Battery Health State Assessment Method and System Based on Large Model and Sequence Matching

Abstract

The present invention discloses a method and system for evaluating the battery health state based on a large model and sequence matching, including: classifying the battery health state according to a numerical range; extracting a general form of charging sequence from the test battery cell, defined as the current sequence; extracting general forms of charging sequences with different median thresholds from known battery cells, defined as candidate sequences; performing sequence matching between the current sequence and each candidate sequence; fine-tuning the large language model using the matched sequences with labels; obtaining the actual charging data of the battery cell to be tested, extracting the general form of the charging sequence as the current sequence, performing sequence matching between the current sequence and each candidate sequence respectively, and using the fine-tuned large language model to obtain the evaluation category of the battery health state. The present invention uses the numerical range of the battery health state as the state evaluation result, avoiding the problem of reduced method confidence caused by outlier estimated values in traditional estimation methods.

Description

Technical Field

[0001] The present invention relates to the technical field of battery state of health assessment, and in particular to a method and system for battery state of health assessment based on large models and sequence matching. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] As the most widely used secondary battery, lithium-ion batteries play a role in electric vehicles, energy storage systems, consumer electronics, and intelligence. However, in actual applications, the combined effects of complex working conditions, environmental temperature, material inconsistencies, etc. may cause non-linear attenuation of the available capacity of lithium-ion batteries, thereby leading to a decline in the overall performance of the batteries.

[0004] The state of health (SOH) of a battery is used to evaluate the aging level of the battery to avoid potential failures and threats. In the prior art, for the assessment of the battery state of health, most methods adopt model-based estimation methods or data-driven estimation methods; model-based estimation methods mainly establish a mathematical model that can accurately simulate the complex mechanism inside the battery to calculate the battery state of health. The defect of this method is that the state value estimated based on the model completely depends on the integrity of the model, real-time data, and parameter identification, making it poor in applicability, flexibility, transferability, and estimation accuracy. Data-driven estimation methods are based on data processing, feature engineering, and deep learning / machine learning algorithms to establish a mapping relationship between aging characteristics and the battery state of health to achieve state estimation. Although this method does not require the establishment of a complex mathematical model, due to the limited number of labels of lithium-ion batteries, data-driven models are often small in scale and have poor model generalization ability.

[0005] The prior art discloses battery state estimation methods based on pre-trained large language models. The pre-trained large language models provide prior knowledge in multiple fields, which can improve the efficiency of model training / fine-tuning and the estimation accuracy. However, these methods all require manual extraction of aging features (such as the peak of the IC curve, relaxation voltage, etc.) to establish the mapping relationship with the state of health, and the prerequisite conditions for feature extraction are harsh, which leads to high general applicability and estimation complexity of the model. At the same time, these methods all use the floating-point number of the state of health as the state estimation result. However, in the case of good overall estimation performance of the model, there will be individual outlier estimation values that reduce the confidence of the method. For ordinary users, in most cases, only the range of the state of health needs to be known, rather than the specific floating-point number of the battery state of health. In addition, in the prior art, one feature corresponds to one floating-point data of the state of health, and the acquisition of one feature requires one charge and discharge cycle; due to the limited charge and discharge cycles of the battery, the number of samples obtained will also be limited. Summary of the Invention

[0006] To solve the above problems, the present invention proposes a method and system for evaluating the state of health of a battery based on a large model and sequence matching, which breaks the mindset of using the floating-point number of the state of health as the state estimation result, and realizes accurate, convenient and reliable evaluation of the state of health of the battery by performing sequence matching on the charging sequence to be measured and different candidate charging sequences through the classification theory and with the help of the fine-tuned large model.

[0007] In some embodiments, the following technical solutions are adopted:

[0008] A method for evaluating the state of health of a battery based on a large model and sequence matching, comprising:

[0009] Classify the state of health of the battery according to the numerical range, and determine the median threshold of the state of health of different categories;

[0010] Based on the historical charging data of the fragments, fuse the voltage and the charging amount to construct a charging sequence in a general form;

[0011] Divide the battery monomers in the historical charging data into test battery monomers and known battery monomers; extract the charging sequence in the general form from the test battery monomers and define it as the current sequence; extract the charging sequences in the general form with different median thresholds from the known battery monomers and define them as candidate sequences;

[0012] Splice the current sequence with each candidate sequence respectively to achieve sequence matching; assign labels to the matching sequences according to whether the current sequence and the spliced candidate sequences belong to the same state of health category;

[0013] Use the matching sequences with labels to fine-tune the large language model;

[0014] Obtain the actual charging data of the battery cell to be tested, extract the charging sequence in general form as the current sequence, perform sequence matching on the current sequence and each candidate sequence respectively, and based on the matching sequences, use the fine-tuned large language model to obtain the evaluation category of the battery health state.

[0015] As a further solution, classify the battery health state according to the numerical range, specifically:

[0016] Classify the battery health state into 5 categories from good to poor, including: 95% < SOH ≤ 100%, 90% < SOH ≤ 95%, 85% < SOH ≤ 90%, 80% < SOH ≤ 85% and SOH < 80%;

[0017] Determine the median threshold of the battery health state under each category.

[0018] As a further solution, construct a charging sequence in general form, specifically:

[0019] Based on the fragmented historical charging data, extract the original voltage sequence V and the charge quantity sequence Q :

[0020] ;

[0021] Couple the voltage sequence V and the charge quantity sequence Q in the time dimension to construct a two-dimensional array E 0:

[0022] ;

[0023] Based on the two-dimensional array E 0, reduce the array dimension by point-by-point differentiation, and then construct a charging sequence in general form E 1:

[0024] ;

[0025] where n is the total number of charging data points, d Q / d V represents point-by-point differentiation, and i = 1, 2, …, n - 1.

[0026] As a further solution, splice the current sequence and each candidate sequence respectively to achieve sequence matching, specifically:

[0027] Determine the voltage range of the current sequence, truncate the complete candidate sequence based on the voltage range, and retain the candidate sequence data within the voltage range.

[0028] As a further solution, use the matching sequences with labels to fine-tune the large language model. Specifically:

[0029] Select BERT-large as the large language model, apply the low-rank adaptation technique to the attention heads and output modules in the model, and use the matching sequences to fine-tune the large language model multiple times;

[0030] The input of the large language model is multiple matching sequences, and the output is the label corresponding to each matching sequence. If the label of the matching sequence is 1, it means that the current sequence and the candidate sequence in this matching sequence belong to the same battery health status category; finally, obtain the battery health status evaluation category of the current sequence.

[0031] As a further solution, use the matching sequences with labels to fine-tune the large language model multiple times, evaluate the large language models obtained by each fine-tuning using evaluation metrics, and select the best large language model.

[0032] As a further solution, obtain the actual charging data of the battery cell to be tested, and use the fine-tuned large language model to obtain the evaluation result of the battery health status. Specifically:

[0033] Based on the actual charging data, construct a charging sequence in general form as the current sequence;

[0034] Truncate the candidate sequences, and splice the current sequence with each candidate sequence respectively to achieve sequence matching;

[0035] Input the matched sequences into the fine-tuned large language model, output the predicted labels, and use the battery health status category corresponding to the candidate sequence in the matching sequences with label 1 as the evaluation result of the battery health status of the battery cell to be tested.

[0036] In some other embodiments, the following technical solutions are adopted:

[0037] A battery health status evaluation system based on a large model and sequence matching includes:

[0038] A category division module, which is used to divide the battery health status according to the numerical range, and determine the median threshold of the battery health status of different categories;

[0039] A charging sequence construction module, which is used to construct a charging sequence in general form by fusing voltage and charge based on the fragmented historical charging data;

[0040] A sequence matching module, which is used to divide the battery cells in the historical charging data into test battery cells and known battery cells; extract the general form of the charging sequence from the test battery cells and define it as the current sequence; extract the general form of the charging sequences with different median thresholds from the known battery cells and define them as candidate sequences; splice the current sequence with each candidate sequence respectively to achieve sequence matching; assign labels to the matching sequences according to whether the current sequence and the spliced candidate sequences belong to the same health status category.

[0041] A model fine-tuning module, which is used to fine-tune the large language model by using the matching sequences with labels.

[0042] A status evaluation module, which is used to obtain the actual charging data of the battery cell to be tested, extract the general form of the charging sequence as the current sequence, perform sequence matching on the current sequence with each candidate sequence respectively, and based on the matching sequences, use the fine-tuned large language model to obtain the evaluation category of the battery health status.

[0043] In some other embodiments, the following technical solutions are adopted:

[0044] A terminal device, which includes a processor and a memory. The processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are suitable for being loaded and executed by the processor to perform the above-mentioned battery health status evaluation method based on a large model and sequence matching.

[0045] In some other embodiments, the following technical solutions are adopted:

[0046] A computer-readable storage medium, in which multiple instructions are stored, and the instructions are suitable for being loaded and executed by the processor of the terminal device to perform the above-mentioned battery health status evaluation method based on a large model and sequence matching.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] (1) Compared with the feature engineering for aging feature extraction in the prior art, the present invention does not require manual feature extraction, but uses charging sequence matching to replace the traditional feature engineering. The input of the large model is a one-dimensional array that fuses voltage and charge, which alleviates the problems such as poor applicability of the status evaluation method caused by complex feature engineering and harsh preconditions, and does not require additional data processing and complicated calculation processes.

[0049] (2) The present invention neither depends on complete charge / discharge data, nor requires the establishment of complex mathematical models and mechanism models. It establishes the mapping relationship between the model input and output through data-driven and large model fine-tuning, and then realizes the health status evaluation.

[0050] The present invention uses a large model to replace the small deep learning / machine learning model. The large model can provide rich prior knowledge, wide applicability, and excellent parallel processing efficiency, solving the problems of low accuracy, poor generalization ability, overfitting, etc. caused by the small parameter scale of the small model.

[0051] (3) Compared with the existing state estimation method that uses the floating-point number of the health state as the state estimation result, the present invention uses the numerical range of the battery health state to replace the floating-point number as the state evaluation result, avoiding the problem of reduced method confidence caused by outlier estimation values in the traditional estimation method; it meets the actual engineering requirements, and ordinary users do not yet need precise floating-point values to judge the battery performance; and using five types of numerical ranges to replace a large number of discrete floating-point values can provide users with a more intuitive state evaluation result.

[0052] The present invention realizes the charging sequence matching by means of the classification theory, abandons the feature engineering, obtains the battery health state evaluation result by matching the sequences, without calculating the feature quantities, reducing the complexity of model processing; in addition, a current sequence can match multiple different candidate sequences to obtain multiple sample data, thereby expanding the sample quantity and greatly improving the evaluation performance on the premise of maintaining similar state evaluation functions.

[0053] Other features and advantages of the additional aspects of the present invention will be partially given in the following description, partially will become obvious from the following description, or will be understood through the practice of this aspect. Description of the Drawings

[0054] Figure 1 It is a flowchart of the battery health state evaluation method based on a large model and sequence matching in an embodiment of the present invention;

[0055] Figure 2 It is the health state decay curve of each battery cell in the public dataset in an embodiment of the present invention;

[0056] Figure 3 It is the candidate sequence curve when the voltage range is [3.7V, 4V] in an embodiment of the present invention;

[0057] Figure 4 It is the battery health state evaluation result based on sequence matching and a large model in an embodiment of the present invention;

[0058] Figure 5 It is a schematic diagram of the battery health state evaluation system based on a large model and sequence matching in an embodiment of the present invention. Detailed Embodiments

[0059] It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in the present invention have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs.

[0060] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they specify the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0061] Embodiment 1

[0062] In one or more embodiments, a method for evaluating the battery health state based on a large model and sequence matching is disclosed, combined with Figure 1 , and specifically includes the following processes:

[0063] S101: Classify the battery health state according to a numerical range.

[0064] In this embodiment, according to the battery health state from good to bad, the battery health state is divided into 5 categories, specifically including: 95% < SOH ≤ 100%, 90% < SOH ≤ 95%, 85% < SOH ≤ 90%, 80% < SOH ≤ 85% and SOH < 80%.

[0065] Then determine the median thresholds for each battery health state category, which are: 97.5%, 92.5%, 87.5%, 82.5% and 77.5% in sequence.

[0066] S102: Based on the historical charging data of the fragments, fuse the voltage and the charging amount to construct a general form of the charging sequence.

[0067] In this embodiment, the charging data of the fragments comes from a public dataset. The public dataset contains a total of 21,720 cycle charge and discharge test data of 181 nickel-cobalt-manganese ternary lithium-ion battery monomers, with a rated capacity of 2.5 Ah, a charging current rate of 3C (7.5 A), a test temperature of 25 °C, and a cycle sample quantity of 21,720, Figure 2 which shows the health state decay curves of each monomer in the public dataset.

[0068] Randomly truncate the complete charging data of each cell in the public dataset at any cycle to generate fragmented charging data. The first 91 cells are composed into a training set, and the remaining 90 cells are composed into a test set.

[0069] The specific process of constructing a charging sequence in general form is as follows:

[0070] Based on the fragmented charging data, extract the original voltage sequence V and the charge sequence Q :

[0071] ;

[0072] Among them, n is the total number of charging data points, which is related to the sampling frequency; and respectively represent the voltage and charge of the nth charging data point.

[0073] Couple the voltage sequence V and the charge sequence Q in the time dimension to construct a two-dimensional array E 0:

[0074] ;

[0075] Based on the two-dimensional array E 0, reduce the array dimension by point-by-point differentiation, and then construct a charging sequence in general form E 1:

[0076] ;

[0077] Among them, d Q / d V is point-by-point differentiation, and i = 1, 2,..., n - 1.

[0078] Both the voltage and the charge are closely related to the electrochemistry, thermodynamics, and aging mechanism of lithium-ion batteries, and are also parameters that can be recorded in ordinary battery management systems. The charging sequence obtained by coupling the two can reflect the battery characteristics to the greatest extent.

[0079] S103: Divide the battery cells of historical charging data into test battery cells and known battery cells; extract the charging sequence in general form from the test battery cells and define it as the current sequence; extract multiple charging sequences in general form with the health state at the median threshold from the known battery cells and define them as candidate sequences.

[0080] In this embodiment, the first cell in the S102 public dataset is used as the known cell, and the remaining cells are used as test battery cells.

[0081] Define the charging sequence of the charging data extracted from the test battery cell as the current sequence; extract the charging sequences when the health states of the known cells are at the median thresholds (97.5%, 92.5%, 87.5%, 82.5%, and 77.5%) and define them as candidate sequences respectively. Figure 3 The candidate sequence curves when the voltage range is [3.7V, 4V] are shown.

[0082] Further, make the candidate sequence consistent with the current sequence in terms of the voltage range by truncation. For example: the voltage range covered by the current sequence is [3.7V, 4.0V]. First, extract the complete candidate sequence, whose voltage range is approximately [2.8V, 4.2V]; second, extract the corresponding candidate sequence according to the voltage range of the current sequence, that is, truncate the candidate sequence; finally, obtain the candidate sequence with the voltage range of [3.7V, 4.0V].

[0083] In this embodiment, there is no need to obtain the battery feature vector. Instead, by constructing a one-dimensional array that fuses voltage and charge, a general form of the charging sequence is obtained. The data acquisition is simple and convenient, and the reliability is high; it alleviates the problems such as poor applicability of the state evaluation method caused by complex feature engineering and harsh preconditions.

[0084] S104: Concatenate the current sequence with each candidate sequence respectively to achieve sequence matching; assign labels to the matching sequences according to whether the current sequence and the concatenated candidate sequence belong to the same health state category.

[0085] In this embodiment, when assigning labels to sequence matching, when the current sequence and the candidate sequence belong to the same health state category, the label is 1, otherwise it is 0.

[0086] In this embodiment, by performing sequence matching between the current sequence and the candidate sequence, one current sequence is matched with five candidate sequences respectively, and five sample data can be obtained, expanding the sample quantity, improving the accuracy and reliability of model evaluation, and solving the problem of limited quantity of the labeled data set in the traditional model.

[0087] S105: Fine-tune the large language model with the matching sequences with labels.

[0088] In this embodiment, based on the nonlinear characteristics of lithium-ion batteries, BERT-large is selected as the large model. The input of the large language model is multiple matching sequences, and the output is the label of each matching sequence. If the label of the matching sequence is 1, then the current sequence and the candidate sequence in this matching sequence belong to the same health state category; finally, the battery health state evaluation category of the current sequence is obtained.

[0089] For the BERT-large model, set the linear layer with an output dimension of 1 as the output module, set the cross-entropy as the loss function, initialize the model parameters, and implement model loading.

[0090] Using the matching sequences in S104, apply the Low-Rank Adaptation (LoRA) technique to the attention heads and output modules in the model, and perform multiple fine-tuning on the large language model; by introducing low-rank matrices, reduce the amount of parameter updates, thereby reducing the computational and storage costs; performing multiple fine-tuning on the large language model can avoid the problem of inaccurate evaluation results caused by model performance fluctuations.

[0091] In this embodiment, accuracy, recall, precision, and F1 score are selected as evaluation indicators. For the large language model obtained from each fine-tuning, calculate its evaluation indicators respectively. Finally, the best large language model can be selected according to the calculation results of the evaluation indicators; as a specific example, the model corresponding to the maximum F1 score can be selected as the best large language model.

[0092] In this embodiment, a total of 20,460 positive samples (labeled 1) and 82,710 negative samples (labeled 0) are generated by assigning labels to the original dataset through sequence matching. Compared with the original 21,720 loop samples, the sample quantity is significantly expanded; BERT-large contains 24 Transformer encoder layers, with a hidden state size of 1024 and a total of 340 million parameters; during the model fine-tuning process, the learning rate is 2e-5, the batch size is 64, the number of fine-tuning rounds is 5, Dropout is 0.3, the gradient decay is 0.01, the learning rate strategy is linear decay, and the optimizer is AdamW.

[0093] S106: Obtain the actual charging data of the battery cell to be tested, extract the charging sequence in general form as the current sequence, perform sequence matching between the current sequence and each candidate sequence respectively, and based on the matching sequence, use the fine-tuned large language model to obtain the evaluation result of the battery health state.

[0094] The specific implementation process is as follows:

[0095] S1061: Based on the actual charging data, construct a charging sequence in general form as the current sequence;

[0096] S1062: Truncate the candidate sequences, and splice the current sequence with each candidate sequence respectively to achieve sequence matching;

[0097] S1063: Input the matched sequence into the fine-tuned large language model, output the predicted label, and use the battery health state category corresponding to the candidate sequence in the matched sequences with a label of 1 as the evaluation result of the battery health state of the battery cell to be tested.

[0098] The method of this embodiment uses the numerical range of the battery health status instead of the floating-point number as the status evaluation result, which can meet the requirements of most ordinary users for the battery health status evaluation results. At the same time, there is no need to calculate the battery characteristic quantity, which saves additional data processing and complicated calculation processes, and improves the evaluation efficiency.

[0099] As a specific verification example, actual charging data is selected from the test set. Figure 4 The battery health status assessment results based on sequence matching and large models are shown. It can be seen that 9937 out of 10068 positive samples were correctly identified, and 41158 out of 41232 negative samples were correctly identified. The overall accuracy rate reached 99.6%, the recall rate reached 98.7%, the precision rate reached 99.3%, and the F1 score reached 99.0%, which fully demonstrated the accuracy and reliability of the battery health status assessment results of the method in this embodiment.

[0100] Embodiment 2

[0101] In one or more embodiments, a battery health status assessment system based on large model and sequence matching is disclosed, combined with Figure 5 , specifically including:

[0102] A classification module is used to classify the battery health status into categories according to the numerical range and determine the median threshold of the battery health status of different categories;

[0103] A charging sequence building module is used to build a general form of charging sequence based on fragmented historical charging data, integrating voltage and charging amount;

[0104] The sequence matching module is used to divide the battery cells of the historical charging data into test battery cells and known battery cells; extract the general form of charging sequence from the test battery cells, which is defined as the current sequence; extract the general form of charging sequences with different median thresholds from the known battery cells, which are defined as candidate sequences; splice the current sequence with each candidate sequence to achieve sequence matching; assign a label to the matching sequence according to whether the current sequence and the spliced ​​candidate sequence belong to the same health status category;

[0105] A model fine-tuning module, which is used to fine-tune the large language model using labeled matching sequences;

[0106] The status assessment module is used to obtain the actual charging data of the battery cell to be tested, extract the general form of the charging sequence as the current sequence, perform sequence matching on the current sequence and each candidate sequence respectively, and obtain the evaluation category of the battery health status based on the matching sequence using a fine-tuned large language model.

[0107] It should be noted that the specific implementation methods of the above-mentioned modules have been described in detail in Embodiment 1, and will not be elaborated here.

[0108] Embodiment 3

[0109] In one or more embodiments, a terminal device is disclosed, which includes a processor and a memory. The processor is used to implement instructions; the memory is used to store multiple instructions, and the instructions are adapted to be loaded and executed by the processor to perform the battery health state evaluation method based on large model and sequence matching described in Embodiment 1.

[0110] It should be understood that in this embodiment, the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), off-the-shelf programmable gate arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0111] The memory may include a read-only memory and a random access memory, and provide instructions and data to the processor. A part of the memory may also include a non-volatile random access memory. For example, the memory may also store information about the device type.

[0112] In the implementation process, each step of the above method may be completed by the integrated logic circuit in the hardware of the processor or the instructions in the form of software.

[0113] Embodiment 4

[0114] In one or more embodiments, a computer-readable storage medium is disclosed, in which multiple instructions are stored, and the instructions are adapted to be loaded and executed by the processor of the terminal device to perform the battery health state evaluation method based on large model and sequence matching described in Embodiment 1.

[0115] Although the specific implementation manners of the present invention have been described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative labor are still within the protection scope of the present invention.

Claims

1. A battery health status assessment method based on large model and sequence matching, characterized in that: include: The battery health status is classified according to the numerical range, and the median threshold of the battery health status of different categories is determined; Based on the fragmented historical charging data, the voltage and charging amount are integrated to construct a general form of charging sequence; The battery cells of the historical charging data are divided into test battery cells and known battery cells; Extracting a general form of charging sequence from the test battery cell, which is defined as the current sequence; extracting a general form of charging sequence with different median thresholds from the known battery cells, which is defined as the candidate sequence; The current sequence is concatenated with each candidate sequence to achieve sequence matching; a label is assigned to the matching sequence according to whether the current sequence and the concatenated candidate sequence belong to the same health status category; Fine-tune the large language model using labeled matching sequences; The actual charging data of the battery cell to be tested is obtained, and a general charging sequence is extracted as the current sequence. The current sequence is matched with each candidate sequence respectively, and based on the matching sequence, a fine-tuned large language model is used to obtain an evaluation category of the battery health status.

2. A battery health status assessment method based on large model and sequence matching as claimed in claim 1, characterized in that: The battery health status is classified according to the numerical range, specifically: The battery health status is divided into 5 categories from good to bad, including: 95%< SOH ≤100%, 90%< SOH ≤95%, 85%< SOH ≤90%, 80%< SOH ≤85% and SOH <80%; Determine the median battery health threshold for each category.

3. A battery health status assessment method based on large model and sequence matching as claimed in claim 1, characterized in that: Construct a charging sequence of general form, specifically: Extract the original voltage sequence based on fragmented historical charging data V and charge sequence Q : ； Couple the voltage sequence described in the time dimension V and charge sequence Q , to construct a two-dimensional array E 0: ； Based on the two-dimensional array E 0, by point-by-point differentiation to reduce the array dimension, and then construct a general form of charging sequence E 1: ； in, n is the total number of charging data points, d Q / d V Represents point-by-point differentiation, i=1,2,…,n-1.

4. A battery health status assessment method based on large model and sequence matching as claimed in claim 1, characterized in that: The current sequence is concatenated with each candidate sequence to achieve sequence matching, specifically: A voltage range of the current sequence is determined, and a complete candidate sequence is truncated based on the voltage range, and candidate sequence data within the voltage range is retained.

5. The battery health status assessment method based on large model and sequence matching according to claim 1, characterized in that: Fine-tune the large language model using labeled matching sequences, specifically: Select BERT-large as the large language model, apply low-rank adaptation technology to the attention head and output module in the model, and use matching sequences to fine-tune the large language model multiple times; The input of the large language model is multiple matching sequences, and the output is a label for each matching sequence. If the label of the matching sequence is 1, the current sequence and the candidate sequence in the matching sequence belong to the same health status category; finally, the battery health status assessment category of the current sequence is obtained.

6. A battery health status assessment method based on large model and sequence matching as claimed in claim 1, characterized in that: The large language model is fine-tuned multiple times using labeled matching sequences, and the large language model obtained through each fine-tuning is evaluated using evaluation indicators to select the best large language model.

7. A battery health status assessment method based on large model and sequence matching as claimed in claim 1, characterized in that: The actual charging data of the battery cell to be tested is obtained, and the evaluation results of the battery health status are obtained by using the fine-tuned large language model, which are as follows: Based on the actual charging data, a general form of charging sequence is constructed as the current sequence; The candidate sequences are truncated and the current sequence is concatenated with each candidate sequence to achieve sequence matching; The matched sequence is input into the fine-tuned large language model, and the predicted label is output. The battery health status category corresponding to the candidate sequence in the matching sequence with label 1 is used as the battery health status assessment result of the battery cell to be tested.

8. A battery health status assessment system based on large model and sequence matching, characterized in that: include: A classification module is used to classify the battery health status into categories according to the numerical range and determine the median threshold of the battery health status of different categories; A charging sequence building module is used to build a general form of charging sequence based on fragmented historical charging data, integrating voltage and charging amount; The sequence matching module is used to divide the battery cells of the historical charging data into test battery cells and known battery cells; extract the general form of charging sequence from the test battery cells, which is defined as the current sequence; extract the general form of charging sequences with different median thresholds from the known battery cells, which are defined as candidate sequences; splice the current sequence with each candidate sequence to achieve sequence matching; assign a label to the matching sequence according to whether the current sequence and the spliced ​​candidate sequence belong to the same health status category; A model fine-tuning module, which is used to fine-tune the large language model using labeled matching sequences; The status assessment module is used to obtain the actual charging data of the battery cell to be tested, extract the general form of the charging sequence as the current sequence, perform sequence matching on the current sequence and each candidate sequence respectively, and obtain the evaluation category of the battery health status based on the matching sequence using a fine-tuned large language model.

9. A terminal device, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions, characterized in that: The instructions are suitable for being loaded by a processor and executing the battery health status assessment method based on large model and sequence matching as described in any one of claims 1-7.

10. A computer-readable storage medium storing a plurality of instructions, characterized in that: The instructions are suitable for being loaded by a processor of a terminal device and executing the battery health status assessment method based on large model and sequence matching as described in any one of claims 1-7.

Images