A lithium-ion battery state-of-health adaptive estimation method

Abstract

The application relates to a lithium ion battery health state adaptive estimation method, which comprises the following steps: 1) performing a battery aging experiment, calibrating the capacity at intervals, and collecting electrochemical impedance spectra at different states of charge; 2) performing time scale identification on the electrochemical impedance spectra and extracting electrochemical impedance time scale features; 3) selecting electrochemical impedance time scale features highly related to the battery health state to form a degradation feature data set; 4) training a health state estimation model based on the degradation feature data set; 5) acquiring the electrochemical impedance spectrum at the current state of charge of the battery in actual application and extracting the same type of electrochemical impedance time scale features as in 3); and 6) inputting the electrochemical impedance time scale features into the health state estimation model to obtain the battery health state estimation result. Compared with the prior art, the application has the advantages of good robustness and suitability for different scenes.

Description

Technical Field

[0001] This invention relates to the technical field of battery management, and in particular to an adaptive estimation method for the state of health of lithium-ion batteries. Background Technology

[0002] Lithium-ion batteries are widely used in transportation, consumer electronics, and energy storage power stations due to their comprehensive advantages such as high energy density, environmental friendliness, and cost-effectiveness. However, after long-term use, the capacity and performance of lithium-ion batteries inevitably degrade, and in severe cases, they may even fail, easily leading to safety accidents. The performance degradation of lithium-ion battery systems seriously affects the driving range of electric vehicles, greatly hindering the further promotion of electric vehicles and the electrification transformation of the transportation sector. Therefore, accurately detecting the health status of batteries is of great significance for improving the safety and reliability of electric vehicles.

[0003] Currently, lithium-ion battery health state estimation methods are mainly divided into two types: The first is a model-based method, which mainly uses battery models such as electrochemical and equivalent circuit models to estimate the health state. This type of estimation method has high accuracy, but some key parameters of the battery model, especially the key parameters of the electrochemical model, are still difficult to obtain accurately. In addition, the accuracy of health state estimation is highly dependent on the battery operating conditions and the model. The second is a data-driven method, which estimates the battery health state based on the relationship between degradation features and health state and combines machine learning algorithms. The degradation features and machine learning algorithms used directly affect the health state estimation performance. The degradation features used in existing research can be divided into the following three types: 1) Measured raw battery data, such as voltage, current, and temperature; these features are relatively easy to obtain, but they are essentially external characteristics of the battery and lack in-depth understanding of the battery's internal state, making them unsuitable for health state estimation under highly dynamic operating conditions. 2) Statistical and geometric features of raw battery data; these features need to be further extracted from the measured raw battery data, but due to the lack of in-depth understanding of the battery's internal state, they are also unsuitable for health state estimation under highly dynamic operating conditions. 3) Features based on differential technology: Differential technology can extract degradation features from the differential curves of the battery's electrical, thermal, and mechanical characteristics. These features usually have certain physical meaning and are related to the battery's internal degradation, enabling effective health state estimation under highly dynamic operating conditions. However, the application scenarios of these features are still strictly limited. For example: 31) In practical applications, batteries are usually charged from different initial SOCs, which inevitably affects the effectiveness of the degradation features of the capacity increment curve and differential voltage curve. This not only poses a serious challenge to the performance of battery health state estimation but also limits the adaptability of health state estimation. In addition, acquiring relevant features is still very time-consuming. 32) Although features based on mechanical characteristics are related to the battery's internal degradation and are beneficial for achieving effective battery health state estimation under highly dynamic operating conditions, the acquisition of mechanical characteristic data requires additional sophisticated equipment, resulting in quite limited application scenarios.

[0004] In summary, existing methods for estimating the state of health of lithium-ion batteries suffer from several problems, including difficulty in obtaining key parameters, a lack of in-depth understanding of the battery's internal state, and a very limited range of applications. They are not suitable for estimating the state of health under highly dynamic operating conditions. Summary of the Invention

[0005] The purpose of this invention is to provide an adaptive estimation method for the health status of lithium-ion batteries in order to overcome the defects of the prior art.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] An adaptive estimation method for the health state of a lithium-ion battery, comprising the following steps:

[0008] 1) Conduct battery aging experiments and perform capacity calibration at certain charge-discharge cycles, while collecting electrochemical impedance spectra under different charging states.

[0009] 2) Identify the time scale of electrochemical impedance spectroscopy and extract the time scale characteristics of electrochemical impedance spectroscopy;

[0010] 3) Select electrochemical impedance timescale features that are highly correlated with battery health status based on Pearson correlation coefficient to form a degradation feature dataset;

[0011] 4) Based on the decay feature dataset, an offline training model for estimating health status based on electrochemical impedance time-scale features and ensemble learning was developed.

[0012] 5) In practical applications, obtain the electrochemical impedance spectrum of the battery under its current state of charge and extract the same type of electrochemical impedance timescale features as in 3).

[0013] 6) Input the electrochemical impedance timescale features from 5) into the trained health state estimation model to obtain the battery health state estimation results;

[0014] The specific process of training the health status estimation model is as follows:

[0015] Input a training set, which includes a reference health state and a decay feature dataset, wherein the reference health state is determined based on a calibrated capacity;

[0016] Define the regression function, the squared loss function, and the number of iterations; initialize the regression function.

[0017] The model's parameters are optimized through iteration to obtain a trained health status estimation model.

[0018] Furthermore, the time-scale characteristics of electrochemical impedance include the central time constant τ of the characteristic peak in the electrochemical impedance relaxation time distribution curve. C The relaxation time distribution γ(τ) corresponding to the center time constant of the characteristic peak c And the interfacial resistance R of the kinetic process corresponding to the characteristic peak.

[0019] Furthermore, the expression for the interfacial resistance in the kinetic process is:

[0020]

[0021] Where, τ L and τ UThese are the lower and upper time constants for each dynamic process, respectively. τ is the relaxation time, and γ(τ) is the relaxation time distribution function.

[0022] Furthermore, the specific steps for performing iterations during the training of the health status estimation model are as follows:

[0023] Predict the residuals of the ensemble estimator updated in the previous round; fit the new base learner and residuals using the loss function to determine the parameters of the new base learner; update the ensemble estimator; repeat the above steps until the number of iterations reaches a preset value to establish the final ensemble estimator, which is the trained health state estimation model; where, when the number of iterations is 1, the ensemble estimator is the initially set regression function.

[0024] Furthermore, during the iteration process, the predicted residuals of the previous round of ensemble estimator are specifically as follows:

[0025]

[0026] in, Let y be the residual of the previous round ensemble estimator for the i-th training sample. i It is the label of the i-th training sample, which is the reference battery health status, F t-1 (x i ) represents the updated ensemble estimator after the (t-1)th iteration, where n is the total number of training samples;

[0027] The learning rate and the parameter set of the representation base learner are:

[0028]

[0029] Where, β t Let a be the learning rate for the t-th training iteration. t To characterize the parameter set of the newly added base learner in the t-th iteration, h(x) i a) is the base learner for the t-th training iteration. Let be the residual of the i-th training sample;

[0030] When the number of iterations reaches the preset value, the established ensemble estimator is:

[0031] F t (x)=F t-1 (x)+β t h(x;a t )

[0032] Among them, F t (x) is the ensemble estimator, F t-1 (x) is the ensemble estimator trained for the (t-1)th time, β tLet h(x) be the learning rate for the t-th training iteration. i ; a) is the base learner for the t-th training iteration.

[0033] Furthermore, the initial regression function is set as follows:

[0034]

[0035] Where F0(x) is the initially set regression function. Indicates y i The mean of y i It is the label of the i-th training sample.

[0036] Furthermore, the squared loss function is:

[0037] L(y,F(x))=(yF(x)) 2 / 2

[0038] Where y represents the label of the training sample, F(x) represents the regression function, and x represents the sample feature vector containing impedance time-scale features.

[0039] Furthermore, the SOC conditions for electrochemical impedance spectroscopy include 30% SOC, 50% SOC, 70% SOC, and 90% SOC.

[0040] Furthermore, the actual electrochemical impedance data were obtained using an electrochemical workstation.

[0041] Furthermore, the actual electrochemical impedance data is obtained through a battery management system with electrochemical impedance testing capabilities.

[0042] Compared with the prior art, the present invention has the following beneficial effects:

[0043] (1) The lithium-ion battery health state estimation method proposed in this invention uses the time scale information of electrochemical impedance as the decay feature. The above features are quantitative features of the internal dynamic process of the battery, which have clear physical meaning and are highly correlated with the battery health state. This enables adaptive health state estimation in a wide SOC range without obtaining the battery SOC and historical operating conditions. Moreover, the estimation method has strong generalization and robustness.

[0044] (2) The adaptive estimation method for the health status of lithium-ion batteries proposed in this invention takes less time to obtain the required electrochemical impedance time scale features and is more flexible. Compared with the existing technology, it has better application prospects for highly dynamic actual operating conditions. Attached Figure Description

[0045] Figure 1 This is a flowchart of the present invention;

[0046] Figure 2 This is a schematic diagram illustrating the capacity degradation of a lithium-ion battery during cycling.

[0047] Figure 3 The evolution of the electrochemical impedance relaxation time distribution curve during the degradation process of lithium-ion batteries;

[0048] Figure 4 A schematic diagram illustrating the time-scale feature extraction of electrochemical impedance of lithium-ion batteries;

[0049] Figure 5 This is an adaptive estimation result of the health status of a lithium-ion battery based on electrochemical impedance timescale information. Detailed Implementation

[0050] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0051] This invention proposes an adaptive estimation method for the health status of lithium-ion batteries, the flowchart of which is shown below. Figure 1 As shown. 1) Perform aging tests and calibrate the capacity at certain intervals, and collect electrochemical impedance spectroscopy under different charging states;

[0052] 2) Identify the electrochemical impedance on a time scale and extract its time-scale characteristics;

[0053] 3) Based on the Pearson correlation coefficient, select electrochemical impedance timescale features that are highly correlated with health status to form a degradation feature dataset;

[0054] 4) Offline training of a health status estimation model based on electrochemical impedance timescale characteristics and ensemble learning;

[0055] 5) In practical applications, obtain the electrochemical impedance of the battery under its current state of charge and extract the time-scale characteristics of the same type of impedance as in 3);

[0056] 6) Input the features extracted in 5) into the health status estimation model to estimate the battery health status.

[0057] In step 1), a battery aging experiment is designed based on the recommended operating temperature and charge / discharge current, and charge / discharge cycle experiments are conducted on the battery. In some embodiments, a 0.5C CC charging method is used, and the discharging method adopts the NEDC condition, which is closer to the actual dynamic vehicle operating conditions. The battery charge / discharge cycle temperature is set to 25°C. Periodic capacity calibration tests and electrochemical impedance tests at different SOCs are performed on the battery every 100 NEDC cycles (in this example, including 30% SOC, 50% SOC, 70% SOC, and 90% SOC). The periodic capacity calibration results are as follows: Figure 2 As shown.

[0058] In section 2), the electrochemical impedance relaxation time distribution curve obtained by time-scale identification of the battery electrochemical impedance is shown in Figure 2. Figure 3 As shown. Figure 4 This diagram illustrates the extraction of time-scale features of battery electrochemical impedance. The time-scale features of electrochemical impedance include the central time constant τ of the characteristic peak in the electrochemical impedance relaxation time distribution curve. C The relaxation time distribution γ(τ) corresponding to the center time constant of the characteristic peak C The characteristic peak and the corresponding kinetic process interface resistance R are calculated using the following formula:

[0059]

[0060] τ L and τ U These are the lower and upper time constants for each dynamic process, respectively. τ is the relaxation time, and γ(τ) is the relaxation time distribution function.

[0061] In step 3), key degradation features are screened based on the Pearson correlation coefficient with battery health status. The larger the absolute value of the correlation coefficient, the stronger the correlation.

[0062] In section 4), the health status estimation model is trained through homogeneous ensemble learning. The ensemble learning employs a least-squares boosting strategy, and the base learner is a regression tree. The specific process for training the adaptive health status estimation model is as follows: input a training set, which includes a reference health status and the electrochemical impedance timescale features selected in section 3); define the regression function, the squared loss function, and the number of iterations; initialize the regression function; perform iterations to optimize the model parameters, resulting in the trained health status estimation model. The specific training process for implementing the least-squares boosting regression tree ensemble is as follows:

[0063] ① Input training set: Where n is the number of training samples, x i It is the feature vector of the i-th sample containing impedance time-scale features, y iIt is the label of the i-th training sample, which is the battery health status.

[0064] ② Define the regression function F(x) and the squared loss function L(y,F(x))=(yF(x)) 2 / 2 and the number of iterations T.

[0065] ③ Initialize the regression function:

[0066] ④ Perform iterations, that is, execute the following for t = 1 to T:

[0067]

[0068]

[0069] F t (x)=F t-1 (x)+β t h(x;a t )

[0070] in It is y i The mean, The current residual is β, which is the learning rate (ranging from 0 to 1), and h(x) is the learning rate. i ; a) is the base learner (i.e., the regression tree) characterized by the parameter set a.

[0071] ⑤ End the iteration; training complete.

[0072] In 5), the electrochemical impedance of the battery under its current state of charge is obtained in practical applications. In laboratory applications, the battery electrochemical impedance can be measured by an electrochemical workstation; in actual vehicle applications, the current battery impedance data can be obtained by a battery management system with electrochemical impedance testing function.

[0073] In step 6), it is not necessary to obtain the battery's state of charge and historical operating conditions. By inputting the features extracted in step 5) into the health state estimation model, adaptive estimation of the health state within a wide state of charge range can be achieved.

[0074] In section 6), based on the above-mentioned adaptive estimation method for battery health state based on electrochemical impedance timescale information, the battery adaptive health state estimation results are as follows: Figure 5 As shown, if the estimated point falls exactly on the dashed line, it indicates that the health status estimation has no error. From Figure 5It can be seen that, for the same degradation state, the data points representing the estimated health state under different charge states are close to each other and are evenly distributed near the dashed line. The estimated MAE of the battery health state under different charge states does not exceed 0.98%, and the RMSE does not exceed 1.17%. Compared with the existing technology, the estimation accuracy is higher and it has adaptability.

[0075] In summary, one embodiment of the present invention is feasible. The battery health state estimation has high accuracy and the proposed estimation method is adaptive within a wide state of charge range, showing good application prospects for highly dynamic actual operating conditions.

[0076] The proposed lithium-ion battery health state estimation method uses electrochemical impedance time-scale information as a decay characteristic. These characteristics are quantitative features of the battery's internal dynamic processes, possessing clear physical meaning and being highly correlated with the battery's health state. This allows for adaptive health state estimation over a wide state-of-charge range without acquiring the battery's state of charge or historical operating conditions. Furthermore, the estimation method exhibits strong generalization and robustness. The method of acquiring the required electrochemical impedance time-scale characteristics is less time-consuming and more flexible, offering better application prospects for highly dynamic real-world operating conditions compared to existing technologies.

[0077] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. An adaptive estimation method for the health status of a lithium-ion battery, characterized in that, The method includes the following steps: 1) Conduct battery aging experiments and perform capacity calibration at certain charge-discharge cycles, while collecting electrochemical impedance spectra under different charging states. 2) Identify the time scale of electrochemical impedance spectroscopy and extract the time scale characteristics of electrochemical impedance spectroscopy; 3) Select electrochemical impedance timescale features that are highly correlated with battery health status based on Pearson correlation coefficient to form a degradation feature dataset; 4) Based on the decay feature dataset, an offline training model for estimating health status based on electrochemical impedance time-scale features and ensemble learning was developed. 5) In practical applications, obtain the electrochemical impedance spectrum of the battery under its current state of charge and extract the same type of electrochemical impedance timescale features as in 3). 6) Input the electrochemical impedance timescale features from 5) into the trained health state estimation model to obtain the battery health state estimation results; The specific process of training the health status estimation model is as follows: Input a training set, which includes a reference health state and a decay feature dataset, wherein the reference health state is determined based on a calibrated capacity; Define the regression function, the squared loss function, and the number of iterations; initialize the regression function. The model's parameters are optimized through iteration to obtain a trained health status estimation model.

2. The adaptive estimation method for the health status of a lithium-ion battery according to claim 1, characterized in that, The time-scale characteristics of electrochemical impedance include the central time constant τ of the characteristic peak in the electrochemical impedance relaxation time distribution curve. C The relaxation time distribution γ(τ) corresponding to the center time constant of the characteristic peak C And the interfacial resistance R of the kinetic process corresponding to the characteristic peak.

3. The adaptive estimation method for the health status of a lithium-ion battery according to claim 2, characterized in that, The expression for the interfacial resistance in a kinetic process is: Where, τ L and τ U These are the lower and upper time constants for each dynamic process, respectively. τ is the relaxation time, and γ(τ) is the relaxation time distribution function.

4. The adaptive estimation method for the health status of a lithium-ion battery according to claim 1, characterized in that, The specific steps for performing iterations during the training of the health status estimation model are as follows: Predict the residuals of the ensemble estimator updated in the previous round; fit the new base learners and residuals using the loss function to determine the parameters of the new base learners; update the ensemble estimator; Repeat the above steps until the number of iterations reaches the preset value, and establish the final ensemble estimator, which is the trained health status estimation model; where the number of iterations is 1, the ensemble estimator is the initially set regression function.

5. The adaptive estimation method for the health status of a lithium-ion battery according to claim 4, characterized in that, During the iteration process, the predicted residuals of the previous round of ensemble estimator are specifically as follows: in, Let y be the residual of the previous round ensemble estimator for the i-th training sample. i It is the label of the i-th training sample, which is the reference battery health status, F t-1 (x i ) represents the updated ensemble estimator after the (t-1)th iteration, where n is the total number of training samples; The learning rate and the parameter set of the representation base learner are: Where, β t Let a be the learning rate for the t-th training iteration. t To characterize the parameter set of the newly added base learner in the t-th iteration, h(x) i a) is the base learner for the t-th training iteration. Let be the residual of the i-th training sample; When the number of iterations reaches the preset value, the established ensemble estimator is: F t (x)=F t-1 (x)+β t h(x;a t ) Among them, F t (x) is the ensemble estimator, F t-1 (x) is the ensemble estimator trained for the (t-1)th time, β t Let h(x) be the learning rate for the t-th training iteration. i ; a) is the base learner for the t-th training iteration.

6. The adaptive estimation method for the health status of a lithium-ion battery according to claim 4, characterized in that, The initial regression function is set as follows: Where F0(x) is the initially set regression function. Indicates y i The mean of y i It is the label of the i-th training sample.

7. The adaptive estimation method for the health status of a lithium-ion battery according to claim 4, characterized in that, The squared loss function is: L(y,F(x))=(y-F(x)) 2 / 2 Where y represents the label of the training sample, F(x) represents the regression function, and x represents the sample feature vector containing impedance time-scale features.

8. The adaptive estimation method for the health status of a lithium-ion battery according to claim 1, characterized in that, The SOC points for electrochemical impedance spectroscopy include 30% SOC, 50% SOC, 70% SOC, and 90% SOC.

9. The adaptive estimation method for the health status of a lithium-ion battery according to claim 1, characterized in that, Actual electrochemical impedance data were obtained using an electrochemical workstation.

10. The adaptive estimation method for the health status of a lithium-ion battery according to claim 1, characterized in that, Actual electrochemical impedance data is obtained through a battery management system with electrochemical impedance testing capabilities.

Images