Lithium metal battery health state estimation method driven by multi-feature data and application

Abstract

The invention discloses a lithium metal battery health state estimation method driven by multi-feature data and application, and belongs to the technical field of battery management. Aiming at the problem of complex nonlinear degradation of the lithium metal battery caused by dendrite growth and interface evolution, cross-band multi-source feature extraction is carried out by fusing grey correlation analysis (GRA) and a simplified equivalent circuit model (SECM): the GRA is utilized to quantify the nonlinear correlation degree of high-frequency features and SOH in an electrochemical impedance spectroscopy (EIS), and five key frequency domain features are screened out; meanwhile, five physical parameter characteristics reflecting the solid-phase diffusion and charge transfer process in a low-frequency region are extracted through SECM. And in combination with multiple working condition variables such as pre-tightening force, charge-discharge rate, SOC and the like, an EIS-SOH estimation model based on GPR (Gaussian process regression) is constructed.

Description

Technical Field

[0001] This invention belongs to the field of lithium metal battery state estimation technology, specifically relating to a multi-feature data-driven method for estimating the health status of lithium metal batteries and its application. Background Technology

[0002] With the strategic deployment of the low-altitude economy globally (e.g., the implementation of the US Federal Aviation Administration's (FAA) Advanced Air Traffic Space Integration Program (2023) and the European Union Aviation Safety Agency's (EASA) Special Airworthiness Certification Framework for VTOL (SC-VTOL)), emerging vehicles such as electric vertical takeoff and landing (eVTOL) aircraft and long-endurance logistics drones are placing disruptive demands on the energy density and cycle stability of energy systems. Lithium metal batteries (LMBs), with their ultra-high theoretical specific capacity (3860 mAh / g) and low electrochemical potential (-3.04 V vs. SHE), can break through the energy density ceiling of existing lithium-ion batteries (LIBs) (>500 Wh / kg), and are expected to become a core option for extending the range and payload capacity of low-altitude vehicles. However, their safety and failure mechanisms severely restrict their application in low-altitude scenarios. The heterogeneous nucleation characteristics of lithium deposition lead to continuous dendrite growth, which can penetrate the separator after a short cycle. Frequent structural reconstruction of the solid electrolyte interphase (SEI) membrane of the negative electrode causes the capacity decay to exhibit strong nonlinearity, and the accumulation of dead lithium leads to safety runaway and cumulative capacity loss. The interplay of these mechanisms makes SOH prediction for LMB more complex than that for LIB. Traditional health management methods based on voltage-capacity models are severely distorted. EIS technology, with its advantages of in-situ non-destructive, multi-frequency sensitivity, and strong mechanism correlation, combined with multi-dimensional degradation feature fusion modeling, is expected to achieve high-precision quantitative estimation of SOH for lithium metal batteries, providing key assurance for the safe operation of low-altitude vehicles.

[0003] Current EIS-based methods for estimating the state of harm (SOH) of lithium-ion batteries mainly fall into two categories: Equivalent Circuit Model (ECM) parameter methods and machine learning fusion methods. ECM parameter methods construct physical mechanism-driven equivalent circuits and extract key parameters such as charge transfer resistance to characterize aging behavior. However, due to significant changes in the physical properties of batteries during cycling (such as interface reaction kinetic drift), accurately constructing a universal ECM model remains challenging. In contrast, machine learning fusion methods directly input full-spectrum EIS data into intelligent algorithms, achieving end-to-end SOH estimation without relying on complex mechanistic models. Data-driven methods offer significant advantages in high-precision and robust SOH assessment. The aforementioned research primarily focuses on lithium-ion battery systems, with insufficient exploration of high-energy-density lithium metal batteries. During cycling, dendrite growth and dynamic interface evolution in lithium metal batteries make traditional ECM modeling difficult. Furthermore, the high dimensionality of full-band EIS data introduces a large amount of non-aging-related noise, increasing computational burden and reducing prediction accuracy. In particular, there is a lack of coupled analysis of the volume expansion effect during lithium metal deposition / stripping.

[0004] For example, CN202510508660.9, "A Fast Estimation Method for State of Harm (SOH) of Lithium-ion Batteries Based on Electrochemical Impedance Spectroscopy," uses Pearson correlation coefficient (PCC) to select two key feature parameters (minimum impedance amplitude and real intercept of the curve) with the strongest linear correlation to SOH from EIS data. Then, it uses multiple linear regression to establish a simple functional relationship between these parameters and SOH. Ultimately, only the high-frequency EIS data of the battery is needed to quickly calculate SOH. The core shortcoming of this existing technology lies in its linearized and simplified approach: it only uses Pearson correlation coefficient to select high-frequency EIS features and employs multiple linear regression modeling. This method cannot capture the complex nonlinear aging mechanism of batteries (especially lithium metal batteries). Furthermore, this technology completely ignores the key influence of multivariate coupled operating conditions such as mechanical preload and charge / discharge rate on battery aging. Moreover, its model and features are only designed for lithium-ion batteries and lack the ability to transfer and generalize to more complex battery systems (such as lithium metal batteries), resulting in severely limited estimation accuracy and robustness in complex practical applications. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by proposing a multi-feature data-driven method and application for estimating the state of health (SOH) of lithium metal batteries. It employs a grey relational-equivalent circuit (GRA) collaborative feature selection and data-driven modeling framework. Key frequency points are selected by quantifying the nonlinear correlation between EIS features and SOH in the mid-to-high frequency band (0.1 Hz–100 kHz). Low-frequency SECM is then used to extract physical parameters of solid-state diffusion and charge transfer (such as SEI impedance and diffusion impedance). A GPR model is constructed based on a 10-dimensional cross-frequency feature set (5 GRA features + 5 SECM parameters) and operating condition variables (preload, rate capability, and SOC) to achieve SOH estimation under multi-variable coupling conditions.

[0006] This invention is implemented as follows: a multi-feature data-driven method for estimating the health status of lithium metal batteries, with the specific steps as follows: Step 1: Apply pre-tightening force to the lithium metal battery. Two sets of experiments with different pre-tightening forces were designed, namely 0 Newtons and x Newtons. Step 2: Set the charge and discharge rates of the lithium metal battery. Two sets of charge and discharge rate experimental conditions were designed, namely 0.2C and 0.5C. Step 3: Conduct charge-discharge cycle tests on the lithium metal battery that has been preloaded and whose charge-discharge rate has been determined. During the charge-discharge cycle tests, perform battery SOC calibration and EIS testing. Step 4: Filter the acquired EIS data; Step 5: Extract dual-path features from the filtered EIS data using both GRA and SECM analysis paths. Step 6: Combine the 5 SECM physical parameters and 5 GRA frequency domain features into a 10-dimensional feature set, and combine them with the working condition variables preload F, ratio C, and SOC to form the model input vector. With SOH as the output variable, construct the SOH estimation model using a Gaussian process regression model. Step 7: Input the fused feature set of the battery under test into the constructed model to obtain the SOH value of the lithium metal battery under test.

[0007] Furthermore, step 3 specifically includes: Step 3-1: Charge the empty-state lithium metal battery with constant current at the charge / discharge rate set in Step 2 until the battery voltage reaches the upper limit cutoff voltage. Step 3-2: Let the lithium metal battery stand for 1 hour at 25℃, calculate the charging capacity of the lithium metal battery, and perform SOH and 100% SOC calibration for this cycle. Step 3-3: Perform EIS testing of lithium metal batteries at 100% SOC using an electrochemical workstation; Steps 3-4: Based on the calculation results of the charging stage capacity, perform constant current discharge at the charge / discharge rate set in step 2, with the discharge amount being half of the charging capacity. Steps 3-5: Let the lithium metal battery stand at 25℃ for 1 hour to perform 50% SOC calibration for this cycle; Steps 3-6: Perform EIS testing of lithium metal batteries at 50% SOC using an electrochemical workstation; Steps 3-7: Perform constant current discharge on the lithium metal battery at the charge / discharge rate set in step 2 until the battery voltage reaches the lower cutoff voltage. Steps 3-8: Let the lithium metal battery stand at 25℃ for 1 hour to perform 0% SOC calibration for this cycle; Steps 3-9: Perform EIS testing of lithium metal batteries at 0% SOC using an electrochemical workstation.

[0008] Furthermore, step 4 involves filtering the acquired initial EIS data, as detailed below: For the mid-to-high frequency range (10 Hz–10 kHz), a Savitzky-Golay filter is used for local smoothing with a window width of 15 frequency points and a polynomial order of 3. For the low-frequency region (0.05Hz–10 Hz), wavelet thresholding is used for noise reduction, employing a soft thresholding function:

[0009] Filtering out high-frequency random noise, where the threshold , σ The standard deviation of noise. N For the number of data points, These are the estimated wavelet coefficients after thresholding. These are the original noisy wavelet coefficients. j For scale parameters, k These are the translation parameters.

[0010] Furthermore, step 5 involves feature extraction from the filtered EIS data, using the following specific method: SECM physical parameter extraction: Based on the simplified equivalent circuit model, the EIS data was fitted to extract five physical parameters, R0, R2, CPE2-T, CPE2-P and W1-P, which reflect the solid-phase diffusion and charge transfer process. The fitting error was controlled within 20%. GRA frequency domain feature screening: The gray correlation degree between 170 features of EIS data, including the real part, imaginary part, and real-imaginary part ratio, and SOH is calculated, with a resolution coefficient ρ=0.2. The 5 mid-to-high frequency features with the highest correlation degree are then screened.

[0011] Further, step 6 involves constructing the EIS-SOH model, with the specific steps as follows: Step 6-1: Based on the EIS features of the lithium metal battery extracted in Step 5, as well as the preload, charge / discharge rate, and state of charge (SOC) parameters, construct a feature dataset characterizing the battery's health status. The feature matrix is ​​defined as X = [R0, R2, CPE2-T, CPE2-P, W1-P, F]. 129 F 136 F 137 F 138 F 139 [, F, C, SOC]; Step 6-2: Model using GPR, its mathematical expression is:

[0012] in, Let be the objective function. gp Represents a Gaussian process. X and For the input sample matrix, a constant basis function is chosen as the mean function. ,Right now: ,in The constant bias term to be optimized is used to eliminate the zero-mean bias after data normalization; the autocorrelation-determined squared exponent kernel ARD-SE is used as the covariance function.

[0013] in, For the squared exponential (SE) kernel function with automatic correlation detection (ARD), x i ,x j For the input feature vector, D For the input feature dimension, For the adaptive length scale of the d-th dimension feature, The signal variance controls the fluctuation amplitude of the control function. The noise variance characterizes the observation noise. The Kronecker delta function is used; when fitting data, the hyperparameters that need to be optimized in GPR include the basis function coefficients. Signal variance Noise variance and length scale These hyperparameters are optimized by maximizing the log-marginal likelihood function:

[0014] in, For log-marginal likelihood, yThis is the observation vector (or target vector). X To train the input matrix, Let be the hyperparameter of the covariance function. K It is the covariance matrix (or kernel matrix). I It is the identity matrix. C For constant terms; Step 6-3: Train the GPR model using the constructed feature matrix and the corresponding SOH values.

[0015] Furthermore, in step 7, the specific steps for estimating the SOH of the lithium metal battery are as follows: Step 7-1: Measure the EIS data of the lithium metal battery to be tested and determine its operating conditions. Extract its feature matrix according to the methods in Steps 4, 5 and 6. Step 7-2: Input the extracted feature matrix into the trained GPR model; Step 7-3: The model outputs the estimated SOH value of the lithium metal battery.

[0016] A computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the above-described method.

[0017] A computer device includes a memory, a processor, and a program stored in the memory and executable thereon, the program being executed by the processor to implement the steps of the method described above.

[0018] The beneficial effects of this application are: This invention uses GRA to quantify the nonlinear correlation between EIS characteristics and SOH in the mid-to-high frequency band (0.1 Hz–100 kHz), selects key frequency points, and combines low-frequency SECM to extract physical parameters of solid-state diffusion and charge transfer (such as SEI impedance, diffusion impedance, etc.). Based on a 10-dimensional cross-band feature set (5 GRA features + 5 SECM parameters) and operating variables (preload, magnification, SOC), a GPR model is constructed to achieve SOH estimation under multivariate coupling conditions.

[0019] This invention, through its proposed multi-feature data-driven framework, achieves for the first time high-precision and robust estimation of the State of Charge (SOH) of lithium metal batteries under the coupled conditions of preload, charge / discharge rate, and state of charge. The average root mean square error is as low as 1.65%, representing an accuracy improvement of over 32% compared to traditional full-feature input methods. Even under extreme operating conditions (such as 0N preload combined with 0.5C high-rate charge / discharge), the estimation error can still be stably controlled within 3%, demonstrating excellent environmental adaptability.

[0020] This invention not only ensures estimation accuracy but also successfully solves the problems of computational efficiency and real-time performance of the estimation model. By using grey relational analysis to select five of the most representative key frequency domain features from the 170 original features and combining them with SECM features, the model training time is significantly reduced from 169.03 seconds in the traditional method to 4.89 seconds, a reduction of 97.1%, which fully meets the stringent requirements of battery management systems for online real-time estimation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart of the multi-feature data-driven method for estimating the health status of lithium metal batteries according to the present invention.

[0023] Figure 2 This is a flowchart of the training and prediction process for the GPR model of the multi-feature data-driven lithium metal battery health state estimation method of the present invention.

[0024] Figure 3 This is a flowchart illustrating the experimental platform setup and single-cell experiment of the multi-feature data-driven lithium metal battery health state estimation method of the present invention.

[0025] Figure 4 The feature correlation ranking chart (top 60) obtained by GRA analysis for the multi-feature data-driven lithium metal battery health state estimation method of this invention is used. Here, F1 to F5 represent the five SECM features, F6 to F60 represent the real part of impedance at 55 frequency measurement points from 10kHz to 0.05Hz, F61 to F115 represent the imaginary part of impedance at 55 frequency measurement points from 10kHz to 0.05Hz, and F116 to F170 represent the ratio of the real part to the imaginary part of impedance at 55 frequency measurement points from 10kHz to 0.05Hz.

[0026] Figure 5 This is a schematic diagram of the SECM model construction for the multi-feature data-driven lithium metal battery health state estimation method of the present invention (the top two figures represent equivalent circuit fitting, and the bottom two figures represent simplified equivalent circuit fitting).

[0027] Figure 6This is an error analysis diagram of the prediction results of the multi-feature data-driven lithium metal battery health state estimation method of the present invention (the naming rule of the 12 variable combinations is based on the combination of three elements: charge / discharge rate, preload and state of charge, and the abbreviation format is uniformly c rate_preload n_SOC).

[0028] Figure 7 This is a comparison chart of SOH prediction results for a multi-feature data-driven lithium metal battery health state estimation method according to the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0030] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.

[0031] like Figure 1 As shown, the multi-feature data-driven lithium metal battery health state estimation method of the present invention includes the following specific steps in the embodiments of this application: Step 1: Apply pre-tightening force to the lithium metal battery using a torque wrench. Two sets of experiments with different pre-tightening forces were designed, namely 0N and 500N.

[0032] Step 2: Set the charge and discharge rates of the lithium metal battery. Two sets of charge and discharge rate experimental conditions were designed, namely 0.2C and 0.5C.

[0033] Step 3: Conduct charge-discharge cycle tests on the lithium metal battery that has been preloaded and whose charge-discharge rate has been determined. During the charge-discharge cycle tests, perform battery SOC calibration and EIS testing.

[0034] Step 4: Filter the acquired EIS data.

[0035] Step 5: Extract dual-path features from the filtered EIS data using both GRA and SECM analysis paths. Step 6: Combine the 5 SECM physical parameters and 5 GRA frequency domain features into a 10-dimensional feature set, and combine them with the operating condition variables (preload F, ratio C, SOC) to form the model input vector. With SOH as the output variable, construct the SOH estimation model using a Gaussian process regression model.

[0036] Step 7: Input the fused feature set of the battery under test into the constructed model to obtain the SOH value of the lithium metal battery under test.

[0037] Furthermore, in this embodiment, step three specifically involves: Step 3-1: Charge the empty-state lithium metal battery with constant current at the charge / discharge rate set in Step 2 until the battery voltage reaches the upper limit cutoff voltage (4.35V).

[0038] Step 3-2: Let the lithium metal battery stand for 1 hour at 25℃, calculate the charging capacity of the lithium metal battery, and perform SOH and 100% SOC calibration for this cycle.

[0039] Step 3-3: Perform EIS testing of lithium metal batteries at 100% SOC using an electrochemical workstation.

[0040] Steps 3-4: Based on the calculation results of the charging stage capacity, perform constant current discharge at the charge / discharge rate set in step 2, with the discharge amount being half of the charging capacity.

[0041] Steps 3-5: Let the lithium metal battery stand for 1 hour at 25℃ to perform the 50% SOC calibration for this cycle.

[0042] Steps 3-6: Perform EIS testing of lithium metal batteries at 50% SOC using an electrochemical workstation.

[0043] Steps 3-7: Perform constant current discharge on the lithium metal battery at the charge / discharge rate set in step 2 until the battery voltage reaches the lower cutoff voltage (3V).

[0044] Steps 3-8: Allow the lithium metal battery to stand at 25°C for 1 hour to perform the 0% SOC calibration for this cycle.

[0045] Steps 3-9: Perform EIS testing of lithium metal batteries at 0% SOC using an electrochemical workstation.

[0046] Furthermore, in this embodiment, step four specifically comprises: Step 4-1: For the mid-to-high frequency range (10 Hz–10 kHz), a Savitzky-Golay filter is used for local smoothing with a window width of 15 frequency points and a polynomial order of 3.

[0047] Step 4-2: For the low-frequency region (0.05Hz–10Hz), wavelet thresholding is used for noise reduction, employing a soft thresholding function:

[0048] Filtering out high-frequency random noise (threshold) , σ The standard deviation of noise. N For the number of data points, These are the estimated wavelet coefficients after thresholding. These are the original noisy wavelet coefficients. j For scale parameters, k (This refers to the translation parameter.)

[0049] Furthermore, in this embodiment, step five specifically includes: Step 5-1: Extraction of SECM physical parameters. Based on the simplified equivalent circuit model, the EIS data is fitted to extract five physical parameters (R0, R2, CPE2-T, CPE2-P, W1-P) that reflect the solid-phase diffusion and charge transfer process. The fitting error is controlled within 20%. Step 5-2: GRA frequency domain feature screening. Calculate the gray correlation degree (resolution coefficient ρ=0.2) between 170 features of EIS data, including the real part, imaginary part, and real-imaginary part ratio, and SOH. Screen the 5 mid-to-high frequency features with the highest correlation degree.

[0050] Furthermore, in this embodiment, step six specifically includes: Step 6-1: Based on the EIS features of the lithium metal battery extracted in Step 5 and multiple physical quantity parameters such as preload, charge / discharge rate, and state of charge, construct a feature dataset characterizing the battery health status. Its feature matrix is ​​defined as X=[R0, R2, CPE2-T, CPE2-P, W1-P, F129, F136, F137, F138, F139, F, C, SOC].

[0051] Step 6-2: Model using GPR, the mathematical expression of which is shown below:

[0052] in, Let be the objective function. gp Represents a Gaussian process. X and For the input sample matrix, a constant basis function is chosen as the mean function. ,Right now: ,in The constant bias term to be optimized is used to eliminate the zero-mean bias after data normalization; the autocorrelation-determined squared exponent kernel ARD-SE is used as the covariance function.

[0053] in, For the squared exponential (SE) kernel function with automatic correlation detection (ARD), x i ,x j For the input feature vector, D For the input feature dimension, For the adaptive length scale of the d-th dimension feature, The signal variance controls the fluctuation amplitude of the control function. The noise variance characterizes the observation noise. The Kronecker delta function is used; when fitting data, the hyperparameters that need to be optimized in GPR include the basis function coefficients. Signal variance Noise variance and length scale These hyperparameters are optimized by maximizing the log-marginal likelihood function:

[0054] in, For log-marginal likelihood, y This is the observation vector (or target vector). X To train the input matrix, Let be the hyperparameter of the covariance function. K It is the covariance matrix (or kernel matrix). I It is the identity matrix. C This is a constant term. This optimization process not only determines the model parameters, but its inherent mechanism also naturally controls the model's complexity.

[0055] Step 6-3: Train the GPR model using the constructed feature matrix and the corresponding SOH values.

[0056] Furthermore, in the embodiments of this application, step seven specifically includes: Step 7-1: Measure the EIS data of the lithium metal battery to be tested and determine its operating conditions. Extract its feature matrix according to the methods in Steps 4, 5 and 6.

[0057] Step 7-2: Input the extracted feature matrix into the trained GPR model.

[0058] Step 7-3: The model outputs the estimated SOH value of the lithium metal battery.

[0059] The present invention will be specifically described below using a lithium-ion battery as an example. Example This example uses a 20Ah lithium metal pouch battery as the experimental subject, employing a 2 (preload force: 0N, 500N) × 2 (charge / discharge rate: 0.2C, 0.5C) experimental design, totaling 4 groups of tests. Each group of tests uses two identical batteries (a total of 8 batteries). All charge / discharge experiments were conducted in a constant temperature (25°C) and low humidity (humidity <30%) environment. The experimental platform and procedures are as follows: Figure 3 As shown. The specific process is as follows: Step 1: Use a torque wrench to apply pre-tightening force to the lithium metal batteries. Apply pre-tightening force to four sets of batteries, with 0N pre-tightening force to two sets and 500N pre-tightening force to the other two sets.

[0060] Step 2: Set the charge / discharge rate for each battery group. For two battery groups with the same preload, set different charge / discharge rates: 0.2C and 0.5C. The operating conditions for the four battery groups are: 0N 0.2C, 0N 0.5C, 500N 0.2C, and 500N 0.5C, respectively.

[0061] Step 3: Perform charge-discharge cycle tests on the four groups of batteries according to the operating conditions determined in Step 1 and Step 2. Record the charging capacity during each charging process to obtain battery health status information. During the discharge process, use an electrochemical workstation to perform EIS tests when the battery SOC is at 100%, 50%, and 0%.

[0062] Step 4: Filter the acquired EIS data.

[0063] Step 5: Perform dual-path feature extraction on the filtered EIS data using both GRA and SECM analysis paths. Perform grayscale correlation analysis on 170 EIS features and sort them in descending order of correlation strength. The top 60 features are listed below. Figure 4 As shown, the construction and characteristics of SECM are as follows: Figure 5 As shown.

[0064] Step Six: Merge the 5 SECM physical parameters and 5 GRA frequency domain features into a 10-dimensional feature set, and combine it with the operating condition variables (preload F, rate of return C, SOC) to form the model input vector, with SOH as the output variable. Select one battery from each group for training, integrate the data from four batteries, and construct the SOH estimation model using a Gaussian process regression model. The model training flowchart is as follows. Figure 2 As shown.

[0065] Step 7: Construct a fused feature set using the EIS data from the remaining four batteries, input it into the trained model, and obtain the SOH prediction result. The prediction result is as follows: Figure 7 As shown, the prediction error is as follows: Figure 6As shown. The final prediction results have an average RMSE of 1.65%, an average MAE of 1.14%, and an average R... 2 It is 0.83.

[0066] The embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. It should be noted that implementations not illustrated or described in the drawings or the main text of the specification are forms known to those skilled in the art and have not been described in detail.

[0067] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented, in whole or in part, as a computer program product, the computer program product includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape) or an optical medium.

[0068] In summary, this invention proposes a multi-feature data-driven health state estimation method for lithium metal batteries. This method innovatively integrates a dual-path feature extraction strategy combining grey relational analysis (GRA) and simplified equivalent circuit model (SECM). By quantifying the nonlinear correlation between mid-to-high frequency electrochemical impedance spectroscopy (EIS) features and state of health (SOH) through GRA, five key frequency domain features are selected. Combined with SECM, five physical parameters reflecting solid-phase diffusion and charge transfer processes in the low-frequency region are extracted. GPR is then used for modeling, effectively solving the challenge of modeling complex nonlinear degradation of lithium metal batteries caused by dendrite growth and interface evolution. Furthermore, by combining multiple operating condition variables such as preload, charge / discharge rate, and state of charge (SOC), an EIS-SOH estimation model based on Gaussian process regression (GPR) is constructed. Its probabilistic output characteristics enable accurate SOH estimation and uncertainty quantification. This provides reliable technical support for battery safety management in dynamic, high-reliability applications such as UAVs and eVTOL.

[0069] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-feature data-driven method for estimating the state of health of lithium metal batteries, characterized in that, The specific steps are as follows: Step 1: Apply pre-tightening force to the lithium metal battery. Two sets of experiments with different pre-tightening forces were designed, namely 0 Newtons and x Newtons. Step 2: Set the charge and discharge rates of the lithium metal battery. Two sets of charge and discharge rate experimental conditions were designed, namely 0.2C and 0.5C. Step 3: Conduct charge-discharge cycle tests on the lithium metal battery that has been preloaded and whose charge-discharge rate has been determined. During the charge-discharge cycle tests, perform battery SOC calibration and EIS testing. Step 4: Filter the acquired EIS data; Step 5: Extract dual-path features from the filtered EIS data using both GRA and SECM analysis paths. Step 6: Combine the 5 SECM physical parameters and 5 GRA frequency domain features into a 10-dimensional feature set, and combine them with the working condition variables preload F, ratio C, and SOC to form the model input vector. With SOH as the output variable, construct the SOH estimation model using a Gaussian process regression model. Step 7: Input the fused feature set of the battery under test into the constructed model to obtain the SOH value of the lithium metal battery under test.

2. The multi-feature data-driven method for estimating the health status of lithium metal batteries according to claim 1, characterized in that, Step 3 specifically involves: Step 3-1: Charge the empty-state lithium metal battery with constant current at the charge / discharge rate set in Step 2 until the battery voltage reaches the upper limit cutoff voltage. Step 3-2: Let the lithium metal battery stand for 1 hour at 25℃, calculate the charging capacity of the lithium metal battery, and perform SOH and 100% SOC calibration for this cycle. Step 3-3: Perform EIS testing of lithium metal batteries at 100% SOC using an electrochemical workstation; Steps 3-4: Based on the calculation results of the charging stage capacity, perform constant current discharge at the charge / discharge rate set in step 2, with the discharge amount being half of the charging capacity. Steps 3-5: Let the lithium metal battery stand at 25℃ for 1 hour to perform 50% SOC calibration for this cycle; Steps 3-6: Perform EIS testing of lithium metal batteries at 50% SOC using an electrochemical workstation; Steps 3-7: Perform constant current discharge on the lithium metal battery at the charge / discharge rate set in step 2 until the battery voltage reaches the lower cutoff voltage. Steps 3-8: Let the lithium metal battery stand at 25℃ for 1 hour to perform 0% SOC calibration for this cycle; Steps 3-9: Perform EIS testing of lithium metal batteries at 0% SOC using an electrochemical workstation.

3. The multi-feature data-driven method for estimating the health status of lithium metal batteries according to claim 1, characterized in that, Step 4 involves filtering the acquired initial EIS data. The specific method is as follows: For the mid-to-high frequency range (10 Hz–10 kHz), a Savitzky-Golay filter is used for local smoothing with a window width of 15 frequency points and a polynomial order of 3. For the low-frequency region (0.05Hz–10 Hz), wavelet thresholding is used for noise reduction, employing a soft thresholding function: Filtering out high-frequency random noise, where the threshold , σ The standard deviation of noise. N For the number of data points, These are the estimated wavelet coefficients after thresholding. These are the original noisy wavelet coefficients. j For scale parameters, k These are the translation parameters.

4. The multi-feature data-driven method for estimating the health status of lithium metal batteries according to claim 1, characterized in that, Step 5 involves feature extraction from the filtered EIS data. The specific method is as follows: SECM physical parameter extraction: Based on the simplified equivalent circuit model, the EIS data was fitted to extract five physical parameters, R0, R2, CPE2-T, CPE2-P and W1-P, which reflect the solid-phase diffusion and charge transfer process. The fitting error was controlled within 20%. GRA frequency domain feature screening: The gray correlation degree between 170 features of EIS data, including the real part, imaginary part, and real-imaginary part ratio, and SOH is calculated, with a resolution coefficient ρ=0.

2. The 5 mid-to-high frequency features with the highest correlation degree are then screened.

5. The multi-feature data-driven method for estimating the health status of lithium metal batteries according to claim 1, characterized in that, Step 6 involves constructing the EIS-SOH model, with the specific steps as follows: Step 6-1: Based on the EIS features of the lithium metal battery extracted in Step 5, as well as the preload, charge / discharge rate, and state of charge (SOC) parameters, construct a feature dataset characterizing the battery's health status. The feature matrix is ​​defined as X = [R0, R2, CPE2-T, CPE2-P, W1-P, F]. 129 F 136 F 137 F 138 F 139 [, F, C, SOC]; Step 6-2: Model using GPR, its mathematical expression is: in, Let be the objective function. gp Represents a Gaussian process. X and For the input sample matrix, a constant basis function is chosen as the mean function. ,Right now: ,in The constant bias term to be optimized is used to eliminate the zero-mean bias after data normalization; the autocorrelation-determined squared exponent kernel ARD-SE is used as the covariance function. in, This is a squared exponential kernel function with automatic correlation detection. x i ,x j For the input feature vector, D For the input feature dimension, For the adaptive length scale of the d-th dimension feature, The signal variance controls the fluctuation amplitude of the control function. The noise variance characterizes the observation noise. The Kronecker delta function is used; when fitting data, the hyperparameters that need to be optimized in GPR include the basis function coefficients. Signal variance Noise variance and length scale These hyperparameters are optimized by maximizing the log-marginal likelihood function: in, For log-marginal likelihood, y This is the observation vector (or target vector). X To train the input matrix, Let be the hyperparameter of the covariance function. K It is the covariance matrix (or kernel matrix). I It is the identity matrix. C For constant terms; Step 6-3: Train the GPR model using the constructed feature matrix and the corresponding SOH values.

6. The multi-feature data-driven method for estimating the state of health of lithium metal batteries according to claim 1, characterized in that, In step 7, the specific steps for estimating the state of harmonics (SOH) of the lithium metal battery are as follows: Step 7-1: Measure the EIS data of the lithium metal battery to be tested and determine its operating conditions. Extract its feature matrix according to the methods in Steps 4, 5 and 6. Step 7-2: Input the extracted feature matrix into the trained GPR model; Step 7-3: The model outputs the estimated SOH value of the lithium metal battery.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the method as described in any one of claims 1-6.

8. A computer device, characterized in that, The computer device includes a memory, a processor, and a program stored in and executable on the memory, the program being executed by the processor to implement the steps of the method as described in any one of claims 1-6.

Images