Battery state of charge estimation method, system, equipment, medium and product
By combining network structure search and knowledge distillation technology in battery state of charge estimation, the deep learning model is optimized, and the applicability problem caused by excessive model in the existing technology is solved, and efficient and accurate battery state of charge estimation is achieved.
Patent Information
- Application Number
- CN202411851873.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-16
- Publication Date
- 2025-03-25
AI Technical Summary
The existing deep learning-based battery state of charge estimation method is too large to run on mobile or embedded devices, resulting in poor applicability.
By combining network structure search algorithm and knowledge distillation technology, the teacher model is trained and optimized to obtain a smaller but highly accurate student model to achieve efficient estimation of battery charge state.
The deployment of battery state of charge estimation model on mobile terminals or embedded devices is realized, which improves the estimation efficiency and accuracy, and solves the problem of poor applicability of existing methods.
Smart Images
Figure CN119689268A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of state of charge estimation, and in particular to a battery state of charge estimation method, system, device, medium and product. Background Art
[0002] Accurate power battery state of charge estimation can provide reliable reference information for the battery management system to ensure that the battery operates within a safe operating range, optimize energy use, and extend battery life. In addition, users can judge the vehicle's endurance based on the state of charge, better plan their trips and alleviate mileage anxiety. In recent years, deep learning technology has shown great application potential and made significant progress in the field of intelligent battery status monitoring due to its automatic feature extraction capabilities and powerful nonlinear fitting capabilities. However, most deep learning models are too large to run on mobile or embedded devices, resulting in poor applicability of deep learning-based battery state estimation methods.
[0003] Therefore, it is necessary to optimize the network structure to reduce the model size and improve the applicability of the model. The current mainstream network structure optimization methods mainly include pruning, knowledge distillation, and network architecture search technology. Among them, pruning technology can reduce the size and computational complexity of the model by removing unimportant structures in the network, but it may cause a certain loss of accuracy; knowledge distillation technology can transfer the knowledge of the teacher model to the student model to obtain a small model with strong representation ability, but choosing different student model structures will affect the learning results; network architecture search technology can automatically search and design neural network structures suitable for specific tasks, but the larger search space makes the search cost higher. Summary of the invention
[0004] The purpose of this application is to provide a battery state of charge estimation method, system, device, medium and product to improve the efficiency and accuracy of battery state of charge estimation.
[0005] To achieve the above objectives, this application provides the following solutions:
[0006] In a first aspect, the present application provides a method for estimating a battery state of charge, comprising:
[0007] Acquiring operating data of the battery to be estimated in a first set period of time; the operating data includes battery terminal voltage, operating current and surface temperature;
[0008] According to the operating data of the battery to be estimated in a set time period, the state of charge of the battery to be estimated is determined using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model.
[0009] Optionally, determining the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set period of time by using a battery state of charge estimation model specifically includes:
[0010] Arrange the operating data of the battery to be estimated in the first set period into n×3 data blocks, where n is the length of the sliding window;
[0011] Normalizing the data block to obtain a processed data block of the battery to be estimated;
[0012] The state of charge of the battery to be estimated is determined according to the processed data block of the battery to be estimated using a battery state of charge estimation model.
[0013] Optionally, the teacher model is trained and optimized using the training data set in combination with the network structure search algorithm and knowledge distillation, including:
[0014] Build a training dataset;
[0015] Processing the training data set to obtain a processed training data set;
[0016] Constructing a teacher model;
[0017] Using the processed training data set to train the teacher model to obtain an initial model for estimating the battery state of charge;
[0018] Based on the battery state of charge estimation initial model, using a network structure search algorithm to determine a learning model structure;
[0019] Based on the learning model structure, knowledge distillation is used to obtain an optimal learning model, and the optimal learning model is used as a battery state of charge estimation model.
[0020] Optionally, processing the training data set to obtain a processed training data set specifically includes:
[0021] Arrange the operating data of the training battery in the second set period into n×3 data blocks, where n is the length of the sliding window;
[0022] Preprocessing the multiple data blocks of the training battery to obtain multiple processed data blocks of the training battery; the preprocessing includes denoising, outlier removal and normalization;
[0023] A processed training data set is obtained based on the multiple processed data blocks and corresponding state of charge labels of the training battery.
[0024] Optionally, preprocessing the multiple data blocks of the training battery to obtain the multiple processed data blocks of the training battery specifically includes:
[0025] Performing denoising processing on the multiple data blocks of the training battery to obtain denoised data blocks;
[0026] Performing outlier elimination on the denoised data block to obtain an outlier-eliminated data block;
[0027] The data blocks after the outliers are removed are normalized to obtain a plurality of processed data blocks of the training battery.
[0028] Optionally, the convolutional neural network model has a first set number of hyperparameters and a second set number of convolutional layers.
[0029] In a second aspect, the present application provides a battery state of charge estimation system, comprising:
[0030] A data acquisition module, used to acquire operating data of the battery to be estimated in a first set period of time; the operating data includes battery terminal voltage, operating current and surface temperature;
[0031] An estimation module is used to determine the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set time period and using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model.
[0032] In a third aspect, the present application provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-described methods for estimating the battery state of charge.
[0033] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for estimating the battery state of charge.
[0034] In a fifth aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned battery state of charge estimation methods.
[0035] According to the specific embodiments provided in this application, this application has the following technical effects:
[0036] The present application provides a battery state of charge estimation method, system, device, medium and product, which obtains the operating data of the battery to be estimated in the first set time period; according to the operating data of the battery to be estimated in the set time period, the battery state of charge estimation model is used to determine the charge state of the battery to be estimated; wherein the battery state of charge estimation model is obtained by training and optimizing the teacher model using a training data set, combined with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in the second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model. The present application combines the advantages of knowledge distillation and network structure search to find the optimal student model in a certain search space and use knowledge distillation to guide the student model to learn, so that a battery state of charge estimation model that meets the accuracy requirements can be obtained with high efficiency, thereby improving the efficiency and accuracy of battery state of charge estimation. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0038] Figure 1 A flowchart of a battery state of charge estimation method provided in one embodiment of the present application;
[0039] Figure 2 A flow chart of the battery state of charge estimation model construction process of the present application;
[0040] Figure 3 A flowchart of determining a learning model using a network structure search algorithm for this application;
[0041] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION
[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0043] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.
[0044] The present application proposes a battery state of charge estimation method, which combines the advantages of knowledge distillation and network structure search, finds the optimal student model in a certain search space and uses knowledge distillation to guide the learning of the student model, so that a small model with satisfactory accuracy can be obtained with high efficiency, solving the problem of limited practical application of existing battery state of charge estimation methods based on deep learning, making it suitable for deployment and application on terminal devices, and improving the practical application value of the method.
[0045] In an exemplary embodiment, Figure 1 As shown, a battery state of charge estimation method is provided, comprising the following steps:
[0046] S1: Acquire operating data of a battery to be estimated in a first set period of time; the operating data includes battery terminal voltage, operating current and surface temperature.
[0047] S2: Determine the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set time period using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge labels; and the teacher model is a convolutional neural network model.
[0048] As an optional implementation, S2 specifically includes:
[0049] The operating data of the battery to be estimated in the first set time period is sorted into n×3 data blocks, where n is the length of the sliding window.
[0050] The data block is normalized to obtain a processed data block of the battery to be estimated.
[0051] The state of charge of the battery to be estimated is determined according to the processed data block of the battery to be estimated using a battery state of charge estimation model.
[0052] As an optional implementation, Figure 2 As shown in the figure, the teacher model is trained and optimized using the training data set, combined with the network structure search algorithm and knowledge distillation, including:
[0053] Step 1: Build a training dataset.
[0054] In practical applications, the battery operation data is collected and the voltage, current, temperature and other data of the training battery in multiple cycles are obtained. Since the battery state of charge estimation model designed in this application is designed based on the convolutional neural network model, in order to make full use of the powerful feature extraction ability of the convolutional neural network model, a simple combination of measurement signals is directly used as input to avoid the complex artificial feature extraction process, and the output is the battery state of charge.
[0055] Step 2: Process the training data set to obtain a processed training data set.
[0056] In practical applications, after data preprocessing, the data is organized into samples that can be used for neural network learning.
[0057] As an optional implementation, step 2 specifically includes:
[0058] Step 21: Arrange the operating data of the training battery in the second set time period into n×3 data blocks; n is the length of the sliding window.
[0059] In practical applications, three characteristic parameters, namely battery terminal voltage, operating current and surface temperature, are used as input. A sliding window of size n is designed to slide on the time series data of each feature with a step size of 1. Subsequences in the same time period (subsequences with a length of n in the second set time period) are extracted from the three parameters to form input data blocks. If a cycle has a total of m sampling points, the cycle can extract (mn)+1 data blocks of size n×3 as model input, and each data block corresponds to an output.
[0060] For example, the bth input is:
[0061] X b =[I t , I t+1 ,…,I t+n-1 ; V t , V t+1 , …, V t+n-1 ; K t , K t+1 , …, K t+n-1 ] T .
[0062] Among them, I t is the working current at time t, I t+1is the operating current at time t+1, I t+n-1 is the operating current at time t+n-1, V t is the battery terminal voltage at time t, V t+1 is the battery terminal voltage at time t+1, V t+n-1 is the battery terminal voltage at time t+n-1, K t is the surface temperature at time t, K t+1 is the surface temperature at time t+1, K t+n-1 is the surface temperature at time t+n-1, and T is the transpose.
[0063] The bth output is: Y b =SOC t+n-1 ;SOC t+n-1 is the state of charge at time t+n-1.
[0064] Step 22: Preprocess the multiple data blocks of the training battery to obtain multiple processed data blocks of the training battery; the preprocessing includes denoising, outlier removal and normalization.
[0065] As an optional implementation, step 22 specifically includes:
[0066] De-noising is performed on the multiple data blocks of the training battery to obtain de-noised data blocks.
[0067] Outliers are removed from the denoised data block to obtain an outlier-removed data block.
[0068] The data blocks after the outliers are removed are normalized to obtain a plurality of processed data blocks of the training battery.
[0069] Step 23: Obtain a processed training data set based on the multiple processed data blocks and corresponding state of charge labels of the training battery.
[0070] Step 3: Build the teacher model.
[0071] Step 4: Use the processed training data set to train the teacher model to obtain an initial model for battery state of charge estimation.
[0072] In practical applications, in order to improve model performance and obtain higher estimation accuracy, the neural network model usually needs to have more parameters and deeper depth. Therefore, firstly, a convolutional neural network model with more parameters and layers is designed as the teacher model, and then the model is trained using the sample set obtained in the first step to obtain an initial battery state of charge estimation model with higher estimation accuracy. The teacher model is represented by TN.
[0073] The convolutional neural network model has a first set number of hyperparameters and a second set number of convolutional layers. The first set number is determined according to the number and size of convolution kernels; the second set number is determined according to the complexity of the battery state of charge estimation problem.
[0074] Step 5: Based on the battery state of charge estimation initial model, a network structure search algorithm is used to determine the learning model structure.
[0075] Step 6: Based on the learning model structure, use knowledge distillation to obtain an optimal learning model, and use the optimal learning model as a battery state of charge estimation model.
[0076] In practical applications, network structure search technology is used to determine the appropriate student model structure, and knowledge distillation technology is combined to guide student model learning.
[0077] Since different teacher models correspond to different optimal student models, it is not possible to simply design the student model by reducing the model structure. However, using network structure search technology, we can always find the most suitable student model for different teacher models to ensure the learning results. The specific steps are:
[0078] (1) First, according to the depth of the teacher model TN, it is divided into N modules, namely:
[0079] TN=TN N o…oTN k+1 oT k o…oTN1.
[0080] Among them, TN k+1 oT k Indicates that the k+1th module and the kth module in the teacher model are connected.
[0081] (2) Divide the hypernetwork into N modules, that is, modularize the search space into N modules, and use the evolutionary algorithm to search the network architecture in each module.
[0082] Step 1: For each module, randomly generate a set of initial network structures k = 1, 2, ..., N, the subscript 0 indicates that this is a network in the initial population, s indicates the network individual number in the corresponding population, s = 1, ..., M, M indicates the population size. This set of network structures is a set of candidate architectures in the corresponding module of the hypernetwork.
[0083] Step 2: In the i-th iteration, training Knowledge distillation is used during training. For the kth module of the hypernetwork, knowledge is extracted from the kth module of the teacher model TN, and the input and output feature maps of this module are used as the input and output of the corresponding module of the hypernetwork to supervise the training of the hypernetwork module. k The output of the jth neuron in the output feature map is y k [j] indicates that in the i-th iteration, the output of the j-th neuron in the k-th module of the hypernetwork candidate architecture is To represent, then in the i-th iteration, the k-th module of the super network has the s-th candidate architecture The loss function can be measured by the difference between the output of the architecture and the output of the kth module of the teacher model:
[0084]
[0085] Among them, A is the number of neurons in the output feature map of the kth module.
[0086] The super network includes all candidate architectures, each candidate architecture is a sub-network of the super network, and the candidate architecture of each module is trained separately. Minimize the parameters of each candidate architecture in each module.
[0087] Step 3: In the i-th iteration, for each module k, select from s candidate architectures Smallest The candidate architecture is propagated as the parent network structure.
[0088] Step 4: Through the crossover operation, a new network structure is generated based on the parent network structure, and each individual is divided into two groups according to a certain probability (p c =0.1) changes its own structure, and a new set of network structures is generated. i=i+1.
[0089] Step 5: Repeat steps 2 to 4 until the number of iterations reaches the set value or Stop when the requirements are met, and finally obtain multiple candidate architectures that have been trained in each module.
[0090] (3) Then, find a sub-network with the best performance in the candidate architecture of each module, concatenate the optimal sub-networks of N modules, and obtain the optimal student model with N modules.
[0091] First, according to the complexity of the problem to be solved, the depth limit d of the student model is designed. max , under the condition of satisfying the number of layers and depth constraints, according to the loss function of each candidate architecture in each module on the validation set Sort the candidate architectures, select the candidate architecture with the smallest loss function result as the optimal sub-network, and connect the optimal sub-networks selected by each module to obtain the student model structure SN. The specific process is as follows Figure 3 shown.
[0092] (4) The structure of the student model can be determined based on the SN output in the previous step. Furthermore, it is hoped that the output results of each module and the final output results of the student model can be close to the output results of each module and the final output results of the teacher model, and the final output results of the student model should be as close to the expected output value as possible. Therefore, the following loss function is designed:
[0093] L = αL1(y in ,s in )+βL2(SOC t , SOC s )+γL3(SOC,SOC s ).
[0094] Among them, L1(y in ,s in ) represents the error between the intermediate modules of the teacher model and the intermediate modules of the student model, which is used to evaluate the ability of each module of the student model to learn from each module of the teacher model. L 1，k，i Indicates the output difference between the kth module of the teacher model and the student model on the i-th input sample; L2 (SOC t , SOC s ) represents the error between the final output of the teacher model and the student model, encouraging the student model to imitate the final output of the teacher model. Its calculation formula is:
[0095]
[0096] I represents the number of samples, SOC t，i Represents the calculation output result of the teacher model for the i-th input sample, SOC s，i Represents the calculation output result of the student model for the i-th sample.
[0097] L3 (SOC, SOC s ) represents the error between the final output of the student model and the expected output, ensuring that the output of the student model is close to the true value, and SOC represents the true output, that is, the target that the student model needs to fit.
[0098]
[0099] α, β, and γ are weight factors used to adjust the influence of each part in the loss function L.
[0100] According to the above loss function, the student model parameters are optimized in combination with the validation set to obtain the optimal student model.
[0101] The trained optimal student model has a small structure and can be deployed on the terminal device for real-time online estimation of the battery state of charge.
[0102] The battery operation data is collected in real time. The computer automatically organizes and converts the collected signals into the model input structure (n×3 data blocks) according to the program, and inputs it into the trained optimal student model to directly estimate the battery state of charge at the current moment.
[0103] Using knowledge distillation technology to compress the model structure can reduce the model size while ensuring the model estimation accuracy, so that the deep learning model for battery state of charge estimation can be actually deployed in resource-constrained environments such as mobile terminals, achieving efficient calculation and prediction.
[0104] Combining network structure search technology with knowledge distillation can ensure that when designing different deep learning models for different battery state estimation, the most suitable student model can always be found to ensure learning effect and efficiency.
[0105] Based on the same inventive concept, the embodiment of the present application also provides a battery state of charge estimation system for implementing the above method. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiment of the battery state of charge estimation system provided below can refer to the limitations of the battery state of charge estimation method above, and will not be repeated here.
[0106] In an exemplary embodiment, a battery state of charge estimation system is provided, comprising:
[0107] The data acquisition module is used to acquire the operating data of the battery to be estimated in a first set time period; the operating data includes the battery terminal voltage, operating current and surface temperature.
[0108] An estimation module is used to determine the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set time period and using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model.
[0109] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the above-mentioned battery state of charge estimation method when executing the computer program.
[0110] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above-mentioned battery state of charge estimation method is implemented.
[0111] In an exemplary embodiment, a computer program product is provided, including a computer program, which implements the above-mentioned battery state of charge estimation method when executed by a processor.
[0112] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a battery state of charge estimation method is implemented.
[0113] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0114] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.
[0115] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).
[0116] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.
[0117] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0118] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.
Claims
1. A method for estimating a battery state of charge, characterized in that: include: Acquiring operating data of the battery to be estimated in a first set period of time; the operating data includes battery terminal voltage, operating current and surface temperature; According to the operating data of the battery to be estimated in a set time period, the state of charge of the battery to be estimated is determined using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model.
2. The battery state of charge estimation method according to claim 1, characterized in that: Determining the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set period of time using a battery state of charge estimation model specifically includes: Arrange the operating data of the battery to be estimated in the first set period into n×3 data blocks, where n is the length of the sliding window; Normalizing the data block to obtain a processed data block of the battery to be estimated; The state of charge of the battery to be estimated is determined according to the processed data block of the battery to be estimated using a battery state of charge estimation model.
3. The battery state of charge estimation method according to claim 1, characterized in that: Using the training data set, combined with the network structure search algorithm and knowledge distillation, the teacher model is trained and optimized, including: Build a training dataset; Processing the training data set to obtain a processed training data set; Constructing a teacher model; Using the processed training data set to train the teacher model to obtain an initial model for estimating the battery state of charge; Based on the battery state of charge estimation initial model, using a network structure search algorithm to determine a learning model structure; Based on the learning model structure, knowledge distillation is used to obtain an optimal learning model, and the optimal learning model is used as a battery state of charge estimation model.
4. The battery state of charge estimation method according to claim 3, characterized in that: Processing the training data set to obtain a processed training data set specifically includes: Arrange the operating data of the training battery in the second set period into n×3 data blocks, where n is the length of the sliding window; Preprocessing the multiple data blocks of the training battery to obtain multiple processed data blocks of the training battery; the preprocessing includes denoising, outlier removal and normalization; A processed training data set is obtained based on the multiple processed data blocks and corresponding state of charge labels of the training battery.
5. The battery state of charge estimation method according to claim 4, characterized in that: Preprocessing the multiple data blocks of the training battery to obtain multiple processed data blocks of the training battery specifically includes: Performing denoising processing on the multiple data blocks of the training battery to obtain denoised data blocks; Performing outlier elimination on the denoised data block to obtain an outlier-eliminated data block; The data blocks after the outliers are removed are normalized to obtain a plurality of processed data blocks of the training battery.
6. The battery state of charge estimation method according to claim 1, characterized in that: The convolutional neural network model has a first set number of hyperparameters and a second set number of convolutional layers.
7. A battery state of charge estimation system, characterized in that: include: A data acquisition module, used to acquire operating data of the battery to be estimated in a first set period of time; the operating data includes battery terminal voltage, operating current and surface temperature; An estimation module is used to determine the state of charge of the battery to be estimated according to the operating data of the battery to be estimated in a set time period and using a battery state of charge estimation model; wherein the battery state of charge estimation model is obtained by training and optimizing a teacher model using a training data set in combination with a network structure search algorithm and knowledge distillation; the training data set includes the operating data of the training battery in a second set time period and the corresponding state of charge label; the teacher model is a convolutional neural network model.
8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the battery state of charge estimation method according to any one of claims 1 to 6.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the battery state of charge estimation method according to any one of claims 1 to 6 is implemented.
10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the battery state of charge estimation method according to any one of claims 1 to 6 is implemented.