CN118244127B - A lithium-ion battery health status assessment method based on graph convolution - Google Patents

Abstract

The invention provides a lithium ion battery health state evaluation method based on graph convolution, which comprises the steps of manufacturing a voltage capacity pixel graph, an electric quantity change rate pixel graph and a capacity difference pixel graph according to battery data and forming a training set; training ResNet a network by using a training set, and obtaining a trained lithium ion battery health state evaluation model; obtaining a feature diagram to be measured according to the data of the power battery to be measured; sending the feature diagram to be tested into a trained lithium ion battery health state evaluation model; outputting a health state evaluation result of the lithium ion battery; the invention comprehensively utilizes the information such as the electric quantity change rate, the capacity difference and the like, can improve the model performance and the generalization capability, and can more accurately evaluate the health condition and the performance of the lithium ion battery.

Description

A lithium-ion battery health status assessment method based on graph convolution

Technical Field

The present invention relates to the technical field of battery health status assessment, and in particular to a lithium-ion battery health status assessment method based on graph convolution.

Background Art

As fossil fuel consumption and carbon emissions become more and more of a concern, lithium-ion batteries (LIBs) are widely used in the field of new energy vehicles, and the state of health (SOH) is one of the important indicators for evaluating the battery status. Typically, when the SOH drops below 80%, it is assumed that the battery has exhausted its maximum service life and must be replaced. In this case, accurately monitoring the changes in SOH becomes critical because it directly marks the performance and life of the battery, highlighting that fast and accurate estimation of the SOH of lithium-ion batteries has become an important research topic. Convolutional neural networks (CNNs) have made breakthroughs in battery health assessment, and their excellent image feature extraction capabilities have attracted much attention and are essential for battery data analysis.

The current problems faced by battery life estimation are mainly the following three points. 1. Battery health is affected by many factors. Accurately representing its characteristics requires a lot of domain knowledge and professional experience, and complex feature engineering is required. 2. Deep learning requires sufficient training data. Since degradation testing usually takes a long time to complete, the number of available training samples is limited. 3. Some batteries may not have obvious visual features, and more in-depth research and analysis of battery characteristics are required. It is difficult to maintain good adaptability on different types of batteries.

In the invention patent of a power battery pack data storage and health assessment method based on visual technology with publication number CN116756351A, it is proposed to use the charging segment data to draw the single cell voltage-capacity curve image, convert the single cell voltage-capacity curve image into a grayscale image and compress it to replace the original floating point data. This method only uses the characteristic image of voltage-capacity; however, the rate and properties of the chemical reaction inside the battery are crucial to understanding the working status of the battery, and more comprehensive battery health status and performance information can be obtained from this.

Summary of the invention

In view of the problems in the prior art, the present invention provides a lithium-ion battery health status assessment method based on graph convolution, which aims to comprehensively utilize information such as capacity difference and the rate of change of actual capacity relative to charging voltage to improve model performance and generalization ability, and more accurately assess the health status and performance of lithium-ion batteries.

A method for evaluating the health status of a lithium-ion battery based on graph convolution comprises the following steps:

Step 1: Training a lithium-ion battery health status assessment model, including the following steps;

Step 1.1: Perform constant current and constant voltage cycle charging on several lithium-ion batteries, and obtain battery data including cycle sequence, charging voltage sequence, actual capacity sequence, and health status;

Step 1.2: Create a voltage capacity pixel map, a charge change rate pixel map, and a capacity difference pixel map based on the battery data. The voltage capacity pixel map is used to record the corresponding relationship between the charging voltage and the actual capacity in each cycle. The charge change rate pixel map is used to record the change rate of the actual capacity relative to the charging voltage in each cycle. The capacity difference pixel map is used to record the capacity change relative to the initial actual capacity of the battery in each cycle.

Step 1.3: Stack the corresponding voltage capacity pixel map, charge change rate pixel map, and capacity difference pixel map to form a three-channel feature map, thereby obtaining a training set consisting of a feature map and a health status;

Step 1.4: Use the training set to train the ResNet50 network and obtain the trained lithium-ion battery health status assessment model;

Step 2: Obtain a characteristic graph to be tested according to the power battery data to be tested;

Step 3: Send the feature map to be tested into the trained lithium-ion battery health status assessment model;

Step 4: Output the health status assessment result of the lithium-ion battery.

Further, the steps of making a voltage capacity pixel map are:

The actual capacity sequence of the cycle is normalized as a whole, a color range is set, black pixels are used to represent the maximum value of the actual capacity, and white pixels are used to represent the minimum value of the actual capacity; then, a 1*100 pixel map is drawn, and the numerical range of the horizontal axis of the pixel map gradually increases from 3.5V to 4.2V and is used to represent the charging voltage. The pixel color is filled according to the actual capacity sequence corresponding to the charging voltage sequence. The rule for filling the pixel color is the correspondence between the actual capacity value and the black to white gradient. Finally, a 1*100 voltage-capacity correspondence map is generated, the voltage-capacity correspondence map is stacked up and down, and a 100*100 voltage-capacity pixel map is obtained.

Further, the steps of making a pixel graph of the rate of change of electric quantity are as follows:

The overall normalization method for the cyclic charge change rate sequence is to use black pixels to represent the maximum value of the charge change rate, and white pixels to represent the minimum value of the charge change rate; then, a 1*100 pixel map is drawn, and the numerical range of the horizontal axis of the pixel map gradually increases from 3.5V to 4.2V and is used to represent the charging voltage. The pixel color is filled according to the charge change rate corresponding to the charging voltage sequence. The rule for filling the pixel color is the correspondence between the actual capacity value and the black to white gradient. A 1*100 voltage charge change rate correspondence map is generated, and the voltage charge change rate correspondence map is stacked up and down to obtain a 100*100 charge change rate pixel map.

Further: power change rate for:

in, Indicates the number of cycles, Described as the slope of the charging voltage as the actual capacity changes, is the change in actual battery capacity, is the change in battery charging voltage, The actual capacity of the battery at the current moment, Indicates the actual capacity of the battery at the previous moment. The current battery charging voltage, The battery charging voltage at the previous moment.

Further, the steps of making a capacity difference pixel map are:

First, construct a list of charging voltage values with a step size of 0.007V, divide 3.5V to 4.2V into 100 intervals, and record the actual capacity difference between the first cycle and the current cycle as , the specific calculation formula is as follows:

in, The difference between the actual capacity of the first cycle and the actual capacity of the nth cycle, Indicates the actual capacity of the lithium battery in the first cycle. Indicates the actual capacity of the lithium battery at the nth cycle;

The calculated actual capacity difference is filled into the capacity difference sequence of the corresponding cycle. The capacity difference sequence of the cycle is normalized as a whole, a color range is set, black pixels are used to represent the maximum capacity difference, and white pixels are used to represent the minimum capacity difference. Then, a 1*100 pixel map is drawn. The numerical range of the horizontal axis of the pixel map gradually increases from 3.5V to 4.2V and is used to represent the charging voltage. The pixel color is filled according to the capacity difference sequence corresponding to the charging voltage sequence. The rule for filling the pixel color is the correspondence between the actual capacity value and the black to white gradient. A 1*100 voltage-capacity difference corresponding map is generated, the voltage-capacity difference is stacked up and down, and a 100*100 capacity difference pixel map is obtained.

Further: The ResNet50 network includes an input layer, a 7x7 convolutional feature extraction layer, a 3x3 maximum pooling layer, a first residual module, a second residual module, a third residual module, a fourth residual module, an Agent Attention module, an Average pool layer, a 1000-d fully connected layer, a softmax layer and an output layer, which are connected in sequence; the abstract features of the image are gradually extracted, and classification prediction is performed through the softmax layer, and the health status assessment result of the lithium-ion battery is output through the output layer.

The beneficial effects of the present invention are as follows: the rate and properties of the internal chemical reactions of lithium-ion batteries are reflected by the rate of change of charge, which is crucial to understanding the working state of lithium-ion batteries. By tracking the actual capacity and the rate of change of charging voltage, more comprehensive information on the health status and performance of the battery can be obtained; at the same time, by comparing the actual capacity difference between the first cycle and the current cycle, important clues to the historical performance and state of the lithium-ion battery are provided, which is helpful to judge the degree of battery aging, the speed of performance degradation and possible failure modes; the present invention comprehensively utilizes information such as the rate of change of charge and the capacity difference to improve the model performance and generalization ability, and more accurately evaluate the health status and performance of lithium-ion batteries.

BRIEF DESCRIPTION OF THE DRAWINGS

Fig. 1 is a flow chart of the present invention;

FIG2 is a schematic diagram of a voltage capacity pixel diagram, a charge change rate pixel diagram, a capacity difference pixel diagram and a characteristic diagram at different cycles in the present invention;

Figure 3 is a block diagram of the ResNet50 network.

DETAILED DESCRIPTION

The present invention is described in detail below in conjunction with the accompanying drawings. The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be interpreted as limiting the present invention. The directional terms such as left, middle, right, top, and bottom in the embodiments of the present invention are only relative concepts or are based on the normal use state of the product, and should not be considered as restrictive.

A method for evaluating the health status of a lithium-ion battery based on graph convolution, as shown in FIG1 , comprises the following steps:

Step 1: Training a lithium-ion battery health status assessment model, including the following steps;

Step 1.1: Perform constant current and constant voltage cycle charging on several lithium-ion batteries, and obtain battery data including cycle sequence, charging voltage sequence, actual capacity sequence, and health status;

Step 1.2: As shown in FIG2, a voltage capacity pixel map, a charge change rate pixel map, and a capacity difference pixel map are prepared according to the battery data. The voltage capacity pixel map is used to record the corresponding relationship between the charging voltage and the actual capacity in each cycle. The charge change rate pixel map is used to record the change rate of the actual capacity relative to the charging voltage in each cycle. The capacity difference pixel map is used to record the capacity change relative to the initial actual capacity of the battery in each cycle.

Step 1.3: Stack the corresponding voltage capacity pixel map, charge change rate pixel map, and capacity difference pixel map to form a three-channel feature map, thereby obtaining a training set consisting of a feature map and a health status;

Step 1.4: Use the training set to train the ResNet50 network and obtain the trained lithium-ion battery health status assessment model;

Step 2: Obtain a characteristic graph to be tested according to the power battery data to be tested;

Step 3: Send the feature map to be tested into the trained lithium-ion battery health status assessment model;

Step 4: Output the health status assessment result of the lithium-ion battery.

in,

Step 1.1 is as follows:

From the same batch of lithium-ion batteries, 7 batteries (model Prospower ICR18650P) with a nominal capacity of 2 Ah and a voltage of 3.6 V were selected, and the same process steps were repeatedly charged and discharged in a constant temperature box at 25°C to obtain the data of battery cross-current constant voltage charging; after data screening and processing, about 7 groups of data were obtained, each group containing 1000 cycles; there were about 100 constant current constant voltage charging data in each cycle; the lithium-ion battery was repeatedly charged and discharged in a constant temperature box at 25°C, and the processing process was as follows: first, it was charged with a constant current (CC) of 1C until the voltage reached 4.2V; then, it was switched to constant voltage (CV) mode for charging until the current dropped to 0.02C; then, the charged lithium-ion battery was left to stand for 30 minutes; then, the lithium-ion battery was discharged at a discharge rate of 3C to a cut-off voltage of 2.5V; finally, the lithium-ion battery was left to stand for 60 minutes;

In order to achieve direct data connection between multiple cycles, the data needs to be smoothed; the specific processing method is as follows: first, calculate the number of data points in each window by dividing the total number of data rows by 100 and determining the number of remaining data points. Then, in each iteration, traverse 100 windows and select the corresponding data subset according to the window size; if there are remaining data points, add an additional data point in each window to ensure that all data points are used; finally, save the processing results in a new data feature file; after smoothing, the data set becomes relatively uniform, and each cycle contains 100 data. Finally, battery data including lithium-ion battery data such as cycle sequence, charging voltage sequence, actual capacity sequence, and health status is obtained;

The present invention defines the state of health SOH as the ratio of the actual capacity of the lithium-ion battery to the initial value, and the formula is:

in Indicates the current maximum charge capacity of the battery, and Indicates the initial capacity.

In step 1.2, the steps to make the voltage-capacity pixel map are:

The cycle sequence, charging voltage sequence and actual capacity sequence in the data features are selected, and the capacity sequence of 1000 cycles is normalized as a whole. The purpose is to eliminate the dimensional differences between different cycles and retain the relative size relationship of the data. The specific formula is:

in, is the normalized data, is the original data, is the minimum value in this set of data. is the maximum value in this set of data;

First, a color range is set, with black pixels representing the maximum value of the actual capacity and white pixels representing the minimum value of the actual capacity. Then, a 1*100 pixel map is drawn, with the value range of the horizontal axis of the pixel map gradually increasing from 3.5V to 4.2V and used to represent the charging voltage. The pixel colors are filled according to the actual capacity sequence corresponding to the charging voltage sequence. The rule for filling the pixel colors is the corresponding relationship between the actual capacity value and the black to white gradient. Finally, a 1*100 voltage-capacity correspondence map is generated. In order to allow the CNN model to learn to represent and extract features, the voltage-capacity correspondence maps are stacked up and down to obtain a 100*100 voltage-capacity pixel map.

The steps of making the pixel graph of the rate of change of electric capacity are as follows: interpolating the original data points by the linear interpolation method, using a function to generate equally spaced voltage values of the same length as the interpolated voltage sequence, and calculating the corresponding actual capacity value by the interpolation function;

Linear interpolation is mainly used to estimate the value of unknown points between given data points. Specifically, it is assumed that the functional relationship between two known data points is linear, and then the value at any position between the two points is estimated by the following formula:

(3)

in, and is the voltage value of the known data point, and is the corresponding actual capacity value, is the voltage to be estimated, is the estimated actual capacity;

Use the power change rate formula to obtain the corresponding value in the sequence, the power change rate for:

(4)

in, Indicates the number of cycles, Described as the slope of the charging voltage as the actual capacity changes, is the change in actual battery capacity, is the change in battery charging voltage, The actual capacity of the battery at the current moment, Indicates the actual capacity of the battery at the previous moment. The current battery charging voltage, The battery charging voltage at the previous moment.

The overall normalization method of the cycle charge change rate sequence is to use black pixels to represent the maximum value of the charge change rate, and white pixels to represent the minimum value of the charge change rate; then, a 1*100 pixel map is drawn, the numerical range of the horizontal axis of the pixel map gradually increases from 3.5V to 4.2V and is used to represent the charging voltage, and the pixel color is filled according to the charge change rate corresponding to the charging voltage sequence. The rule for filling the pixel color is the corresponding relationship between the actual capacity value and the black to white gradient, and a 1*100 voltage charge change rate corresponding map is generated. The voltage charge change rate corresponding map is stacked up and down to obtain a 100*100 charge change rate pixel map;

The steps for making the capacity difference pixel map are as follows: first, build a voltage value list with a step size of 0.007V, divide 3.5V to 4.2V into 100 intervals, and use one-dimensional linear interpolation to interpolate the actual capacity and estimate the actual capacity corresponding to the voltage; record the actual capacity difference between the first cycle and the current cycle as , the specific calculation formula is as follows:

in, The difference between the actual capacity of the first cycle and the actual capacity of the nth cycle, Indicates the actual capacity of the lithium battery in the first cycle. Indicates the actual capacity of the lithium battery at the nth cycle;

The calculated actual capacity difference is filled into the capacity difference sequence of the corresponding cycle. The capacity difference sequence of 1000 cycles is normalized as a whole, a color range is set, black pixels are used to represent the maximum value of the capacity difference, and white pixels are used to represent the minimum value of the capacity difference. Then, a 1*100 pixel map is drawn. The value range of the horizontal axis of the pixel map gradually increases from 3.5V to 4.2V and is used to represent the charging voltage. The pixel color is filled according to the capacity difference sequence corresponding to the charging voltage sequence. The rule for filling the pixel color is the corresponding relationship between the actual capacity value and the black to white gradient. A 1*100 voltage-capacity difference corresponding map is generated, and the voltage-capacity difference is stacked up and down to obtain a 100*100 capacity difference pixel map;

By observing the feature map, it can be seen that with the increase in the number of cycles, the difference in charging capacity (actual capacity) between 3.5V and 4.2V gradually narrows; since the charging current remains constant and the charging time is proportional to the charging capacity, the time required for the battery voltage to rise from 3.5V to 4.2V gradually decreases; it can be observed that the color of the pixel representing the actual capacity in the voltage-capacity pixel map gradually becomes brighter, and the pixel area representing the high actual capacity area in the image gradually shrinks; in the charge change rate pixel map, the dark area gradually becomes lighter and gradually gathers; in the capacity difference pixel map, the low voltage part has a small change, and the light-colored part of the high voltage gradually becomes darker; the obtained voltage-capacity pixel map, charge change rate pixel map, and capacity difference pixel map are stacked together to form a three-channel feature map, and the actual capacity in each charging cycle is taken to represent the health state SoH, and this is used as the label variable, and finally a data set of 7,000 images is obtained.

Step 1.3 is as follows:

The obtained dataset is divided into training set, test set and validation set in a ratio of 7:2:1, and the training set is sent to the ResNet50 network with the Agent Attention module added for training; In order to achieve the best performance of the ResNet50 network model, it is necessary to upsample the feature image and convert it to the most suitable size, that is, 224*224*3. We chose bilinear interpolation as the upsampling method because it can keep the image smooth while increasing the image size, and the calculation amount is relatively small; Bilinear interpolation is a method of estimating the value of a new pixel based on the weighted average of the four closest pixels in the image; Imagine that we want to interpolate at position (x, y):

First, find the four nearest pixels around the position (x, y) (usually top left, top right, bottom left, and bottom right);

Then, according to the distance of the target position relative to these four pixels, its weight in the horizontal and vertical directions is calculated;

Finally, the intensity of these four pixels is weighted averaged according to their distance to obtain the intensity value of the new pixel; assuming the target position is (x, y), its weight in the horizontal direction is , the weight in the vertical direction is ;

Assume that the intensities of the upper left, upper right, lower left, and lower right pixels are , , , , then the intensity of the new pixel can be expressed by the following formula:

in , , and are the coordinates of the upper left pixel;

In addition, as shown in FIG3 , the ResNet50 network includes an input layer, a 7x7 convolution feature extraction layer, a 3x3 maximum pooling layer, a first residual module, a second residual module, a third residual module, a fourth residual module, an AgentAttention module, an Average pool layer, a 1000-d fully connected layer, a softmax layer, and an output layer connected in sequence; the abstract features of the image are gradually extracted, and the classification prediction is performed through the softmax layer, and the health status assessment result of the lithium-ion battery is output through the output layer; the specific process is as follows:

(1) The obtained 224*224*3 feature map is sent to the "input layer", which is the Input in Figure 3. The input layer is the starting point of the neural network, responsible for receiving the original data and passing it to the next layer for processing, providing input for subsequent convolution, pooling and other operations. The input layer does not process the data, so the output result is the same as the input data;

(2) The output data in the input module is sent to the "7x7 convolution feature extraction layer". This layer uses a 7x7 convolution kernel to perceive image information in a larger range. The 64 filters can extract rich features. The step size of 2 reduces the size of the feature map, which helps to reduce the computational burden while retaining important information. The output result is 112*112*64;

(3) The output of the "7x7 convolutional feature extraction layer" is sent to the "3x3 maximum pooling layer". This layer uses a 3x3 maximum pooling operation with a step size of 2. It retains the main features of the image while reducing the size of the feature map, thereby reducing the computational complexity. Using a pooling operation with a step size of 2 can further reduce the size of the feature map and help prevent overfitting. The output result is 56*56*64;

(4) The output of the "3x3 maximum pooling layer" is sent to the "first residual module", which is composed of a series of convolution operations, using convolution kernels of 1x1, 3x3, and 1x1 sizes, and 64 and 256 filters. The 1x1 convolution is used to reduce the dimension and increase nonlinearity, the 3x3 convolution is used to capture local features and texture information, and then a 1x1 convolution is used to restore the dimension of the feature. By stacking multiple such convolution modules, the network can learn richer and more complex feature expressions, thereby improving the classification accuracy and generalization ability of the model. The output result is 56*56*256;

(5) The output of the "first residual module" is sent to the "second residual module". This module uses 1x1 and 3x3 convolution operations, as well as 128 and 512 filters. This module helps the network learn richer and more diverse image features, improves the network's image representation ability and classification performance. The role of the convolution kernel here is still 1x1 convolution for dimensionality reduction and increasing nonlinearity, and 3x3 convolution is used to capture local features and texture information. The output result is 28*28*512;

(6) The output of the "second residual module" is sent to the "third residual module". This module uses 1x1 and 3x3 convolution kernels and 256 and 1024 filters. It aims to gradually extract more abstract and complex features in the image through multi-level convolution operations. These abstract features have higher semantic information, which helps the network better distinguish between images of different categories and improve the generalization ability and robustness of the model. The output of this layer is 14*14*1024;

(7) The output of the third residual module is fed into the fourth residual module, which uses 1x1 and 3x3 convolution kernels and 512 and 2048 filters to further deepen the network and increase its nonlinearity to extract more advanced and abstract image features. Through the combination of 1x1 convolution and 3x3 convolution, the network can better capture image features of different scales and levels. This layer enables the network to better understand the input image and make more accurate classification decisions. The output of this layer is 7*7*2048;

(8) The output of the fourth residual module is fed into the Agent Attention module. The main function of the attention module is to significantly reduce the computational complexity while maintaining the global context modeling capability, so that the model can process large-scale data more effectively. The agent feature vector is generated through a simple pooling strategy. The Agent Attention module can maintain high efficiency when processing high-resolution images and long sequence data, and can better capture global information. Since the Agent Attention module keeps the input and output dimensions unchanged, the output is still 7*7*2048;

(9) The output of the "Agent Attention module" is sent to the "Average pool layer". Average pooling achieves feature downsampling by taking the average of each area of each feature map, while retaining the main feature information, further extracting the main features in the image and reducing the computational burden. Such operations can improve the generalization ability of the model and reduce the risk of overfitting. The output of this layer is 1*1*2048.

(10) The output of the "Average pool layer" is sent to the "1000-d fully connected layer". The main function of this layer is to flatten the features extracted by the previous convolution and pooling layers and connect them to the last layer of the neural network. The fully connected layer connects all neurons in the previous layer with all neurons in the current layer through the weight matrix to achieve feature combination. This design enables the network to learn the complex relationship between input features and output labels, thereby accurately evaluating the health status of the battery. The output of this layer is 1*1*1000;

(11) The output of the "1000-d fully connected layer" is sent to the "softmax layer". The softmax function converts the original scores of the network output into probability values, so that the predicted probability of each category is between 0 and 1, and the sum of the probabilities of all categories is 1. This design allows the model to make classification decisions by comparing the probabilities of different categories, and can quantify the model's confidence in different categories. The output of this layer is a probability vector, each element of which represents the predicted probability of the lithium-ion battery SOH;

(12) The output of the "softmax layer" is sent to the "output layer". The purpose of this layer is to receive the category probability distribution after processing by the softmax layer and output the neural network's inference results on the input data in an understandable form. The output result is a category label, indicating which category the network believes the input data belongs to.

Compared with existing technologies:

In a power battery pack data storage and health assessment method based on visual technology, the publication number is CN116756351A. Since the model cannot capture the relationship between the capacities of multiple cycles and the information provided by the features is limited and single, the model may tend to learn simple rules, resulting in underfitting, which limits the generalization ability of the model and makes it impossible to accurately predict or classify data. Therefore, the present invention comprehensively utilizes information such as capacity difference and the rate of change of actual capacity relative to charging voltage to improve model performance and generalization ability, and more accurately evaluate the health status and performance of lithium-ion batteries. Since the present invention adopts a three-channel feature map formed by a combination of three features, richer features can be obtained during model training, and the complex patterns and relationships in the data can be better captured, thereby improving the characterization ability of the model.

The present invention adopts image features for performance estimation instead of directly using the provided raw data. Image processing technology provides automatic feature extraction, data augmentation and adaptability, thereby effectively completing the battery health assessment task. First, image processing technology can help to automatically extract features related to the battery health status without the need for manual complex feature engineering. Second, image processing technology can be used for data augmentation. By performing transformations such as rotation, scaling, and translation on the image, more diverse training data can be generated, which helps to improve the generalization ability of the model. Finally, the image features of the battery can better complicate the relationship between features, which makes the model more widely applicable.

In addition, the present invention improves the resnet50 network, and the method is suitable for estimating the battery life of electric vehicles in a collaborative vehicle infrastructure system; although in actual applications, the operating conditions may be different from the conditions of the laboratory test platform, this method can be extended to other fields; the proposed method will further enhance the application in fields such as large-scale power battery systems and energy storage systems; by adopting the present invention, the estimation of the battery health state can be achieved. As a new lithium-ion battery health state estimation method, the present invention can be widely used in the field of battery state estimation.

The basic principles, main features and advantages of the present invention are shown and described above. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims (2)

1. A lithium ion battery health state evaluation method based on graph convolution is characterized in that: the method comprises the following steps:

Step 1: training a lithium ion battery health state evaluation model, comprising the following steps of;

step 1.1: constant-current constant-voltage cyclic charging is carried out on a plurality of lithium ion batteries, and battery data comprising a cyclic sequence, a charging voltage sequence, an actual capacity sequence and a health state are obtained;

Step 1.2: manufacturing a voltage capacity pixel map, an electric quantity change rate pixel map and a capacity difference pixel map according to battery data, wherein the voltage capacity pixel map is used for recording the corresponding relation between charging voltage and actual capacity in each cycle, and the electric quantity change rate pixel map is used for recording the change rate of the actual capacity relative to the charging voltage in each cycle; the capacity difference pixel map is used for recording capacity variation relative to the initial actual capacity of the battery in each cycle;

The method for manufacturing the voltage capacity pixel map comprises the following steps of:

Carrying out integral normalization on the circulated actual capacity sequence, setting a color range, using black pixels to represent the maximum value of the actual capacity, and using white pixels to represent the minimum value of the actual capacity; then, drawing a1 x 100 pixel map, gradually increasing the numerical range of the abscissa of the pixel map from 3.5V to 4.2V and representing charging voltage, filling pixel colors according to an actual capacity sequence corresponding to the charging voltage sequence, wherein the rule of filling the pixel colors is a corresponding relation between actual capacity value and black-to-white gradual change, finally generating a1 x 100 voltage capacity map, stacking the voltage capacity map up and down, and obtaining a 100 x 100 voltage capacity pixel map;

the method for manufacturing the electric quantity change rate pixel map comprises the following steps of:

The method comprises the steps of integrally normalizing a cyclic electric quantity change rate sequence, wherein black pixels are used for representing the maximum value of the electric quantity change rate, and white pixels are used for representing the minimum value of the electric quantity change rate; then, a 1x 100 pixel diagram is drawn, the numerical range of the abscissa of the pixel diagram gradually increases from 3.5V to 4.2V and is used for representing charging voltage, pixel colors are filled according to the electric quantity change rate corresponding to a charging voltage sequence, the rule of the filled pixel colors is the corresponding relation between the actual capacity value and black-to-white gradual change, a 1x 100 voltage electric quantity change rate corresponding diagram is generated, the voltage electric quantity change rate corresponding diagram is stacked up and down, and the electric quantity change rate pixel diagram of 100x 100 is obtained; the electric quantity change rate F2 is:

wherein n represents the number of cycles, Described as the slope of the change in the charging voltage with the actual capacity, Δq is the amount of change in the actual capacity of the battery, Δv is the amount of change in the charging voltage of the battery,The actual capacity of the battery at the present moment,Indicating the actual capacity of the battery at the previous moment,The charge voltage of the battery at the present moment,The charging voltage of the battery at the previous moment;

The step of making the capacity difference pixel map is as follows:

Firstly, a charging voltage value list is constructed, the step size is 0.007V, 3.5V to 4.2V are divided into 100 intervals, the actual capacity difference between the first cycle and the current cycle is recorded as F3, and a specific calculation formula is as follows:

F3＝Q^1,n(V)＝Q¹(V)-Qⁿ(V) (5)

Wherein, the difference between the actual capacity of the first cycle of Q ^1,n (V) and the actual capacity of the nth cycle, Q ¹ (V) represents the actual capacity of the lithium battery of the 1 st cycle, and Q ⁿ (V) represents the actual capacity of the lithium battery of the nth cycle;

Filling the calculated actual capacity difference value into a capacity difference sequence of a corresponding cycle, adopting integral normalization of the capacity difference sequence of the cycle, setting a color range, using black pixels to represent the maximum value of the capacity difference, using white pixels to represent the minimum value of the capacity difference, then drawing a 1 x 100 pixel diagram, gradually increasing the numerical range of the abscissa of the pixel diagram from 3.5V to 4.2V and using the pixel diagram to represent charging voltage, filling pixel colors according to the capacity difference sequence corresponding to the charging voltage sequence, generating a 1 x 100 voltage capacity difference corresponding diagram according to the rule that the actual capacity value is in a corresponding relation with black-to-white gradual change, stacking the voltage capacity differences up and down, and obtaining a 100 x 100 capacity difference pixel diagram;

step 1.3: stacking the corresponding voltage capacity pixel diagram, the corresponding electric quantity change rate pixel diagram and the corresponding capacity difference pixel diagram to form a three-channel characteristic diagram, thereby obtaining a training set consisting of the characteristic diagram and the health state;

Step 1.4: training ResNet a network by using a training set, and obtaining a trained lithium ion battery health state evaluation model;

step 2: obtaining a feature diagram to be measured according to the data of the power battery to be measured;

step 3: sending the feature diagram to be tested into a trained lithium ion battery health state evaluation model;

step 4: and outputting the health state evaluation result of the lithium ion battery.

2. The method for evaluating the health state of a lithium ion battery based on graph convolution according to claim 1, wherein the method comprises the following steps: the ResNet network comprises an input layer, a 7x7 convolution feature extraction layer, a 3x3 maximum pooling layer, a first residual error module, a second residual error module, a third residual error module, a fourth residual error module, an Agent Attention module, an Average pool layer, a 1000-d full connection layer, a softmax layer and an output layer which are sequentially connected; the abstract features of the images are gradually extracted, classification prediction is carried out through a softmax layer, and the health state evaluation result of the lithium ion battery is output through an output layer.