A multi-feature lithium battery state of health online estimation method and device

Abstract

The application provides a multi-feature lithium battery health state online estimation method and device, comprising the following steps: step S1, acquiring first feature F1, second feature F2, third feature F3, fourth feature F4, fifth feature F5 and sixth feature F6; step S2, taking the first to sixth features in step S1 as an input vector X=[F1, F2, F3, F4, F5, F6], establishing a hypothesis function formula of a multiple linear regression model: Y=θ0+θ1F1+θ2F2+…+θ6F6, and preliminarily determining a parameter vector φ=[θ0, θ1, …, θ6] by using a gradient descent algorithm; step S3, randomly generating N initialization particles according to the parameter vector φ, updating each initialization particle by using Gaussian white noise, and updating the parameter vector and the predicted value of the multiple linear regression model after importance sampling and resampling. The application extracts multiple features from different angles to construct a more accurate multiple linear regression model, and updates the model parameters online, thereby improving the efficiency and accuracy of online estimation.

Description

Technical Field

[0001] This invention relates to a method and apparatus for online estimation of the health status of a lithium battery with multiple features. Background Technology

[0002] Effective management and control of lithium batteries is one of the core technologies that distinguishes new energy vehicles from ordinary gasoline vehicles. With the increase in the number of charge-discharge cycles during use, the aging phenomenon that occurs in lithium batteries has a significant impact on the lifespan of lithium-ion batteries and the safe driving range of electric vehicles. The state of health (SOH) of a lithium battery is an important indicator for identifying the degree of aging of lithium-ion batteries and a key parameter of the lithium-ion battery management system. Therefore, predicting the health status of lithium-ion batteries can help avoid potential safety accidents and extend battery life.

[0003] Methods for estimating the state of harmonics (SOH) of batteries can be broadly categorized into model-based and data-driven approaches. Model-based methods construct models by analyzing the internal electrochemical reaction mechanisms and external electrical characteristics of the battery. These methods can be further subdivided into electrochemical models, equivalent circuit models, and empirical models, and then the aging degree of the battery is estimated based on the model. This method offers good real-time performance and strong adaptability, but it is very difficult to combine all influencing factors, easily leading to complex models and high computational costs. Data-driven methods, on the other hand, construct a mapping relationship between feature vectors during the aging process and the battery's SOH by mining them. Traditional machine learning methods such as neural networks, support vector regression, and Gaussian regression have been applied to SOH estimation. These methods do not require in-depth understanding of the internal aging mechanisms of lithium batteries and have relatively low computational costs. However, these methods are still trained offline, and when the sample size is small, the estimation effect of data-driven methods is generally poor, often exhibiting a high dependence on data during use. Summary of the Invention

[0004] This invention proposes a multi-feature online estimation method and device for the health status of lithium batteries. It extracts multiple features from different perspectives to construct a more accurate multiple linear regression model and updates the model parameters online, thereby improving the efficiency and accuracy of online estimation.

[0005] This invention is achieved through the following technical solution:

[0006] A multi-feature online estimation method for the state of health of lithium batteries includes the following steps:

[0007] Step S1: Through the aging cycle test of the lithium battery, obtain the charging time of the constant current charging mode as the first feature F1, the initial voltage of the constant current charging mode as the second feature F2, the cumulative charging voltage of the constant current charging mode as the third feature F3, the initial slope of the constant current charging mode as the fourth feature F4, the average voltage of the lithium battery as the fifth feature F5, and the slope at the end of the constant current charging mode as the sixth feature F6.

[0008] Step S2: Form a sample vector X = [F1, F2, F3, F4, F5, F6] from the first to the sixth features. Use the sample vector X as the input vector to establish the hypothesis function formula of the multiple linear regression model: Y = θ0 + θ1F1 + θ2F2 + ... + θ6F6. Use the gradient descent algorithm to initially determine the parameter vector φ = [θ0, θ1, ..., θ6], where θ0 is the intercept, and the parameters θ1-θ6 are the regression coefficients corresponding to the first to the sixth features. Y is the estimated health status of the lithium battery.

[0009] Step S3: Randomly generate N initial particles based on the parameter vector φ, and update each initial particle using Gaussian white noise. Then, perform importance sampling and resampling to update the parameter vector and the predicted values ​​of the multiple linear regression model.

[0010] Furthermore, step S3 specifically includes:

[0011] S31. Randomly generate N initial particles based on the parameter vector φ. i = 1, 2, ..., N;

[0012] S32, based on φ k =φ k-1 +u k Update particles Among them, u k It is Gaussian white noise, where k refers to the number of particle updates;

[0013] S33. Obtain the weights by performing importance sampling according to the following formula. Normalize according to the following formula When N th >N eff When δ is reached, resampling is performed, where δ v For standard deviation, y k These are reference values ​​for health status. N th =2N / 3;

[0014] S34. Update the parameter vector φ using the following formula. k The predicted value Y of the multiple linear regression model k :

[0015] Furthermore, the health status reference value y k The calculation process is as follows: through Calculate the current capacity of the lithium battery, then calculate y. k =C current / C initial Where I(τ) represents the magnitude of the current, and t f Represents the charging time, η is the coulombic efficiency, and C initial This refers to the initial nominal capacity of the lithium battery.

[0016] Furthermore, in step S1, before proceeding to step S2, the first to sixth features need to be normalized according to the following formula: F norl =(F l -min(F l )) / (max(F l )-min(F l )), l=1,2,…,6.

[0017] Furthermore, in step S2, the cost function of the multiple linear regression model is: Among them, X (i) Y(X) represents the i-th sample vector, m is the number of sample vectors, and Y(X) represents the number of sample vectors. (i) y represents the hypothesis function value obtained when the i-th sample vector is input. (i) This represents the true battery health value corresponding to the i-th sample vector.

[0018] Furthermore, in step S2, the gradient descent algorithm is as follows: Where α is the learning rate.

[0019] Furthermore, in step S1, the aging cycle test of the lithium battery refers to the test conducted using size 5, 6, and 7 lithium batteries from the NASA dataset under constant temperature conditions at room temperature.

[0020] Furthermore, in step S1, the third feature F3 = ∑|U c (i+1)-U c (i)|, where, U c (i) represents the charging voltage.

[0021] Furthermore, in step S1, the fifth feature Among them, U d (i) represents the terminal voltage of the lithium battery.

[0022] This invention is also achieved through the following technical solutions:

[0023] A multi-feature online health state estimation device for lithium batteries, comprising:

[0024] Feature acquisition module: used to acquire the following features through aging cycle experiments of lithium battery: charging time of constant current charging mode as first feature F1, initial voltage of constant current charging mode as second feature F2, cumulative charging voltage of constant current charging mode as third feature F3, initial slope of constant current charging mode as fourth feature F4, average voltage of lithium battery as fifth feature F5, and slope at the end of constant current charging mode as sixth feature F6.

[0025] Model building module: Used to establish the hypothesis function formula of the multiple linear regression model with the first to sixth features as input vector X = [F1, F2, F3, F4, F5, F6]: Y = θ0 + θ1F1 + θ2F2 + ... + θ6F6, and use the gradient descent algorithm to initially determine the parameter vector φ = [θ0, θ1, ..., θ6], where θ0 is the intercept, and the parameters θ1-θ6 are the regression coefficients corresponding to the first to sixth features. Y is the estimated health status of the lithium battery.

[0026] The parameter update module is used to randomly generate N initial particles based on the parameter vector φ, update each initial particle using Gaussian white noise, and then update the parameter vector and the predicted values ​​of the multiple linear regression model after importance sampling and resampling.

[0027] The present invention has the following beneficial effects:

[0028] 1. This invention first obtains the first to sixth features from different perspectives, then uses the first to sixth features as input vectors to establish a multiple linear regression model. The gradient descent algorithm is used to initially determine the parameter vector, and then N initial particles are randomly generated according to the parameter vector φ. Gaussian white noise is used to update each initial particle. Finally, importance sampling and resampling are performed to update the parameter vector and the predicted value of the multiple linear regression model. This can build a more accurate model, and all types of features are easy to extract. It also enables online updating of model parameters, improves the dynamic performance of the model, and thus improves the efficiency and accuracy of online estimation. Attached Figure Description

[0029] The present invention will now be described in further detail with reference to the accompanying drawings.

[0030] Figure 1 This is a route diagram for the present invention.

[0031] Figures 2-1 to 2-6 The relationships between the first to sixth characteristics and the number of aging cycles are shown respectively. Detailed Implementation

[0032] like Figure 1 As shown, the online estimation method for the health status of a lithium battery with multiple features includes the following steps:

[0033] Step S1: The No. 5, No. 6 and No. 7 lithium batteries in the NASA dataset were subjected to aging cycle test under constant temperature conditions at room temperature. The lithium battery charging process includes two processes: constant current charging and constant voltage charging. In constant current charging mode, the charging current is 1.5A. When the voltage reaches 4.2V, the charging continues in constant voltage charging mode.

[0034] Figure 2-1 It is shown that as the number of cycles increases, the charging time of the constant current charging mode continuously decreases, and the polarization degree of the lithium battery also gradually decreases. It can be seen that the charging time of the constant current charging mode is related to the health status of the lithium battery. Therefore, the charging time of the constant current charging mode is obtained as the first feature F1.

[0035] Figure 2-2 It is shown that the initial voltage of the constant current charging mode increases continuously with the increase of the number of cycles, while the rate of increase decreases continuously until it reaches the critical point of the initial voltage. Therefore, the initial voltage of the constant current charging mode is obtained as the second feature F2.

[0036] Figure 2-3 The cumulative charging voltage in constant current charging mode is shown to be related to the number of cycles. Therefore, the cumulative charging voltage in constant current charging mode is obtained as the third feature F3, which can be expressed by the following formula: F3=∑|U c (i+1)-U c (i)|, where, U c (i) represents the charging voltage;

[0037] Figure 2-4 It shows that during the charging process, the initial slope of the constant current charging mode changes continuously. At the beginning of the aging cycle, the initial slope continues to decrease. After reaching a certain critical value, the initial slope slowly increases and tends to stabilize. Therefore, the initial slope of the constant current charging mode is obtained as the fourth feature F4.

[0038] As the number of charging cycles increases, the internal resistance of the lithium battery continuously increases; therefore, as... Figure 2-5 As shown, the average voltage is also increasing, so the average voltage of the lithium battery is taken as the fifth feature F5. The fifth feature F5 can be expressed by the following formula: Among them, U d (i) represents the terminal voltage of the lithium battery;

[0039] The slope at the end of the constant current charging mode indicates that throughout the mode, the curve's slope gradually decreases and then stabilizes in a certain region. The change in the slope at the end of the curve reflects the increase in the lithium battery's internal resistance. Figure 2-6 As shown, the slope at the end of the constant current charging mode is obtained as the sixth feature F6.

[0040] The six features mentioned above are on different orders of magnitude and have different units of measurement, making them incomparable. Therefore, these six features are normalized using the following formula: F norl =(F l -min(F l )) / (max(F l )-min(F l ), l=1,2,…,6;

[0041] Step S2: Form a sample vector X = [F1, F2, F3, F4, F5, F6] from the first to sixth features described in step S1. Use the sample vector X as the input vector to establish the hypothesis function formula of the multiple linear regression model: Y = θ0 + θ1F1 + θ2F2 + ... + θ6F6, where θ0 is the intercept, and the parameters θ1-θ6 are the regression coefficients corresponding to the first to sixth features. Y is the estimated health status of the lithium battery.

[0042] The cost function of the multiple linear regression model is Among them, X (i) Y(X) represents the i-th sample vector, m is the number of sample vectors, and Y(X) represents the number of sample vectors. (i) y represents the hypothesis function value obtained when the i-th sample vector is input. (i) The true battery health value corresponding to the i-th sample vector is obtained by minimizing the cost function and solving for the parameter vector φ = [θ0, θ1, ..., θ6] in the multiple linear regression equation.

[0043] Then, the following gradient descent algorithm is used. To initially determine the parameter vector φ = [θ0, θ1, ..., θ6], α is the learning rate, used to control the magnitude of the descent. That is, to find θ for the cost function J(θ0,θ1,…,θ6) j Partial derivatives of (j=0,1,…,6);

[0044] Step S3: Randomly generate N initial particles based on the parameter vector φ, and update each initial particle using Gaussian white noise. Then, perform importance sampling and resampling to update the parameter vector and the predicted values ​​of the multiple linear regression model. Specifically, this includes:

[0045] S31. Randomly generate N initial particles based on the parameter vector φ. i = 1, 2, ... N, and the dimension of each particle is 7;

[0046] S32, based on φ k =φ k-1 +u k Update particles Among them, u k It is Gaussian white noise, where k refers to the number of particle updates;

[0047] S33. Obtain the weights by performing importance sampling according to the following formula. Normalize according to the following formula Where, δ v The standard deviation is a known parameter that can take the value 1. k This is a health status reference value, calculated as follows: through... Calculate the current capacity of the lithium battery, then calculate y. k =C current / C initial Where I(τ) represents the magnitude of the current, and t f Represents the charging time, η is the coulombic efficiency, and C initial This refers to the initial nominal capacity of the lithium battery.

[0048] When N th >N eff If so, resampling will be performed. N th =2N / 3;

[0049] S34. Update the parameter vector φ using the following formula. k The predicted value Y of the multiple linear regression model k :

[0050]

[0051] Correspondingly, a multi-feature online lithium battery health state estimation device includes:

[0052] Feature acquisition module: used to acquire the following features through aging cycle experiments of lithium battery: charging time of constant current charging mode as first feature F1, initial voltage of constant current charging mode as second feature F2, cumulative charging voltage of constant current charging mode as third feature F3, initial slope of constant current charging mode as fourth feature F4, average voltage of lithium battery as fifth feature F5, and slope at the end of constant current charging mode as sixth feature F6.

[0053] Model building module: Used to establish the hypothesis function formula of the multiple linear regression model with the first to sixth features as input vector X = [F1, F2, F3, F4, F5, F6]: Y = θ0 + θ1F1 + θ2F2 + ... + θ6F6, and use the gradient descent algorithm to initially determine the parameter vector φ = [θ0, θ1, ..., θ6], where θ0 is the intercept, and the parameters θ1-θ6 are the regression coefficients corresponding to the first to sixth features. Y is the estimated health status of the lithium battery.

[0054] The parameter update module is used to randomly generate N initial particles based on the parameter vector φ, update each initial particle using Gaussian white noise, and then update the parameter vector and the predicted values ​​of the multiple linear regression model after importance sampling and resampling.

[0055] The above description is merely a preferred embodiment of the present invention and should not be construed as limiting the scope of the present invention. All equivalent changes and modifications made in accordance with the scope of the patent application and the contents of the specification of the present invention should still fall within the scope of the patent of the present invention.

Claims

1. A multi-feature online estimation method for the state of health of a lithium battery, characterized in that: Includes the following steps: Step S1: Obtain the charging time in constant current charging mode as the first feature through an aging cycle experiment of the lithium battery. The initial voltage of the constant current charging mode is used as the second characteristic. The cumulative charging voltage in constant current charging mode is used as a third feature. The initial slope of the constant current charging mode is used as the fourth feature. The average voltage of lithium batteries is the fifth characteristic. The slope at the end of the constant current charging mode is used as the sixth feature. ; Step S2: Form a sample vector from the first to the sixth features. , sample vector The hypothesis function formula for establishing a multiple linear regression model is as follows: And the gradient descent algorithm is used to initially determine the parameter vector. , For intercept, parameter The regression coefficients correspond to the first through sixth features. This is the valuation of the health status of lithium batteries; Step S3: Based on the parameter vector Randomly generated Each initial particle is initialized and updated using Gaussian white noise. Then, importance sampling and resampling are performed to update the parameter vector and the predicted values ​​of the multiple linear regression model. Step S3 specifically includes: S31, Based on the parameter vector Randomly generated Initialized particles , ; S32, based on Update particles ,in, It is Gaussian white noise. This refers to the number of particle updates. S33. Obtain the weights by performing importance sampling according to the following formula. Normalize according to the following formula. ,when If this occurs, resampling will be performed, where... For standard deviation, These are reference values ​​for health status. , ; S34. Update the parameter vector using the following formula. and the predicted values ​​of the multiple linear regression model : .

2. The online estimation method for the health status of a lithium battery with multiple features according to claim 1, characterized in that: The health status reference value The calculation process is as follows: through Calculate the current capacity of the lithium battery, then calculate ,in, Indicates the magnitude of the current. Indicates charging time. For Coulomb efficiency, This refers to the initial nominal capacity of the lithium battery.

3. The online estimation method for the health status of a lithium battery with multiple features according to claim 1 or 2, characterized in that: In step S1, before proceeding to step S2, the first to sixth features need to be normalized according to the following formula: , .

4. The online estimation method for the health status of a multi-feature lithium battery according to claim 1 or 2, characterized in that: In step S2, the cost function of the multiple linear regression model is: ,in, Indicates the first A sample vector, The number of sample vectors. Indicates the input number The hypothesis function value obtained when there are 100 sample vectors. For the first The true battery health value corresponding to each sample vector.

5. The online estimation method for the health status of a lithium battery with multiple features according to claim 4, characterized in that: In step S2, the gradient descent algorithm is as follows: ,in, This is the learning rate.

6. The online estimation method for the health status of a multi-feature lithium battery according to claim 1 or 2, characterized in that: In step S1, the aging cycle test of the lithium battery refers to the test conducted using lithium batteries No. 5, No. 6 and No. 7 from the NASA dataset under constant temperature conditions at room temperature.

7. The online estimation method for the health status of a multi-feature lithium battery according to claim 1 or 2, characterized in that: In step S1, the third feature ,in, This is the charging voltage.

8. The online method for estimating the health status of a lithium battery with multiple features according to claim 1 or 2, characterized in that: In step S1, the fifth feature ,in, This refers to the terminal voltage of the lithium battery.

9. A multi-feature online health state estimation device for lithium batteries, characterized in that: include: Feature acquisition module: used to obtain the charging time in constant current charging mode as the first feature through aging cycle experiments of lithium batteries. The initial voltage of the constant current charging mode is used as the second characteristic. The cumulative charging voltage in constant current charging mode is used as a third feature. The initial slope of the constant current charging mode is used as the fourth feature. The average voltage of lithium batteries is the fifth characteristic. The slope at the end of the constant current charging mode is used as the sixth feature. ; Model building module: used to input vectors from the first to the sixth features. The hypothesis function formula for establishing a multiple linear regression model is as follows: And the gradient descent algorithm is used to initially determine the parameter vector. , For intercept, parameter The regression coefficients correspond to the first through sixth features. This is the valuation of the health status of lithium batteries; Parameter update module: used to update parameters based on the parameter vector. Randomly generated Each initial particle is initialized and updated using Gaussian white noise. Then, importance sampling and resampling are performed to update the parameter vector and the predicted values ​​of the multiple linear regression model. The parameter update module is implemented according to the following steps: based on the parameter vector Randomly generated Initialized particles , ;based on Update particles ,in, It is Gaussian white noise. This refers to the number of particle updates; the weights are obtained by importance sampling according to the following formula. Normalize according to the following formula. ,when If this occurs, resampling will be performed, where... For standard deviation, These are reference values ​​for health status. , The parameter vector is updated using the following formula. and the predicted values ​​of the multiple linear regression model : .

Images