CN108805217A - A kind of health state of lithium ion battery method of estimation and system based on support vector machines - Google Patents

Abstract

The invention discloses a kind of health state of lithium ion battery method of estimation and system based on support vector machines, including：Determine the input variable and output variable of Support vector regression prediction；Input variable and output variable are divided into training set data group and test set data group；Regression model foundation is carried out to the training set data after normalization, obtains regression function；Test set data are brought into regression model after training, to predict the electricity being filled with when constant-current charging of battery reaches blanking voltage；The electricity that constant-current charge in training set data to blanking voltage is filled with is fitted with the current test capacity obtained after volume test, the equation that the electricity that test set data prediction obtains is brought into after fitting is obtained into current predictive capacity, to estimate cell health state.The present invention can reach the electricity that blanking voltage is filled with to the lithium ion battery constant-current charge under various constant-current charge environment and predict there is wide applicability.

Description

A method and system for estimating the state of health of a lithium-ion battery based on a support vector machine

technical field

The invention belongs to the technical field of estimation of the state of health of lithium-ion batteries, and in particular relates to a method and system for estimating the state of health of lithium-ion batteries based on a support vector machine.

Background technique

Compared with fuel vehicles, electric vehicles have greater advantages in terms of energy saving, emission reduction and environmental protection. Therefore, various countries are scrambling to introduce various incentives to vigorously promote the development of electric vehicles. Lithium batteries have the advantages of high specific energy, long service life, high rated voltage, high power tolerance, and low self-discharge rate, and have been favored by electric vehicle manufacturers since their listing. The rapid development of electric vehicles has made people put forward higher requirements on the performance of lithium batteries, so the state of health (State Of Health; SOH) estimation of lithium batteries has attracted much attention. Due to the complexity of the battery aging mechanism, it is difficult to quickly and accurately estimate the battery SOH.

A typical method of judging SOH is to estimate its value based on a decrease in capacity or an increase in resistance. Among them, the extended Kalman filter method is considered to be the most reliable method for estimating SOH online. However, the accuracy of this method depends on the parameters of the built model, and the accuracy is easily affected by the accuracy of the model. Therefore, a convenient, simple and accurate method for obtaining the SOH value has attracted much attention.

It can be seen from the battery cycle data that under the same charging conditions, the amount of electricity Q charged by constant current charging to the cut-off voltage of the battery decreases with the decrease of SOH and there is a high degree of correlation between the two. Due to the actual charging situation, the lithium battery cannot charge the terminal voltage to the cut-off voltage every time the constant current is charged. Therefore, accurately predicting the charge Q of the battery constant current charging to the cut-off voltage under the same charging conditions becomes a key step in predicting the SOH value of lithium batteries. At the same time, due to the hysteresis of the battery charging system, after the lithium battery reaches the charging cut-off voltage in the constant current charging process, the charging related equipment will take action, which will inevitably cause the battery to overshoot and cause damage to the battery, thereby shortening the battery life. service life. In order to solve the above problems, it is necessary to predict in advance the charge Q of the battery constant current charging to the cut-off voltage to prevent the terminal voltage from exceeding the cut-off voltage and to further estimate the battery SOH.

SOH is generally defined as Among them, c _M is the current test capacity of the battery, and c ₀ is the rated capacity of the battery. Since the test cycle is long and wasteful when testing the c _M value, it is an important task of the battery management system to estimate the SOH of the battery accurately, reliably and conveniently.

Contents of the invention

The object of the present invention is to solve the above-mentioned problems, and propose a method and system for estimating the state of health of a lithium-ion battery based on a support vector machine algorithm. In this method, based on the historical data, working status and environmental data of the lithium battery, the support vector machine makes an accurate prediction of the amount of electricity Q charged by the constant current charging of the lithium battery to the cut-off voltage. The parameter selection is optimized accordingly, which improves the accuracy of parameter selection. It can be seen from the experiment that under the same charging conditions, the amount of electricity charged by constant current charging of lithium batteries to the cut-off voltage decreases as the SOH value decreases, and there is a high degree of correlation between the two. Therefore, according to the relevant data of the previous test, the relationship between the test set constant-current charging quantity _Q11 and the SOH value can be fitted by the least square method, so that the constant-current charging quantity Q11 predicted by the above-mentioned support vector machine algorithm has _a significant impact on the current lithium battery. The SOH of the battery is estimated.

In order to achieve the above object, the present invention adopts the following technical solutions:

The first object of the present invention is to disclose a method for estimating the state of health of a lithium-ion battery based on a support vector machine, comprising:

Carry out cyclic charging and discharging experiments on lithium batteries, record the historical data of various working states of lithium batteries in real time, and determine the input variables and output variables for support vector machine regression prediction;

Divide the input variables and output variables into two groups: the training set data group and the test set data group;

Normalize the training set data set and the test set data set;

Establish a regression model on the normalized training set data to obtain a regression function;

Select RBF as the kernel function, and use the grid search method to select the optimal combination of RBF kernel function parameters: the width parameter of the kernel function, the penalty coefficient and the loss function;

Determine the optimal parameter regression model according to the optimal RBF kernel function parameter combination;

Bring the test set data into the trained regression model to predict the charge Q ₁ when the constant current charging of the battery reaches the cut-off voltage;

Under the same constant-current charging condition, use the least squares method to fit the electric quantity Q ₁₁ charged from the constant-current charge in the training set data to the cut-off voltage and the current test capacity Q ₀ obtained after the capacity test, and use the least squares method to fit the test set The Q ₁ obtained from the data prediction is brought into the fitted equation to obtain the current predicted capacity c _M , thereby estimating the state of health of the battery.

Further, the input variables and output variables for determining the regression prediction of the support vector machine are specifically:

Record the historical data of various working states of lithium batteries in real time, including the number of cycles, cycle charge and discharge current, discharge depth, temperature and the current of the next constant current charge, as input variables for support vector machine regression prediction;

After the set cycle period, conduct a constant current charging and capacity test on the lithium battery, and count the charged power Q when the battery constant current charging reaches the cut-off voltage as the output variable of the support vector machine regression prediction.

Further, the normalization processing of the training set data group and the test set data group is specifically:

The normalized sample value is the ratio of the difference between the sample value and the minimum value of the sample to the difference between the maximum value and the minimum value of the sample value.

Furthermore, when the performance of the parameter training prediction model is the same, the parameter combination with a relatively small penalty coefficient is preferred.

Further, the regression model is established on the normalized training set data to obtain a regression function; specifically:

Let the training set sample pair containing m training samples be in, is the input column vector of the ith training sample, is the corresponding output value;

The linear regression function established in the high-dimensional feature space is:

Define the ε linear insensitive loss function as:

In order to find an optimal classification surface so that all training samples have the smallest error from the optimal classification surface, the slack variables ξ _i , ξ _i ^* need to be satisfied as follows:

in, is a nonlinear mapping function, is the regression coefficient vector, b is the threshold; is the predicted value returned by the regression function, y is the corresponding real value, ε is a set number greater than 0; C is the penalty factor, and the larger the C, the greater the penalty for the sample whose training error is greater than ε, and ε specifies the regression function The smaller the ε, the smaller the error of the regression function.

Further, the constraint relational expression is solved, specifically:

Using the Lagrangian dual theory to transform the constraint relation into a dual problem;

By the Lagrangian duality, the minimum problem in the original Lagrangian function is transformed into a maximum problem;

Use the SMO algorithm to solve the Lagrangian multipliers in the dual problem; and then obtain the regression coefficient vector and the threshold b to determine the regression function.

Further, the regression function is specifically:

in, are the optimal solutions of the Lagrangian multipliers corresponding to sample i, respectively, is the kernel function, and b ^* is the threshold obtained by solving the optimal solution of Lagrangian multipliers.

Further, the RBF is selected as the kernel function, and the grid search method is used to select the optimal RBF kernel function parameter combination; specifically:

Step 1: Carry out cross-validation on various possible RBF kernel function parameter combination values, and find out the combination pair with the highest cross-validation accuracy; arrange and combine the possible values of each parameter, and list all possible combination results to generate " grid";

Step 2: Combine the parameters of each RBF kernel function for SVM training, and use cross-validation to evaluate the performance;

Step 3: After the fitting function has tried all the parameter combinations, return an optimal classification surface, and automatically adjust to the optimal parameter combination, so that the error of all training samples from the optimal classification surface is minimized.

The second object of the present invention is to disclose a system for estimating the state of health of a lithium-ion battery based on a support vector machine, including a server, the server including a memory, a processor and a computer program stored on the memory and operable on the processor, When the processor executes the program, the following steps are implemented:

Determine input variables and output variables for SVM regression prediction;

Divide the input variables and output variables into two groups: the training set data group and the test set data group;

Normalize the training set data set and the test set data set;

Establish a regression model on the normalized training set data to obtain a regression function;

Select RBF as the kernel function, and use the grid search method to select the optimal combination of RBF kernel function parameters: the width parameter of the kernel function, the penalty coefficient and the loss function;

Determine the optimal parameter regression model according to the optimal RBF kernel function parameter combination;

Bring the test set data into the trained regression model to predict the charge Q ₁ when the constant current charging of the battery reaches the cut-off voltage;

Under the same constant-current charging condition, use the least squares method to fit the electric quantity Q ₁₁ charged from the constant-current charge in the training set data to the cut-off voltage and the current test capacity Q ₀ obtained after the capacity test, and use the least squares method to fit the test set The Q ₁ obtained from the data prediction is brought into the fitted equation to obtain the current predicted capacity c _M , thereby estimating the state of health of the battery.

The third object of the present invention is to disclose a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the following steps are performed:

Determine input variables and output variables for SVM regression prediction;

Divide the input variables and output variables into two groups: the training set data group and the test set data group;

Normalize the training set data set and the test set data set;

Establish a regression model on the normalized training set data to obtain a regression function;

Select RBF as the kernel function, and use the grid search method to select the optimal combination of RBF kernel function parameters: the width parameter of the kernel function, the penalty coefficient and the loss function;

Determine the optimal parameter regression model according to the optimal RBF kernel function parameter combination;

Bring the test set data into the trained regression model to predict the charge Q ₁ when the constant current charging of the battery reaches the cut-off voltage;

Under the same constant-current charging condition, use the least squares method to fit the electric quantity Q ₁₁ charged from the constant-current charge in the training set data to the cut-off voltage and the current test capacity Q ₀ obtained after the capacity test, and use the least squares method to fit the test set The Q ₁ obtained from the data prediction is brought into the fitted equation to obtain the current predicted capacity c _M , thereby estimating the state of health of the battery.

Beneficial effects of the present invention:

(1) There is no need for a lithium battery of a specific material, and it is possible to predict the amount _of electricity Q1 charged into the lithium-ion battery under various constant-current charging environments when the constant-current charging reaches the cut-off voltage, and has wide applicability;

(2) There is no need to study the complex electrochemical reaction mechanism of lithium batteries, which simplifies the calculation process;

(3) It is possible to estimate the SOH of the lithium battery based on the charging data, without the need for a separate capacity test, shortening the long test cycle and reducing waste;

(4) Considering the practical application, the battery does not have to be charged to the cut-off voltage every time.

(5) It can determine the allowable charging capacity of the battery, prevent the occurrence of overcharging during the constant current charging process of the battery, and effectively protect the charging equipment and the lithium battery.

Description of drawings

Figure 1 is a flowchart of battery SOH estimation;

Figure 2 is a constant current and constant voltage charging diagram of a lithium battery under different SOH values;

Fig. 3 (a) is the electric quantity Q ₁₁ that is charged into the electric quantity Q that the constant current charging of training set reaches the cut-off voltage prediction figure;

Fig. 3(b) is a prediction diagram of the electric quantity Q ₁ charged into the test set when the constant current charging reaches the cut-off voltage;

Fig. 3(c) is a fitting relation diagram of the electric quantity Q ₁₁ charged by the constant current in the training set and the current test capacity Q ₀ obtained from the capacity test.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

The invention discloses a method for estimating the state of health of a lithium-ion battery based on a support vector machine, as shown in Figure 1, comprising the following steps:

Step (1): During the cycle charge and discharge process of the lithium battery, record the historical data of the various working states of the lithium battery in real time, including the cycle number N, cycle charge and discharge current I _charge and I _discharge , discharge depth D, temperature T _em and the current size I of the constant current charging to be carried out next time are used as input variables for the regression prediction of the support vector machine.

After going through a certain cycle period, carry out a constant current charging and capacity test on the lithium battery, and count the amount of electricity Q charged by the constant current charging of the battery to the cut-off voltage, which is used as the output variable of the regression prediction.

Step (2): The input and output variable groups are divided into a training set data set and a test set data set, wherein the training set is used to train the regression model, and the test set is used for regression prediction and evaluation of the performance of the model. Since the input and output exist in pairs, the input and output of the first half can be used as the training set, and the input and output of the second half can be used as the prediction set.

And adjust the data to a format that meets the requirements of the Libsvm toolbox.

Step (3): Normalize the obtained input and output variable sets to speed up the cross-validation method to find the optimal solution and improve the prediction accuracy.

Before building the regression model, the data is normalized first, that is, the input and output variables are uniformly normalized to the value range of [0,1] according to the formula (1).

Among them, x _i and y _i are the values before and after conversion respectively, and MaxValue and MinValue are the maximum value and minimum value of the sample respectively.

Step (4): Establish a regression model from the normalized training set data obtained in the previous step to obtain a regression function.

Without loss of generality, let the training set sample pair containing m training samples be in,

is the input column vector of the ith training sample, is the corresponding output value. In the practical application of this prediction, for such linear inseparable problems, it is necessary to map the samples of the original input space to the high-dimensional feature space H through the nonlinear mapping Φ:R ^d →H, and then map the high-dimensional feature space H The optimal classification surface is constructed in order to minimize the error of all training samples from the optimal classification surface. Assume that the linear regression function established in the high-dimensional feature space is shown in formula (2).

in, is a nonlinear mapping function, is the regression coefficient vector, is the threshold.

Define the ε linear insensitive loss function as:

Among them, f(x) is the predicted value returned by the regression function, y is the corresponding real value and ε is a predetermined number greater than 0. This function means that if The difference between y and y is less than or equal to ε, then the loss is equal to 0.

In order to find an optimal classification surface, the error of all training samples from the optimal classification surface is minimized, but considering that a small number of samples cannot meet the constraints, the optimal classification surface cannot be found. For such cases, under the conditions of satisfying the above constraints, the minimum, need to introduce slack variables ξ _i , ξ _i ^* , and search for The problem of b is described in mathematical language, which is shown in formula (4):

Among them, C is the penalty factor. The larger the C, the greater the penalty for the sample whose training error is greater than ε. ε specifies the error requirement of the regression function. The smaller the ε, the smaller the error of the regression function.

The solution to the above problem is a convex quadratic programming problem. Due to the complexity of the calculation, it is relatively difficult to solve it directly. Although it can be solved directly with the ready-made optimization calculation package, in comparison, a more efficient method will be adopted. That is, according to the Lagrangian dual theory, formula (4) is transformed into a dual problem. The advantage of this is that: one dual problem is often easier to solve; the two can naturally introduce kernel functions, and then be extended to nonlinear classification problems. The constructed Lagrangian function is shown in formula (5).

Among them, a=(a ₁ ,a ₂ ,...,a _m ) ^T , a ^* =(a ₁ ^* ,a ₂ ^* ,...,a _m ^* ) ^T , η=(η ₁ ,η ₂ ,...,η _m ) ^T , η ^* = (η ₁ ^* ,η ₂ ^* ,...,η _m ^* ) ^T is a Lagrange multiplier vector.

To solve this dual problem, it is divided into three steps: First, let L(ω,b,ξ,ξ*,a,a*,η,η*) be minimized with respect to ω,b and ξ,ξ*, and then find For the maxima of a, a*, η, η*, finally the SMO algorithm is used to solve the Lagrangian multipliers in the dual problem.

First fix a, a*, η, η*, to minimize L with respect to ω, b and ξ, ξ*, we take partial derivatives for ω, b and ξ, ξ* respectively, that is and is equal to zero, so that formula (6)

Due to the Lagrangian duality, bringing formula (6) into the original Lagrangian function (5) can transform the minimum value problem into a maximization problem, as shown in formula (7)

in, is the kernel function.

Suppose the optimal solution obtained when solving formula (7) is then there is

Among them, N _nsv is the number σ of support vectors.

Therefore, the regression function can be transformed into

Among them, only some parameters is not equal to 0, and its corresponding sample _xi is the support vector in the problem.

Step (5): It can be seen from formula (9) that in order to solve the regression function problem of the support vector machine algorithm, an appropriate kernel function must be selected. The commonly used kernel functions include: 1. Linear kernel function; 2. d-order polynomial kernel function; 3. Radial basis kernel function (RBF); 4. Sigmoid kernel function. Compared with other kernel functions, the RBF kernel function is relatively widely used, whether it is a small sample or a large sample, high-dimensional or low-dimensional, etc., the RBF kernel function is applicable. It has the following advantages over other functions:

1) The RBF kernel function can map a sample to a higher-dimensional space, and the linear kernel function is a special case of RBF, that is, if RBF is considered, then there is no need to consider the linear kernel function.

2) Compared with the polynomial kernel function, RBF needs to determine fewer parameters, and the number of kernel function parameters directly affects the complexity of the function. In addition, when the order of the polynomial is relatively high, the element values of the kernel matrix will tend to be infinitely large or infinitely small. The RBF will reduce the numerical calculation difficulty in the above problems.

3) For some parameters, RBF and Sigmoid have similar performance.

Therefore, the RBF kernel function is selected in the prediction of the amount of electricity charged when the constant current charging reaches the cut-off voltage, and its expression is shown in formula (10).

Among them, σ is the width parameter of the function, which controls the radial range of the function.

Since the kernel function model parameters have a great influence on the performance of the model, when using the Libsvm toolbox, it is necessary to select a better combination of RBF kernel function parameters: the width parameter σ of the kernel function, the penalty coefficient C and the loss function P. Considering that there are relatively few parameters to be selected, the complexity of the "grid search method" is not much different from that of advanced algorithms, and it has the advantages of high parallelism and the ability to avoid overfitting. Therefore, we apply the "grid search method" in the parameter selection. The specific implementation steps are as follows:

1. Try various possible combination values, and then perform cross-validation to find the combination pair that makes the cross-validation accuracy the highest. Arrange and combine the possible values of each parameter, and list all possible combinations to generate a "grid".

Second, each combination is then used for SVM training and performance is evaluated using cross-validation.

3. After the fitting function tries all the parameter combinations, it returns a suitable classification surface and automatically adjusts to the best parameter combination.

Step (6): Use the optimal parameters obtained by the "grid search method" to train the prediction model. When the performance of the models is the same, in order to reduce the calculation time, the parameter combination with a relatively small penalty factor C is preferred.

Step (7): Bring the test set data into the trained regression model, thereby predicting the charge Q1 _of the battery constant current charging to the cut-off voltage to further predict the SOH value, and the accuracy of the final test set prediction results degree to make an evaluation.

Step (8): Under the same constant current charging conditions, use the least squares method to simulate the electric quantity Q ₁₁ that is charged when the constant current charging reaches the cut-off voltage in the training set data and the current test capacity Q ₀ obtained after the capacity test. Then put the power Q ₁ predicted by the test set data into the fitted equation to get the predicted current capacity c _M , according to Make an estimate of the battery SOH.

Figure ₃ (a) and ₃ (b). The fitting relationship between the constant current charging power Q ₁₁ and the current test capacity Q ₀ obtained from the capacity test under the same charging conditions in the training set is shown in Figure 3(c).

The accuracy _of the prediction results of _Q11 and Q1 from the constant current charging to the cut-off voltage of the training set and the test set is evaluated by the mean square error E and the coefficient of determination R2. ^The accuracy of the training set and the test set for the battery SOH estimation is evaluated by the absolute error The specific results are shown in Table 1. From the specific evaluation index, based on the support vector machine constant current charging to the cut-off voltage charging Q1 prediction and SOH estimation method has _a higher prediction accuracy.

Table 1 Prediction of the amount of electricity Q charged from constant current charging to the cut-off voltage of lithium batteries and the accuracy evaluation of SOH estimation methods

E. R ² SOH estimated absolute error rate mean Training set 0.00065566 0.99271 0.48% test set 0.00086036 0.99112 0.56%

It can be seen from the experiment that under the same charging conditions, the charge Q of lithium battery constant current charging to reach the cut-off voltage is shortened as the SOH value decreases, and there is a high degree of correlation between the two. Therefore, according to the relevant data of the previous test, the relationship between _Q11 and SOH value can be fitted by the least square method, so that the constant current charging quantity Q1 predicted by the above-mentioned support vector machine algorithm can be used to estimate the SOH _of the current lithium battery .

Figure 2 shows the constant current and constant voltage charging diagram of lithium batteries under different SOH values. It can be seen from Figure 2 that under the same constant current charging conditions, the constant current charging power decreases with the decrease of SOH value.

Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it does not limit the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims (10)

1. A lithium ion battery health state estimation method based on a support vector machine is characterized by comprising the following steps:

carrying out a cyclic charge and discharge experiment on the lithium battery, recording historical data of various working states of the lithium battery in real time, and determining an input variable and an output variable of regression prediction of a support vector machine;

the input variables and the output variables are divided into two groups: a training set data set and a test set data set;

carrying out normalization processing on the training set data group and the test set data group;

carrying out regression model establishment on the normalized training set data to obtain a regression function;

selecting RBF as kernel function, selecting optimal RBF kernel function parameter combination by using grid search method: width parameters of kernel functions, penalty coefficients and loss functions;

determining an optimal parameter regression model according to the optimal RBF kernel function parameter combination;

the data of the test set is substituted into the regression model after training, so that the electric quantity Q charged when the constant current charging of the battery reaches the cut-off voltage is predicted₁；

Under the same constant current charging condition, the constant current in the training set data is charged to the electric quantity Q charged by the cut-off voltage₁₁And the current testing capacity Q obtained after the capacity testing₀Performing fitting by using a least square method, and predicting the Q obtained by the data of the test set₁The fitted equation is substituted to obtain the current predicted capacity c_MAnd thus the state of health of the battery is estimated.

2. The method for estimating the state of health of a lithium ion battery based on a support vector machine according to claim 1, wherein the determining of the input variables and the output variables of the regression prediction of the support vector machine is specifically:

recording historical data of various working states of the lithium battery in real time, wherein the historical data comprises cycle times, cycle charging and discharging current, discharging depth, temperature and current of constant current charging to be carried out next time, and the historical data is used as an input variable for regression prediction of a support vector machine;

after a set cycle period, performing constant current charging and capacity testing on the lithium battery once, and counting the electric quantity Q charged when the constant current charging of the battery reaches a cut-off voltage, wherein the electric quantity Q is used as an output variable for regression prediction of the support vector machine.

3. The lithium ion battery health state estimation method based on the support vector machine according to claim 1, wherein the normalization processing is performed on the training set data group and the test set data group, specifically:

the normalized sample value is the ratio of the difference between the sample value and the minimum value of the sample to the difference between the maximum value and the minimum value of the sample value.

4. The lithium ion battery state of health estimation method based on support vector machine of claim 1, characterized in that, when the performance of the parameter training prediction model is the same, the parameter combination with relatively small penalty coefficient is selected preferentially.

5. The lithium ion battery state of health estimation method based on support vector machine of claim 1, characterized in that, the normalized training set data is subjected to regression model establishment to obtain a regression function; the method specifically comprises the following steps:

let the training set sample pair containing m training samples beWherein,is the input column vector for the ith training sample, is the corresponding output value;

the linear regression function established in the high-dimensional feature space is set as:

define the epsilon linear insensitive loss function as:

in order to find an optimal classification surface to minimize the error of all training samples from the optimal classification surface, a relaxation variable ξ is introduced_i,ξ_i ^*Then, the following constraint relation needs to be satisfied:

wherein,in order to be a non-linear mapping function,is a regression coefficient vector, b is a threshold;a predicted value returned by the regression function is used, y is a corresponding true value, and epsilon is a set number larger than 0; and C is a penalty factor, wherein the larger C is the penalty larger for the sample with the training error larger than epsilon, epsilon specifies the error requirement of the regression function, and the smaller epsilon is the error of the regression function.

6. The lithium ion battery state of health estimation method based on a support vector machine according to claim 5, characterized in that the constraint relation is solved, specifically:

converting the constraint relation into a dual problem by adopting a Lagrange dual theory;

converting the minimum problem in the original Lagrange function into a maximization problem by Lagrange duality;

solving a Lagrange multiplier in the dual problem by adopting an SMO algorithm; further obtain the regression coefficient vectorAnd a threshold value b, determining a regression function.

7. The lithium ion battery state of health estimation method based on a support vector machine according to claim 6, characterized in that the regression function is specifically:

wherein,respectively the optimal solutions of the lagrange multipliers corresponding to sample i,is a kernel function, b^*The obtained threshold value is solved according to the optimal solution of the Lagrange multiplier.

8. The method according to claim 1, wherein the RBFs are selected as kernel functions, and an optimal RBF kernel function parameter combination is selected by using a grid search method; the method specifically comprises the following steps:

the method comprises the following steps: performing cross validation on various possible RBF kernel function parameter combination values, and finding out a combination pair which enables the cross validation accuracy to be highest; arranging and combining the possible values of each parameter, and listing all possible combination results to generate a 'grid';

step two: using each RBF kernel function parameter combination for SVM training, and evaluating the performance by using cross validation;

step three: and after the fitting function tries all parameter combinations, returning to an optimal classification surface, and automatically adjusting to the optimal parameter combination to ensure that the error of all training samples from the optimal classification surface is minimum.

9. A system for estimating state of health of a lithium ion battery based on a support vector machine, comprising a server including a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the following steps when executing the program:

determining an input variable and an output variable of the regression prediction of the support vector machine;

the input variables and the output variables are divided into two groups: a training set data set and a test set data set;

carrying out normalization processing on the training set data group and the test set data group;

carrying out regression model establishment on the normalized training set data to obtain a regression function;

selecting RBF as kernel function, selecting optimal RBF kernel function parameter combination by using grid search method: width parameters of kernel functions, penalty coefficients and loss functions;

determining an optimal parameter regression model according to the optimal RBF kernel function parameter combination;

the data of the test set is substituted into the regression model after training, so that the electric quantity Q charged when the constant current charging of the battery reaches the cut-off voltage is predicted₁；

Under the same constant current charging condition, the constant current in the training set data is charged to the electric quantity Q charged by the cut-off voltage₁₁And the current testing capacity Q obtained after the capacity testing₀Performing fitting by using a least square method, and predicting the Q obtained by the data of the test set₁The fitted equation is substituted to obtain the current predicted capacity c_MAnd thus the state of health of the battery is estimated.

10. A computer-readable storage medium on which a computer program is stored, the program, when executed by a processor, performing the steps of:

determining an input variable and an output variable of the regression prediction of the support vector machine;

the input variables and the output variables are divided into two groups: a training set data set and a test set data set;

carrying out normalization processing on the training set data group and the test set data group;

carrying out regression model establishment on the normalized training set data to obtain a regression function;

selecting RBF as kernel function, selecting optimal RBF kernel function parameter combination by using grid search method: width parameters of kernel functions, penalty coefficients and loss functions;

determining an optimal parameter regression model according to the optimal RBF kernel function parameter combination;

the data of the test set is substituted into the regression model after training, so that the electric quantity Q charged when the constant current charging of the battery reaches the cut-off voltage is predicted₁；

Under the same constant current charging condition, the constant current in the training set data is charged to the electric quantity Q charged by the cut-off voltage₁₁And the current testing capacity Q obtained after the capacity testing₀Performing fitting by using a least square method, and predicting the Q obtained by the data of the test set₁The fitted equation is substituted to obtain the current predicted capacity c_MAnd thus the state of health of the battery is estimated.