Lithium battery online SOH estimation method based on dynamic electrochemical impedance spectroscopy data

Abstract

The invention relates to a lithium battery online SOH (state of health) estimation method based on dynamic electrochemical impedance spectroscopy data, which comprises the following steps of: applying small signal disturbance to a key SOC (state of charge) point of a lithium battery charge-discharge cycle to obtain medium-high frequency electrochemical impedance spectroscopy data of each stage of the charge-discharge cycle; performing relaxation time analysis on the medium-high frequency electrochemical impedance spectroscopy data, and analyzing a mapping relation between a frequency domain and a time constant domain through deconvolution to obtain impedance distribution characteristics in the time constant domain; aiming at the impedance distribution characteristics, extracting a key characteristic subset representing the SOH change of the battery through characteristic importance evaluation, correlation analysis and redundancy control; and inputting the feature subset into a pre-constructed Gaussian process regression model to realize online SOH estimation of the lithium battery. According to the method, the dynamic electrochemical impedance spectroscopy is introduced into the lithium battery SOH estimation process, so that real-time monitoring and efficient evaluation in the battery operation process are realized.

Description

Technical Field

[0001] This invention relates to the field of battery energy storage state estimation, and in particular to an online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data. Background Technology

[0002] Electrochemical impedance spectroscopy (EIS) has become a key tool for estimating the State of Health (SOH) of batteries due to its non-invasiveness, high sensitivity, and ability to resolve multi-frequency processes. EIS-based SOH estimation methods are mainly divided into two categories. The first category is model-driven, whose core idea is to construct an equivalent circuit model, obtain physically meaningful parameters through fitting, and establish a mapping relationship between the parameters and SOH. The second category is data-driven methods. These methods do not rely on a specific electrochemical model but directly extract statistical or data-driven features from the EIS spectrum, such as impedance amplitude, phase angle, and impedance differences at specific frequencies, and then use machine learning, deep learning, and other algorithms to construct an SOH estimation model. These methods have the advantages of flexible modeling and strong adaptability, but lack explicit physical interpretation, and the model's generalization ability is highly correlated with feature selection.

[0003] Currently, most mainstream SOH estimation methods rely on full-frequency EIS data acquired under steady-state conditions. Traditional steady-state EIS testing requires the battery to be allowed to settle sufficiently to ensure the system reaches a steady state, thus obtaining a reliable frequency domain response. However, this process is time-consuming, limiting its real-time performance and engineering deployment efficiency. In contrast, dynamic EIS can acquire impedance information in real time during battery operation, possessing greater potential for online monitoring. However, the measurement conditions of Dynamic Electrochemical Impedance Spectroscopy (DEIS) violate the fundamental assumption that impedance analysis requires a "quasi-steady-state" state, especially at low frequencies, easily leading to signal distortion and characteristic drift. This results in insufficient accuracy and stability of the impedance data, thus limiting its direct application in SOH estimation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology by providing an online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data. This method aims to solve or partially solve the problem that existing DEIS measurement conditions violate the basic assumption that impedance analysis must be in a "quasi-steady state," especially in the low-frequency range where signal distortion and characteristic drift are easily caused.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] One aspect of the present invention provides an online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data, comprising the following steps:

[0007] Small signal perturbations were applied at the critical SOC point of the lithium battery charge-discharge cycle to obtain mid-to-high frequency electrochemical impedance spectroscopy data for each stage of the charge-discharge cycle.

[0008] Relaxation time analysis was performed on the aforementioned mid-to-high frequency electrochemical impedance spectroscopy data. By deconvolution, the mapping relationship between the frequency domain and the time constant domain was analyzed, and the impedance distribution characteristics in the time constant domain were obtained.

[0009] Based on the impedance distribution characteristics, a subset of key features characterizing the SOH changes of the battery is extracted through feature importance assessment, correlation analysis, and redundancy control.

[0010] The feature subset is input into a pre-built Gaussian process regression model to achieve online SOH estimation of lithium batteries.

[0011] As a preferred technical solution, the process of performing relaxation time analysis on the mid-to-high frequency electrochemical impedance spectroscopy data and obtaining the impedance distribution characteristics in the time constant domain by deconvolution to resolve the mapping relationship between the frequency domain and the time constant domain includes the following steps:

[0012] Data with frequencies greater than a threshold in electrochemical impedance spectroscopy (EIS) data are selected as mid-to-high frequency EIS data.

[0013] By treating the electrochemical system as a superposition of several parallel processes on a logarithmic time constant scale, a continuous physical-mathematical model of DRT is constructed.

[0014] For the aforementioned DRT continuous physical-mathematical model, the inversion problem is transformed into a least squares problem;

[0015] For the aforementioned least squares problem, under the premise of suppressing the overall amplitude of the discrete value vector of the DRT to be determined, the mapping relationship between the frequency domain and the time constant domain is analyzed by deconvolution, and the time constant and the corresponding relaxation intensity in the time constant domain are obtained as impedance distribution characteristics.

[0016] As a preferred technical solution, the DRT continuous physical-mathematical model is modeled as follows:

[0017]

[0018] Among them, Z * (ω) represents the complex impedance, R ∞ and R pγ(logτ) represents the mid-to-high frequency internal resistance and polarization resistance, respectively; v is the angular frequency; γ(logτ) is the DRT function describing the process distribution intensity of the system on the logarithmic time scale to be inverted; τ is the time constant; and j is the imaginary unit.

[0019] As a preferred technical solution, the least squares problem is modeled as follows:

[0020]

[0021] Where A represents the response matrix of the kernel function at the measurement frequency and the discrete time constant, x is the discrete value vector of the DRT to be determined, b is the impedance observation vector of the real or imaginary part, and λ is the regularization parameter used to balance the fitting accuracy and the smoothness of the solution.

[0022] As a preferred technical solution, the process of extracting a subset of key features characterizing the SOH change of the battery based on the impedance distribution characteristics through feature importance assessment, correlation analysis, and redundancy control includes the following steps:

[0023] Based on the impedance distribution characteristics, an initial set of candidate features is constructed;

[0024] For each candidate feature in the initial set of candidate features, an SOH estimation model is constructed, and the score of each candidate feature is calculated by combining the model's coefficient of determination, root mean square error, and mean absolute error.

[0025] The scores of each candidate feature are ranked, and the top-ranked features are selected as a subset of key features. A new SOH estimation model is constructed, and a new score is obtained by combining the model complexity, model determination coefficient, root mean square error, and mean absolute error. New features are continuously added to the subset of key features until the score increment for two consecutive rounds is less than the threshold.

[0026] As a preferred technical solution, the process of inputting the feature subset into a pre-constructed Gaussian process regression model to achieve online SOH estimation of lithium batteries includes the following steps:

[0027] The target variable is modeled as the output of a latent function described by a set of input features. The radial basis function is used as the kernel function. The Gaussian process regression model is trained using a pre-acquired training set to learn the nonlinear mapping relationship between the key feature subset and battery capacity decay, and then validated on test samples.

[0028] Obtain the variable values ​​corresponding to the key feature subset, and use the trained Gaussian process regression model to achieve online SOH estimation of lithium batteries.

[0029] As a preferred technical solution, the mid-to-high frequency electrochemical impedance spectroscopy data for each stage include impedance data from charging to the critical SOC point, impedance data after 15 minutes of rest following charging, and impedance data from discharging to the critical SOC point.

[0030] As a preferred technical solution, the key SOC point is when the battery SOC is at 50% of its total capacity.

[0031] In another aspect, an electronic device is provided, comprising: one or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the aforementioned online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data.

[0032] In another aspect, the present invention provides a computer-readable storage medium including one or more programs executable by one or more processors of an electronic device, said one or more programs including instructions for performing the aforementioned online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data.

[0033] Compared with the prior art, the present invention has at least one of the following beneficial effects:

[0034] (1) Improving the real-time performance of SOH estimation: Traditional steady-state electrochemical impedance spectroscopy (EIS) requires the battery to be fully settled to reach a quasi-steady state, which is time-consuming and cannot meet the real-time monitoring needs during battery operation. While dynamic electrochemical impedance spectroscopy (DEIS) can collect data in real time, the low-frequency band is prone to signal distortion due to non-quasi-steady-state conditions, resulting in insufficient accuracy and stability of impedance data. In this invention, a small signal perturbation is applied at the critical SOC point of the lithium battery charge-discharge cycle during data acquisition, and only electrochemical impedance spectroscopy data in the mid-to-high frequency band are collected to avoid the risk of signal distortion in the low-frequency band. During data processing, relaxation time analysis is performed on the mid-to-high frequency data. By establishing a continuous physical-mathematical model and introducing a regularization strategy, the oscillation and overfitting of the inversion results are suppressed, and the stability and physical rationality of the impedance distribution characteristics are improved, thereby improving the real-time performance of SOH estimation, meeting the needs of online monitoring, overcoming the low-frequency signal distortion problem of dynamic EIS, and ensuring the accuracy of mid-to-high frequency impedance data.

[0035] (2) Avoid overfitting / underfitting: The performance of data-driven SOH estimation methods is highly dependent on feature selection. Traditional feature extraction is prone to introducing noise and redundant information, resulting in weak model generalization ability, low estimation accuracy, and difficulty in balancing feature effectiveness and dimensionality compression requirements. This invention adopts a multi-stage feature screening and optimization algorithm to construct a full-process feature optimization mechanism. After preliminary scoring and screening by combining the model's coefficient of determination, root mean square error and mean absolute error, and secondary scoring and forward addition strategy by combining model complexity, a subset of key features highly related to SOH is extracted. This maximizes the retention of effective features highly related to battery SOH decay while compressing feature dimensionality. Attached Figure Description

[0036] Figure 1 This is a flowchart of the online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data in the embodiments;

[0037] Figure 2 This is a flowchart illustrating the extraction of a key feature subset characterizing changes in battery SOH in the embodiments;

[0038] Figure 3 This is a schematic diagram of the electronic device in the embodiment. Detailed Implementation

[0039] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0040] Example 1

[0041] Currently, as the number of cycles in lithium-ion batteries increases during use, their electrode materials and interface structures gradually degrade, leading to capacity decay and increased internal resistance. This makes it impossible to accurately assess the state of charge (SOH) of the battery during operation. To address this issue, this embodiment provides an online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data. (See [link to relevant documentation]). Figure 1 The method includes the following steps:

[0042] Step S1: Apply a small signal perturbation at the critical SOC point of the lithium battery charge-discharge cycle and collect mid-to-high frequency electrochemical impedance spectroscopy data.

[0043] The mid-to-high frequency electrochemical impedance spectroscopy data specifically include impedance data when charged to 50% SOC, impedance data after 15 minutes of rest following charging, and impedance data when discharged to 50% SOC. The data at 50% SOC is dynamic electrochemical impedance spectroscopy data.

[0044] Step S2: Perform relaxation time analysis (DRT) on the electrochemical impedance spectroscopy data to obtain the impedance distribution characteristics in the time constant domain.

[0045] In the DRT analysis process, the collected mid-to-high frequency impedance data is used as input, and the mapping relationship between the frequency domain and the time constant domain is analyzed by deconvolution, so as to obtain the impedance contribution distribution under different time constants.

[0046] Specifically, step S2 includes steps S2-1 to S2-5:

[0047] S2-1, Select only the mid-to-high frequency band from the collected EIS data as the input frequency band of DRT to avoid distortion caused by low-frequency non-quasi-steady state under dynamic operating conditions; the mid-to-high frequency band satisfies f≥8.81Hz.

[0048] S2-2, Establishing a continuous physical-mathematical model for DRT: The electrochemical system is considered as a superposition of several parallel processes on a logarithmic time constant scale, and the complex impedance is expressed as a weighted integral over logarithmic time as follows:

[0049]

[0050] Among them, R ∞ and R p These represent the internal resistance and polarization resistance, respectively. ω is the angular frequency, and G(logτ) is the DRT function to be inverted, used to describe the process distribution intensity of the system on a logarithmic time scale.

[0051] S2-3, because directly solving γ(logτ) in S2-2 is extremely sensitive to small perturbations, a 0th-order Tikhonov regularization strategy is introduced to suppress oscillations and overfitting in the inversion results. This transforms the inversion into a regularized least squares problem as follows:

[0052]

[0053] in, This represents the response matrix of the kernel function at the measurement frequency and the discrete time constant. Let be the discrete value vector of the DRT to be determined. It is the impedance vector observed for the real or imaginary part, and λ is the regularization parameter used to balance the fitting accuracy and the smoothness of the solution.

[0054] S2-4 uses the analytical equivalent solution form of the 0th-order Tikhonov equation to obtain the discrete relaxation distribution vector. By suppressing the overall amplitude of x, it achieves a dual improvement in the stability and physical rationality of the inversion result, obtaining the distribution of γ(logτ) as a function of the discrete time constant τ. Its analytical solution form is as follows:

[0055] x=(A T A+λ 2 I) -1 A b b

[0056] S2-5 uses (τ,γ(logτ)) as a candidate input for the subsequent feature selection stage, and the subsequent feature selection method automatically locates the informative time constant point within the entire curve range.

[0057] Step S3: Use the Multi-Stage Feature Optimization Algorithm (MAOS) to filter and fuse the impedance features in the DRT results.

[0058] By comprehensively screening the time constant domain features from multiple perspectives (feature importance assessment, correlation analysis, redundancy control, etc.), noise and redundant information are eliminated, and a subset of key features that can significantly reflect the changes in battery SOH are extracted.

[0059] For details, see Figure 2 Step S3 includes steps S3-1 to S3-4:

[0060] S3-1, From the DRT results output in step S2, take the discrete time constant set τ and the corresponding relaxation intensity γ(logτ) to form the initial candidate feature set C.

[0061] S3-2, for each feature c i Independently construct the SOH estimation model for ∈C, evaluate its estimation performance, and calculate the comprehensive score as follows:

[0062]

[0063] in, RMSE norm MAE norm These are the normalized results for the model's coefficient of determination, root mean square error, and mean absolute error, respectively. α, β, and γ are the corresponding weighting coefficients (satisfying α + β + γ = 1).

[0064] S3-3: Based on the scores in S3-2, the top k1 features with the highest scores are selected as candidates to enter the second stage. For the candidate features selected in S3-2, construct a set of all features with scores not exceeding k... max Combinations of each dimension are used to construct SOH estimation models and evaluate their overall performance as follows:

[0065]

[0066] Where λ is the model complexity penalty coefficient, |C * | indicates the number of features in the selected feature subset. The design is intended to explicitly penalize the increase in the number of features.

[0067] S3-4 employs a progressive forward addition strategy, continuously attempting to add new features c. new Add to current subset C * This ultimately yields a compressed and stable optimal feature subset. If the score improvement ΔJ in two consecutive rounds is less than a set threshold △J... min If the expansion fails, the expansion terminates. The comprehensive scoring function for each step of the augmented combination is calculated as follows:

[0068] ΔJ=J fianl (C * ∪{c new})-J final (C * )>△J min

[0069] Step S4: Input the feature subset into the Gaussian process regression (GPR) model to estimate the SOH of the lithium battery.

[0070] The GPR model automatically learns the nonlinear mapping relationship between input features and battery capacity decay on the training set and validates it on the test set. By comparing the estimation results based on dynamic EIS and static EIS, the differences in SOH estimation accuracy and generalization performance between the two types of impedance data are evaluated.

[0071] Specifically, step S4 includes steps S4-1 to S4-5:

[0072] S4-1, assume that the target variable (e.g., capacity or internal resistance) can be described by a set of input features as the output of a latent function f(x). This latent function is assumed to follow a Gaussian process as follows:

[0073] f(x)~GP(0,k(x,x′))

[0074] Typically, the mean function is assumed to be 0, thus focusing on the selection of the kernel function and parameter optimization. k(x,x′) represents the covariance function, and GP represents the Gaussian process regression (GPR) model.

[0075] S4-2, a radial basis function kernel is selected as the kernel function, as follows:

[0076]

[0077] in, denoted by the signal variance, and l by the length scale, both are hyperparameters used to adjust the intensity of the influence of changes in input variables on the output.

[0078] S4-3, for a given training set, where y = [y1, ..., yn ] T Hyperparameters can be estimated by maximizing the log-marginal likelihood function of the training samples, as follows:

[0079]

[0080] Where θ is the set of hyperparameters of the kernel function, and K is the covariance matrix composed of kernel functions. Here, I represents the noise variance, and I is the model complexity penalty coefficient, representing the number of features in the selected feature subset.

[0081] S4-4, for any input x * Output value y * The training set output y forms the following Gaussian distribution:

[0082]

[0083] Where K(X,x) * K(x) represents the covariance between the test samples and the training samples. * ,x * Let be the covariance of the test sample itself. Based on this, the posterior estimated distribution of the test sample can be obtained as follows:

[0084]

[0085] Wherein, the estimated mean μ * and variance They are respectively:

[0086]

[0087]

[0088] S4-5 applies the trained Gaussian process regression model to online SOH estimation of lithium batteries. New dynamic electrochemical impedance spectroscopy data are collected in real time during battery operation. After relaxation time analysis and feature selection, an input feature vector is formed and fed into the trained Gaussian process regression model, outputting the predicted SOH result for the battery.

[0089] In summary, this method constructs battery degradation models for normal battery conditions and typical degradation modes. First, impedance data of the battery at different stages of the charging process are obtained using a dynamic EIS testing device. Then, the DRT method is used to extract electrochemical characteristics under multiple time constants, and feature curves for different degradation modes are established through a feature selection algorithm. Finally, the GPR model is used to predict the state of harmonics (SOH) of the battery and identify degradation modes.

[0090] This invention introduces dynamic electrochemical impedance spectroscopy (EIS) into the SOH estimation process of lithium batteries, achieving real-time monitoring and efficient evaluation during battery operation, overcoming the bottlenecks of traditional steady-state EIS which relies on static setup and has long testing times. This method effectively suppresses signal distortion caused by non-steady-state conditions in the low-frequency band by extracting and modeling features of the dynamic EIS signal, improving data utilization and estimation accuracy. Simultaneously, this invention avoids complex static setup and full-frequency testing processes, significantly improving the real-time performance and online applicability of SOH estimation, and possessing greater engineering deployment value. Furthermore, the method of this invention has both universality and scalability, and can be widely applied to lithium battery energy storage systems of different types and application scenarios, which is of great significance for improving the intelligence level of battery management systems (BMS) and ensuring the safe and reliable operation of energy storage and electric vehicles.

[0091] Example 2

[0092] An electronic device includes: one or more processors and a memory, the memory storing one or more programs, the one or more programs including instructions for executing the online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data as described in Example 1.

[0093] like Figure 3 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The method described herein. Of course, in addition to software implementation, this invention does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. That is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0094] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0095] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0096] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for online SOH estimation of lithium batteries based on dynamic electrochemical impedance spectroscopy data, characterized in that, Includes the following steps: Small signal perturbations were applied at the critical SOC point of the lithium battery charge-discharge cycle to obtain mid-to-high frequency electrochemical impedance spectroscopy data for each stage of the charge-discharge cycle. Relaxation time analysis was performed on the aforementioned mid-to-high frequency electrochemical impedance spectroscopy data. By deconvolution, the mapping relationship between the frequency domain and the time constant domain was analyzed, and the impedance distribution characteristics in the time constant domain were obtained. Based on the impedance distribution characteristics, a subset of key features characterizing the SOH changes of the battery is extracted through feature importance assessment, correlation analysis, and redundancy control. The feature subset is input into a pre-built Gaussian process regression model to achieve online SOH estimation of lithium batteries.

2. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 1, characterized in that, The process of performing relaxation time analysis on the mid-to-high frequency electrochemical impedance spectroscopy data, and obtaining the impedance distribution characteristics in the time constant domain by deconvolution to resolve the mapping relationship between the frequency domain and the time constant domain, includes the following steps: Data with frequencies greater than a threshold in electrochemical impedance spectroscopy (EIS) data are selected as mid-to-high frequency EIS data. By treating the electrochemical system as a superposition of several parallel processes on a logarithmic time constant scale, a continuous physical-mathematical model of DRT is constructed. For the aforementioned DRT continuous physical-mathematical model, the inversion problem is transformed into a least squares problem; For the aforementioned least squares problem, under the premise of suppressing the overall amplitude of the discrete value vector of the DRT to be determined, the mapping relationship between the frequency domain and the time constant domain is analyzed by deconvolution, and the time constant and the corresponding relaxation intensity in the time constant domain are obtained as impedance distribution characteristics.

3. The method for online SOH estimation of lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 2, characterized in that, The DRT continuous physical-mathematical model is modeled as follows: Among them, Z * (ω) represents the complex impedance, R ∞ and R p ω represents the internal resistance and polarization resistance, respectively; ω is the angular frequency; γ(logτ) is the DRT function describing the intensity of the process distribution of the system on the logarithmic time scale to be inverted; τ is the time constant; and j is the imaginary unit.

4. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 2, characterized in that, The least squares problem is modeled as follows: Where A represents the response matrix of the kernel function at the measurement frequency and the discrete time constant, x is the discrete value vector of the DRT to be determined, b is the impedance observation vector of the real or imaginary part, and λ is the regularization parameter used to balance the fitting accuracy and the smoothness of the solution.

5. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 1, characterized in that, The process of extracting a subset of key features characterizing battery SOH changes based on the impedance distribution characteristics through feature importance assessment, correlation analysis, and redundancy control includes the following steps: Based on the impedance distribution characteristics, an initial set of candidate features is constructed; For each candidate feature in the initial set of candidate features, an SOH estimation model is constructed, and the score of each candidate feature is calculated by combining the model's coefficient of determination, root mean square error, and mean absolute error. The scores of each candidate feature are ranked, and the top-ranked features are selected as a subset of key features. A new SOH estimation model is constructed, and a new score is obtained by combining the model complexity, model determination coefficient, root mean square error, and mean absolute error. New features are continuously added to the subset of key features until the score increment for two consecutive rounds is less than the threshold.

6. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 1, characterized in that, The process of inputting the feature subset into a pre-constructed Gaussian process regression model to achieve online SOH estimation of lithium batteries includes the following steps: The target variable is modeled as the output of a latent function described by a set of input features. The radial basis function is used as the kernel function. The Gaussian process regression model is trained using a pre-acquired training set to learn the nonlinear mapping relationship between the key feature subset and battery capacity decay, and then validated on test samples. Obtain the variable values ​​corresponding to the key feature subset, and use the trained Gaussian process regression model to achieve online SOH estimation of lithium batteries.

7. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 1, characterized in that, The mid-to-high frequency electrochemical impedance spectroscopy data for each stage include impedance data from charging to the critical SOC point, impedance data after 15 minutes of rest following charging, and impedance data from discharging to the critical SOC point.

8. The online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data according to claim 1, characterized in that, The critical SOC point is when the battery SOC is at 50% of its total capacity.

9. An electronic device, characterized in that, include: One or more processors and a memory, wherein the memory stores one or more programs, the one or more programs including instructions for executing the online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, Includes one or more programs executed by one or more processors of an electronic device, said one or more programs including instructions for executing the online SOH estimation method for lithium batteries based on dynamic electrochemical impedance spectroscopy data as described in any one of claims 1-8.

Images