Search

CN-121978532-A - Lithium battery SOC rapid estimation method under low-temperature working condition

CN121978532ACN 121978532 ACN121978532 ACN 121978532ACN-121978532-A

Abstract

The invention belongs to the field of lithium battery SOC estimation, and particularly relates to a lithium battery SOC rapid estimation method under a low-temperature working condition, which comprises the following steps: S1, constructing a sample data set, S2, screening key features based on ReliefF algorithm, S3, estimating SOC based on a machine learning regression model, S4, training and estimating the model, and by combining ReliefF feature screening and the machine learning model, realizing high-precision and quick estimation of the SOC of the lithium battery in a low-temperature environment, and effectively solving the problems of low precision and complex calculation of the traditional method under severe cold working conditions.

Inventors

  • YANG YI
  • HU NIAOQING
  • CHENG ZHE
  • ZHANG LUN

Assignees

  • 中国人民解放军国防科技大学

Dates

Publication Date
20260505
Application Date
20260120

Claims (10)

  1. 1. The lithium battery SOC rapid estimation method under the low-temperature working condition is characterized by comprising the following steps of: S1, constructing a sample data set: Processing the time sequence signals by adopting a sliding window method for setting the size and the step length of a window, independently calculating the statistical characteristics of the voltage and the current of each window, generating a high-dimensional characteristic vector, and taking an SOC value corresponding to the last time stamp of the window as a target label of the characteristic vector, thereby constructing a sample data set consisting of the characteristic vector and the SOC label; s2, key feature screening based on ReliefF algorithm: Extracting all high-dimensional feature vectors from the sample data set constructed in the step S1 to form a feature set, carrying out feature importance assessment on the feature set by adopting ReliefF algorithm, and screening out features with weights higher than a set threshold value according to feature weight ranking obtained by calculation to form a core feature subset for SOC estimation; S3, SOC estimation based on a machine learning regression model: Taking the core feature subset obtained in the step S2 as input, and adopting at least one machine learning regression model to carry out SOC estimation, wherein the machine learning regression model comprises any one or combination of a linear regression model, a support vector regression model, a Lasso regression model, a random forest model, a BP neural network model or LightGBM model; S4, model training and evaluation: Training the model of S3 on the low-temperature working condition data set, evaluating the model precision by using the mean square error and the decision coefficient, and comprehensively evaluating the model performance by combining the training time.
  2. 2. The method for rapidly estimating SOC of a lithium battery under a low temperature condition as set forth in claim 1, wherein the step of estimating the feature importance using ReliefF algorithm in S2 includes: s21, initial weights are distributed for all the features in the feature set; S22, randomly selecting a sample R from the sample data set, and searching k nearest neighbors in the same type sample and different types of samples respectively; S23, iteratively updating the weight of each feature according to the distance difference of the sample R, the similar neighbors and the different similar neighbors on each feature and the category prior probability; And S24, after the preset iteration times are reached, sorting according to the final weights of the features, and screening out features with weights higher than the set threshold value to form the core feature subset.
  3. 3. The method for rapidly estimating the SOC of the lithium battery under the low-temperature working condition as set forth in claim 2, wherein the step of estimating the feature importance by using ReliefF algorithm in S2 includes: S21, resetting the initial weight of each feature to 0 for a feature set containing n features; s22, randomly selecting one sample R from the sample dataset comprising m samples, which has a feature vector (f 1 , f 2 , F n ), searching k nearest neighbors H j in the same kind of sample of the sample R, and searching k nearest neighbors M j in different kinds of samples; S23, updating the weight of each feature according to the feature difference between the sample R and the neighbor of the sample R, wherein for each feature f i , the weight updating model is as follows: Wherein, P (C) is the prior probability of the class C, that is, the ratio of the number of samples of the class C to the total number of samples of the class to which the non-R belongs, f i is the ith feature of the sample R, H ji is the ith feature of the jth same-class nearest neighbor H j , M ji is the ith feature of the jth different-class nearest neighbor M j , k is the number of same-class/different-class nearest neighbors, z is the sampling number, and the calculation formulas of the discrete features diff (f i ,R, H j ) and diff (f i ,R, M j ) are as follows: Wherein H j is the j-th similar nearest neighbor, and M j is the j-th different similar nearest neighbor; And S24, after the preset iteration times are reached, sorting according to the final weights of the features, and screening out features with weights higher than the set threshold value to form the core feature subset.
  4. 4. The method for rapidly estimating SOC of a lithium battery under a low temperature condition as set forth in any one of claims 1 to 3, wherein in S1: The multi-dimensional statistical features extracted from the voltage and current time sequence signals are 18 dimensions, and the sum of the multi-dimensional statistical features forms 36-dimensional initial features, wherein the 18-dimensional statistical features comprise maximum values, minimum values, average values, median, upper quartiles, lower quartiles, variances, standard deviations, kurtosis, skewness, root mean square, waveform factors, peak factors, pulse factors, margin factors, average values of absolute values, energy and variation coefficients; The window size of the sliding window method is set to 500 data points, and the step size is set to 100 data points.
  5. 5. The method for rapidly estimating SOC of a lithium battery under low-temperature conditions as claimed in any one of claims 1 to 3, wherein in the third step, the machine learning regression model used includes at least LightGBM model.
  6. 6. A method for rapidly estimating SOC of a lithium battery under low temperature conditions according to any one of claims 1 to 3, wherein the low temperature conditions involved in S1 and S4 are 0 ℃ environments; the construction of the sample dataset and the training and evaluation of the model are based on battery test data in a 0 ℃ environment.
  7. 7. The method of claim 6, wherein the model training and evaluation in S4 is based on a specific low temperature dataset collected from a plurality of driving Cycle tests in a 0 ℃ environment and divided into a training set and an independent test set according to driving Cycle types, wherein the training set comprises data of US06, HWFET, UDDS, LA, neural Network and Cycle 4 driving cycles, and the test set comprises data of Cycle 1, cycle 2 and Cycle 3 driving cycles.
  8. 8. The method of claim 7, wherein the data set in the environment of 0 ℃ has a sampling time step of 0.1 seconds, and the battery is charged to 4.2V with a constant current of 1C rate and is charged to a constant voltage with a cutoff current of 50mA until the current is lower than the cutoff current after each driving cycle test.
  9. 9. A lithium battery SOC rapid estimation method device under a low-temperature working condition, characterized in that the device comprises: a sample data set construction module, configured to perform S1: Processing the time sequence signals by adopting a sliding window method for setting the size and the step length of a window, independently calculating the statistical characteristics of the voltage and the current of each window, generating a high-dimensional characteristic vector, and taking an SOC value corresponding to the last time stamp of the window as a target label of the characteristic vector, thereby constructing a sample data set consisting of the characteristic vector and the SOC label; The key feature screening module is used for S2, namely key feature screening based on ReliefF algorithm: Extracting all high-dimensional feature vectors from the sample data set constructed in the step S1 to form a feature set, carrying out feature importance assessment on the feature set by adopting ReliefF algorithm, and screening out features with weights higher than a set threshold value according to feature weight ranking obtained by calculation to form a core feature subset for SOC estimation; the SOC estimation module is used for carrying out S3, namely SOC estimation based on a machine learning regression model: Taking the core feature subset obtained in the step S2 as input, and adopting at least one machine learning regression model to carry out SOC estimation, wherein the machine learning regression model comprises any one or combination of a linear regression model, a support vector regression model, a Lasso regression model, a random forest model, a BP neural network model or LightGBM model; the model training and evaluating module is used for carrying out S4: Training the model of S3 on the low-temperature working condition data set, evaluating the model precision by using the mean square error and the decision coefficient, and comprehensively evaluating the model performance by combining the training time.
  10. 10. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any of claims 1 to 8 when the computer program is executed.

Description

Lithium battery SOC rapid estimation method under low-temperature working condition Technical Field The invention belongs to the field of lithium battery SOC estimation, and particularly relates to a lithium battery SOC rapid estimation method under a low-temperature working condition. Background The lithium ion battery has become a core power source of the electric automobile due to the advantages of high energy density, long cycle life and the like, and the state of charge (SOC) is used as a key index for reflecting the residual electric quantity of the battery, so that the estimation accuracy directly determines the running safety, the cruising reliability and the service life of the battery of the electric automobile, and meanwhile, the mileage anxiety of a user can be effectively relieved, and the abuse behaviors of the battery such as overcharging, overdischarging and the like are avoided. At present, an SOC estimation method mainly comprises a model method, a data driving method and a mixing method, and although the prior art has a certain progress, the method has significant limitations that a part of the model method relies on complex electrochemical mechanism modeling and has strict requirements on parameter matching, and the traditional data driving method relies on massive monitoring data, and does not conduct targeted screening on redundant characteristics in original voltage and current signals, so that the model calculation complexity is high, the instantaneity is poor, and the dynamic operation scene of an electric automobile is difficult to adapt. The electrochemical characteristics of the lithium ion battery can be obviously changed in a low-temperature environment (such as 0 ℃), the ion diffusion rate is reduced, the internal resistance is increased, the difficulty of SOC estimation is further increased, and the problem of accuracy downslide easily occurs in the conventional method in the scene. Therefore, developing an SOC estimation scheme that combines low-temperature adaptability, estimation accuracy and calculation efficiency becomes a key requirement for promoting large-scale application of electric vehicles under complex climate conditions. Based on the background, the patent screens the high-dimensional characteristics of the battery voltage and current signals by adopting ReliefF algorithm, eliminates redundant information and focuses on key characteristics which are strongly related to the SOC, and then combines various machine learning methods to perform contrast optimization, so that the balance between high precision and high efficiency is finally realized. Disclosure of Invention The invention aims to provide a lithium battery SOC rapid estimation method under a low-temperature working condition. By combining ReliefF feature screening and a machine learning model, high-precision and rapid estimation of the lithium battery SOC is realized in a low-temperature environment, and the problems of low precision and complex calculation of the traditional method under severe cold working conditions are effectively solved. The invention provides a lithium battery SOC rapid estimation method under a low-temperature working condition, which comprises the following steps: S1, constructing a sample data set: Processing the time sequence signals by adopting a sliding window method for setting the size and the step length of a window, independently calculating the statistical characteristics of the voltage and the current of each window, generating a high-dimensional characteristic vector, and taking an SOC value corresponding to the last time stamp of the window as a target label of the characteristic vector, thereby constructing a sample data set consisting of the characteristic vector and the SOC label; s2, key feature screening based on ReliefF algorithm: Extracting all high-dimensional feature vectors from the sample data set constructed in the step S1 to form a feature set, carrying out feature importance assessment on the feature set by adopting ReliefF algorithm, and screening out features with weights higher than a set threshold value according to feature weight ranking obtained by calculation to form a core feature subset for SOC estimation; S3, SOC estimation based on a machine learning regression model: Taking the core feature subset obtained in the step S2 as input, and adopting at least one machine learning regression model to carry out SOC estimation, wherein the machine learning regression model comprises any one or combination of a linear regression model, a support vector regression model, a Lasso regression model, a random forest model, a BP neural network model or LightGBM model; S4, model training and evaluation: Training the model of S3 on the low-temperature working condition data set, evaluating the model precision by using the mean square error and the decision coefficient, and comprehensively evaluating the model performance by combining the training time. Further,