A SOH Evaluation Method Based on Time-Frequency Feature Fusion and LSTM

Abstract

This invention belongs to the field of lithium battery technology and relates to a State of Health (SOH) assessment method based on time-frequency feature fusion and LSTM. The method includes: extracting time-domain health indicators from lithium battery charge-discharge cycle data (full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle); performing wavelet packet decomposition on the lithium battery charge-discharge cycle data to obtain sub-frequency bands; calculating the sum of energy of each sub-frequency band using Passevar's theorem as the frequency domain feature signal; calculating the Pearson correlation coefficient between the time-domain health indicators and SOH, and filtering the time-domain health indicators based on the magnitude of the Pearson correlation coefficient; using the filtered time-domain health indicators as input to an LSTM network model; training the LSTM network model and optimizing it using the Adam algorithm; and outputting the lithium battery SOH estimate from the fully connected layer of the LSTM network model. Its beneficial effect is that it solves the potential reliability problem of certain health indicators under different operating conditions.

Description

Technical Field

[0001] This invention relates to the field of lithium battery technology, and in particular to a SOH evaluation method based on time-frequency feature fusion and LSTM. Background Technology

[0002] As one of the most widely used battery technologies, lithium-ion batteries have experienced rapid development in recent years. With increasing battery usage time and charge / discharge cycles, the active materials inside the battery gradually degrade, leading to performance degradation. This is manifested in reduced charge / discharge efficiency, capacity decay, and poor thermal management. This inevitable degradation can even trigger catastrophic events. IEEE Standard 1188-1996 clearly stipulates that a power battery should be replaced when its capacity drops to 80%, i.e., when its State of Health (SOH) is less than 80%. As an indispensable indicator for battery management systems, accurate SOH estimation helps predict the remaining battery life. This is crucial for developing battery replacement plans, optimizing battery maintenance strategies, and managing the overall battery system.

[0003] To avoid the limitations of model-driven methods that heavily rely on their accurate description of the real world, research on data-driven methods for estimating battery state of health (SOH) is flourishing. "Journal of Energy Storage, 2023; 74(Part B), 109370" proposes using random segments of the charging curve as input, employing a multi-layer long short-term memory network with an attention mechanism as the base learner. Multiple base learners are trained using different parts of the charging curve segments, and a backpropagation neural network is finally used to integrate the health estimates from each base 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 with the advantages of Transformer in parallel processing and global context understanding. LSTM (Long Short-Term Memory) networks, through their unique gating mechanism and in-depth mining and analysis of large amounts of data, can effectively process data with time-series characteristics, such as battery voltage, current, and temperature. This gives LSTM networks a natural advantage in predicting battery state of harmonics (SOH), enabling them to more accurately predict future trends and outcomes.

[0004] However, selecting battery health indicators for battery charging and discharging operations often requires complete time-series signals, such as the constant-current charging time and constant-voltage charging time within a single charge-discharge cycle. In practice, however, the charging and discharging process of lithium batteries often cannot strictly adhere to ideal constant-current charging, constant-voltage charging, and constant-current discharging cycles. Currently, most battery health indicators are time-domain features, which allow us to understand the global characteristics of the battery but cannot determine local behavior. In some charging cycles and superimposed cycles with variable charge-discharge modes, these characteristic parameters cannot be fully captured, thus requiring greater reliance on raw data. With the evolution of lithium-ion battery operating modes, health characteristics that previously showed a strong linear correlation with SOH (State of Health) may become invalid, as time-domain health indicators are directly susceptible to changes in charging and discharging rates, temperature fluctuations, and operating mode transitions. Summary of the Invention

[0005] Technical problems to be solved

[0006] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a SOH evaluation method based on time-frequency feature fusion and LSTM, which solves the technical problem of how to avoid the failure of health indicators caused by changes in battery operating mode under actual working conditions, based on accurate estimation of lithium battery SOH.

[0007] Technical solution

[0008] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0009] In a first aspect, the present invention provides a SOH evaluation method based on time-frequency feature fusion and LSTM, comprising:

[0010] Time-domain health indicators were extracted from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, to reflect the aging degree and health status of the battery.

[0011] Sub-frequency bands are obtained by wavelet packet decomposition of lithium battery charge-discharge cycle data;

[0012] The sum of the energies of each sub-band is calculated using Passevar's theorem as the frequency domain characteristic signal;

[0013] Calculate the Pearson correlation coefficient between time-domain health indicators and SOH, and screen time-domain health indicators based on the magnitude of the Pearson correlation coefficient;

[0014] The selected time-domain health indicators are used as input to the LSTM network model;

[0015] The LSTM network model is trained and then optimized using the Adam algorithm.

[0016] The fully connected layer of the LSTM network model outputs estimated SOH values ​​for lithium batteries under three lithium battery operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle.

[0017] Optionally, time-domain health indicators are extracted from the lithium battery charge-discharge cycle data of three types of operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle. These indicators include:

[0018] Voltage, current, and temperature parameters are obtained from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, respectively, to extract time-domain health indicators. These 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.

[0019] Optionally, wavelet packet decomposition is performed on the lithium battery charge-discharge cycle data to obtain sub-frequency bands, including:

[0020] The wavelet packet decomposition has 3 decomposition layers, with 8 nodes in the third layer. The frequency bands of voltage, current and temperature parameters in the lithium battery charge and discharge cycle data are decomposed by a tree diagram to obtain 8 different sub-frequency bands.

[0021] Optionally, Passevar's theorem is used to calculate the sum of the energies of each sub-band as the frequency domain characteristic signal, including:

[0022] Let d(j,k) be the wavelet packet decomposition coefficient of the k-th sub-band of the j-th layer, and N be the length of the lithium battery charge-discharge cycle data. Then the energy E(j,k) of the k-th sub-band of the j-th layer is:

[0023] E(j,k)=∑[d(j,k,n)]², n=1,2,...,N;

[0024] Where d(j,k,n) represents the nth coefficient in the kth sub-band of the jth layer.

[0025] Optionally, the Pearson correlation coefficient between the time-domain health index and SOH can be calculated according to the following formula. PCC :

[0026] ;

[0027] Where X and Y represent the reference values ​​for the time-domain health index and SOH, respectively. and These represent the average values ​​of the time-domain health index and SOH, respectively.

[0028] Optionally, time-domain health indicators can be screened based on the magnitude of the Pearson correlation coefficient, including:

[0029] If the Pearson correlation coefficient is less than the preset value, the corresponding time-domain health indicator will be removed.

[0030] The default value is 0.9.

[0031] Optionally, the method further includes:

[0032] Set the hyperparameters of the LSTM network model, including the number of hidden layer neurons, the number of iterations, and the learning rate;

[0033] The hidden layer has 100 neurons, the LSTM network model is trained in 1000 iterations per training cycle, the initial learning rate is 0.01, and the learning rate decrease factor is 0.1 / 100 times.

[0034] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any of the first aspects above.

[0035] Thirdly, the present invention provides a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any of the first aspects above. Beneficial effects

[0036] The beneficial effects of this invention are as follows: This invention provides a State of Health (SOH) assessment method based on time-frequency feature fusion and LSTM, which fuses the time-domain and frequency-domain features of lithium battery charge-discharge operation data to estimate the SOH of lithium batteries across multiple time scales and partial cycles. Wavelet packet decomposition technology is used to analyze the characteristics of voltage, current, and temperature signals during the charge-discharge cycle of lithium-ion batteries. Passevar's theorem is applied to quantify the energy distribution across frequency bands, revealing the dynamic energy characteristics under various operating conditions. This provides interpretable input for the network model and elucidates the potential mechanisms of energy changes. The sum of energy at different frequencies after wavelet packet decomposition is used as the frequency-domain feature signal, capturing the overall energy characteristics of current, voltage, and temperature signals within the frequency range represented by each layer after decomposition. This addresses the limitations of idealized constant charge-discharge cycles experiencing partial or variable cycle modes in practical applications. To better reflect actual operating conditions, the SOH prediction performance under three different scenarios—complete cycle, partial cycle, and variable partial cycle—is evaluated, resolving potential reliability issues of certain health indicators under different operating conditions. Attached Figure Description

[0037] Figure 1A flowchart illustrating a SOH evaluation method based on time-frequency feature fusion and LSTM provided in an embodiment of the present invention;

[0038] Figure 2 This is a schematic diagram of the SOH estimation process provided in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of wavelet packet decomposition provided in an embodiment of the present invention. Detailed Implementation

[0040] To better explain and facilitate understanding of the present invention, it will be described in detail below with reference to the accompanying drawings and specific embodiments. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a clearer and more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0041] Firstly, referring to Figure 1 and Figure 2 This embodiment provides a SOH evaluation method based on time-frequency feature fusion and LSTM, including:

[0042] S1 extracts time-domain health indicators from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, to reflect the aging degree and health status of the battery.

[0043] S2, wavelet packet decomposition is performed on the lithium battery charge and discharge cycle data to obtain sub-frequency bands.

[0044] S3 uses Passevar's theorem to calculate the sum of the energies of each sub-band as the frequency domain characteristic signal.

[0045] S4. Calculate the Pearson correlation coefficient between the time-domain health indicators and SOH, and screen the time-domain health indicators based on the magnitude of the Pearson correlation coefficient.

[0046] S5 uses the selected time-domain health indicators as input to the LSTM network model.

[0047] S6. Train the LSTM network model and optimize it using the Adam algorithm.

[0048] The S7, LSTM network model's fully connected layer outputs estimated SOH values ​​for lithium batteries under three lithium battery operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle.

[0049] The technical solution proposed in this invention employs wavelet packet decomposition to extract frequency domain features. Compared with Fast Fourier Transform, the extracted frequency domain health indicators show stronger correlation, providing interpretable input for the network model. This solves the problem of health indicators failing due to changes in lithium battery operating conditions during actual operation. To verify and evaluate the performance of the proposed solution, this invention designs three operating modes for lithium-ion batteries: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle. The actual prediction error throughout the cycle remains below 2%, with root mean square errors of 0.41%, 0.80%, and 0.63% for the three operating modes, respectively.

[0050] Optionally, time-domain health indicators are extracted from the lithium battery charge-discharge cycle data of three types of operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle. These indicators include:

[0051] Voltage, current, and temperature parameters are obtained from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, respectively, to extract time-domain health indicators. These 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.

[0052] These indicators reflect the performance changes of the battery during the charging and discharging process from different aspects, comprehensively cover the health-related characteristics of the battery, effectively quantify battery aging and health status, and provide rich and valuable data to more accurately assess the actual state of lithium batteries.

[0053] Optionally, combined Figure 3 Wavelet packet decomposition was performed on the lithium battery charge-discharge cycle data to obtain sub-frequency bands, including:

[0054] The wavelet packet decomposition has 3 decomposition layers, with 8 nodes in the third layer. The frequency bands of voltage, current and temperature parameters in the lithium battery charge and discharge cycle data are decomposed by a tree diagram to obtain 8 different sub-frequency bands.

[0055] For current and voltage signals, the low-frequency components likely represent the average current and voltage levels during charging and discharging, thus better reflecting the battery's health characteristics. The high-frequency components primarily reflect instantaneous fluctuations or noise during the lithium battery's charging and discharging process. For temperature signals, the low-frequency components may represent the average temperature change during battery operation, while the high-frequency components may reflect localized temperature fluctuations or thermal noise within the battery.

[0056] This refined frequency band decomposition method enables the analysis of lithium battery charge-discharge cycle data at different frequency levels, uncovering battery state information contained in different frequency components. This helps to gain a deeper understanding of the changes in electrical characteristics during battery charge-discharge processes, thus laying the foundation for extracting more accurate and effective frequency domain feature signals and improving the accuracy of battery state assessment.

[0057] Optionally, Passevar's theorem is used to calculate the sum of the energies of each sub-band as the frequency domain characteristic signal, including:

[0058] Let d(j,k) be the wavelet packet decomposition coefficient of the k-th sub-band of the j-th layer, and N be the length of the lithium battery charge-discharge cycle data. Then the energy E(j,k) of the k-th sub-band of the j-th layer is:

[0059] E(j,k)=∑[d(j,k,n)]², n=1,2,...,N;

[0060] Where d(j,k,n) represents the nth coefficient in the kth sub-band of the jth layer.

[0061] The time-domain health indicators are constant current charging time, constant voltage charging time, and thermoelectric ratio, while the frequency-domain characteristic signals are current signal from 0Hz to 0.625Hz, voltage signal from 0.625Hz to 1.25Hz, and temperature signal from 0.625Hz to 1.25Hz.

[0062] By employing Passevar's theorem to calculate the sum of the energies of sub-bands as the frequency domain characteristic signal, the contribution of each sub-band to the overall signal energy is accurately quantified, enabling the extraction of the energy distribution characteristics of lithium battery charging and discharging signals in different frequency domains. These frequency domain characteristic signals can reflect the internal physical changes of the battery under different operating conditions.

[0063] Optionally, the Pearson correlation coefficient between the time-domain health index and SOH can be calculated according to the following formula. PCC :

[0064] ;

[0065] Where X and Y represent the reference values ​​for the time-domain health index and SOH, respectively. and These represent the average values ​​of the time-domain health index and SOH, respectively.

[0066] Optionally, time-domain health indicators can be screened based on the magnitude of the Pearson correlation coefficient, including:

[0067] If the Pearson correlation coefficient is less than the preset value, the corresponding time-domain health indicator will be removed.

[0068] The default value is 0.9.

[0069] Identify and remove health characteristic variables that are weakly correlated with SOH in order to reduce data dimensionality and improve analysis efficiency.

[0070] Optionally, the method further includes:

[0071] Set the hyperparameters of the LSTM network model, including the number of hidden layer neurons, the number of iterations, and the learning rate;

[0072] The hidden layer has 100 neurons, the LSTM network model is trained in 1000 iterations per training cycle, the initial learning rate is 0.01, and the learning rate decrease factor is 0.1 / 100 times.

[0073] Properly setting the hyperparameters of the LSTM network model, such as the number of hidden layer neurons, the number of iterations, and the learning rate, helps optimize the performance of the LSTM network model. Appropriate settings of the number of hidden layer neurons, the number of iterations, and the learning rate reduction factor enable the model to accurately learn the potential relationship between time-domain health indicators and SOH in complex lithium battery data, improving the model's generalization ability and prediction accuracy. This results in more stable and efficient output of SOH estimates for lithium batteries under different operating conditions, enhancing the model's effectiveness and reliability in practical applications.

[0074] In a second aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any of the first aspects above.

[0075] Thirdly, embodiments of the present invention provide a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any of the first aspects above.

[0076] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0077] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

[0078] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A SOH evaluation method based on time-frequency feature fusion and LSTM, characterized in that, include: Time-domain health indicators were extracted from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, to reflect the aging degree and health status of the battery. Sub-frequency bands are obtained by wavelet packet decomposition of lithium battery charge-discharge cycle data; The sum of the energies of each sub-band is calculated using Passevar's theorem as the frequency domain characteristic signal; Calculate the Pearson correlation coefficient between time-domain health indicators and SOH, and screen time-domain health indicators based on the magnitude of the Pearson correlation coefficient; The selected time-domain health indicators and frequency-domain feature signals are used as inputs to the LSTM network model; The LSTM network model is trained and then optimized using the Adam algorithm. The fully connected layer of the LSTM network model outputs estimated SOH values ​​for lithium batteries under three lithium battery operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle.

2. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 1, characterized in that, Time-domain health indicators were extracted from lithium battery charge-discharge cycle data of three different operating conditions: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle. These indicators included: Voltage, current, and temperature parameters are obtained from three types of lithium battery charge-discharge cycle data: full charge-discharge cycle, partial charge cycle, and partial charge-discharge cycle, respectively, to extract time-domain health indicators. These 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, characterized in that, Wavelet packet decomposition was performed on the charge-discharge cycle data of lithium batteries to obtain sub-frequency bands, including: The wavelet packet decomposition has 3 decomposition layers, with 8 nodes in the third layer. The frequency bands of voltage, current and temperature parameters in the lithium battery charge and discharge cycle data are decomposed by 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, characterized in that, The sum of the energies of each sub-band is calculated using Passevar's theorem as the frequency domain characteristic signal, including: Let d(j,k) be the wavelet packet decomposition coefficient of the k-th sub-band of the j-th layer, and N be the length of the lithium battery charge-discharge cycle data. Then the energy E(j,k) of the k-th sub-band of the j-th 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 sub-band of the jth layer.

5. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 4, characterized in that, The Pearson correlation coefficient between the time-domain health index and SOH is calculated using the following formula. PCC : ； Where X and Y represent the reference values ​​for the time-domain health index and SOH, respectively. and These represent the average values ​​of the 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, Time-domain health indicators were screened based on the magnitude of the Pearson correlation coefficient, including: If the Pearson correlation coefficient is less than the preset value, the corresponding time-domain health indicator will be removed. The default value is 0.

9.

7. The SOH evaluation method based on time-frequency feature fusion and LSTM according to claim 6, characterized in that, The method further includes: Set the hyperparameters of the LSTM network model, including the number of hidden layer neurons, the number of iterations, and the learning rate; The hidden layer has 100 neurons, the LSTM network model is trained in 1000 iterations per training cycle, the initial learning rate is 0.01, and the learning rate decrease 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 the processor, it implements the SOH evaluation method based on time-frequency feature fusion and LSTM as described in any one of claims 1 to 7.

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

Images