Lithium battery health state estimation method based on multi-coefficient fusion

Abstract

The invention provides a lithium battery health state estimation method based on multi-coefficient fusion, so as to improve the problem of insufficient consideration of feature distribution in traditional correlation analysis and improve the evaluation of the service life and safety of a lithium battery. According to the brand-new method, disclosed lithium battery charging data is converted into a 67-dimensional manual feature sequence data set, multiple relations between features of all dimensions and labels are determined through a Pearson's correlation coefficient, a Spearman correlation coefficient and mutual information, feature selection is achieved according to an analytic hierarchy process, and therefore dimensionality reduction is achieved; and the time required for model training is reduced, and the lithium battery health state is estimated through a classical deep learning model residual network, so that the lithium battery health state is quickly analyzed. The lithium battery health state estimation method is superior to a traditional method in the aspect of lithium battery health state estimation, the lithium battery health state estimation error is remarkably reduced, and important practical significance and guidance are provided for lithium battery service life and safety evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery management, and in particular to a method for estimating the health status of a lithium battery. Background Art

[0002] In modern battery technology, health state estimation is crucial to the reliability and performance of battery systems and is widely used in electric vehicles, renewable energy storage, and portable electronic devices. It can provide real-time battery degradation information, predict remaining life, optimize charging strategies, and prevent failures, thereby extending battery life and improving efficiency and safety. However, the battery health state is affected by multiple factors such as electrode chemistry, corrosion, and electrolyte aging. The degradation process is complex and nonlinear, and it is difficult to measure directly. Common estimation methods include experimental methods (such as capacity testing and impedance measurement), model methods (such as electrochemical models and equivalent circuit models), and data-driven methods. Although the experimental method is simple to implement, it is limited by the test environment. The model method can deeply analyze the working principle of the battery but the calculation is complex. The data-driven method analyzes battery historical data through machine learning, provides high accuracy and real-time performance, and has become the current research focus and promoted the innovation of battery management technology. In terms of time series data preprocessing, dimensionality reduction is a common operation. The method based on multi-coefficient fusion has shown advantages in dimensionality reduction and error rate of lithium battery sequence features, so that it can quickly detect the current health state of lithium batteries while only requiring 10-dimensional features, providing a novel idea for lithium battery health state estimation. Summary of the invention

[0003] The purpose of the present invention is to provide a lithium battery health status estimation method based on multi-coefficient fusion, combining three correlation analysis methods, namely Pearson correlation coefficient, Spearman phase relationship, and mutual information, and hierarchical analysis method to calculate the weights of each correlation analysis method, introduce residual network training and perform estimation, and obtain a lithium battery health status estimation with a low error rate.

[0004] In order to achieve the above object, the present invention provides a method for estimating the health status of a lithium battery, comprising:

[0005] Step 1. Convert the publicly available lithium battery charging data into a 67-dimensional manual feature sequence dataset and divide it into a training set and a test set;

[0006] Step 2. Calculate the features and labels of each dimension in the divided training set using three coefficient analysis methods: Pearson correlation coefficient, Spearman correlation coefficient, and mutual information, and sort them according to the calculated correlation values;

[0007] Step 3. The sorted sequence data is trained by increasing the dimensions one by one, and the obtained errors are quantified and analyzed to construct a judgment matrix;

[0008] Step 4. The judgment matrix constructed by error quantification analysis is calculated by analytic hierarchy process to obtain the weight of each analysis method;

[0009] Step 5. Calculate and sort the weights of each analysis method according to the correlation value of the same dimension, and select the strongly correlated dimension features according to the sorting results;

[0010] Step 6. Import the selected dimensional features in the training set into the residual network model for iterative training;

[0011] Step 7. Input the selected dimensional features in the charging sequence data of the lithium battery to be tested into the trained residual network model and output its health status result.

[0012] In step 1, the preprocessing of the input 67-dimensional manual sequence data is performed and it is divided into a training set and a test set. The division ratio is that among a group of batteries with the same charging strategy, the sequence data of one battery is selected as the test set, and the rest are used as the training set.

[0013] In step 2, the three correlation indicators of Pearson correlation coefficient, Spearman correlation coefficient and mutual information are calculated for the features and labels of each dimension in the divided training set, and they are sorted according to the calculated correlation values.

[0014] In step 3, the model is trained by gradually increasing the dimensions according to the sorted feature sequence. In each training stage, new feature dimensions are gradually introduced, and their contribution to the overall performance is evaluated through the model. Avoid the interference of redundant features and optimize the learning process of the model. After the training is completed, the error between the model prediction results and the real data is quantitatively analyzed, and the judgment matrix is ​​further constructed by calculating the distribution and change trend of the error value. This judgment matrix not only reflects the contribution of each feature to the error, but also provides a quantitative basis for the subsequent weight allocation. By analyzing the judgment matrix, the most influential features can be accurately identified, and the feature selection and model adjustment strategies can be optimized, thereby improving the overall prediction accuracy and the generalization ability of the model.

[0015] In step 4, the weights of each analysis method are calculated by the hierarchical analysis method based on the judgment matrix constructed by the error quantification analysis.

[0016] In step 5, the correlation values ​​of the same dimension are weighted and sorted according to the weights of each analysis method, and the strongly correlated dimension features are selected based on the sorting results.

[0017] In step 6, the dimension features selected in the training set are imported into the residual network model for training.

[0018] In step 7, the test set is input into the trained residual network model, and the estimated result of the health status of the lithium battery is output. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of a lithium battery health status estimation method based on multi-coefficient fusion proposed by the present invention;

[0020] Figure 2 It is a schematic diagram of the residual network model used in the lithium battery health status estimation method proposed in the present invention;

[0021] Figure 3 The method proposed in the present invention uses the 10-dimensional sequence features selected by the hierarchical analysis method to draw a scatter plot of a randomly selected battery based on the 10-dimensional sequence features;

[0022] Figure 4 It is a bar graph of the mean absolute error and mean absolute percentage error obtained by the method proposed in the present invention for the test set in the published lithium battery 67-dimensional manual feature sequence data set. DETAILED DESCRIPTION

[0023] The following describes the embodiments of the present invention by means of specific examples.

[0024] See also Figure 1 , a schematic diagram of a lithium battery health status estimation method based on multi-coefficient fusion proposed in the present invention. The Pearson correlation coefficient mentioned in the figure is calculated by the formula:

[0025]

[0026] Where X i represents the extracted n-th dimension artificial feature, Y represents the battery health status value to be predicted, cov(X i ,Y) represents the covariance between them, σ Y are their respective variances, and ρ represents the correlation value. The Spearman correlation coefficient mentioned in the figure is calculated by:

[0027]

[0028] where d i For X i , Y i The mutual information mentioned in the figure is calculated by the formula:

[0029]

[0030] Where P(x,y) is the joint probability distribution of variables X and Y, P(x) and P(y) are the marginal probability distributions of variables X and Y respectively, and I(X;Y) represents the correlation value. The AHP steps mentioned in the figure include:

[0031] Step 1. Calculate the Pearson correlation coefficient PC, Spearman correlation coefficient SC and mutual information MI of each dimension feature according to the above three formulas to form a coefficient matrix Cor;

[0032] Step 2. To eliminate the differences in correlation values ​​calculated by the three correlation coefficient methods and the fusion differences caused by different data ranges, PC, SC and MI are normalized by the normalization formula:

[0033]

[0034] In the formula, x is the correlation value, min(x) and max(x) are the minimum and maximum values ​​of the correlation values, and x′ is the normalized correlation value after processing.

[0035] Step 3. Calculate the weight matrix W according to the hierarchical analysis method, and calculate the score S = Cor × W of each dimension by weighted average method.

[0036] Step 4. Add dimension features in turn according to the scores of each dimension to estimate the battery health status.

[0037] Step 5. Select the dimensional feature corresponding to the minimum value in the battery health status training estimation error as the final selected feature.

[0038] like Figure 2 As shown, for the above Figure 4 Figure 1 shows the residual network structure of the machine learning model trained in , which contains a preprocessing module, five residual modules, and a predictor. Each residual block consists of two convolutional layers, a batch normalization layer, a pooling layer, and key shortcut connections, while the predictor consists of two linear layers with a ReLU activation layer in between.

[0039] like Figure 3 As shown, a scatter plot of 10 dimensional features selected according to the Pearson correlation coefficient, Spearman correlation coefficient, and mutual information level hierarchical analysis method includes 10 dimensional features of charge temperature skewness, CC segment voltage kurtosis, capacity during 3.9-4.0V, standard deviation of charging voltage, charging voltage kurtosis, charge temperature kurtosis, average voltage during 3.9-4.0V, cumulative charge energy, cumulative charge capacity, and CC segment energy.

[0040] like Figure 4As shown in the figure, the mean absolute error (MAE) of the first 10 dimensional features after sorting based on the selected 10 dimensional features, as well as the correlation values ​​calculated by the Pearson correlation coefficient, the Spearman correlation coefficient, and the mutual information, and the mean absolute percentage error (MAPE) of the left figure and the right figure, respectively. The calculation formula for the mean absolute error (MAE) is:

[0041]

[0042] y in the formula i and Represent the true value and estimated value of the battery health status respectively, and N represents the number of samples. The mean absolute percentage error MAPE calculation formula is:

[0043]

[0044] y in the formula i and They represent the true value and estimated value of the battery health status respectively, and N represents the number of samples.

Claims

1. A lithium battery health status estimation method based on multi-coefficient fusion, comprising the following steps: Step 1. Convert the publicly available lithium battery charging data into a 67-dimensional manual feature sequence dataset and divide it into a training set and a test set; Step 2. Calculate the features and labels of each dimension in the divided training set using three coefficient analysis methods: Pearson correlation coefficient, Spearman correlation coefficient, and mutual information, and sort them according to the calculated correlation values; Step 3. The sorted sequence data is trained by increasing the dimensions one by one, and the obtained errors are quantified and analyzed to construct a judgment matrix; Step 4. The judgment matrix constructed by error quantification analysis is calculated by analytic hierarchy process to obtain the weight of each analysis method; Step 5. Calculate and sort the weights of each analysis method according to the correlation value of the same dimension, and select the strongly correlated dimension features according to the sorting results; Step 6. Import the selected dimensional features in the training set into the residual network model for iterative training; Step 7. Input the selected dimensional features in the charging sequence data of the lithium battery to be tested into the trained residual network model and output its health status result.

2. A lithium battery health status estimation method based on multi-coefficient fusion according to claim 1, characterized in that: The input 67-dimensional manual sequence data is preprocessed and divided into a training set and a test set. The division ratio is that among a group of batteries with the same charging strategy, the sequence data of one battery is selected as the test set, and the rest are used as the training set.

3. The method for estimating the health status of a lithium battery based on multi-coefficient fusion according to claim 1, characterized in that: For the features and labels of each dimension in the divided training set, the three correlation indicators of Pearson correlation coefficient, Spearman correlation coefficient and mutual information are calculated respectively, and they are sorted according to the calculated correlation values.

4. According to claim 1, a lithium battery health state estimation method based on multi-coefficient fusion is characterized in that, according to the sorted feature sequence, the model training is carried out in a way of gradually increasing the dimension, and the obtained error is quantitatively analyzed to construct a judgment matrix, and the steps include: 4-1. Input the sorted feature sequence into the model for training to obtain the error matrix under different correlation coefficient methods. 4-2. Calculate five quantization intervals based on the minimum error and maximum error in the error matrix. 4-3. According to the five quantitative intervals in step 4-2, find the error intervals corresponding to the Pearson correlation coefficient and the Spearman correlation coefficient, use the Pearson error interval as a comparison, determine the importance of each correlation coefficient method according to the error size, and thus construct a judgment matrix.

5. The method for estimating the health status of a lithium battery based on multi-coefficient fusion according to claim 1, characterized in that: According to the judgment matrix constructed by error quantification analysis, the weights of each analysis method are calculated by analytic hierarchy process.

6. The method for estimating the health status of a lithium battery based on multi-coefficient fusion according to claim 1, characterized in that: According to the weight of each analysis method, the correlation values ​​of the same dimension are weighted and sorted, and the dimension features that are strongly correlated with the label are selected based on the sorting results.

7. The method for estimating the health status of a lithium battery based on multi-coefficient fusion according to claim 1, characterized in that: Import the dimension features selected in the training set into the residual network model for training.

8. The method for estimating the health status of a lithium battery based on multi-coefficient fusion according to claim 1, characterized in that: The test set is input into the trained residual network model to output the estimated result of the health status of the lithium battery.

9. A lithium battery health status estimation method according to claim 2, characterized in that: The 67-dimensional manual sequence data includes 9 groups of battery sequence data with different charging strategies.

Images