Method for training battery state prediction model and related device

Abstract

The application discloses a battery state prediction model training method and related device in the field of artificial intelligence. In the technical scheme provided by the application, the first model is self-supervised trained according to the first battery characteristic information of the target battery, the first model comprising a first pre-training representation model and a first prediction model; the battery state prediction model is supervised trained according to the second battery characteristic information and the first label information of the target battery, the battery state prediction model comprising a feature vector extraction model and a second prediction model, and the initial parameters of the feature vector extraction model before supervised training comprising the parameters of the first pre-training representation model obtained after self-supervised training of the first model. The technical scheme of the application can predict the battery state of the real vehicle using real vehicle data, and reduces the training cost of the battery state prediction model.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to training methods and related devices for battery state prediction models. Background Technology

[0002] Artificial intelligence (AI) is the theory, methods, technology, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. Research in the field of AI includes robotics, natural language processing, computer vision, decision-making and reasoning, human-computer interaction, recommendation and search, and fundamental AI theories.

[0003] Electric vehicles (EVs) have become a major direction in modern automotive development due to their advantages of low pollution and high performance. Meanwhile, battery state prediction for EVs has also become a hot topic of interest in the industry.

[0004] Currently, the models used to predict various battery states (such as SOC information, SOH information, fault information, or remaining life information) are end-to-end prediction models, and they are all trained from zero initial state, so the requirements for the amount of labeled data are extremely high. Summary of the Invention

[0005] This application provides a training method and related apparatus for a battery state prediction model, which enables the prediction of the battery state of an EV using real vehicle data and reduces the training cost of the battery state prediction model.

[0006] In a first aspect, this application provides a training method for a battery state prediction model. The training method includes: acquiring first battery feature information of a target battery within a first time period, wherein the battery feature information of the target battery includes one or more of the following: current information of the target battery, voltage information of the target battery, temperature information of the target battery, and state of charge (SOC) of the target battery; performing self-supervised training on a first model based on the first battery feature information, wherein the first model includes a first pre-trained representation model and a first prediction model, wherein the input of the first prediction model includes the output of the first pre-trained representation model, the first pre-trained representation model is used to determine the representation vector of the input battery feature information, and the first prediction model is used to determine the target battery feature information corresponding to the input representation vector; acquiring second battery feature information and first label information of the target battery, wherein the first label information is used to indicate the battery state information corresponding to the second battery feature information; and performing supervised training on the battery state prediction model based on the second battery feature information and the first label information, wherein the battery state prediction model includes a feature vector extraction model and a second prediction model, wherein the input of the second prediction model includes the output of the feature vector extraction model, and the initial parameters of the feature vector extraction model include the parameters of the first pre-trained representation model obtained after performing the self-supervised training on the first model.

[0007] In this method, because the pre-trained representation model, trained using unlabeled data, learns the correlation between battery feature information and battery state information in its parameters, the requirement for labeled data can be reduced when training the battery state prediction model, whose initialization parameters include the pre-trained representation model. In other words, a battery state prediction model can be trained using a small amount or even existing labeled data.

[0008] Optionally, the battery feature information in this method can be raw battery data features, such as raw battery data features collected by sensors. For example, battery feature information can include current information, voltage information, temperature information, and SOC information. This allows the trained battery state model to directly predict the state based on the raw battery data, that is, it can achieve battery state prediction based on real-time battery data.

[0009] In conjunction with the first aspect, in a first possible implementation, the structure of the second prediction model is the same as that of the first prediction model, and the initial parameters of the second prediction model before the supervised training include the parameters of the first prediction model obtained after the self-supervised training of the first model.

[0010] In this implementation, the structure of the second prediction model is the same as that of the first prediction model, and the initial parameters of the second prediction model include the parameters of the first prediction model after self-supervised training. The first prediction model, after self-supervised training, has learned the correlation between the representation vector of battery feature information and the battery feature information itself. This is equivalent to pre-training the second prediction model before training it, thus improving the training efficiency of the battery state prediction model.

[0011] In conjunction with the first aspect or the first possible implementation, in the second possible implementation, the first battery feature information includes N battery feature information of the target battery at N times within the first time period, where N is a positive integer. The first pre-trained representation model includes a first encoder and a second encoder. The input of the first encoder includes the battery feature information, and the input of the second encoder includes the output of the first encoder.

[0012] In a third possible implementation, in conjunction with the second possible implementation, the step of performing self-supervised training of the first model based on the first battery feature information includes: The N battery feature information is divided into n subsets, each subset containing at least one battery feature information. The time corresponding to any battery feature information in the i-th subset of the n subsets within the first time period is earlier than the time corresponding to any battery feature information in the (i+1)-th subset of the n subsets within the first time period. n is a positive integer less than or equal to N, and i is a positive integer less than n. The battery feature information sequence obtained by arranging all battery feature information in each subset of the n subsets in chronological order is input into the first encoder to obtain a representation vector corresponding to each subset. The battery feature information sequence obtained by arranging the n representation vectors corresponding to the n subsets in chronological order or in ascending order of SOC value is input into the second encoder to obtain a first representation vector of the first battery feature information. The first representation vector is input into the first prediction model to obtain the target battery feature information corresponding to the first representation vector. The parameters of the first model are adjusted based on the target battery feature information and the first battery feature information.

[0013] In this implementation, N battery feature information points within a first time period are divided into n subsets. The battery feature information sequence obtained by arranging all battery feature information in each subset according to their corresponding chronological order is input into a first encoder to obtain a representation vector corresponding to each subset. The representation vectors corresponding to each subset are then input into a second encoder in chronological order or in ascending order of SOC value to obtain a first representation vector for the first battery feature information. The first representation vector is then input into a first prediction model to obtain the target battery feature information corresponding to the first representation vector. The parameters of the first model are trained based on the target battery feature information and the first battery feature information, thereby improving the accuracy of the first model and reducing training costs.

[0014] In conjunction with the third possible implementation, in the fourth possible implementation, dividing the N battery feature information into n subsets includes: dividing the battery feature values ​​into the n subsets according to the SOC value, wherein all battery feature information contained in each of the n subsets has the same SOC value.

[0015] In a fifth possible implementation, which combines the second, third, or fourth possible implementations, the target battery is a battery pack, and the first battery feature information and / or the second battery feature information also include the voltage information of each individual cell in the battery pack.

[0016] In conjunction with the first aspect, in the sixth possible implementation, the first battery feature information includes M battery feature information of the target battery at M time points within the first time period, where M is a positive integer; the first pre-trained representation model includes a variational encoder; the first prediction model includes a variational decoder; and the second prediction model includes a regression model or a classification model.

[0017] In the seventh possible implementation, in conjunction with the sixth possible implementation, the step of performing self-supervised training of the first model based on the first battery feature information includes: dividing the M battery feature information into m subsets, each of the m subsets containing at least one battery feature information, wherein the time corresponding to any battery feature information in the j-th subset of the m subsets in the first time interval is earlier than the time corresponding to any battery feature information in the (j+1)-th subset of the m subsets in the first time interval, where m is a positive integer and m is less than or equal to M, and j is a positive integer and j is less than m; determining the representation vector corresponding to each subset based on all battery feature information in each subset of the m subsets; calculating the average value of the m representation vectors corresponding to the m subsets; inputting the average value into the first pre-trained representation model to obtain a second representation vector of the first battery feature information; inputting the second representation vector into the first prediction model to obtain the target battery feature information corresponding to the second representation vector; and adjusting the parameters of the first model based on the target battery feature information and the first battery feature information.

[0018] In this implementation, the M battery feature information within the first time period are divided into m subsets. A representation vector is determined for each subset based on all battery feature information within each subset. The average value of the m representation vectors corresponding to the m subsets is calculated. This average value is input into a first pre-trained representation model to obtain a second representation vector for the first battery feature information. The second representation vector is input into a first prediction model to obtain the target battery feature information corresponding to the second representation vector. The parameters of the first model are trained based on the target battery feature information and the first battery feature information, thereby improving the accuracy of the first model and reducing training costs.

[0019] In the eighth possible implementation, which combines the seventh possible implementation, the number of battery feature information in any one of the m subsets is equal to the number of battery feature information in any other subset of the m subsets.

[0020] In a ninth possible implementation, which combines the sixth, seventh, or eighth possible implementations, the target battery is a single cell, and the first battery characteristic information and / or the second battery characteristic information includes the insulation resistance information of the single cell and / or the electrochemical impedance spectroscopy (EIS) of the single cell.

[0021] In combination with the first aspect or any of the above possible implementations, in the tenth possible implementation, the target battery feature information includes: the SOC sequence of the target battery in the first time period, the voltage sequence of the target battery in the first time period, the charging duration sequence of each individual cell in the target battery in the first time period, the charging mode of the target battery or the SOC value of the target battery at a random mask in the first time period, wherein the charging mode includes a fast charging mode or a slow charging mode.

[0022] In combination with the first aspect or any of the above possible implementations, in the eleventh possible implementation, the battery status information includes battery health indicators (SOH), fault information, or remaining lifespan information.

[0023] Secondly, this application provides a method for predicting battery state, the method comprising: acquiring battery feature information of a battery to be predicted; and using a battery state prediction model to determine the battery state information of the battery to be predicted based on the battery feature information, wherein the battery state prediction model is a battery state prediction model trained using the training method described in the first aspect or any of the possible implementations thereof.

[0024] In this method, the battery feature information of the battery to be predicted can be directly obtained from the data reported by the EV, so that the EV can predict the battery status based on the real vehicle data.

[0025] Thirdly, this application provides a training apparatus for a battery state prediction model. The apparatus may include various modules for implementing the method in the first aspect, and these modules may be implemented by software and / or hardware.

[0026] Fourthly, this application provides a battery state prediction device, which may include various modules for implementing the method in the second aspect, and these modules may be implemented by software and / or hardware.

[0027] Fifthly, this application provides a training apparatus for a battery state prediction model. The apparatus may include a processor coupled to a memory. The memory stores program code, and the processor executes the program code in the memory to implement the method in the first aspect or any of its implementations.

[0028] Optionally, the device may also include the memory.

[0029] In a sixth aspect, this application provides a battery state prediction apparatus. The apparatus may include a processor coupled to a memory. The memory stores program code, and the processor executes the program code in the memory to implement the method of the second aspect.

[0030] Optionally, the device may also include the memory.

[0031] In a seventh aspect, this application provides a chip including at least one processor and a communication interface, the communication interface and the at least one processor being interconnected via a line, the at least one processor being configured to run a computer program or instructions to perform the method as described in the first aspect or the second aspect or any possible implementation thereof.

[0032] Eighthly, this application provides a computer-readable medium storing program code for execution by a device, the program code including methods for performing the methods described in the first aspect or the second aspect or any possible implementation thereof.

[0033] Ninthly, this application provides a computer program product containing instructions that, when run on a computer, cause the computer to perform the method as described in the first aspect or the second aspect or any possible implementation thereof.

[0034] In a tenth aspect, this application provides a computing device including at least one processor and a communication interface, the communication interface and the at least one processor being interconnected via a line, the communication interface communicating with a target system, and the at least one processor being configured to run computer programs or instructions to perform the methods described in the first aspect or the second aspect or any possible implementation thereof.

[0035] Eleventhly, this application provides a computing system including at least one processor and a communication interface, the communication interface and the at least one processor being interconnected via a line, the communication interface communicating with a target system, and the at least one processor being configured to run computer programs or instructions to perform the methods described in the first aspect or the second aspect or any possible implementation thereof.

[0036] The battery state prediction model training method provided in this application introduces the concept of embedded representation and pre-training pipeline in deep learning. It proposes to represent the state of the battery in any detection segment as an abstract battery representation vector, making full use of the unlabeled data reported by the battery pack, and training the battery state prediction model through at least one pre-training task designed according to the electrochemical characteristics of the battery.

[0037] For the battery state prediction model after training, either transfer learning can be used to perform end-to-end data-driven tasks for all subsequent batteries, or the pre-trained representation model in the battery state prediction model can be used to directly extract the battery representation vector for subsequent battery state prediction tasks. This reduces the dependence on a large amount of manually labeled data, so that the EV's prediction of battery state only depends on the general data reported by the EV. There is no need to perform feature transformation on the general data reported by the EV, nor is there any need to modify the EV hardware, thus reducing costs.

[0038] The training method for the battery state prediction model provided in this application can be adapted to EV battery packs of different manufacturers and structures or individual cells of different material systems through transfer learning.

[0039] The pre-trained representation model in the battery state prediction model provided in this application can use a transformer encoder model. By designing a two-layer inherited encoder model, the model can make full use of the samples at each time point, handle time series of variable length, and has a small computational cost.

[0040] When the amount of real-vehicle data reported by EVs is large enough, it can cover multiple characteristics such as voltage, temperature and current. By expanding the scale of the battery state prediction model, the nonlinear relationship between input data can be fully explored, thus achieving the effect of covering complex operating conditions and complex internal structures of battery packs. Attached Figure Description

[0041] Figure 1 A schematic diagram of a system architecture provided for an embodiment of this application;

[0042] Figure 2 A schematic diagram of a chip hardware structure provided for an embodiment of this application;

[0043] Figure 3 A schematic diagram of another system architecture provided for embodiments of this application;

[0044] Figure 4 This is a flowchart illustrating a training method for a battery state prediction model according to an embodiment of this application.

[0045] Figure 5 This is a flowchart illustrating the training method of a battery pack SOH prediction model according to an embodiment of this application.

[0046] Figure 6 This is a flowchart illustrating a method for training a fault prediction model for a battery pack according to an embodiment of this application.

[0047] Figure 7 This is a flowchart illustrating the training method for a single-cell SOH prediction model according to an embodiment of this application.

[0048] Figure 8 This is a flowchart illustrating a training method for a fault prediction model of a single battery cell according to an embodiment of this application.

[0049] Figure 9 This is a schematic flowchart illustrating a battery state prediction method according to an embodiment of this application.

[0050] Figure 10 This is a schematic structural diagram of a training device for a battery state prediction model according to an embodiment of this application;

[0051] Figure 11 This is a schematic structural diagram of a battery state prediction device according to an embodiment of this application;

[0052] Figure 12 This is a schematic diagram of the structure of an apparatus provided in one embodiment of this application. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0054] Figure 1 This is a schematic diagram of a system architecture provided for an embodiment of this application. See also... Figure 1 The data acquisition device 160 is used to collect battery feature information of the target battery and battery status label information corresponding to some battery features and store them in the database 130. Among them, battery feature information without corresponding battery status label information is called first battery feature information, battery feature information with corresponding battery status label information is called second battery feature information, and the battery status label information corresponding to the second battery feature information is called first label information. The training device 120 generates a battery status prediction model 101 based on the first battery feature information, second battery feature information and first label information maintained in the database 130. The battery status model can also be called a battery status prediction rule.

[0055] The method by which training device 120 obtains battery state prediction model 101 based on first battery feature information, second battery feature information, and first label information can be found in [reference needed]. Figures 4 to 8 The embodiment shown in any of the figures is as follows.

[0056] The battery state prediction model 101 obtained from training device 120 can be applied to different systems or devices, such as execution device 110.

[0057] The execution device 110 is equipped with an I / O interface 112 for data interaction with external devices. The "user" can input the characteristic information of the battery to be predicted to the I / O interface 112 through the client device 140.

[0058] The execution device 110 can call data, code, etc. in the data storage system 150, and can also store data, instructions, etc. in the data storage system 150.

[0059] The calculation module 111 uses the battery state prediction model 101 to process the battery feature information of the battery to be predicted, thereby obtaining the battery state information of the battery to be predicted.

[0060] Finally, I / O interface 112 returns the processing result to client device 140 for use by the user.

[0061] exist Figure 1 In the illustrated scenario, the user can manually specify the battery feature information to be predicted in the execution device 110, for example, by operating through the interface provided by the I / O interface 112. Alternatively, the client device 140 can automatically input the battery feature information to be predicted into the I / O interface 112 and obtain battery status information. If the client device 140 requires user authorization to automatically input the battery feature information, the user can set the corresponding permissions in the client device 140. The user can view the battery status information output by the execution device 110 on the client device 140, which can be presented in various ways such as display, sound, or actions. The client device 140 can also act as a data acquisition terminal, storing the collected battery feature information and tag information into the database 130.

[0062] It is worth noting that Figure 1 This is merely a schematic diagram of a system architecture provided in an embodiment of this application. The positional relationships between the devices, components, modules, etc., shown in the diagram do not constitute any limitation. For example, in Figure 1 In this context, the data storage system 150 is an external memory relative to the execution device 110. In other cases, the data storage system 150 may also be placed within the execution device 110.

[0063] Figure 2 This is a schematic diagram of a chip hardware structure provided for an embodiment of this application. See also... Figure 2The neural network processing unit (NPU) is mounted as a coprocessor on the host central processing unit (Host CPU), and tasks are assigned by the Host CPU. The core of the NPU is the arithmetic circuit 20, and the controller 204 controls the arithmetic circuit 203 to retrieve data from the memory (weight memory 202 and / or input memory 201) and perform calculations.

[0064] In some implementations, the arithmetic circuit 203 internally includes multiple process engines (PEs). In some implementations, the arithmetic circuit 203 is a two-dimensional pulsating array. The arithmetic circuit 203 can also be a one-dimensional pulsating array or other electronic circuitry capable of performing mathematical operations such as multiplication or addition. In some implementations, the arithmetic circuit 203 is a general-purpose matrix processor.

[0065] For example, suppose we have an input matrix A, a weight matrix B, and an output matrix C. The arithmetic circuit retrieves the corresponding data of matrix B from the weight memory 202 and caches it in each PE of the arithmetic circuit 203. The arithmetic circuit retrieves the data of matrix A from the input memory 201 and performs matrix operations with matrix B. The partial result or the final result of the obtained matrix is ​​stored in the accumulator 208.

[0066] The vector computation unit 207 can further process the output of the arithmetic circuit, such as vector multiplication, vector addition, exponentiation, logarithmic operations, size comparisons, etc. For example, the vector computation unit 207 can be used for network computation in non-convolutional / non-FC layers of neural networks, such as pooling, batch normalization, and local response normalization.

[0067] In some implementations, vector computation unit 207 can store the processed output vector in unified memory 206. For example, vector computation unit 207 can apply a nonlinear function to the output of arithmetic circuit 203, such as a vector of accumulated values, to generate activation values. In some implementations, vector computation unit 207 can generate normalized values, merged values, or both. In some implementations, the processed output vector can be used as activation input to arithmetic circuit 203, for example, for use in subsequent layers of a neural network.

[0068] The unified memory 206 is used to store input data and output data.

[0069] The direct memory access controller (DMAC) 205 moves input data from external memory to input memory 201 and / or unified memory 206, stores weight data from external memory into weight memory 202, and stores data from unified memory 206 into external memory.

[0070] The bus interface unit (BIU) 210 is used to enable interaction between the main CPU, DMAC and instruction fetch memory 209 via the bus.

[0071] The instruction fetch buffer 209, which is connected to the controller 204, is used to store the instructions used by the controller 204.

[0072] The controller 204 is used to call instructions cached in the instruction fetch memory 209 to control the operation of the computation accelerator. The unified memory 206, input memory 201, weight memory 202, and instruction fetch memory 209 are all on-chip memories. External memory is proprietary to this NPU hardware architecture.

[0073] In some implementations, Figure 2 The chip shown can achieve Figures 4 to 8 The method shown in the figure is used to obtain a battery state prediction model. As an example, this method can be accomplished by the main CPU and NPU working together.

[0074] In some other implementations, Figure 2 The chip shown can achieve Figure 9 The method shown is used to obtain battery state information of the battery to be predicted. As an example, this method can be accomplished by the main CPU and NPU working together.

[0075] Figure 3 A schematic diagram of another system architecture provided for embodiments of this application. See also Figure 3 The computing device 310 is implemented by one or more servers, and optionally, it can work with other computing devices, such as data storage, routers and load balancers; the computing device 310 can be deployed on a physical site or distributed across multiple physical sites.

[0076] exist Figure 3 In the system shown, computing device 310 can use data from data storage system 350 or call program code from data storage system 150 to implement [the system]. Figures 4 to 8The method shown in any figure. As an example, computing device 310 can be... Figure 1 Training equipment 120.

[0077] Or, in Figure 3 In the system shown, computing device 310 can use data from data storage system 350 or call program code from data storage system 150 to implement [the system]. Figure 9 The method shown.

[0078] Users can interact with computing device 310 by operating their respective user devices (such as local device 301 and local device 302). Each local device can represent any computing device, such as a personal computer, computer workstation, smartphone, tablet, smart camera, smart car or other type of cellular phone, media consumption device, wearable device, set-top box, game console, etc.

[0079] Each user's local device can interact with the computing device 310 through a communication network of any communication mechanism or communication standard. The communication network can be a wide area network, a local area network, a point-to-point connection, or any combination thereof.

[0080] In another implementation, one or more aspects of computing device 310 may be implemented by each local device. For example, local device 301 may provide local data or feedback calculation results to computing device 310.

[0081] It should be noted that all the functions of computing device 310 can also be implemented by local devices. For example, local device 301 implements the functions of computing device 310 and provides services to its own users, or provides services to users of local device 302.

[0082] Figure 4 This is a flowchart illustrating a training method for a battery state prediction model according to an embodiment of this application, as shown below. Figure 4 As shown, the method includes at least S401 to S404.

[0083] S401, Obtain the first battery characteristic information of the target battery in the first time period.

[0084] The target battery can be a battery pack containing multiple individual cells, or it can be a single individual cell. The first time period can be any time period, and the duration of the first time period can be predetermined.

[0085] It is understood that the first battery feature information of the target battery within the first time period may include the battery feature information of the target battery at each time point in at least one time point within the first time period.

[0086] As an example, when the target battery is a battery pack, the first battery characteristic information may include the battery pack's current information, battery pack's voltage information, battery pack's temperature information, the voltage information of each individual cell in the battery pack, and the battery pack's state of charge (SOC) index, etc.

[0087] As another example, when the target battery is a single cell, the first battery characteristic information may include the voltage information, temperature information, current information, SOC information, insulation resistance information, or electrochemical impedance spectroscopy (EIS) of the single cell.

[0088] In some implementations, the first battery characteristic information of the target battery can be obtained directly from the data reported by the original battery management system (BMS) of the EV to which the target battery belongs.

[0089] S402, the first model is self-supervised trained based on the first battery feature information. The first model includes a first pre-trained representation model and a first prediction model. The input of the first prediction model includes the output of the first pre-trained representation model. The first pre-trained representation model is used to determine the representation vector of the input battery feature information, and the first prediction model is used to determine the target battery feature information corresponding to the input representation vector.

[0090] In some implementations, the first pre-trained representation model includes a first encoder and a second encoder. The input of the first encoder includes first battery feature information, and the input of the second encoder includes the output of the first encoder. The first prediction model includes a regressor or a classifier, etc.

[0091] In some other implementations, the first pre-trained representation model includes a variational auto-encoder (VAE), and the first prediction model includes a variational decoder.

[0092] When the first pre-trained representation model includes a two-level encoder, for example, the first encoder and the second encoder in the first pre-trained representation model can be a multi-layer inherited transformer structure.

[0093] In some implementations of this embodiment, the self-supervised training may include at least one of the following training tasks: predicting the SOC sequence of the target battery in the first time period, i.e., the target battery feature information is the SOC sequence of the target battery in the first time period; predicting the voltage sequence of the target battery in the first time period, i.e., the target battery feature information is the voltage sequence of the target battery in the first time period; predicting the charging time of each SOC segment of the target battery in the first time period, i.e., the target battery feature information is the charging time of each SOC segment of the target battery in the first time period; predicting the charging mode of the target battery in the first time period, i.e., the target battery feature information is whether the charging mode of the target battery in the first time period is fast charging mode or slow charging mode; or, predicting the SOC value at the random mask of the target battery in the first time period, i.e., the target battery feature information is the SOC value at the random mask of the target battery in the first time period.

[0094] In this embodiment, one implementation method for self-supervised training of the first model based on the first battery feature information may include steps 1 to 5. In this implementation method, the first battery feature information includes N battery feature information of the target battery at N time points within a first time period, where N is a positive integer.

[0095] Step 1: Divide the N battery feature information within the first time period into n subsets. Each subset contains at least one battery feature information. The time corresponding to any battery feature information in the i-th subset of the n subsets in the first time period is earlier than the time corresponding to any battery feature information in the (i+1)-th subset of the n subsets in the first time period. n is a positive integer and n is less than or equal to N, and i is a positive integer and i is less than n.

[0096] As an example, the N battery feature values ​​can be divided into n subsets based on the SOC values ​​of the N battery feature information within the first time period. Each of the n subsets contains all battery feature information with the same SOC value.

[0097] As another example, the N battery feature values ​​can be divided into n subsets based on the voltage values ​​of the N battery feature information within the first time period. Each subset contains all battery feature information with the same voltage value.

[0098] Step 2: Input the battery feature information sequence obtained by arranging all battery feature information in each of the n subsets in the corresponding time sequence into the first encoder to obtain the representation vector corresponding to each subset.

[0099] Step 3: The battery feature information sequence obtained by arranging the n representation vectors corresponding to the n subsets in chronological order or in order of SOC value from low to high is input into the second encoder to obtain the first representation vector of the first battery feature information.

[0100] Step 4: Input the first representation vector into the first prediction model to obtain the target battery feature information corresponding to the first representation vector.

[0101] Step 5: Train the parameters of the first model based on the target battery feature information and the first battery feature information.

[0102] In this embodiment, another way to perform self-supervised training of the first model based on the first battery feature information may include steps 6 to 11. The first battery feature information may include M battery feature information of the target battery at M time points within a first time period, where M is a positive integer.

[0103] Step 6: Divide the M battery feature information within the first time period into m subsets. Each of the m subsets contains at least one battery feature information. The time corresponding to any battery feature information in the j-th subset of the m subsets in the first time period is earlier than the time corresponding to any battery feature information in the (j+1)-th subset of the m subsets in the first time period. m is a positive integer and m is less than or equal to M, and j is a positive integer and j is less than m.

[0104] As an example, the number of battery feature information in each of the m subsets can be preset, and the number of battery feature information in any one of the m subsets is equal to the number of battery feature information in any other one of the m subsets. The M battery feature information in the first time period is divided according to the preset number of battery feature information in each subset. If the number of battery feature information in any one of the m subsets does not reach the preset number of battery feature information in each subset, the missing battery feature information is recorded as 0.

[0105] Step 7: Determine the representation vector corresponding to each subset based on all battery feature information in each of the m subsets.

[0106] Step 8: Calculate the average value of the m representation vectors corresponding to the m subsets.

[0107] Step 9: Input the average value into the first pre-trained representation model to obtain the second representation vector of the first battery feature information.

[0108] Step 10: Input the second representation vector into the first prediction model to obtain the target battery feature information corresponding to the second representation vector.

[0109] Step 11: Train the parameters of the first model based on the target battery feature information and the first battery feature information.

[0110] S403, acquire the second battery feature information and the first tag information, the first tag information is used to indicate the battery status information corresponding to the second battery feature information.

[0111] As an example, the target battery can be a battery pack, and the second battery characteristic information can include the battery pack's current information, battery pack voltage information, battery pack temperature information, voltage information of each individual cell in the battery pack, and the battery pack's SOC value, etc.

[0112] As another example, the target battery can be a single cell, and the second battery characteristic information can include the voltage information, temperature information, current information, SOC information, insulation resistance information, or EIS of the single cell.

[0113] In this embodiment, the second battery feature information and the first battery feature information can be the same battery feature information or different battery feature information.

[0114] In some implementations of this embodiment, the battery state information corresponding to the second battery feature information may include the target battery's SOH information, fault information, or remaining lifespan information, etc.

[0115] In this embodiment, the second battery feature information and the first tag information together can be referred to as training data.

[0116] S404, Supervised training of the battery state prediction model is performed based on the second battery feature information and the first label information. The battery state prediction model includes a feature vector extraction model and a second prediction model. The input of the second prediction model includes the output of the feature vector extraction model. The initial parameters of the feature vector extraction model before supervised training include the parameters of the first pre-trained representation model obtained after self-supervised training of the first model.

[0117] In some implementations, the model parameters of the first pre-trained representation model obtained after self-supervised training in the first model can be used to initialize all or part of the parameters in the feature vector extraction model in the battery state prediction model. Then, the second battery feature information and the first label information are used to further train the feature vector extraction model and the battery state prediction model to obtain the battery state prediction model.

[0118] In some other possible implementations, the first pre-trained representation model obtained after self-supervised training of the first model can be used as the feature vector extraction model in the battery state prediction model. The second battery feature information is input into the feature vector extraction model to obtain the representation vector of the second battery feature information. The second prediction model is trained using the second battery feature information, the representation vector of the second battery feature information and the first label information to obtain the battery state prediction model.

[0119] In this embodiment, when the battery state information includes SOH information, the battery state prediction model can be called the SOH prediction model; when the battery state information includes fault information, the battery state prediction model can be called the fault prediction model.

[0120] In the technical solution proposed in this application, the first battery feature information for self-supervised training of the first model and the second battery feature information for supervised training of the second model can both be directly obtained from the data reported by EV. The amount of first label information used for supervised training of the second model is greatly reduced compared with the amount of label data used for training the battery state prediction model in the prior art, thereby reducing the training cost of the battery state prediction model.

[0121] Because the pre-trained representation model of the battery state prediction model is trained using only real vehicle data during the training process, when using the battery state prediction model to predict the battery state of an EV, the battery state of the EV can be predicted using only real vehicle data.

[0122] The following example, taking the target battery as the battery pack, the first pre-trained representation model including the SOC encoder and the SOC sequence encoder, the first prediction model as a classifier or regressor, and the battery state prediction model as the SOH prediction model, introduces an exemplary training method for the battery state prediction model of this application.

[0123] Figure 5 This is a flowchart illustrating the training method of a battery pack SOH prediction model according to an embodiment of this application. Figure 5 As shown, the method includes at least S501 to S508.

[0124] S501, acquire all first battery feature information of the battery pack within the first detection time.

[0125] Optionally, the first battery characteristic information of the battery pack during the first detection period may include the current information of the battery pack, the voltage information of the battery pack, the temperature information of the battery pack, the voltage information of each individual cell in the battery pack, and the SOC information of the battery pack.

[0126] The first battery characteristic information of the battery pack can be obtained based on the data reported by the EV's original BMS.

[0127] S502, the first detection duration is divided into K time segments according to the SOC value of the battery pack. The SOC value of the first battery feature information in each time segment is the same, and K is a positive integer.

[0128] The first detection period is divided into K time segments based on the SOC value of the battery pack within the first detection period. Each time segment corresponds to a SOC value. The SOC value of the first battery feature information in each time segment is the same. Each time segment includes at least one first battery feature information. The time segment can also be called an SOC frame.

[0129] The time corresponding to any first battery feature information in the xth time segment of the K time segments within the first detection duration is earlier than the time corresponding to any first battery feature information in the (x+1)th time segment of the K time segments within the first detection duration, where x is a positive integer and x is less than K.

[0130] S503: Input the first battery feature information of each of the K time segments into the SOC encoder to obtain the representation vector corresponding to that time segment.

[0131] Each of the K time segments corresponds to a representation vector, which is used to characterize the first battery feature information of the corresponding time segment.

[0132] The first detection duration has K time segments corresponding to K representation vectors.

[0133] S504, the representation vectors corresponding to each of the K time segments are sequentially input into the SOC sequence encoder in chronological order to obtain the representation vector corresponding to the first detection duration.

[0134] The SOC sequence encoder obtains the representation vector corresponding to the first detection duration based on the representation vectors of K time segments in the first detection duration. This representation vector corresponding to the first detection duration is used to characterize the battery feature information of the battery pack within the first detection duration.

[0135] S505, input the representation vector corresponding to the first detection duration into the regressor or classifier to obtain the predicted output result of the battery pack within the first detection duration.

[0136] S506, self-supervised training of a first pre-trained representation model and a first prediction model is performed using a pre-training task, wherein the first pre-trained representation model includes a SOC encoder and a SOC sequence encoder, and the first prediction model includes a regressor or a classifier.

[0137] Optionally, various pre-training tasks can be designed for self-supervised training. The designed pre-training tasks may include predicting the SOC sequence of the battery pack during the first detection period, predicting the voltage sequence of the battery pack during the first detection period, predicting the charging time of each SOC segment of the battery pack during the first detection period, predicting the charging mode of the battery pack during the first detection period, and predicting the SOC value of the battery pack at the random mask during the first detection period.

[0138] Understandably, when performing self-supervised training based on multiple pre-training tasks, different pre-training tasks can correspond to the same pre-trained representation model and different first prediction models. Alternatively, if multiple pre-training tasks are designed, a corresponding first prediction model can be designed for each pre-training task, and then self-supervised training can be performed sequentially on the same pre-training task and its corresponding first prediction model for each pre-training task.

[0139] For example, when designing 5 pre-training tasks, a corresponding first prediction model can be designed for each pre-training task, resulting in 5 first prediction models. Then, the model consisting of the pre-trained representation model and the first prediction model corresponding to the first pre-training task is trained using the first pre-training task among these 5 pre-training tasks. Next, the model consisting of the trained pre-trained representation model and the first prediction model corresponding to the second pre-training task is self-supervised using the second pre-training task. This process continues until the model consisting of the pre-trained representation model obtained from the fourth self-supervised training and the first prediction model corresponding to the last training task is self-supervised.

[0140] S507, Obtain the second battery feature information of the battery pack and the SOH tag information corresponding to the second battery feature information.

[0141] Optionally, the second battery characteristic information of the battery pack may include the battery pack's current information, battery pack's voltage information, battery pack's temperature information, the voltage information of each individual cell in the battery pack, and the battery pack's SOC information, etc.

[0142] S508, supervised training of the battery state prediction model is performed based on the second battery feature information of the battery pack and the SOH label information corresponding to the second battery feature information. The battery state prediction model includes a feature vector extraction model and a second prediction model.

[0143] In some possible implementations, the parameters of the first pre-trained representation model obtained after self-supervised training are used to initialize the parameters of the feature vector extraction model in the battery state prediction model. Then, the second battery feature information and the SOH label information corresponding to the second battery feature information are used to further train the feature vector extraction model and the second prediction model to obtain the battery state prediction model.

[0144] In some other possible implementations, the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model. The second battery feature information is input into the feature vector extraction model to obtain the representation vector of the second battery feature information. The second prediction model is trained using the second battery feature information, the representation vector of the second battery feature information, and the SOH label information corresponding to the second battery feature information to obtain the battery state prediction model.

[0145] according to Figure 5 The battery state prediction model obtained by the training method shown can be used to predict the SOH information of the battery pack.

[0146] In the technical solution proposed in this application, all first battery feature information of the battery pack within a first detection time period is divided into K time segments based on the SOC value of the battery pack. The battery feature information of each of the K time segments is input into the SOC encoder in the first pre-trained representation model to obtain the representation vector corresponding to each time segment. The representation vectors corresponding to each of the K time segments are then sequentially input into the SOC sequence encoder in the first pre-trained representation model to obtain the representation vector corresponding to the first detection time period. The representation vector corresponding to the first detection time period is then input into the first prediction model to obtain the prediction output result of the battery pack within the first detection time period. The first pre-trained representation model and the first prediction model are self-supervised using a pre-training task. The battery state prediction model is supervisedly trained based on the second battery feature information of the battery pack and the SOH label information corresponding to the second battery feature information, thus obtaining the SOH prediction model of the battery pack. The first and second battery feature information used in the training process of the SOH prediction model of the battery pack can be directly obtained from the data reported by the EV, reducing the training cost. Furthermore, the EV can predict the SOH information of the battery pack based on real vehicle data.

[0147] The following example, taking the target battery as the battery pack, the first pre-trained representation model including the SOC encoder and the SOC sequence encoder, the first prediction model as a classifier or regressor, and the battery state prediction model as a fault prediction model, introduces an exemplary training method for the battery state prediction model of this application.

[0148] Figure 6This is a flowchart illustrating a method for training a fault prediction model for a battery pack according to an embodiment of this application. Figure 6 As shown, the method includes at least S601 to S608.

[0149] S601, Obtain all first battery feature information of the battery pack within the first detection time.

[0150] S602, the first detection duration is divided into L time segments according to the SOC value of the battery pack. The SOC value of the first battery feature information in each time segment is the same, and L is a positive integer.

[0151] S603, input the first battery feature information of each of the L time segments into the SOC encoder to obtain the representation vector corresponding to that time segment.

[0152] S604, the representation vector corresponding to each time segment is input into the SOC sequence encoder in chronological order to obtain the representation vector corresponding to the first detection duration.

[0153] S605, input the representation vector corresponding to the first detection duration into the regressor or classifier to obtain the pre-training output result.

[0154] S606, using a pre-training task to perform self-supervised training on a first pre-trained representation model and a first prediction model, the first pre-trained representation model including a SOC encoder and a SOC sequence encoder, and the first prediction model including a regressor or a classifier.

[0155] It should be noted that S601 to S606 can be referred to as S501 to S506, and will not be repeated here.

[0156] S607, obtain the second battery characteristic information of the battery pack and the fault tag information corresponding to the second battery characteristic information.

[0157] Optionally, the second battery characteristic information of the battery pack may include the battery pack's current information, battery pack's voltage information, battery pack's temperature information, the voltage information of each individual cell in the battery pack, and the battery pack's SOC information, etc.

[0158] S608, supervised training of the battery state prediction model is performed based on the second battery feature information and the fault label information corresponding to the second battery feature information of the battery pack. The battery state prediction model includes a feature vector extraction model and a second prediction model.

[0159] In some possible implementations, the parameters of the first pre-trained representation model obtained after self-supervised training are used to initialize the parameters of the feature vector extraction model in the battery state prediction model. Then, the second battery feature information and the fault label information corresponding to the second battery feature information are used to further train the feature vector extraction model and the second prediction model to obtain the battery state prediction model.

[0160] In some other possible implementations, the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model. The second battery feature information is input into the feature vector extraction model to obtain the representation vector of the second battery feature information. The second prediction model is trained using the second battery feature information, the representation vector of the second battery feature information, and the fault label information corresponding to the second battery feature information to obtain the battery state prediction model.

[0161] according to Figure 6 The battery state prediction model obtained by the training method shown can be used to predict battery pack fault information.

[0162] In the technical solution proposed in this application, all first battery feature information of the battery pack within a first detection time period is divided into L time segments based on the SOC value of the battery pack. The battery feature information of each of the L time segments is input into the SOC encoder in the first pre-trained representation model to obtain the representation vector corresponding to each time segment. The representation vectors corresponding to each of the L time segments are then sequentially input into the SOC sequence encoder in the first pre-trained representation model in chronological order to obtain the representation vector corresponding to the first detection time period. The representation vector corresponding to the first detection time period is then input into the first prediction model to obtain the prediction output result of the battery pack within the first detection time period. The first pre-trained representation model and the first prediction model are self-supervised using a pre-training task. The battery state prediction model is supervisedly trained based on the second battery feature information and the fault label information corresponding to the second battery feature information, resulting in the fault prediction model of the battery pack. The first and second battery feature information used in the training process of the fault prediction model of the battery pack can be directly obtained from the data reported by the EV, reducing the training cost. Furthermore, the EV can predict the fault information of the battery pack based on real vehicle data.

[0163] The following example uses the target battery as a single cell, the first pre-trained representation model including a variational encoder, the first prediction model as a variational decoder, and the battery state prediction model as a SOH prediction model to introduce an exemplary training method for the battery state prediction model of this application.

[0164] Figure 7This is a flowchart illustrating the training method for a single-cell SOH prediction model according to an embodiment of this application. Figure 7 As shown, the method includes at least S701 to S708.

[0165] S701, acquire all first battery feature information of a single cell within the first detection time.

[0166] Optionally, the first battery characteristic information of a single cell within the first detection period may include the voltage information, temperature information, current information, SOC information, insulation resistance information, or EIS information of the single cell, etc.

[0167] The first battery characteristic information of the battery cell can be obtained based on the data reported by the original BMS of the EV.

[0168] S702, the first detection duration is divided into Q time window frames, and all the first battery feature information in each time window frame constitutes the representation vector corresponding to that time window frame, where Q is a positive integer.

[0169] In some possible implementations, the number of first battery feature information in each of the Q time window frames can be preset, and the number of first battery feature information in any one of the Q time window frames is equal to the number of first battery feature information in any other one of the Q time window frames. The first battery feature information within the first detection duration is divided according to the preset number of first battery feature information in each sub-time window frame. If the number of first battery feature information in any one of the Q time window frames does not reach the preset number of first battery feature information in each time window frame, the missing first battery feature information is recorded as 0, and each time window frame contains at least one first battery feature information.

[0170] The time corresponding to any first battery feature information in the q-th time window frame of the Q time window frames within the first detection duration is earlier than the time corresponding to any first battery feature information in the (q+1)-th time window frame of the Q time window frames within the first detection duration, where q is a positive integer less than Q.

[0171] S703, calculate the average of the Q representation vectors corresponding to the Q time window frames.

[0172] Each of the Q time window frames corresponds to a representation vector, which is used to characterize the individual cell feature information of the corresponding time window frame.

[0173] S704, the average value of the calculated Q representation vectors is input into the variational encoder to obtain the representation vector corresponding to the first detection duration.

[0174] The variational encoder obtains the representation vector corresponding to the first detection duration based on the average of Q representation vectors. This representation vector corresponding to the first detection duration is used to characterize the feature information of a single battery cell within the first detection duration.

[0175] S705 inputs the representation vector corresponding to the first detection duration into the variational decoder to obtain the prediction output result of a single cell within the first detection duration.

[0176] S706, a pre-training task is used to perform self-supervised training on a first pre-trained representation model and a first prediction model, wherein the first pre-trained representation model includes a variational encoder and the first prediction model includes a variational decoder.

[0177] This step can be referred to in S506, and will not be repeated here.

[0178] S707, obtain the second battery feature information of the single cell and the SOH tag information corresponding to the second battery feature information.

[0179] Optionally, the second battery characteristic information of a single cell during the first detection period may include the voltage information, temperature information, current information, SOC information, insulation resistance information, or EIS information of the single cell.

[0180] S708, supervised training of the battery state prediction model is performed based on the second battery feature information of the single cell and the SOH label information corresponding to the second battery feature information. The battery state prediction model includes a feature vector extraction model and a second prediction model.

[0181] In some possible implementations, the parameters of the first pre-trained representation model obtained after self-supervised training are used to initialize the parameters of the feature vector extraction model in the battery state prediction model. Then, the second battery feature information and the SOH label information corresponding to the second battery feature information are used to further train the feature vector extraction model and the second prediction model to obtain the battery state prediction model.

[0182] In some other possible implementations, the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model. The second battery feature information is input into the feature vector extraction model to obtain the representation vector of the second battery feature information. The second prediction model is trained using the second battery feature information, the representation vector of the second battery feature information, and the SOH label information corresponding to the second battery feature information to obtain the battery state prediction model.

[0183] according to Figure 7 The battery state prediction model obtained by the training method shown can be used to predict the SOH information of a single battery cell.

[0184] In the technical solution proposed in this application, all first battery feature information of a single battery cell within a first detection time period is divided into Q time window frames. All first battery feature information within each time window frame forms a representation vector corresponding to that time window frame. The average value of the Q representation vectors corresponding to the Q time window frames is calculated, and this average value is input into a first pre-trained representation model to obtain the representation vector corresponding to the first detection time period. The representation vector corresponding to the first detection time period is then input into a first prediction model to obtain the predicted output result of the single battery cell within the first detection time period. A pre-training task is used to perform self-supervised training on the first pre-trained representation model and the first prediction model. Based on the second battery feature information of the single battery cell and the SOH label information corresponding to the second battery feature information, a supervised training is performed on the battery state prediction model to obtain the SOH prediction model for the single battery cell. The first and second battery feature information used in training the SOH prediction model for the single battery cell can be directly obtained from the data reported by the EV, reducing training costs. Furthermore, the EV can predict the SOH information of the single battery cell based on real-vehicle data.

[0185] The following example, taking the target battery as a single cell, the first pre-trained representation model including a variational encoder, the first prediction model being a variational decoder, and the battery state prediction model being a fault prediction model, introduces an exemplary training method for the battery state prediction model of this application.

[0186] Figure 8 This is a flowchart illustrating a training method for a fault prediction model of a single battery cell according to an embodiment of this application. Figure 8 As shown, the method includes at least S801 to S808.

[0187] S801, acquire all first battery feature information of a single cell within the first detection time.

[0188] S802, the first detection duration is divided into W time window frames, and all the first battery feature information in each time window frame constitutes the representation vector corresponding to that time window frame.

[0189] S803 calculates the average value of the W representation vectors corresponding to the W time window frames.

[0190] S804 inputs the average value of the calculated W representation vectors into the variational encoder to obtain the representation vector corresponding to the first detection duration.

[0191] S805 inputs the representation vector corresponding to the first detection duration into the variational decoder to obtain the predicted target battery feature information.

[0192] S806, using a pre-training task to perform self-supervised training on a first pre-trained representation model and a first prediction model, the first pre-trained representation model including a variational encoder and the first prediction model including a variational decoder.

[0193] It should be noted that S801 to S806 can be referred to as S701 to S706, and will not be repeated here.

[0194] S807, obtain the second battery characteristic information of the single cell and the fault tag information corresponding to the second battery characteristic information.

[0195] Optionally, the second battery characteristic information of a single cell during the first detection period may include the voltage information, temperature information, current information, SOC information, insulation resistance information, or EIS information of the single cell.

[0196] S808, based on the second battery feature information of the individual battery cell and the fault label information corresponding to the second battery feature information, performs supervised training on the battery state prediction model. The battery state prediction model includes a feature vector extraction model and a second prediction model.

[0197] In some possible implementations, the parameters of the first pre-trained representation model obtained after self-supervised training are used to initialize the parameters of the feature vector extraction model in the battery state prediction model. Then, the second battery feature information and the fault label information corresponding to the second battery feature information are used to further train the feature vector extraction model and the second prediction model to obtain the battery state prediction model.

[0198] In some other possible implementations, the first pre-trained representation model obtained after self-supervised training is used as the feature vector extraction model in the battery state prediction model. The second battery feature information is input into the feature vector extraction model to obtain the representation vector of the second battery feature information. The second prediction model is trained using the second battery feature information, the representation vector of the second battery feature information, and the fault label information corresponding to the second battery feature information to obtain the battery state prediction model.

[0199] according to Figure 8 The battery state prediction model obtained by the training method shown can be used to predict fault information of individual battery cells.

[0200] In the technical solution proposed in this application, all first battery feature information of a single battery cell within a first detection period is divided into W time window frames. All first battery feature information within each time window frame forms a representation vector corresponding to that time window frame. The average value of the W representation vectors corresponding to the W time window frames is calculated, and this average value is input into a first pre-trained representation model to obtain the representation vector corresponding to the first detection period. The representation vector corresponding to the first detection period is then input into a first prediction model to obtain the predicted output result of the single battery cell within the first detection period. A pre-training task is used to perform self-supervised training on the first pre-trained representation model and the first prediction model. Supervised training is performed on the battery state prediction model based on the second battery feature information of the single battery cell and the fault label information corresponding to the second battery feature information, resulting in a fault prediction model for the single battery cell. The first and second battery feature information used in training the fault prediction model for the single battery cell can be directly obtained from the data reported by the EV, reducing training costs. Furthermore, the EV can predict the fault information of the single battery cell based on real-vehicle data.

[0201] Figure 9 This is a flowchart illustrating a battery state prediction method according to an embodiment of this application, as shown below. Figure 9 As shown, the method includes at least S901 to S902.

[0202] S901, Obtain the characteristic information of the battery to be predicted.

[0203] The battery to be predicted can be a battery pack containing multiple individual cells, or it can be a single individual cell.

[0204] As an example, when the battery to be predicted is a battery pack, the characteristic information of the battery to be predicted may include the current information of the battery pack, the voltage information of the battery pack, the temperature information of the battery pack, the voltage information of each individual cell in the battery pack, and the SOC information of the battery pack, etc.

[0205] As another example, when the battery to be predicted is a single cell, the characteristic information of the battery to be predicted may include the voltage information, temperature information, current information, SOC information, insulation resistance information, or EIS information of the single cell, etc.

[0206] In some implementations, the characteristic information of the battery to be predicted can be obtained directly from the data reported by the original BMS of the EV to which the battery to be predicted belongs.

[0207] S902 uses a battery state prediction model to determine the battery state information of the battery to be predicted based on the characteristic information of the battery to be predicted.

[0208] By inputting the characteristic information of the battery to be predicted into the battery state prediction model, the battery state information of the battery to be predicted can be obtained. The state information of the battery to be predicted may include the SOH information and / or fault information and / or remaining life information of the battery to be predicted.

[0209] Understandably, battery state prediction models can be based on... Figures 4 to 8 The training method shown is used to obtain the results.

[0210] For example, when the battery to be predicted is a battery pack, use Figure 5 The battery state prediction model obtained through the training method shown can be used to predict the SOH information of the battery pack; using Figure 6 The battery state prediction model obtained by the training method shown can be used to predict battery pack fault information.

[0211] For example, when the battery to be predicted is a single cell, use Figure 7 The battery state prediction model obtained through the training method shown can be used to predict the SOH information of a single battery cell; using Figure 8 The battery state prediction model obtained by the training method shown can be used to predict the fault information of individual battery cells.

[0212] In the technical solution proposed in this application, the characteristic information of the battery to be predicted is input into the system based on... Figures 4 to 8 In the battery state prediction model trained by the training method described in any of the embodiments, the battery state information of the battery to be predicted is obtained. The battery feature information of the battery to be predicted can be directly obtained from the data reported by the EV, enabling the EV to predict the battery state based on real-vehicle data.

[0213] Figure 10 This is a schematic structural diagram of a training device for a battery state prediction model according to an embodiment of this application. Figure 10 As shown, the device 1000 may include an acquisition module 1001 and a training module 1002.

[0214] In the embodiments of this application, any module in the acquisition module and the training module can be implemented entirely or partially through software and / or hardware. The software-implemented portion can run on a processor to perform the corresponding function, while the hardware-implemented portion can be a component of the processor.

[0215] In one implementation, the device 1000 can be used to implement the above. Figure 4 The method is shown. For example, the acquisition module 1001 is used to implement S401 and S403, and the training module 1002 is used to implement S402 and S404.

[0216] In another implementation, device 1000 can be used to implement the above. Figure 5 The method is shown. For example, the acquisition module 1001 is used to implement S501 and S507, and the training module 1002 is used to implement S506 and S508.

[0217] In another implementation, device 1000 can be used to implement the above. Figure 6 The method is shown. For example, the acquisition module 1001 is used to implement S601 and S607, and the training module 1002 is used to implement S606 and S608.

[0218] In another implementation, device 1000 can be used to implement the above. Figure 7 The method is shown. For example, the acquisition module 1001 is used to implement S701 and S707, and the training module 1002 is used to implement S706 and S708.

[0219] In another implementation, device 1000 can be used to implement the above. Figure 8 The method is shown. For example, the acquisition module 1001 is used to implement S801 and S807, and the training module 1002 is used to implement S806 and S808.

[0220] Figure 11 This is a schematic structural diagram of a battery state prediction device according to an embodiment of this application. Figure 11 As shown, the device 1100 may include an acquisition module 1101 and a processing module 1102.

[0221] In the embodiments of this application, any module in the acquisition module and the processing module can be implemented entirely or partially through software and / or hardware. The software-implemented portion can run on a processor to perform the corresponding function, while the hardware-implemented portion can be a component of the processor.

[0222] In one implementation, device 1100 can be used to implement the above. Figure 9 The method is shown. For example, the acquisition module 1101 is used to implement S901, and the processing module 1102 is used to implement S902.

[0223] Figure 12 This is a schematic diagram of the structure of an apparatus provided in one embodiment of this application. Figure 12 The apparatus 1200 shown can be used to perform the method described in any of the foregoing embodiments.

[0224] like Figure 12As shown, the device 1200 of this embodiment includes: a memory 1201, a processor 1202, a communication interface 1203, and a bus 1204. The memory 1201, the processor 1202, and the communication interface 1203 are interconnected via the bus 1204.

[0225] The memory 1201 can be a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1201 can store programs, and when the program stored in the memory 1201 is executed by the processor 1202, the processor 1202 can use it to execute... Figures 4 to 9 The steps of the method shown.

[0226] The processor 1202 may be a general-purpose central processing unit (CPU), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, used to execute related programs to implement the battery state prediction model training method and battery state prediction method of the method embodiments of this application.

[0227] The processor 1202 can also be an integrated circuit chip with signal processing capabilities. In implementation, each step of the method in the various embodiments of this application can be accomplished by the integrated logic circuitry in the hardware of the processor 1202 or by instructions in software form.

[0228] The processor 1202 described above can also be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0229] The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory 1201. The processor 1202 reads the information in memory 1201 and, in conjunction with its hardware, completes the functions required by the various methods in the embodiments of this application. For example, it can execute... Figures 4 to 9 The various steps / functions of the illustrated embodiment.

[0230] The communication interface 1203 can use, but is not limited to, transceivers to enable communication between the device 1200 and other devices or communication networks.

[0231] Bus 1204 may include a pathway for transmitting information between various components of device 1200 (e.g., memory 1201, processor 1202, communication interface 1203).

[0232] It should be understood that the device 1200 shown in the embodiments of this application may be an electronic device, or it may be a chip configured in an electronic device.

[0233] It should be understood that the processor in the embodiments of this application can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc.

[0234] It should also be understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous DRAM (DDR SDRAM), enhanced synchronous DRAM (ESDRAM), synchronous linked DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0235] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0236] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0237] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0238] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0239] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0240] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0241] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0242] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0243] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0244] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0245] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A training method for a battery state prediction model, characterized in that, The training method includes: Obtain first battery characteristic information of the target battery within a first time period. The battery characteristic information of the target battery includes one or more of the following: current information of the target battery, voltage information of the target battery, temperature information of the target battery, and state of charge (SOC) index of the target battery. The first model is self-supervised based on the first battery feature information. The first model includes a first pre-trained representation model and a first prediction model. The input of the first prediction model includes the output of the first pre-trained representation model. The first pre-trained representation model is used to determine the representation vector of the input battery feature information, and the first prediction model is used to determine the target battery feature information corresponding to the input representation vector. The self-supervised training includes at least one of the following training tasks: predicting the SOC sequence of the target battery in a first time period; predicting the voltage sequence of the target battery in the first time period; predicting the charging time of each SOC segment of the target battery in the first time period; predicting the charging mode of the target battery in the first time period; or, predicting the SOC value of the target battery at a random mask in the first time period. The second battery feature information and the first tag information of the target battery are obtained. The first tag information is used to indicate the battery status information corresponding to the second battery feature information. The second battery feature information is obtained from the data reported by the original battery management system. The battery state prediction model is trained under supervision based on the second battery feature information and the first label information. The battery state prediction model includes a feature vector extraction model and a second prediction model. The input of the second prediction model includes the output of the feature vector extraction model. The initial parameters of the feature vector extraction model include the parameters of the first pre-trained representation model obtained after the first model is trained under supervision.

2. The training method according to claim 1, characterized in that, The structure of the second prediction model is the same as that of the first prediction model, and the initial parameters of the second prediction model before the supervised training include the parameters of the first prediction model obtained after the self-supervised training of the first model.

3. The training method according to claim 1 or 2, characterized in that, The first battery feature information includes N battery feature information of the target battery at N time points within the first time period, where N is a positive integer. The first pre-trained representation model includes a first encoder and a second encoder. The input of the first encoder includes the battery feature information, and the input of the second encoder includes the output of the first encoder.

4. The training method according to claim 3, characterized in that, The step of performing self-supervised training of the first model based on the first battery feature information includes: The N battery feature information is divided into n subsets, and each subset contains at least one battery feature information. The time corresponding to any battery feature information in the i-th subset of the n subsets in the first time period is earlier than the time corresponding to any battery feature information in the (i+1)-th subset of the n subsets in the first time period. n is a positive integer and n is less than or equal to N, and i is a positive integer and i is less than n. The battery feature information sequence obtained by arranging all battery feature information in each of the n subsets in the corresponding time sequence is input into the first encoder to obtain the representation vector corresponding to each subset. The battery feature information sequence obtained by arranging the n representation vectors corresponding to the n subsets in chronological order or in order of SOC value from low to high is input into the second encoder to obtain the first representation vector of the first battery feature information. Input the first representation vector into the first prediction model to obtain the target battery feature information corresponding to the first representation vector; The parameters of the first model are adjusted based on the target battery feature information and the first battery feature information.

5. The training method according to claim 4, characterized in that, The step of dividing the N battery feature information into n subsets includes: The battery feature values ​​are divided into n subsets according to the SOC value, and all battery feature information contained in each of the n subsets has the same SOC value.

6. The training method according to claim 4 or 5, characterized in that, The target battery is a battery pack, and the first battery feature information and / or the second battery feature information further include: the voltage information of each individual cell in the battery pack.

7. The training method according to claim 1, characterized in that, The first battery feature information includes M battery feature information of the target battery at M time points within a first time period, where M is a positive integer. The first pre-trained representation model includes a variational encoder, the first prediction model includes a variational decoder, and the second prediction model includes a regression model or a classification model.

8. The training method according to claim 7, characterized in that, The step of performing self-supervised training of the first model based on the first battery feature information includes: The M battery feature information is divided into m subsets. Each of the m subsets contains at least one battery feature information. The time corresponding to any battery feature information in the j-th subset of the m subsets in the first time period is earlier than the time corresponding to any battery feature information in the (j+1)-th subset of the m subsets in the first time period. m is a positive integer and m is less than or equal to M, and j is a positive integer and j is less than m. The representation vector corresponding to each subset is determined based on all battery feature information in each of the m subsets; Calculate the average of the m representation vectors corresponding to the m subsets; The average value is input into the first pre-trained representation model to obtain a second representation vector of the first battery feature information; The second representation vector is input into the first prediction model to obtain the target battery feature information corresponding to the second representation vector; The parameters of the first model are adjusted based on the target battery feature information and the first battery feature information.

9. The training method according to claim 8, characterized in that, The number of battery feature information in any one of the m subsets is equal to the number of battery feature information in any other subset of the m subsets.

10. The training method according to any one of claims 7 to 9, characterized in that, The target battery is a single cell, and the first battery characteristic information and / or the second battery characteristic information further include: the insulation resistance information of the single cell and / or the electrochemical impedance spectroscopy (EIS) of the single cell.

11. The training method according to any one of claims 1, 2, 4, 5, 7-9, characterized in that, The target battery feature information includes: the SOC sequence of the target battery in the first time period, the voltage sequence of the target battery in the first time period, the charging time sequence of each individual cell in the target battery in the first time period, the charging mode of the target battery or the SOC value of the target battery at a random mask in the first time period, wherein the charging mode includes a fast charging mode or a slow charging mode; the target battery feature information corresponds to the training task included in the self-supervised training.

12. The method according to any one of claims 1, 2, 4, 5, 7-9, characterized in that, The battery status information includes battery health indicators (SOH), fault information, or remaining lifespan information.

13. A method for predicting battery state, characterized in that, include: Obtain the characteristic information of the battery to be predicted; The battery state prediction model is used to determine the battery state information of the battery to be predicted based on the feature information of the battery to be predicted. The battery state prediction model is a battery state prediction model trained using the training method as described in any one of claims 1 to 12.

14. A training device for a battery state prediction model, characterized in that, It includes various functional modules for implementing the method of any one of claims 1 to 12.

15. A device for predicting battery state, characterized in that, It includes various functional modules for implementing the method of claim 13.

16. A training device for a battery state prediction model, characterized in that, include: Memory and processor; The memory is used to store program instructions; The processor is used to invoke program instructions in the memory to execute the method as described in any one of claims 1 to 12.

17. A device for predicting battery state, characterized in that, include: Memory and processor; The memory is used to store program instructions; The processor is used to invoke program instructions in the memory to execute the method as described in claim 13.

18. A chip, characterized in that, It includes at least one processor and a communication interface, the communication interface and the at least one processor being interconnected via a line, the at least one processor being configured to run a computer program or instructions to perform the method as claimed in any one of claims 1 to 12 or the method as claimed in claim 13.

19. A computer-readable medium, characterized in that, The computer-readable medium stores program code for computer execution, the program code including instructions for performing the method as claimed in any one of claims 1 to 12 or the method as claimed in claim 13.

20. A computer program product, characterized in that, The computer program product includes instructions that, when executed, cause a computer to perform the method of any one of claims 1 to 12 or the method of claim 13.

Images