A self-supervised-based battery feature extraction method, system, device and medium

Abstract

The application provides a battery operation and maintenance technical field, and relates to a battery feature extraction method, system, device and medium based on self-supervision, which comprises the following steps: step S1, creating a self-supervision learning battery feature extraction model based on a neural network; step S2, obtaining a large amount of battery charging and discharging data, and randomly cutting each piece of the charging and discharging data in the same charging and discharging cycle into two equal-length charging and discharging sub-data; step S3, inputting each piece of the charging and discharging sub-data into the battery feature extraction model for self-supervision learning training; and step S4, extracting battery features by using an encoder in the trained battery feature extraction model. The application has the advantage that the battery feature extraction efficiency is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of battery operation and maintenance technology, and in particular to a battery feature extraction method, system, device and medium based on self-supervision. Background Art

[0002] As new energy vehicles become increasingly popular, we can obtain massive amounts of battery charge and discharge data. In order to determine the health of the battery and perform better battery operation and maintenance, it is necessary to extract battery features from the charge and discharge data. For battery feature extraction, the traditional method is to train the model with a supervised learning method based on a neural network, and then extract battery features based on the trained model. However, supervised learning technology requires labeling massive amounts of charge and discharge data, which takes a lot of time and resources. For example, to build a SOH (battery health) prediction task, it is necessary to collect charge and discharge data with characteristics such as voltage, current, and temperature, and also know the true SOH labels of these charge and discharge data. However, obtaining the true SOH labels often requires another standard charge and discharge process to obtain them, which is very cumbersome and inefficient.

[0003] Therefore, how to provide a battery feature extraction method, system, device and medium based on self-supervision to improve the efficiency of battery feature extraction has become a technical problem that needs to be solved urgently. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a battery feature extraction method, system, device and medium based on self-supervision to improve the efficiency of battery feature extraction.

[0005] In a first aspect, the present invention provides a battery feature extraction method based on self-supervision, comprising the following steps:

[0006] Step S1: creating a battery feature extraction model for self-supervised learning based on a neural network;

[0007] Step S2: obtaining a large amount of battery charge and discharge data, and randomly cutting each piece of charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments;

[0008] Step S3: inputting the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training;

[0009] Step S4: extract battery features using the trained encoder in the battery feature extraction model.

[0010] Furthermore, in step S1, the battery feature extraction model includes an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer;

[0011] The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer.

[0012] Furthermore, the first predictor is used to perform supervised learning on labeled charging and discharging data, and the loss function adopts the MSE function; the second predictor is used to perform high-dimensional feature projection on the output of the first embedding layer, and the loss function adopts the cosine similarity function.

[0013] Furthermore, the step S3 is specifically as follows:

[0014] The charge and discharge electronic data are input into the battery feature extraction model for self-supervised learning training until the set convergence conditions are met. The encoder in the trained battery feature extraction model is connected to the downstream task, and the training results are then verified using the labeled charge and discharge data.

[0015] In a second aspect, the present invention provides a battery feature extraction system based on self-supervision, comprising the following modules:

[0016] A battery feature extraction model creation module is used to create a battery feature extraction model based on a neural network and self-supervised learning;

[0017] A charge and discharge data clipping module is used to obtain a large amount of battery charge and discharge data and randomly clip each charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments;

[0018] A model training module, configured to input the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training;

[0019] The battery feature extraction module is used to extract battery features using the encoder in the trained battery feature extraction model.

[0020] Furthermore, in the battery feature extraction model creation module, the battery feature extraction model includes an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer;

[0021] The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer.

[0022] Further, the first predictor is used for supervised learning on the labeled charging and discharging data, and a loss function adopts an MSE function; the second predictor is used for high-dimensional feature projection on the output of the first embedding layer, and a loss function adopts a cosine similarity function.

[0023] Further, the model training module is specifically used for:

[0024] The charging and discharging data are input into the battery feature extraction model for self-supervised learning training until a set convergence condition is met, the encoder in the trained battery feature extraction model is connected to a downstream task, and then the training result is verified by using the labeled charging and discharging data.

[0025] In a third aspect, the present application provides a battery feature extraction device based on self-supervision, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, and the processor implements the method of the first aspect when executing the program.

[0026] In a fourth aspect, the present application provides a battery feature extraction medium based on self-supervision, which stores a computer program, and the program is executed by a processor to implement the method of the first aspect.

[0027] The one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

[0028] The battery feature extraction model based on self-supervised learning is created by a neural network, each piece of charging and discharging data in the same charging and discharging cycle is randomly cut into two equal-length charging and discharging data, and then the battery feature extraction model is input for self-supervised learning training, and then the encoder in the trained battery feature extraction model is connected to a downstream task for battery feature extraction, that is, self-supervised learning is used instead of traditional supervised learning, and the training of the battery feature extraction model can be completed without manual labeling of the charging and discharging data, thereby greatly improving the battery feature extraction efficiency.

[0029] The above description is only a summary of the technical solutions of the present application, in order to more clearly understand the technical means of the present application, the content of the specification can be implemented, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0030] The present application will be further described below with reference to the accompanying drawings and embodiments.

[0031] Figure 1 is a flowchart of a battery feature extraction method based on self-supervision.

[0032] Figure 2It is a structural diagram of a battery feature extraction system based on self-supervision of the present invention.

[0033] Figure 3 It is a structural schematic diagram of a battery feature extraction device based on self-supervision of the present invention.

[0034] Figure 4 It is a structural schematic diagram of a battery feature extraction medium based on self-supervision of the present invention.

[0035] Figure 5 It is a structural diagram of the battery feature extraction model of the present invention. DETAILED DESCRIPTION

[0036] The embodiments of the present application improve the efficiency of battery feature extraction by providing a battery feature extraction method, system, device and medium based on self-supervision.

[0037] The technical solution in the embodiments of the present application has the following overall idea: the battery feature extraction model of self-supervised learning created based on a neural network is trained using charging and discharging data. There is no need to manually label the charging and discharging data before training. The encoder in the trained battery feature extraction model is then used to extract battery features to improve the efficiency of battery feature extraction.

[0038] Example 1

[0039] This embodiment provides a battery feature extraction method based on self-supervision, such as Figure 1 、 5 As shown, the following steps are included:

[0040] Step S1: creating a battery feature extraction model for self-supervised learning based on a neural network;

[0041] Step S2: obtaining a large amount of battery charge and discharge data, and randomly cutting each piece of charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments;

[0042] Step S3: inputting the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training;

[0043] Step S4: extract battery features using the trained encoder in the battery feature extraction model.

[0044] In step S1, the battery feature extraction model includes an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer;

[0045] The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer.

[0046] The first predictor is used for supervised learning of labeled charging and discharging data, and the loss function adopts the MSE function, with the formula L=Lcontrast+λLlabel, that is, the first predictor is used for weak supervision tasks and does not work for unlabeled charging and discharging data; the second predictor is used to perform high-dimensional feature projection on the output of the first embedding layer, and the loss function adopts the cosine similarity function. The second predictor is used for contrastive proxy tasks, that is, it is hoped that the two features embedding1 and embedding3 are as similar as possible; the encoder is used to extract unique representations of charging and discharging data.

[0047] This paper constructs a reasonable proxy task and loss function to enable the battery feature extraction model to autonomously discover the underlying structure in the data. Because the proxy task constructed in this paper is primarily designed to enable the network's encoder to extract high-dimensional representations, and the predictor is designed to assist the encoder in learning, only the encoder needs to be retained for fine-tuning when migrating downstream tasks.

[0048] The step S3 is specifically as follows:

[0049] Dividing the charge and discharge electronic data into a training set and a validation set according to a preset ratio, and setting a convergence condition and an accuracy threshold;

[0050] The charge and discharge data are fed into a battery feature extraction model for self-supervised training until a predetermined convergence condition is met. The encoder in the trained battery feature extraction model is then connected to a downstream task, and the training results are verified using labeled charge and discharge data. The convergence condition may be that the loss value of a predetermined loss function is less than a predetermined threshold.

[0051] During specific training, assume that the two segments of charging and discharging electronic data cropped from a charging and discharging data are input1 and input2. Since input1 and input2 come from the same charging and discharging cycle, their potential features should be similar. Therefore, the embedding1 obtained by input1 after the encoder and the embedding3 obtained by input2 after the encoder and the second predictor should be similar.

[0052] Example 2

[0053] This embodiment provides a battery feature extraction system based on self-supervision, such as Figure 2 、 5 As shown, it includes the following modules:

[0054] A battery feature extraction model creation module is used to create a battery feature extraction model based on a neural network and self-supervised learning;

[0055] A charge and discharge data clipping module is used to obtain a large amount of battery charge and discharge data and randomly clip each charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments;

[0056] A model training module, configured to input the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training;

[0057] The battery feature extraction module is used to extract battery features using the encoder in the trained battery feature extraction model.

[0058] In the battery feature extraction model creation module, the battery feature extraction model includes an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer;

[0059] The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer.

[0060] The first predictor is used for supervised learning of labeled charging and discharging data, and the loss function adopts the MSE function, with the formula L=Lcontrast+λLlabel, that is, the first predictor is used for weak supervision tasks and does not work for unlabeled charging and discharging data; the second predictor is used to perform high-dimensional feature projection on the output of the first embedding layer, and the loss function adopts the cosine similarity function. The second predictor is used for contrastive proxy tasks, that is, it is hoped that the two features embedding1 and embedding3 are as similar as possible; the encoder is used to extract unique representations of charging and discharging data.

[0061] This paper constructs a reasonable proxy task and loss function to enable the battery feature extraction model to autonomously discover the underlying structure in the data. Because the proxy task constructed in this paper is primarily designed to enable the network's encoder to extract high-dimensional representations, and the predictor is designed to assist the encoder in learning, only the encoder needs to be retained for fine-tuning when migrating downstream tasks.

[0062] The model training module is specifically used for:

[0063] The charge and discharge data are fed into a battery feature extraction model for self-supervised training until a predetermined convergence condition is met. The encoder in the trained battery feature extraction model is then connected to a downstream task, and the training results are verified using labeled charge and discharge data. The convergence condition may be that the loss value of a predetermined loss function is less than a predetermined threshold.

[0064] During specific training, assume that the two segments of charging and discharging electronic data cropped from a charging and discharging data are input1 and input2. Since input1 and input2 come from the same charging and discharging cycle, their potential features should be similar. Therefore, the embedding1 obtained by input1 after the encoder and the embedding3 obtained by input2 after the encoder and the second predictor should be similar.

[0065] Based on the same inventive concept, this application provides an electronic device embodiment corresponding to the first embodiment, see the third embodiment for details.

[0066] Example 3

[0067] This embodiment provides a battery feature extraction device based on self-supervision, such as Figure 3 As shown, it includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, any implementation method in the first embodiment can be implemented.

[0068] Since the electronic device described in this embodiment is the device used to implement the method in Example 1 of this application, based on the method described in Example 1 of this application, those skilled in the art will be able to understand the specific implementation of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiment of this application will not be described in detail here. As long as the device used by those skilled in the art to implement the method in the embodiment of this application falls within the scope of protection to be provided by this application.

[0069] Based on the same inventive concept, this application provides a storage medium corresponding to Example 1, see Example 4 for details.

[0070] Example 4

[0071] This embodiment provides a battery feature extraction medium based on self-supervision, such as Figure 4 As shown, a computer program is stored thereon, and when the computer program is executed by a processor, any implementation method in Example 1 can be implemented.

[0072] Since the storage medium described in this embodiment is the storage medium used to implement the method in Example 1 of this application, based on the method described in Example 1 of this application, those skilled in the art will be able to understand the specific implementation and various variations of the storage medium of this embodiment, so how the storage medium implements the method in the embodiment of this application will not be described in detail here. As long as those skilled in the art implement the storage medium used in the method in the embodiment of this application, it falls within the scope of protection of this application.

[0073] The technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0074] A self-supervised learning battery feature extraction model is created through a neural network. After randomly cutting each charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge electronic data, the data is input into the battery feature extraction model for self-supervised learning training. The encoder in the trained battery feature extraction model is then connected to the downstream task for battery feature extraction. That is, self-supervised learning replaces traditional supervised learning, and the battery feature extraction model can be trained without manually labeling the charge and discharge data, thereby greatly improving the efficiency of battery feature extraction.

[0075] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as the combination of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0077] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0079] Although the specific embodiments of the present invention are described above, those skilled in the art should understand that the specific embodiments described are merely illustrative and are not intended to limit the scope of the present invention. Equivalent modifications and changes made by those skilled in the art in accordance with the spirit of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A battery feature extraction method based on self-supervision, characterized by: The steps include: Step S1, creating a self-supervised learning battery feature extraction model based on a neural network; the battery feature extraction model includes an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer; The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer; Step S2: obtaining a large amount of battery charge and discharge data, and randomly cutting each piece of charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments; Step S3: inputting the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training; Step S4: extract battery features using the trained encoder in the battery feature extraction model.

2. The battery feature extraction method based on self-supervision according to claim 1, characterized in that: The first predictor is used to perform supervised learning on labeled charging and discharging data, and the loss function adopts the MSE function; the second predictor is used to perform high-dimensional feature projection on the output of the first embedding layer, and the loss function adopts the cosine similarity function.

3. The battery feature extraction method based on self-supervision according to claim 1, characterized in that: The step S3 is specifically as follows: The charge and discharge electronic data are input into the battery feature extraction model for self-supervised learning training until the set convergence conditions are met. The encoder in the trained battery feature extraction model is connected to the downstream task, and the training results are then verified using the labeled charge and discharge data.

4. A battery feature extraction system based on self-supervision, characterized by: Includes the following modules: A battery feature extraction model creation module, configured to create a self-supervised learning battery feature extraction model based on a neural network; the battery feature extraction model comprising an encoder, a first predictor, a second predictor, a first embedding layer, a second embedding layer, and a third embedding layer; The output end of the encoder is connected to the input end of the first embedding layer and the input end of the second embedding layer respectively; the output end of the first embedding layer is connected to the input end of the first predictor and the second predictor respectively; the output end of the second predictor is connected to the third embedding layer; A charge and discharge data clipping module is used to obtain a large amount of battery charge and discharge data and randomly clip each charge and discharge data in the same charge and discharge cycle into two equal-length charge and discharge data segments; A model training module, configured to input the charge and discharge electronic data into a battery feature extraction model for self-supervised learning training; The battery feature extraction module is used to extract battery features using the encoder in the trained battery feature extraction model.

5. The battery feature extraction system based on self-supervision according to claim 4, characterized in that: The first predictor is used to perform supervised learning on labeled charging and discharging data, and the loss function adopts the MSE function; the second predictor is used to perform high-dimensional feature projection on the output of the first embedding layer, and the loss function adopts the cosine similarity function.

6. The battery feature extraction system based on self-supervision according to claim 4, characterized in that: The model training module is specifically used for: The charge and discharge electronic data are input into the battery feature extraction model for self-supervised learning training until the set convergence conditions are met. The encoder in the trained battery feature extraction model is connected to the downstream task, and the training results are then verified using the labeled charge and discharge data.

7. A battery feature extraction device based on self-supervision, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 3 is implemented.

8. A battery feature extraction medium based on self-supervision, having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

Images