SOH prediction method and system based on multi-feature extraction and optimized SVR

Abstract

The invention discloses an SOH prediction method and system based on multi-feature extraction and optimized SVR, and belongs to the field of battery SOH prediction. The method comprises the following steps: acquiring constant-current charging data of a lithium battery, and dividing constant-current charging voltage into different initial voltages; extracting battery voltage change and energy change under different initial voltages when the charging capacity is equal, and screening to form an input feature set; dividing the input feature set into a training set and a test set; a DBO-SVR prediction model is obtained after an SVR model is optimized based on a DBO algorithm, and the training set is adopted for training; and inputting the test set into the trained DBO-SVR prediction model to obtain an SOH prediction value. According to the method, the problem that the existing feature acquisition time is too long is improved, the flexibility and rapidity of SOH prediction are improved, and meanwhile, the SVR model is optimized based on the DBO to improve the prediction precision.

Description

Technical Field

[0001] The present invention belongs to the field of battery SOH prediction. Specifically, the present invention relates to a method and system for SOH prediction based on multi-feature extraction and optimized SVR. Background Art

[0002] In real-world usage scenarios, lithium batteries inevitably undergo charging and discharging processes, which cause irreversible changes to the state of health (SOH) of the battery, thereby affecting the safety and stability of its operation. Therefore, accurately predicting the SOH of lithium batteries is crucial for practical applications.

[0003] Currently, a large number of machine learning algorithms have been used for battery capacity or state of health estimation, such as Gaussian process regression, support vector machine regression (SVR) models, random forests, neural networks, etc. Support vector machines are widely used because they have good model generalization ability, can handle the non-linear relationship between input variables and output variables, and have good robustness compared to other data-driven methods. However, there is a problem of kernel parameter selection, and the quality of the parameters directly affects the prediction accuracy of the support vector machine method. While many scholars have proposed data-driven prediction models, they have also extracted many feature factors from lithium battery charge and discharge data. Many scholars have extracted features such as peaks and peak areas from the IC (Increment capacity, IC) curve. However, these extracted features often require fixed starting conditions and long extraction times, and the extracted feature factors are not considered easy to implement in practical engineering applications.

[0004] Therefore, the present invention proposes a method and system for SOH prediction based on multi-feature extraction and optimized SVR. Summary of the Invention

[0005] The present invention aims to overcome the deficiencies of the prior art and proposes a method and system for SOH prediction based on multi-feature extraction and optimized SVR to achieve the following objectives: to improve the problem of too long existing feature acquisition time, thereby improving the flexibility and rapidity of SOH prediction, and at the same time optimizing the support vector machine regression model based on DBO to improve the prediction accuracy.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a method for SOH prediction based on multi-feature extraction and optimized SVR, the method comprising the following steps:

[0007] Step S1, obtaining the constant current charging data of the lithium battery and dividing the constant current charging voltage into different initial voltages;

[0008] Step S2, extracting the battery voltage change and energy change at equal charging capacity under different initial voltages, screening them, and then forming an input feature set; dividing the input feature set into a training set and a test set;

[0009] Step S3: Optimize the SVR model based on the DBO algorithm to obtain the DBO-SVR prediction model, and use the training set for training;

[0010] Step S4: Input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

[0011] Preferably, the step S1 includes:

[0012] Starting from the first selected initial voltage V start , select a voltage as the initial voltage V every 0.01V voltage, that is: n

[0013] V n = V start + 0.0l(n - 1);

[0014] Where n represents the ordinal number, n = 1, 2, 3,....

[0015] Preferably, the step S2 includes:

[0016] Step S21: Calculate the charging capacity according to the charging current and charging time data during the constant current charging of the lithium battery;

[0017] Step S22: Extract the battery voltage change and energy change at the same charging capacity under different initial voltages as input features;

[0018] Step S23: Calculate the Pearson correlation coefficient between the features under different initial voltages and the battery SOH and compare it with a preset threshold. Take all the features in the initial voltage segment where the features are greater than the preset threshold as input features, that is:

[0019]

[0020] Where F ij represents the i-th feature corresponding to the j-th initial voltage, represents the average value of the corresponding feature; SOH ij represents the SOH under the i-th feature corresponding to the j-th initial voltage, represents its average value; PCC ij represents the Pearson correlation coefficient between the i-th feature corresponding to the j-th initial voltage and SOH;

[0021] Step S24: The selected input features form a set and are divided into a training set and a test set.

[0022] Preferably, the equal charging capacity is set to the power of 5% of the rated capacity of the battery.

[0023] Preferably, in step S3, the DBO algorithm is used to optimize the SVR penalty factor and kernel function parameters. The specific steps include:

[0024] Step S31: Initialize the position of the dung beetle and randomly generate an initial solution;

[0025] Step S32: Establish a fitness function and evaluate the fitness, calculate the fitness value of each dung beetle individual and sort them;

[0026] Step S33: Simulate the behavior of each dung beetle individual, update the position after the behavior ends, and calculate the new fitness value;

[0027] Step S34: Update the optimal solution, update the individual optimal solution, and update the global optimal position;

[0028] Step S35: Determine whether the maximum number of iterations is reached. If so, obtain the global optimal solution at this time, and use the position of the global optimal solution as the optimal penalty factor and kernel function parameters of the SVR model; otherwise, return to step S32.

[0029] Preferably, the fitness function in step S32 is established based on the root mean square error, that is:

[0030]

[0031] where, y i represents the true value of SOH; represents the predicted value of SOH; n represents the total number of samples in the training set.

[0032] Preferably, in step S33, the behaviors of the dung beetle individuals include rolling the ball, dancing, breeding, foraging, and stealing, where:

[0033] (1) The position change of the dung beetle responsible for rolling the ball is as follows:

[0034] x t (t + 1) = x t (t) + α × k × x t (t - 1) + b × Δx

[0035] Δx = |x i (t) - X ω |

[0036] where, t is the number of iterations; x i is the position of the i-th dung beetle; k is the deflection coefficient;

[0037] b is a random number between (0, 1); α is a natural coefficient, ranging from -1 to 1; X ω is the global worst position;

[0038] (2) When encountering an obstacle, the dung beetle will reorient itself through dancing behavior, and its position update formula is as follows:

[0039] x i (t + 1) = x i (t) + tan(θ)|x i (t) - x i (t - 1)|

[0040] where θ is the deflection angle;

[0041] (3) The update formula for the reproductive behavior is as follows:

[0042] Ub * = min(X * × (1 - R), Ub)

[0043] Lb * = max(X * × (1 - R), Lb)

[0044] where X * is the current best position: Ub * and Lb * are the lower and upper bounds of the spawning area respectively; Ub and Lb are the lower and upper bounds of the optimization problem respectively; R = 1 - t / Tmax, where Tmax represents the maximum number of iterations;

[0045] During the iteration process, the position of the egg mass changes dynamically, and its definition is as follows:

[0046]

[0047] where B i is the position of the egg mass; b 1 , b 2 is a 1×D random variable, and D represents the dimension of the optimization problem;

[0048] (4) Establish an optimal foraging area to guide the foraging behavior of the young dung beetles, where the optimal foraging area is defined as follows:

[0049] Ub b = min(X b × (1 + R), Ub)

[0050] Lb b = max(X b × (1 - R), Lb)

[0051] where X b is the current best position: Ub b and Lb bThey are the lower bound and the lower bound of the optimal foraging area respectively; Ub and Lb are the lower bound and the lower bound of the optimization problem respectively; R = 1 - t / Tmax, where Tmax represents the maximum number of iterations;

[0052] The position update formula of the corresponding small dung beetle is as follows:

[0053] x i (t + 1) = x i (t) + C 1 ×(x i (t) - Lb b ) + C 2 ×(x i (t) - Ub b )

[0054] where C 1 is a random number obeying the Gaussian distribution; C 2 is a random variable in (0, 1);

[0055] (5) The update formula of the stealing behavior is as follows:

[0056] x i (t + 1) = X b + S × g × (∣x i (t) - x * ∣ + ∣x i (t0 - x b ∣)

[0057] where g is a random variable obeying the Gaussian distribution with a mean of 0 and a variance of 1; S is a constant. Preferably, the relationship between the input and output of the SVR model is expressed as:

[0058] y i (x i ) = ω · φ(x i ) + b

[0059] where x i is the i-th input feature vector, y i is the corresponding output vector; ω and b represent the weight and the intercept respectively; φ represents the mapping of the input data in the feature space;

[0060] Introducing slack variables and we get:

[0061]

[0062] where C is the penalty factor, and ε (ε > 0) is the maximum allowable error of the regression;

[0063] By introducing Lagrange multipliers and kernel functions, after minimizing the Lagrangian function, the SVR expression of the nonlinear mapping is obtained, that is:

[0064]

[0065] where, α i and are Lagrange multipliers, and K(x i , x) = φ(x i )φ(x j ) is the kernel function.

[0066] Preferably, the kernel function is selected as the radial basis function RBF, and the formula is expressed as follows:

[0067]

[0068] where, σ represents the kernel function parameter.

[0069] Meanwhile, this application also proposes a SOH prediction system based on multi-feature extraction and optimized SVR. The system includes: a data preprocessing module, a prediction model optimization and training module, and a prediction model testing module. Among them:

[0070] The data preprocessing module is used to obtain battery constant current charging data, initial voltage division, multi-feature extraction, and the construction of training sets and test sets;

[0071] The prediction model optimization and training module is used to optimize the SVR model using the DBO algorithm to obtain the DBO-SVR prediction model, and train it based on the training set;

[0072] The prediction model testing module is used to input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

[0073] The technical effect of the present invention is: The present invention improves the problem of the existing overly long feature acquisition time. In the charging stage where features are more easily extracted, more initial voltage segments are considered to obtain a shorter feature acquisition time, thereby improving the flexibility and rapidity of SOH prediction. At the same time, the support vector machine regression is optimized based on the DBO algorithm, improving the prediction accuracy. Description of the Drawings

[0074] Figure 1 is the flowchart of the SOH prediction method based on multi-feature extraction and optimized SVR provided by the embodiment of the present invention;

[0075] Figure 2 is the schematic diagram of the SOH prediction result curve at the initial voltage of 3.7V provided by the embodiment of the present invention;

[0076] Figure 3 Schematic diagram of the SOH prediction result curve at the initial voltage of 3.76V provided by the embodiment of the present invention;

[0077] Figure 4 Schematic diagram of the SOH prediction result curve at the initial voltage of 3.81V provided by the embodiment of the present invention;

[0078] Figure 5 Schematic diagram of the SOH prediction result curve at the initial voltage of 3.86V provided by the embodiment of the present invention;

[0079] Figure 6 Schematic diagram of the SOH prediction result curve at the initial voltage of 3.9V provided by the embodiment of the present invention. Detailed implementation manners

[0080] The following is a detailed description of the specific implementation manners of the present invention by referring to the accompanying drawings and describing the embodiments, aiming to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solutions of the present invention, and facilitate its implementation. It should be noted that the terms "first", "second", etc. used in this application are only for conveniently describing the technical solutions to distinguish different components, and do not limit this application. To make the technical solutions of the present invention clearer, the present invention is explained and illustrated through the following embodiments.

[0081] This application provides a method for predicting SOH based on multi-feature extraction and optimized SVR. First, the charging voltage of the constant-current charging (constant-current charging, abbreviated as CC) curve is divided into different initial charging voltages, then features are extracted from different initial voltages, and finally the optimized support vector machine regression is used to estimate SOH. As Figure 1 shown, the method of this application includes the following steps:

[0082] Step S1, obtain the constant-current charging data of the lithium battery and divide the constant-current charging voltage into different initial voltages;

[0083] Step S2, extract the battery voltage change and energy change at the same charging capacity under different initial voltages, and after screening, form an input feature set; divide the input feature set into a training set and a test set;

[0084] Step S3, optimize the SVR model based on the DBO (dung beetle optimization algorithm) algorithm to obtain the DBO-SVR prediction model, and use the training set for training;

[0085] Step S4, input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

[0086] The following combines Figure 1The method flow chart shown below specifically describes each method step of the present application.

[0087] Step S1: Obtain the constant-current charging data of the lithium battery and divide the constant-current charging voltage into different initial voltages. Relatively speaking, the data during the charging stage is easier to collect and extract features. The constant-current charging data includes data such as charging voltage, charging current, and charging time.

[0088] Specifically, step S1 of the present application includes:

[0089] Starting from the first selected initial voltage V start , select a voltage as the initial voltage V every 0.01V n , that is:

[0090] V n = V start + 0.01(n - 1);

[0091] Where n represents the ordinal number, n = 1, 2, 3,....

[0092] Step S2: Extract the battery voltage change and energy change at equal charging capacity under different initial voltages, and after screening, form an input feature set; divide the input feature set into a training set and a test set.

[0093] Specifically, step S2 of the present application includes:

[0094] Step S21: Calculate the charging capacity according to the charging current and charging time data during the constant-current charging of the lithium battery.

[0095] Step S22: Extract the battery voltage change and energy change at equal charging capacity under different initial voltages as input features.

[0096] The equal charging capacity in the present application is set to 5% of the rated capacity, and it can be flexibly selected according to needs during specific implementation. Charging 5% of the rated capacity only takes 3 minutes for 1C constant-current charging, significantly reducing the feature extraction time and improving the rapidity and flexibility of SOH prediction. After charging 5% of the rated capacity into the battery under different initial voltages, the corresponding voltage change and energy change are used as input features F 1n and F 2n , correspondingly, set x = [F 1n F 2n as the input feature vector.

[0097] Based on this, the present application simultaneously extracts features for multiple initial voltage segments, with faster efficiency and shorter feature extraction time, thereby achieving the flexibility and rapidity of SOH prediction.

[0098] Step S23: Calculate the Pearson correlation coefficient between the features at different initial voltages and the battery SOH, and compare it with a preset threshold. All the features in the initial voltage segment where the feature is greater than the preset threshold are used as input features, that is:

[0099]

[0100] where F ij represents the i-th feature corresponding to the j-th initial voltage, represents the average value of the corresponding feature; SOH ij represents the SOH under the i-th feature corresponding to the j-th initial voltage, represents its average value; PCC ij represents the Pearson correlation coefficient between the i-th feature corresponding to the j-th initial voltage and SOH.

[0101] Through data screening based on the Pearson correlation coefficient, this application realizes the lightweight of a large amount of feature data extracted from multiple initial voltage segments, thereby improving the flexibility and rapidity of SOH prediction. At the same time, the data screening also eliminates feature data with low correlation, improves the reliability of the input feature data, and thus improves the prediction accuracy of subsequent SOH prediction.

[0102] Step S24: The selected input features form a set and are divided into a training set and a test set. The training set is used for model training, and the test set is used for evaluating the model performance.

[0103] Step S3: Optimize the SVR model based on the DBO (Dung Beetle Optimization Algorithm) algorithm to obtain the DBO-SVR prediction model, and use the training set for training.

[0104] First, construct an SVR model as the SOH prediction model. The relationship between the input and output of the SVR model is expressed as:

[0105] y i (x i ) = ω·φ(x i ) + b

[0106] where x i is the i-th input feature vector, y i is the corresponding output vector; ω and b represent the weight and intercept respectively; φ represents the mapping of the input data in the feature space;

[0107] Introduce slack variables and to get:

[0108]

[0109] Among them, C is the penalty factor, and ε (ε > 0) is the maximum allowable error of regression;

[0110] By introducing Lagrange multipliers and kernel functions, after minimizing the Lagrangian function, the SVR expression of the nonlinear mapping is obtained, that is:

[0111]

[0112] Among them, α i and are Lagrange multipliers, and K(x i , x) = φ(x i )φ(x j ) is the kernel function.

[0113] The kernel function selects the radial basis function RBF, and the formula is expressed as follows:

[0114]

[0115] Among them, σ represents the kernel function parameter.

[0116] The dung beetle optimization algorithm searches for the optimal solution by simulating the search behavior of the dung beetle colony. Correspondingly, this application uses the dung beetle optimization algorithm to optimize the penalty factor and kernel function parameters of the support vector machine regression prediction model to obtain the optimal dung beetle position as the optimal penalty factor and kernel function parameters. Specifically, the DBO optimization steps in step S3 of this application include:

[0117] Step S31: Set the DBO parameters, including the population size, the upper and lower limits of the parameters to be optimized, the maximum number of iterations, and the dimension of the fitness function, so as to initialize the positions of the dung beetles and randomly generate the initial solution.

[0118] Step S32: Establish the fitness function and evaluate the fitness, calculate the fitness values of each dung beetle individual and sort them, so as to obtain the current best position of the dung beetles. The fitness function of this application is established based on the root mean square error, that is:

[0119]

[0120] Among them, Fitness represents the fitness value. The smaller the fitness value, the better the model performance; y i represents the true SOH value; represents the predicted SOH value; n represents the total number of samples in the training set.

[0121] Step S33: Simulate the behavior of each dung beetle individual, update the position after the behavior ends, and calculate the new fitness value.

[0122] Among them, the behaviors of dung beetle individuals include ball rolling, dancing, reproduction, foraging, and stealing, where:

[0123] (1) The position change of the dung beetle responsible for ball rolling is as follows:

[0124] x t (t + 1) = x t (t0 + α × k × x t (t - 1) + b × Δx

[0125] Δx = |x i (t) - X ω |

[0126] Among them, t is the number of iterations; x i is the position of the i-th dung beetle; k is the deflection coefficient;

[0127] b is a random number between (0, 1); α is a natural coefficient, ranging from -1 to 1; X ω is the globally worst position;

[0128] (2) When encountering an obstacle, the dung beetle will reorient its direction through dancing behavior, and its position update formula is as follows:

[0129] x i (t + 1) = x i (t) + tan(θ)|x i (t0 - x i (t - 1)|

[0130] Among them, θ is the deflection angle;

[0131] (3) The update formula for reproduction behavior is as follows:

[0132] Ub * = min(X * × (1 - R0, Ub0

[0133] Lb * = max(X * × (1 - R), Lb)

[0134] Among them, X * is the current best position: Ub * and Lb * are the lower and upper bounds of the spawning area respectively; Ub and Lb are the lower and upper bounds of the optimization problem respectively; R = 1 - t / Tmax, where Tmax represents the maximum number of iterations;

[0135] During the iteration process, the position of the egg ball changes dynamically, and its definition is as follows:

[0136]

[0137] Among them, B i is the position of the oosphere; b 1 , b 2 is a 1×D random variable, where D represents the dimension of the optimization problem;

[0138] (4) Establish the optimal foraging area to guide the foraging behavior of the dung beetles, where the optimal foraging area is defined as follows:

[0139] Ub b = min(X b ×(1 + R), Ub)

[0140] Lb b = max(X b ×(1 - R), Lb)

[0141] Among them, X b is the current best position: Ub b and Lb b are the lower and upper bounds of the best foraging area respectively; Ub and Lb are the lower and upper bounds of the optimization problem respectively; R = 1 - t / Tmax, where Tmax represents the maximum number of iterations;

[0142] The corresponding position update formula for the dung beetles is as follows:

[0143] x i (t + 1) = x i (t) + C 1 ×(x i (t) - Lb b ) + C 2 ×(x i (t) - Ub b )

[0144] Among them, C 1 is a random number obeying the Gaussian distribution; C 2 is a random variable in (0, 1);

[0145] (5) The update formula for the stealing behavior is as follows:

[0146] x i (t + 1) = X b + S×g×(|x i (t) - x * | + |x i (t) - x b |)

[0147] Among them, g is a random variable obeying the Gaussian distribution with a mean of 0 and a variance of 1; S is a constant.

[0148] Step S34: Update the optimal solution, update the individual optimal solution, and update the global optimal position;

[0149] Step S35: Determine whether the maximum number of iterations is reached. If so, obtain the global optimal solution at this time, and use the position of the global optimal solution as the optimal penalty factor and kernel function parameters of the SVR model; otherwise, return to the above-mentioned Step S32.

[0150] After the SVR model is optimized by the DBO algorithm, a DBO-SVR prediction model with optimal model parameters can be obtained. Then, the training set is used for training to train the optimal support vector machine. The trained DBO-SVR prediction model with optimal parameters ensures the SOH prediction accuracy.

[0151] Step S4: Input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

[0152] Embodiment 1:

[0153] The data in this example uses the battery test database open by the University of Maryland, USA. Batteries CS2_35 and CS2_37 are selected as the research objects. The steps of the SOH estimation model and SOH prediction model for optimizing the support vector machine by extracting multi-segment features in the charging stage are as follows:

[0154] 1. Extract features F 1n and F 2n from different initial voltages of different charging cycle data, analyze their correlations with SOH, screen out the initial voltage segments where the features with high correlations are located, and use the features under the corresponding initial voltage segments as the input data set. In this embodiment, the initial voltage range with high correlation is between [3.7, 3.9], and the data in this interval is used as the input data set.

[0155] 2. Set the DBO parameters. The population size of the dung beetle algorithm is 30; the upper limits of the parameters to be optimized are all set to 50, and the lower limits are all set to 0.001; the maximum number of iterations is set to 50, and the dimension dim of the fitness function is 2.

[0156] 3. Construct a support vector machine regression model and optimize it using the DBO algorithm; use the input features related to CS2_35 as the training set, and the input features related to CS2_37 as the test set. In this embodiment, 717 cycle data of battery CS2_35 are used as the training set, and 758 cycle data of battery CS2_37 are used as the test set. After the prediction model training is completed, input the test set vector into the already trained support vector machine regression model to obtain the output prediction value, such as Figures 2 to 6The following shows the SOH prediction curve graph of this embodiment under different initial voltages. It can be seen from the graph that the fitting degree between the predicted value curve and the true value curve is high, indicating that the prediction accuracy of this application is high.

[0157] Table 1 shows the specific prediction error index values under different initial voltages.

[0158]

[0159]

[0160] It can be seen from Table 1 that the maximum RMSE (root mean square error) does not exceed 3.06%, the MAE (mean absolute error) does not exceed 2.22%, and the R 2 (coefficient of determination) is greater than 0.85, indicating that this embodiment has high prediction accuracy.

[0161] According to the method of this application, this application also proposes an SOH prediction system based on multi-feature extraction and optimized SVR. The system includes: a data preprocessing module, a prediction model optimization and training module, and a prediction model testing module, where:

[0162] The data preprocessing module is used to obtain battery constant current charging data, initial voltage division, multi-feature extraction, and the construction of training sets and test sets;

[0163] The prediction model optimization and training module is used to optimize the SVR model using the DBO algorithm to obtain the DBO-SVR prediction model, and train it based on the training set;

[0164] The prediction model testing module is used to input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

[0165] In summary, the present invention first divides the charging voltage of the constant current charging curve into different initial charging voltages, then extracts features at different starting voltages, and finally uses the optimized SVR to estimate the SOH. The present invention improves the problems of long characteristic acquisition time and relatively fixed characteristic starting voltage of existing lithium batteries, improves the flexibility and rapidity of lithium battery SOH estimation, and at the same time improves the SOH prediction accuracy.

[0166] The present invention has been described exemplarily in combination with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above methods. As long as various non-substantive improvements are made by adopting the method concept and technical solution of the present invention; or without improvement, the above concept and technical solution of the present invention are directly applied to other occasions, all are within the protection scope of the present invention.

Claims

1. The SOH prediction method based on multi-feature extraction and optimized SVR is characterized by: The method comprises the following steps: Step S1, obtaining constant current charging data of the lithium battery, and dividing the constant current charging voltage into different initial voltages; Step S2, extracting the battery voltage change and energy change at equal charging capacity under different initial voltages and filtering them to form an input feature set; dividing the input feature set into a training set and a test set; Step S3, optimizing the SVR model based on the DBO algorithm to obtain a DBO-SVR prediction model, and using the training set for training; Step S4: input the test set into the trained DBO-SVR prediction model to obtain the SOH prediction value.

2. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 1 is characterized in that: The step S1 comprises: From the first initial voltage V start At the beginning, a voltage is selected every 0.01V as the initial voltage V n ,Right now: V n =V start +0.0l(n-1); Here, n represents an ordinal number, n=1, 2, 3, ...

3. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 1 or 2, characterized in that: The step S2 comprises: Step S21, calculating the charging capacity according to the charging current and charging time data during constant current charging of the lithium battery; Step S22, extracting battery voltage change and energy change at equal charging capacity under different initial voltages as input features; Step S23, calculate the Pearson correlation coefficient between the features at different initial voltages and the battery SOH and compare them with a preset threshold, and take all features in the initial voltage segment where the features are greater than the preset threshold as input features, that is: Among them, F ij represents the i-th feature corresponding to the j-th initial voltage, Represents the average value of the corresponding feature; SOH ij represents the SOH under the i-th feature corresponding to the j-th initial voltage, Indicates its average value; PCC ij represents the Pearson correlation coefficient between the i-th feature corresponding to the j-th initial voltage and SOH; Step S24: The screened input features form a set and are divided into a training set and a test set.

4. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 3 is characterized in that: The equal charging capacity is set to be 5% of the rated capacity of the battery.

5. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 1 is characterized in that: In step S3, the DBO algorithm is used to optimize the SVR penalty factor and kernel function parameters, and the specific steps include: Step S31, initializing the position of the dung beetle and randomly generating an initial solution; Step S32, establishing a fitness function and evaluating the fitness, calculating the fitness value of each dung beetle individual and sorting them; Step S33, performing behavior simulation on each dung beetle individual, updating the position after the behavior ends, and calculating a new fitness value; Step S34, update the optimal solution, update the individual optimal solution, and update the global optimal position; Step S35, determine whether the maximum number of iterations has been reached, if so, obtain the global optimal solution at this time, and use the position of the global optimal solution as the optimal penalty factor and kernel function parameter of the SVR model; otherwise, return to step S32.

6. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 5 is characterized in that: The fitness function of step S32 is established based on the root mean square error, that is: Among them, y i Indicates the true value of SOH; represents the predicted value of SOH; n represents the total number of samples in the training set.

7. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 5 is characterized in that: In step S33, the behaviors of the dung beetle individuals include rolling, dancing, breeding, foraging, and stealing, wherein: (1) The position of the dung beetle responsible for rolling the ball changes as follows: x t (t+1)=x t (t)+α×k×x t (t-1)+b×Δx Δx=|x i (t)-X ω | Where t is the number of iterations; x i is the position of the i-th dung beetle; k is the deflection coefficient; b is a random number between (0,1); α is a natural coefficient, ranging from -1 to 1; X ω is the worst position globally; (2) When encountering an obstacle, the dung beetle will reorient itself through dancing. The position update formula is as follows: x i (t+1)=x i (t)+tan(θ)|x i (t)-x i (t-1)| Where, θ is the deflection angle; (3) The updating formula of reproductive behavior is as follows: Ub * =min(X * ×(1-R),Ub) Lb * =max(X * ×(1-R),Lb) Among them, X * The current best position: Ub * and Lb * are the lower and upper bounds of the spawning area, respectively; Ub and Lb are the upper and lower bounds of the optimization problem, respectively; R = 1-t / Tmax, Tmax represents the maximum number of iterations; During the iteration process, the position of the egg ball changes dynamically, which is defined as follows: Among them, B i is the position of the egg; b1, b2 are 1×D random variables, D represents the dimension of the optimization problem; (4) Establishing an optimal foraging area to guide the foraging behavior of dung beetles, where the optimal foraging area is defined as follows: Ub b =min(X b ×(1+R),Ub) Lb b =max(X b ×(1-R),Lb) Among them, X b The current best position: Ub b and Lb b are the upper and lower bounds of the optimal foraging area, respectively; Ub and Lb are the upper and lower bounds of the optimization problem, respectively; R = 1-t / Tmax, Tmax represents the maximum number of iterations; The corresponding position update formula of the dung beetle is as follows: x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b ) Among them, C1 is a random number that obeys Gaussian distribution; C2 is a random variable in (0,1); (5) The update formula for stealing behavior is as follows: x i (t+1)=X b +S×g×(∣x i (t)-x * ∣+∣x i (t)-x b ∣) Where g is a random variable that obeys a Gaussian distribution with a mean of 0 and an equation of 1; S is a constant.

8. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 1 is characterized in that: The relationship between the input and output of the SVR model is expressed as: y i (x i )=ω·φ(x i )+b Among them, x i The i-th input feature vector, y i is the corresponding output vector; ω and b represent the weight and intercept respectively; φ represents the mapping of input data in the feature space; Introducing slack variables and have to: Among them, C is the penalty factor, ε (ε>0) is the maximum allowable error of regression; By introducing Lagrange multipliers and kernel functions, after minimizing the Lagrange function, the SVR expression of the nonlinear mapping is obtained, namely: Among them, α i and is the Lagrange multiplier, K(x i ,x)=φ(x i )φ(x j ) is the kernel function.

9. The SOH prediction method based on multi-feature extraction and SVR optimization according to claim 8 is characterized in that: The kernel function uses radial basis function RBF, and the formula is as follows: Among them, σ represents the kernel function parameter.

10. A SOH prediction system based on multi-feature extraction and SVR optimization according to any one of claims 1 to 9, characterized in that: The system includes: a data preprocessing module, a prediction model optimization and training module, and a prediction model testing module, wherein: The data preprocessing module is used to obtain battery constant current charging data, initial voltage division, multi-feature extraction, and construction of training sets and test sets; The prediction model optimization training module is used to optimize the SVR model using the DBO algorithm to obtain a DBO-SVR prediction model, and perform training based on a training set; The prediction model testing module is used to input the test set into the trained DBO-SVR prediction model to measure the SOH prediction value.

Images