Language model-based battery health state prediction method, device and product

Abstract

The invention provides a method, a device and a product for predicting the state of health of a battery based on a language model, and the method comprises the steps: obtaining the time-varying data of the electrical parameters of a target battery at an application side, carrying out the processing, and obtaining the input data according with the input characteristics of the language model, thereby achieving the prediction of the state of health of the battery. The battery health state corresponding to the target battery is obtained through the output characteristics of the target language model, and the pre-trained model backbone in the basic language model is frozen and the adjustable component is trained at the training side by taking the physical principle of the battery as a physical constraint condition, so that the target language model is obtained. By utilizing the generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of the physical constraint, the demand for training data in the model training process is reduced, the modeling process is optimized, and the modeling efficiency is improved, so that the effectiveness of the battery health state prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular to a method, device and product for predicting the health status of a battery based on a language model. Background Art

[0002] During the application of batteries, it is necessary to estimate their health status to evaluate the degree of battery degradation, so as to determine the battery's performance, service life and safety. The battery health status cannot be measured directly and needs to be estimated through raw data. A feasible data-driven health status estimation method requires data from the entire life cycle of the battery to establish a mapping relationship, but this method requires too much data and a long collection cycle, and the data collection efficiency is not high. The established model may not be applicable to the current development of battery technology. In summary, the current battery health status prediction method has a long modeling process and a large amount of data, which makes modeling difficult and the obtained model is not effective. Therefore, a battery health status prediction method based on a language model and a corresponding model training method can be designed to simplify the required data volume, optimize the modeling process, and improve the modeling efficiency, thereby improving the effectiveness of battery health status prediction. Summary of the invention

[0003] The embodiments of the present invention provide a battery health status prediction method, device and product based on a language model to solve the problem that the current battery health status prediction method has a long modeling process, a large amount of data, resulting in great difficulty in modeling, and the obtained model is not very effective.

[0004] In order to achieve the above objectives, in a first aspect, the present application provides a method for predicting a battery health state based on a language model, the method comprising:

[0005] Determine target battery data;

[0006] Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time;

[0007] Processing the target battery data to obtain a first input feature corresponding to a target language model input form;

[0008] The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone;

[0009] Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model;

[0010] Inputting the first input feature and the second input feature into the target language model to obtain a battery health state corresponding to the target battery;

[0011] And, the prompt learning includes:

[0012] Determining battery training data, and inputting the battery training data into the basic language model;

[0013] Wherein, the battery training data includes an input charging curve corresponding to each battery;

[0014] Identify physical constraints;

[0015] The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery;

[0016] Determining an optimization objective function according to the physical constraints;

[0017] generating an output charging curve according to the optimization objective function;

[0018] In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

[0019] As an optional implementation manner, the determining of the physical constraint condition includes:

[0020] According to the first-order RC equivalent circuit model, the first formula is determined;

[0021] The first formula is used to indicate the electrical behavior of the first-order RC equivalent circuit model, and the first formula includes:

[0022]

[0023] Wherein, u is the terminal voltage, I is the equivalent current, θ1(T) and θ2(T) are functions associated with temperature and used to indicate the electrical behavior of the first-order RC equivalent circuit model;

[0024] And, determining the optimization objective function according to the physical constraint condition comprises:

[0025] According to the second formula, determine the unsupervised loss function;

[0026] Wherein, the second formula includes:

[0027]

[0028] Among them, the optimization objective function is used to indicate the second-order norm based on the physical constraint condition formed by the output values ​​corresponding to each sequence on the output charging curve and the corresponding input values, and the optimization condition of minimizing the sum of the product of the first optimization coefficient λ and the second-order norm based on the first formula and the physical constraint condition.

[0029] As an optional implementation manner, generating an output charging curve according to the optimization objective function includes:

[0030] Determining a coding sequence according to the battery training data;

[0031] Covering the preset data in the coding sequence to generate target coding data;

[0032] According to the optimization objective function, the target encoding data is expanded to generate an output charging curve.

[0033] As an optional implementation, the first input feature includes a sequence word, the battery training data includes an input charging curve, and the determining the battery training data and inputting the battery training data into the basic language model includes:

[0034] Obtaining input charging curves in the target battery data set, and determining corresponding time series data according to each of the input charging curves;

[0035] Determine an input vector according to the time series data, and determine corresponding text representation sequence features through the input translation layer according to a multi-head attention mechanism;

[0036] According to the text representation sequence features, sequence words are determined, and the sequence words are input into the model backbone of the basic language model.

[0037] As an optional implementation, the method further includes:

[0038] By means of the physical constraint condition, the input charging curve of the input is constrained, and according to the constrained input charging curve and the back propagation characteristics, the component structure and component characteristics of the input translation layer are obtained through training;

[0039] The propagation path of the back-propagation feature includes: the output mapping layer to the output feature layer of the model backbone to the model backbone body to the input translation layer;

[0040] The complete output charging curve output by the feature reconstruction layer is constrained through the equivalent circuit model, and prompt words are obtained through training based on the constrained complete output charging curve. In addition, the model backbone of the basic language model is input through the prompt words and the sequence words to obtain the output features of the current model backbone, and the component structure and component features of the feature reconstruction layer and the output mapping layer are obtained through training based on the output features.

[0041] In a second aspect, the present application provides a battery health status prediction device based on a language model, the device comprising:

[0042] A determination module, used for determining target battery data;

[0043] Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time;

[0044] A processing module, used for processing the target battery data to obtain a first input feature corresponding to the input form of a target language model;

[0045] The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone;

[0046] Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model;

[0047] The processing module is further configured to input the first input feature and the second input feature into the target language model to obtain a battery health state corresponding to the target battery;

[0048] And, the specific manner in which the processing module performs prompt learning includes:

[0049] Determining battery training data, and inputting the battery training data into the basic language model;

[0050] Wherein, the battery training data includes an input charging curve corresponding to each battery;

[0051] Identify physical constraints;

[0052] The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery;

[0053] Determining an optimization objective function according to the physical constraints;

[0054] generating an output charging curve according to the optimization objective function;

[0055] In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

[0056] In a third aspect, the present application provides an electronic device, characterized in that the electronic device includes a memory and a processor, the memory stores at least one program, and the at least one program is executed by the processor to implement the method described in the first aspect.

[0057] In a fourth aspect, the present application provides a computer-readable storage medium, wherein at least one program is stored in the storage medium, and the at least one program is executed by a processor to implement the method described in the first aspect.

[0058] In a fifth aspect, the present application provides a computer program product, comprising a computer program, characterized in that when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0059] The battery health status prediction method based on the language model provided in the present application obtains data on the change of electrical parameters of the target battery over time on the application side, and processes the data in the form of input features that conform to the language model as the input data of the target language model, inputs the input data into the target language model obtained by learning and training through physical constraint conditions, and obtains the battery health status corresponding to the target battery through the output features of the target language model. On the training side, the physical principles of the battery can be used as physical constraints to freeze the pre-trained model backbone in the basic language model and train the adjustable components to obtain the target language model. Thus, the generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process, and the corresponding output features are obtained through the language model of physical constraints, and then a complete prediction curve or battery health status is generated through the output features. Thus, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 A flowchart of a method for predicting battery health status based on a language model provided in an embodiment of the present application;

[0061] Figure 2 A schematic diagram of the structure of a target language model provided in an embodiment of the present application;

[0062] Figure 3 A schematic diagram of the structure of an equivalent circuit model provided in an embodiment of the present application;

[0063] Figure 4 A schematic diagram of the effect of a battery health status prediction method based on a language model provided in an embodiment of the present application. DETAILED DESCRIPTION

[0064] An important direction of battery management systems is the state of health (SOH) estimation, which cannot be measured directly but can only be estimated to assess the degree of battery degradation. Accurate estimation of SOH is essential for safe and efficient energy management and the development of new generation batteries. The rapid development of lithium-ion batteries (LIBs) has led to their widespread application in various fields. Taking LIBs as an example, existing data-driven SOH estimation methods require lifetime data to establish a mapping relationship for each LIB. Due to the extremely long life cycle of modern lithium-ion batteries, to obtain lifetime training data, heavy discharge aging experiments must be carried out, which often take months to years, which greatly hinders their application in battery management. This also greatly hinders the development of battery management systems (BMS) and new battery technologies. New SOH estimation methods based on transfer learning have alleviated the burden of data collection to a certain extent; however, their assumptions on the target LIB dataset are far from perfect. For example, some methods require 25% of the target LIB dataset to be labeled, while other methods require features of unlabeled target LIB data. In this case, obtaining the target LIB dataset still requires a lot of effort.

[0065] In contrast, this application assumes that only one sample of the target LIB is accessed each time, which is more in line with reality. For example, charging a new mobile phone once a day, the BMS can only collect one unlabeled charging characteristic curve. It would be meaningful if this application could use the health information in the existing samples to update the model instead of spending a lot of time and effort to collect a large amount of data to update.

[0066] In addition to their great success in language modeling, base models have also achieved great success in areas such as computer vision, time series, audio, and graphics. Base models are usually defined as models trained on a wide range of data and can be adapted to a variety of downstream tasks. For general time series data such as battery charging data, there are two typical modes: collect enough data and train a large model from scratch, or explore the feasibility of pre-training large language models (LLMs), i.e. reprogramming the input types and / or fine-tuning the LLM.

[0067] Since the time required to collect datasets cannot keep up with the development of new batteries, it is difficult to sustain the practice of collecting a large dataset and training a large LIB model. Therefore, this application takes advantage of the strong generalization ability of LLM and proposes a target language model. This application designs a translation layer to reprogram battery charging data into a text prototype representation that is more suitable for the function of the language model to unify data of different modes and adapt LLM to battery tasks. To further enhance the generalization ability of the model, this application proposes a novel physical guided test time prompt adjustment (PGTPT) strategy to fully explore the health information in a single test by learning adaptive prompts on the fly. PGTPT optimizes prompts by guiding LLM to generate a complete battery charging curve. This strategy can also effectively alleviate the cumulative error caused by the time distribution offset caused by changes in the internal mechanism of the battery. GPT4Battery can achieve state-of-the-art accuracy on a wide range of benchmarks without the need for additional training data or annotations.

[0068] It should be noted that in the implementation examples of the present application, GPT4Battery, that is, the model architecture corresponding to the present application, should be understood as an implementation form of a target language model.

[0069] Taking LIB state estimation as an example, the motivation for LLM (Large Language Model)-driven and physics-guided test-time learning is explained below.

[0070] SOH is an indicator for assessing the degree of battery degradation. Although definitions vary, it is generally accepted that the ratio of current capacity to initial capacity has a great impact on battery management tasks, such as driving range estimation and lifetime prediction. This work also focuses on the definition of capacity as a health indicator.

[0071]

[0072] Among them, Qnominal is the nominal capacity of a brand new battery, and Qcycle is the capacity of a fully charged battery in the current cycle. However, capacity measurement requires the battery to be fully charged or discharged according to a specific procedure, which is not realistic for batteries in use. This has inspired the study of estimating SOH using partial charge curves.

[0073] In the field of battery research, there are two main approaches to estimating the state of a battery: modeling-based and data-driven. Modeling-based approaches involve building a simulation model comparable to a LIB and using parameter estimation to identify the state of the real battery. The two most studied models are the equivalent circuit model (ECM) and the pseudo two-dimensional (P2D) model. These models help to gain insight into the internal dynamics of the battery, such as lithium ion diffusion and electrochemical kinetics. Although these models provide a detailed explanation of the working principle of lithium-ion batteries, a trade-off must be made between accuracy and computational efficiency in online applications, because detailed models lead to complex partial differential equations, and solving partial differential equations requires high memory and computing power. In contrast, data-driven methods have both high accuracy and extremely short inference time. However, data-driven methods still have two disadvantages: the generalization ability is heavily dependent on data and lacks interpretability. First, unlike general time series data, obtaining battery data requires extremely expensive degradation experiments, which often take months to years, making the collection of training data a time-consuming and resource-consuming labor. Second, since the usage time often spans months to years, the internal mechanisms of the battery often change, such as side reactions and the stability of the solid electrolyte interface. Therefore, battery data is likely to involve changes in time distribution, making purely data-driven models less accurate in the middle and late stages of the battery life cycle.

[0074] Specifically, to overcome the limitations of pure data-driven methods, it is necessary to build a model with high generality and interpretability. Recent studies have evaluated the excellent performance of LLM (language pre-training) on ​​zero-point time series tasks, which inspired this application to use the generalization ability of LLM to save tedious data collection. In terms of interpretability, physical-informed machine learning (PIML) technology has studied several integrated architectures based on physics and machine learning. These methods combine physical models, such as Newman's pseudo-two-dimensional (P2D), with basic neural networks as an unsupervised loss or use physical models to generate training data for machine learning models. In order to better combine the universality of LLM and the interpretability of physical-based models, the goal of this application is to embed the knowledge and rules in the physics-based model into LLM in the form of prompts. At the same time, few previous works have attached importance to the inherent health information in the partial charging curve of the test sample, which reflects the latest and most realistic battery status. If guided by the correct physical rules, this sample can be a good supplement to the shortage of data.

[0075] Therefore, this application designs a physical guided test time prompt adjustment (PGTPT) algorithm to learn adaptive prompts using a single test sample. The prompts are optimized by guiding the LLM to generate a complete battery charging curve, thereby enhancing the generalization ability of the model and making up for the lack of interpretability of pure data-driven models.

[0076] This application proposes a target language model, which may also be referred to as GPT4Battery in subsequent implementations. Compared with the prior art in this field, this model is a basic model for cross-battery SOH estimation, which reduces the burden of months to years of degradation experiments in data collection. This application uses a single test sample to instantly learn adaptive prompts, thereby enhancing the generalization and adaptability of the model in real-world scenarios. This application has conducted extensive experiments on five widely recognized LIB datasets. GPT4Battery has achieved state-of-the-art accuracy in cross-battery SOH estimation, which is comparable to domain adaptation or fine-tuning methods that require additional training data. It can be specifically illustrated through implementation methods.

[0077] The technical concept of the present application is to obtain data on the changes of electrical parameters of the target battery over time on the application side, and process the data in the form of input features that conform to the language model as the input data of the target language model, input the input data into the target language model obtained by learning and training through physical constraints, and obtain the battery health status corresponding to the target battery through the output features of the target language model. On the training side, the physical principles of the battery can be used as physical constraints to freeze the pre-trained model backbone in the basic language model and train the adjustable components to obtain the target language model. Thus, the generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process, and the corresponding output features are obtained through the language model of physical constraints, and then a complete prediction curve or battery health status is generated through the output features. Thus, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction.

[0078] To further illustrate the various embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, which are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these contents, a person of ordinary skill in the art should be able to understand other possible implementations and advantages of the present invention. The components in the figures are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0079] The present invention will now be further described with reference to the accompanying drawings and specific implementation methods.

[0080] Embodiment 1:

[0081] See also Figure 1 , Figure 1 A flow chart of a method for predicting battery health status based on a language model provided in an embodiment of the present application. Figure 1 As shown, the present application provides a method for predicting battery health status based on a language model, the method comprising:

[0082] Determine target battery data;

[0083] Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time;

[0084] Processing the target battery data to obtain a first input feature corresponding to a target language model input form;

[0085] The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone;

[0086] Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model;

[0087] The first input feature and the second input feature are input into the target language model to obtain a battery health state corresponding to the target battery.

[0088] See also Figure 2 , Figure 2 A schematic diagram of the structure of a target language model provided in an embodiment of the present application is shown in FIG. Figure 2 As shown, as an optional implementation, the adjustable component includes an input translation layer, an output mapping layer, and a feature reconstruction layer;

[0089] Among them, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model.

[0090] The model corresponding to the method of the present application includes an input translation layer, an output mapping layer, a feature reconstruction layer, and a pre-trained frozen LLM structure, i.e., four main parts of the model backbone, and the training process only adjusts the three adjustable components without optimizing the specific features of the pre-trained model backbone. During the training process, all three adjustable components are jointly trained with the prompter. During the test, the prompt is only optimized by guiding the LLM to reconstruct the complete battery charging curve. The details can be explained in conjunction with the relevant descriptions of each implementation method.

[0091] By designing three adjustable components, namely the input translation layer, the output mapping layer and the feature reconstruction layer, in actual training, the features corresponding to the pre-trained model backbone are no longer adjusted, but each adjustable component is trained to obtain more accurate output features to improve the accuracy of health status prediction. The generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process, and the corresponding output features are obtained through the language model of physical constraints, and then a complete prediction curve or battery health status is generated through the output features. Therefore, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction.

[0092] And, the prompt learning includes:

[0093] Determining battery training data, and inputting the battery training data into the basic language model;

[0094] Wherein, the battery training data includes an input charging curve corresponding to each battery;

[0095] Identify physical constraints;

[0096] The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery;

[0097] Determining an optimization objective function according to the physical constraints;

[0098] generating an output charging curve according to the optimization objective function;

[0099] In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

[0100] In summary, corresponding strategies are designed on the training side and the application side. By obtaining data on the changes of electrical parameters of the target battery over time on the application side, and processing the data in the form of input features that conform to the language model as the input data of the target language model, the input data is input into the target language model obtained by learning and training through physical constraints, and the battery health status corresponding to the target battery is obtained through the output features of the target language model. On the training side, the physical principles of the battery can be used as physical constraints to freeze the pre-trained model backbone in the basic language model and train the adjustable components to obtain the target language model. In this way, the generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process, and the corresponding output features are obtained through the language model with physical constraints, and then a complete prediction curve or battery health status is generated through the output features. Therefore, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction.

[0101] As an optional implementation, the battery training data includes an input charging curve, and the determining the battery training data and inputting the battery training data into the basic language model includes:

[0102] Obtaining input charging curves in the target battery data set, and determining corresponding time series data according to each of the input charging curves;

[0103] According to the time series data, determining corresponding text representation sequence features through the input translation layer;

[0104] According to the text representation sequence features, sequence words are determined, and the sequence words are input into the model backbone of the basic language model.

[0105] As an optional implementation, the first input feature includes a sequence word, the battery training data includes an input charging curve, and the determining the battery training data and inputting the battery training data into the basic language model includes:

[0106] Obtaining input charging curves in the target battery data set, and determining corresponding time series data according to each of the input charging curves;

[0107] Determine an input vector according to the time series data, and determine corresponding text representation sequence features through the input translation layer according to a multi-head attention mechanism;

[0108] According to the text representation sequence features, sequence words are determined, and the sequence words are input into the model backbone of the basic language model.

[0109] Specifically, the present application can formalize the cross-battery SOH estimation problem. Given a well-collected battery dataset, X∈R, representing a partial constant current input charging curve with a time step of T, y represents the SOH label. The present application includes a sample of the battery life cycle consisting of S tags, that is, S={(x1,y1),(x2,y2),...,(xS,yS)}. For new batteries, the present application obtains target features in sequence through daily charging and discharging, denoted by T={(x1,y1),x2,...,xT}. The goal of the present application is to estimate each target yt, where y1 can be regarded as a new battery with SOH=100%.

[0110] This application uses the input voltage sampling window ΔV as univariate time series data. During constant current charging, the LIB charge is obtained by integrating the current at each time step. This application expands the current into a d-dimensional vector and normalizes it with the nominal capacity Qnominal of the LIB to adapt it to different LIBs. Then, this application represents this series of data as X∈R^1×T. Further, continuous timestamps can be aggregated into overlapping patch markers of length Lp, so the total number of blocks of the input sequence is Where S represents the horizontal sliding step size, and T is the number of samples on the aforementioned charge and discharge curve. Given these patches, this application can obtain patch embedding XP∈R^P×D, where D is the dimension of word embedding, which is convenient for subsequent translation.

[0111] Now the present application obtains the patch embedding form of the input sequence. In order to align the modal consistency between time series data and language and activate the time series understanding ability of LLM, the present application reprograms the patch embedding into the pre-trained word embedding space of LLM. However, for the word embedding space E∈R^V×D of LLM, V is the vocabulary. In a large language model, the vocabulary may be very large. Directly using E will produce a huge and possibly dense reprogramming space. In order to improve computational efficiency and increase the difficulty of capturing relevant source tags, the present application linearly represents it as E'∈R^V'×D, where V'<<V, thereby reducing the amount of data. Then, the present application uses a multi-head cross-attention layer consisting of a query matrix Q, a keyword matrix K, and a value matrix V to align the time series patches and text prototypes. In this way, the text prototype can learn clues in the language that can represent relevant local patch information, so that each attention head can capture different relevant features.

[0112]

[0113] Thus, this application has successfully re-encoded the input battery data into sequence embeddings, effectively adapting the modalities of time series data and natural language to fully exploit the capabilities of LLM.

[0114] By regularizing and blocking the input charging curve, the corresponding time series data can be obtained. The multi-head attention mechanism of the input translation layer is combined with the curve and prompt word features to determine the corresponding text representation sequence features, so as to determine the sequence words according to the text representation sequence features, and input the sequence words into the model backbone of the basic language model to execute subsequent output and training processes. Thus, the generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process, and the corresponding output features are obtained through the language model of physical constraints, and then a complete prediction curve or battery health status is generated through the output features. Thus, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction.

[0115] See also Figure 3 , Figure 3 This is a schematic diagram of the structure of an equivalent circuit model provided in an embodiment of the present application. Figure 3 As shown, as an optional implementation, the determining of the physical constraint condition includes:

[0116] According to the first-order RC equivalent circuit model, the first formula is determined;

[0117] The first formula is used to indicate the electrical behavior of the first-order RC equivalent circuit model, and the first formula includes:

[0118]

[0119] Wherein, u is the terminal voltage, I is the equivalent current, θ1(T) and θ2(T) are functions associated with temperature and used to indicate the electrical behavior of the first-order RC equivalent circuit model;

[0120] And, determining the optimization objective function according to the physical constraint condition comprises:

[0121] According to the second formula, determine the unsupervised loss function;

[0122] Wherein, the second formula includes:

[0123]

[0124] Among them, the optimization objective function is used to indicate the second-order norm based on the physical constraint condition formed by the output values ​​corresponding to each sequence on the output charging curve and the corresponding input values, and the optimization condition of minimizing the sum of the product of the first optimization coefficient λ and the second-order norm based on the first formula and the physical constraint condition.

[0125] For details, please refer to the relevant descriptions in the following of this application. The electrical behavior of the equivalent circuit model is determined by the first formula as a constraint of the physical conditions of the battery, and the corresponding loss function is established through the second-order norm corresponding to the electrical behavior and the second-order norm related to the test data. The corresponding output charging curve can be generated according to the loss function, and the model can be optimized and the prompt words can be trained, so that the model has physical interpretability, and unsupervised learning with partial data masking can be achieved. For details, please refer to the following implementation method.

[0126] See also Figure 4 , Figure 4 A schematic diagram of the effect of a battery health status prediction method based on a language model provided in an embodiment of the present application, as an optional implementation, generating an output charging curve according to the optimization objective function, including:

[0127] Determining a coding sequence according to the battery training data;

[0128] Covering the preset data in the coding sequence to generate target coding data;

[0129] According to the optimization objective function, the target encoding data is expanded to generate an output charging curve.

[0130] Therefore, by masking part of the data and guiding the missing data to generate a complete output charging curve, and making the model physically interpretable, through unsupervised learning or partial data self-supervised learning, the amount of data required is reduced, the level of training automation is improved, and the difficulty of data acquisition is reduced.

[0131] As an optional implementation, the physical constraint condition includes an equivalent circuit model, and the prompt learning of each adjustable component of the basic language model is performed through the physical constraint condition to train and obtain the target language model, including:

[0132] By means of the equivalent circuit model, the input charging curve of the input is constrained, and according to the constrained input charging curve and the back propagation characteristics, the component structure and component characteristics of the input translation layer are obtained through training;

[0133] The propagation path of the back-propagation feature includes: the output mapping layer to the output feature layer of the model backbone to the model backbone body to the input translation layer;

[0134] The complete output charging curve output by the feature reconstruction layer is constrained through the equivalent circuit model, and prompt words are obtained through training based on the constrained complete output charging curve. The model backbone of the basic language model is input through the prompt words and the sequence words to obtain the output features of the current model backbone, and the component structure and component features of the feature reconstruction layer and the output mapping layer are obtained through training based on the output features.

[0135] Specifically, in real-world battery usage scenarios, in some feasible implementations, only one test sample is obtained each time. If the corresponding curve is used to describe the health characteristics of the battery, it will be a process of several years. The present application guides the LLM to adapt to the downstream battery SOH estimation task by re-encoding the input into a text prototype representation and a linear output projection. Specifically, according to the needs of actual application scenarios and to reduce the cumulative error caused by changes in time distribution, the present application uses a physical model-guided test time prompt learning (PGTPT) method, which optimizes the prompts of the language model by guiding the LLM to generate a complete battery charging curve. For details, please refer to the relevant implementation methods.

[0136] Compared to directly fine-tuning the model, hint learning can learn outside the pre-trained model by simply modifying the context of the model input. Therefore, this technique does not distort the pre-trained features or lead to domain-specific behaviors that would lose the out-of-distribution generalization ability of LLM.

[0137] Test-time training was originally proposed as a strategy to improve the generalization ability of models under distribution shift. This application improves the generalization ability of LLM on cross-battery tasks by optimizing hints at test time.

[0138]

[0139] F represents the model. It is important to emphasize that no additional labels or data are required except for the zero-point test sample. This means it is an unsupervised learning objective. Now, let this application first introduce the ECM model.

[0140] The equivalent circuit model (ECM) is a widely used model to describe the electrical behavior of a battery in terms of voltage, current, resistance, and capacitance. The first-order Thevenin model is considered to accurately and adequately simulate the state of a battery and is simple and computationally efficient. In this model, the electrical behavior of a battery can be represented by a voltage source with its internal ohmic resistance represented as R. The parallel branches shown in the figure are used to simulate the polarization effect of the battery. u and I represent the terminal voltage and current that can be collected when in use, respectively. According to Kirchhoff's laws, the electrical behavior of a battery can be characterized by the physical equation:

[0141] uOCV=uR+up+u

[0142]

[0143] The circuit parameters can be found in Figure 4 The relevant locations are shown.

[0144] According to recent studies, these coefficients can be written as a function of temperature, namely:

[0145]

[0146] This means that the terminal voltage u, current I and temperature should in principle follow the physical state differential equation (ODE) function of the battery. The equivalent circuit model (ECM) has good qualities, such as easy implementation and relatively accurate description of battery characteristics, which prompted this application to use it as a physical constraint.

[0147] Since the label cannot be adjusted at test time, this application selects an unsupervised loss to make timely adjustments. In battery research, the complete charge / discharge curve from the lower voltage limit to the upper voltage limit can describe the aging pattern of the lithium-ion battery (LIB), so in theory the accurate health of the battery can be determined. Inspired by this, this application hopes to guide the pre-trained language model (LLM) to generate a complete battery charging curve from a partially obscured charging curve when designing the target. This curve can be used as test time training (TTT) for self-supervised learning. Specifically, this application inputs a test sample Then re-encode into the marked representation The present application randomly masks a portion of the mark in the form of MaskrZ(i). The present application hopes to use these masked portions to generate a complete battery charging curve Where T′>T. However, without any annotation of the complete curve, it is rarely possible to generate a meaningful curve. Now the present application can use the physical laws in the equivalent circuit model (ECM) to guide the pre-trained language model (LLM) to generate a charging curve that conforms to the physical laws.

[0148] In summary, the optimization goal of this application can be written as:

[0149]

[0150] In the present application, GPT-2 can be selected as the pre-trained large language model (LLM) backbone. In the process of training the source data set, the reconstruction layer, output layer and all tag embeddings are trained at the same time. It should be noted that when updating the output linear layer, the present application uses the target lithium-ion battery (LIB) features of the first cycle (SOH=100%). When transferring to the target LIB data set, the present application also uses the features of the first cycle (SOH=100%) to update the output linear layer. During testing, the present application uses a single test sample to further optimize the prompt mark to guide the LLM to reconstruct a complete battery charging curve that conforms to physical principles, that is, the complete battery charging curve of the 1-RC ECM model.

[0151] The equivalent circuit model is used to explain the physical mechanism of the battery, so as to constrain the relevant parameters and obtain the component structure and component characteristics of each adjustable component, so as to obtain more accurate output features to improve the accuracy of health status prediction. The generalization ability of the pre-trained model backbone and the generalization ability brought by the interpretability of physical constraints are utilized to reduce the demand for training data in the model training process. The corresponding output features are obtained through the language model of physical constraints, and then the complete prediction curve or battery health status is generated through the output features. Therefore, the required data volume can be simplified, the modeling process can be optimized, and the modeling efficiency can be improved, thereby improving the effectiveness of battery health status prediction.

[0152] In some implementation scenarios, the method provided by this application can be tested with some common data sets provided by existing research. According to the cross-dataset verification results, the accuracy of the method provided by this application in cross-battery SOH estimation can be comparable to that of domain adaptation or fine-tuning methods that require additional training data. In addition, this application can also analyze the impact of each component design method on the effect of this application based on ablation experiments.

[0153] For PGTPT, the present application initializes the prompt to the default manually created prompt "Analyze series data" and optimizes the corresponding text token in the text input embedding space. The present application optimizes the prompt to minimize the PGTPT loss to 1 step, using the AdamW optimizer with a learning rate of 0.0001. After the present application obtains the output feature O~∈R^P×D, the present application can convert O~ into a one-dimensional tensor of length P×D, input it into a fully connected layer, and linearly project it downward to SOH y^, and linearly project it upward to the complete battery charging curve

[0154] As mentioned above, in the present application, the input charging curve of the input is constrained by the physical constraint condition, and the component structure and component characteristics of the input translation layer are trained based on the constrained input charging curve and the back propagation characteristics;

[0155] The propagation path of the back-propagation feature includes: the output mapping layer to the output feature layer of the model backbone to the model backbone body to the input translation layer;

[0156] The complete output charging curve output by the feature reconstruction layer is constrained through the equivalent circuit model, and prompt words are obtained through training based on the constrained complete output charging curve. In addition, the model backbone of the basic language model is input through the prompt words and the sequence words to obtain the output features of the current model backbone, and the component structure and component features of the feature reconstruction layer and the output mapping layer are obtained through training based on the output features.

[0157] Embodiment 2:

[0158] The present application provides a battery health status prediction device based on a language model, the device comprising:

[0159] A determination module, used for determining target battery data;

[0160] Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time;

[0161] A processing module, used for processing the target battery data to obtain a first input feature corresponding to the input form of a target language model;

[0162] The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone;

[0163] Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model;

[0164] The processing module is further used to input the first input feature and the second input feature into the target language model to obtain a battery health state corresponding to the target battery;

[0165] And, the processing module performs prompt learning in a manner including:

[0166] Determining battery training data, and inputting the battery training data into the basic language model;

[0167] Wherein, the battery training data includes an input charging curve corresponding to each battery;

[0168] Identify physical constraints;

[0169] The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery;

[0170] Determining an optimization objective function according to the physical constraints;

[0171] generating an output charging curve according to the optimization objective function;

[0172] In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

[0173] As an optional implementation manner, the specific manner in which the determination module determines the physical constraint condition includes:

[0174] According to the first-order RC equivalent circuit model, the first formula is determined;

[0175] The first formula is used to indicate the electrical behavior of the first-order RC equivalent circuit model, and the first formula includes:

[0176]

[0177] Wherein, u is the terminal voltage, I is the equivalent current, θ1(T) and θ2(T) are functions associated with temperature and used to indicate the electrical behavior of the first-order RC equivalent circuit model;

[0178] And, determining the optimization objective function according to the physical constraint condition comprises:

[0179] According to the second formula, determine the unsupervised loss function;

[0180] Wherein, the second formula includes:

[0181]

[0182] Among them, the optimization objective function is used to indicate the second-order norm based on the physical constraint condition formed by the output values ​​corresponding to each sequence on the output charging curve and the corresponding input values, and the optimization condition of minimizing the sum of the product of the first optimization coefficient λ and the second-order norm based on the first formula and the physical constraint condition.

[0183] As an optional implementation manner, the processing module generates a specific manner of outputting a charging curve according to the optimization objective function, including:

[0184] Determining a coding sequence according to the battery training data;

[0185] Covering the preset data in the coding sequence to generate target coding data;

[0186] According to the optimization objective function, the target encoding data is expanded to generate an output charging curve.

[0187] As an optional implementation, the first input feature includes a sequence word, the battery training data includes an input charging curve, and the processing module determines the battery training data, and a specific manner of inputting the battery training data into the basic language model includes:

[0188] Obtaining input charging curves in the target battery data set, and determining corresponding time series data according to each of the input charging curves;

[0189] Determine an input vector according to the time series data, and determine corresponding text representation sequence features through the input translation layer according to a multi-head attention mechanism;

[0190] According to the text representation sequence features, sequence words are determined, and the sequence words are input into the model backbone of the basic language model.

[0191] As an optional implementation manner, the processing module is further used for:

[0192] By means of the physical constraint condition, the input charging curve of the input is constrained, and according to the constrained input charging curve and the back propagation characteristics, the component structure and component characteristics of the input translation layer are obtained through training;

[0193] The propagation path of the back-propagation feature includes: the output mapping layer to the output feature layer of the model backbone to the model backbone body to the input translation layer;

[0194] The complete output charging curve output by the feature reconstruction layer is constrained through the equivalent circuit model, and prompt words are obtained through training based on the constrained complete output charging curve. In addition, the model backbone of the basic language model is input through the prompt words and the sequence words to obtain the output features of the current model backbone, and the component structure and component features of the feature reconstruction layer and the output mapping layer are obtained through training based on the output features.

[0195] For the various implementation methods and technical effects on the device side, please refer to the relevant description on the corresponding method side, which will not be repeated here.

[0196] Embodiment three:

[0197] The present invention also provides an electronic device, which includes a processor, a memory, a bus, and a computer program stored in the memory and executable on the processor. The processor includes one or more processing cores. The memory is connected to the processor via a bus. The memory is used to store program instructions. When the processor executes the computer program, the steps in the above-mentioned method embodiment of the first embodiment of the present invention are implemented.

[0198] Further, as an executable solution, the electronic device may be a computer unit, and the computer unit may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The computer unit may include, but is not limited to, a processor and a memory. Those skilled in the art may understand that the composition structure of the above-mentioned computer unit is only an example of a computer unit and does not constitute a limitation on the computer unit, and may include more or less components than the above-mentioned, or a combination of certain components, or different components. For example, the computer unit may also include input and output devices, network access devices, buses, etc., which are not limited in the embodiments of the present invention.

[0199] Further, as an executable solution, the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer unit, and various interfaces and lines are used to connect the various parts of the entire computer unit.

[0200] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the computer unit by running or executing the computer program and / or module stored in the memory, and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system and at least one application required for a function; the data storage area can store data created according to the use of the mobile phone, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0201] Embodiment 4:

[0202] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above method in the embodiment of the present invention are implemented.

[0203] If the module / unit integrated in the computer unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory) and software distribution medium, etc. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction.

[0204] Embodiment five:

[0205] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the steps of the method described above are implemented.

[0206] Although the present invention has been specifically shown and described in conjunction with the preferred embodiments, it should be understood by those skilled in the art that various changes may be made to the present invention in form and details without departing from the spirit and scope of the present invention as defined by the appended claims, all of which are within the scope of protection of the present invention.

Claims

1. A battery health status prediction method based on a language model, characterized in that: The method comprises: Determine target battery data; Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time; Processing the target battery data to obtain a first input feature corresponding to a target language model input form; The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone; Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model; Inputting the first input feature and the second input feature into the target language model to obtain a battery health state corresponding to the target battery; And, the prompt learning includes: Determining battery training data, and inputting the battery training data into the basic language model; Wherein, the battery training data includes an input charging curve corresponding to each battery; Identify physical constraints; The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery; Determining an optimization objective function according to the physical constraints; generating an output charging curve according to the optimization objective function; In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

2. The method according to claim 1, characterized in that The determining of the physical constraint conditions comprises: According to the first-order RC equivalent circuit model, the first formula is determined; The first formula is used to indicate the electrical behavior of the first-order RC equivalent circuit model, and the first formula includes: Wherein, u is the terminal voltage, I is the equivalent current, θ1(T) and θ2(T) are functions associated with temperature and used to indicate the electrical behavior of the first-order RC equivalent circuit model; And, determining the optimization objective function according to the physical constraint condition comprises: According to the second formula, determine the unsupervised loss function; Wherein, the second formula includes: Among them, the optimization objective function is used to indicate the second-order norm based on the physical constraint condition formed by the output values ​​corresponding to each sequence on the output charging curve and the corresponding input values, and the optimization condition of minimizing the sum of the product of the first optimization coefficient λ and the second-order norm based on the first formula and the physical constraint condition.

3. The method according to claim 2, characterized in that The step of generating an output charging curve according to the optimization objective function includes: Determining a coding sequence according to the battery training data; Covering the preset data in the coding sequence to generate target coding data; According to the optimization objective function, the target encoding data is expanded to generate an output charging curve.

4. The method according to claim 1, characterized in that: The first input feature includes a sequence word, the battery training data includes an input charging curve, and the determining the battery training data and inputting the battery training data into a basic language model include: Obtaining input charging curves in the target battery data set, and determining corresponding time series data according to each of the input charging curves; Determine an input vector according to the time series data, and determine corresponding text representation sequence features through the input translation layer according to a multi-head attention mechanism; According to the text representation sequence features, sequence words are determined, and the sequence words are input into the model backbone of the basic language model.

5. The method according to claim 1, characterized in that The method further comprises: By means of the physical constraint condition, the input charging curve of the input is constrained, and according to the constrained input charging curve and the back propagation characteristics, the component structure and component characteristics of the input translation layer are obtained through training; The propagation path of the back-propagation feature includes: the output mapping layer to the output feature layer of the model backbone to the model backbone body to the input translation layer; The complete output charging curve output by the feature reconstruction layer is constrained through the equivalent circuit model, and prompt words are obtained through training based on the constrained complete output charging curve. In addition, the model backbone of the basic language model is input through the prompt words and the sequence words to obtain the output features of the current model backbone, and the component structure and component features of the feature reconstruction layer and the output mapping layer are obtained through training based on the output features.

6. A battery health status prediction device based on a language model, characterized in that: The device comprises: A determination module, used for determining target battery data; Wherein, the target battery data includes the change of at least one electrical parameter in the target battery over time; A processing module, used for processing the target battery data to obtain a first input feature corresponding to the input form of a target language model; The target language model is obtained by training a basic language model through prompt learning, the basic language model includes a pre-trained model backbone, an adjustable component and a prompter, the pre-trained model backbone includes a GPT-2 model backbone, the adjustable component includes an input translation layer, an output mapping layer and a feature reconstruction layer, the prompter is used to generate a prompt word, and the prompt word is used to indicate a second input feature of the model backbone; Furthermore, the input translation layer is used to obtain the corresponding text representation sequence features according to the input features corresponding to the battery training data, the feature reconstruction layer is used to obtain the complete output charging curve according to the output features of the target language model, and the complete output charging curve is used to optimize the prompt words corresponding to the prompt learning process, and the output mapping layer is used to obtain the battery health status of the corresponding battery according to the output features of the target language model; The processing module is further configured to input the first input feature and the second input feature into the target language model to obtain a battery health state corresponding to the target battery; And, the specific manner in which the processing module performs prompt learning includes: Determining battery training data, and inputting the battery training data into the basic language model; Wherein, the battery training data includes an input charging curve corresponding to each battery; Identify physical constraints; The physical constraint condition is used to indicate the constraint conditions of the physical model and electrical parameters of the target battery; Determining an optimization objective function according to the physical constraints; generating an output charging curve according to the optimization objective function; In the training phase, the component structures and component characteristics of each of the adjustable components and the prompter are adjusted according to the output charging curve; and in the testing phase, the component structures and component characteristics of the prompter are adjusted according to the output charging curve.

7. The device according to claim 6, characterized in that The specific manner in which the processing module generates the output charging curve according to the optimization objective function includes: Determining a coding sequence according to the battery training data; Covering the preset data in the coding sequence to generate target coding data; According to the optimization objective function, the target encoding data is expanded to generate an output charging curve.

8. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores at least one program, and the at least one program is executed by the processor to implement the method according to any one of claims 1 to 5.

9. A computer-readable storage medium, characterized in that: The storage medium stores at least one program, and the at least one program is executed by a processor to implement the method according to any one of claims 1 to 5.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images