Modeling method of popular regularization support vector machine for lithium battery SOC prediction

Abstract

The invention discloses a modeling method of a manifold regularization support vector machine (SVM) for predicting a state of charge (SOC) of a lithium battery, which comprises the following steps of: 1, collecting and preprocessing data; 2, constructing a manifold regularization LSSVM (least square support vector machine) objective function; 3, constructing a data neighborhood relation by using a KNN method; 4, solving a target function to construct a prediction model; and 5, model error calculation and weight updating are carried out. Compared with the prior art, the method has the advantages that an internal manifold structure in data is effectively captured by introducing a manifold learning technology, the fitting capability of a model to complex data is improved, in addition, the influence of data noise on a prediction result is effectively inhibited by combining a robust weight error processing mechanism, and the prediction accuracy is improved. The stability of the model in a complex dynamic environment is improved, the robustness of the model is enhanced, the method can better adapt to complex time-varying working condition requirements, and a new effective method is provided for SOC prediction of the lithium battery.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium batteries, and in particular to a popular regularized support vector machine modeling method for lithium battery SOC prediction. Background Art

[0002] With the rapid development of electric vehicles and renewable energy storage systems, accurate SOC prediction provides a basis for efficient and stable operation of lithium batteries, formulation of battery balancing strategies and intelligent charging, etc. It can effectively prevent battery overcharging or over-discharging, extend battery life, and improve energy utilization efficiency. However, the SOC prediction of lithium batteries is affected by many factors, including battery aging, temperature changes, and changes in charge and discharge rates. These factors lead to the dynamic and complex operating environment of lithium batteries and the existence of uncertain interference. Traditional methods are difficult to achieve ideal prediction effects. Therefore, it is particularly important to develop a method that can accurately predict the SOC of lithium batteries under complex time-varying conditions. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a new popular regularized LSSVM robust modeling method for lithium battery SOC prediction, aiming at improving the accuracy and robustness of the SOC prediction model, in view of the problem that the prediction performance of the existing lithium battery SOC prediction method deteriorates when facing data noise and complex dynamic environment.

[0004] To solve the above technical problems, the present invention provides a technical solution: a popular regularized support vector machine modeling method for lithium battery SOC prediction, which includes the following steps:

[0005] Step 1: Data collection and preprocessing: Collect data sets during the continuous operation of lithium batteries, including key parameters such as voltage, current, and temperature, and perform data normalization preprocessing operations to eliminate the impact of inconsistent dimensions on model training;

[0006] Step 2: Construct the objective function of the popular regularized LSSVM: introduce the popular regularized learning technology to enhance the model's adaptability to complex data by mining the intrinsic manifold structure in the data. At the same time, combine the weight error processing mechanism to improve the model's robustness to noisy data.

[0007] Step 3: Use KNN method to build data neighborhood relationship: Use K-nearest neighbor (KNN) algorithm to build neighborhood relationship between data points and calculate Laplace matrix to describe the local geometric structure of data;

[0008] Step 4: Solve the objective function and build a prediction model: Use the Lagrange multiplier method to solve the objective function containing the popular regularization and weight error terms, and build a SOC robust prediction model of the popular regularized LSSVM;

[0009] Step 5: Model error calculation and weight update: Calculate the model error and iteratively update the weights in the model based on the error feedback until the preset convergence condition is reached.

[0010] Furthermore, in step 1, the current, voltage and temperature data of the lithium battery during continuous operation collected in the experiment are used as input samples xi, and the SOC data obtained by the ampere-hour integration method is used as training label samples yi. The data length is N, and these data are normalized to ensure that they are in the same order of magnitude to facilitate subsequent model training and optimization.

[0011] Furthermore, in step 2, the popular regularization term and the weight error term are introduced into the LSSVM modeling, so that the established model can maintain the essential characteristics of the data and improve the robustness of the model. The robust model of the popular regularized LSSVM is constructed as follows:

[0012]

[0013] Among them, Y T =[y1-e1,y2-e2…y n -e n ]; γ and ψ are regularization factors obtained by cross-validation method; v i ∈(0,1] is the weight of the error, which is obtained through iterative updating.

[0014] Furthermore, in step 3, the Laplace matrix is ​​solved: the Laplace matrix L can be obtained by L = DW, where D is a diagonal matrix, W ij is the element of the similarity matrix W, representing the data point x i and x j The calculation formula for this matrix is:

[0015]

[0016] Among them, ind(x i ,x j )=1, which represents x i and x j are adjacent points to each other; ρ is an adjustable parameter obtained by cross-validation.

[0017] Furthermore, in step 4, the Lagrange multiplier method is used to solve the optimization problem. In order to solve the optimization problem of the objective function, the Lagrange function is constructed as follows:

[0018]

[0019] Among them, α i(i=1,…n) is the Lagrange multiplier. According to the KKT condition, we can get:

[0020]

[0021] Among them, L i is the i-th row of the Laplacian matrix;

[0022] By combining the above, we can get:

[0023]

[0024] The above formula can be further derived as:

[0025]

[0026] in, It is the kernel function;

[0027] The above formula can be further transformed to obtain:

[0028]

[0029] Among them, Y * =[Y n×1 -ψM -1 LY n×1 ],Y n×1 =[y1,y2,…,y n ] T , H=[1,…,1], α=[α1,…α n ] T ,

[0030] Therefore, the parameters α and b can be obtained by the following formula:

[0031]

[0032] Furthermore, in step 4, the popular regularized LSSVM robust prediction output model can be obtained as:

[0033] y=[K(x1,x i ),…,K(x n ,x i )]α+b;

[0034] in, α=(Ω+M -1 ) -1 (Y*-H T b).

[0035] Furthermore, in step 5, the weight of the robust term can be obtained as:

[0036] Error weight v i It is solved based on the error of the corresponding sample and can be obtained by the following formula:

[0037]

[0038] Wherein, c1 and c2 are constant coefficients, s=IQR / (2×0.6745) is a statistical variable; IQR is the interquartile range, that is, the difference between the value at the 75% position and the value at the 25% position of the error sorting sequence.

[0039] After adopting the above structure, the present invention has the following advantages: by introducing popular learning technology, the intrinsic manifold structure in the data is effectively captured, and the model's fitting ability for complex data is improved. In addition, by combining the robust weight error processing mechanism, the influence of data noise on the prediction results is effectively suppressed, the stability of the model in complex dynamic environments is improved, and the robustness of the model is enhanced. This method can better adapt to the needs of complex time-varying working conditions and provides a new and effective method for SOC prediction of lithium batteries. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A schematic diagram of a popular regularized LSSVM robust modeling method for a popular regularized support vector machine modeling method for lithium battery SOC prediction;

[0041] Figure 2 Schematic diagram of added noise data for a popular regularized support vector machine modeling method for lithium battery SOC prediction;

[0042] Figure 3 A schematic diagram of lithium battery SOC prediction results of a popular regularized support vector machine modeling method for lithium battery SOC prediction;

[0043] Figure 4 A schematic diagram of lithium battery SOC prediction error of a popular regularized support vector machine modeling method for lithium battery SOC prediction; DETAILED DESCRIPTION

[0044] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0045] Combined with Figure 1-4 ,like Figure 1 As shown in the figure, this paper provides a robust prediction method for SOC of lithium batteries under the popular regularized LSSVM framework. The richer the sample data collected in the experiment, the higher the prediction accuracy of the model established later. The specific steps include the following:

[0046] Step 1: Data collection and preprocessing. Collect data sets during the continuous operation of lithium batteries, including key parameters such as voltage, current, and temperature, and perform data normalization preprocessing operations to eliminate the impact of inconsistent dimensions on model training.

[0047] In step 1, the current, voltage and temperature data of the lithium battery during continuous operation collected in the experiment are used as input samples xi, and the SOC data obtained by the ampere-hour integration method is used as training label samples yi. The data length is N, and these data are normalized to ensure that they are in the same order of magnitude to facilitate subsequent model training and optimization.

[0048] Step 2: Construct the popular regularized LSSVM objective function and introduce popular regularized learning technology to enhance the model's adaptability to complex data by mining the intrinsic manifold structure in the data. At the same time, combine the weight error processing mechanism to improve the model's robustness to noisy data.

[0049] The purpose of constructing the objective function of the manifold regularized LSSVM includes the following aspects: The first term is the global structural risk. Minimizing this term can improve generalization ability and prevent overfitting. The second term is the error adjustment term, which enables the model to maintain good robustness even in the presence of noise or outliers. The weight is the value obtained from the error distribution of the corresponding sample. The third term is manifold regularization. Introducing manifold regularization in the objective function can enable the established model to maintain the essential structure of the data and improve the approximation ability of the model.

[0050] In step 2, the popular regularization term and the weight error term are introduced into the LSSVM modeling, so that the established model can maintain the essential characteristics of the data and improve the robustness of the model. The robust model of the popular regularized LSSVM is:

[0051]

[0052] Among them, Y T =[y1-e1,y2-e2…y n -e n ]; γ and ψ are regularization factors obtained by cross-validation method; v i ∈(0,1] is the weight of the error, which is obtained through iterative updating.

[0053] Step 3: Use the KNN method to construct the neighborhood relationship of data. The K-nearest neighbor (KNN) algorithm is used to construct the neighborhood relationship between data points and calculate the Laplace matrix to describe the local geometric structure of the data.

[0054] In step 3, solve the Laplace matrix: The Laplace matrix L can be obtained by L = DW, where D is the degree matrix, W ijis the element of the similarity matrix W, representing the data point x i and x j The calculation formula for this matrix is:

[0055]

[0056] Among them, ind(x i ,x j )=1, which represents x i and x j are adjacent points to each other; ρ is an adjustable parameter obtained by cross-validation.

[0057] Step 4: Solve the objective function and build a prediction model. The Lagrange multiplier method is used to solve the objective function containing the popular regularization and weight error terms, and a SOC robust prediction model of the popular regularized LSSVM is constructed.

[0058] In step 4, the Lagrange multiplier method is used to solve the optimization problem. In order to solve the optimization problem of the objective function, the Lagrange function is constructed as follows:

[0059]

[0060] Among them, α i (i=1,…n) is the Lagrange multiplier. According to the KKT condition, we can get:

[0061]

[0062] Among them, L i is the i-th row of the Laplacian matrix.

[0063] By combining the above, we can get:

[0064]

[0065] The above formula can be further derived as:

[0066]

[0067] in, It is the kernel function.

[0068] The above formula can be further transformed to obtain:

[0069]

[0070] Among them, Y * =[Y n×1 -ψM -1 LY n×1 ],Y n×1 =[y1,y2,...,yn ] T , H=[1,…,1], α=[α1,…α n ] T ,

[0071] Therefore, the parameters α and b can be obtained by the following formula:

[0072]

[0073] The popular regularized LSSVM robust prediction output model can be obtained as:

[0074] y=[K(x1,x i ),…,K(x n ,x i )]α+b;

[0075] in, α=(Ω+M -1 ) -1 (Y*-H T b).

[0076] Step 5: Model error calculation and weight update: Calculate the model error and iteratively update the weights in the model based on the error feedback until the preset convergence condition is reached.

[0077] In step 5, the weight of the robust term can be obtained as:

[0078] Error weight v i It is solved based on the error of the corresponding sample and can be obtained by the following formula:

[0079]

[0080] Wherein, c1 and c2 are constant coefficients, s=IQR / (2×0.6745) is a statistical variable; IQR is the interquartile range, that is, the difference between the value at the 75% position and the value at the 25% position of the error sorting sequence.

[0081] 1. The connection between data and models

[0082] Data input: Collect data related to the problem, including feature variables and target variables. Perform normalization preprocessing on the data to improve the training efficiency and prediction accuracy of the model.

[0083] 2. Model Training

[0084] Initialize model parameters: Assign initial values ​​to the model's weight parameters. These initial values ​​can be random or set based on some prior knowledge.

[0085] Constructing the objective function: The objective function usually consists of two parts: the squared loss function and the regularization term. The squared loss function is used to quantify the difference between the model's predicted value and the actual value, while the regularization term is used to control the complexity of the model.

[0086] Solve the optimization problem: Use the Lagrange multiplier method to solve the linear equations to obtain the parameters of the model.

[0087] 3. Calculation Error

[0088] Comparison of predicted values ​​and actual values: Use the trained model to predict the training data and get the predicted values. Compare the predicted values ​​with the actual values ​​and calculate the prediction error.

[0089] These indicators help evaluate the performance of the model and provide a basis for subsequent model optimization.

[0090] 4. Weight Iteration Update

[0091] Based on the calculation error, it is determined whether the model meets the modeling requirements. If the modeling requirements are met, the calculation is terminated. If the requirements are not met, the weight calculation formula is used to calculate the weight, and then the model is solved.

[0092] Repeat the above error calculation and parameter updating steps until the predetermined modeling error is reached, and then the model training is terminated.

[0093] Prediction results:

[0094] In the experiment, the sampling interval of the system was precisely set to 1s to capture the subtle characteristics of battery performance over time. Here, the battery discharge condition was set to DST, and a total of 5,000 data were collected, with the first 4,500 data used for model building and verification. Among them, the first 3,000 data were used as training sets to build the model framework and adjust model parameters, and the remaining 1,500 data were used as test sets to evaluate modeling capabilities. In order to test and verify the robustness of the proposed method, we artificially added various types of noise and outliers to the training and verification data sets. Specifically, it is a mixed perturbation data set composed of Gaussian noise, Rayleigh noise, and abnormal interference values, which is used to simulate extreme or erroneous measurement situations that may be encountered in actual applications. Then, the test set or newly collected data is input into the trained model, and the model calculates and predicts the battery SOC value based on the input data.

[0095] In a specific embodiment, this implementation provides a lithium battery SOC robust prediction method under the popular regularized LSSVM framework, and the specific steps include the following:

[0096] Step 1: Data collection and preprocessing.

[0097] Preprocess the data, use the current, voltage and temperature data of the lithium battery collected in the experiment as the input sample xi, and use the SOC data obtained by the ampere-hour integration method as the training label sample yi, with a data length of N; and normalize these data. (Please provide the normalized calculation expression)

[0098] The ampere-hour integration method is a method for calculating the battery state of charge (SOC). The basic principle is that the change in battery charge is proportional to the integral of the current. By integrating the battery current, the battery SOC can be calculated in real time. The specific formula is:

[0099]

[0100] SOC is the current state of charge of the battery; SOC0 is the initial state of charge of the battery; I is the battery charge and discharge current, t is the time; C is the total capacity of the battery.

[0101] Use the maximum-minimum normalization method: It is a commonly used data preprocessing technique that linearly maps each eigenvalue to [0,1]. Specifically, each eigenvalue x is converted to a new value x', where the minimum value of the feature is mapped to 0 and the maximum value is mapped to 1.

[0102] Assuming there is a feature x, whose minimum value is xmin and maximum value is xmax, the mathematical expression of maximum-minimum normalization is:

[0103]

[0104] Step 2: Construct the popular regularized LSSVM objective function.

[0105] Introducing the popular regularization term The use of regularization in the objective function allows the established model to maintain the essential characteristics of the data and improve the approximation of the model through the similarity between the data. In addition, the weight error term is introduced The data is weighted by statistically analyzing the training error, giving smaller weights to data points that may be noise or outliers, and larger weights to normal data points, which improves the robustness of the model. The objective function of the constructed popular regularized LSSVM model is:

[0106]

[0107] Among them, Y T =[y1-e1,y2-e2…y n -e n ]; γ and ψ are regularization factors obtained by cross-validation method; v i∈(0,1] is the weight of the error, which is obtained through iterative updating.

[0108] Step 3: Use KNN method to build data neighborhood relationship

[0109] The Laplace matrix L can be obtained by L = DW, where D is a diagonal matrix, W ij is the element of the similarity matrix W, representing the data point x i and x j The calculation formula for this matrix is:

[0110]

[0111] Among them, ind(x i ,x j )=1, which represents x i and x j are adjacent points to each other, and ρ is an adjustable parameter obtained by cross-validation.

[0112] Step 4: Solve the objective function to build a prediction model;

[0113] In order to solve the optimization problem of the objective function, the Lagrangian function is constructed as:

[0114]

[0115] Among them, α i (i=1,…n) is the Lagrange multiplier, which can be obtained according to the KKT condition.

[0116]

[0117] Among them, L i is the i-th row of the Laplacian matrix.

[0118] By combining the above four equations, we can get:

[0119]

[0120] We can further get:

[0121]

[0122] in It is the kernel function.

[0123] Further deformation of the above can be obtained:

[0124]

[0125] Among them, Y * =[Yn×1 -ψM -1 LY n×1 ],Y n×1 =[y1,y2,…,y n ] T , H=[1,…,1], α=[α1,…α n ] T ,

[0126] The parameters α and b can be obtained by the following formula:

[0127]

[0128] Based on the above derivation, the SOC robust prediction model of the popular regularized LSSVM can be constructed as follows:

[0129] y=[K(x1,x i ),…,K(x n ,x i )]α+b

[0130] in, α=(Ω+M -1 ) -1 (Y*-H T b).

[0131] Step 5: Model error calculation and weight update.

[0132] Error weight v i It is solved based on the error of the corresponding sample and can be obtained by the following formula:

[0133]

[0134] Among them, c1, c2 are constant coefficients, and s is a statistical variable.

[0135] s can be obtained by the following formula:

[0136] s=IQR / (2×0.6745)

[0137] Among them, IQR is the interquartile range, that is, the difference between the value at the 75% position and the value at the 25% position of the error sorting sequence.

[0138] The present invention takes lithium iron phosphate battery cells as the object, carries out DST operating condition experiments under room temperature conditions, and uses DST operating condition data to predict the SOC of lithium batteries. First, the current, voltage and temperature of the lithium battery are selected as input data, and the data is normalized. In order to fully verify the reliability and robustness of the constructed model, we incorporate an appropriate amount of interference noise into the experimental data, aiming to simulate various complex working conditions that may be encountered in the real world, so as to ensure that the model can also perform well in practical applications. The experimental results show that despite the presence of interference, the model can still predict the SOC of lithium batteries relatively accurately. The prediction error is controlled within a reasonable range, and the prediction results are highly consistent with the actual situation. Figure 2 and 3 As shown in the figure, the predicted curve closely follows the actual SOC changes, which fully demonstrates the effectiveness and practicality of the model.

[0139] The popular regularized LSSVM robust modeling method for lithium battery SOC prediction provided by the present invention can make full use of the advantages of popular regularization technology and weighted theory to effectively build a high-precision prediction model for lithium battery SOC, and can show excellent prediction performance even in a data environment with noise interference. The introduction of this technology provides a new solution for the accurate prediction of lithium battery SOC, and is expected to be widely used in electric vehicles, energy storage systems and other fields.

[0140] The present invention and its implementation methods are described above, and such description is not restrictive, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by it and, without departing from the purpose of the invention, designs a structure and an implementation method similar to the technical solution without creativity, they shall all fall within the protection scope of the present invention.

Claims

1. A popular regularized support vector machine modeling method for lithium battery SOC prediction, characterized by: It includes the following steps: Step 1: Data collection and preprocessing: Collect data sets during the continuous operation of lithium batteries, including key parameters such as voltage, current, and temperature, and perform data normalization preprocessing operations to eliminate the impact of inconsistent dimensions on model training; Step 2: Construct the objective function of the popular regularized LSSVM: introduce the popular regularized learning technology to enhance the model's adaptability to complex data by mining the intrinsic manifold structure in the data. At the same time, combine the weight error processing mechanism to improve the model's robustness to noisy data. Step 3: Use KNN method to build data neighborhood relationship: Use K-nearest neighbor (KNN) algorithm to build neighborhood relationship between data points and calculate Laplace matrix to describe the local geometric structure of data; Step 4: Solve the objective function and build a prediction model: Use the Lagrange multiplier method to solve the objective function containing the popular regularization and weight error terms, and build a SOC robust prediction model of the popular regularized LSSVM; Step 5: Model error calculation and weight update: Calculate the model error and iteratively update the weights in the model based on the error feedback until the preset convergence condition is reached.

2. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: In step 1, the current, voltage and temperature data of the lithium battery during continuous operation collected in the experiment are used as input samples xi, and the SOC data obtained by the ampere-hour integration method is used as training label samples yi. The data length is N, and these data are normalized to ensure that they are in the same order of magnitude to facilitate subsequent model training and optimization.

3. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: In step 2, the popular regularization term and the weight error term are introduced into the LSSVM modeling, so that the established model can maintain the essential characteristics of the data and improve the robustness of the model. The robust model of the popular regularized LSSVM is constructed as follows: s.t.y i =w T φ(x i )+b+e i i=1,2…,n; Among them, Y T =[y1-e1,y2-e2…y n -e n ]; γ and ψ are regularization factors obtained by cross-validation method; v i ∈(0,1] is the weight of the error, which is obtained through iterative updating.

4. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: Solve the Laplace matrix in step 3: The Laplace matrix L can be obtained by L = DW, where D is a diagonal matrix, W ij is the element of the similarity matrix W, representing the data point x i and x j The calculation formula for this matrix is: Among them, ind(x i ,x j )=1, which represents x i and x j are adjacent points to each other; ρ is an adjustable parameter obtained by cross-validation.

5. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: In step 4, the Lagrange multiplier method is used to solve the optimization problem. In order to solve the optimization problem of the objective function, the Lagrange function is constructed as follows: Among them, α i (i=1,…n) is the Lagrange multiplier. According to the KKT condition, we can get: Among them, L i is the i-th row of the Laplacian matrix; By combining the above, we can get: The above formula can be further derived as: in, It is the kernel function; The above formula can be further transformed to obtain: Among them, Y * =[Y n×1 -ψM -1 LY n×1 ],Y n×1 =[y1,y2,…,y n ] T , H=[1,…,1],α=[α1,…α n ] T , Therefore, the parameters α and b can be obtained by the following formula:

6. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: In step 4, the popular regularized LSSVM robust prediction output model can be obtained as: y=[K(x1,x i ),…,K(x n ,x i )]α+b; Among them, α=(Ω+M -1 ) -1 (Y*-H T b).

7. The popular regularized support vector machine modeling method for lithium battery SOC prediction according to claim 1, characterized in that: In step 5, the weight of the robust term can be obtained as: Error weight v i It is solved based on the error of the corresponding sample and can be obtained by the following formula: Wherein, c1 and c2 are constant coefficients, s=IQR / (2×0.6745) is a statistical variable; IQR is the interquartile range, that is, the difference between the value at the 75% position and the value at the 25% position of the error sorting sequence.

Images