Lithium ion battery health state diagnosis method based on equivalent circuit feature screening

Abstract

The invention belongs to the field of energy storage battery health state estimation, and particularly relates to a method for estimating the health state of an energy storage battery by using a machine learning algorithm and a data model based on equivalent circuit feature screening. According to the method, equivalent circuit model features and data driving features are combined, redundant information is reduced through feature screening, SOH estimation precision and calculation efficiency are improved, the generalization ability of the model is enhanced, and the model can adapt to different working conditions and environment changes. Compared with a pure data driving method, the method introduces ECM features, so that the prediction result is more suitable for the physical degradation mechanism of the battery, and the interpretability of the model is improved. In addition, optimization of feature screening reduces calculation complexity, so that SOH estimation is more efficient, and real-time or online prediction requirements can be met. By fusing the advantages of data driving and physical modeling, the method has both reliability and accuracy, and a better solution is provided for lithium ion battery health management.

Description

Technical Field

[0001] The present invention belongs to the field of state of health estimation of energy storage batteries, and more specifically, relates to a method for estimating the state of health of energy storage batteries based on equivalent circuit feature screening and using machine learning algorithms and data models. Background Art

[0002] With the significant increase in the popularity of lithium-ion batteries (LIBs) in contemporary society, the safety risks and range anxiety problems caused by them have become increasingly prominent. The performance degradation and aging process of LIBs easily lead to system failures, posing potential risks of property damage and personal injury. The phenomenon of range anxiety mostly stems from the mismatch between the mileage displayed on the dashboard and the actual remaining capacity of the battery. Constructing an accurate battery state of health diagnosis system has become the key path to solving the above technical bottlenecks. The field of battery state of health prediction (SOH) has received extensive attention. Currently, the mainstream SOH scalar characterization methods are mainly constructed based on historical cycle discharge capacity or internal resistance parameters. Specifically, in application scenarios centered on energy supply such as electric vehicles, the capacity parameter is often used as the SOH indicator factor; while the internal resistance parameter is mostly used in application scenarios such as hybrid electric vehicles with the core requirement of power output. To achieve the above goals, three types of solutions based on model-driven, data-driven, and hybrid methods have been proposed in the prior art. However, currently, the SOH estimation of lithium-ion batteries faces problems such as limited prediction accuracy, high computational complexity, insufficient generalization ability, and lack of physical interpretability of data-driven methods. The existence of redundant features may affect the model performance, resulting in low computational efficiency. Although pure data-driven methods can provide certain prediction capabilities, it is difficult to reveal the physical degradation mechanism of the battery state of health. Summary of the Invention

[0003] To address these problems, this method combines equivalent circuit model (ECM) features and data-driven features, reduces redundant information through feature screening, improves the accuracy and computational efficiency of SOH estimation, and at the same time enhances the generalization ability of the model, enabling it to adapt to different working conditions and environmental changes. Compared with traditional black-box models, this method introduces physical modeling, making the prediction results more consistent with the actual degradation process of the battery, enhancing interpretability, and reducing computational complexity by optimizing feature input, enabling it to meet real-time or online prediction requirements. By integrating the advantages of data-driven and physical modeling, this method has been optimized in terms of reliability, accuracy, and applicability, providing an efficient and feasible solution for lithium-ion battery health management.

[0004] A method for diagnosing the state of health of lithium-ion batteries based on equivalent circuit feature screening disclosed by the present invention includes the following steps:

[0005] Step 1: Select a battery data set with high data quality, sufficient samples, and strong representativeness;

[0006] Step 2: Preprocess the data and extract features;

[0007] Step 3: Input the features into a machine learning model to conduct the first assessment of the battery health state;

[0008] Step 4: Establish an equivalent circuit model and extract relevant parameters, couple the parameters with the features of the incremental capacity curve (ICA), and further screen the features;

[0009] Step 5: Input the screened feature pairs to conduct a second assessment of the battery health state.

[0010] In the above technical solution, Step 1 specifically includes:

[0011] Among them, the battery dataset adopted in the present invention is from the Zenodo database and includes charge-discharge data and EIS (electrochemical impedance spectroscopy) data in the frequency domain. The research selects the graphite / NCA 18650 battery (Dataset_1) as the object, whose nominal capacity is 3.5 Ah, rated voltage is 3.6 V, charging cut-off voltage is 4.2 V, and discharging cut-off voltage is 2.65 V.

[0012] In the above technical solution, Step 2 specifically includes:

[0013] Step 2.1: The target value of the present invention is the battery health state data, which can be expressed as:

[0014]

[0015] Q n is the capacity when the battery is fully charged currently (state of charge SOC = 100), and Q 0 is the rated capacity of the battery;

[0016] Step 2.2: The Z-score method is mainly used to screen out outliers for the treatment of outliers, and the data beyond the threshold is replaced by the average value before and after;

[0017] Step 2.3: Extract the features in the charging data, including: constant current charging time, constant voltage charging time, initial voltage, average voltage, constant current charging time / constant voltage charging time, constant current charging time / total charging time, and thus 6 different features are obtained;

[0018] Step 2.4: Process the charging data to obtain the incremental capacity curve;

[0019] Step 2.5: If the peak disappears in the curve, then by finding the point with the minimum absolute value of the derivative in the voltage window, it is used as the value of the peak;

[0020] Step 2.6: Select the peak value and peak voltage of each peak of the incremental capacity curve after dealing with the disappearance of peaks (there are three peaks in total in the curve) as the other 6 features for inputting into the machine learning model.

[0021] In the above technical solution, step 3 specifically includes:

[0022] Step 3.1: Input the above-extracted 12 features into the machine learning model, where the machine learning model includes: LightGBM model, XGBoost model, GPR model, and Ridge model;

[0023] Step 3.2: Under different temperature conditions, set different training and testing paths, and adopt the cross-validation strategy: at 25°C, use the four batteries 25C03, 25C06, 25C11, and 25C19. Each battery is used as the test set, and the remaining batteries are used as the training set; and so on. At 35°C and 45°C, follow the above paths, only with different batteries. The batteries used at 35°C are 35C01 and 35C02; the batteries used at 45°C are 45C13, 45C16, 45C21, and 45C27 respectively; this cross-validation framework effectively ensures the generalization ability of the model for batteries of the same model but different individuals; the input is the above-obtained 12 features, and the output is the battery health state;

[0024] Step 3.3: Use common evaluation metrics in machine learning, including: mean squared error (MSE), root mean squared error (RMSE), and coefficient of determination (R²).

[0025] In the above technical solution, step 4 specifically includes:

[0026] Step 4.1: Extract the impedance data of the battery: including a total of 100 data points within the frequency range of 50 mHz to 10 kHz, and extraction is performed every 25 cycles (impedance test period);

[0027] Step 4.2: According to the obtained impedance data, use the module of the python library impedance.py to establish and fit the equivalent circuit model, and the equivalent circuit model is modeled according to Figure 3 for modeling;

[0028] Step 4.3: Extract several parameters in the equivalent circuit model that are most relevant to battery decay: R sei , C sei , C ct , R ct , Z w ;

[0029] Step 4.4: Perform Pearson correlation analysis on the 12 features extracted in Step 2 and these 5 parameters. Sum and average the Pearson coefficients of each feature with respect to the parameters and sort them. Remove the features ranked at the back or with an average value less than 0.6 to obtain the filtered features.

[0030] In the above technical solution, Step 5 specifically includes:

[0031] Step 5.1: Adopt the cross - validation strategy as in the solution of Step 3 above. The input is the filtered features, and the output is the battery health state.

[0032] Step 5.2: Evaluate the results using the same machine - learning evaluation metrics as in Step 3 above.

[0033] Since the present invention adopts the above technical means, it has the following beneficial effects:

[0034] This method combines the equivalent - circuit - model features and data - driven features, uses feature screening to reduce redundant information, improves the accuracy and computational efficiency of SOH estimation, and simultaneously enhances the generalization ability of the model, enabling it to adapt to different working conditions and environmental changes. Compared with the pure data - driven method, this method introduces ECM features, making the prediction results more in line with the physical degradation mechanism of the battery and improving the interpretability of the model. In addition, the optimization of feature screening reduces the computational complexity, making the SOH estimation more efficient and meeting the requirements of real - time or online prediction. By integrating the advantages of data - driven and physical modeling, this method combines reliability and accuracy, providing a better solution for the health management of lithium - ion batteries. Brief Description of the Drawings

[0035] Figure 1 It is a flowchart of the implementation steps of the estimation method provided by the embodiment of the present invention.

[0036] Figure 2 It is a battery attenuation curve graph of the data set provided by the embodiment of the present invention.

[0037] Figure 3 It is a schematic diagram of establishing an equivalent - circuit model provided by the embodiment of the present invention.

[0038] Figure 4 It is a comparison graph of the SOH estimated value and the actual value of the 25C11 battery under the condition of 25°C provided by the embodiment of the present invention. Detailed Embodiment

[0039] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. Components of the embodiments of the present invention generally described and illustrated in the figures herein can be arranged and designed in a variety of different configurations.

[0040] Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed present invention, but merely represents selected embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0041] It should be noted that like reference numerals and letters denote like items in the following drawings. Therefore, once an item is defined in one drawing, it does not require further definition and explanation in subsequent drawings. It should be noted that, without conflict, the features in the embodiments of the present invention can be combined with each other.

[0042] The present invention discloses a method for diagnosing the state of health of a lithium-ion battery based on equivalent circuit feature screening (hereinafter referred to as: estimation method). The estimation method includes the following steps:

[0043] Step 1: Select a battery data set with high data quality, sufficient samples, and strong representativeness;

[0044] Step 2: Preprocess the data and extract features;

[0045] Step 3: Input the features into a machine learning model to conduct a first evaluation of the state of health of the battery;

[0046] Step 4: Establish an equivalent circuit model and extract relevant parameters, couple the parameters with ICA features for further feature screening;

[0047] Step 5: Input the screened feature pairs to conduct a second evaluation of the state of health of the battery.

[0048] Embodiment 1

[0049] A method for diagnosing the state of health of a lithium-ion battery based on equivalent circuit feature screening includes the following steps:

[0050] Step 1: Select a battery data set with high data quality, sufficient samples, and strong representativeness.

[0051] In the above technical solution, Step 1 specifically includes:

[0052] Among them, the battery dataset used in the present invention is from the Zenodo database, including charge-discharge data and EIS (electrochemical impedance spectroscopy) data in the frequency domain. The graphite / NCA 18650 battery (Dataset_1) is selected as the research object, with a nominal capacity of 3.5 Ah, a rated voltage of 3.6 V, a charging cut-off voltage of 4.2 V, and a discharging cut-off voltage of 2.65 V. The charge-discharge protocol of this battery covers three rates of 0.5C, 1C, and 2C, and the cycling temperature is set in a wide temperature range from -20°C to 25°C. The specific test scheme includes capacity calibration cycles, dynamic working condition cycles, and EIS tests at different cycle intervals. The charging protocol of the lithium battery adopts the standard constant current-constant voltage (CC-CV) strategy. The charging process is divided into three parts: (I) constant current charging stage, (II) constant voltage charging stage, (III) post-charging rest stage. The specific process is as follows: First, the battery is charged at a constant current from 0.25C (0.875 A) to 1C (3.5 A) to 4.2 V, and then switched to constant voltage charging at 4.2 V until the current drops to 0.05C (0.175 A), with a sampling interval of 10 s. After charging, it is left to stand for 30 minutes, with a sampling interval of 2 minutes. The discharging protocol adopts a 1C constant current discharging mode, and the discharging is terminated when the voltage drops to 2.65 V. The cycling temperature (25°C, 35°C, 45°C) is controlled by a high-precision thermostat (±0.2°C). For impedance data, the test frequency range of the electrochemical impedance spectroscopy (EIS) of the lithium-ion battery is 50 mHz to 10 kHz (20 data points are collected per decade), and the amplitude of the excitation signal potential is 20 mV. The EIS test is carried out under steady-state conditions after standing for 30 minutes after full charge and is performed at intervals of every 25 charge-discharge cycles, as Figure 2 shown, the capacities of the 10 groups of commercial lithium-ion batteries used in this study all show a significant attenuation trend with the increase in the number of cycles.

[0053] Step 2. Preprocess the data and extract features.

[0054] In the above technical solution, step 2 specifically includes:

[0055] Step 2.1. The target value of the present invention is the state of health data of the battery, which can be expressed as:

[0056]

[0057] Qn is the capacity of the battery when it is currently fully charged (state of charge SOC = 100), and Q0 is the rated capacity of the battery;

[0058] Step 2.2. The main method for dealing with outliers is to use the Z-score method to screen out outliers, and the data exceeding the threshold is replaced by the average value before and after.

[0059] Step 2.3, extract features from the charging data, including: constant current charging time, constant voltage charging time, initial voltage, average voltage, constant current charging time / constant voltage charging time, constant current charging time / total charging time, thus obtaining 6 different features;

[0060] Step 2.4, process the charging data to obtain the incremental capacity curve; the incremental capacity analysis method converts the voltage platform of the traditional constant current charging into easily observable peaks and valleys on the incremental capacity curve to correspond to the aging state and aging mechanism inside the battery. The specific calculation method is as follows: The incremental capacity is equal to the ratio of the difference between the charge and the voltage at the previous and next moments:

[0061]

[0062] Among them, Q k Q is the amount of electricity charged at the kth moment. k-1 The amount of electricity charged at the k-1th moment. V k is the voltage value at the kth moment, V k-1 is the voltage value at the k-1th moment;

[0063] Step 2.5, if the peak disappears in the curve, find the point with the smallest absolute value of the derivative in the voltage window and take it as the peak value;

[0064] Step 2.6. Select the peak value and peak voltage of each peak (the curve has three peaks in total) of the incremental capacity curve after the peak disappearance is processed as the other 6 features of the input machine learning model.

[0065] Step 3: Input the features into the machine learning model to perform the first assessment of the battery health status.

[0066] In the above technical solution, step 3 specifically includes:

[0067] Step 3.1: Input the 12 features extracted above into the machine learning model. The machine learning models include: LightGBM model, XGBoost model, GPR model and Ridge model. The principles of these models are as follows:

[0068] Ridge Model:

[0069] Ridge regression is a regularization method for linear regression, which aims to solve the problem that ordinary least squares (OLS) is prone to overfitting in feature multicollinearity or high-dimensional data. Its core principle is to impose L2 regularization constraints on the coefficients of the model, thereby reducing the complexity of the model and improving generalization ability.

[0070] XGBoost (Extreme gradient boosting) model:

[0071] XGBoost is an optimized implementation based on Gradient Boosting Decision Trees (GBDT). It adopts the Boosting ensemble learning framework and uses the weighted combination of multiple decision trees to improve the prediction ability of the model. The basic idea is to train a tree, calculate the residuals (i.e., the errors between the predicted values and the true values); train the next tree to fit the residuals of the previous tree; iterate sequentially until the preset number of trees or error convergence is reached.

[0072] LightGBM (Light gradient boosting machine) model:

[0073] LightGBM is also a variant of GBDT, designed specifically to address the limitations of GBDT when dealing with massive data, effectively improving its practical application value and operation efficiency. The decision tree model of LightGBM adopts a splitting method based on the leaf growth strategy. The computational complexity of this strategy is relatively low, but overfitting needs to be suppressed by restricting the tree depth and the minimum sample size of leaf nodes. In addition, LightGBM adopts a histogram-based decision tree algorithm, which discretizes the feature values into several "buckets", and searches for split points within these "buckets", significantly reducing the storage requirements and computational consumption. The parallelization framework of this algorithm includes three optimization strategies: feature parallelism when the feature dimension is high, data parallelism when the data scale is large, and voting parallelism when both the feature dimension and the number of votes are high.

[0074] Step 3.2: Under different temperature conditions, set different training and testing paths and adopt the cross-validation strategy. At 25°C, four batteries, namely 25C03, 25C06, 25C11, and 25C19, will be used. Each battery serves as the test set, and the remaining batteries serve as the training set; and so on. At 35°C and 45°C, the same paths are followed, but the batteries are different. The batteries used at 35°C are 35C01 and 35C02; the batteries used at 45°C are 45C13, 45C16, 45C21, and 45C27 respectively. This cross-validation framework effectively ensures the generalization ability of the model for different individual batteries of the same model. The input is the 12 features obtained above, and the output is the battery health status.

[0075] Step 3.3: Use common evaluation metrics in machine learning, including: Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Coefficient of Determination (R²).

[0076] Step 4: Establish an equivalent circuit model and extract relevant parameters, couple the parameters with ICA features, and further perform feature screening;

[0077] In the above technical solution, Step 4 specifically includes:

[0078] Step 4.1: Extract the impedance data of the battery, including 100 data points within the frequency range of 50 mHz to 10 kHz, and extract once every 25 cycles (impedance test period).

[0079] Step 4.2: Based on the obtained impedance data, establish and fit an equivalent circuit model through the python impedance.py module. The equivalent circuit model is modeled according to Figure 3 for modeling.

[0080] Step 4.3: Extract several parameters in the equivalent circuit model that are most relevant to battery degradation: R sei , C sei , C ct , R ct , Z w , R sei / / C sei and R ct / / C ct represent the loss of lithium inventory (LAM); the impedance Z w is used to describe the loss of active material (LAM).

[0081] Step 4.4: Conduct a Pearson correlation analysis on the 12 features extracted in Step 2 and these 5 parameters. Sum and average the Pearson coefficients of each feature for the parameters and sort them. Remove the features with lower rankings or an average value less than 0.6 to obtain the filtered features.

[0082] Step 5: Input the filtered feature pairs to re-evaluate the state of health of the battery.

[0083] In the above technical solution, Step 5 specifically includes:

[0084] Step 5.1: Adopt the cross-validation strategy as in the solution in Step 3 above. The input is the filtered features, and the output is the state of health of the battery.

[0085] Step 5.2: Evaluate the results using the same machine learning evaluation metrics as in Step 3 above.

[0086] The present invention introduces physical modeling, making the prediction results more conform to the actual degradation process of the battery, improving interpretability, and reducing the computational complexity by optimizing the feature input, enabling it to meet real-time or online prediction requirements. By integrating the advantages of data-driven and physical modeling, this method optimizes in terms of reliability, accuracy, and applicability, providing an efficient and feasible solution for the health management of lithium-ion batteries.

[0087] It should be noted and understood that various modifications and improvements can be made to the present invention described in detail above without departing from the spirit and scope of the present invention as claimed in the appended claims. Therefore, the scope of the claimed technical solution is not limited by any specific exemplary teachings given.

[0088] The applicant declares that the present invention illustrates the detailed method of the present invention through the above embodiments, but the present invention is not limited to the above detailed method, that is, it does not mean that the present invention must rely on the above detailed method to be implemented. Those skilled in the art should understand that any improvement to the present invention, the equivalent replacement of each raw material of the product of the present invention, the addition of auxiliary components, the selection of specific methods, etc., all fall within the protection scope and the disclosure scope of the present invention.

Claims

1. A lithium-ion battery health status diagnosis method based on equivalent circuit feature screening, characterized in that: The following steps are involved: Step 1: Select a battery data set with high data quality, sufficient samples, and strong representativeness; Step 2: Preprocess the data and extract features; Step 3: Input the features into the machine learning model to make a first assessment of the battery health status; Step 4: Establish an equivalent circuit model and extract relevant parameters, couple the parameters with ICA features, and further perform feature screening; Step 5: Input the filtered feature pairs and re-evaluate the battery health status.

2. A lithium-ion battery health status diagnosis method based on equivalent circuit feature screening according to claim 1, characterized in that: Step 1 Specific include: The battery data set used in this paper comes from the Zenodo database, which contains charge and discharge data and EIS (electrochemical impedance spectroscopy) data in the frequency domain. The study selected graphite / NCA 18650 battery (Dataset_1) as the object, with a nominal capacity of 3.5 Ah, a rated voltage of 3.6 V, a charge cut-off voltage of 4.2 V, and a discharge cut-off voltage of 2.65 V.

3. A lithium-ion battery health status diagnosis method based on equivalent circuit feature screening according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1: The target value of the present invention is the health status data of the battery, which can be expressed as: 4.Q n is the capacity of the battery when it is currently fully charged (state of charge SOC = 100), and Q0 is the rated capacity of the battery; Step 2.2: The Z-score method is mainly used to filter out outliers, and the data exceeding the threshold is replaced by the average value before and after; Step 2.3, extract features from the charging data, including: constant current charging time, constant voltage charging time, initial voltage, average voltage, constant current charging time / constant voltage charging time, constant current charging time / total charging time, thus obtaining 6 different features; Step 2.4, processing the charging data to obtain an incremental capacity curve; Step 2.5, if the peak disappears in the curve, find the point with the smallest absolute value of the derivative in the voltage window and take it as the peak value; Step 2.

6. Select the peak value and peak voltage of each peak (the curve has three peaks in total) of the incremental capacity curve after the peak disappearance is processed as the other 6 features of the input machine learning model.

5. The method for diagnosing the health status of a lithium-ion battery based on equivalent circuit feature screening according to claim 1, characterized in that: Step 3 specifically includes: Step 3.1, input the 12 features extracted above into the machine learning model, the machine learning model includes: LightGBM model, XGBoost model, GPR model and Ridge model; Step 3.2, under different temperature conditions, set different training and testing paths, and adopt a cross-validation strategy: at 25°C, four batteries, 25C03, 25C06, 25C11, and 25C19, will be used, each battery as a test set, and the remaining batteries as training sets; and so on, 35°C and 45°C also follow the above path, except that the batteries are different. The batteries used for 35°C are 35C01 and 35C02; the batteries used for 45°C are 45C13, 45C16, 45C21, and 45C27; this cross-validation framework effectively ensures the generalization ability of the model for different individual batteries of the same model; the input is the 12 features obtained above, and the output is the battery health status; Step 3.3: Use common evaluation indicators for machine learning, including mean square error (MSE), root mean square error (RMSE), and coefficient of determination (R²).

6. The method for diagnosing the health status of a lithium-ion battery based on equivalent circuit feature screening according to claim 1, characterized in that: Step 4 specifically includes: Step 4.1, extracting the impedance data of the battery: including a total of 100 data points within the frequency range of 50 mHz to 10 kHz, and extracting once every 25 cycles (impedance test cycles); Step 4.2, according to the obtained impedance data, an equivalent circuit model is established and fitted through the module of the python library impedance.py, and the equivalent circuit model is modeled according to FIG3; Step 4.3: Extract the parameters most relevant to battery degradation in the equivalent circuit model: R sei , C sei , C ct , R ct , Z w ; Step 4.4: Perform Pearson correlation analysis on the 12 features extracted in step 2 and the five parameters. Add the Pearson coefficients of each feature to the parameters, take the average and sort them. Remove the features that are ranked at the bottom or have an average value less than 0.6 to obtain the screened features.

7. A lithium-ion battery health status diagnosis method based on equivalent circuit feature screening according to claim 1, characterized in that: Step 5 specifically includes: Step 5.1: Similar to the solution in step 3 above, a cross-validation strategy is adopted, with the input being the filtered features and the output being the battery health status; Step 5.

2. Evaluate the results using the same machine learning evaluation metrics as in step 3 above.

Images