Lithium battery health condition prediction method

Abstract

The invention discloses a lithium battery health condition prediction method, which comprises the steps of applying frequency-adjustable sinusoidal current excitation to a to-be-detected lithium battery, and obtaining historical impedance spectrum data of the lithium battery; inputting the historical impedance spectrum data of the lithium battery into a preset'improved second-order R-CPE + Warburg-C 'equivalent circuit model, and automatically fitting to obtain five health indexes at least including ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, a Warburg coefficient Y0 and shunt capacitance C; and inputting the five health indexes serving as input features into the trained XGBoost regression model to obtain the SOH predicted value of the lithium battery to be tested. According to the method, original impedance data are subjected to dimensionality reduction to obtain features with physical significance, and single-sample inference is performed; and the computing resource demand is far lower than that of a deep network, so that the lithium battery health state can be quickly and accurately evaluated.

Description

Technical Field

[0001] This invention relates to the field of lithium battery electrochemical technology analysis, and in particular to a method for predicting the health status of lithium batteries. Background Technology

[0002] Lithium-ion batteries play an irreplaceable role in electric vehicles and energy storage. Currently, from a technical standpoint, the development, large-scale production, and application of lithium-ion batteries face no major challenges. However, the complexity and instability of lithium-ion batteries themselves hinder quality control and practical application, causing battery testing levels to lag behind production and application. Common online monitoring methods for lithium-ion batteries involve measuring voltage, current, and temperature during operation or cycling. However, these parameters often exhibit lag and cannot promptly reflect changes in the battery's internal condition. When internal faults or drastic changes occur, the provided warning window is very short, significantly reducing reaction time and making it difficult to make and execute correct decisions within the safety window. Electrochemical impedance spectroscopy (EIS) of lithium-ion batteries has been found to provide rich information about their internal state, showing a close correlation with performance changes and potential failure modes. By analyzing EIS data, the state of lithium-ion batteries can be more accurately determined and estimated.

[0003] Many companies have begun to use EIS to predict the health status of lithium batteries. Among related technologies, Chinese invention patent CN121027892A discloses a dynamic estimation method for the state of health (SOH) of lithium batteries based on electrochemical impedance spectroscopy (EIS) and an ensemble learning algorithm. This method intelligently divides the entire battery lifecycle into three stages by calculating the capacity decay rate (CAR): a lightweight Extreme Learning Machine (ELM) is used for mild decay, a more accurate Convolutional Neural Network (CNN) is switched for moderate decay, and a Support Vector Machine (SVM) is used to ensure stability for deep decay. This method solves the problem of a single model struggling to maintain high accuracy throughout the entire battery lifecycle by dynamically selecting the optimal algorithm. However, this method still suffers from high computational resource requirements and the difficulty in balancing these requirements. Summary of the Invention

[0004] To at least address one of the problems existing in the prior art, the present invention provides a method for predicting the health status of lithium batteries that requires less computing resources.

[0005] To achieve the objective of this invention, a method for predicting the health status of a lithium battery is provided, comprising: A frequency-adjustable sinusoidal current excitation was applied to the lithium battery under test to obtain historical impedance spectrum data of the lithium battery. The historical impedance spectrum data of the lithium battery is input into the preset "improved second-order R-CPE + Warburg-C" equivalent circuit model, and the model is automatically fitted to obtain five health indicators, including at least ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, Warburg coefficient Y0 and parallel capacitance C. Five health indicators are used as input features and fed into a trained XGBoost regression model to obtain the predicted state of health (SOH) value of the lithium battery under test.

[0006] Furthermore, the objective function of the XGBoost regression model includes first-order and second-order Taylor expansions of the loss function and a regularization term to prevent overfitting.

[0007] Furthermore, the amplitude of the sinusoidal current excitation satisfies the linear response condition, and the frequency sweep range is 0.1 Hz–10 kHz.

[0008] Furthermore, the equivalent circuit model includes a series inductor L, used to characterize high-frequency inductive reactance; The ohmic resistor R0 is used to characterize the electrolyte and contact resistance; The first ZARC unit is formed by the first resistor R1 and the first CPE1 connected in parallel, and is used to characterize the SEI impedance; The second ZARC unit, which is formed by the second resistor R2 and the second CPE2 connected in parallel, is used to characterize the charge transfer impedance; The Warburg unit W is connected in parallel with the third CPE3 and then in series with the second ZARC unit to characterize the low-frequency diffusion process and the additional capacitance effect.

[0009] Furthermore, the automatic fitting was completed in batches using Python scripts, and the fitting results were verified by Kramers-Kronig, removing unqualified data.

[0010] Furthermore, the XGBoost regression model is trained in the following manner: Obtain n sets of GEIS data for lithium batteries with known state of equilibrium (SOH), and calibrate the true SOH value using the capacity method; A feature vector is constructed using the five health indicators, and a training set is formed using the true values ​​of SOH as labels. Using an additive model and forward distribution optimization, T CART regression trees are iteratively constructed, and the objective function is minimized:

[0011] in, Let be the objective function. It is a decision tree. , These are the first and second derivatives of the loss function with respect to the predicted values, respectively. , The regularization coefficient is... The number of leaf nodes. Leaf weight, Indicates the first The set of indices of all samples in a leaf node.

[0012] The present invention also provides a lithium battery health status prediction system.

[0013] The present invention also provides a computer device.

[0014] The present invention also provides a computer-readable storage medium.

[0015] Compared with the prior art, the beneficial effects that the present invention can achieve include: (1) This method has a fast computation speed and low system complexity. Compared with the traditional neural network system XGBoost, it has a faster computation speed and lower system complexity, and can quickly and accurately calculate the health status of different batteries.

[0016] (2) This method achieves dimensionality reduction of the data based on equivalent circuit fitting, transforming dozens of data points in the impedance data into several equivalent circuit characteristic values, and assigning physical meaning to each point of the disordered impedance data using the equivalent circuit method. While improving the calculation speed, it constructs a framework to explain the correlation between data point displacement and aging in the physical world.

[0017] (3) This method achieves batch processing of impedance data through program fitting. The pre-written equivalent circuit is suitable for replacement, meeting the fitting requirements of impedance data of different types of batteries.

[0018] (4) This method uses feature value extraction to analyze the correlation between different feature values ​​and battery health status, thereby further improving the accuracy of the analysis. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of historical impedance spectrum data of a lithium battery, which is a method for predicting the health status of a lithium battery proposed in an embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of the equivalent circuit fitting procedure for a lithium battery health status prediction method proposed in an embodiment of the present invention.

[0021] Figure 3 The diagram shows the equivalent circuit structure of the fitting procedure for a lithium battery health status prediction method proposed in this embodiment of the invention.

[0022] Figure 4This is a schematic diagram of the XGBoost algorithm in a lithium battery health status prediction method proposed in an embodiment of the present invention.

[0023] Figure 5 This is a comparison chart of the prediction results and actual health status of a lithium battery health status prediction method proposed in an embodiment of the present invention.

[0024] Figure 6 This is a flowchart illustrating the steps of the prediction method proposed in an embodiment of the present invention. Detailed Implementation

[0025] The embodiments of the present invention will now be described in detail. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are only set for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0026] This invention provides a lightweight, high-precision method for predicting the health status of lithium batteries, comprising the following steps: Step (1): Apply a small sinusoidal current excitation to the battery in the range of 0.1 Hz–10 kHz to obtain historical impedance spectrum data (GEIS data) of lithium battery.

[0027] In one embodiment, a frequency-adjustable sinusoidal current excitation is applied to the lithium battery under test to obtain historical impedance spectrum data of the lithium battery. Figure 1 The image shows the acquired historical impedance spectrum data of the lithium battery, obtained by cyclically charging and discharging the lithium battery using the Xinwei charge-discharge device, and is defined as cyclic charge-discharge data. The Xinwei charge-discharge device is configured to: generate a corresponding amplitude of charge-discharge current and rest interval based on the characteristics of the lithium battery under test and the parameters of the previous charge-discharge cycle, and to control the charge-discharge rate and battery relaxation time.

[0028] The cyclic charge-discharge data are obtained by recording the lithium battery electrochemical impedance spectroscopy under different variables, using different states of charge and number of cycles as variables, during the cyclic charge-discharge process.

[0029] In one embodiment, the cyclic charge-discharge data is extracted at ten-cycle intervals, collecting the charge input in the last cycle to compare with the initial capacity and determine the battery's true health condition. Then, the following operations are performed: 1. Measure and record the electrochemical impedance spectroscopy (EIS) of the lithium battery at full charge. During the EIS measurement, the excitation current output i(t) = A1sin(ωx + φ1) is injected into the battery and the sampling resistor R. The frequency scan is 0.1 Hz–10 kHz, with 10 points taken every tenth harmonic. The amplitude A1 is taken as the current corresponding to 0.1 C of the battery's rated capacity to ensure linear response. ω, x, and φ1 represent the angular frequency, time, and initial phase of the excitation current, respectively.

[0030] 2. The lithium battery is discharged by controlling the discharge through the Xinwei charging and discharging device, and the amount of electricity discharged is 10% of the full charge in the current state; 3. Allow 30 minutes for the internal reaction process of the battery to stabilize before continuing to measure the electrochemical impedance spectroscopy of the lithium battery; 4. Repeat steps 2 and 3 until the lithium battery is depleted.

[0031] Through steps 1 to 4, a total of 11 sets of impedance spectrum data are obtained. Repeat the charge-discharge cycle and execute steps 1 to 4 until sufficient data is obtained.

[0032] Step (2): Equivalent circuit fitting: Input the GEIS data into the preset "improved second-order R-CPE+Warburg-C" equivalent circuit model, and automatically fit to obtain five health indicators including at least the ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, Warburg coefficient Y0 and parallel capacitance C.

[0033] The fitting scripts were completed in batches using Python and filtered for Kramers-Kronig residuals ≤1%.

[0034] The parameters of different components in the equivalent circuit can reflect the internal electrochemical behavior of the battery, and the equivalent circuit parameters can serve as characteristic values ​​reflecting the battery's health status. Planning the equivalent circuit can not only simplify the original impedance data but also extract characteristic values ​​reflecting the internal electrochemical behavior of the battery. Figure 2 This diagram shows a Python program that pre-writes equivalent circuits for fitting impedance spectra and determines the parameter sizes of each equivalent circuit element based on the input impedance spectrum fitting results. The first 11 lines of code in the diagram represent the environment configuration required for fitting. The fitting process does not require manual input of impedance data; simply import the data to a specified address, set the storage address for the fitting results, and run the program to achieve batch fitting. All address-related code is shown in lines 12 to 14 of the diagram. The last line of code displays all the equivalent components used in the equivalent circuit fitting.

[0035] The equivalent circuit includes: inductor L0, ohmic resistor R0, first ZARC (R1∥CPE1), second ZARC (R2∥CPE2), Warburg cell W, and capacitor C, with Warburg cell W and capacitor C connected in parallel. Batch fitting was performed using the Python-Levenberg-Marquardt algorithm. Initial values ​​were obtained from a global optimization algorithm. After fitting, the KK residual (KK, Kramers-Kronig, used to determine the reliability of experimentally obtained electrochemical impedance data) was calculated. If the residual > 1%, the data point was discarded.

[0036] Figure 3 Three different equivalent circuit structures and their fitting results are shown. (a), (b), and (c) represent three different equivalent circuits. Figures (b) and (c) respectively show the first-order and second-order R-CPE models commonly used in lithium battery equivalent circuit fitting; Figure (a) is an improvement on the second-order R-CPE model, with a capacitor connected in parallel across the Warburg diffusion impedance at the end to compensate for the internal capacitive characteristics of the lithium battery at low frequencies. Clearly, Figure 3 (a) The corresponding equivalent circuit shows the best fitting effect, indicating that paralleling a capacitor on the Warburg unit with low-frequency impedance can effectively improve the fitting accuracy. Therefore, the mentioned embodiment adopts... Figure 3 (a) Equivalent circuit fitting impedance data.

[0037] Step (3): XGBoost Regression Model Training and Prediction: Using the five health indicators as input features, input the trained XGBoost regression model to obtain the predicted state of health (SOH) value of the lithium battery under test. The XGBoost training set is calibrated with the SOH true value using the capacity method, and a CART tree is constructed using additive serial methods to balance accuracy and overfitting suppression.

[0038] The prediction principle and structure of the XGBoost algorithm are as follows: Figure 4 As shown in the figure, the input feature represented by X is input into the first decision tree H1(x), and then the error ε1 of the first decision tree is used as the second decision tree H2(x) through Taylor expansion. This process is repeated until the last decision tree H2(x) is reached. n (x), the final prediction is the sum of the outputs of all trees. The XGBoost algorithm uses the residual approximation term from the boosting tree algorithm. In the current model's negative regression problem of the loss function, it utilizes the Taylor second-order expansion and regularization term of the loss function to improve model accuracy. XGBoost also uses the second-order gradient as "weights," locally approximating the original loss surface as a weighted least squares problem. The larger the second-order gradient, the greater the curvature of the surface near the sample, and the model will force the learning speed of that region to decrease, thereby reducing the risk of overfitting.

[0039] XGBoost belongs to the Boosting algorithm family. It connects multiple CART learning trees in a chain using an additive model and forward distribution optimization. Although the trees must be generated serially, the additive model allows for parallel inference: all leaf indices can be looked up at once. Therefore, XGBoost can achieve a single-sample inference latency of less than 0.1ms on GPUs / multi-core CPUs. In the corresponding embodiment, each learning tree outputs its own prediction result and error. The error is used as input correction for the next level learning tree, and the error of the next level learning tree is corrected and reduced, achieving high-precision prediction.

[0040] In one embodiment, the XGBoost regression model is trained in the following manner: Obtain n sets of GEIS data for lithium batteries with known state of equilibrium (SOH), and calibrate the true SOH value using the capacity method; Constructing a feature vector using five health indicators Using the true value of SOH as the label , forming a training set D={( , )}; Using an additive model and forward distribution optimization, M CART regression trees are iteratively constructed, and the objective function is minimized:

[0041] in The objective function is the structure score. It is a complete decision tree. , These are the first and second derivatives of the loss function with respect to the predicted values, respectively. It is a counting ordinal number. Indicates the first The set of indices of all samples in each leaf node. , The regularization coefficient is... The number of leaf nodes. Here, T represents the leaf weight, and T is the number of leaf nodes.

[0042] Step (4): Feature extraction There are five parameters for each component obtained through equivalent circuit processing (i.e., the five health indicators mentioned above). The correlation between each parameter and the battery's state of health (SOH) is not entirely the same, the degree of aging corresponding to each part is not entirely the same, and the trends and distinguishability of change also have certain differences. In one embodiment, the parameter R0, which reflects the ohmic internal resistance of the lithium battery, accounts for the highest proportion among all parameters, indicating that the ohmic internal resistance R0 has the closest relationship with the SOH during the prediction process compared to other parameters.

[0043] Step (5): Comparison of prediction results.

[0044] The predicted health status of lithium batteries is compared with the actual health status. Regression analysis is used to determine the accuracy of the prediction, show the trend of predicted SOH, and calculate the prediction evaluation index.

[0045] When the microprocessor hardware of the device (battery data acquisition device) in this embodiment is powered on, it initializes the peripherals in a unified manner. During the test, it exchanges instructions and data with the host computer via USB 2.0 and schedules the peripherals in real time to realize the timing control and gain adaptation of the entire electrochemical impedance spectroscopy process.

[0046] The apparatus in this embodiment describes a method for electrochemical impedance spectroscopy testing of lithium batteries, and the specific steps include: S1. After the battery data acquisition device is powered on, it is connected to the host computer via a USB Type-C cable. The experimental parameters are set and the experiment is started. The experimental data is transmitted back to the host computer via USB Type-C.

[0047] S2. The host computer is connected to the charging and discharging device. The battery is connected to the charging and discharging device, and the charging and discharging rate is controlled to adjust the battery's state of charge and the number of charging and discharging cycles.

[0048] S3. Repeat steps S1 and S2 to construct a historical impedance spectrum dataset for lithium batteries. Considering the large amount of data required for training the algorithm, at least 100 sets of impedance data should be available.

[0049] S4. The data obtained in step S3 are used to perform batch fitting of the dataset impedance through a pre-written equivalent circuit fitting program. The equivalent circuit structure is custom-built before the input impedance data, and bad data is automatically screened out according to KK residual ≤1%, thereby extracting health characteristics that can reflect the internal electrochemical process of the battery.

[0050] S5 and XGBoost train multiple CART trees sequentially using addition, and use second-order Taylor expansion of the loss and regularization to fit the residuals into weighted least squares. Automatic speed reduction is used to prevent overfitting in areas with large curvature, and the error is gradually reduced to achieve high-precision prediction.

[0051] S6. Extract the equivalent circuit element parameters as battery health status feature values, qualitatively analyze the relationship between different feature values ​​and lithium battery health status, determine the prediction weight of different feature values, and sort the prediction weights of different feature values.

[0052] S7. Perform regression analysis on the health status prediction results, and use the evaluation indicators of the regression analysis to judge the accuracy of the prediction results, thereby achieving the prediction of health status.

[0053] Figure 5This graph displays a comparison between the predicted SOH and the actual SOH in the lithium battery prediction set. The straight line formed by the green dots represents the position where the predicted SOH should appear when the prediction is completely accurate. The closer the predicted SOH represented by the black dots is to this straight line, the better the prediction effect.

[0054] In one embodiment, if the evaluation indicators fail to meet the standards and cannot accurately predict health status, the following steps are performed: A1. Reconstruct several possible equivalent circuits and input them into the fitting software; A2. Randomly select a set of data from the dataset as fitting samples for several equivalent circuits; A3. Compare all fitting results and select the most accurate equivalent circuit; A4. Input the equivalent circuit with the best fitting result into the fitting software, and repeat S4-S7 until the evaluation index of the regression analysis is acceptable.

[0055] In one embodiment, a lithium battery health status prediction system is provided for implementing the method described in the foregoing embodiments, the system comprising the following modules: The data acquisition module is used to apply a frequency-adjustable sinusoidal current excitation to the lithium battery under test and acquire historical impedance spectrum data of the lithium battery. The input feature fitting module is used to input the historical impedance spectrum data of the lithium battery into a preset "improved second-order R-CPE + Warburg-C" equivalent circuit model, and automatically fit to obtain five health indicators including at least ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, Warburg coefficient Y0 and parallel capacitance C. The prediction module takes five health indicators as input features and inputs them into the trained XGBoost regression model to obtain the predicted state of health (SOH) value of the lithium battery under test.

[0056] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the methods described in the foregoing embodiments.

[0057] In one embodiment, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program that, when executed by a processor, implements the methods described in the foregoing embodiments.

[0058] The system, device, and medium described herein possess the technical effects achievable by the methods described in the foregoing embodiments.

[0059] The present invention has been described in detail above with reference to preferred embodiments. However, the present invention is not limited to the described embodiments. Those skilled in the art can make equivalent modifications, alterations or substitutions to these embodiments without departing from the spirit of the present invention, and such modifications, alterations or substitutions all fall within the protection scope defined by the claims of this application.

Claims

1. A method for predicting the health status of a lithium battery, characterized in that, Includes the following steps: A frequency-adjustable sinusoidal current excitation was applied to the lithium battery under test to obtain historical impedance spectrum data of the lithium battery. The historical impedance spectrum data of the lithium battery is input into the preset "improved second-order R-CPE + Warburg-C" equivalent circuit model, and the model is automatically fitted to obtain five health indicators, including at least ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, Warburg coefficient Y0 and parallel capacitance C. Five health indicators are used as input features and fed into a trained XGBoost regression model to obtain the predicted state of health (SOH) value of the lithium battery under test.

2. The method for predicting the health status of a lithium battery according to claim 1, characterized in that, The amplitude of the sinusoidal current excitation satisfies the linear response condition, and the frequency scanning range is 0.1 Hz–10 kHz.

3. The method for predicting the health status of a lithium battery according to claim 1, characterized in that, The equivalent circuit model includes: The series inductance L is used to characterize the high-frequency inductive reactance; The ohmic resistor R0 is used to characterize the electrolyte and contact resistance; The first ZARC unit is formed by connecting the first resistor R1 and the first CPE1 in parallel, and is used to characterize the SEI impedance; The second ZARC unit is formed by connecting the second resistor R2 in parallel with the second CPE2, and is used to characterize the charge transfer impedance. The Warburg unit W is connected in parallel with the third CPE3 and then in series with the second ZARC unit to characterize the low-frequency diffusion process and the additional capacitance effect.

4. The method for predicting the health status of a lithium battery according to claim 1, characterized in that, Automatic fitting was performed in batches using Python scripts, and the fitting results were validated by Kramers-Kronig to remove unqualified data.

5. A method for predicting the health status of a lithium battery according to any one of claims 1-4, characterized in that, The XGBoost regression model is trained in the following way: Obtain n sets of historical impedance spectrum data of lithium batteries with known state of equilibrium (SOH), and calibrate the true value of SOH using the capacity method; A feature vector is constructed using five health indicators, and the SOH true value is used as the label to form a training set; An additive model and forward distribution optimization are used to iteratively construct M CART regression trees while minimizing the objective function.

6. The lithium battery health status prediction method according to claim 5, characterized in that, in, The objective function of the XGBoost regression model includes first-order and second-order Taylor expansions of the loss function and a regularization term to prevent overfitting.

7. The lithium battery health status prediction method according to claim 6, characterized in that, The expression for the objective function is: in, Let be the objective function. It is a decision tree. , These are the first and second derivatives of the loss function with respect to the predicted values, respectively. , The regularization coefficient is... The number of leaf nodes. Leaf weights Indicates the first The set of indices of all samples in a leaf node.

8. A lithium battery health status prediction system, characterized in that, The system for implementing the method according to any one of claims 1-7, the system comprising the following modules: The data acquisition module is used to apply a frequency-adjustable sinusoidal current excitation to the lithium battery under test and acquire historical impedance spectrum data of the lithium battery. The input feature fitting module is used to input the historical impedance spectrum data of the lithium battery into a preset "improved second-order R-CPE + Warburg-C" equivalent circuit model, and automatically fit to obtain five health indicators including at least ohmic internal resistance R0, SEI impedance Z1, charge transfer impedance Z2, Warburg coefficient Y0 and parallel capacitance C. The prediction module takes five health indicators as input features and inputs them into the trained XGBoost regression model to obtain the predicted state of health (SOH) value of the lithium battery under test.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method according to any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-7.

Images