A lithium ion battery soh estimation method based on eis geometric characteristics

Abstract

This invention relates to the field of battery testing technology, specifically disclosing a lithium-ion battery SOH estimation method based on EIS geometric features, comprising the following steps: S1, EIS data acquisition and preprocessing; S2, extracting effective geometric feature statistics from the preprocessed EIS data; S3, inputting the effective geometric feature statistics into a machine learning model for training, and evaluating the trained model; S4, using the trained model to predict the battery health state, and outputting the predicted SOH value of the battery health state. This invention employs the above-mentioned lithium-ion battery SOH estimation method based on EIS geometric features, which significantly improves the SOH prediction efficiency and accuracy by simplifying feature dimensions and eliminating irrelevant information. Furthermore, the model is adaptable to multiple temperature ranges, has high confidence probability, and can meet the battery SOH monitoring needs of scenarios such as new energy vehicles and energy storage systems, demonstrating strong applicability.

Description

Technical Field

[0001] This invention relates to the field of battery testing technology, and in particular to a method for estimating the state of oxygen (SOH) of lithium-ion batteries based on EIS geometric features. Background Technology

[0002] In the field of battery testing technology, State of Health (SOH) is a core indicator for measuring the degree of battery performance degradation. Its value directly determines the range, safety performance and service life of new energy equipment, and is also the underlying parameter support for ensuring the stable operation of key equipment such as energy storage systems and new energy vehicles.

[0003] Electrochemical impedance spectroscopy (EIS) is an effective method for analyzing battery aging mechanisms. As the active material in a battery decreases and its performance declines, the EIS curve also changes accordingly. Therefore, using EIS data to estimate state of harmonics (SOH) is a practical approach. Typically, researchers use EIS data to train machine learning models, such as neural networks or support vector machines, to capture and estimate the changes in battery SOH.

[0004] Traditionally, the EIS impedance data measured in each iteration is used for model training. However, the large volume of complete EIS data significantly impacts training and estimation time. Furthermore, the complex information contained in complete EIS data increases the time and accuracy of training and estimation. Summary of the Invention

[0005] The purpose of this invention is to provide a lithium-ion battery SOH estimation method based on EIS geometric features, which solves the problems of low estimation efficiency and low accuracy in the prior art.

[0006] To achieve the above objectives, this invention provides a method for estimating the state of harm (SOH) of lithium-ion batteries based on EIS geometric features, comprising the following steps:

[0007] S1 and EIS data acquisition and preprocessing;

[0008] S2. Extract effective geometric feature statistics from the preprocessed EIS data;

[0009] S3. Input the effective geometric feature statistics into the machine learning model for training, and evaluate the trained model;

[0010] S4. Use the trained model to predict the battery health status and output the predicted SOH value of the battery health status.

[0011] Preferably, S1 includes the following steps:

[0012] S11. Obtain lithium-ion batteries from the same batch with consistent initial performance and conduct cycle aging tests under the same experimental conditions.

[0013] S12. Perform a capacity test and an EIS test once for each cycle of the experiment to obtain the original capacity data and electrochemical impedance spectroscopy (EIS) data of the battery.

[0014] S13. Perform Kramers-Kronig test on the obtained original battery capacity data and electrochemical impedance spectroscopy (EIS) data to determine whether the EIS data meets the assumptions of linearity, causality, and stability, and remove outlier data.

[0015] Preferably, the true SOH value of the battery health status is obtained by calculating the ratio of the original battery capacity to the initial battery capacity in S12.

[0016] Preferably, S2 includes the following steps:

[0017] S21. Plot the pre-processed electrochemical impedance spectroscopy (EIS) data into an EIS curve. Define a standard line in the EIS curve and calculate the vertical distance from all data points on the EIS curve to the standard line as distance characteristic data.

[0018] S22. Perform statistical analysis on the distance feature data of each loop and extract multiple geometric feature statistics from each loop;

[0019] S23. Perform correlation coefficient analysis on the extracted geometric feature statistics, and select the geometric feature statistics that are correlated with the true SOH value of the battery health state as effective geometric feature statistics for predicting the battery SOH value.

[0020] Preferably, the geometric characteristic statistics in S22 include the mean, standard deviation, maximum value, minimum value, 25th percentile, 50th percentile, 75th percentile, interquartile range, and range.

[0021] Preferably, the mean in the geometric feature statistics is used to reflect the average offset level of the distance feature data;

[0022] Standard deviation is used to describe the dispersion of distance feature data;

[0023] The maximum and minimum values ​​are the maximum and minimum values ​​in the distance feature data, respectively.

[0024] The 25th percentile is the position of the data point after all distance feature data are sorted from smallest to largest. The value at the location;

[0025] The 50th percentile is the value located in the middle position after all distance feature data are sorted from smallest to largest, and is used to reflect the medium level of the data.

[0026] The 75th percentile is the position of the data point after all distance feature data are sorted from smallest to largest. The value at the location;

[0027] The interquartile range (IMR) is the difference between the 75th percentile and the 50th percentile, reflecting the distance between the median and the characteristic data. The degree of dispersion of numerical values ​​is used to identify outliers in distance feature data;

[0028] The range is the difference between the maximum and minimum values ​​of the feature data, reflecting the overall fluctuation range of the data.

[0029] Preferably, S3 specifically involves: inputting the extracted effective geometric feature statistics into the machine learning model for training, and evaluating the trained machine learning model, with evaluation metrics including mean absolute error. Maximum absolute error Root mean square error Mean square error Confidence probability and coefficient of determination .

[0030] Preferably, the machine learning model uses the PSO-GPR fusion prediction model. The PSO model iteratively updates the velocity and position of particles by tracking their own historical best position and the historical best position of the entire swarm, ultimately causing the swarm to converge to the optimal solution. The velocity update formula is as follows:

[0031] ;

[0032] in, For the first Individual particles The speed of time, Inertial weights; and The cognitive coefficient is used to control the particle's learning of its own historical best and the group's best, respectively. and A random number within the range [0,1]; This is the particle's optimal position. The globally optimal position;

[0033] The displacement update formula is as follows:

[0034] ;

[0035] in, For the first Individual particles Location at any given moment For the first An object in The speed of time;

[0036] The GPR model is a nonparametric model that uses a Gaussian process prior to perform regression analysis on the data. It is used to capture complex nonlinear relationships and provide uncertainty estimates, including the joint Gaussian distribution, posterior distribution, and covariance function. The formula for the joint Gaussian distribution is as follows:

[0037] ;

[0038] in, For the observed values, For predicted values, for n A covariance matrix of order-4 symmetric positive definiteness. = For test points With training set The covariance matrix between the inputs, For noise variance;

[0039] The posterior distribution formula is as follows:

[0040] ;

[0041] in, For the mean difference, For variance, ;

[0042] Covariance function:

[0043] ;

[0044] in, For signal variance, It is a length scale.

[0045] Therefore, the present invention employs the above-mentioned method for estimating the SOH of lithium-ion batteries based on EIS geometric features, and the beneficial effects are as follows:

[0046] (1) The present invention first processes the EIS data to extract geometric features for model training. The number of extracted geometric features is much smaller than the original EIS data. At the same time, by conducting statistical analysis on the geometric features, information unrelated to SOH estimation is removed. This reduces the interference of irrelevant information on the estimation results and significantly improves the efficiency and accuracy of SOH estimation.

[0047] (2) The PSO-GPR fusion prediction model obtained by training based on geometric features in this invention has a wide applicable temperature range. Within the specified temperature range, the model can effectively estimate the state of harm (SOH) of lithium-ion batteries, and the model estimation results are satisfactory within this temperature range. , It meets the requirements for battery SOH estimation under different temperature environments and has strong applicability.

[0048] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0049] Figure 1 This is an overall flowchart of an embodiment of the lithium-ion battery SOH estimation method based on EIS geometric features according to the present invention;

[0050] Figure 2 This is a schematic diagram illustrating the acquisition of the original geometric features of the EIS in an embodiment of the lithium-ion battery SOH estimation method based on EIS geometric features according to the present invention.

[0051] Figure 3 This is a PSO-GPR fusion prediction model for SOH estimation in lithium-ion batteries, based on EIS geometric features, according to an embodiment of the present invention.

[0052] Figure 4 This is an embodiment of a lithium-ion battery SOH estimation method based on EIS geometric features according to the present invention. A comparison of impedance characteristic estimation and geometric characteristic estimation, where (a) is... The following is a graph showing the results of impedance characteristic estimation: (b) The following is a graph showing the results of geometric feature estimation;

[0053] Figure 5 This is an embodiment of a lithium-ion battery SOH estimation method based on EIS geometric features according to the present invention. A comparison of impedance characteristic estimation and geometric characteristic estimation, where (a) is... The following is a graph showing the results of impedance characteristic estimation: (b) The following is a graph showing the results of geometric feature estimation. Detailed Implementation

[0054] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0055] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0056] like Figure 1 As shown, a method for estimating the state of harm (SOH) of a lithium-ion battery based on EIS geometric features includes the following steps:

[0057] S1, EIS data acquisition and preprocessing includes the following steps:

[0058] S11. Obtain lithium-ion batteries from the same batch with consistent initial performance and conduct cycle aging tests under the same experimental conditions.

[0059] S12. Perform a capacity test and an EIS test once for each cycle of the experiment to obtain the original capacity data and electrochemical impedance spectroscopy (EIS) data of the battery; by calculating the ratio of the original capacity of the battery to the initial capacity of the battery, the true SOH value of the battery health state is obtained.

[0060] S13. Perform Kramers-Kronig test on the obtained original battery capacity data and electrochemical impedance spectroscopy (EIS) data to determine whether the EIS data meets the assumptions of linearity, causality, and stability, and remove outlier data to ensure the reliability of subsequent analyses.

[0061] S2, such as Figure 2 As shown, extracting effective geometric feature statistics from preprocessed EIS data includes the following steps:

[0062] S21. Plot the preprocessed electrochemical impedance spectroscopy (EIS) data into an EIS curve. Define a standard line in the EIS curve and calculate the vertical distance from all data points on the EIS curve to the standard line as distance characteristic data.

[0063] S22. Perform statistical analysis on the distance feature data of each cycle, and extract multiple geometric feature statistics from each cycle. The geometric feature statistics include nine statistics: mean, standard deviation, maximum value, minimum value, 25th percentile, 50th percentile, 75th percentile, interquartile range, and range.

[0064] ① The mean in the geometric feature statistics of this invention is the average value of all data, used to reflect the average offset level of the distance feature data. The specific calculation formula is as follows:

[0065] ;

[0066] in, X i The vertical distance from the data point on the EIS curve to the standard line. n This represents the total number of data points on the EIS curve. i This represents the current data point.

[0067] ②Standard deviation (std) is used to describe the dispersion of distance feature data. The larger the value, the more dispersed the data. The specific formula is:

[0068] .

[0069] ③ The maximum value (max) is the maximum value in the distance feature data, and the specific formula is:

[0070] .

[0071] ④ The minimum value (min) is the minimum value in the distance feature data, and the specific formula is as follows:

[0072] .

[0073] ⑤ 25th percentile (25p): After sorting all distance feature data from smallest to largest, we have The data is less than or equal to this value (lower quartile). The calculation logic for the 25th percentile is that, in the sorted distance feature data, the data located at... The value at the location.

[0074] ⑥ 50th percentile / median (50p): The value located in the middle after sorting all distance feature data from smallest to largest, used to reflect the median level of the data; the calculation logic is: if If it is an odd number, its value is the th order after sorting. One value; if If it is an even number, the value is the first (or second) after sorting. and The average of the values.

[0075] ⑦ 75th percentile (75p): After sorting all distance feature data from smallest to largest, we have... The data is less than or equal to this value (upper quartile). The 75th percentile is calculated based on the distance feature data in the sorted data. The value at the location.

[0076] ⑧ Interquartile range ( IQR The difference between the 75th percentile and the 50th percentile reflects the distance from the median of the characteristic data. The degree of dispersion of numerical values ​​can be used to identify outliers in distance feature data. The specific formula is as follows:

[0077] .

[0078] in, It is the 75th percentile. It is the 25th percentile.

[0079] ⑨ Range ( Range The distance is the difference between the maximum and minimum values ​​of the feature data, reflecting the overall fluctuation range of the data. The specific formula is:

[0080] .

[0081] S23. Perform correlation coefficient analysis on the extracted geometric feature statistics to verify that the information contained therein is related to SOH, and select the geometric feature statistics that are correlated with the true SOH value of the battery health state as effective geometric feature statistics for predicting the battery SOH value.

[0082] S3. The extracted effective geometric feature statistics are input into the machine learning model for training. The machine learning model used in this invention is a PSO-GPR fusion prediction model, that is, a fusion model of particle swarm optimization (PSO) and Gaussian process regression (GPR). The trained machine learning model is evaluated to verify the effectiveness of this training. The evaluation index includes mean absolute error. Maximum absolute error Root mean square error Mean square error Confidence probability and coefficient of determination The model training is effective when all evaluation metrics reach the preset performance threshold within a training cycle.

[0083] like Figure 3 As shown, in this invention, the PSO model treats each potential solution to the problem as a "particle" in the search space. Each particle has two attributes: velocity and position. The algorithm iteratively updates the velocity and position of each particle by tracking its own historical best position and the historical best position of the entire swarm, ultimately causing the swarm to converge to a global or local optimum. The velocity update formula is as follows:

[0084] ;

[0085] in, For the first Individual particles The speed of time, Inertial weights; and The cognitive coefficient is used to control the particle's learning of its own historical best and the group's best, respectively. and A random number within the range [0,1]; This is the particle's optimal position. This is the globally optimal position.

[0086] The displacement update formula is as follows:

[0087] ;

[0088] in, For the first Individual particles Location at any given moment For the first An object in The speed of time.

[0089] The Gaussian process regression (GPR) model is a nonparametric model that uses a Gaussian process prior to perform regression analysis on data. It is used to solve regression problems, capture complex nonlinear relationships, and provide uncertainty estimation. It mainly consists of three core components: the joint Gaussian distribution, the posterior distribution, and the covariance function. The formula for the joint Gaussian distribution is as follows:

[0090] ;

[0091] in, For the observed values, For predicted values, for n A covariance matrix of order-4 symmetric positive definiteness. = For test points With training set The covariance matrix between the inputs, This represents the noise variance.

[0092] The posterior distribution formula is as follows:

[0093] ;

[0094] in, For the mean difference, For variance, .

[0095] Covariance function:

[0096] ;

[0097] in, For signal variance, It is a length scale.

[0098] The training process for the PSO and GPR fusion model is as follows:

[0099] Step 1: Data preprocessing and parameter initialization. Normalize the data and remove outlier data points. Divide the data into training and testing parts.

[0100] Step 2: PSO particle swarm initialization, setting the initial position and velocity of the particle swarm, and setting the fitness function.

[0101] Step 3: Train the GPR model using the hyperparameters of the current particle, and fit the Gaussian process model using the training data.

[0102] Step 4: Use the trained GPR model to predict the training data, calculate the error between the predicted value and the true value, and use it as the fitness value.

[0103] Step 5: Record the historical best position of each particle, and select the particle with the best fitness among all particles as the global optimum.

[0104] Step 6: Update the particle's velocity and position using the PSO algorithm to find a new combination of GPR hyperparameters. If the condition is not met, proceed to step 3 to continue iterating. If the termination condition is met, proceed to the next step.

[0105] Step 7: Output the GPR hyperparameter combination corresponding to the globally optimal particle, retrain the GPR model, and construct the final PSO-GPR fusion prediction model.

[0106] S4. Input the test data into the trained PSO-GPR fusion prediction model to predict the battery health state and output the predicted SOH value of the battery health state.

[0107] Example 1

[0108] To verify that the geometric features proposed in this invention can effectively estimate the state of harmonics (SOH) of lithium-ion batteries, four coin cells were used for a cycle test. The battery was charged and discharged, and a capacity test and an EIS test were performed after each cycle. The obtained data were estimated and analyzed, and the specific estimation results are shown in Table 1:

[0109] Table 1 Comparison of lower impedance characteristic estimation and geometric characteristic estimation results

[0110]

[0111] As shown in Table 1, the estimation results using impedance features have larger errors and require longer training times, while the results using geometric features show significant improvements over impedance features in all aspects. It decreased by 75.63%. This represents an improvement of 116.48%, while training time has been reduced by 20.6037 seconds. From Figure 4 It can also be seen intuitively that the results of geometric features are far better than those of impedance features.

[0112] To investigate the performance of this invention under high-temperature conditions, two button batteries were used for a cycle test. The battery was charged and discharged, and a capacity test and an EIS test were performed after each cycle. The obtained data were estimated and analyzed, and the specific estimation results are shown in Table 2.

[0113] Table 2 Comparison of lower impedance characteristic estimation and geometric characteristic estimation results

[0114]

[0115] As can be seen from the table, under high-temperature conditions, the estimation results using impedance features have larger errors and longer training times, while the results estimated using geometric features show significant improvements over impedance features in all aspects. It decreased by 78.64%. This represents a 12.36% improvement, while training time has been reduced by 2.3995 seconds. From Figure 5 It can also be seen intuitively that the results of the distance characteristic are far better than those of the impedance characteristic.

[0116] Therefore, the present invention adopts the above-mentioned lithium-ion battery SOH estimation method based on EIS geometric features, which can effectively solve the problems of low efficiency and poor accuracy of traditional lithium-ion battery SOH estimation. By simplifying features and reducing interference, it significantly improves prediction efficiency and accuracy, meets the needs of different scenarios, provides a reliable solution for battery health monitoring of new energy equipment, and has high practical value.

[0117] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A lithium-ion battery SOH estimation method based on EIS geometric characteristics, characterized in that, Includes the following steps: S1 and EIS data acquisition and preprocessing; S2. Extract effective geometric feature statistics from the preprocessed EIS data, including the following steps: S21. Plot the pre-processed electrochemical impedance spectroscopy (EIS) data into an EIS curve. Define a standard line in the EIS curve and calculate the vertical distance from all data points on the EIS curve to the standard line as distance characteristic data. S22. Perform statistical analysis on the distance feature data of each cycle and extract multiple geometric feature statistics from each cycle; the geometric feature statistics include mean, standard deviation, maximum value, minimum value, 25th percentile, 50th percentile, 75th percentile, interquartile range, and range; S23. Perform correlation coefficient analysis on the extracted geometric feature statistics, and select the geometric feature statistics that are correlated with the true SOH value of the battery health state as effective geometric feature statistics for predicting the battery SOH value. S3. Input the effective geometric feature statistics into the machine learning model for training, and evaluate the trained model. The machine learning model uses the PSO-GPR fusion prediction model. The PSO model updates the velocity and position by iterating the particles to track their own historical best position and the historical best position of the entire group, and finally makes the group converge to the optimal solution. S4. Use the trained model to predict the battery health status and output the predicted SOH value of the battery health status.

2. The lithium-ion battery SOH estimation method based on EIS geometric characteristics of claim 1, wherein, S1 includes the following steps: S11. Obtain lithium-ion batteries from the same batch with consistent initial performance and conduct cycle aging tests under the same experimental conditions. S12. Perform a capacity test and an EIS test once for each cycle of the experiment to obtain the original capacity data and electrochemical impedance spectroscopy (EIS) data of the battery. S13. Perform Kramers-Kronig test on the obtained original battery capacity data and electrochemical impedance spectroscopy (EIS) data to determine whether the EIS data meets the assumptions of linearity, causality, and stability, and remove outlier data.

3. The lithium-ion battery SOH estimation method based on EIS geometric characteristics of claim 2, characterized in that, The true SOH value of the battery's health status is obtained by calculating the ratio of the battery's original capacity to its initial capacity in S12.

4. The lithium-ion battery SOH estimation method based on EIS geometric characteristics of claim 3, characterized in that, The mean in geometric feature statistics is used to reflect the average offset level of distance feature data; Standard deviation is used to describe the dispersion of distance feature data; The maximum and minimum values ​​are the maximum and minimum values ​​in the distance feature data, respectively. The 25th percentile is the position of the data point after all distance feature data are sorted from smallest to largest. The value at the location; The 50th percentile is the value located in the middle position after all distance feature data are sorted from smallest to largest, and is used to reflect the medium level of the data. The 75th percentile is the value at the position of the sorted distance feature data from small to large position. The interquartile range is the difference between the 75th percentile and the 50th percentile, reflecting the dispersion of the numerical characteristic data in the middle of the data, for identifying outliers in the numerical characteristic data; The range is the difference between the maximum and minimum values ​​of the feature data, reflecting the overall fluctuation range of the data.

5. The lithium-ion battery SOH estimation method based on EIS geometric characteristics of claim 1, wherein, S3 specifically involves: inputting the extracted effective geometric feature statistics into the machine learning model for training, and evaluating the trained machine learning model using metrics including mean absolute error. Maximum absolute error Root mean square error Mean square error Confidence probability and coefficient of determination .

6. The lithium-ion battery SOH estimation method based on EIS geometric characteristics of claim 5, wherein, The speed update formula is as follows: ； in, For the first Individual particles The speed of time Inertial weights; and The cognitive coefficient is used to control the particle's learning of its own historical best and the group's best, respectively. and A random number within the range [0,1]; This is the particle's optimal position. The globally optimal position; The displacement update formula is as follows: ； in, For the first Individual particles Location at any given moment For the first An object in The speed of time; The GPR model is a nonparametric model that uses a Gaussian process prior to perform regression analysis on the data. It is used to capture complex nonlinear relationships and provide uncertainty estimates, including the joint Gaussian distribution, posterior distribution, and covariance function. The formula for the joint Gaussian distribution is as follows: ； in, For the observed values, For predicted values, for A symmetric positive definite covariance matrix of order 1. = For test points With training set The covariance matrix between the inputs, For noise variance; The posterior distribution formula is as follows: ； wherein is the mean difference, is the variance, ; Covariance function: ； wherein, is the signal variance, is the length scale.

Images