WO2024016496A1

WO2024016496A1 - Method and apparatus for estimating soh of lithium battery

Info

Publication number: WO2024016496A1
Application number: PCT/CN2022/125898
Authority: WO
Inventors: 许恩永; 郑伟光; 曹成虎; 冯高山; 王善超; 徐小红; 郭葵; 覃记荣; 姜峰; 何水龙
Original assignee: 东风柳州汽车有限公司; 广西科技大学; 桂林电子科技大学
Priority date: 2022-07-20
Filing date: 2022-10-18
Publication date: 2024-01-25
Also published as: CN115219938A

Abstract

A method for estimating the SOH of a lithium battery. The method comprises the following steps: collecting data of a lithium battery, extracting health factors to construct feature vectors, and generating training samples and test samples (S100); determining an estimation algorithm for the SOH of the lithium battery to be an ant lion optimization algorithm, and defining algorithm parameters and an output parameter group (S110); improving the estimation algorithm to output an optimal output parameter group (S120), wherein the improvement refers to controlling search balance in different iteration stages by means of adjusting weight values corresponding to the random walking of an elite ant lion and an ordinary ant lion, so as to output the optimal output parameter group; and on the basis of the optimal output parameter group, predicting a test sample set by means of a support vector regression model, so as to output an estimated value of the SOH (S130). By using the estimation method, outputting an optimal parameter group with fewer iterations can be supported, and the generalization capability and the fitting capability for model training are improved with a lower cost, such that the SOH of a lithium ion battery is estimated accurately in real time, and the accuracy and the convergence precision are improved.

Description

A method and device for estimating SOH status of lithium batteries

Technical field

The present invention relates to the field of battery technology, and specifically, to a method and device for estimating the SOH state of a lithium battery.

Background technique

Lithium-ion batteries are one of the most widely used energy storage devices at present. As they are continuously charged and discharged, the battery capacity will continue to decrease and the battery will continue to age. This indicator of battery aging is SOH (state of health). , battery health status). Monitoring the battery SOH status can know the battery aging status in time to facilitate timely maintenance of the battery pack and battery replacement. In practical applications, SOH is difficult to measure directly through sensors and requires characteristic parameters to characterize. The capacity of lithium-ion batteries is widely considered to be an indicator of SOH. Currently, there are three main SOH estimation methods used in the existing technology: direct measurement method, model-based method and data-driven method. Direct measurement methods mainly include ampere-hour integration method and coulomb counting method. By directly measuring battery capacity, internal resistance and other aging characteristics offline, the estimated SOH value is calculated by formula. This measurement method has very limited application scenarios and cannot be measured online in actual battery application scenarios. It can only be measured offline using complex equipment in a laboratory environment. The model-based method comprehensively considers battery material characteristics, internal chemical mechanisms and load conditions to construct a battery aging model for SOH estimation, which mainly includes electrochemical models, equivalent circuit models and empirical degradation models. However, model-based methods are relatively complex and easily interfered by external dynamic factors, resulting in low accuracy and weak robustness. The data-driven method does not need to consider electrochemical reactions, complex external factors and complex models. It only needs to mine battery degradation rules from battery cycle data, find health factors that can map SOH, establish non-linear mapping relationships, and achieve accurate SOH estimate. The current main data-driven methods are: 1) BP (Back Propagation Neural Network) neural network estimates SOH; 2) SOH estimation method based on Ant Lion Optimization Algorithm Support Vector Regression (ALO-SVR).

Although the SOH estimation method based on the support vector regression (ALO-SVR) of the Ant Lion optimization algorithm overcomes the accuracy and robustness problems of direct measurement methods and model-based methods, it is prone to falling into local minima during the iterative process. The problem of optimal solutions leads to large errors in estimation results, certain limitations in the scope of application, and the convergence accuracy of the algorithm is not high enough.

Contents of the invention

The purpose of the invention is to solve the current technical problems of large error in ALO-SVR estimation results, high requirements on the number of iterations, and limited application scope.

In the first aspect, in order to achieve the above purpose, this application provides a method for estimating the SOH state of lithium batteries, which includes the following steps:

Lithium battery data collection, extract health factors to construct feature vectors, generate training samples and test samples, where health factors are elements related to the battery capacity decline trend; health factors include: equal voltage rise time, average discharge voltage and equal voltage drop time; training The proportions of samples and test samples are 60% and 40% respectively.

Determine the prediction method for the SOH state of the lithium battery, define the algorithm parameters and output parameter group; determine the prediction method as the Ant Lion optimization algorithm;

Improve the prediction algorithm and output the optimal output parameter group. The improvement specifically refers to: by adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions, control the search balance in different iteration stages, and output the optimal output parameter group, corresponding to different The search balance in the iterative stage is: global search is the main one in the early and middle stages of the iteration, and local search is the main one in the late iteration stage;

Combined with the optimal output parameter group, the test sample set is predicted through a support vector regression model, and the SOH estimated value is output.

Among them, the definition algorithm parameters include:

The population size of ants and ant lions is set to Agents_no=30,

The dimension of the fitness function is set to dim=2,

The maximum number of iterations is set to Max_iter=100,

The output parameter group is a space vector (c, σ) representing the positions of the ants and ant lions. The upper limit of c and σ is set to 100, and the lower limit is set to 0.01.

Among them, the search balance in different iteration stages is controlled by the corresponding weight values of the random walks of elite ant lions and ordinary ant lions: the inertial weight γ of ordinary ant lions is nonlinearly dynamically adjusted according to the number of iterations. The calculation method is:

Among them, γ _max is the maximum value of inertia weight, and γ _min is the minimum value of inertia weight.

Furthermore, before outputting the optimal output parameter group, it also includes calculating the fitness value and sorting the moderate values of ants and ant lions. The maximum value is set to the elite ant lion, and the corresponding parameter is the optimal output parameter group. Among them, the fitness value of The calculation method is the fitness function between the actual value and the predicted value of the mean square error (MSE), and its expression is:

in

is the i-th predicted value, yi is the _i -th actual value, and n is the number of training sets.

Further, prediction of the test sample set through the support vector regression model includes prediction result analysis, including using absolute error and root mean square error as evaluation criteria. The calculation method is as follows:

Among them: MAE is the mean absolute error, RMSE is the root mean square error,

and _yi are the predicted value and actual value of the i-th SOH respectively, and n is the number of samples.

This application also provides a device for estimating the SOH status of lithium batteries, including:

Data processing module: used to collect lithium battery data, extract health factors to construct feature vectors, and generate training samples and test samples; the health factors collected by the data processing module include: equal voltage rise time, average discharge voltage and equal voltage drop time; data processing The proportions of module control samples and test samples are 60% and 40% respectively.

Algorithm control module: Prediction algorithm for determining the SOH state of lithium batteries, defining algorithm parameters and output parameter groups; defining algorithm parameters includes defining the population number of ants and ant lions, the dimensionality of the fitness function, and the maximum number of iterations; output parameters The group includes the space vector (c, σ), the upper limit of c and σ is set to 100, and the lower limit is set to 0.01;

Algorithm optimization module: used to improve the prediction algorithm and output the optimal output parameter group. The improvements include adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions;

Model training and data output module: used to predict the test sample set through a support vector regression model and output the SOH estimate.

Furthermore, the calculation method for adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions defined by the algorithm optimization module is:

Further, the algorithm control module includes a fitness value control unit, which is used to calculate the fitness value for the output of the optimal output parameter group, sort the fitness values of ants and ant lions, set the maximum value to the elite ant lion, and set the corresponding parameters to the best Optimal output parameter group, in which the calculation method of fitness value supports the fitness function between the mean square error (MSE) actual value and the predicted value, and its expression is:

in

According to the present invention, the optimized ant lion algorithm is used to optimize the support vector regression (IALO-SVR), support fewer iterations to output the optimal parameter group, and improve the generalization ability and fitting of model training at less cost. Ability to achieve accurate and real-time estimation of lithium-ion battery SOH, improving accuracy and convergence precision.

Description of drawings

Figure 1 is a step diagram of a method for estimating the SOH state of a lithium battery according to an embodiment of the present invention;

Figure 2 is a logic flow chart of a method for estimating the SOH state of a lithium battery provided according to an embodiment of the present invention;

Figure 3 is a schematic diagram of the ordinary ant lion inertia weight adjustment process in the method for estimating the SOH state of lithium batteries provided according to an embodiment of the present invention;

Figure 4 is a schematic diagram of the change curve of the actual SOH of the B05 battery and three health factors in the method for estimating the SOH state of the lithium battery provided according to an embodiment of the present invention;

Figure 5 is a comparison diagram of SOH estimation results using different methods in the estimation method of lithium battery SOH state provided according to an embodiment of the present invention;

Figure 6 is a comparison diagram of SOH estimation errors using different methods in the estimation method of lithium battery SOH state provided according to an embodiment of the present invention;

Figure 7 is a comparison diagram of the SOH estimation mean absolute error (MAE) and root mean square error (RMSE) using different methods in the method for estimating the SOH state of lithium batteries provided according to an embodiment of the present invention;

Figure 8 is a structural diagram of a device for estimating the SOH state of a lithium battery provided according to an embodiment of the present invention.

Detailed ways

The technical basis of this application is the application of the mathematical model of vector regression: collecting data from lithium batteries for model training (SVR), and then outputting the prediction of the SOH state of the lithium batteries through correlation analysis. During the process of model training, the model needs to be continuously optimized. The current mature approach is to construct the Ant Lion Optimization (ALO) algorithm, through which the algorithm outputs a parameter set to provide optimal parameters for the vector regression training model. In the ALO algorithm, ordinary ant lions mainly affect the global search ability of the algorithm, and elite ant lions mainly affect the local optimization ability of the algorithm. When updating the ant position, the weight of both is 0.5. The optimization accuracy controlled by this algorithm cannot meet the requirements.

Therefore, this application optimizes the ALO algorithm. The main idea is to adjust the global search capability and local optimization capability in different periods: focus on the global search capability in the early and middle stages of the iteration to increase the diversity of the population and avoid falling into the local minimum. Optimal, and a stronger local search is performed in the later stages of the iteration to improve the convergence speed, and ultimately achieve the output of optimal parameters with fewer iterations. The SOH state of the lithium battery is estimated through the optimized Ant Lion algorithm optimized support vector regression (IALO-SVR).

The specific implementation manner of the present invention will be described in detail below with reference to the accompanying drawings.

Figure 1 is a step diagram of a method for estimating the SOH state of a lithium battery provided by an embodiment of the present application. Combined with it, Figure 2 is a specific logic flow chart. As shown in the figure, it includes the following steps:

Step S100: Collect lithium battery data, extract health factors to construct feature vectors, and generate training samples and test samples;

This step is shown as step S210 in Figure 2, which is the sample processing stage.

First of all, in terms of lithium battery data collection, health factors related to the decline trend of lithium battery capacity can be extracted as samples through charge and discharge data such as voltage, current, and internal resistance. In the scheme proposed by this application, the data collected can be: equal voltage rise time, average discharge voltage and equal voltage drop time. These three data can well reflect the decline of SOH, as shown in Figure 4:

HF1: Equal voltage rise time. The relationship between the equal voltage rise time and the actual SOH is shown in the P410 part of Figure 4. As the number of cycles increases, the downward trends of the two are roughly the same.

HF2: Average discharge voltage. The P420 part of Figure 4 shows the changing trend of the average discharge voltage and actual SOH. There is a strong correlation between the actual SOH and the number of cycles in which the average discharge voltage is above 80%, so the average discharge voltage can be used as a health factor for SOH estimation.

HF3: With equal voltage drop time, the voltage of 4.2v is discharged to the cut-off voltage with a constant current. The comparison between the time taken and the actual SOH is shown in the P430 part of Figure 4. The two have very similar downward trends.

Therefore, in practical applications, one of them can be selected as a data sample according to the actual situation.

In step S212 shown in Figure 2, the sample set is classified. For the collected data, the first 60% of the samples are used as the training sample set, and the last 40% are used as the test sample set.

Step S110: Determine the prediction algorithm for the SOH state of the lithium battery, and define the algorithm parameters and output parameter group; this step is shown in step S220 in Figure 2.

First, in this step, the algorithm parameters and output parameter group are also determined:

The algorithm used in this application is the ant lion optimization algorithm. The definition of the algorithm parameters includes setting the population number of ants and ant lions to Agents_no=30, the dimension of the fitness function to dim=2, and the maximum number of iterations to Max_iter= 100, the upper and lower limits of c and σ are set to 100 and 0.01 respectively;

The output parameter group is a set of spatial vectors (c, σ). Each group of parameters reflects the position of each ant and ant lion. This parameter is finally output to the training model.

The algorithm process of this step requires the data of the sample set. The algorithm construction is as follows:

1) First, as shown in S221 of Figure 2, the positions x _{n and d} of the ant lions and ants are randomly initialized. The initialization method is shown in formula (1). Each position x _{n and d} represents a solution to the problem. Finally, the data will calculate the corresponding fitness value according to the fitness function, and the one with the best fitness value will be selected as the elite ant lion, recorded as _RE :

X _n,d =L _d +rand(U _d -L _d ) (1)

Where N is the number of ant lions and ants, D is the number of variables, n is a natural number less than or equal to 1, d is a natural number less than or equal to D, L _d is the lower limit of the d variable, U _d is the d variable upper limit.

2) Next, the ALO algorithm simulates the walking process of ants through a random function, and its random wandering step set can be expressed as:

X(t)＝{0, cumsum[2r(t ₁ )-1],..., cumsum[2r(t _k )-1]} (2)

Among them, cumsum is used to calculate the cumulative sum, t _k represents the k-th iteration, k is the maximum number of iterations; r(t) is a random function, defined as:

where rand is a random number between 0 and 1.

The standardized position of the i-th variable at the t-th iteration can be expressed as:

where a _i and d _i represent the minimum and maximum values of the i-th variable random walk,

and

The minimum and maximum values of the random walk for the i-th variable at the t-th iteration.

3) Assume

and

The defined hypersphere is the search space for random walks of ants. The random walk space will continue to shrink to speed up the capture process. This process is defined as:

where c ^t and d ^t are the minimum and maximum values of all variables at the t-th iteration,

is the position of the i-th ant lion at the t-th iteration, t and T are the current iteration number and the maximum iteration number respectively.

4) As shown in step S222 of Figure 2, the ant lion with the best fitness value in each iteration is saved as an elite ant lion, and then an ordinary ant lion is randomly selected through roulette. Each ant randomly swims around ordinary ant lions and elite ant lions, and its position update can be expressed as:

is the position of the i-th ant at the t-th iteration,

The position of the common antlion selected for the roulette wheel at iteration t,

The position of the elite ant lion selected for roulette at the t-th iteration, the random walk position is updated from equation (2) to equation (4). In this way, the positions of the ant lion and elite ant lion are obtained.

After obtaining the position, you need to determine the optimal parameters, that is, set the fitness function: set the fitness function between the actual value and the predicted value of the mean square error (MSE), and its expression is:

in

As shown in S223 of Figure 2, the fitness value is calculated according to Equation (20), and the fitness values of ants and ant lions are sorted. The one with the largest value becomes the elite ant lion, and the corresponding parameters are used as the optimal parameters.

5) If the fitness value of the ant is higher than that of the ant lion that preyed on it, then the position of the ant lion will be updated to the position of the prey ant, the trap will be rebuilt at the location of the prey ant, and the new trap will have a higher Opportunity to catch ants, the rules are expressed as follows:

The above steps construct the traditional ALO algorithm. Among them, ordinary ant lions mainly affect the global search ability of the algorithm, and elite ant lions mainly affect the local optimization ability of the algorithm. During the ALO algorithm process, the weight of both ants is 0.5.

Step S120: Improve the ant lion algorithm and output the optimal output parameter set. The improvement plan in this application is: by adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions, control the search balance in different iteration stages, and output the optimal Output parameter group. The control method of search balance is as follows: in the early and middle stages of iteration, global search is the main one, and in the late iteration stage, local search is the main one;

In order to realize this control method, this application adopts the method of dynamically adjusting the inertia weight γ of ordinary ant lions non-linearly with the number of iterations. The calculation formula is:

Among them, γ _max =0.9 and γ _min =0.1 are the maximum and minimum values of the inertia weight respectively. Then combined with formula (9) and formula (10), the improved ant lion position can be expressed as:

Figure 3 shows the changing trend of the ordinary ant lion inertia weight γ with iteration. Utilizing the nonlinear dynamic adjustment of the ordinary ant lion inertia weight γ, IALO has strong global search capabilities in the early and middle stages of iteration to avoid falling into local optimality. Then, the local search capability is enhanced in the later stage to ensure that the algorithm can find the local optimal solution.

The implementation process is shown in step S224 in Figure 2. During the iterative process, the ordinary ant lion inertia weight γ is adjusted and the ant lion position is updated, an improved ant lion optimization algorithm (IALO) is constructed, and the optimal parameter group (c, σ) is finally output. .

In the above steps, the algorithms involved in the estimation method of lithium battery SOH status are constructed and optimized. Next is the application in model training:

Step S130: Use the support vector regression model to predict the test sample set and output the SOH estimated value.

Support vector regression (SVR) is an extension of support vector machine in the field of regression. Due to its good generalization ability and fast convergence speed, it is often used to solve nonlinear and small sample problems. Due to the long battery health experiment period and small sample size, SVR is very suitable for SOH prediction. At the same time, the optimal parameter set is obtained through IASO, and C and σ control the generalization ability and fitting ability in the training of the SVR model, that is, controlling the accuracy of prediction and verification accuracy.

The training model is defined as follows:

Given a sample set

Where x _i is the i-th input feature vector, y _i is the corresponding output value, and n is the total number of sample sets. The expression of support vector regression is:

f(x)＝ωφ(x)+b (13)

where f(x) is the output value, x, ω, and b are the input value, weight and intercept respectively.

By introducing slack variables

and

Available:

Among them, c is the penalty factor that affects the generalization ability of the SVR model, and ε is the maximum allowable error of regression, which can be obtained by introducing the Lagrangian operator:

where i=1,2,…,n,

and α _i are Lagrangian operators. By introducing the kernel function, the SVR equation is transformed into:

Among them, K( _xi , x _j )=φ( _xi )φ(x _j ) is the kernel function, x _i and x _j are the input vectors in the training sample set and the test sample set respectively. The radial basis (RBF) function is selected as the kernel function, which is defined as:

where σ is the bandwidth of the RBF kernel function, which affects the fitting ability of the SVR model.

Since c and σ are optimal parameters calculated and generated in IALO, the prediction error and verification accuracy are better controlled in the SVR model.

In order to verify the performance of the solution of this application, the mean absolute error (MAE) or root mean square error (RMSE) can be used for evaluation. They can all be calculated by:

The mean absolute error algorithm is:

The root mean square error algorithm is:

in

Run the algorithm until the end of the iteration, use the corresponding elite antlion output parameters (c, σ), and then use the output parameters to estimate the battery SOH.

This application also provides three methods (Back-Propagation neural network) BP, ALO-SVR and IALO-SVR to compare the SOH of three sets of batteries, corresponding to three lithium batteries, labeled B05, B06 and B07 respectively.

Figure 5 shows the SOH estimation results of different methods, where P510, P520, and P530 are the estimation results of B05, B06, and B07 respectively. When the first 60% of the samples are used as the training set, it can be seen from the figure that the IALO-SVR estimation SOH is the most accurate, while the other two methods do not fit well enough, especially BP. IALO-SVR estimates the SOH of B05, B06, and B07 more accurately, while for BP, the estimation ability is seriously degraded when encountering data fluctuations, resulting in a larger deviation in the final estimated SOH. ALO-SVR has better estimation accuracy in the early stage, but also gradually deviates from the true SOH in the later stage. In the later period, the prediction accuracy of IALO-SVR is still very high compared with ALO-SVR, which is caused by the dynamic adjustment of the inertial weight of ordinary antlion.

Figure 6 shows the SOH estimation errors of the three methods. Figures P610, P620, and P630 are the estimation errors of B05, B06, and B07 respectively. The estimation errors are all controlled within 2% without large fluctuations.

Figure 7 plots the performance evaluation indicators of the three algorithms, namely mean absolute error (MAE) and root mean square error (RMSE). The smaller the corresponding value, that is, the smaller the error, the better the performance. Compared with ALO-SVR and BP, the estimation performance of IALO-SVR is significantly improved. For the three battery types, the maximum MAE and RMSE of IALO-SVR are 0.64% and 0.70%, respectively. Compared with ALO-SVR, the MAE and RMSE of IALO-SVR are reduced by 0.998 and 1.394 on average. In addition, compared with BP, the MAE and RMSE of IALO-SVR are also greatly reduced, averaging 2.1422 and 2.5587. In addition, we can also see from the figure that in the SOH estimation performance of B05, the difference between BP and ALO-SVR and IALO-SVR is relatively small, while the SOH estimation performance of B06 and B07 lags significantly behind IALO-SVR. This It is because the SOH curves of B06 and B07 fluctuate greatly, so the fitting ability and generalization ability of BP and ALO-SVR are not as good as IALO-SVR.

Figure 8 is a structural diagram of the evaluation device provided by the present invention. As shown in the figure, it includes the following modules:

P810 data processing module: used for lithium battery data collection, extracting health factors to construct feature vectors, and generating training samples and test samples;

The health factors collected in this module include: equal voltage rise time, average discharge voltage and equal voltage drop time; the proportions of generated training samples and test samples are 60% and 40% respectively.

P820 algorithm control module: a prediction algorithm for determining the SOH state of lithium batteries, defining algorithm parameters and output parameter groups; the defined algorithm parameters include defining the population size of ants and ant lions, the dimensionality of the fitness function, and the maximum number of iterations ;The output parameter group includes a space vector (c, σ), the upper limit of c and σ is set to 100, and the lower limit is set to 0.01;

The algorithm control module also includes the P821 fitness value control unit, which is used to calculate the fitness value for the output of the optimal output parameter group, sort the fitness values of ants and ant lions, set the maximum value to the elite ant lion, and the corresponding parameters are optimal Output parameter group, in which the calculation method of fitness value supports the fitness function between the mean square error (MSE) actual value and the predicted value, and its expression is:

in

P830 algorithm optimization module: used to improve the prediction algorithm and output the optimal output parameter group. The improvement includes adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions;

The calculation method for adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions defined by the algorithm optimization module is:

Among them, γ _max is the maximum value of inertia weight and is the minimum value of inertia weight.

P840 model training and data output module: used to predict the test sample set through the support vector regression model and output the SOH estimated value.

Through the prediction scheme for lithium battery SOH status proposed in this application, the number of iterations can be reduced, the SOH prediction accuracy can be effectively improved, and the safety and stability of the battery management system can be improved.

The above disclosures are only a few specific embodiments of the present invention. However, the present invention is not limited thereto. Any changes that can be thought of by those skilled in the art should fall within the protection scope of the present invention.

Claims

A method for estimating SOH status of lithium batteries, which is characterized by including the following steps:

Collect lithium battery data, extract health factors to construct feature vectors, and generate training samples and test samples. The health factors are elements related to the battery capacity decline trend;

The prediction method for determining the SOH state of lithium batteries is the Ant Lion optimization algorithm, which defines algorithm parameters and output parameter groups;

Improve the prediction method and output the optimal output parameter group. The improvement means: by adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions, control the search balance in different iteration stages, and output the optimal output parameter group, The search balance in different iteration stages is: global search is the main focus in the early and middle stages of the iteration, and local search is the main focus in the late iteration stage;

Combined with the optimal output parameter group, the test sample set is predicted through a support vector regression model, and the SOH estimated value is output.
The method for estimating the SOH state of a lithium battery according to claim 1, wherein the defined algorithm parameters include:

The population size of ants and ant lions is set to Agents_no=30,

The dimension of the fitness function is set to dim=2,

The maximum number of iterations is set to Max_iter=100,

The output parameter group is a space vector (C, σ) representing the positions of ants and ant lions. The upper limit of C and σ is set to 100, and the lower limit is set to 0.01.
The method for estimating the SOH state of lithium batteries according to claim 2, characterized in that the weight value corresponding to the random walk of elite ant lions and ordinary ant lions controls the search balance in different iteration stages: non-linear according to the number of iterations Dynamically adjust the inertia weight γ of the ordinary ant lion. The calculation method is:

Among them, γ max is the maximum value of inertia weight, and γ min is the minimum value of inertia weight.
The method for estimating SOH status of lithium batteries according to claim 1, characterized in that before outputting the optimal output parameter group, the method further includes calculating fitness values, sorting the fitness values of ants and ant lions, and the maximum value is set to For Elite Ant Lion, the corresponding parameter is the optimal output parameter group, in which the fitness value is calculated as the fitness function between the mean square error (MSE) actual value and the predicted value, and its expression is:

in
is the i-th predicted value, yi is the i -th actual value, and n is the number of training sets.
The method for estimating the SOH state of a lithium battery according to claim 1, wherein the health factors include: equal voltage rise time, average discharge voltage and equal voltage drop time;

The proportions of training samples and test samples are 60% and 40% respectively.
The method for estimating the SOH state of a lithium battery according to claim 1, wherein the prediction of the test sample set through a support vector regression model includes prediction result analysis, including using absolute error and root mean square error. As the evaluation criteria, the calculation method is as follows:

Among them: MAE is the mean absolute error, RMSE is the root mean square error,
and yi are the predicted value and actual value of the i-th SOH respectively, and n is the number of samples.
A device for estimating the SOH state of a lithium battery, which is characterized by including:

Data processing module: used to collect lithium battery data, extract health factors to construct feature vectors, and generate training samples and test samples;

Algorithm control module: a prediction algorithm for determining the SOH state of lithium batteries, defining algorithm parameters and output parameter groups; the defined algorithm parameters include defining the population size of ants and ant lions, the dimensionality of the fitness function, and the maximum number of iterations; The output parameter group includes a space vector (C, σ), the upper limit of C and σ is set to 100, and the lower limit is set to 0.01;

Algorithm optimization module: used to improve the prediction algorithm and output the optimal output parameter group. The improvement includes adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions;

Model training and data output module: used to predict the test sample set through a support vector regression model and output the SOH estimate.
The device for estimating the SOH state of a lithium battery according to claim 7, wherein the calculation method defined by the algorithm optimization module for adjusting the weight values corresponding to the random walks of elite ant lions and ordinary ant lions is:

Among them, γ max is the maximum value of the inertia weight, γ min is the minimum value of the inertia weight, and γ max + γ min =1.
The device for estimating the SOH state of a lithium battery according to claim 7, wherein the algorithm control module includes a fitness value control unit for calculating the fitness value for outputting the optimal output parameter group, for ants and The fitness value of ant lions is sorted, and the maximum value is set as the elite ant lion, and the corresponding parameter is the optimal output parameter group. Among them, the calculation method of the fitness value supports the fitness between the mean square error (MSE) actual value and the predicted value. function, its expression is:

in
is the i-th predicted value, yi is the i -th actual value, and n is the number of training sets.
A device for estimating the SOH state of a lithium battery according to claim 7, wherein the health factors collected by the data processing module include: equal voltage rise time, average discharge voltage and equal voltage drop time;

The data processing module controls the proportions of the samples and test samples to be 60% and 40% respectively.