Lithium ion battery multi-fault grading diagnosis method and system based on large language model

Abstract

The invention discloses a lithium ion battery multi-fault grading diagnosis method and system based on a large language model, and relates to the field of lithium ion battery safety control. The disproportionation degree of a battery pack to be detected is calculated based on a battery signal; performing down-sampling on the battery signal based on the disproportionation degree to obtain a down-sampled battery signal; constructing a fault classification task prompt word based on the down-sampled battery signal, and inputting the fault classification task prompt word into a fault classification large model to obtain a fault type and a battery signal sequence corresponding to the fault; calculating a voltage variance quantile based on the battery signal sequence corresponding to the fault, and constructing a fault grading task cue word based on the voltage variance quantile; and inputting the fault grading task cue word into the fault grading large model to obtain a fault grade. With the help of the excellent nonlinear fitting capability of the large model, tiny faults are accurately captured under complex working conditions; and meanwhile, the multi-task parallel processing capability of the large model is exerted, and fault classification and diagnosis functions are realized under the same base model.

Description

Technical Field

[0001] The present invention relates to the field of safety control of lithium-ion batteries, and particularly to a multi-fault hierarchical diagnosis method and system for lithium-ion batteries based on large language models. Background Art

[0002] The statements in this section merely provide background technical information related to the present disclosure and do not necessarily constitute prior art.

[0003] Lithium-ion batteries are the core of electric vehicles and energy storage systems. Their frequent safety accidents have seriously hindered the further development of new energy systems. Fault diagnosis is an effective means to ensure battery safety. However, lithium-ion batteries are a highly complex non-linear time-varying system, and early fault characteristics are tiny and hidden. Existing fault diagnosis methods are limited by the model size and have poor non-linear fitting ability. Under complex working conditions, misdiagnosis and missed diagnosis are serious, and it is difficult to prevent and control fault risks.

[0004] Existing methods, such as a micro-short circuit fault diagnosis method based on clustering algorithms, use the IC curve under constant current charging conditions as the feature, and detect micro-short circuit faults by means of data processing methods such as principal component analysis and machine learning methods such as unsupervised clustering. However, this method is only effective under constant current charging conditions, which greatly limits the application range. Existing methods, such as a diagnosis method based on deep learning, use small neural network models such as CNN and GRU to predict the voltage curve, and diagnose faults according to the residual between the predicted voltage and the true voltage. However, limited by the model scale, the voltage prediction step of this method is limited, and it can only detect relatively obvious voltage anomalies and cannot diagnose early battery faults. Existing methods, such as a fault diagnosis method based on a mechanism model, build a lithium intercalation expansion compensation model of the battery by means of electrochemistry mechanism, evaluate the lithium plating state of the battery, and then diagnose faults. However, the generalization ability of the mechanism model is extremely poor, and it can only be used on specific batteries, which is difficult to meet the requirements of practical applications. Existing methods, such as a fault diagnosis method based on neural networks, realize the scoring of battery anomaly degree through multiple trainings on simulation data, and diagnose the batteries exceeding the threshold as faults. However, this method is only effective for individual fault types included in the training set and lacks generalization. It can be seen that existing methods are greatly limited by the model scale, and lack effectiveness, generalization, and reliability, and cannot meet the requirements of diagnosing early tiny faults. Summary of the Invention

[0005] To overcome the deficiencies of the above-mentioned prior art, the present invention provides a multi-fault hierarchical diagnosis method and system for lithium-ion batteries based on large language models, which accurately capture tiny faults under complex working conditions by virtue of the excellent non-linear fitting ability of the large language model; at the same time, give play to the multi-task parallel processing ability of the large language model to realize the functions of fault classification and grading under the same base model.

[0006] To achieve the above object, one or more embodiments of the present invention provide the following technical solutions: In a first aspect, the present invention provides a multi-fault hierarchical diagnosis method for lithium-ion batteries based on a large language model, including: Obtain battery signals within the battery pack to be tested, and calculate the disproportion degree of the battery pack to be tested based on the battery signals; Downsample the battery signals based on the disproportion degree to obtain the downsampled battery signals; Construct a fault classification task prompt word based on the downsampled battery signals, and input the fault classification task prompt word into a fault classification large model to obtain the fault type and the battery signal sequence corresponding to the fault; Calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct a fault grading task prompt word based on the voltage variance quantile; input the fault grading task prompt word into a fault grading large model to obtain the fault level.

[0007] In a further technical solution, the disproportion degree includes a voltage difference degree and a current deviation degree.

[0008] In a further technical solution, the voltage difference degree is calculated based on the characteristic voltage matrix, expressed as:

[0009] where VDD represents the voltage difference degree, represents the improved correlation coefficient, represents the th battery, represents the th battery, represents the improved multiplication operation.

[0010] In a further technical solution, the calculation formula for the current deviation degree is:

[0011] where, represents the current sequence, represents the current sequence after normalization and adding a square wave signal, represents the scaling coefficient, represents the current deviation degree.

[0012] In a further technical solution, the voltage difference degree threshold and the current deviation degree threshold are determined by the quantile method, and the sampling frequency is determined based on the disproportion degree, expressed as:

[0013] where, represents the downsampled frequency, represents the original frequency, represents the voltage difference threshold, represents the current deviation threshold.

[0014] In a further technical solution, the fault classification task prompt words include task type, sampling frequency, data length, current data, voltage data, and temperature data, and the fault grading task prompt words include task type, number of batteries, quantile, fault type, voltage difference degree, and current deviation degree.

[0015] In a further technical solution, the fault classification large model and the fault grading large model are obtained by fine-tuning the large language model through the mutually exclusive low-rank adaptation fine-tuning method.

[0016] In a second aspect, the present invention provides a multi-fault grading diagnosis system for lithium-ion batteries based on a large language model, including: A data acquisition module, which is configured to: acquire battery signals within a battery pack to be measured, and calculate the disproportion degree of the battery pack to be measured based on the battery signals; A downsampling module, which is configured to: perform downsampling on the battery signals based on the disproportion degree to obtain the downsampled battery signals; A fault classification module, which is configured to: construct fault classification task prompt words based on the downsampled battery signals, and input the fault classification task prompt words into the fault classification large model to obtain the fault type and the battery signal sequence corresponding to the fault; A fault grading module, which is configured to: calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct fault grading task prompt words based on the voltage variance quantile; input the fault grading task prompt words into the fault grading large model to obtain the fault level.

[0017] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the method for multi-fault grading diagnosis of lithium-ion batteries based on a large language model as described in the first aspect.

[0018] In a fourth aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, it implements the steps in the method for multi-fault grading diagnosis of lithium-ion batteries based on a large language model as described in the first aspect.

[0019] The above one or more technical solutions have the following beneficial effects: Large language models have a large number of parameters and complex structures, so they inherently have excellent non-linear fitting ability and multi-task parallel ability, and have good performance in many fields. However, large language models also have problems such as high computational complexity and slow computational speed, which hinder their application and practice in the field of battery fault diagnosis. To address this problem, the present invention first proposes a downsampling method for fault signals based on disproportionation degree, and invents a new feature of disproportionation degree. This feature reflects the fault probability in different time periods by evaluating the overall abnormal degree of multiple battery cells in the battery module within different time periods. Based on the disproportionation degree, the sampling frequency in the low fault probability interval is reduced, and the length of the fault signal is compressed to less than 1 / 200. Furthermore, the present invention proposes a fine-tuning method for large models for fault diagnosis tasks. Using the M-LoRA technology, the large language model is fine-tuned based on high-quality fault samples, enabling it to have both fault classification and grading functions. Finally, the present invention proposes a fault diagnosis framework based on large models, integrating three functional modules: data downsampling, fault classification, and fault grading. First, the downsampling results are used for fault classification, and then fault grading is performed based on the fault type and the original fault signal, achieving a one-stop completion of the full-chain tasks of fault classification and grading.

[0020] The present invention gives full play to the powerful non-linear fitting ability of large models to achieve accurate classification and grading diagnosis of minor faults under complex working conditions. Compared with traditional small model methods, the fault recognition accuracy is significantly improved, and the situations of misdiagnosis and missed diagnosis are significantly reduced.

[0021] The present invention utilizes the multi-task parallel processing ability of large models to achieve fault classification and fault grading on the basis of the same large model, simplifying the fault diagnosis process.

[0022] By designing a downsampling method based on disproportionation degree, the present invention realizes a significant reduction in signal length without losing fault information, reduces the computational complexity of large models, and provides the possibility for online diagnosis.

[0023] The present invention first proposes the mutually exclusive low-rank adaptation fine-tuning method (M-LoRA). By constructing two mutually orthogonal low-rank matrices, the parameter update is decoupled in the training of fault classification and fault grading tasks to reduce mutual interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The schematic diagrams forming a part of the present invention are used to provide a further understanding of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention.

[0025] Figure 1 is a flowchart of the multi-fault grading diagnosis method for lithium-ion batteries according to an embodiment of the present invention; Figure 2 is a signal curve diagram under the UDDS working condition according to an embodiment of the present invention; Figure 3 is the signal curve diagram of the embodiment of the present invention under the charging condition; Figure 4 is the voltage difference degree curve of the embodiment of the present invention under the UDDS condition; Figure 5 is the current deviation degree curve of the embodiment of the present invention under the UDDS condition; Figure 6 is the voltage difference degree curve of the embodiment of the present invention under the charging condition; Figure 7 is the current deviation degree curve of the embodiment of the present invention under the charging condition; Figure 8 is the parameter curve after sampling of the embodiment of the present invention under the UDDS condition; Figure 9 is the parameter curve after sampling of the embodiment of the present invention under the charging condition. Detailed implementation manners

[0026] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field to which the present invention belongs.

[0027] It should be noted that the terms used herein are only for describing specific implementation manners and are not intended to limit the exemplary implementation manners according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0028] Without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0029] Embodiment 1 As Figure 1 shown, this embodiment discloses a multi-fault hierarchical diagnosis method for lithium-ion batteries based on large language models, and the method includes the following steps: S1: Obtain the battery signals in the battery pack to be tested, and calculate the disproportion degree of the battery pack to be tested based on the battery signals; In this embodiment, the battery signals include voltage signals, current signals, and temperature signals. First, a new feature, namely the disproportion degree, is defined, and the disproportion degree is composed of the voltage difference degree and the current deviation degree.

[0030] The voltage difference degree evaluates the overall dissimilation degree of the voltages of multiple different batteries in the battery pack. The higher its value, the greater the voltage difference between the batteries and the higher the probability of failure. The calculation method of the voltage difference degree is as follows: First, normalize the voltage of the battery, and then add a square wave signal to the normalized battery signal, as shown in Equation (1): (1) where, represents the normalized voltage of the th battery at time , represents the original voltage of the th battery at time , represents the rated voltage range of the battery, represents the normalized battery voltage sequence, represents the total length of the battery sequence (in the time direction), represents the voltage sequence of the th battery after adding the square wave signal, represents a constant used to adjust the amplitude of the square wave signal; represents the sign function, which is 1 when , 0 when , and -1 when .

[0031] The original correlation coefficient between battery voltages is severely affected by noise. A small noise in a single battery can easily cause a significant drop in the correlation coefficient. For example, the voltage sequences of three sampling points of two batteries are 3.2V, 3.21V, 3.2V and 3.2V, 3.19V, 3.2V respectively. The difference between these two voltage sequences is extremely small, but the correlation coefficient is -1, which is very likely to cause misdiagnosis. Therefore, in this embodiment, adding a square wave to the original data (normalized battery signal) is equivalent to adding an initial bias, eliminating the effect of small noise and increasing the robustness of the algorithm.

[0032] Then, select the battery with the largest voltage range and the largest voltage variance from the battery pack as the sentinel battery, and its voltage constitutes the characteristic voltage matrix, as shown in Equation (2): (2) where, represents the sequence containing the highest voltage, represents the sequence containing the lowest voltage, represents the sequence containing the largest voltage range, represents the sequence containing the battery with the largest voltage contrast, represents the sequence containing the battery with the smallest voltage contrast, represents the characteristic voltage matrix.

[0033] Based on the characteristic voltage matrix, the voltage difference degree VDD can be calculated as shown in Equation (3): (3) where VDD represents the voltage difference degree, and its value is the maximum improved correlation coefficient between different voltage sequences in the characteristic voltage matrix; represents the improved correlation coefficient, which replaces the dot product in the traditional correlation coefficient with an improved multiplication operation; represents the th battery cell, represents the th battery cell, represents the improved multiplication operation, which strengthens the influence of elements with different signs on the result by introducing coefficients and exponential terms; represents the standard deviation of the sequence , represents the standard deviation of the sequence .

[0034] The current deviation degree characterizes the degree of association between the voltage change in the battery pack and factors other than current. The voltage change of normal batteries is mainly affected by current, while the voltage change of faulty batteries is affected by faults. Therefore, the current deviation degree can also reflect the probability of faults and is more sensitive to module-level faults. The calculation of the current deviation degree is shown in Equation (4): (4) where, represents the current sequence, represents the current sequence after normalization and adding a square wave signal, represents the scaling coefficient used to amplify the current signal; represents the current deviation degree, and its value is the ratio of the voltage range to the current range; represents the maximum value of the current, represents the minimum value of the current.

[0035] S2: Downsample the battery signal based on the disproportionation degree to obtain the downsampled battery signal; In this embodiment, the voltage difference degree threshold and the current deviation degree threshold are determined by the quantile method. The method for determining the sampling frequency based on the disproportionation degree is shown in Equation (5): (5) where, represents the downsampled frequency, represents the original frequency, represents the voltage difference degree threshold, represents the current deviation degree threshold. When both the voltage difference degree and the current deviation degree exceed the threshold, a low sampling frequency is adopted. When one of them exceeds the threshold, a medium sampling frequency is adopted. When both do not exceed the threshold, a high sampling frequency is adopted.

[0036] Open - circuit, short - circuit faults, etc. are sudden faults, and a higher sampling frequency is required to capture fault characteristics. Inconsistent self - discharge, etc. are slow - changing faults, and a lower sampling frequency can obtain fault characteristics. The current divergence reflects the abnormal degree of the voltage signal and is an important response to short - term faults. When the divergence is high, it means that there are likely sudden faults such as poor short - circuit contact, and a higher data frequency is required for diagnosis. When the divergence is low, there are no sudden faults, and only faults with relatively gentle development such as self - discharge may exist. At this time, using a lower data frequency does not affect the diagnostic result.

[0037] S3: Construct a fault classification task prompt word based on the down - sampled battery signals, and input the fault classification task prompt word into the fault classification large - model to obtain the fault type and the battery signal sequence corresponding to the fault. In this embodiment, based on the down - sampled battery signals, namely voltage, current, and temperature signals, a fault classification task prompt word is constructed. The fault classification task prompt word includes the task type, sampling frequency, data length, current data (current signal), voltage data (voltage signal), and temperature data (temperature signal), and its structure is shown in Table 1: Table 1 Structure of Fault Classification Task Prompt Word

[0038] The expected output of the large - model is the fault type of the battery, and an example of the output result is: "Battery 1: normal, Battery 2: internal short - circuit fault, Battery 3: normal, …".

[0039] Using the fault classification task prompt word as the input and the fault type as the output to construct a fault classification training set, and based on the fault classification training set, using the mutually exclusive low - rank adaptation fine - tuning method (M - LoRa, mutex Low - Rank Adaptation) to fine - tune the large - model can endow the large - model with the fault classification function and obtain the fault classification large - model. This embodiment is verified on the Qwen2.5 7b model.

[0040] The fault classification large - model classifies according to the input fault classification task prompt word, and outputs the fault type and the data corresponding to the fault.

[0041] S4: Calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct a fault grading task prompt word based on the voltage variance quantile; input the fault grading task prompt word into the fault grading large - model to obtain the fault grade.

[0042] In this embodiment, the fault type of the input battery data is returned during fault classification. If there is a fault, this segment is a fault segment, and the data before downsampling of this segment, that is, the original battery signal data, is obtained. Based on the data within the diagnosed fault segment, the variance sequence of the battery voltage is calculated, and the formula is as follows: (6) Where, represents the average voltage within the battery module at time represents the voltage variance within the battery module at time represents the historical variance sequence of the battery module.

[0043] Calculate the voltage variance quantiles, as shown in Equation (7): (7) Where, represents the first quantile, represents the second quantile, represents the third quantile. First, arrange the variance sequence of the voltage from largest to smallest. The first, second, and third quantiles are the variance values at the 25%, 50%, and 75% positions respectively. If the position is not an integer, it is obtained based on the difference between the values on the left and right sides.

[0044] Construct a fault classification task prompt word based on the voltage variance quantiles. The fault classification task prompt word includes task type, number of batteries, quantiles, fault type, voltage difference degree VDD, and current deviation degree CDD, and its structure is shown in Table 2: Table 2 Structure of Fault Classification Task Prompt Word

[0045] Fine-tune the fault classification function of the large model based on the fault data of the original frequency. The expected output is the fault level, and the fault level can be determined according to factors such as voltage fluctuation. An example is: The output result label example is "Battery 2: Internal short circuit fault level 3".

[0046] Construct a fault classification training set with the fault classification task prompt word as the input and the fault level as the output. Based on the fault classification training set, use the mutually exclusive low-rank adaptation fine-tuning method (M-LoRA) to fine-tune the large model to endow the large language model with the fault classification function and obtain the fault classification large model.

[0047] Therefore, this application uses a battery multi-fault classification and diagnosis framework based on a large model to complete fault classification and quantitative diagnosis. First, the input signal is downsampled based on the degree of disproportion to obtain a low-frequency battery signal sequence. Then, based on the low-frequency signal, a fault classification task prompt word is constructed, and the prompt word is input into the large model to complete the fault classification task and obtain the fault types of all batteries in the battery pack. Finally, based on the data within the diagnosed fault segment, the voltage variance quantile is calculated, and a fault classification prompt word is constructed. The prompt word is input into the large model to complete the fault classification.

[0048] The present invention first proposes a mutually exclusive low-rank adaptation and fine-tuning method (M-LoRA), which decouples the parameter updates in the fault classification and fault classification tasks by constructing two mutually orthogonal low-rank matrices to reduce mutual interference.

[0049] The core idea is to design two LoRa matrices to satisfy the orthogonality constraint where is the Frobenius inner product.

[0050] The two matrices are the trainable parameters corresponding to the fault classification task and the fault classification task respectively. Satisfying the orthogonality condition can ensure non-overlapping update directions for the two tasks, and the gradients will not interfere with each other during the backpropagation process. The specific fine-tuning method is as follows: (1) Construct a low-rank left factor and a low-rank right factor such that and freeze all other parameters.

[0051] (2) Update the large language model LLM parameters to where are the LLM parameters, .

[0052] (3) Calculate and project the gradients of different low-rank matrices during the training of the two tasks respectively.

[0053] (8) where and are the corrected gradients of the two low-rank matrices respectively, while and are the original gradients of the two low-rank matrices with respect to the Loss function respectively.

[0054] (4) Alternately update the matrix parameters (9) where is the learning rate.

[0055] Through the above method, the gradient directions of the two tasks are projected onto orthogonal directions to ensure that the gradient propagation does not affect the effects of the two tasks.

[0056] The following is an illustration for specific application examples.

[0057] As Figure 2 、 Figure 3 shown, there are the voltage data (obtained from simulation experiments) of 6 series-connected single cells of an existing electric vehicle battery pack, and the signal curves under the UDDS (Urban Dynamometer Driving Schedule) and charging conditions.

[0058] Among them, under the UDDS condition, an external short circuit fault occurred in battery 2 from 200 to 400 s, resulting in a voltage drop of about 0.01 V. An internal short circuit fault occurred in battery 2 from 600 to 800 s, resulting in a voltage drop of 0.05 V and a temperature increase of 2°C. A poor contact fault occurred in battery 3 from 1000 to 1200 s, resulting in voltage fluctuations. Under the constant current charging condition, there is a self-discharge fault in battery 1, and the voltage rises slowly. There is an inconsistency fault in battery 4 s, resulting in a higher SOC than other batteries.

[0059] The results of the disproportionation degree under the UUDS condition are as Figure 4 、 Figure 5 shown. During the periods from 200 to 400 s and from 1000 to 1200 s, both the voltage difference degree and the current deviation degree increased significantly. Therefore, the highest sampling frequency was adopted during this period. During the period from 600 to 800 s, only one parameter exceeded the threshold, so the second highest sampling frequency was adopted. In addition, within most of the segments, both indicators did not exceed the threshold, so a lower sampling frequency was adopted.

[0060] The calculation results of the two parameters under the constant current charging condition are as Figure 6 、 Figure 7 shown. Throughout the sampling interval, both of the two characteristic parameters are low and do not exceed the threshold, so a lower sampling frequency is adopted for both.

[0061] The signal downsampling results under the two conditions are as Figure 8 、 Figure 9 shown. The number of signal points after downsampling is 72 and 170 respectively.

[0062] Based on the downsampled signal, prompt words are constructed for fault classification. The diagnostic results are shown in Table 3. It can be seen that all faults are successfully detected and accurately classified.

[0063] Table 3 Fault Classification Results

[0064] Construct a prompt word based on the data integrity signal within the fault segment for fault classification. The classification results are shown in Table 4. It can be seen that all faults are accurately classified.

[0065] Table 4 Fault Classification Results

[0066] Embodiment 2 This embodiment discloses a multi-fault classification and diagnosis system for lithium-ion batteries based on a large language model, including: A data acquisition module configured to: acquire battery signals within the battery pack to be tested, and calculate the disproportion degree of the battery pack to be tested based on the battery signals; A downsampling module configured to: downsample the battery signals based on the disproportion degree to obtain the downsampled battery signals; A fault classification module configured to: construct a fault classification task prompt word based on the downsampled battery signals, and input the fault classification task prompt word into a fault classification large model to obtain the fault type and the battery signal sequence corresponding to the fault; A fault classification module configured to: calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct a fault classification task prompt word based on the voltage variance quantile; input the fault classification task prompt word into a fault classification large model to obtain the fault level.

[0067] Embodiment 3 The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the method in Embodiment 1 are implemented.

[0068] Embodiment 4 The purpose of this embodiment is to provide a computer-readable storage medium. A computer-readable storage medium stores a computer program, and when the program is executed by a processor, the steps of the method in Embodiment 1 are executed.

[0069] The steps involved in the devices in the above Embodiments 3 and 4 correspond to those in Method Embodiment 1. For specific implementation manners, reference may be made to the relevant description part of Embodiment 1. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0070] Those skilled in the art should understand that the various modules or steps of the present invention described above can be implemented by a general-purpose computer device. Optionally, they can be implemented by program codes executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be separately fabricated into individual integrated circuit modules, or multiple modules or steps among them can be fabricated into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0071] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0072] 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 based on the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A multi-fault hierarchical diagnosis method for lithium-ion batteries based on large language models, characterized in that Including: Obtain the battery signals within the battery pack to be tested, and calculate the disproportionation degree of the battery pack to be tested based on the battery signals; Perform downsampling on the battery signals based on the disproportionation degree to obtain the downsampled battery signals; Construct a fault classification task prompt word based on the downsampled battery signals, and input the fault classification task prompt word into the fault classification large model to obtain the fault type and the battery signal sequence corresponding to the fault; Calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct a fault grading task prompt word based on the voltage variance quantile; Input the fault grading task prompt word into the fault grading large model to obtain the fault level.

2. The multi-fault hierarchical diagnosis method for lithium-ion batteries based on large language models according to claim 1, wherein, The disproportionation degree includes voltage difference degree and current deviation degree.

3. The method for multi-fault hierarchical diagnosis of lithium-ion batteries based on large language models according to claim 2, wherein Calculate the voltage difference degree based on the characteristic voltage matrix, expressed as: Among them, VDD represents the voltage difference degree, represents the improvement correlation coefficient, represents the n-th battery, represents the m-th battery, represents the improved multiplication operation.

4. The method for multi-fault hierarchical diagnosis of lithium-ion batteries based on large language models according to claim 2, wherein, The calculation formula for the current deviation degree is: Among them, represents the current sequence, represents the current sequence after normalization and adding a square wave signal, represents the scaling factor, represents the current deviation degree.

5. The multi-fault hierarchical diagnosis method for lithium-ion batteries based on large language models according to claim 1, wherein, Determine the voltage difference degree threshold and the current deviation degree threshold through the quantile method, and determine the sampling frequency based on the disproportionation degree, expressed as: Among them, represents the frequency after downsampling, represents the original frequency, represents the voltage difference threshold, represents the current deviation threshold.

6. The multi-fault hierarchical diagnosis method for lithium-ion batteries based on large language models according to claim 1, wherein The fault classification task prompt word includes task type, sampling frequency, data length, current data, voltage data, and temperature data. The fault grading task prompt word includes task type, number of batteries, quantile, fault type, voltage difference degree, and current deviation degree.

7. The method for multi-fault hierarchical diagnosis of lithium-ion batteries based on large language models according to claim 1, wherein The fault classification large model and the fault grading large model are obtained by fine-tuning the large language model through the mutually exclusive low-rank adaptation fine-tuning method.

8. A multi-fault hierarchical diagnosis system for lithium-ion batteries based on large language models, characterized in that, Including: A data acquisition module, which is configured to: obtain the battery signals within the battery pack to be tested, and calculate the disproportionation degree of the battery pack to be tested based on the battery signals; A downsampling module, which is configured to: perform downsampling on the battery signals based on the disproportionation degree to obtain the downsampled battery signals; A fault classification module, which is configured to: construct a fault classification task prompt word based on the downsampled battery signals, and input the fault classification task prompt word into the fault classification large model to obtain the fault type and the battery signal sequence corresponding to the fault; A fault grading module, which is configured to: calculate the voltage variance quantile based on the battery signal sequence corresponding to the fault, and construct a fault grading task prompt word based on the voltage variance quantile; Input the fault grading task prompt word into the fault grading large model to obtain the fault level.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the multi-fault grading diagnosis method for lithium-ion batteries based on the large language model described in any one of claims 1-7.

10. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the multi-fault grading diagnosis method for lithium-ion batteries based on the large language model described in any one of claims 1-7.

Images