Lithium battery state-of-health estimation method based on Grubrum mixing field

Abstract

The invention provides a lithium battery state-of-health estimation method based on a Grubrum mixing field, so as to improve the problem of insufficient fusion of multiple data sources in an actual application scene of a lithium battery, improve the service life of the lithium battery and improve the safety evaluation of the lithium battery. According to the innovative method provided by the invention, based on the preprocessing of the time sequence data of the lithium battery data set, the time sequence data is converted into the image data set through the Grubm mixing field, different sequences are effectively fused and imported into the traditional deep learning model Resnet, and the model estimation performance and generalization ability are remarkably improved. The method has an obvious effect in the aspect of lithium battery health state estimation, the error of lithium battery health state estimation is remarkably reduced, and the potential of multi-mode fusion in the aspect of robust battery health prediction is emphasized. And important practical and guiding significance is provided for evaluating the service life and the safety of the lithium battery.

Description

Technical Field

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

[0002] Lithium batteries are widely used in many fields, such as electric vehicles, portable electronic devices, etc. And lithium batteries usually use fast charging with a large current rate to shorten the charging time. However, using a large current will accelerate battery aging and may cause high temperatures, increasing the safety risk of thermal runaway. Therefore, accurately estimating the state of health of a lithium battery is crucial for ensuring the safe and reliable operation of the battery and optimizing the battery management system.

[0003] Early estimations of the state of health of batteries mostly used methods based on physical models and simple measurements, but these methods have certain limitations. In recent years, data-driven methods have become a research hotspot. Researchers use a large amount of battery operation data and rely on advanced technologies such as neural networks and regression analysis to explore the complex relationship between battery performance and the state of health. At the same time, more and more research has begun to combine the internal aging mechanism of the battery and integrate it into the model to improve the estimation accuracy. For example, by analyzing the influence of factors such as chemical reactions and material structure changes inside the battery on battery performance and integrating the information into the state-of-health estimation model, but this method has a heavy computational burden and is difficult to perform large-scale real-time monitoring and rapid prediction. Although certain progress has been made, there are still many problems at present, such as uneven data quality, insufficient adaptability of the model due to complex and changeable actual application scenarios, and high computational costs of some models that are difficult to meet the real-time requirements. The estimation of the state of health of a lithium battery based on the image coding of the Gram mixed field shows advantages. It can effectively fuse multi-source time series data, generate color image data, and use excellent deep learning models in the field of image recognition for operation, providing a new idea for the estimation of the state of health of a lithium battery. Summary of the Invention

[0004] The purpose of the present invention is to provide a method for estimating the state of health of a lithium battery based on the Gram mixed field, which combines the charging current, voltage, and temperature information in a lithium battery pack with 9 different charging strategies, performs image coding of the Gram mixed field on it, generates image data, and imports it into the Resnet model for training and estimation to obtain the estimation advantage of the state of health of lithium batteries with different charging strategies and reduce the error rate of the estimation of the state of health of lithium batteries.

[0005] To achieve the above purpose, the present invention provides a method for estimating the state of health of a lithium battery, including:

[0006] Step 1. Generate color image data by performing image coding of the Gram mixed field on the current, voltage, temperature, and time series data in the publicly available lithium battery charging data;

[0007] Step 2. Divide the generated color image data into a training set and a test set;

[0008] Step 3. Build the backbone network Resnet, set the pre-training parameters, and perform iterative training on the Resnet network model.

[0009] Step 4. Input the image data set of the lithium battery to be measured into the trained Resnet network model, and output the estimated result of its health state.

[0010] In Step 1, perform Gram addition field and Gram subtraction field calculations on the lithium battery time series data to generate a matrix, then perform numerical mapping, convert it into a grayscale image, and superimpose the grayscale images to generate color image data.

[0011] In Step 2, for the image data set generated after preprocessing the lithium battery time series data, divide it into a training set and a test set. The division ratio is that among the batteries with the same charging strategy, select the image data of one battery as the test set, and the rest as the training set.

[0012] In Step 3, build the backbone network Resnet, which lies in the bipartite process of ResBlock. By connecting the correlation between shallow features and deep features, and then connecting the MLP fully connected layer network, the performance of the model is improved.

[0013] In Step 4, input the generated image data test set into the trained Resnet network model, and output the estimated result of the lithium battery health state. Description of the Drawings

[0014] Figure 1 is a schematic flow chart of a method for estimating the health state of a lithium battery based on a Gram mixed field proposed by the present invention;

[0015] Figure 2 is a schematic diagram of the Gram mixed field image coding method proposed by the present invention;

[0016] Figure 3 is a schematic diagram of the color images corresponding to different charge and discharge cycle data of the battery generated by the image coding proposed by the present invention;

[0017] Figure 4 is a schematic diagram of the backbone network model Resnet model proposed by the present invention;

[0018] Figure 5 is a schematic diagram of ResBlock in the backbone network proposed by the present invention; Detailed Embodiments

[0019] The following describes the implementation manners of the present invention through specific examples.

[0020] Please refer to Figure 1 , which is a schematic flowchart of a method for estimating the state of health of a lithium battery based on a Gram hybrid field proposed by the present invention.

[0021] As Figure 2 shown, specifically, the data preprocessing of converting the lithium battery time series data through the image encoding of the Gram hybrid field to generate a color image, and the important steps include:

[0022] Step 1. Normalize the time series data X(1×128) of the three dimensions of charging current, voltage, and temperature, and the normalization formula:

[0023]

[0024] Step 2. Calculate the matrices GASF and GADF through the Gram addition field and Gram subtraction field formulas for the normalized data, and then calculate the weighted matrix A, where the formula is:

[0025]

[0026] Step 3. Numerically map the calculated matrix A to the interval [0, 255] to obtain the matrix GAF-Hybrid i,j , and its mapping formula:

[0027]

[0028] Step 4. The calculated matrix GAF-Hybrid i,j , where the value range is [0, 255], can be converted into a 128×128 grayscale image, and the three grayscale images of current, voltage, and temperature are superimposed according to the red, green, and blue channels to form a color image for training and testing by a deep learning model.

[0029] As Figure 3 shown, a backbone network based on Resnet is built, and finally an MLP multi-layer perceptron model structure is connected.

[0030] As Figure 4 shown, color images encoded by the Gram hybrid field corresponding to different charge and discharge cycle data of the battery are generated.

[0031] As Figure 5 shown, the internal structure of the core module ResBlock in the Resnet model includes the original shallow data and the data after passing through the convolutional layer.

Claims

1. A method for estimating the health status of a lithium battery based on a Gram mixing field, characterized in that: The steps include: Step 1. Perform a Gram mixture field preprocessing process on the voltage, current, and temperature time series features of the time series data of each lithium battery charging cycle, and convert them into a color image data set with three channels. Step 2. Convert all into color image datasets, and use the charging cycles of a single battery in each group of batteries as the test set and the charging cycles of the remaining batteries as the training set. Step 3. Build the traditional backbone network Resnet, import the preprocessed lithium battery training set, and perform iterative training on it. Step 4. Input the preprocessed lithium battery test set into the trained Resnet network model and output the estimated results.

2. A method for estimating the health status of a lithium battery based on a Gram mixing field according to claim 1, characterized in that the Gram mixing field calculation and channel merging of lithium battery time series data to generate a color image comprises the following steps: 2-1. The time series data of lithium battery charging current, voltage and temperature, whose dimensions are all [1,128], are generated into matrices with dimensions [128,128] through Gram field addition and Gram field subtraction operations. Each set of data contains two matrices, for a total of 6 matrices. 2-2. After processing, the three sets of data with dimensions [128,128] are weighted summed with weights 0.5 and 0.5 to generate a matrix with data dimensions [128,128]. 2-3. Perform data processing on the weighted sum matrix, map its values ​​to the interval [0,255], and generate three grayscale images of the current, voltage and temperature data of each charging cycle of the lithium battery by grayscale image generation. 2-4. The three grayscale images generated by each charging cycle are superimposed on the red, yellow, and blue channels to generate a color image with a dimension of [3,128,128], where 3 represents the number of channels and 128 represents the width and height of the image, respectively.

3. The method for estimating the health status of a lithium battery based on a Gram mixing field according to claim 1, characterized in that: The traditional deep learning model Resnet is built, in which the core module is the traditional ResBlock, which is the binary branch process of ResBlock and contains 6 ResBlocks.

4. The method for estimating the health status of a lithium battery based on a Gram mixing field according to claim 1, characterized in that: The Resnet network in the model structure is connected to two MLP networks consisting of linear layers, and the generated image training set is trained. The test set is input into the trained model to output the estimated result of the health status of the lithium battery.

5. The method for estimating the health status of a lithium battery based on a Gram mixing field according to claim 1, characterized in that: The time series characteristics of voltage, current and temperature of the time series data of the charging cycle of the lithium battery include 9 groups of battery sequence data with different charging strategies.

Images