge cycle.

[0025] In the technical solution of the present application, wavelet packet decomposition is used to realize frequency domain feature extraction. Compared with fast Fourier transform, the extracted frequency domain health indicators have stronger correlation, providing interpretable input for the network model. The problem of invalidation of health indicators caused by changes in lithium battery working conditions in actual operation mode is solved. In order to verify the performance of the proposed evaluation scheme, three operation modes of lithium ion battery are designed: full charge-discharge cycle, partial charge cycle and partial charge-discharge cycle. The actual prediction error of the whole cycle is kept below 2%, and the root mean square error under the three operation modes is 0.41%, 0.80% and 0.63% respectively.

[0026] Optionally, time-domain health indicators are extracted from the three types of lithium battery charge-discharge cycle working condition data, including: Voltage, current and temperature parameters are obtained from the three types of lithium battery charge-discharge cycle working condition data to extract time-domain health indicators, including constant-current charging time, constant-voltage charging time, maximum charging temperature, average charging temperature, total battery input energy, degree of heat accumulation, and thermal-electricity ratio.

[0027] These indicators reflect the performance changes of the battery during the charge-discharge process from different aspects, comprehensively cover the health-related characteristics of the battery, effectively quantify the battery aging and health status, provide rich and valuable data basis, and thus more accurately evaluate the actual state of the lithium battery.

[0028] Optionally, in combination with Figure 3 The lithium battery charge-discharge cycle working condition data is decomposed by wavelet packet to obtain sub-frequency bands, including: The decomposition level of wavelet packet decomposition is 3, and the third layer has 8 nodes. The frequency bands of voltage, current and temperature parameters in the lithium battery charge-discharge cycle working condition data are decomposed by tree diagram to obtain 8 different sub-frequency bands.

[0029] For current and voltage signals, the low-frequency part may represent the average current and voltage level of the battery during the charge-discharge process, thus better reflecting the health characteristics of the battery. The high-frequency part mainly reflects the instantaneous fluctuations or noise in the lithium battery charge-discharge process. For temperature signals, the low-frequency part may represent the average temperature change of the battery during the working process, while the high-frequency part may reflect the local temperature fluctuations or thermal noise inside the battery.

[0030] This fine frequency band decomposition method can analyze the lithium battery charge-discharge cycle working condition data at different frequency levels and extract the battery state information contained in different frequency components. It helps to better understand the change rules of electrical characteristics during the battery charge-discharge process, thereby laying a foundation for extracting more accurate and effective frequency domain feature signals and improving the accuracy of battery state evaluation.

[0031] Optionally, the sum of the energy of each sub-frequency band is calculated as the frequency domain feature signal using the Parseval theorem, including: Let d(j, k) be the wavelet packet decomposition coefficient of the jth layer and the kth sub-frequency band, and N be the length of the lithium battery charge-discharge cycle data. Then the energy E(j, k) of the jth layer and the kth sub-frequency band is: E(j, k)=∑[d(j, k, n)]², n=1,2,...,N; wherein d(j, k, n) represents the nth coefficient in the jth layer and the kth sub-band.

[0032] The time-domain health indicators are constant-current charging time, constant-voltage charging time and thermoelectric ratio, and the frequency-domain characteristic signals are current signals of 0 Hz-0.625 Hz, voltage signals of 0.625 Hz-1.25 Hz and temperature signals of 0.625 Hz-1.25 Hz.

[0033] The sum of the energy of the sub-band calculated by the Parseval theorem is used as the frequency-domain characteristic signal, which accurately quantifies the contribution of each sub-band to the overall signal energy and can mine the energy distribution characteristics of the lithium battery charging and discharging signal in different frequency domains. These frequency-domain characteristic signals can reflect the internal physical changes of the battery under different working conditions.

[0034] Optionally, the Pearson correlation coefficient of the time-domain health indicators and the SOH is calculated according to the following formula PCC : ; wherein X and Y represent the reference values of the time-domain health indicators and the SOH, and respectively represent the average values of the time-domain health indicators and the SOH.

[0035] Optionally, the time-domain health indicators are screened according to the size of the Pearson correlation coefficient, including: in the case where the Pearson correlation coefficient is less than a preset value, the corresponding time-domain health indicator is removed; wherein the preset value is 0.9.

[0036] The health characteristic variables with weak correlation with the SOH are identified and removed to reduce the data dimension and improve the analysis efficiency.

[0037] Optionally, the method further includes: setting the hyperparameters of the LSTM network model, including the number of hidden layer neurons, the number of iterations and the learning rate; The number of hidden layer neurons is 100, the number of iterations for each training of the LSTM network model is 1000, the initial learning rate is 0.01, and the learning rate reduction factor is 0.1 / 100 times.

[0038] Reasonably setting the hyperparameters of the LSTM network model, such as the number of hidden layer neurons, the number of iterations, and the learning rate, can help optimize the performance of the LSTM network model. Reasonable setting of the number of hidden layer neurons, the number of iterations, and the learning rate reduction factor enables the model to accurately learn the potential relationship between the time domain health indicators and the SOH in complex lithium battery data, improves the generalization ability and prediction accuracy of the model, more stably and efficiently outputs the SOH estimation value of the lithium battery under different working conditions, and improves the effectiveness and reliability of the model in practical application.

[0039] In a second aspect, an embodiment of the present application provides a computer readable storage medium having a computer program stored thereon, wherein the program implements the SOH evaluation method based on time-frequency feature fusion and LSTM when executed.

[0040] In a third aspect, an embodiment of the present application provides a storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, and the program implements the SOH evaluation method based on time-frequency feature fusion and LSTM when executed by the processor.

[0041] Those skilled in the art will appreciate that embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage devices, etc.) containing computer usable program code.

[0042] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application should also include these modifications and variations.

[0043] Although the embodiments of the present application have been shown and described above, it should be understood that the above-described embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments within the scope of the present application.

Claims

1. A SOH evaluation method based on time-frequency feature fusion and LSTM, characterized in that: include: Time domain health indicators are extracted from three types of lithium battery charge and discharge cycle data: full charge and discharge cycle, partial charge cycle, and partial charge and discharge cycle, to reflect the aging degree and health status of the battery; Perform wavelet packet decomposition on the lithium battery charge and discharge cycle data to obtain sub-bands; Parseval's theorem is used to calculate the sum of the energy of each sub-band as the frequency domain feature signal; Calculate the Pearson correlation coefficient between the time domain health index and SOH, and screen the time domain health index based on the size of the Pearson correlation coefficient; The filtered time domain health indicators are used as the input of the LSTM network model; Train the LSTM network model and use the Adam algorithm to optimize the LSTM network model; The fully connected layer of the LSTM network model outputs the estimated SOH value of the lithium battery under three lithium battery operating conditions: full charge and discharge cycle, partial charge cycle, and partial charge and discharge cycle.

2. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 1 is characterized in that: Time domain health indicators are extracted from three types of lithium battery charge and discharge cycle data: full charge and discharge cycle, partial charge cycle, and partial charge and discharge cycle, including: Voltage, current and temperature parameters are obtained from three types of lithium battery charge and discharge cycle data: full charge and discharge cycle, partial charge cycle and partial charge and discharge cycle, to extract time domain health indicators. The time domain health indicators include constant current charging time, constant voltage charging time, maximum charging temperature, average charging temperature, total electrical energy input to the battery, degree of heat accumulation and thermoelectric ratio.

3. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 2 is characterized in that: The sub-bands are obtained by performing wavelet packet decomposition on the lithium battery charge and discharge cycle data, including: The number of decomposition layers of wavelet packet decomposition is 3, and the third layer has a total of 8 nodes. The frequency bands of the voltage, current and temperature parameters in the lithium battery charge and discharge cycle data are decomposed through a tree diagram to obtain 8 different sub-frequency bands.

4. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 3 is characterized in that: Parseval's theorem is used to calculate the sum of the energy of each sub-band as the frequency domain feature signal, including: Assume d(j,k) is the wavelet packet decomposition coefficient of the kth sub-band of the jth layer, N is the length of the lithium battery charge and discharge cycle data, then the energy E(j,k) of the kth sub-band of the jth layer is: E(j,k)=∑[d(j,k,n)]², n=1,2,...,N; Where d(j,k,n) represents the nth coefficient in the kth subband of the jth layer.

5. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 4 is characterized in that: The Pearson correlation coefficient between the time domain health index and SOH is calculated according to the following formula: PCC : ； Where X and Y represent the reference values ​​of time domain health index and SOH respectively, and They represent the average values ​​of time domain health index and SOH respectively.

6. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 5, characterized in that: Screen the time domain health indicators based on the size of the Pearson correlation coefficient, including: When the Pearson correlation coefficient is less than the preset value, the corresponding time domain health indicator is eliminated; Among them, the default value is 0.

9.

7. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 6 is characterized in that: The method further comprises: Set the hyperparameters of the LSTM network model, including the number of hidden layer neurons, number of iterations, and learning rate; The number of neurons in the hidden layer is 100, the number of iterations of each training of the LSTM network model is 1000, the initial learning rate is 0.01, and the learning rate reduction factor is 0.1 / 100 times.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any one of claims 1 to 7 is implemented.

9. A storage device comprising a storage medium and a processor, wherein the storage medium stores a computer program, wherein: When the processor executes the computer program, the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any one of claims 1 to 7 is implemented.

Images