Method and system for estimating health state of lithium ion battery

Abstract

The invention provides a lithium ion battery health state estimation method, and belongs to the field of battery management, and the method comprises the steps: obtaining the EIS data, the current temperature and the current SOC of a to-be-detected lithium ion battery; the EIS data are analyzed through a DRT method, a DRT curve is obtained, health indexes are extracted from the DRT curve, and the health indexes comprise the peak amplitude, the time constant corresponding to the peak value, the peak area, the full width at half maximum, the weighted average time constant, the weighted standard deviation time constant, the time constant skewness and the time constant kurtosis; combining the health index, the current temperature and the current SOC to obtain an input feature vector, inputting the input feature vector into the trained SOH estimation model, and estimating the SOH of the lithium ion battery to be measured; the invention further provides an estimation system. The eight health indexes are closely related to aging mechanisms such as SEI membrane growth, active substance loss and impedance increase of the battery, high-quality input is provided for the model, and SOH estimation precision is improved; the temperature and the SOC serve as key input, and the influence of working condition changes on EIS measurement is effectively compensated.

Description

Technical Field

[0001] The present invention relates to the technical field of battery management, and in particular to a method and system for estimating the health status of a lithium-ion battery. Background Art

[0002] With the popularity of applications such as electric vehicles and energy storage systems, the use of lithium-ion batteries is becoming more and more widespread. The state of health (SOH) of a battery is a key indicator for measuring its performance and remaining life, and is directly related to the safety and reliability of the equipment. Therefore, the development of accurate and reliable SOH estimation methods is crucial for battery management systems (BMS). At present, the methods for estimating SOH mainly include model-based physical methods and data-driven methods. Although physical methods can explain the internal mechanism, the models are complex, the computational effort is large, and they are difficult to apply online. Data-driven methods, especially those based on machine learning, have attracted widespread attention because they do not require complex electrochemical models.

[0003] Electrochemical impedance spectroscopy (EIS) is a powerful non-destructive testing technology that can provide rich information about the electrochemical processes inside the battery (such as charge transfer, ion diffusion, SEI film changes, etc.), which is closely related to the aging state of the battery. However, traditional EIS analysis methods face many challenges. On the one hand, the raw EIS data has high dimensions and complex morphology, making it difficult to directly extract effective features that have a clear monotonic relationship with SOH; on the other hand, the impedance characteristics of the battery are affected by the coupling of multiple factors such as operating temperature and state of charge (SOC), resulting in large performance fluctuations of existing models under different operating conditions, lack of stability and robustness, which poses a huge challenge to the accurate estimation of SOH.

[0004] In the prior art, the Chinese invention patent application with publication number CN118311434A, "Method and system for estimating SOH of lithium-ion batteries based on electrochemical impedance spectroscopy", is based on electrochemical impedance spectroscopy (EIS). It interprets and extracts features from electrochemical impedance spectroscopy data through an improved relaxation time distribution algorithm. The constructed neural network model fuses the DRT features calculated based on the relaxation time DRT with the DRT features automatically extracted by the neural network model. However, this method has two problems: (1) It only extracts the peak area of ​​the DRT curve. When the battery ages, the core physical and chemical degradation mechanisms include lithium ion loss, active material loss, and impedance increase. The above aging mechanisms occur simultaneously and affect each other. They act together on the electrochemical steps that are not passed. The final manifestation on the DRT curve is the systematic evolution of multiple morphological parameters of multiple peaks at the same time. Looking only at the single feature of the peak area will inevitably affect the accuracy of the battery SOH estimation. (2) It does not consider the coupling effect of the battery impedance characteristics on the operating temperature and the state of charge, resulting in large fluctuations in the estimation performance of this method under different working conditions, lacking stability and robustness.

[0005] The Chinese invention patent application with publication number CN118311464A, "A method and system for estimating the state of health of a battery," predicts the relaxation time distribution of the battery in its current state by obtaining the impedance of the battery in several specific frequency bands, and estimates the health state of the battery under the cycle based on the predicted relaxation time distribution. The problem with this method is that it extracts the battery impedance spectrum after each charge and discharge, which is a partial EIS, and then calculates the relaxation time distribution curve of the partial EIS, extracting only one polarization resistance, namely the peak area. Since only the single feature of the peak area is considered, the accuracy of the battery SOH estimation is bound to be affected.

[0006] Therefore, this field urgently needs a new SOH estimation method that can effectively process high-dimensional EIS data, decouple the influence of multiple factors, and maintain high accuracy and high stability under a wide range of working conditions. Summary of the Invention

[0007] The technical problem to be solved by the present invention is how to solve the problems of low accuracy, difficulty in feature extraction and poor robustness under different operating conditions in existing lithium-ion battery SOH estimation methods.

[0008] The present invention solves the above technical problems through the following technical solutions: a method for estimating the health status of a lithium-ion battery, the method comprising:

[0009] Obtain EIS data, current temperature and current SOC of the lithium-ion battery to be tested;

[0010] The DRT method was used to analyze EIS data to obtain the DRT curve, from which health indicators were extracted. The health indicators included peak amplitude, time constant corresponding to the peak, peak area, full width at half maximum, weighted average time constant, weighted standard deviation time constant, time constant skewness, and time constant kurtosis.

[0011] The health index, current temperature and current SOC are combined to obtain an input feature vector, which is input into the trained SOH estimation model to estimate the SOH of the lithium-ion battery to be tested.

[0012] Beneficial effects: The present invention uses DRT analysis instead of directly using raw EIS data, which can separate overlapping electrochemical processes. The eight key health indicators extracted are effective health characteristics of the battery electrochemical process. These indicators are closely related to the battery's SEI film growth, active material loss, and impedance increase and other aging mechanisms. Therefore, they are more physically meaningful and can systematically capture key information related to battery aging, providing high-quality input for subsequent models and improving the accuracy of SOH estimation. By taking temperature and SOC as key inputs, the impact of operating condition changes on EIS measurements is effectively compensated, so that the model can maintain a consistent low error level under a wide range of operating environments, showing excellent robustness.

[0013] Preferably, the process of analyzing EIS data using the DRT method includes:

[0014] Fit the EIS data to obtain the relaxation time distribution function F(τ), use a set of time constants to represent the battery impedance spectrum Z(ω), and obtain the DRT curve by solving the minimization error function. The error function is:

[0015]

[0016] in, ω is the frequency, τ is the time constant, and λ is the regularization parameter.

[0017] Preferably, among the health indicators, the peak amplitude represents the intensity or dominance of the electrochemical process of the battery at the characteristic reaction rate; the time constant corresponding to the peak represents the characteristic reaction rate of the battery electrochemical process; the peak area is proportional to the total polarization resistance of the battery electrochemical process, and represents the contribution of the electrochemical process to the total impedance; the full width at half maximum represents the uniformity or dispersion of the distribution of the time constant of the electrochemical process; the weighted average time constant reflects the central position of the DRT curve; the weighted standard deviation time constant represents the dispersion of the distribution of the reaction rate of the electrochemical process; the time constant skewness represents the asymmetry of the peak distribution of the DRT curve; the time constant kurtosis represents the sharpness of the peak of the DRT curve or the degree of "tail heaviness".

[0018] Beneficial Effects: The eight health indicators extracted by the present invention can reflect the significant changes in the electrochemical properties of lithium-ion batteries with age, as well as their collective ability to comprehensively describe the evolution of DRT peaks. These specific parameters quantify key aspects of DRT peaks, such as their amplitude (peak amplitude), position (Tau at the peak), width (FWHM, weighted normalized Tau) and overall shape (Tau skew, Tau kurtosis), which systematically evolve with degradation mechanisms such as SEI layer growth, active material loss and impedance increase.

[0019] Preferably, in the process of extracting health indicators from the DRT curve, after determining the boundary of the DRT curve peak, the maximum value of the function F(τ) in the interval is found to obtain the peak amplitude, and the abscissa corresponding to the peak amplitude is the time constant corresponding to the peak; the DRT curve peak is numerically integrated on the logarithmic time axis to obtain the peak area; the full width at half maximum FWHM is calculated as follows: FWHM = |τ2-τ1|, τ1 and τ2 are two time constant points corresponding to half the peak amplitude; each time constant point τ i As a value, its corresponding function value F(τ i ) as the weight, and perform weighted averaging to obtain the weighted average time constant; calculate the standard deviation of the weighted average time constant to obtain the weighted standard deviation time constant; calculate the standardized third-order central moment of the weighted average time constant to obtain the time constant skewness; calculate the standardized fourth-order central moment of the weighted average time constant and then subtract 3 to obtain the time constant kurtosis.

[0020] Preferably, the calculation formula of peak area HI3 is:

[0021]

[0022] Among them, τ start , τ end They are the starting and ending time constant points of the DRT curve respectively.

[0023] Preferably, the calculation formula of the weighted average time constant HI5 is:

[0024]

[0025] The calculation formula of the weighted standard deviation time constant HI6 is:

[0026]

[0027] The calculation formula of time constant skewness HI7 is:

[0028]

[0029] The calculation formula of time constant kurtosis HI8 is:

[0030]

[0031] Preferably, the training process of the trained SOH estimation model includes:

[0032] Obtain EIS data of lithium-ion batteries at different SOH, SOC and ambient temperatures, preprocess the EIS data, and obtain preprocessed EIS data;

[0033] The DRT method is used to analyze the preprocessed EIS data to obtain the DRT curve. The health index is extracted from the DRT curve, and the input feature vector is obtained by combining the health index, ambient temperature and SOC.

[0034] The input feature vector is used as input and the battery SOH is used as the output label to train the SOH estimation model. When the loss function is minimized, the trained SOH estimation model is obtained.

[0035] Preferably, the EIS data is preprocessed by using a Savitzky-Golay filtering algorithm to smooth and denoise the EIS data.

[0036] Preferably, the SOH estimation model is an AE model, a CNN model, or an MLP model, and the SOH estimation model includes:

[0037] An encoder, which maps the input feature vector into a feature representation in a low-dimensional latent space;

[0038] A decoder that maps feature representations to battery SOH estimates.

[0039] Beneficial effects: When the SOH estimation model adopts the AE model, the introduction of the autoencoder model, through its powerful nonlinear mapping and feature learning capabilities, can discover deep features in the data that are intrinsically related to the SOH, effectively filter out noise interference, and thus improve the generalization ability of the model.

[0040] The present invention also provides a lithium-ion battery health status estimation system, the system comprising:

[0041] Data acquisition module, used to obtain EIS data, current temperature and current SOC of the lithium-ion battery to be tested;

[0042] The feature extraction module is used to analyze EIS data using the DRT method to obtain the DRT curve and extract health indicators from the DRT curve. The health indicators include peak amplitude, time constant corresponding to the peak, peak area, full width at half maximum, weighted average time constant, weighted standard deviation time constant, time constant skewness and time constant kurtosis;

[0043] The SOH estimation module is used to combine the health index, current temperature and current SOC to obtain an input feature vector, input the input feature vector into the trained SOH estimation model, and estimate the SOH of the lithium-ion battery to be tested. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a method for estimating the health status of a lithium-ion battery provided in Example 1 of the present invention;

[0045] Figure 2 Schematic diagram of the DRT curve changing with SOH using the method for estimating the health status of a lithium-ion battery in Example 3 of the present invention;

[0046] Figure 3 This is a scatter plot of the corresponding relationship between the weighted average time constant in the health indicator and the SOH at different operating points using the method for estimating the health status of lithium-ion batteries in Example 3 of the present invention;

[0047] Figure 4 This is a violin plot of the distribution of absolute errors estimated by the AE model under different SOC conditions using the method for estimating the health status of lithium-ion batteries in Example 3 of the present invention;

[0048] Figure 5 This is a violin plot of the distribution of absolute errors estimated by the AE model under different temperature conditions using the method for estimating the health status of lithium-ion batteries in Example 3 of the present invention;

[0049] Figure 6 This is a radar chart comparing the errors of the AE model with those of the CNN and MLP models under different temperature and SOC combinations using the lithium-ion battery health status estimation method in Example 3 of the present invention;

[0050] Figure 7 This is a schematic diagram of a lithium-ion battery health status estimation system provided by Example 2 of the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the following describes the technical solutions of the present invention in a clear and complete manner with reference to specific embodiments and the accompanying drawings. It is apparent that the embodiments described are only a portion of the embodiments of the present invention, not all of them. All other embodiments obtained by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] Explanation of relevant terms:

[0053] The state of health (SOH) of a battery is used to quantify the capacity decay of a lithium-ion battery. It is calculated by the ratio of the actual battery capacity to the nominal capacity. The calculation formula is:

[0054] SOH=C present / C initial ×100%

[0055] Among them, C present is the maximum capacity of the battery available under current conditions, C initial This is the initial capacity of a new battery.

[0056] Electrochemical Impedance Spectroscopy (EIS) data refers to a series of battery impedance values ​​obtained from a battery impedance spectrum test, used to study the electrochemical reactions within the battery. The battery impedance spectrum is obtained by applying small-amplitude AC sinusoidal potential waves of varying frequencies to the battery and measuring how the ratio of the AC potential to the current signal (i.e., the system's impedance) changes with the frequency of the sinusoidal wave.

[0057] State of Charge (SOC) is a key indicator for measuring the remaining power of a battery. It is defined as the ratio of the battery's current remaining capacity to its rated capacity when fully charged.

[0058] The Distribution of Relaxation Times (DRT) determines the relaxation time distribution in a material by measuring the frequency response of an electrochemical system.

[0059] Example 1

[0060] like Figure 1 As shown, this embodiment provides a method for estimating the health status of a lithium-ion battery, comprising the following steps:

[0061] Step 1: Obtain EIS data, current temperature, and current SOC of the lithium-ion battery to be tested.

[0062] Step 2: Use the DRT method to analyze the EIS data to obtain the DRT curve, and extract 8 health indicators (HI) from the DRT curve. The health indicators include peak amplitude (Peak Amplitude) HI1 (the highest point value of the peak on the DRT curve, reflecting the intensity of the electrochemical process), time constant corresponding to the peak (Tau at Peak) HI2 (the time constant corresponding to the DRT peak position, reflecting the reaction rate of the electrochemical process), peak area (Area) HI3 (the total area under the DRT peak, reflecting the total relaxation intensity of the electrochemical process), full width at half maximum (FWHM) HI4 (the width at half the peak height, reflecting the uniformity of the electrochemical process distribution), weighted average time constant (Weighted Mean Tau) HI5 (the weighted average time constant of the DRT peak, representing the center position of the peak), weighted standard deviation time constant (Weighted Std Tau) HI6 (the weighted standard deviation of the time constant of the DRT peak, representing the degree of peak dispersion), time constant skewness (Tau Skew) HI7 (the asymmetry of the DRT peak) and time constant kurtosis (Tau Kurtosis) HI8 (sharpness of DRT peak).

[0063] Here we focus on the reasons for extracting the above 8 health indicators to estimate battery SOH:

[0064] The root cause of battery aging, or the decline of state of health (SOH), can be attributed to several core physical and chemical degradation mechanisms. These include:

[0065] (1) Loss of Lithium Inventory (LLI): During the cycling process, some available lithium ions are “fixed” and can no longer participate in charge and discharge, for example, they are consumed in the growth of the SEI film. This is one of the main reasons for capacity decay.

[0066] (2) Loss of Active Material (LAM): The electrode material (such as NMC for the positive electrode and graphite for the negative electrode) itself undergoes structural damage, peels off from the current collector, or dissolves, resulting in a reduction in the "space" that can store lithium ions.

[0067] (3) Impedance Growth: The internal resistance of the battery, especially the resistance to electrochemical reactions, increases with use. This not only reduces the power performance of the battery, but also leads to increased heat generation, further accelerating other aging processes.

[0068] Each “peak” in the DRT corresponds to one or more specific electrochemical steps within the battery. As the battery ages, the kinetics of these steps change, causing the morphology of the DRT peak to change.

[0069] (1) Medium and high frequency peaks: usually related to SEI film and contact impedance

[0070] Corresponding electrochemical process: This peak primarily reflects the process of lithium ions passing through the solid electrolyte interface (SEI) film. LLI and impedance increase: During each charge and discharge, the SEI film undergoes slight dissolution and regeneration, a process that continuously consumes electrolyte and active lithium ions (LLI). This causes the SEI film to continuously thicken and become less dense, thereby increasing the resistance to lithium ion penetration.

[0071] Indicator manifestation: Peak area (Area): The increase in SEI film impedance directly leads to an increase in the area of ​​this peak. Full width at half maximum (FWHM) / weighted standard deviation (Weighted Std Tau): The growth of SEI film is often uneven, resulting in differences in impedance in different areas, which will cause the DRT peak to become wider. Time constant corresponding to the peak (Tau at Peak): As the SEI film becomes thicker, it takes longer for lithium ions to pass through it, so the peak position may shift to the right (the time constant increases). Skew (Skew): Uneven growth may also cause the peak shape to become asymmetric.

[0072] (2) Intermediate frequency peak: usually related to the charge transfer process

[0073] Corresponding electrochemical process: This is the core electrochemical reaction—charge transfer. Increased LAM and impedance: The surface of the active material particles may be covered by an excessively thick SEI film, or the active material itself may undergo microstructural damage (LAM). This reduces the "active sites" for effective charge transfer, making the reaction more difficult and increasing the charge transfer impedance. Peak area (Area): An increase in charge transfer impedance will significantly increase the area of ​​this peak. This is one of the most frequently observed features in SOH estimation.

[0074] Indicator: Time constant corresponding to the peak (Tau at Peak) / Weighted Mean Tau: The reaction kinetics slows down, meaning that the reaction takes longer to complete, which causes the center position of the peak to shift significantly to the right.

[0075] Peak Amplitude: With aging, the peak amplitude may decrease while the peak broadens to maintain or increase the total area (total impedance).

[0076] (3) Low-frequency peak: Usually related to the diffusion process, it mainly reflects the diffusion process of lithium ions within the solid phase particles of the electrode active material. LAM: Long-term cycling can cause microcracks in the active material particles (such as graphite) or even breakage. These cracks increase the diffusion path of lithium ions within the particles, making it more tortuous and difficult. The electrode pores may also be blocked by byproducts, hindering the infiltration of the electrolyte, thereby affecting diffusion.

[0077] Indicator: Peak area (Area): The increase in diffusion impedance leads to an increase in peak area.

[0078] Time constant corresponding to the peak (Tau at Peak): The diffusion path becomes longer and more difficult, which means it takes longer, so the peak will shift to the right.

[0079] Full Width at Half Maximum (FWHM) / Weighted Std Tau: Particle breakage and pore blockage are non-uniform across the electrode, causing the distribution of diffusion paths to become very non-uniform, and thus the peak becomes significantly broadened.

[0080] Skew / Kurtosis: Complex structural damage can cause the distribution of the diffusion process to deviate significantly from the ideal state, resulting in a very asymmetric peak shape or the appearance of multiple secondary peaks, which can be captured by skewness and kurtosis.

[0081] This shows that battery aging is not simply a matter of increasing a certain impedance, but rather a complex, unevenly distributed process. For example, the thickening and uneven growth of the SEI film can lead to an increase in the DRT peak area, a widening of the full width at half maximum, and even a shift (a change in skewness) responsible for this process. The loss of active material can lead to a weakening of certain reaction intensities (a decrease in amplitude). A slowdown in the charge transfer process is reflected in a rightward shift in the time constant corresponding to the peak.

[0082] In step 2, the relaxation time distribution method is used to perform in-depth analysis on the pre-processed EIS data. The DRT technique can decompose the complex impedance spectrum into the time constant (τ) domain, thereby separating the multiple overlapping electrochemical processes in the EIS and forming a series of DRT peaks that are easier to analyze. The process of analyzing EIS data using the DRT method includes:

[0083] Fit the EIS data to obtain the relaxation time distribution function F(τ), use a set of time constants to represent the battery impedance spectrum Z(ω), and obtain the DRT curve by solving the minimization error function. The error function is:

[0084]

[0085] in, ω is the frequency, τ is the time constant, and λ is the regularization parameter.

[0086] Among the health indicators, the peak amplitude (HI1) indicates the intensity or dominance of the electrochemical process at the battery's characteristic reaction rate. The time constant (HI2) corresponding to the peak indicates the characteristic reaction rate of the battery's electrochemical process. As the battery ages, the internal kinetics slow, and this time constant typically shifts to the right, meaning it becomes longer. The peak area (HI3) is proportional to the total polarization resistance of the battery's electrochemical process and indicates the contribution of the electrochemical process to the total impedance. The full width at half maximum (HI4) indicates the uniformity or dispersion of the distribution of the electrochemical time constant. A wider peak indicates a wider distribution of reaction rates involved in the process, and a more uneven process, which is often associated with uneven aging. The weighted average time constant (HI5) reflects the center position of the DRT curve. The weighted standard deviation time constant (HI6) indicates the dispersion of the electrochemical reaction rate distribution. A larger standard deviation indicates a more dispersed reaction rate distribution. The time constant skewness (HI7) indicates the asymmetry of the DRT curve peak distribution. Aging of the SEI film may create a "tail" next to the main reaction, causing the peak shape to skew to one side. Skewness quantifies this phenomenon. A time constant skewness greater than zero is right-skewed. The time constant kurtosis HI8 characterizes the sharpness of the peak of the DRT curve or the degree of "tail heaviness". It can reflect the number of extreme values ​​in the distribution, that is, the number of parts with extremely different reaction rates from the central rate.

[0087] In the process of extracting health indicators from the DRT curve, after determining the boundary of the DRT curve peak, find the maximum value of the function F(τ) in the interval to obtain the peak amplitude HI1. The horizontal coordinate corresponding to the peak amplitude is the time constant HI2 corresponding to the peak; the DRT curve peak is numerically integrated on the logarithmic time axis. Among discrete data points, the Riemann sum is usually used for approximate calculation to obtain the peak area HI3.

[0088] The calculation formula for the peak amplitude HI1 is: HI1 = max(F(τ)), and the calculation formula for the time constant HI2 corresponding to the peak is: HI2 = τ peak ,whereF(τ peak )=max(F(τ)).

[0089] The calculation formula of peak area HI3 is:

[0090]

[0091] Among them, τ start , τ end They are the starting and ending time constant points of the DRT curve respectively.

[0092] The full width at half maximum (FWHM) is calculated as: FWHM = |τ2-τ1|, where τ1 and τ2 are two time constant points corresponding to half the peak amplitude.

[0093] At each time constant point τ i As a value, its corresponding function value F(τ i ) is the weight, and the weighted average is performed to obtain the weighted average time constant; the calculation formula of the weighted average time constant HI5 is:

[0094]

[0095] The weighted standard deviation time constant is obtained by calculating the standard deviation of the weighted average time constant; the calculation formula of the weighted standard deviation time constant HI6 is:

[0096]

[0097] The time constant skewness is obtained by calculating the normalized third-order central moment of the weighted average time constant. The calculation formula of the time constant skewness HI7 is:

[0098]

[0099] The time constant kurtosis is calculated by calculating the normalized fourth-order central moment of the weighted average time constant and subtracting 3 from it, which is then compared with the normal distribution (kurtosis is 3). The time constant kurtosis HI8 is calculated as:

[0100]

[0101] Step 3: Combine the health index, current temperature, and current SOC to obtain an input feature vector, input the input feature vector into the trained SOH estimation model, and estimate the SOH of the lithium-ion battery to be tested.

[0102] The SOH estimation model may adopt an AE model, a CNN model, or an MLP model. The training process of the trained SOH estimation model includes:

[0103] Step 3.1, obtain the EIS data of lithium-ion batteries at different SOH, SOC and ambient temperatures, preprocess the EIS data, and obtain preprocessed EIS data; in actual work, by cyclically charging and discharging several new batteries, the batteries are aged to different SOH levels between 100% and 80%, and at each SOH level, EIS measurements are performed under different SOC (for example, 5% to 95%) and different temperatures (for example, 15°C, 25°C, 35°C). The way to preprocess the EIS data is: use the Savitzky-Golay filtering algorithm to smooth and denoise the EIS data. The preprocessed EIS data and the corresponding SOH, SOC, and ambient temperature work are used as a data set, and the data set is divided into a training set and a test set.

[0104] Step 3.2: Use the DRT method to analyze the preprocessed EIS data to obtain the DRT curve, extract the health index from the DRT curve, and combine the health index, ambient temperature, and SOC to obtain the input feature vector.

[0105] Step 3.3: Using the input feature vector as input and the battery SOH as the output label, train the SOH estimation model. When the loss function is minimized, the trained SOH estimation model is obtained. In this embodiment, the SOH estimation model uses the AE model, which performs better than the CNN model and the MLP model. The introduction of the autoencoder model, through its powerful nonlinear mapping and feature learning capabilities, can discover deep features in the data that are intrinsically related to SOH, effectively filtering out noise interference, thereby improving the model's generalization ability and providing a precise, robust, and valuable technical solution for battery management systems.

[0106] During training, the AE model is trained using the training set and fed into the test set. The AE model is an autoencoder regression model consisting of an encoder and a decoder. The encoder maps the high-dimensional input feature vector into a feature representation in an information-dense low-dimensional latent space, while the decoder maps the feature representation into the final battery SOH estimate. The AE model is trained using a training dataset with known SOH labels, and network parameters are optimized by minimizing the error between the predicted SOH and the true SOH.

[0107] The present invention uses DRT analysis instead of directly using raw EIS data, which can separate overlapping electrochemical processes. The eight key health indicators extracted are effective health characteristics of the battery electrochemical process. These indicators are closely related to the battery's aging mechanisms such as SEI film growth, active material loss and impedance increase. Therefore, they are more physically meaningful and can systematically capture key information related to battery aging, providing high-quality input for subsequent models and improving the accuracy of SOH estimation. By taking temperature and SOC as key inputs, the impact of operating condition changes on EIS measurements is effectively compensated, so that the model can maintain a consistent low error level under a wide range of operating environments, showing excellent robustness.

[0108] Example 2

[0109] like Figure 7 As shown, this embodiment provides a system for estimating the health status of a lithium-ion battery, including:

[0110] The data acquisition module is used to obtain the EIS data, current temperature and current SOC of the lithium-ion battery to be tested.

[0111] The feature extraction module is used to analyze EIS data using the DRT method to obtain the DRT curve and extract health indicators from the DRT curve. The health indicators include peak amplitude, time constant corresponding to the peak, peak area, full width at half maximum, weighted average time constant, weighted standard deviation time constant, time constant skewness and time constant kurtosis.

[0112] The process of analyzing EIS data using the DRT method includes:

[0113] Fit the EIS data to obtain the relaxation time distribution function F(τ), use a set of time constants to represent the battery impedance spectrum Z(ω), and obtain the DRT curve by solving the minimization error function. The error function is:

[0114]

[0115] in, ω is the frequency, τ is the time constant, and λ is the regularization parameter.

[0116] Among the health indicators, the peak amplitude represents the intensity or dominance of the electrochemical process of the battery at the characteristic reaction rate; the time constant corresponding to the peak represents the characteristic reaction rate of the battery's electrochemical process; the peak area is proportional to the total polarization resistance of the battery's electrochemical process, representing the contribution of the electrochemical process to the total impedance; the full width at half maximum represents the uniformity or dispersion of the distribution of the time constant of the electrochemical process; the weighted average time constant reflects the central position of the DRT curve; the weighted standard deviation time constant represents the dispersion of the distribution of the reaction rate of the electrochemical process; the time constant skewness represents the asymmetry of the peak distribution of the DRT curve; the time constant kurtosis represents the sharpness of the peak of the DRT curve or the degree of "tail heaviness".

[0117] In the process of extracting health indicators from the DRT curve, after determining the boundary of the DRT curve peak, find the maximum value of the function F(τ) in the interval to obtain the peak amplitude. The horizontal coordinate corresponding to the peak amplitude is the time constant corresponding to the peak. The DRT curve peak is numerically integrated on the logarithmic time axis to obtain the peak area. The calculation method of the full width at half maximum (FWHM) is: FWHM = |τ2-τ1|, τ1 and τ2 are the two time constant points corresponding to half the peak amplitude. At each time constant point τ i As a value, its corresponding function value F(τ i ) as the weight, and perform weighted averaging to obtain the weighted average time constant; calculate the standard deviation of the weighted average time constant to obtain the weighted standard deviation time constant; calculate the standardized third-order central moment of the weighted average time constant to obtain the time constant skewness; calculate the standardized fourth-order central moment of the weighted average time constant and then subtract 3 to obtain the time constant kurtosis.

[0118] The calculation formula of peak area HI3 is:

[0119]

[0120] Among them, τ start , τ end They are the starting and ending time constant points of the DRT curve respectively.

[0121] The calculation formula of the weighted average time constant HI5 is:

[0122]

[0123] The calculation formula of the weighted standard deviation time constant HI6 is:

[0124]

[0125] The calculation formula of time constant skewness HI7 is:

[0126]

[0127] The calculation formula of time constant kurtosis HI8 is:

[0128]

[0129] The SOH estimation module is used to combine the health index, current temperature and current SOC to obtain an input feature vector, input the input feature vector into the trained SOH estimation model, and estimate the SOH of the lithium-ion battery to be tested.

[0130] The training process of the trained SOH estimation model includes:

[0131] EIS data of lithium-ion batteries at different SOH, SOC and ambient temperatures are obtained, and the EIS data are preprocessed to obtain preprocessed EIS data; the EIS data are preprocessed by using a Savitzky-Golay filtering algorithm to smooth and denoise the EIS data.

[0132] The DRT method is used to analyze the preprocessed EIS data to obtain the DRT curve. The health index is extracted from the DRT curve, and the input feature vector is obtained by combining the health index, ambient temperature and SOC.

[0133] The input feature vector is used as input and the battery SOH is used as the output label to train the SOH estimation model. When the loss function is minimized, the trained SOH estimation model is obtained.

[0134] The SOH estimation model is an AE model, a CNN model, or an MLP model. The SOH estimation model includes:

[0135] An encoder, which maps the input feature vector into a feature representation in a low-dimensional latent space;

[0136] A decoder that maps feature representations to battery SOH estimates.

[0137] Example 3

[0138] This example aims to verify the SOH estimation method of the present invention. The data used is from public literature and contains 25 new NMC 811 cylindrical batteries with a rated capacity of 5Ah. These batteries are aged to different SOH levels between 100% and 80% through cyclic charge and discharge. At each SOH level, EIS measurements are performed under different SOC (5% to 95%) and temperature conditions (15°C, 25°C, 35°C). In the experiment, part of the battery data is used as a test set, and the rest is used as a training set.

[0139] The specific estimation process is as follows: First, all EIS data are subjected to Savitzky-Golay filtering to smooth noise. Next, DRT analysis is performed on each EIS data set using the MATLAB DRT toolbox, and the eight health indicators described above are extracted from the generated DRT spectra. Figure 2 and Figure 3 The systematic changes of DRT curve with SOH and the good correlation between key health indicators and SOH are demonstrated, verifying the effectiveness of the features.

[0140] Subsequently, an autoencoder (AE) model was constructed, with the extracted 8 health indicators and the corresponding temperature and SOC values, a total of 10 variables, as model input, the battery's true SOH as the output label, and trained using the training set.

[0141] In the model evaluation phase, the test set data was input into the trained model. The results showed that the overall prediction accuracy of this method was very high (RMSE = 2.6542, MAE = 1.8870). Figure 4 and Figure 5 As shown in Figure 2, the performance analysis of different working conditions shows that the model is stable in most working conditions. Finally, by comparing with CNN and MLP models, as shown in Figure 2, Figure 6 As shown, it can be clearly seen that the AE model of the present invention has generally lower errors and more balanced performance at all test operating points. Through the description of this embodiment, it can be confirmed that the method proposed by the present invention can effectively, accurately and stably estimate the SOH of lithium-ion batteries, and has strong practicality.

[0142] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for estimating the health status of a lithium-ion battery, characterized by: Methods include: Obtain EIS data, current temperature and current SOC of the lithium-ion battery to be tested; The DRT method was used to analyze EIS data to obtain the DRT curve, from which health indicators were extracted. The health indicators included peak amplitude, time constant corresponding to the peak, peak area, full width at half maximum, weighted average time constant, weighted standard deviation time constant, time constant skewness, and time constant kurtosis. The health index, current temperature and current SOC are combined to obtain an input feature vector, which is input into the trained SOH estimation model to estimate the SOH of the lithium-ion battery to be tested.

2. The method for estimating the health status of a lithium-ion battery according to claim 1, wherein: The process of analyzing EIS data using the DRT method includes: Fit the EIS data to obtain the relaxation time distribution function F(τ), use a set of time constants to represent the battery impedance spectrum Z(ω), and obtain the DRT curve by solving the minimization error function. The error function is: in, ω is the frequency, τ is the time constant, and λ is the regularization parameter.

3. The method for estimating the health status of a lithium-ion battery according to claim 1, wherein: Among the health indicators, the peak amplitude represents the intensity or dominance of the electrochemical process of the battery at the characteristic reaction rate; the time constant corresponding to the peak represents the characteristic reaction rate of the battery electrochemical process; the peak area is proportional to the total polarization resistance of the battery electrochemical process, and represents the contribution of the electrochemical process to the total impedance; the full width at half maximum represents the uniformity or dispersion of the distribution of the time constant of the electrochemical process; the weighted average time constant reflects the central position of the DRT curve; the weighted standard deviation time constant represents the dispersion of the distribution of the reaction rate of the electrochemical process; the time constant skewness represents the asymmetry of the peak distribution of the DRT curve; the time constant kurtosis represents the sharpness of the peak of the DRT curve or the degree of "tail heaviness".

4. The method for estimating the health status of a lithium-ion battery according to claim 1 or 3, characterized in that: In the process of extracting health indicators from the DRT curve, after determining the boundary of the DRT curve peak, find the maximum value of the function F(τ) in the interval to obtain the peak amplitude. The horizontal coordinate corresponding to the peak amplitude is the time constant corresponding to the peak. The DRT curve peak is numerically integrated on the logarithmic time axis to obtain the peak area. The calculation method of the full width at half maximum (FWHM) is: FWHM = |τ2-τ1|, τ1 and τ2 are the two time constant points corresponding to half the peak amplitude. At each time constant point τ i As a value, its corresponding function value F(τ i ) as the weight, and perform weighted averaging to obtain the weighted average time constant; calculate the standard deviation of the weighted average time constant to obtain the weighted standard deviation time constant; calculate the standardized third-order central moment of the weighted average time constant to obtain the time constant skewness; calculate the standardized fourth-order central moment of the weighted average time constant and then subtract 3 to obtain the time constant kurtosis.

5. The method for estimating the health status of a lithium-ion battery according to claim 4, wherein: The calculation formula of peak area HI3 is: Among them, τ start , τ end They are the starting and ending time constant points of the DRT curve respectively.

6. The method for estimating the health status of a lithium-ion battery according to claim 4, wherein: The calculation formula of the weighted average time constant HI5 is: The calculation formula of the weighted standard deviation time constant HI6 is: The calculation formula of time constant skewness HI7 is: The calculation formula of time constant kurtosis HI8 is:

7. The method for estimating the health status of a lithium-ion battery according to claim 1, wherein: The training process of the trained SOH estimation model includes: Obtain EIS data of lithium-ion batteries at different SOH, SOC and ambient temperatures, preprocess the EIS data, and obtain preprocessed EIS data; The DRT method is used to analyze the preprocessed EIS data to obtain the DRT curve. The health index is extracted from the DRT curve, and the input feature vector is obtained by combining the health index, ambient temperature and SOC. The input feature vector is used as input and the battery SOH is used as the output label to train the SOH estimation model. When the loss function is minimized, the trained SOH estimation model is obtained.

8. The method for estimating the health status of a lithium-ion battery according to claim 7, wherein: The method of preprocessing EIS data is to use Savitzky-Golay filtering algorithm to smooth and denoise the EIS data.

9. The method for estimating the health status of a lithium-ion battery according to claim 7, wherein: The SOH estimation model is an AE model, a CNN model, or an MLP model. The SOH estimation model includes: An encoder, which maps the input feature vector into a feature representation in a low-dimensional latent space; A decoder that maps feature representations to battery SOH estimates.

10. A lithium-ion battery health status estimation system, characterized by: The system includes: Data acquisition module, used to obtain EIS data, current temperature and current SOC of the lithium-ion battery to be tested; The feature extraction module is used to analyze EIS data using the DRT method to obtain the DRT curve and extract health indicators from the DRT curve. The health indicators include peak amplitude, time constant corresponding to the peak, peak area, full width at half maximum, weighted average time constant, weighted standard deviation time constant, time constant skewness and time constant kurtosis; The SOH estimation module is used to combine the health index, current temperature and current SOC to obtain an input feature vector, input the input feature vector into the trained SOH estimation model, and estimate the SOH of the lithium-ion battery to be tested.

Images