A method and device for predicting the health status of electric vehicle batteries

Abstract

The present application provides a method and device for predicting the health status of an electric vehicle battery, the method comprising: constructing an original model based on a multi-scale convolutional neural network; wherein the original model at least comprises a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer; training the original model through training data and model training configuration data to obtain a battery capacity prediction model; inputting the acquired sensor sampling data into the battery capacity prediction model to evaluate the battery health status and obtain corresponding results. The method and device can accurately predict the battery health status without requiring complete charge and discharge cycle data, avoiding the training and prediction difficulties caused by the lack of complete charge and discharge data, improving the applicability and accuracy, and thus better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles.

Description

Technical Field

[0001] The present application relates to the field of battery safety technology, and more specifically, to a method and device for predicting the health status of an electric vehicle battery. Background Art

[0002] In the field of new energy vehicles, accurately assessing the health status of the battery is the key to ensuring vehicle performance, extending battery life, and improving user experience. Existing technologies usually require the use of data from a complete charge and discharge cycle to train and predict the battery health status of the battery. However, in practice, it is found that users often choose different charging modes and charging times according to their needs and the conditions of the charging facilities, and rarely fully charge or completely discharge the battery. Such a usage pattern will cause the collected data to lack a complete charge and discharge cycle, thus affecting the prediction effect. It can be seen that the existing technology relies on complete charge and discharge cycle data to predict the battery health status, which has low accuracy and low practicality. Summary of the invention

[0003] The purpose of the embodiments of the present application is to provide a method and device for predicting the health status of an electric vehicle battery, which can accurately predict the battery health status without the need for complete charge and discharge cycle data, thereby avoiding the training and prediction difficulties caused by the lack of complete charge and discharge data, and improving the applicability and accuracy, thereby better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles.

[0004] The first aspect of the present application provides a method for predicting the health status of an electric vehicle battery, comprising:

[0005] Obtain training data and model training configuration data;

[0006] Constructing an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer;

[0007] The original model is trained by using the training data and the model training configuration data to obtain a battery capacity prediction model;

[0008] Deploying the battery capacity prediction model onto an onboard hardware platform of a target vehicle;

[0009] During the charging process of the target vehicle, sensor sampling data is acquired according to a preset collection time period and a preset number of sampling points;

[0010] Calculating a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data;

[0011] Calculating the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value;

[0012] The battery health status assessment result including the battery SOH value is output.

[0013] In the above implementation process, the method can accurately predict the battery health status without the need for complete charge and discharge cycle data, avoiding the training and prediction difficulties caused by the lack of complete charge and discharge data, improving the applicability and accuracy, and thus better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles.

[0014] Furthermore, the obtaining of training data and model training configuration data includes:

[0015] During the charging process of the electric vehicle, a battery data set is obtained through a sensor according to a preset collection time period and a preset number of sampling points;

[0016] Performing labeling processing on the battery data group to obtain labeled data;

[0017] Preprocessing the labeled data to obtain training data;

[0018] Obtain model training configuration data; wherein the model training configuration data at least includes a loss function, a learning rate optimization algorithm, and a number of iterations.

[0019] Further, the sensor includes at least a current sensor, a voltage sensor and a temperature sensor;

[0020] The battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data;

[0021] The loss function is a mean square error loss function;

[0022] The learning rate optimization algorithm is the Adam adaptive learning rate optimization algorithm.

[0023] Further, the labeling process is performed on the battery data group to obtain the labeling data, including:

[0024] Calculate the current actual battery capacity corresponding to each group of sensor collected data according to the battery data group;

[0025] Capacity annotation processing is performed on each group of sensor collected data according to the current actual battery capacity to obtain annotated data.

[0026] Further, the calculation of the current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data includes:

[0027] Performing maximum value and average value pooling processing on the sensor sampling data simultaneously through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data;

[0028] Concatenate the average value pooling output data and the maximum value pooling output data to obtain pooling layer output data;

[0029] Performing a self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain a weight score corresponding to the output data of the pooling layer;

[0030] Merging the pooling layer output data and the weight score through the channel merging layer of the battery capacity prediction model to obtain channel merging layer output data;

[0031] Perform convolution calculation on the output data of the channel merging layer according to a preset convolution kernel size, a preset activation function, and a multi-scale convolution layer of the battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size;

[0032] Merging the convolution data to obtain convolution layer output data;

[0033] The actual capacity value is predicted based on the output layer of the battery capacity prediction model and the output data of the convolution layer to obtain the current actual capacity prediction value.

[0034] A second aspect of the present application provides a device for predicting the health status of a battery in an electric vehicle, the device comprising:

[0035] A first acquisition unit, used to acquire training data and model training configuration data;

[0036] A construction unit, used to construct an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer;

[0037] A training unit, used to train the original model using the training data and the model training configuration data to obtain a battery capacity prediction model;

[0038] A model deployment unit, used to deploy the battery capacity prediction model onto an on-board hardware platform of a target vehicle;

[0039] A second acquisition unit is used to acquire sensor sampling data according to a preset collection time period and a preset number of sampling points during the charging process of the target vehicle;

[0040] A first calculation unit, configured to calculate a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data;

[0041] A second calculation unit, used to calculate the battery SOH value according to a preset battery rated capacity and the current actual capacity prediction value;

[0042] The output unit is used to output the battery health status assessment result including the battery SOH value.

[0043] Furthermore, the first acquiring unit includes:

[0044] A first acquisition subunit is used to acquire a battery data group through a sensor according to a preset collection time period and a preset number of sampling points during the charging process of the electric vehicle;

[0045] A labeling subunit, used for labeling the battery data group to obtain labeling data;

[0046] A labeling subunit, used for preprocessing the labeling data to obtain training data;

[0047] The second acquisition subunit is used to acquire model training configuration data; wherein the model training configuration data at least includes a loss function, a learning rate optimization algorithm and a number of iterations.

[0048] Further, the sensor includes at least a current sensor, a voltage sensor and a temperature sensor;

[0049] The battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data;

[0050] The loss function is a mean square error loss function;

[0051] The learning rate optimization algorithm is the Adam adaptive learning rate optimization algorithm.

[0052] Furthermore, the marking subunit includes:

[0053] A calculation module, used for calculating the current actual battery capacity corresponding to each group of sensor collected data according to the battery data group;

[0054] The labeling module is used to perform capacity labeling processing on each group of sensor collected data according to the current actual battery capacity to obtain labeling data.

[0055] Furthermore, the first computing unit includes:

[0056] A processing subunit, configured to simultaneously perform maximum value and average value pooling processing on the sensor sampling data through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data;

[0057] A splicing subunit, used for splicing the average pooling output data and the maximum pooling output data to obtain pooling layer output data;

[0058] A calculation subunit, configured to perform a self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain a weight score corresponding to the output data of the pooling layer;

[0059] A merging subunit, used for merging the pooling layer output data and the weight score through the channel merging layer of the battery capacity prediction model to obtain channel merging layer output data;

[0060] The computing subunit is further configured to perform convolution calculation on the output data of the channel merging layer according to a preset convolution kernel size, a preset activation function, and a multi-scale convolution layer of the battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size;

[0061] The merging subunit is further used to merge the convolution data to obtain convolution layer output data;

[0062] The prediction subunit is used to predict the actual capacity value based on the output layer of the battery capacity prediction model and the output data of the convolution layer to obtain the current actual capacity prediction value.

[0063] The third aspect of the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the electric vehicle battery health status prediction method described in any one of the first aspect of the present application.

[0064] A fourth aspect of the present application provides a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, executes the electric vehicle battery health status prediction method described in any one of the first aspect of the present application.

[0065] The beneficial effects of the present application are: using pooling kernels of different scales to process input data can not only retain important information of the original data, but also effectively remove noise and irrelevant details, and extract more comprehensive features while reducing the data dimension;

[0066] The channel attention mechanism can adaptively assign weights to each channel and learn the importance of different channel features. By capturing the correlation and importance between channels, the network can dynamically adjust the contribution of each channel, better utilize multi-channel feature information, and improve prediction accuracy.

[0067] The multi-scale convolution layer uses convolution kernels of different sizes and combines them with two-dimensional convolution to simultaneously extract complex nonlinear features of different granularities, further enhancing the feature extraction capability of the model.

[0068] Compared with the traditional SOH estimation method based on deep learning, the network structure of the present invention is relatively simple, with fewer parameters, which is more conducive to the deployment and real-time reasoning of the model on vehicle-mounted embedded devices, meeting the needs of practical applications;

[0069] During the charging process of new energy vehicles, the SOH value of the lithium battery can be estimated by using current, voltage and temperature sensors to collect data from N sampling points within L time, without the need for complete charge and discharge cycle data, making data collection simpler and more flexible, and can be easily applied to different charging modes and closer to actual application scenarios; at the same time, it can also avoid the training and prediction difficulties caused by the lack of complete charge and discharge data, thereby better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 A schematic diagram of a flow chart of a method for predicting the health status of an electric vehicle battery provided in an embodiment of the present application;

[0072] Figure 2 An overall block diagram of a method for predicting the health status of an electric vehicle battery provided in an embodiment of the present application;

[0073] Figure 3 A schematic diagram of a flow chart of another electric vehicle battery health status prediction method provided in an embodiment of the present application;

[0074] Figure 4 A schematic diagram of an example flow chart of a method for predicting the health status of an electric vehicle battery provided in an embodiment of the present application;

[0075] Figure 5A schematic diagram of the structure of a device for predicting the health status of an electric vehicle battery provided in an embodiment of the present application;

[0076] Figure 6 A schematic diagram of the structure of another electric vehicle battery health status prediction device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0077] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0078] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0079] Example 1

[0080] Please see Figure 1 , Figure 1 A flowchart of a method for predicting the health status of an electric vehicle battery provided in this embodiment. The method for predicting the health status of an electric vehicle battery includes:

[0081] S101. Obtain training data and model training configuration data.

[0082] In this embodiment, the method can be applied to the health status prediction scenario of lithium batteries.

[0083] S102. Construct an original model based on a multi-scale convolutional neural network.

[0084] In this embodiment, the original model includes at least a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer.

[0085] In this embodiment, the convolutional neural network has the characteristics of local perception through the convolution operation, and can effectively capture local patterns and features in the input data. In addition, the convolutional neural network is usually composed of multiple convolutional layers and pooling layers stacked alternately. The stacking of these hierarchical structures enables the network to extract and abstract features in the data layer by layer, and gradually establish a hierarchical representation. Such a hierarchical representation can better understand the complex structure and hierarchical information of the data.

[0086] Traditional convolution usually uses a fixed-size convolution kernel to extract the features of the input data, which cannot fully capture the complex features of the health status of lithium batteries, and there is a complex nonlinear relationship between the health status of lithium batteries and voltage, current and temperature. Traditional convolution is difficult to effectively extract these complex feature information. This application proposes a method of using multiple convolution kernels of different sizes to process the input in the same level of network to capture feature information of different scales. In the forward propagation process, the input data is convolved with each convolution kernel of different sizes, and the outputs of convolution kernels of all scales are spliced ​​together as the final output, thereby solving the above problems.

[0087] S103: Train the original model using the training data and the model training configuration data to obtain a battery capacity prediction model.

[0088] S104: deploying the battery capacity prediction model to the on-board hardware platform of the target vehicle.

[0089] S105. During the charging process of the target vehicle, sensor sampling data is acquired according to a preset collection time period and a preset number of sampling points.

[0090] S106: Calculate a current actual capacity prediction value of the target vehicle's electric battery according to the battery capacity prediction model and the sensor sampling data.

[0091] S107: Calculate the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value.

[0092] S108. Output the battery health status assessment result including the battery SOH value.

[0093] Please see Figure 2 , Figure 2 The overall block diagram of a method for predicting the health status of an electric vehicle battery is shown. The method is a method for estimating the health status of a new energy vehicle lithium battery based on a multi-scale convolutional neural network, and specifically achieves corresponding functions by training a new energy vehicle lithium battery health status estimation model using a specific model training method.

[0094] In this embodiment, the execution subject of the method may be a computing device such as a computer or a server, and no limitation is made in this embodiment.

[0095] In this embodiment, the execution subject of the method may also be a smart device such as a smart phone, a tablet computer, etc., which is not limited in this embodiment.

[0096] It can be seen that the electric vehicle battery health status prediction method described in this embodiment can accurately predict the battery health status without the need for complete charge and discharge cycle data, avoiding the training and prediction difficulties caused by the lack of complete charge and discharge data, and improving the applicability and accuracy, thereby better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles.

[0097] Example 2

[0098] Please see Figure 3 , Figure 3 A flowchart of a method for predicting the health status of an electric vehicle battery provided in this embodiment. The method for predicting the health status of an electric vehicle battery includes:

[0099] S201. During the charging process of the electric vehicle, a battery data group is acquired through a sensor according to a preset collection time period and a preset number of sampling points. The battery data group includes at least one group of sensor collection data.

[0100] In this embodiment, the sensor includes at least a current sensor, a voltage sensor and a temperature sensor.

[0101] In this embodiment, the battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data;

[0102] In this embodiment, the battery health state (SOH) is an important indicator of the battery system, which is defined as the ratio of the current capacity to the rated capacity, and is usually used to describe the health of the battery. Among them, SOH is usually expressed as a percentage or a value between 0 and 1, where 100% or 1 indicates that the battery health state is intact, while 0% and 0 indicate that the battery has failed.

[0103] In this embodiment, when charging a new energy vehicle, the method uses a current sensor, a voltage sensor, and a temperature sensor to randomly collect battery voltage, current, and temperature data at N sampling points within a time period of L. k,s ,i k,s ,t k,s}, represents the data of the sth collection process in the kth charging cycle. This collection method only collects data in a small period of time, which is easier to collect during the actual charging process of new energy vehicles.

[0104] S202: Calculate the current actual battery capacity corresponding to each group of sensor collected data according to the battery data group.

[0105] In this embodiment, in a complete charging process, the method can collect multiple sets of data, and the label corresponding to each set of data is the current actual capacity. The corresponding label calculates the current actual capacity by the ampere-hour integration method:

[0106] (1)

[0107] Wherein, k and t represent the number of cycles and charging time, respectively;

[0108] c k is the actual capacity of the kth cycle;

[0109] is the current of the kth cycle.

[0110] Finally, the method obtains the data set (input {u k,s ,i k,s ,t k,s}, output c k ).

[0111] S203 , performing capacity labeling processing on each group of sensor collected data according to the current actual battery capacity to obtain labeled data.

[0112] In this embodiment, the method prepares to train a novel SOH estimation model for new energy vehicle lithium batteries based on a multi-scale convolutional neural network.

[0113] In this embodiment, the method first labels the acquired sensor data so that the labeled data can be input into the model as a training set to train the model. The model training can adopt a supervised learning method.

[0114] S204: Preprocess the labeled data to obtain training data.

[0115] In this embodiment, the preprocessing may be to fill the missing values ​​and abnormal values ​​in the data by linear interpolation, and then set the batch size for each training.

[0116] S205: Obtain model training configuration data.

[0117] In this embodiment, the model training configuration data includes at least a loss function, a learning rate optimization algorithm, and the number of iterations.

[0118] In this embodiment, the loss function is a mean square error loss function, and the learning rate optimization algorithm is an Adam adaptive learning rate optimization algorithm.

[0119] In this embodiment, the mean square error loss function may be:

[0120] (2)

[0121] Where n represents the number of samples, is the true label, are the estimates of the model.

[0122] In this embodiment, Adam is used as a commonly used adaptive learning rate optimization algorithm. This method uses Adam to perform gradient update during the model training process.

[0123] In this embodiment, the method sets the number of iterations in the model training configuration data, and the number of iterations is used to stop training when the model reaches the set number of iterations.

[0124] S206. Construct an original model based on a multi-scale convolutional neural network.

[0125] In this embodiment, the original model includes at least a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer.

[0126] ① About the multi-scale pooling layer:

[0127] The pooling operation can compress the data by taking the maximum value, average value, etc., retaining the main feature information in the data and removing some unimportant details, so that the model pays more attention to the important information in the data and improves the generalization ability and stability of the model.

[0128] However, traditional pooling operations usually only set one pooling kernel to process, or only use the average value or the maximum value, which will not fully extract data features. To solve this problem, this method can achieve a perfect balance between reducing data noise and fully retaining important information.

[0129] The specific process of the multi-scale pooling layer is to use three sizes of pooling kernels to perform maximum and average pooling on the input at the same time. , expressed as:

[0130] (3)

[0131] (4)

[0132] in, and They represent the output after average pooling and maximum pooling when the pooling kernel is i, respectively. i is set to {3, 5, 7}. By using boundary padding, the output length is guaranteed to be consistent after being processed by different pooling kernels. The same operation is performed on current and temperature:

[0133] (5)

[0134] (6)

[0135] (7)

[0136] (8)

[0137] in, , They represent the average value pooling output and the maximum value pooling output when the current passes through the pooling kernel size i respectively;

[0138] , They respectively represent the average pooling output and the maximum pooling output when the temperature passes through the pooling kernel size i.

[0139] Concatenate the above outputs to get , when i={3,5,7}, Merge to get the final output of the multi-scale pooling layer .

[0140] ② About the channel attention layer

[0141] The output obtained by the above multi-scale pooling layer It is a three-channel matrix. In order to calculate the weight of each channel, this method constructs a channel attention layer to adaptively assign weights to each channel to ensure the integrity of the transmitted information.

[0142] First, this method The self-attention mechanism is used separately. The self-attention mechanism can learn the correlation and importance between the features in each channel. The features of different channels correspond to information of different scales. Therefore, the self-attention mechanism can better capture the important features within each channel. The calculation process is as follows:

[0143] (9)

[0144] (10)

[0145] in, , , is a learnable weight matrix, self-attention score Computes the weighted sum of a vector of values. For the dimension, is the output of self-attention. Perform the same operation to obtain Then Flatten them separately to get , that is, expand it to get a one-dimensional vector, Use the fully connected layer to get its corresponding weights:

[0146] (11)

[0147] (12)

[0148] (13)

[0149] in , , is the learnable weight matrix, , , is the bias vector. is the Sigmoid activation function, which is used to compress the weights to the range of (0, 1). , , is the obtained weight score.

[0150] ③ About channel merging layer

[0151] The function of this layer is to combine the weights calculated by the channel attention layer with the output of the multi-scale pooling layer. Multiply and combine, the calculation process is as follows:

[0152] (14)

[0153] in This is the output of the channel merging layer. The pooling layer not only removes interference information such as noise, but also retains important feature information to the maximum extent.

[0154] ④ About multi-scale convolutional layers

[0155] The traditional convolution layer only sets the convolution kernel of one size, which will lead to insufficient feature extraction. In order to solve this problem, a multi-scale convolution layer is introduced. Similar to the multi-scale pooling layer, this layer also sets three sizes of convolution kernels to fully capture feature information, and uses two-dimensional convolution globally to capture the correlation between features.

[0156] First of all, Use convolution kernels of size i={3, 5, 7} to calculate:

[0157] (15)

[0158] (16)

[0159] (17)

[0160] in , , They represent the output when the convolution kernel sizes are {3, 5, 7} respectively. , , They represent convolution kernels of sizes {3, 5, 7} respectively. represents the RELU activation function. Then , , Flatten it to get , , , and then merge them to get the final output of this layer:

[0161] (18)

[0162] Represents the final output of the multi-scale convolutional layer.

[0163] ⑤About the output layer

[0164] In order for the model to output the predicted current actual capacity value, the output layer uses a fully connected network to output the predicted actual capacity value, as shown below:

[0165] (19)

[0166] in, , Represents the weight matrix and bias vector, using Reduce the risk of overfitting of neural networks, Represents the final predicted capacity.

[0167] S207. Train the original model using the training data and the model training configuration data to obtain a battery capacity prediction model.

[0168] Please see Figure 4 , Figure 4 An example flow chart of a method for predicting the health status of an electric vehicle battery is shown.

[0169] S208: deploy the battery capacity prediction model on the on-board hardware platform of the target vehicle.

[0170] In this embodiment, the method saves the final model parameters at the end of model training as a deployment model for vehicle-mounted SOH estimation.

[0171] S209. During the charging process of the target vehicle, sensor sampling data is acquired according to a preset collection time period and a preset number of sampling points.

[0172] S210, performing maximum value and average value pooling processing on the sensor sampling data simultaneously through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data.

[0173] S211. Concatenate the average pooling output data and the maximum pooling output data to obtain pooling layer output data.

[0174] S212. Perform self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain the weight score corresponding to the output data of the pooling layer.

[0175] S213. The pooling layer output data and the weight score are merged through the channel merging layer of the battery capacity prediction model to obtain the channel merging layer output data.

[0176] S214. Perform convolution calculation on the output data of the channel merging layer according to the preset convolution kernel size, the preset activation function and the multi-scale convolution layer of the battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size.

[0177] S215: Merge the convolution data to obtain convolution layer output data.

[0178] S216: Predict the actual capacity value based on the output layer and convolution layer output data of the battery capacity prediction model to obtain the current actual capacity prediction value.

[0179] S217: Calculate the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value.

[0180] S218. Output the battery health status assessment result including the battery SOH value.

[0181] In this embodiment, after the method deploys the trained vehicle-mounted SOH estimation model to the vehicle-mounted hardware platform, when the model reads the battery voltage, current and temperature (u, i, t) at N sampling points within L time during the charging process, the model can directly calculate the current actual capacity value of the lithium battery, that is:

[0182] (20)

[0183] Where NN represents the trained network, Represents the predicted capacity value, and then calculates The ratio of the rated capacity c of the lithium battery can give the SOH value:

[0184] (twenty one)

[0185] In this embodiment, the execution subject of the method may be a computing device such as a computer or a server, and no limitation is made in this embodiment.

[0186] In this embodiment, the execution subject of the method may also be a smart device such as a smart phone, a tablet computer, etc., which is not limited in this embodiment.

[0187] It can be seen that the electric vehicle battery health status prediction method described in this embodiment can use three pooling kernels of different sizes to perform maximum pooling and average pooling operations on the input data, so as to better capture the complex features in the battery health status data, and can not only retain the important information of the original data, but also effectively remove noise and irrelevant details.

[0188] After extracting features in the multi-scale pooling layer, this method can also adaptively assign weights to each channel through the channel attention layer, thereby effectively learning the importance of different channel features. At the same time, through the self-attention mechanism, the network can capture the correlation and importance between the features of each channel, thereby dynamically adjusting the contribution of each channel, thereby ensuring that the network better utilizes multi-channel feature information and improving the accuracy of the final estimation result.

[0189] In addition to multi-scale pooling and channel attention mechanisms, this method also introduces a multi-scale convolution layer. This layer uses three different sizes of convolution kernels (3x3, 5x5, 7x7) combined with two-dimensional convolution to more comprehensively capture the complex nonlinear features in the battery health status data. Compared with the traditional single-scale convolution layer, this multi-scale convolution can simultaneously extract features of different granularities, further enhancing the model's feature extraction capabilities.

[0190] Finally, this method only needs to collect partial data during the charging process, and does not require complete charging and discharging history data, so this method can greatly reduce the time and cost of data collection. This method based on partial data training is more practical and efficient than the traditional method that requires complete data. Applying this method in the battery management system can obtain battery SOH information more quickly, providing a basis for the optimal management and use of the battery.

[0191] Example 3

[0192] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a device for predicting the health status of an electric vehicle battery provided in this embodiment. Figure 5 As shown, the electric vehicle battery health status prediction device includes:

[0193] A first acquisition unit 310, used to acquire training data and model training configuration data;

[0194] A construction unit 320 is used to construct an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer;

[0195] A training unit 330, configured to train the original model using training data and model training configuration data to obtain a battery capacity prediction model;

[0196] A model deployment unit 340, used to deploy the battery capacity prediction model to an on-board hardware platform of a target vehicle;

[0197] The second acquisition unit 350 is used to acquire sensor sampling data according to a preset collection time period and a preset number of sampling points during the charging process of the target vehicle;

[0198] A first calculation unit 360, for calculating a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data;

[0199] The second calculation unit 370 is used to calculate the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value;

[0200] The output unit 380 is used to output the battery health status assessment result including the battery SOH value.

[0201] In this embodiment, the explanation of the electric vehicle battery health status prediction device can refer to the description in Example 1 or Example 2, and will not be repeated in this embodiment.

[0202] It can be seen that the electric vehicle battery health status prediction device described in this embodiment can accurately predict the battery health status without the need for complete charge and discharge cycle data, avoiding the training and prediction difficulties caused by the lack of complete charge and discharge data, and improving the applicability and accuracy, thereby better adapting to various charging habits and environmental conditions in the actual use of new energy vehicles.

[0203] Example 4

[0204] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a device for predicting the health status of an electric vehicle battery provided in this embodiment. Figure 6 As shown, the electric vehicle battery health status prediction device includes:

[0205] A first acquisition unit 310, used to acquire training data and model training configuration data;

[0206] A construction unit 320 is used to construct an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer;

[0207] A training unit 330, configured to train the original model using training data and model training configuration data to obtain a battery capacity prediction model;

[0208] A model deployment unit 340, used to deploy the battery capacity prediction model to an on-board hardware platform of a target vehicle;

[0209] The second acquisition unit 350 is used to acquire sensor sampling data according to a preset collection time period and a preset number of sampling points during the charging process of the target vehicle;

[0210] A first calculation unit 360, for calculating a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data;

[0211] The second calculation unit 370 is used to calculate the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value;

[0212] The output unit 380 is used to output the battery health status assessment result including the battery SOH value.

[0213] As an optional implementation manner, the first acquiring unit 310 includes:

[0214] The first acquisition subunit 311 is used to acquire a battery data group through a sensor according to a preset collection time period and a preset number of sampling points during the charging process of the electric vehicle;

[0215] The labeling subunit 312 is used to label the battery data group to obtain labeling data;

[0216] The labeling subunit 313 is used to pre-process the labeling data to obtain training data;

[0217] The second acquisition subunit 314 is used to acquire model training configuration data; wherein the model training configuration data at least includes a loss function, a learning rate optimization algorithm, and a number of iterations.

[0218] In this embodiment, the sensor includes at least a current sensor, a voltage sensor and a temperature sensor;

[0219] The battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data;

[0220] The loss function is the mean square error loss function;

[0221] The learning rate optimization algorithm is the Adam adaptive learning rate optimization algorithm.

[0222] As an optional implementation, the marking subunit 312 includes:

[0223] A calculation module, used to calculate the current actual battery capacity corresponding to each group of sensor collection data according to the battery data group;

[0224] The labeling module is used to perform capacity labeling processing on each group of sensor collected data according to the current actual battery capacity to obtain labeled data.

[0225] As an optional implementation, the first calculation unit 360 includes:

[0226] The processing subunit 361 is used to simultaneously perform maximum value and average value pooling processing on the sensor sampling data through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data;

[0227] A concatenation subunit 362 is used to concatenate the average pooling output data and the maximum pooling output data to obtain pooling layer output data;

[0228] The calculation subunit 363 is used to perform a self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain a weight score corresponding to the output data of the pooling layer;

[0229] A merging subunit 364 is used to merge the pooling layer output data and the weight score through the channel merging layer of the battery capacity prediction model to obtain the channel merging layer output data;

[0230] The calculation subunit 363 is further used to perform convolution calculation on the output data of the channel merging layer according to a preset convolution kernel size, a preset activation function and a multi-scale convolution layer of a battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size;

[0231] The merging subunit 364 is further used to merge the convolution data to obtain convolution layer output data;

[0232] 365, used to predict the actual capacity value based on the output layer and convolution layer output data of the battery capacity prediction model to obtain the current actual capacity prediction value.

[0233] In this embodiment, the explanation of the electric vehicle battery health status prediction device can refer to the description in Example 1 or Example 2, and will not be repeated in this embodiment.

[0234] It can be seen that the electric vehicle battery health status prediction device described in this embodiment can use three pooling kernels of different sizes to perform maximum pooling and average pooling operations on the input data, so as to better capture the complex features in the battery health status data, retain the important information of the original data, and effectively remove noise and irrelevant details.

[0235] After extracting features in the multi-scale pooling layer, this method can also adaptively assign weights to each channel through the channel attention layer, thereby effectively learning the importance of different channel features. At the same time, through the self-attention mechanism, the network can capture the correlation and importance between the features of each channel, thereby dynamically adjusting the contribution of each channel, thereby ensuring that the network better utilizes multi-channel feature information and improving the accuracy of the final estimation result.

[0236] In addition to multi-scale pooling and channel attention mechanisms, this method also introduces a multi-scale convolution layer. This layer uses three different sizes of convolution kernels (3x3, 5x5, 7x7) combined with two-dimensional convolution to more comprehensively capture the complex nonlinear features in the battery health status data. Compared with the traditional single-scale convolution layer, this multi-scale convolution can simultaneously extract features of different granularities, further enhancing the model's feature extraction capabilities.

[0237] Finally, this method only needs to collect partial data during the charging process, and does not require complete charging and discharging history data, so this method can greatly reduce the time and cost of data collection. This method based on partial data training is more practical and efficient than the traditional method that requires complete data. Applying this method in the battery management system can obtain battery SOH information more quickly, providing a basis for the optimal management and use of the battery.

[0238] An embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the electric vehicle battery health status prediction method in embodiment 1 or embodiment 2 of the present application.

[0239] An embodiment of the present application provides a computer-readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the electric vehicle battery health status prediction method in embodiment 1 or embodiment 2 of the present application is executed.

[0240] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0241] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0242] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0243] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0244] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0245] 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.

Claims

1. A method for predicting the health status of an electric vehicle battery, characterized in that: include: Obtain training data and model training configuration data; Constructing an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer; The original model is trained by using the training data and the model training configuration data to obtain a battery capacity prediction model; Deploying the battery capacity prediction model onto an onboard hardware platform of a target vehicle; During the charging process of the target vehicle, sensor sampling data is acquired according to a preset collection time period and a preset number of sampling points; Calculating a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data; Calculating the battery SOH value according to the preset battery rated capacity and the current actual capacity prediction value; Outputting a battery health status assessment result including the battery SOH value; The step of calculating the current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data includes: Performing maximum value and average value pooling processing on the sensor sampling data simultaneously through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data; Concatenate the average value pooling output data and the maximum value pooling output data to obtain pooling layer output data; Performing a self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain a weight score corresponding to the output data of the pooling layer; Merging the pooling layer output data and the weight score through the channel merging layer of the battery capacity prediction model to obtain channel merging layer output data; Perform convolution calculation on the output data of the channel merging layer according to a preset convolution kernel size, a preset activation function, and a multi-scale convolution layer of the battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size; Merging the convolution data to obtain convolution layer output data; The actual capacity value is predicted based on the output layer of the battery capacity prediction model and the output data of the convolution layer to obtain the current actual capacity prediction value.

2. The electric vehicle battery health status prediction method according to claim 1, characterized in that: The obtaining of training data and model training configuration data includes: During the charging process of the electric vehicle, a battery data set is obtained through a sensor according to a preset collection time period and a preset number of sampling points; Performing labeling processing on the battery data group to obtain labeled data; Preprocessing the labeled data to obtain training data; Obtain model training configuration data; wherein the model training configuration data at least includes a loss function, a learning rate optimization algorithm, and a number of iterations.

3. The electric vehicle battery health status prediction method according to claim 2, characterized in that: The sensor includes at least a current sensor, a voltage sensor and a temperature sensor; The battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data; The loss function is a mean square error loss function; The learning rate optimization algorithm is the Adam adaptive learning rate optimization algorithm.

4. The electric vehicle battery health status prediction method according to claim 3, characterized in that: The labeling process of the battery data group to obtain the labeling data includes: Calculate the current actual battery capacity corresponding to each group of sensor collected data according to the battery data group; Capacity annotation processing is performed on each group of sensor collected data according to the current actual battery capacity to obtain annotated data.

5. An electric vehicle battery health status prediction device, characterized in that: The electric vehicle battery health status prediction device comprises: A first acquisition unit, used to acquire training data and model training configuration data; A construction unit, used to construct an original model based on a multi-scale convolutional neural network; wherein the original model at least includes a multi-scale pooling layer, a channel attention layer, a channel merging layer, a multi-scale convolution layer and an output layer; A training unit, used to train the original model using the training data and the model training configuration data to obtain a battery capacity prediction model; A model deployment unit, used to deploy the battery capacity prediction model onto an on-board hardware platform of a target vehicle; A second acquisition unit is used to acquire sensor sampling data according to a preset collection time period and a preset number of sampling points during the charging process of the target vehicle; A first calculation unit, configured to calculate a current actual capacity prediction value of the electric battery of the target vehicle according to the battery capacity prediction model and the sensor sampling data; A second calculation unit, used to calculate the battery SOH value according to a preset battery rated capacity and the current actual capacity prediction value; An output unit, used to output a battery health status assessment result including the battery SOH value; Wherein, the first calculation unit includes: A processing subunit, configured to simultaneously perform maximum value and average value pooling processing on the sensor sampling data through the multi-scale pooling layer of the battery capacity prediction model to obtain average value pooling output data and maximum value pooling output data; A splicing subunit, used for splicing the average pooling output data and the maximum pooling output data to obtain pooling layer output data; A calculation subunit, configured to perform a self-attention mechanism calculation on the output data of the pooling layer through the channel attention layer of the battery capacity prediction model to obtain a weight score corresponding to the output data of the pooling layer; A merging subunit, used for merging the pooling layer output data and the weight score through the channel merging layer of the battery capacity prediction model to obtain channel merging layer output data; The computing subunit is further configured to perform convolution calculation on the output data of the channel merging layer according to a preset convolution kernel size, a preset activation function, and a multi-scale convolution layer of the battery capacity prediction model to obtain convolution data corresponding to the preset convolution kernel size; The merging subunit is further used to merge the convolution data to obtain convolution layer output data; The prediction subunit is used to predict the actual capacity value based on the output layer of the battery capacity prediction model and the output data of the convolution layer to obtain the current actual capacity prediction value.

6. The electric vehicle battery health status prediction device according to claim 5, characterized in that: The first acquiring unit includes: A first acquisition subunit is used to acquire a battery data group through a sensor according to a preset collection time period and a preset number of sampling points during the charging process of the electric vehicle; A labeling subunit, used for labeling the battery data group to obtain labeling data; A labeling subunit, used for preprocessing the labeling data to obtain training data; The second acquisition subunit is used to acquire model training configuration data; wherein the model training configuration data at least includes a loss function, a learning rate optimization algorithm and a number of iterations.

7. The electric vehicle battery health status prediction device according to claim 6, characterized in that: The sensor includes at least a current sensor, a voltage sensor and a temperature sensor; The battery data group includes at least one group of sensor collected data, and each group of sensor collected data includes at least voltage data, current data and temperature data; The loss function is a mean square error loss function; The learning rate optimization algorithm is the Adam adaptive learning rate optimization algorithm.

8. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the electric vehicle battery health status prediction method according to any one of claims 1 to 4.

9. A readable storage medium, characterized in that: The readable storage medium stores computer program instructions, and when the computer program instructions are read and executed by a processor, the electric vehicle battery health status prediction method according to any one of claims 1 to 4 is executed.

Images