SOH prediction method and system based on multi-head CNN-LSTM model

Abstract

The invention discloses an SOH prediction method and system based on a multi-head CNN-LSTM model, and a readable storage medium. The method comprises the following steps: testing and collecting electrochemical impedance spectroscopy data of a battery; a multi-head CNN-LSTM model is constructed and trained; and inputting the electrochemical impedance spectroscopy data of the battery into the multi-head CNN-LSTM model for SOH prediction, and outputting a prediction result. According to the method, the multi-head CNN model and the LSTM model are combined, and the SOH is jointly predicted based on the EIS data. EIS is regarded as a characteristic, and a real part and an imaginary part of impedance data corresponding to each frequency are regarded as characteristic values. A multi-head CNN model is used for extracting spatial features of an impedance spectrum, an LSTM model is used for extracting a time sequence change rule of the impedance spectrum, and SOH is jointly predicted. According to the method, the SOH prediction precision can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of electrochemical technology, and more specifically, to a SOH prediction method, system and readable storage medium based on a multi-head CNN-LSTM model. Background Art

[0002] Electrochemical Impedance Spectroscopy (EIS) is a test method used to characterize the electrochemical process inside a battery. EIS can be used to test the impedance spectrum of the battery, including real impedance and imaginary impedance. The attenuation of the battery's state of health (SOH) will lead to an increase in the battery's internal resistance and AC impedance, and the impedance spectrum can reflect the state of the battery's SOH. A large number of studies have shown that the impedance spectrum obtained by EIS testing can be used to predict the battery's SOH.

[0003] Chinese patent CN112611976A published on April 6, 2021 provides a method for estimating the health status of a power battery based on a double difference curve. For a battery management system, accurately predicting the health status of the battery can effectively ensure the reliability and safety of the system and provide a reference for the rational use of the battery in the future. This solution provides a method for estimating the health status of a power battery based on a double difference curve. By making full use of the information of the incremental capacity curve and the temperature difference curve, the feature vector corresponding to each cycle is obtained, and the battery state of health (SOH) is estimated through a one-dimensional convolutional neural network (CNN) and an improved recurrent neural network (RNN) model. The one-dimensional CNN is used for feature extraction, and the improved RNN is used for time series prediction. At the same time, the shortcomings of the traditional RNN are overcome, thereby obtaining a method for estimating the health status of a battery that makes full use of process information and estimates accurately. However, the scheme is relatively simple in terms of feature extraction, resulting in the accuracy of the prediction of the battery SOH still needs to be improved. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a SOH prediction method, system and readable storage medium based on a multi-head CNN-LSTM model.

[0005] A first aspect of the present invention provides a SOH prediction method based on a multi-head CNN-LSTM model, the method comprising the following steps:

[0006] Test and collect battery electrochemical impedance spectroscopy data;

[0007] Build and train a multi-head CNN-LSTM model;

[0008] The battery electrochemical impedance spectroscopy data is input into the multi-head CNN-LSTM model for SOH prediction, and the prediction results are output.

[0009] Preferably, the multi-head CNN-LSTM model includes: an input layer, a multi-head CNN model, an LSTM model, and a linear regression layer;

[0010] The input layer is used to input the battery electrochemical impedance spectroscopy data into the multi-head CNN model;

[0011] The multi-head CNN model extracts EIS spatial features based on the battery electrochemical impedance spectroscopy data and outputs the EIS spatial features to the LSTM model;

[0012] The LSTM model extracts the time series variation law of EIS according to the spatial features of EIS, and outputs the time series variation law of EIS to the linear regression layer;

[0013] The linear regression layer predicts SOH based on the spatial characteristics of EIS and the time series variation law of EIS, and outputs the prediction results.

[0014] Preferably, the collecting of battery electrochemical impedance spectroscopy data is specifically:

[0015] Collecting battery electrochemical impedance spectroscopy data at a preset frequency interval;

[0016] When collecting battery electrochemical impedance spectroscopy data, the real and imaginary impedance data of the battery electrochemical impedance spectroscopy data corresponding to each frequency are recorded; as well as the target value to be predicted, that is, the battery capacity data.

[0017] Preferably, the number of heads of the multi-head CNN model is 2.

[0018] Preferably, the multi-head CNN model includes two parallel sub-CNNs, each sub-CNN is a double-layer convolution calculation structure, the activation function is a relu function, the number of filters is set to 32, and the kernel size is 1.

[0019] Preferably, the LSTM model includes two LSTM layers connected in series, the number of neurons in the LSTM layer is 64, and the activation function is the relu function.

[0020] Preferably, the activation function of the linear regression layer is a linear function.

[0021] Preferably, the loss function of the multi-head CNN-LSTM model is mean square error, the optimizer is rmsprop, and the evaluation indicator metrics is mean absolute error.

[0022] The second aspect of the present invention provides a SOH prediction system based on a multi-head CNN-LSTM model, comprising a memory and a processor, wherein the memory includes a SOH prediction method program based on the multi-head CNN-LSTM model, and when the SOH prediction method program based on the multi-head CNN-LSTM model is executed by the processor, the steps of the SOH prediction method based on the multi-head CNN-LSTM model are implemented.

[0023] The third aspect of the present invention provides a readable storage medium, which includes a SOH prediction method program based on a multi-head CNN-LSTM model. When the SOH prediction method program based on the multi-head CNN-LSTM model is executed by a processor, the steps of the SOH prediction method based on the multi-head CNN-LSTM model are implemented.

[0024] Compared with the prior art, the beneficial effect of the technical solution of the present invention is as follows: the present invention combines the multi-head CNN and LSTM models to jointly predict SOH based on EIS data. EIS is regarded as a feature, and the real and imaginary parts of the impedance data corresponding to each frequency are used as feature values. The multi-head CNN model is used to extract the spatial features of the impedance spectrum, and the LSTM model is used to extract the time series variation law of the impedance spectrum, and jointly predict SOH. The present invention can effectively improve the accuracy of SOH prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 This is a flow chart of a SOH prediction method based on a multi-head CNN-LSTM model described in Example 1.

[0026] Figure 2 This is the structure diagram of the multi-head CNN-LSTM model.

[0027] Figure 3 This is the basic structure diagram of CNN.

[0028] Figure 4 This is the structure diagram of the multi-head CNN model.

[0029] Figure 5 This is the structure diagram of the LSTM model.

[0030] Figure 6 This is a sample diagram of SOH prediction results based on the multi-head CNN-LSTM model. DETAILED DESCRIPTION

[0031] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0032] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0033] Example 1

[0034] like Figure 1 As shown, this embodiment discloses a SOH prediction method based on a multi-head CNN-LSTM model, and the method includes the following steps:

[0035] Test and collect battery electrochemical impedance spectroscopy data;

[0036] Build and train a multi-head CNN-LSTM model;

[0037] The battery electrochemical impedance spectroscopy data is input into the multi-head CNN-LSTM model for SOH prediction, and the prediction results are output.

[0038] It should be noted that this application improves the traditional EIS SOH prediction model and constructs a multi-head CNN-LSTM model for SOH prediction to improve the accuracy of SOH prediction.

[0039] According to an embodiment of the present invention, the multi-head CNN-LSTM model includes: an input layer, a multi-head CNN model, an LSTM model, and a linear regression layer;

[0040] The input layer is used to input the battery electrochemical impedance spectroscopy data into the multi-head CNN model;

[0041] The multi-head CNN model extracts EIS spatial features based on the battery electrochemical impedance spectroscopy data and outputs the EIS spatial features to the LSTM model;

[0042] The LSTM model extracts the time series variation law of EIS according to the spatial features of EIS, and outputs the time series variation law of EIS to the linear regression layer;

[0043] The linear regression layer predicts SOH based on the spatial characteristics of EIS and the time series variation law of EIS, and outputs the prediction results.

[0044] It should be noted that in the model using EIS to predict SOH, the convolution neural network (CNN) can extract the spatial features of EIS. The multi-head convolution neural network is an improved convolution neural network structure. Unlike the traditional CNN that only uses a single convolution kernel, the multi-head CNN uses multiple different convolution kernels to extract different features, generate multiple feature maps, and finally fuse these feature maps to obtain an expression of the input. Using multiple convolution kernels to extract different feature expressions, the multi-head CNN can obtain a more comprehensive understanding of the input and enhance the generalization ability and stability of the model.

[0045] In addition, the Long Short-Term Memory (LSTM) network can learn the time series variation patterns of EIS.

[0046] In summary, the combined model of multi-head CNN and LSTM can learn the complex mapping relationship between EIS and SOH, so the combination of the above models can be used to predict SOH based on battery EIS data.

[0047] This application combines the multi-head CNN and LSTM models to jointly predict SOH based on EIS data. EIS is regarded as a feature, and the real and imaginary parts of the impedance data corresponding to each frequency are used as feature values. The multi-head CNN model is used to extract the spatial features of the impedance spectrum, and the LSTM model is used to extract the time series change law of the impedance spectrum to jointly predict SOH.

[0048] In this embodiment, the multi-head CNN-LSTM model is as follows Figure 2 As shown. Figure 2 In the input part, real1 to real60 represent the real data of the impedance data corresponding to the high frequency to the low frequency, and imaginary1 to imaginary60 represent the imaginary data of the impedance data corresponding to the high frequency to the low frequency;

[0049] Next is the multi-head CNN structure, the activation function is relu, and the number of heads of the multi-head CNN is 2;

[0050] After that, the multi-head CNN calculation results need to be concatenated, that is, the Concatenate operation;

[0051] Then, the concatenated layer of the multi-head CNN is connected to a 2-layer LSTM model structure, and the activation function of the LSTM model is relu;

[0052] Finally, after passing through a linear regression layer, SOH is output.

[0053] According to an embodiment of the present invention, the acquisition of battery electrochemical impedance spectroscopy data is specifically:

[0054] Collecting battery electrochemical impedance spectroscopy data at a preset frequency interval;

[0055] When collecting battery electrochemical impedance spectroscopy data, the real and imaginary impedance data of the battery electrochemical impedance spectroscopy data corresponding to each frequency are recorded; as well as the target value to be predicted, that is, the battery capacity data.

[0056] It should be noted that testing and collecting battery electrochemical impedance spectrum data requires professional impedance spectrum testing instruments and needs to cover a certain frequency range. The frequency range can be set from 10-2HZ to 105HZ. Multiple frequency values ​​are selected at certain intervals in the middle of the frequency range. The frequency values ​​are generally selected from 5 to 60. In this embodiment, 60 are selected as a sample.

[0057] When collecting data, it is necessary to record the real and imaginary data of the impedance corresponding to each frequency, as well as the target value that needs to be predicted, which is the corresponding battery capacity data.

[0058] According to an embodiment of the present invention, the number of heads of the multi-head CNN model is 2.

[0059] According to an embodiment of the present invention, the multi-head CNN model includes two parallel sub-CNNs, each sub-CNN is a double-layer convolution calculation structure, the activation function is a relu function, the number of filters is set to 32, and the kernel size is 1.

[0060] It should be noted that Multi-Head CNN refers to the use of multiple parallel convolution channels or branches in the CNN model.

[0061] First, we introduce the basic structure of the CNN model. The basic structure of the CNN model is as follows: Figure 3 shown.

[0062] The CNN model structure mainly includes the following components:

[0063] (1) Input layer: input data

[0064] (2) Convolutional layer: Use convolution kernels to extract data features

[0065] (3) Pooling layer: downsampling to reduce parameters

[0066] (4) Fully connected layer: maps features to classification output

[0067] (5) Output layer: output prediction results

[0068] In the actual process of building a prediction model, the module involving CNN can only select some calculation methods in the basic structure of CNN, such as convolution calculation or a combination of convolution calculation and pooling calculation, or a combination of multiple calculation methods.

[0069] The algorithm principles of multi-head CNN include the following points:

[0070] (1) Building a multi-head structure

[0071] In the CNN model, multiple parallel convolution branches or channels are constructed, which are called multi-head structures.

[0072] (2) Extracting multiple features

[0073] Let each convolution head learn different types of feature representations, such as through different convolution kernels, convolution layer parameters, etc. In this way, diverse features can be obtained.

[0074] (3) Feature Fusion

[0075] The feature maps extracted by each convolution head are fused using concatenation or weighted summation. The fused feature contains richer information than a single feature.

[0076] (4) Final prediction

[0077] Input the fused feature map into the subsequent model structure to complete the final prediction task.

[0078] The multi-head CNN model described in this embodiment is as follows Figure 4 As shown in the figure, the multi-head CNN model contains two sub-CNNs, each of which has a double-layer convolution calculation structure, the activation function is the relu function, the number of filters is set to 32, and the kernel size is 1.

[0079] According to an embodiment of the present invention, the LSTM model includes two LSTM layers connected in series, the number of neurons in the LSTM layer is 64, and the activation function is the relu function.

[0080] It should be noted that LSTM is a special RNN that can learn long-term dependencies and capture dynamic changes in time series data. The LSTM structure in this embodiment includes two LSTM layers, the number of neurons in the LSTM layer is 64, and the activation function is the relu function. The LSTM layer can learn the long-term dependencies of the input data and identify the dynamic change pattern of the time series.

[0081] like Figure 5 As shown in the figure, the LSTM model consists of the following components:

[0082] (1) Input gate: determines which parts of the current input information are allowed to enter the state unit.

[0083] (2) Forget gate: determines which information in the state unit needs to be forgotten or removed.

[0084] (3) Status unit: The core component used to store historical status information.

[0085] (4) Output gate: determines the information output to the next hidden layer.

[0086] (5) Activation function (Sigmoid, Tanh): used for nonlinear transformation.

[0087] (6) Output layer: LSTM outputs data corresponding to the input data.

[0088] LSTM controls the flow of information entry, retention and output by opening and closing the input gate, forget gate and output gate, and realizes the memory and processing of time series information. Compared with simple RNN, LSTM can better capture the long-term dependencies in time series data, process longer time series, and has more powerful modeling capabilities.

[0089] According to an embodiment of the present invention, the activation function of the linear regression layer is a linear function.

[0090] According to an embodiment of the present invention, the loss function of the multi-head CNN-LSTM model is mean square error (MSE); the optimizer is rmsprop; and the evaluation indicator metrics is mean absolute error (MAE), that is, the sum of the absolute values ​​of the differences between the target value and the predicted value.

[0091] As a specific embodiment, the following is described in conjunction with specific data:

[0092] The present invention tests the effect on a public data set, the data comes from https: / / github.com / YunweiZhang / ML-identify-battery-degradation, and uses the impedance data therein to predict the battery capacity. The data set is divided into a training set and a test set in a ratio of 8:2, a multi-head CNN-LSTM model is trained on the training set, and the effect is tested on the test set. The test data has a total of 120 features, the feature numbers are 1 to 120, the first 60 features are the real part data of the impedance data corresponding to the frequency from high to low, and the last 60 features are the imaginary part data of the impedance data corresponding to the frequency from high to low. The battery EIS predicts the battery SOH partial data set as shown in Table 1.

[0093] Table 1 Battery EIS prediction battery SOH data set sample (partial)

[0094] f1 f2 f3 …… f119 f120 SOH 0.3847 0.39156 0.39684 …… 0.29026 0.32795 37.2027 0.38886 0.39343 0.40066 …… 0.28064 0.323 36.2230 0.39038 0.396 0.4025 …… 0.28773 0.32955 35.5893 0.39194 0.39643 0.40406 …… 0.29768 0.33673 35.1081 0.39225 0.3976 0.4054 …… 0.28396 0.32562 34.7673

[0095] Among them, f1 to f120 represent the EIS frequency characteristics, and the SOH here represents the maximum capacity of the battery and also reflects the health of the battery.

[0096] The mean of the test set is 30.5773, the standard deviation is 4.0661, and the mean of the prediction set is 30.3889, the standard deviation is 4.1317, which shows that the data distribution of the prediction results is basically consistent with the data distribution of the real data.

[0097] The root mean square error between the prediction set and the test set is only 0.3716, indicating that the model has a good prediction effect.

[0098] The prediction results are shown in Table 2:

[0099] Table 2 Battery EIS prediction battery SOH prediction result example (partial)

[0100]

[0101]

[0102] The prediction results are as follows: Figure 6 As shown, Figure 6 The solid line in the upper middle represents the true value, and the dotted line represents the predicted value. The numbers below are the true value, and the numbers above are the predicted value. The dotted line below represents the error value, and the numbers are the error. From the overlap degree of the broken lines and the error value, we can see that the prediction effect is good.

[0103] Example 2

[0104] This embodiment discloses a SOH prediction system based on a multi-head CNN-LSTM model, including a memory and a processor, wherein the memory includes a SOH prediction method program based on the multi-head CNN-LSTM model, and when the SOH prediction method program based on the multi-head CNN-LSTM model is executed by the processor, the method steps described in Example 1 are implemented.

[0105] Example 3

[0106] The present embodiment discloses a readable storage medium, which includes a SOH prediction method program based on a multi-head CNN-LSTM model. When the SOH prediction method program based on the multi-head CNN-LSTM model is executed by a processor, the steps of a SOH prediction method based on a multi-head CNN-LSTM model as described in Example 1 are implemented.

[0107] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0108] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0109] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0110] Those skilled in the art can understand that: all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, and other media that can store program codes.

[0111] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

Claims

1. A SOH prediction method based on a multi-head CNN-LSTM model, characterized in that: The method comprises the following steps: Test and collect battery electrochemical impedance spectroscopy data; Build and train a multi-head CNN-LSTM model; The battery electrochemical impedance spectroscopy data is input into the multi-head CNN-LSTM model for SOH prediction, and the prediction results are output.

2. The SOH prediction method based on the multi-head CNN-LSTM model according to claim 1 is characterized in that: The multi-head CNN-LSTM model includes: an input layer, a multi-head CNN model, an LSTM model, and a linear regression layer; The input layer is used to input the battery electrochemical impedance spectroscopy data into the multi-head CNN model; The multi-head CNN model extracts EIS spatial features based on the battery electrochemical impedance spectroscopy data and outputs the EIS spatial features to the LSTM model; The LSTM model extracts the time series variation law of EIS according to the spatial features of EIS, and outputs the time series variation law of EIS to the linear regression layer; The linear regression layer predicts SOH based on the spatial characteristics of EIS and the time series variation law of EIS, and outputs the prediction results.

3. The SOH prediction method based on the multi-head CNN-LSTM model according to claim 1 is characterized in that: The collecting of battery electrochemical impedance spectroscopy data is specifically as follows: Collecting battery electrochemical impedance spectroscopy data at a preset frequency interval; When collecting battery electrochemical impedance spectroscopy data, the real and imaginary impedance data of the battery electrochemical impedance spectroscopy data corresponding to each frequency are recorded; as well as the target value to be predicted, that is, the battery capacity data.

4. The SOH prediction method based on a multi-head CNN-LSTM model according to claim 1, characterized in that: The number of heads of the multi-head CNN model is 2.

5. The SOH prediction method based on a multi-head CNN-LSTM model according to claim 1, characterized in that: The multi-head CNN model includes two parallel sub-CNNs, each sub-CNN is a double-layer convolution calculation structure, the activation function is the relu function, the number of filters is set to 32, and the kernel size is 1.

6. The SOH prediction method based on a multi-head CNN-LSTM model according to claim 1, characterized in that: The LSTM model includes two LSTM layers connected in series. The number of neurons in the LSTM layer is 64, and the activation function is the relu function.

7. The SOH prediction method based on a multi-head CNN-LSTM model according to claim 1, characterized in that: The activation function of the linear regression layer is a linear function.

8. The SOH prediction method based on a multi-head CNN-LSTM model according to claim 1, characterized in that: The loss function of the multi-head CNN-LSTM model is mean square error, the optimizer is rmsprop, and the evaluation metric is mean absolute error.

9. A SOH prediction system based on a multi-head CNN-LSTM model, characterized in that: It includes a memory and a processor, wherein the memory includes a SOH prediction method program based on a multi-head CNN-LSTM model, and when the SOH prediction method program based on the multi-head CNN-LSTM model is executed by the processor, the method steps described in any one of claims 1 to 8 are implemented.

10. A readable storage medium, characterized in that: The readable storage medium includes a SOH prediction method program based on a multi-head CNN-LSTM model. When the SOH prediction method program based on a multi-head CNN-LSTM model is executed by a processor, the steps of a SOH prediction method based on a multi-head CNN-LSTM model as described in any one of claims 1 to 8 are implemented.

Images