A method for predicting the life of lithium-ion batteries based on capacity regeneration phenomenon

Abstract

The present invention discloses a lithium-ion battery life prediction method for capacity regeneration phenomenon, and belongs to the technical field of battery life prediction. The present invention is aimed at the fact that the existing technology cannot predict the battery life for the battery capacity regeneration phenomenon. The present invention obtains the health factor data and cycle number of the lithium-ion battery; decomposes the health factor into a first type of modal component and a second type of modal component using a variational mode decomposition method; inputs the first type of modal component into a first prediction model, and inputs the second type of modal component into a second prediction model to obtain a health factor prediction value; inputs the health factor prediction value and the cycle number into a degradation relationship model to obtain a capacity prediction value of the lithium-ion battery, and then obtains the life value of the lithium-ion battery. The present invention solves the problem that the capacity regeneration fluctuation of battery data during the degradation process of lithium-ion batteries leads to weak generalization ability and low prediction accuracy of traditional data-driven methods.

Description

Technical Field

[0001] The present invention relates to the field of lithium ion battery life prediction, and in particular to a lithium ion battery life prediction method targeting capacity regeneration phenomenon. Background Art

[0002] Lithium-ion batteries have the advantages of high energy density, long cycle life, safety and reliability, and are widely used in mobile electronic devices, medical equipment, transportation, power grid energy storage systems and other fields, and are gradually expanding to military communications, aerospace and other fields. Battery health status prediction is an important part of the battery management system. Accurately predicting the battery life is of key significance to the battery and the entire system.

[0003] The data-driven method is the mainstream prediction method at present. It collects data during the battery charging and discharging process and analyzes the implicit degradation rules to make predictions. It does not need to understand the complex chemical mechanism inside the battery, and the model building is simpler. Commonly used basic methods include support vector machine, relevance vector machine, neural network, autoregressive model algorithm, Gaussian process regression (Gaussian Process Regression, GPR), etc. Support vector machine is the most commonly used method in the field of lithium-ion battery life prediction, but its computational complexity is high and it is sensitive to the selection of parameters and kernel functions, which easily leads to low prediction accuracy. As the network structure of the neural network changes, network training becomes difficult and it is easy to overfit. Most of the Gaussian process regression models in existing literature use the conjugate gradient descent algorithm. In the case of small samples, it is highly dependent on the initial value, and there is no theoretical basis for the initial value setting. Therefore, only a single basic algorithm is used, which has problems such as weak generalization ability and low prediction accuracy.

[0004] At present, data-driven methods based on fusion are gradually being applied in the field of lithium-ion battery life prediction. Statistical methods, such as particle filtering algorithms, are used to integrate with other single basic algorithms. However, there are problems of particle degradation and sample depletion in the particle filtering algorithm, which affect the accuracy of the prediction. In addition, in view of the nonlinear and non-stationary characteristics of battery data, existing methods use wavelet denoising methods to process data. However, they are only simple denoising of the original data, and do not fully consider the impact of complex characteristics in the battery degradation process on the prediction of lithium-ion battery life, that is, the capacity regeneration phenomenon in the process of lithium-ion battery decline. The use of variational mode decomposition algorithm can overcome the shortcomings of the wavelet denoising algorithm, but the mode aliasing phenomenon is serious, resulting in low prediction accuracy.

[0005] In view of the problem that direct prediction of battery life is difficult to achieve, it is very important to find a health factor to replace the battery degradation state. The traditional data prediction method based on battery voltage, current, and capacity has a large amount of data and a lot of redundant information, resulting in poor prediction accuracy. At present, most of the battery life predictions use capacity data as a direct health factor to establish a mapping relationship between the number of battery cycles and battery capacity. Although the number of battery cycles is correlated with capacity, the generalization performance of the model is not strong and the prediction accuracy is poor. When there is less training data, it can only guarantee good short-term prediction results, and the long-term prediction results are very poor. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a lithium-ion battery life prediction method for the capacity regeneration phenomenon, which solves the problem that the capacity regeneration fluctuation of battery data during the degradation process of lithium-ion batteries leads to weak generalization ability and low prediction accuracy of traditional data-driven methods.

[0007] The present invention provides a lithium ion battery life prediction method for capacity regeneration phenomenon, comprising the following steps:

[0008] S1. Obtain health factor data and cycle number data of lithium-ion batteries;

[0009] S2. Decomposing the health factor into a first type modal component and a second type modal component by using a variational modal decomposition method, wherein the first type modal component reflects the overall degradation trend of the lithium-ion battery, and the second type modal component reflects the capacity regeneration phenomenon of the lithium-ion battery;

[0010] S3, inputting the first type of modal components into a first prediction model, inputting the second type of modal components into a second prediction model, and obtaining a health factor prediction value according to an output result of the first prediction model and an output result of the second prediction model;

[0011] S4. Input the predicted value of the health factor and the number of cycles into a degradation relationship model to obtain a predicted value of the capacity of the lithium-ion battery, and then obtain a life value of the lithium-ion battery.

[0012] Furthermore, the health factor is the discharge time with constant voltage drop.

[0013] Furthermore, the first prediction model is a kernel function k Matern (r) Gaussian regression model, the kernel function k Matern (r) is:

[0014]

[0015] The second prediction model is a kernel function k PER(r) Gaussian process regression model, the kernel function k PER (r) is:

[0016]

[0017] In the formula, r = ||xx′||, δ is the sample variance, l and v are the hyperparameters of the kernel function, K v is the Bessel function and p is the period.

[0018] Furthermore, the method for constructing the degradation relationship model includes:

[0019] A Gaussian process regression model is established, wherein the hyperparameters of the Gaussian process regression model include an inertia weight factor ω, a learning factor c 1 and c 2 ;

[0020] The hyperparameters of the Gaussian process regression model are optimized by using a dynamic adaptive immune particle swarm algorithm to obtain a degradation relationship model, wherein the input data of the degradation relationship model are the health factor and the number of cycles, and the output is the capacity of the lithium-ion battery.

[0021] Furthermore, the inertia weight factor ω optimized by the dynamic immune adaptive particle swarm algorithm is:

[0022]

[0023] Among them, k max is the maximum number of iterations, k is the number of iterations, ω max is the maximum value of the inertia weight factor, ω min is the minimum value of the inertia weight factor.

[0024] Furthermore, the learning factor c optimized by the dynamic immune adaptive particle swarm algorithm is 1 and c 2 They are:

[0025]

[0026]

[0027] Among them, c 1max , c 1min , c 2max , c 2min They are c 1 、c 2 The maximum and minimum values ​​of .

[0028] As described above, compared with the prior art, the present invention has the following effects:

[0029] 1. The present invention extracts the isovoltage drop discharge time series as a health factor for predicting RUL, which has good effects in both the short and long term.

[0030] 2. The present invention adopts the VMD algorithm to enable the model to fully learn the intrinsic information of the data, and the prediction model established by using different kernel functions is more purposeful. Among them, the Matern kernel function can better predict the linearly changing data component IMF1, and the periodic kernel function can better predict the periodically changing data component IMF2~IMF5, so that the prediction model can accurately capture the long-term Xiangjiang death and short-term regeneration phenomena, and improve the prediction accuracy.

[0031] 3. The improved particle swarm algorithm of the present invention can search in a gradually enhanced manner from global to local, has nonlinear adaptive change capability, and has stronger convergence, can find the global optimal solution more accurately, and has better optimization effect than the conjugate gradient algorithm in the prior art, and finally obtains a better prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The present invention is based on the DAIPSO-GPR model flow chart;

[0033] Figure 2 This is a schematic diagram of equal voltage drop discharge time extraction of the present invention;

[0034] Figure 3 is the decomposition result of the variational modal decomposition of the B0005 battery health factor of the present invention, Figure 3 a is the modal component decomposition result, Figure 3 b is the spectrum of the modal component;

[0035] Figure 4 It is the prediction result of each modal component of the health factor of batteries B0005, B0006, B0007 and B0018 of the present invention; Figure 4 a is the prediction result of IMF1 component of B0005 battery, Figure 4 b is the prediction result of IMF1 component of B0006 battery, Figure 4 c is the prediction result of IMF1 component of B0007 battery, Figure 4 d is the prediction result of IMF1 component of B0018 battery, Figure 4 e is the prediction result of the detailed component of 0005 battery, Figure 4 f is the prediction result of the B0006 battery detail component, Figure 4 g is the prediction result of B0007 battery detail component, Figure 4 h is the prediction result of the B0018 battery detail component;

[0036] Figure 5 is a flow chart of DAIPSO optimization of GPR of the present invention;

[0037] Figure 6 This is a result diagram of the lithium-ion battery life prediction based on VMD-DAIPSO-GPR in the present invention. Figure 6 a is the prediction result of B0005 battery, Figure 6 b is the prediction result of B0006 battery, Figure 6 c is the prediction result of B0018 battery, Figure 6 d is the prediction result of B0007 battery;

[0038] Figure 7 This is the error diagram of lithium-ion battery life prediction based on VMD-DAIPSO-GPR in the present invention. Figure 7 a is the prediction error of B0005 battery, Figure 7 b is the prediction error of B0006 battery, Figure 7 c is the prediction error of B0018 battery, Figure 7 d is the prediction error of B0007 battery; DETAILED DESCRIPTION

[0039] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0040] It should be noted that the illustrations provided in the following embodiments are only used to illustrate the basic concept of the present invention in a schematic manner, and thus the illustrations only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0041] like Figure 1 As shown, this embodiment provides a lithium-ion battery life prediction method for capacity regeneration phenomenon, comprising the following steps:

[0042] S1. Obtain health factor data and cycle number data of a lithium-ion battery, wherein the health factor (Health Indicator, HI) is the discharge time with constant voltage drop.

[0043] In a specific embodiment, a lithium battery cycle aging test experiment was conducted on four batteries (B0005, B0006, B0007, B0018), and data was recorded for the entire life cycle of the four batteries. The experiment was charged at a current of 0.75C. When the battery voltage reached 4.2V, it was switched to constant voltage charging until the charging current dropped below 20mA. The discharge test current was 1C until the battery voltage dropped to 2.7V, 2.5V, 2.2V and 2.5V respectively.

[0044] As the battery ages, its actual capacity tends to decrease overall, but there is a capacity regeneration phenomenon, so the discharge voltage curve of the lithium-ion battery will change with the cycle. The greater the number of cycles, the smaller the SoH and the shorter the discharge time. The time difference from a higher voltage value to a lower voltage in each discharge stage of the lithium-ion battery is called the equal voltage drop discharge time. The battery has the longest discharge time near the nominal voltage. The nominal voltage of the lithium-ion battery is selected to be 3.6V to 3.7V. Combined with the actual discharge voltage curve, the equal voltage drop discharge range selected in this embodiment is [3.5, 3.8]. Figure 2 The figure shows the schematic diagram of the discharge time of the first discharge voltage curve of the B0005 lithium-ion battery with constant voltage drop. In order to further analyze and verify the correlation between capacity and HI, the Pearson correlation coefficient is used to calculate the correlation coefficient r, which is:

[0045]

[0046] Among them, x i represents the discharge time with constant voltage drop, y i Represents the battery capacity. Table 1 lists the correlation coefficients between HI and capacity of the four batteries.

[0047] Table 1 Correlation coefficients of discharge time and capacity of lithium-ion batteries with equal voltage drop

[0048]

[0049] It can be seen from the table that there is a strong correlation between the extracted HI and the capacity. Therefore, the use of the equal pressure drop time series as the input data of the lithium-ion battery RUL prediction model in this embodiment is reliable.

[0050] The number of cycles, equal voltage drop discharge time, and capacity data are combined to form training data and test data.

[0051] S2. Decomposing the health factor into a first type modal component and a second type modal component by using a variational modal decomposition method, wherein the first type modal component reflects the overall degradation trend of the lithium-ion battery, and the second type modal component reflects the capacity regeneration phenomenon of the lithium-ion battery;

[0052] Variational mode decomposition (VMD) is an adaptive, completely non-recursive method of modal variation and signal processing. It solves the problems of endpoint effects and modal component aliasing in empirical mode decomposition, and has a more solid mathematical theoretical basis. It can reduce the non-stationarity of time series with high complexity and strong nonlinearity, and decompose to obtain relatively stable subsequences containing multiple different frequency scales, which is suitable for non-stationary sequences.

[0053] In this embodiment, the method of decomposing the health factor into the first type of modal component and the second type of modal component by using the variational modal decomposition method includes:

[0054] S21. Establish a variational mode decomposition model as follows:

[0055]

[0056] In the formula, represents the partial derivative with respect to time t, δ(t) is the impulse function, u n (t) is the modal component, ω n is the center frequency of each component.

[0057] S22, solving the variational mode decomposition model;

[0058] In order to ensure the reconstruction accuracy of the signal and strengthen the effect of the constraint conditions, the quadratic penalty term and the Lagrange multiplication operator are introduced, and the augmented Lagrangian function expression is as follows:

[0059]

[0060] Where λ is the Lagrange multiplication operator and α is the quadratic penalty term.

[0061] The alternating direction multiplier method (ADMM) is used to solve the above variational problem and find the saddle point of the augmented Lagrangian function. Update by formula (5) and formula (5)

[0062]

[0063]

[0064] Update λ according to formula (6) k+1 :

[0065]

[0066] Until it converges to satisfy the formula: When , the iteration stops and n modal components are obtained. u n(t), f(t) are the results of Fourier isometric transformation.

[0067] S23. In this embodiment, the health factor data obtained in step S1 is subjected to variational modal decomposition by the above method to obtain five modal components IMF1 to IMF5 with different variation patterns, such as Figure 3 As shown, it is the decomposition result of B0005 battery. Figure 3 a is the time domain curve of the modal component, Figure 3 b is the spectrum diagram of each modal component. It can be seen from the figure that the low-frequency component IMF1 can better show the overall trend of battery degradation and has monotonic stability. Other modal components show the curve fluctuation caused by the capacity regeneration phenomenon and show a certain periodicity. The modal component IMF1 is monotonic and stable, reflecting the overall degradation trend of the battery. Therefore, the modal component IMF1 is set as the first type of modal component, and the components IMF2~IMF5 that reflect the capacity regeneration phenomenon and show a certain periodicity are set as the second type of modal components.

[0068] S3, inputting the first type of modal components into a first prediction model, inputting the second type of modal components into a second prediction model, and obtaining a health factor prediction value according to an output result of the first prediction model and an output result of the second prediction model;

[0069] The first prediction model is a Gaussian process regression (GPR) prediction model of the degradation trend component, which takes the first type of modal component as input to obtain the first health factor prediction value. The kernel function of the first prediction model is the Matern covariance function, that is, formula (7);

[0070]

[0071] The second prediction model is a Gaussian process regression (GPR) prediction model of detail components, which takes the second type of modal component as input to obtain a second health factor prediction value. The kernel function of the second prediction model is a periodic covariance kernel function, that is, formula (8):

[0072]

[0073] In the formula, r = ||xx′||, δ is the sample variance, l and v are the hyperparameters of the kernel function, K v is the Bessel function and p is the period.

[0074] The first health factor prediction value and the second health factor prediction value are superimposed to obtain the final health factor prediction value, such as Figure 4 As shown, the prediction results of each modal component of B0005 battery.

[0075] The method for constructing the Gaussian process regression model includes:

[0076] The basic theory is Bayesian linear regression, and the sample set is X={x 1 , x 2 , …, x N}, its basic model is formula (9)

[0077] f(x)=x T w,y=f(x)+ε=x T w+ε (9)

[0078] Among them, x is the input vector, w is the weight vector, f represents the functional relationship, y is the observation value, and ε is the noise, which obeys the Gaussian distribution, that is,

[0079] The posterior probability is obtained by using the prior probability and likelihood function. Formula (10) is the Bayesian formula:

[0080]

[0081] Among them, the likelihood function is:

[0082]

[0083] In Bayesian theory, the prior probability distribution of w is given in advance:

[0084] p(w)=N(0,∑ p ) (12)

[0085] It also satisfies the Gaussian distribution, so the posterior probability is:

[0086] p(w|X,y)∝p(y|X,w)p(w) (13)

[0087] Through the concept of conjugation, we can know that p(w|X, y) also follows a Gaussian distribution. When the posterior probability is known, given the input x ＊ , then regression prediction output can be performed, f ＊ Represents the predicted value, and the conclusion is

[0088]

[0089] In formula (14), w is the weight vector of the linear regression model, and the mean Covariance∑ w =A -1 ,

[0090] When the target is a nonlinear problem, the kernel function method is introduced and Φ(X) is used to represent the original X, that is, Therefore, the probability of the output value of (11) can be further deduced as follows:

[0091]

[0092] Let the covariance function K = Φ T ∑ P Φ,K(x ＊ , X) = Φ(x ＊ )∑ p Φ(X)=K ＊ , substituted into the above formula (15), that is, expressed in the form of kernel function, the final predicted mean of GPR is The prediction variance is Where I is the identity matrix.

[0093] The hyperparameters are obtained by maximum likelihood estimation. In this embodiment, the log-likelihood function is used, which is expressed as:

[0094]

[0095] For the NASA battery data set, there are 168 sets of data for batteries B0005, B0006, and B0007, and 132 sets of data for battery B0018. The first 50% of the data for the four batteries is used as the training set for the model, that is, groups 1-80 of batteries B0005, B0006, and B0007 are used as the training set, and groups 1-60 of batteries B0018 are used as the training set. The remaining data is used as the test set to predict the RUL of the battery. Figure 4 Shown are the prediction results of each modal component of B0005, B0006, and B0007 batteries.

[0096] S4. Input the predicted value of the health factor into the degradation relationship model to obtain the predicted value of the capacity of the lithium-ion battery, and then obtain the life value of the lithium-ion battery.

[0097] The degradation relationship model is obtained by optimizing the hyperparameters of the Gaussian process regression model using a dynamic immune adaptive particle swarm algorithm. The degradation relationship model takes the health factor prediction value obtained in step S3 as input, and outputs the battery capacity prediction value.

[0098] like Figure 5 As shown, the method for establishing the degradation relationship model includes:

[0099] S41. Establish a Gaussian process regression (GPR) model and determine the kernel function;

[0100] The Gaussian process regression (GPR) model is:

[0101]

[0102] The sample set is X={x 1, x 2 , …, x N}, f ＊ represents the predicted value, w is the weight vector, and the mean Covariance∑ w =A -1 ,

[0103] Introduce the kernel function and use Φ(X) to represent the original x, that is, Therefore, the above GPR model becomes:

[0104]

[0105] Let the covariance function K = Φ T ∑ P Φ,K(x ＊ , X) = Φ(x ＊ )∑ p Φ(X)=K ＊ , put it into the above formula, that is, express it in the form of kernel function, and finally get the predicted mean of GPR The prediction variance is Where I is the identity matrix.

[0106] The kernel function is the Matern covariance function, that is:

[0107]

[0108] S42, initialize the particle population and the hyperparameters of the GPR model, and set the maximum number of iterations; in this embodiment, the number of particle swarms is set to 20, the number of iterations is set to 100, the value range of the hyperparameters in the kernel function of the GPR model is set to [-10^3, 10^3], and the maximum inertia weight ω is set to max =0.9, minimum inertia weight ω min =0.4, maximum learning factor c 1max =c 2max =2.2, minimum learning factor c 1min =c 2min =0.2.

[0109] S43, input the training set to train the GPR model;

[0110] S44, update the particle position and speed, and calculate the particle fitness to obtain the historical optimal position of a single particle and the global optimal solution of all particles in this iteration process; determine whether the stopping optimization condition is reached. If the stopping condition is reached, the GPR model with the optimal hyperparameters is obtained. If not, the hyperparameters of the GPR model are optimized using the improved immune particle swarm algorithm. After obtaining the GPR model with new hyperparameters, step S44 is executed. After 100 iterations, the optimal solution of each hyperparameter within the range is determined to obtain a trained degradation relationship model.

[0111] The updating method of the particle's velocity and position is shown in formulas (17) and (18):

[0112] v id =wv id +c 1 r 1 (p id -x id )+c 2 r 2 (p gd -x id ) (17)

[0113] x id =x id +v id (18)

[0114] The optimization of the hyperparameters of the Gaussian process regression model includes the optimization of the inertia weight and the optimization of the learning factor, so as to realize the dynamic adaptive search capability. The optimization method of the inertia weight ω is realized by formula (19):

[0115]

[0116] Among them, k max is the maximum number of iterations, k is the number of iterations;

[0117] The learning factor c 1 、c 2 They represent the coefficients of the particle tracking itself and the optimal historical value of the group respectively. The optimization method is shown in equations (20) and (21):

[0118]

[0119]

[0120] Among them, c 1max , c 1min , c 2max , c 2min They are c 1 、c 2 The maximum and minimum values ​​of .

[0121] The concentration selection mechanism of the immune particle group is shown in formula (22), which optimizes the hyperparameters of the Gaussian process regression model.

[0122]

[0123] S46. Input the test set into the GPR model with the optimal hyperparameters and verify the prediction results.

[0124] Methods for obtaining the life value of a lithium-ion battery based on the capacity prediction value of the lithium-ion battery include:

[0125] A SOH prediction curve for lithium-ion batteries, that is, a lithium-ion battery life prediction result curve, is established. According to existing data statistics, the failure threshold of lithium-ion batteries is generally 70% of their rated capacity. Therefore, in this embodiment, the failure threshold is set to 70% of the lithium-ion battery capacity, and the cycle corresponding to the failure is the failure life. In the prediction curve, the cycle difference from the prediction point to the failure life is the remaining life of the battery.

[0126] In this embodiment, the life prediction result curves of the four types of batteries are as follows: Figure 6 As shown in the figure, it is generally believed that the failure threshold of lithium-ion batteries is 70% of the rated capacity. In this data set, the failure threshold is 1.4Ah, and the cycle corresponding to the failure threshold is the failure life (End of Life, EoL). The failure life in the prediction curve is the predicted failure life (End of Prediction, EoP). The RUL of a battery refers to the cycle difference from the prediction point to the failure life. Figure 7 Absolute error plot of the prediction results for each battery lifetime prediction.

[0127] RMSE and MAPE are used as rating indicators for statistical analysis. Table 2 is a quantitative representation of the prediction error of the model. The maximum RMSE of the prediction error of the four batteries optimized by DAIPSO GPR is 1.53%, the minimum is 1.36%, the maximum MAPE is 0.77%, the minimum is 0.63%, and the RUL errorThe error does not exceed 2. The quantitative analysis of the error proves that the method proposed in this paper has a high prediction accuracy. From the perspective of the algorithm, first of all, this paper uses the VMD algorithm to reduce the complexity of the data, so that the model can easily and fully learn the intrinsic information of the data. The VMD algorithm decomposes HI and obtains the component IMF1 that can reflect the overall downward trend and the component IMF2~IMF5 that reflects the capacity regeneration fluctuation. The prediction model established with different covariance kernel functions is more purposeful. Among them, the Matern kernel function can better predict the linearly changing data component IMF1, and the periodic kernel function can better predict the periodically changing data component IMF2~IMF5. Therefore, the model can capture the capacity regeneration fluctuation well and obtain a more accurate prediction result. Secondly, compared with the gradient descent method used in the traditional model, the improved particle swarm algorithm can search in a gradually enhanced manner from global to local, has nonlinear adaptive change capabilities, and has stronger convergence, and can more accurately find the global optimal solution. In summary, from the perspective of error quantitative analysis and algorithm analysis, it can be proved that the battery life prediction method of this embodiment can perform well in RMSE and MAPE.

[0128] Table 2 Model prediction error

[0129]

[0130] The above embodiments are merely illustrative of the principles and effects of the present invention, and are not intended to limit the present invention. Anyone familiar with the art may modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by a person of ordinary skill in the art without departing from the spirit and technical concept disclosed by the present invention shall still be covered by the claims of the present invention.

Claims

1. A method for predicting the life of a lithium-ion battery for capacity regeneration, characterized in that: The steps include: S1. Obtaining health factor data and cycle number of a lithium-ion battery, wherein the health factor is a discharge time with a constant voltage drop; S2. Decomposing the health factor into a first type modal component and a second type modal component by using a variational modal decomposition method, wherein the first type modal component reflects the overall degradation trend of the lithium-ion battery, and the second type modal component reflects the capacity regeneration phenomenon of the lithium-ion battery; S3, inputting the first type of modal components into a first prediction model, inputting the second type of modal components into a second prediction model, and obtaining a health factor prediction value according to an output result of the first prediction model and an output result of the second prediction model; The first prediction model is a kernel function Gaussian regression model, the kernel function for: ； The second prediction model is a kernel function Gaussian process regression model, the kernel function for: ； In the formula, , is the sample variance, , is the hyperparameter of the kernel function, is the Bessel function, For the cycle; S4. Input the predicted value of the health factor and the number of cycles into a degradation relationship model to obtain a predicted value of the capacity of the lithium-ion battery, and then obtain a life value of the lithium-ion battery.

2. A lithium-ion battery life prediction method for capacity regeneration according to claim 1, characterized in that: The method for constructing the degradation relationship model includes: A Gaussian process regression model is established, wherein the hyperparameters of the Gaussian process regression model include an inertia weight factor , learning factor and ; The hyperparameters of the Gaussian process regression model are optimized by using a dynamic adaptive immune particle swarm algorithm to obtain a degradation relationship model, wherein the input data of the degradation relationship model are the health factor and the number of cycles, and the output is the capacity of the lithium-ion battery.

3. A lithium-ion battery life prediction method for capacity regeneration according to claim 2, characterized in that: The dynamic immune adaptive particle swarm algorithm optimizes the inertia weight factor for: ； in, is the maximum number of iterations, k is the number of iterations, is the maximum value of the inertia weight factor, is the minimum value of the inertia weight factor.

4. A lithium-ion battery life prediction method for capacity regeneration according to claim 3, characterized in that: The learning factor optimized by the dynamic immune adaptive particle swarm algorithm and They are: ； ； in, , , , They are , The maximum and minimum values ​​of .

Images