A multi-feature fusion method for lithium battery SOC estimation considering stress

Abstract

The present invention relates to a lithium battery SOC estimation method for multi-feature fusion taking stress into account, which obtains lithium battery feature information under various test conditions, constructs a decomposition model, performs variable mode decomposition on the lithium battery feature information under various conditions based on the decomposition model, obtains modal components corresponding to the lithium battery feature information, constructs a screening model, screens the lithium battery feature information and its corresponding modal components based on the screening model, obtains an input feature combination set, constructs a lithium battery SOC estimation model, inputs each lithium battery feature information and modal components corresponding to the input feature combination set into the lithium battery SOC estimation model for training, outputs a lithium battery SOC estimation value, and realizes accurate prediction of the remaining power of the lithium battery under non-steady-state conditions, which not only helps to optimize battery management strategies and extend battery life, but also has important significance for improving the overall safety and reliability of electric vehicles and energy storage systems.

Description

Technical Field

[0001] The present invention relates to the technical field of new energy vehicle battery estimation, and in particular to a lithium battery SOC estimation method that takes stress into account and integrates multiple features. Background Art

[0002] Under the background of green and low-carbon energy development, new energy vehicles are developing rapidly. Lithium batteries have been widely favored by the new energy industry due to their advantages such as high specific energy, low self-discharge, long cycle life and fast charging speed. However, the safety of lithium batteries in use has become one of the primary problems that need to be solved in their large-scale application process. BMS is born out of demand. Lithium battery SOC represents the state of charge of the battery and is one of the important parameters of BMS, affecting the battery's charge and discharge control strategy and safety. With the development of advanced sensing technology, using data collected by multi-sensor to improve the estimation accuracy of battery SOC is one of the important methods in the current SOC estimation strategy. The existing collected signals mainly include voltage, current, temperature, stress, etc. Among them, stress signals are difficult to capture due to their uncertainty and instantaneousness, and the battery mechanical characteristics are difficult to capture. Conventional analysis methods are difficult to describe the relationship between this instantaneous non-steady-state signal and SOC, and it is difficult to meet the needs of SOC estimation.

[0003] With the development of the times and the advancement of science and technology, the existing technology has achieved certain results in combining machine learning and battery mechanical information to estimate battery SOC. However, the existing research still has key research deficiencies such as the simple structure of the deep learning model and the unclear relationship between the multivariate signal factors and the current state of the battery. First, the deep learning model in the existing technology is a simple form with a fully connected layer. The stress signal is directly filtered and put into the model for training, but the filtering causes some original non-steady-state signal data loss, affecting the SOC estimation accuracy. In addition, due to the uncertainty and instantaneousness of the mechanical signal, the characteristics are difficult to capture. Conventional analysis methods are difficult to describe the relationship between this instantaneous non-steady-state signal and the SOC of the lithium battery. Whether the battery SOC estimation taking into account the mechanical characteristics is more accurate, more efficient, and more robust is analyzed. In view of the above shortcomings, a lithium battery SOC estimation method taking into account multiple feature fusion of stress is proposed to solve the above problems. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a lithium battery SOC estimation method that takes stress into account and integrates multiple features, which has the advantages of good lithium battery SOC estimation accuracy and solves the problem of poor lithium battery SOC estimation accuracy.

[0005] To achieve the above object, the present invention provides the following technical solution: a lithium battery SOC estimation method taking into account stress multi-feature fusion, comprising the following specific steps:

[0006] S1: Obtain lithium battery characteristic information under various test conditions;

[0007] S2: Construct a decomposition model, and perform variable modal decomposition on the characteristic information of lithium batteries under various working conditions based on the decomposition model to obtain the modal components corresponding to the characteristic information of lithium batteries;

[0008] S3: construct a screening model, and screen the lithium battery feature information and its corresponding modal components based on the screening model to obtain an input feature combination set;

[0009] S4: Construct a lithium battery SOC estimation model, input the feature information and modal components of each lithium battery corresponding to the input feature combination set into the lithium battery SOC estimation model for training, and output the lithium battery SOC estimation value.

[0010] Furthermore, in step S1, characteristic information of lithium batteries under a plurality of test conditions is obtained. There are four plurality of test conditions, namely FUDS, DST, US06 and BJDST. The characteristic information of lithium batteries includes voltage, current, temperature and stress information.

[0011] Furthermore, the decomposition model is constructed in step S2, and variable modal decomposition is performed on the characteristic information of the lithium battery under various working conditions based on the decomposition model to obtain the modal components corresponding to the characteristic information of the lithium battery, including the following sub-steps:

[0012] S2.1: Construct a decomposition model based on variable mode decomposition algorithm;

[0013] S2.2: Optimize the parameters of the decomposition model based on the whale optimization algorithm, iteratively update to obtain the hyperparameter combination, and configure the hyperparameter combination into the decomposition model.

[0014] Further, the step S2.2 performs parameter optimization on the decomposition model based on the whale optimization algorithm, iterates and updates to obtain a hyperparameter combination, and configures the hyperparameter combination into the decomposition model, including the following sub-steps:

[0015] S2.2.1: Initialize the whale population position;

[0016] S2.2.2: Input the voltage and stress information under various test conditions into the decomposition model to perform variable mode decomposition and obtain the modal components corresponding to the voltage and stress information;

[0017] S2.2.3: Based on the original signal sequence, calculate the fitness through position search, and iteratively update the initial position of the whale population to obtain the optimal hyperparameter combination ;

[0018] S2.2.4: Combining the best hyperparameters Configured into the decomposition model, the voltage and stress information under various test conditions are modally decomposed to obtain the modal components corresponding to the voltage and stress information.

[0019] Furthermore, the step S3 constructs a screening model, and based on the screening model, screens the lithium battery feature information and its corresponding modal components to obtain an input feature combination set, including the following sub-steps:

[0020] S3.1: Calculate the maximum information coefficient between the modal components corresponding to voltage and stress and the SOC of the lithium battery;

[0021] S3.2: Based on the energy moment calculation formula of the modal components corresponding to voltage and stress, calculate the energy moment proportion and energy entropy of the modal component signals corresponding to voltage and stress respectively;

[0022] S3.3: According to the principle that the larger the maximum information coefficient, the stronger the correlation; the smaller the energy entropy, the more ordered the system, and the higher the characteristic purity, the maximum information coefficient and energy entropy are combined to select the characteristic mode component signals corresponding to the voltage and stress with close correlation and stable energy;

[0023] S3.4: Based on Gaussian process regression, establish the maximum likelihood function of voltage, stress and its corresponding characteristic modal component signal, current, temperature and lithium battery SOC;

[0024] S3.5: Determine the optimal input feature combination based on the Bayesian Information Criterion.

[0025] Further, the calculation expression of the maximum information coefficient between the modal components corresponding to the calculated voltage and stress and the lithium battery SOC in step S3.1 is:

[0026] ;

[0027] ;

[0028] , , ;

[0029] In the formula, is the set of modal components corresponding to the voltage, is the set of modal components corresponding to stress, , ... are the modal components corresponding to the voltage under FUDS, DST, US06 and BJDST conditions, , ... is the modal component corresponding to the stress under FUDS, DST, US06 and BJDST conditions, and the number of decomposed modes , is the optimal hyperparameter combination obtained by optimizing the decomposition in step S2.2.3 , that is, the number of variable mode decomposition layers Decide, is a random variable and random variables The maximum information coefficient between The value range of is (0, 1). If The closer the value of is to 0, the more and random variables The less closely the relationship between The closer the value of is to 1, the more and random variables The closer the relationship between for Two variables in , The maximum information coefficient of (IMFV, SOC), for The maximum information coefficient of two variables X, Y (IMFS, SOC), for The number of grid divisions in the axis direction, The number of grid divisions in the Y-axis direction. For the sample The dimension calculation formula is: .

[0030] Further, the energy moment calculation formula of the modal components corresponding to the voltage and stress described in step S3.2 is used to calculate the energy moment proportion and energy entropy of the modal component signals corresponding to the voltage and stress respectively:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] In the formula, is the energy moment of the lithium battery SOC, For time point, For the last time point, is a positive integer, For the voltage modal components, , is the voltage residual The energy moment, is the sampling frequency, For the energy torch of stress components, is the stress residual The energy moment, , is the set of voltage modal components and residual energy moments and stress modal components and residual energy moments, , is a positive integer, For the The proportion of energy moments, For the The energy entropy of an energy moment.

[0040] Furthermore, in step S3.4, based on Gaussian process regression, the calculation expression of the maximum likelihood function of voltage, stress and its corresponding characteristic modal component signal, current, temperature and lithium battery SOC is established as follows:

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] ;

[0046] ;

[0047] In the formula, is the mean function, is the square exponential covariance function, and the mean of the standard Gaussian process regression is 0; is the signal variance, which represents the amplitude of the covariance function; is the long-scale parameter, which determines the smoothness of the covariance function. is noise and satisfies normal distribution, and its variance is , is the identity matrix, SOC time series calibrated by ampere-hour integration method , the sorting method is arranged according to the observation time, is the input vector sequence , the hyperparameter set is , is the maximum likelihood function equation.

[0048] Further, the calculation expression for determining the best input feature combination based on the Bayesian Information Criterion in step S3.5 is:

[0049] ;

[0050] ;

[0051] ;

[0052] In the formula, The maximum likelihood estimation equation is obtained for Gaussian process regression, is the number of variable parameters, is the sample size, is the BIC factor, is the qth actual value, is the output value of the qth model evaluation process, is the variance of the difference between the actual value and the estimated value.

[0053] Further, the lithium battery SOC estimation model is constructed in step 4, and each lithium battery feature information and modal component corresponding to the input feature combination set is input into the lithium battery SOC estimation model for training, and the lithium battery SOC estimation value is output, including the following sub-steps:

[0054] Step 4.1: Construct a lithium battery SOC estimation model based on a gated recurrent unit network;

[0055] Step 4.2: Construct a data set. The original current, temperature data signal, voltage, pressure and their decomposed signals are screened by the feature screening model to form a feature combination as the input combination of the SOC estimation model. The SOC calculated by the ampere-hour integration method is used as the output, which together constitutes the training data set;

[0056] Step 4.3: Input the data set into the lithium battery SOC estimation model for training, and obtain the trained lithium battery SOC estimation model for outputting the lithium battery SOC estimation value.

[0057] Compared with the prior art, the technical solution of this application has the following beneficial effects:

[0058] This stress-based multi-feature fusion lithium battery SOC estimation method significantly improves the accuracy, efficiency and robustness of lithium battery SOC estimation through fine signal decomposition, feature screening and fusion, and the construction of an efficient estimation model. It effectively removes noise and redundant information in the original signal and extracts key features that are highly correlated with the lithium battery SOC, thereby achieving accurate prediction of the remaining power of the lithium battery under non-steady-state conditions. This not only helps to optimize battery management strategies and extend battery life, but is also of great significance for improving the overall safety and reliability of electric vehicles and energy storage systems, and provides strong support for the widespread application of new energy technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 It is a flow chart of the present invention;

[0060] Figure 2 A decomposition model is constructed for the present invention, and variable modal decomposition is performed on the characteristic information of the lithium battery under various working conditions based on the decomposition model to obtain a modal component flow chart corresponding to the characteristic information of the lithium battery;

[0061] Figure 3 The present invention optimizes the parameters of the decomposition model based on the whale optimization algorithm, iteratively updates to obtain the hyperparameter combination, and configures the hyperparameter combination into the decomposition model. Flowchart;

[0062] Figure 4 A screening model is constructed for the present invention, and the lithium battery feature information and its corresponding modal components are screened based on the screening model to obtain an input feature combination set flow chart;

[0063] Figure 5 A lithium battery SOC estimation model is built for the present invention, and each lithium battery feature information and modal component corresponding to the input feature combination set is input into the lithium battery SOC estimation model for training, and a lithium battery SOC estimation value flow chart is output. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] See also Figure 1-5 In this embodiment, a lithium battery SOC estimation method taking into account stress and multi-feature fusion includes the following specific steps:

[0066] S1: Obtain lithium battery characteristic information under various test conditions;

[0067] S2: Construct a decomposition model, and perform variable modal decomposition on the characteristic information of lithium batteries under various working conditions based on the decomposition model to obtain the modal components corresponding to the characteristic information of lithium batteries;

[0068] S3: construct a screening model, and screen the lithium battery feature information and its corresponding modal components based on the screening model to obtain an input feature combination set;

[0069] S4: Construct a lithium battery SOC estimation model, input the feature information and modal components of each lithium battery corresponding to the input feature combination set into the lithium battery SOC estimation model for training, and output the lithium battery SOC estimation value.

[0070] In step S1, characteristic information of lithium batteries under various test conditions is obtained. There are four various test conditions, namely FUDS, DST, US06 and BJDST. The characteristic information of lithium batteries includes voltage, current, temperature and stress information.

[0071] In step S2, a decomposition model is constructed, and variable modal decomposition is performed on the characteristic information of the lithium battery under various working conditions based on the decomposition model to obtain modal components corresponding to the characteristic information of the lithium battery, including the following sub-steps:

[0072] S2.1: Construct a decomposition model based on variable mode decomposition algorithm;

[0073] S2.2: Optimize the parameters of the decomposition model based on the whale optimization algorithm, iteratively update to obtain the hyperparameter combination, and configure the hyperparameter combination into the decomposition model.

[0074] In step S2.2, the decomposition model is optimized based on the whale optimization algorithm, the hyperparameter combination is obtained by iterative update, and the hyperparameter combination is configured in the decomposition model, including the following sub-steps:

[0075] S2.2.1: Initialize the whale population position;

[0076] S2.2.2: Input the voltage and stress information under various test conditions into the decomposition model to perform variable mode decomposition and obtain the modal components corresponding to the voltage and stress information;

[0077] S2.2.3: Based on the original signal sequence, calculate the fitness through position search, and iteratively update the initial position of the whale population to obtain the optimal hyperparameter combination ;

[0078] S2.2.4: Combining the best hyperparameters Configured into the decomposition model, the voltage and stress information under various test conditions are modally decomposed to obtain the modal components corresponding to the voltage and stress information;

[0079] Variable mode decomposition can decompose non-stationary signals and has good adaptability to the local time-frequency structure of the signal. On the one hand, it removes the noise of the original signal, and on the other hand, it can accurately separate the modal functions of different time scales in the data and explore the characteristic laws of the signal time domain scale. The calculation expression is:

[0080] ;

[0081] ;

[0082] ;

[0083] ;

[0084] In the formula, is the number of variable modal decomposition layers, which determines the number of modal components obtained by decomposition. is the voltage modal component obtained by decomposing the voltage signal, is the frequency center of the voltage modal component, is the number of decomposed voltage modal components, is the stress modal component obtained by decomposing stress information, is the frequency center of the stress modal component, is the number of decomposed stress modal components, is the gradient, is the Dirac function, For time point, To measure voltage, To measure stress information;

[0085] Using Lagrangian operator and the quadratic penalty factor Improve the above formula, the quadratic penalty factor It plays a role in balancing the signal-to-noise ratio and controlling the decomposition accuracy. It can obtain the Lagrangian constraint equations corresponding to the voltage and stress respectively, and solve the optimal solution of the finite variable problem. , , and To update, the calculation expression is:

[0086] ;

[0087] ;

[0088] ;

[0089] ;

[0090] ;

[0091] ;

[0092] The variable mode decomposition has a large amount of calculation. When processing long-term signals, the decomposition results may be unstable. The optimal value of the parameter is found through the optimization mechanism of the whale optimization algorithm. The calculation expression is:

[0093] ;

[0094] ;

[0095] ;

[0096] ;

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] In the formula, is the number of iterations, is the current optimal position vector, and are coefficient vectors, is the distance between the optimal whale and the current whale, and is a random number between 0 and 1, is the amount that decreases linearly from 2 to 0, is the maximum number of iterations, is the distance between the first whale and its prey, is a constant parameter that controls the shape of the helix. is a random number, The value range is [-1, 1], is the current optimal solution position, is the current location, is a random number, The value range of is [0, 1], is a randomly selected whale position, when When the whales are forced to move away from their prey, they use randomly generated Updates the positions of all whales to find more suitable prey.

[0104] In step S3, a screening model is constructed, and the lithium battery feature information and its corresponding modal components are screened based on the screening model to obtain an input feature combination set, which includes the following sub-steps:

[0105] S3.1: Calculate the maximum information coefficient between the modal components corresponding to voltage and stress and the SOC of the lithium battery. The calculation expression is:

[0106] ;

[0107] ;

[0108] , , ;

[0109] In the formula, is the set of modal components corresponding to the voltage, is the set of modal components corresponding to stress, , ... are the modal components corresponding to the voltage under FUDS, DST, US06 and BJDST conditions, , ... is the modal component corresponding to the stress under FUDS, DST, US06 and BJDST conditions, and the number of decomposed modes , is the optimal hyperparameter combination obtained by optimizing the decomposition in step S2.2.3 , that is, the number of variable mode decomposition layers Decide, is a random variable and random variables The maximum information coefficient between The value range of is (0, 1). If The closer the value of is to 0, the more and random variables The less closely the relationship between The closer the value of is to 1, the more and random variables The closer the correlation is, the more variables , Discretize in two-dimensional space, and then divide the current two-dimensional space along Axis and The axis is divided into a certain number of intervals, and then the scattered points are observed to fall into each square. for Two variables in , The maximum information coefficient of (IMFV, SOC), for The maximum information coefficient of two variables X, Y (IMFS, SOC), for The number of grid divisions in the axis direction, The number of grid divisions in the Y-axis direction. For the sample The dimension calculation formula is: ;

[0110] S3.2: Based on the energy moment calculation formula of the modal components corresponding to voltage and stress, the energy moment proportion and energy entropy of the modal component signals corresponding to voltage and stress are calculated respectively. The calculation expressions are:

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] ;

[0117] ;

[0118] ;

[0119] In the formula, is the energy moment of the lithium battery SOC, For time point, For the last time point, is a positive integer, For the voltage modal components, , is the voltage residual The energy moment, is the sampling frequency, For the energy torch of stress components, is the stress residual The energy moment, , is the set of voltage modal components and residual energy moments and stress modal components and residual energy moments, , is a positive integer, For the The proportion of energy moments, For the The energy entropy of each energy moment;

[0120] S3.3: According to the principle that the larger the maximum information coefficient, the stronger the correlation; the smaller the energy entropy, the more ordered the system, and the higher the characteristic purity, the maximum information coefficient and energy entropy are combined to select the characteristic mode component signals corresponding to the voltage and stress with close correlation and stable energy;

[0121] S3.4: Based on Gaussian process regression, the maximum likelihood function of voltage, stress and its corresponding characteristic modal component signals, current, temperature and lithium battery SOC is established to prepare for determining the optimal feature combination based on the Bayesian information criterion. The calculation expression is:

[0122] ;

[0123] ;

[0124] ;

[0125] ;

[0126] ;

[0127] ;

[0128] In the formula, is the mean function, is the square exponential covariance function, and the mean of the standard Gaussian process regression is 0; is the signal variance, which represents the amplitude of the covariance function; is the long-scale parameter, which determines the smoothness of the covariance function. is noise and satisfies normal distribution, and its variance is , is the identity matrix, SOC time series calibrated by ampere-hour integration method , the sorting method is arranged according to the observation time, is the input vector sequence , the hyperparameter set is , is the maximum likelihood function equation;

[0129] S3.5: Determine the best input feature combination based on the Bayesian information criterion. The smaller the Bayesian factor, the better the feature combination. The calculation expression is:

[0130] ;

[0131] ;

[0132] ;

[0133] In the formula, The maximum likelihood estimation equation is obtained for Gaussian process regression, is the number of variable parameters, is the sample size, is the BIC factor, is the qth actual value, is the output value of the qth model evaluation process, is the variance of the difference between the actual value and the estimated value.

[0134] In step 4, a lithium battery SOC estimation model is constructed, and each lithium battery feature information and modal component corresponding to the input feature combination set is input into the lithium battery SOC estimation model for training, and the lithium battery SOC estimation value is output, including the following sub-steps:

[0135] Step 4.1: Construct a lithium battery SOC estimation model based on a gated recurrent unit network;

[0136] Step 4.2: Construct a data set. The original current, temperature data signal, voltage, pressure and their decomposed signals are screened by the feature screening model to form a feature combination as the input combination of the lithium battery SOC estimation model. The SOC calculated by the ampere-hour integration method is used as the output, which together constitutes the training data set;

[0137] Step 4.3: Input the data set into the lithium battery SOC estimation model for training, and obtain the trained lithium battery SOC estimation model for outputting the lithium battery SOC estimation value.

[0138] Step 4: The structure of the lithium battery SOC estimation model includes an input layer, a convolution layer, a recurrent neural network layer, an attention mechanism layer, and an output layer. The input layer is used to receive the raw data of the lithium battery SOC estimation model, the convolution layer is used to extract the raw data features, the recurrent neural network layer is used to learn the raw data features, the attention mechanism layer is used to adjust the weights of the raw data features, and the output layer is used to output the lithium battery SOC estimation value.

[0139] Step 4.2 constructs a data set. The original current, temperature data signal, voltage, pressure and their decomposed signals are screened by the feature screening model to form a feature combination as the input combination of the lithium battery SOC estimation model. The SOC calculated by the ampere-hour integration method is used as the output. Together, they constitute a training data set. The calculation expression is:

[0140] ;

[0141] ;

[0142] ;

[0143] ;

[0144] ;

[0145] ;

[0146] ;

[0147] ;

[0148] ;

[0149] In the formula, The time step is Where , The time step is The current at is the sampling interval, is the current maximum capacity of the battery, is the input vector set, is the voltage under working condition, is the current under working condition, is the temperature under working condition, is the stress under working condition, is the characteristic voltage modal component under working conditions, is the characteristic stress modal component under the working condition, for Activation function, To reset the gate, To update the gate, is the input of the recurrent neural network layer, is the weight value, is the temporary hidden state of the hidden layer in the recurrent neural network layer, is the output of the hidden layer in the recurrent neural network layer, , , is the bias term, is the hyperbolic tangent activation function, is the normalization function, is the bias term, is the attention weight, is the attention weight of the ith input at time t, is the context vector of the attention mechanism, is an adjustable parameter used to balance the influence of the state of the recurrent neural network layer itself and the context vector. is the hidden state after correction;

[0150] Four evaluation criteria for evaluating the performance of lithium battery SOC estimation model, namely the coefficient of determination , mean absolute error , mean square error , root mean square error , the calculation formula is:

[0151] ;

[0152] ;

[0153] ;

[0154] .

[0155] The working principle of the above embodiment is:

[0156] This stress-based multi-feature fusion lithium battery SOC estimation method significantly improves the accuracy, efficiency and robustness of lithium battery SOC estimation through fine signal decomposition, feature screening and fusion, and the construction of an efficient estimation model. It effectively removes noise and redundant information in the original signal and extracts key features that are highly correlated with the lithium battery SOC, thereby achieving accurate prediction of the remaining power of the lithium battery under non-steady-state conditions. This not only helps to optimize battery management strategies and extend battery life, but is also of great significance for improving the overall safety and reliability of electric vehicles and energy storage systems, and provides strong support for the widespread application of new energy technologies.

[0157] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0158] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A lithium battery SOC estimation method taking into account stress multi-feature fusion, characterized in that: The specific steps include: S1: Obtain lithium battery characteristic information under various test conditions; S2: Construct a decomposition model, and perform variable modal decomposition on the characteristic information of lithium batteries under various working conditions based on the decomposition model to obtain the modal components corresponding to the characteristic information of lithium batteries; S3: construct a screening model, and screen the lithium battery feature information and its corresponding modal components based on the screening model to obtain an input feature combination set; S3.1: Calculate the maximum information coefficient between the modal components corresponding to voltage and stress and the SOC of the lithium battery; S3.2: Based on the energy moment calculation formula of the modal components corresponding to voltage and stress, calculate the energy moment proportion and energy entropy of the modal component signals corresponding to voltage and stress respectively; S3.3: According to the principle that the larger the maximum information coefficient, the stronger the correlation; the smaller the energy entropy, the more ordered the system, and the higher the characteristic purity, the maximum information coefficient and energy entropy are combined to select the characteristic mode component signals corresponding to the voltage and stress with close correlation and stable energy; S3.4: Based on Gaussian process regression, establish the maximum likelihood function of voltage, stress and its corresponding characteristic modal component signal, current, temperature and lithium battery SOC; S3.5: Determine the best input feature combination based on the Bayesian Information Criterion; S4: Construct a lithium battery SOC estimation model, input the feature information and modal components of each lithium battery corresponding to the input feature combination set into the lithium battery SOC estimation model for training, and output the lithium battery SOC estimation value.

2. A lithium battery SOC estimation method taking into account stress and multi-feature fusion as claimed in claim 1, characterized in that: In step S1, characteristic information of lithium batteries under a variety of test conditions is obtained. There are four kinds of test conditions, namely FUDS, DST, US06 and BJDST. The characteristic information of lithium batteries includes voltage, current, temperature and stress information.

3. The lithium battery SOC estimation method of multi-feature fusion taking into account stress as claimed in claim 1, characterized in that: The step S2 constructs a decomposition model, performs variable modal decomposition on the characteristic information of the lithium battery under various working conditions based on the decomposition model, and obtains the modal components corresponding to the characteristic information of the lithium battery, including the following sub-steps: S2.1: Construct a decomposition model based on variable mode decomposition algorithm; S2.2: Optimize the parameters of the decomposition model based on the whale optimization algorithm, iteratively update to obtain the hyperparameter combination, and configure the hyperparameter combination into the decomposition model.

4. A lithium battery SOC estimation method taking into account stress multi-feature fusion as claimed in claim 3, characterized in that: The step S2.2 described in the above method optimizes the parameters of the decomposition model based on the whale optimization algorithm, iterates and updates to obtain the hyperparameter combination, and configures the hyperparameter combination into the decomposition model, including the following sub-steps: S2.2.1: Initialize the whale population position; S2.2.2: Input the voltage and stress information under various test conditions into the decomposition model to perform variable mode decomposition and obtain the modal components corresponding to the voltage and stress information; S2.2.3: Based on the original signal sequence, calculate the fitness through position search, and iteratively update the initial position of the whale population to obtain the optimal hyperparameter combination ; S2.2.4: Combining the best hyperparameters Configured into the decomposition model, the voltage and stress information under various test conditions are modally decomposed to obtain the modal components corresponding to the voltage and stress information.

5. The lithium battery SOC estimation method of multi-feature fusion taking stress into account according to claim 1, characterized in that: The calculation expression for the maximum information coefficient between the modal components corresponding to the calculated voltage and stress and the lithium battery SOC described in step S3.1 is: ； ； ， ， In the formula, is the set of modal components corresponding to the voltage, is the set of modal components corresponding to stress, , ... are the modal components corresponding to the voltage under FUDS, DST, US06 and BJDST conditions, , ... is the modal component corresponding to the stress under FUDS, DST, US06 and BJDST conditions, and the number of decomposed modes , is the optimal hyperparameter combination obtained by optimizing the decomposition in step S2.2.3 , that is, the number of variable mode decomposition layers Decide, is a random variable and random variables The maximum information coefficient between The value range of is (0, 1). If The closer the value of is to 0, the more and random variables The less closely the relationship between The closer the value of is to 1, the more and random variables The closer the relationship between for Two variables in , The maximum information coefficient of (IMFV, SOC), for The maximum information coefficient of two variables X, Y (IMFS, SOC), for The number of grid divisions in the axis direction, The number of grid divisions in the Y-axis direction. For the sample The dimension calculation formula is: .

6. The lithium battery SOC estimation method of multi-feature fusion taking into account stress as claimed in claim 1, characterized in that: Based on the energy moment calculation formula of the modal components corresponding to the voltage and stress described in step S3.2, the calculation expressions for the energy moment proportion and energy entropy of the modal component signals corresponding to the voltage and stress are respectively calculated as follows: In the formula, is the energy moment of the lithium battery SOC, For time point, For the last time point, is a positive integer, For the voltage modal components, , is the voltage residual The energy moment, is the sampling frequency, For the energy torch of stress components, is the stress residual The energy moment, , is the set of voltage modal components and residual energy moments and stress modal components and residual energy moments, , is a positive integer, For the The proportion of energy moments, For the The energy entropy of an energy moment.

7. The lithium battery SOC estimation method of claim 1 taking into account stress and multi-feature fusion, characterized in that: In step S3.4, based on Gaussian process regression, the calculation expression of the maximum likelihood function of voltage, stress and its corresponding characteristic modal component signal, current, temperature and lithium battery SOC is established as follows: In the formula, is the mean function, is the square exponential covariance function, and the mean of the standard Gaussian process regression is 0; is the signal variance, which represents the amplitude of the covariance function; is the long-scale parameter, which determines the smoothness of the covariance function. is noise and satisfies normal distribution, and its variance is , is the identity matrix, SOC time series calibrated by ampere-hour integration method , the sorting method is arranged according to the observation time, is the input vector sequence , the hyperparameter set is , is the maximum likelihood function equation.

8. The lithium battery SOC estimation method of multi-feature fusion taking into account stress as claimed in claim 1, characterized in that: The calculation expression for determining the best input feature combination based on the Bayesian Information Criterion in step S3.5 is: In the formula, The maximum likelihood estimation equation is obtained for Gaussian process regression, is the number of variable parameters, is the sample size, is the BIC factor, is the qth actual value, is the output value of the qth model evaluation process, is the variance of the difference between the actual value and the estimated value.

9. The lithium battery SOC estimation method of multi-feature fusion taking into account stress as claimed in claim 1, characterized in that: The lithium battery SOC estimation model is constructed in step 4, and each lithium battery feature information and modal component corresponding to the input feature combination set is input into the lithium battery SOC estimation model for training, and the lithium battery SOC estimation value is output, including the following sub-steps: Step 4.1: Construct a lithium battery SOC estimation model based on a gated recurrent unit network; Step 4.2: Construct a data set. The original current, temperature data signal, voltage, pressure and their decomposed signals are screened by the feature screening model to form a feature combination as the input combination of the SOC estimation model. The SOC calculated by the ampere-hour integration method is used as the output, which together constitutes the training data set; Step 4.3: Input the data set into the lithium battery SOC estimation model for training, and obtain the trained lithium battery SOC estimation model for outputting the lithium battery SOC estimation value.

Images