Power battery multi-state collaborative estimation method and system based on large language model

Abstract

The application belongs to the technical field of lithium ion batteries, and provides a power battery multi-state cooperative estimation method and system based on a large language model, which initializes the large language model; extracts aging characteristics under random charging behavior, extracts window extreme voltage and cumulative charge as aging characteristics; based on the initialized large language model, the input and output of the model are determined according to the multi-state cooperative estimation task, the extracted aging characteristics are fused with the battery state parameters to serve as input features together, and the input type is adjusted; the input and output of the base model are used to improve the large language model, and an output layer module is added to the large language model to adapt to the multi-state cooperative estimation task; based on the improved large language model and the actual data set, the model is fine-tuned by using a low-rank adaptive method; and the target fragmented charging data is processed based on the fine-tuned model to perform multi-state cooperative estimation. The application can realize multi-state cooperative precise estimation.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of lithium ion batteries, and particularly relates to a power battery multi-state cooperative estimation method and system based on a large language model. BACKGROUND

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

[0003] With the large-scale promotion of electric vehicles, lithium ion batteries are generally considered to be a key energy storage technology that is expected to achieve sustainable development. Due to its high energy density, low self-discharge rate, long cycle life, lightweight, and other characteristics, it is widely used in various fields. However, in engineering practice, the performance of lithium ion batteries will inevitably and irreversibly decline due to factors such as extreme working conditions, abnormal charging / discharging, internal material structure degradation, high temperature, misuse, etc. There are many safety hazards. This performance decline will significantly reduce the endurance time of electric vehicles or electronic devices, increase the operating burden, and thus accelerate battery aging, and even cause safety accidents such as overheating, fire or explosion. Therefore, safe management of lithium ion batteries is the only way to alleviate the endurance anxiety of electric vehicles and avoid high-risk risks such as thermal runaway.

[0004] State estimation technology in battery management systems can effectively evaluate the safety of lithium ion batteries and ensure their reliable operation. Today, the key states of lithium ion batteries include the state of health (SOH), the state of charge (SOC), and the remaining useful life (RUL), which are different in definition, function, and time scale. At the same time, due to the fact that lithium ion batteries are a dynamic, time-varying and nonlinear electrochemical system, there is a complex coupling relationship between the three states. Therefore, compared to the estimation technology for a single state, multi-state cooperative estimation is a more promising solution.

[0005] The current stage, the method of lithium ion battery multi-state collaborative estimation includes: definition method, model-based estimation method and data-driven estimation method. In the traditional definition method, the health state is defined as the percentage of the ratio of the available capacity to the rated capacity, the state of charge is defined as the percentage of the ratio of the current remaining capacity to the available capacity, and the remaining service life is defined as the number of cycles remaining from the current to the preset health threshold. The defects of this method are: in engineering practice, the available capacity required to calculate the health state and the state of charge cannot be accurately obtained, and the remaining service life cannot be directly calculated; the model-based method estimates the battery state by constructing a mathematical model that can reflect the internal electrochemical reaction mechanism of the battery, mainly including empirical model, equivalent circuit model and electrochemical model. The defects of this method are: there are significant limitations in flexibility, precision and dynamic response, which makes it difficult to be stably deployed and adapt in practice; the data-driven method does not need to deeply understand the internal reaction mechanism of the lithium ion battery, and realizes estimation by establishing the mapping relationship between the battery state and the input features. This method is widely used and has high precision, but the strict prerequisite of feature engineering, the model is prone to overfitting, and the collaborative estimation accuracy is low, which limits its further development, so the technical means and application scope of this method still need to be strengthened. SUMMARY

[0006] The present application provides a kind of based on large language model's power battery multi-state collaborative estimation method and system to solve the above problems, can alleviate the obstacle that fragmented charging behavior brings to feature engineering in engineering practice, efficiently, flexibly and reliably extract the aging characteristics used for health state estimation, and establish the mapping relationship between characteristics and state by fine-tuning 3B parameter scale large language model. The present application solves the problem that the flexibility of feature engineering is poor due to random charging behavior in engineering practice, clarifies the complex coupling relationship between the multi-state of lithium ion battery by means of large language model, and then realizes multi-state collaborative accurate estimation.

[0007] According to some embodiments, the present application adopts the following technical scheme:

[0008] A kind of based on large language model's power battery multi-state collaborative estimation method, comprising the following steps:

[0009] Initialize large language model;

[0010] Extract aging characteristics under random charging behavior, extract window extreme voltage and cumulative charge as aging characteristics;

[0011] Based on the initialized large language model, determine the input and output of the model according to the multi-state collaborative estimation task, fuse the battery state parameters with the extracted aging characteristics as input features, and adjust the input type;

[0012] Based on the input and output of the model, the large language model is improved, and an output layer module is added to the large language model to adapt to the multi-state collaborative estimation task.

[0013] Based on the improved large language model and the actual data set, the model is fine-tuned by using a low-rank adaptation method.

[0014] Based on the fine-tuned model, the target fragmentation charging data is processed, and multi-state collaborative estimation is performed.

[0015] As an alternative embodiment, the process of extracting aging characteristics under random charging behavior includes:

[0016] Obtain the fragmentation charging data of the power battery in a certain cycle, and filter and combine the fragmentation charging data to form voltage-charge data;

[0017] Based on the voltage-charge data, the average slope corresponding to each window is iteratively calculated by using a two-dimensional sliding window method;

[0018] Summarize the average slopes corresponding to all windows, and record the minimum average slope and the corresponding window information;

[0019] Based on the window information, extract the corresponding aging characteristics in the cycle, including the window extreme voltage and the cumulative charge.

[0020] As an alternative embodiment, the process of obtaining the fragmentation charging data of the power battery in a certain cycle and filtering and combining the fragmentation charging data to form voltage-charge data includes:

[0021] Obtain the voltage, current and charge data of the power battery, eliminate irrelevant parameter data, and retain the voltage and charge data;

[0022] Based on the voltage and charge data, combine the voltage-charge two-dimensional data with the charge as the x direction and the voltage as the y direction.

[0023] As an alternative embodiment, the process of iteratively calculating the average slope corresponding to each window based on the voltage-charge data by using a two-dimensional sliding window method includes:

[0024] Based on the voltage-charge data, a two-dimensional sliding window method is used, with a set voltage length ΔV as the sliding window size and the voltage rising direction as the window sliding direction;

[0025] According to the set sliding window size and sliding direction, the window slides point by point, records the window information and data point information corresponding to each sliding, and ends the sliding until all data points are covered;

[0026] Based on the recorded window information and the data point information in the window, if a window contains multiple data points, the slope between two points is calculated point by point in the window, and n-1 slope values {m1, m2, …, mn-1} are summarized, n being the total number of data points in the window. n-1};

[0027] Based on the n-1 summarized slope values, the average slope corresponding to the window is calculated, and the average slopes corresponding to all windows are iteratively calculated based on this; wherein the average slope M i corresponding to the i-th window is defined as:

[0028]

[0029] As an optional implementation, the process of summarizing the average slopes corresponding to all windows and recording the minimum average slope and the window information corresponding thereto includes: based on the average slopes of the windows, all average slopes {M1, M2, …} are summarized.

[0030] The minimum average slope M min and the sliding window information corresponding thereto are extracted, including the voltage values and the charge values of the data points in the window.

[0031] As an optional implementation, the window extreme voltage is the maximum voltage value V max and the minimum voltage value V min corresponding to the upper and lower boundaries of the window, and the cumulative charge Q c is the difference between the charge value corresponding to the last data point in the window and the charge value corresponding to the first data point in the window.

[0032] As an optional implementation, a large language model with a 3B parameter scale is selected as the large language model.

[0033] As an optional implementation, the input and output of the model are determined according to the multi-state collaborative estimation task, and the process of adjusting the input includes: setting the model input as the maximum voltage value V max , the minimum voltage value V min , the cumulative charge Q c , the voltage value V t at the current time, the current value I t at the current time, and the last time health state estimation value SOH t-1 , and setting the model output as the health state, the state of charge, and the remaining useful life.

[0034] Based on the model input and output, the input type is adjusted to a string type, and the output type is adjusted to a floating point type.

[0035] As an alternative embodiment, the process of improving the large language model includes: adding an output layer module to the large language model based on the model output type and dimension, the output layer module including multiple linear layers and their activation functions and normalization layers to change the dimension from 2048 to 2, i.e., the health state and the state of charge;

[0036] Based on the output layer module, a lightweight model is added to realize the remaining useful life prediction when the health state estimate accumulates to a set value.

[0037] As an alternative embodiment, the process of fine-tuning the model using the low-rank adaptation method includes: taking the historical acquisition data of the actual power battery as an actual data set, dividing the actual data set into a training set and a test set, the training set being used for fine-tuning the large language model, and the test set being used for verifying the performance of the large language model;

[0038] The low-rank adaptation method is applied to the query matrix W Q and the value matrix W V of the multi-head attention mechanism of the Qwen layer of the large language model.

[0039] Based on the low-rank adaptation method, the required hyperparameters are set, including a learning rate of 1e-4, a batch size of 128, and a number of training rounds of 50.

[0040] As an alternative embodiment, in the process of performing multi-state collaborative estimation, the root mean square error is used as an index to evaluate the performance of the health state and state of charge estimation, and the difference d R between the estimated value and the real value of the number of cycles corresponding to the preset health threshold is used as an index to evaluate the performance of the remaining useful life prediction.

[0041] A power battery multi-state collaborative estimation system based on a large language model, comprising:

[0042] A model initialization module configured to initialize the large language model;

[0043] A feature extraction module configured to extract aging features under random charging behavior, and extract window extreme voltage and cumulative charge as aging features;

[0044] An input feature determination module configured to determine the input and output of the model based on the initialized large language model according to the multi-state collaborative estimation task, and to fuse the battery state parameters extracted by the feature extraction module as input features together with the aging features, and to adjust the input type;

[0045] A model improvement module configured to improve the large language model based on the input and output of the model, and to add an output layer module to the large language model to adapt to the multi-state collaborative estimation task;

[0046] The model fine-tuning module is configured to fine-tune the model by using a low-rank adaptation method based on the improved large language model and the actual data set;

[0047] The multi-state collaborative estimation module is configured to process the target fragmented charging data based on the fine-tuned model to perform multi-state collaborative estimation.

[0048] The feature extraction module comprises:

[0049] The data preprocessing module is configured to obtain fragmented charging data of the power battery within a cycle, and filter and combine the fragmented charging data to form voltage-charge data;

[0050] The iterative calculation module is configured to iteratively calculate the average slope corresponding to each window based on the voltage-charge data by using a two-dimensional sliding window method;

[0051] The summary module is configured to summarize the average slopes corresponding to all windows, and record the minimum average slope and the window information corresponding thereto;

[0052] The feature extraction module is configured to extract the aging characteristics corresponding to the cycle based on the window information, including the window extreme voltage and the cumulative charge.

[0053] Compared with the prior art, the present application has the following advantages:

[0054] (1) Compared with the definition method based on complete charging and discharging data, the present application is aimed at engineering practical application and does not require complete charging and discharging data, but can realize reliable extraction of aging characteristics under random charging behavior, supporting multi-state collaborative estimation.

[0055] (2) Compared with the state estimation method based on the model, the present application does not need to deeply understand the internal electrochemical reaction mechanism of the lithium ion battery, nor does it need to establish a complex mathematical model or an electrochemical model, but only relies on the nonlinear fitting ability of the large language model to establish the mapping relationship between the multi-state and the input characteristics, greatly improving the flexibility, precision and dynamic response ability of the method.

[0056] (3) Compared with the feature engineering method for extracting aging characteristics in the prior art, the present application fully considers the typical characteristics of fragmented charging and discharging of electric vehicles in engineering practice, and realizes effective, efficient and reliable extraction of aging characteristics under random charging behavior only relying on voltage and other conventional parameters, breaking through the long-standing barriers of poor flexibility, strict prerequisites and difficulty in effective implementation of existing methods, and without additional data processing and complex calculation process.

[0057] (4) Compared with the existing data-driven battery multi-state collaborative estimation method, the application solves the problem that the traditional machine learning / deep learning model can only achieve single state accurate estimation on a specific data set, and uses a large language model with a large parameter scale and model fine-tuning technology to clarify the complex coupling relationship between the multi-state of the battery, and improves the universality, accuracy and robustness of the method.

[0058] (5) Compared with the mainstream battery multi-state collaborative estimation method, the feature engineering in the application is efficient and easy to implement, and the estimation performance is stable and reliable, does not involve the electrochemical reaction mechanism and aging mechanism of lithium ion batteries, can adapt to the complex operating conditions of electric vehicles in engineering practice, and has good development potential and expandability.

[0059] In order to make the above-mentioned objects, features and advantages of the application more obvious and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0060] The drawings accompanying the specification of the application serve to provide a further understanding of the application, and the schematic embodiments of the application and their descriptions serve to explain the application and do not constitute an improper limitation on the application.

[0061] Figure 1 is a flowchart of a power battery aging feature extraction method of an embodiment;

[0062] Figure 2 is a battery health state degradation curve in data of an embodiment;

[0063] Figure 3 is a voltage-charge curve of a lithium ion battery Cell#1 of an embodiment;

[0064] Figure 4 is a flowchart of a power battery multi-state collaborative estimation method of an embodiment;

[0065] Figure 5 is an estimation result of the battery health state of an embodiment;

[0066] Figure 6 is an estimation result of the battery state of charge of an embodiment;

[0067] Figure 7 is a prediction result of the remaining useful life of the battery of an embodiment. DETAILED DESCRIPTION

[0068] The application will be further described below in combination with the drawings and embodiments.

[0069] It should be noted that the following detailed description is illustrative only, and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0070] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, steps, operations, elements, components, and / or groups thereof, but do not preclude the presence or addition of one or more other features, steps, operations, elements, components, and / or groups thereof.

[0071] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0072] Embodiment one

[0073] As shown in the figure, the embodiment provides a power battery aging feature extraction method. In this embodiment, lithium ion batteries in electric vehicles are taken as examples for detailed description. Figure 1

[0074] The method comprises the following steps:

[0075] Fragmentation charging data of lithium ion batteries in electric vehicles within a certain cycle are acquired, and voltage-charging capacity data are screened and combined; specifically including:

[0076] Fragmentation charging data of lithium ion batteries within a certain cycle are acquired by a battery management system mounted in an electric vehicle, including voltage, current and charging capacity, etc.; irrelevant parameter data are eliminated, and data of voltage and charging capacity are retained;

[0077] Based on the data of voltage and charging capacity, voltage-charging capacity two-dimensional data are combined with charging capacity as x direction and voltage as y direction.

[0078] Based on the voltage-charging capacity data, two-dimensional sliding window technology is adopted to iteratively calculate average slopes corresponding to each window; specifically including:

[0079] Based on the voltage-charging capacity data, two-dimensional sliding window technology is adopted, a certain fixed voltage length ΔV is taken as the sliding window size, and the voltage rising direction (i.e. y direction) is taken as the window sliding direction. In this embodiment, the voltage length ΔV is 0.1 V.

[0080] Based on the set sliding window size and sliding direction, the window slides point by point, records window information and data point information corresponding to each sliding, and ends the sliding until all data points are covered. ​

[0081] Based on the recorded window information and the data point information within the window, taking the i-th window as an example, assuming that it contains n data points, the slope between two points is calculated point by point within the window, and a total of n-1 slope values {m1, m2, …, mn-1} are obtained. n-1};

[0082] Based on the n-1 slope values, the average slope corresponding to the window is calculated, and the average slope corresponding to all windows is iteratively calculated according to the average slope; wherein the average slope M i corresponding to the i-th window is defined as:

[0083]

[0084] Based on the average slope of each window, all average slopes are summarized and the minimum average slope and its corresponding window information are recorded; specifically including:

[0085] Based on the average slope of each window, all average slopes {M1, M2, …} are summarized.

[0086] The minimum average slope M min and the corresponding sliding window information are extracted, including the voltage value and the charge value of each data point in the window.

[0087] Based on the window information, the aging characteristics corresponding to the cycle are extracted, including the window extreme voltage and the cumulative charge; wherein the window extreme voltage is defined as the maximum voltage value V max and the minimum voltage value V min corresponding to the upper and lower boundaries of the window, and the cumulative charge Q c is defined as the difference between the charge value corresponding to the last data point and the charge value corresponding to the first data point in the window.

[0088] The following specific examples illustrate the method and effect of embodiment one:

[0089] 1. Obtain the fragmentation charging data of lithium ion batteries in an electric vehicle in a certain cycle, and screen and combine voltage-charge data;

[0090] The lithium ion battery data set contains a total of 9 battery monomers, named Cell#1, #2, #3, #4, #5, #6, #7, #8 and #9 respectively. The rated capacity of the monomer is 2.5 Ah. The battery health state degradation curve of the above monomers is shown in Figure 2 . In order to simulate the random charging behavior of electric vehicles in engineering practice, the original data in this case is fragmented and truncated. Specifically, within any cycle, the original complete data is randomly truncated into charging fragments, and the voltage range of the fragment is 0.2V-0.5V.

[0091] The irrelevant parameter data is removed, and the voltage and charge data are reserved.

[0092] Based on the voltage and charge data, the voltage-charge two-dimensional data is combined with the charge as the x direction and the voltage as the y direction. The voltage-charge curve of the lithium ion battery Cell#1 is shown as Figure 3 .

[0093] 2. Based on the voltage-charge data, the average slope of each window is iteratively calculated by using the two-dimensional sliding window technology.

[0094] Based on the voltage-charge data, the two-dimensional sliding window technology is used, a certain fixed voltage length ΔV = 0.1V is used as the sliding window size, and the voltage rising direction (y direction) is used as the window sliding direction.

[0095] Based on the set sliding window size and sliding direction, the window slides point by point, records the window information and data point information corresponding to each sliding, and ends until all data points are covered.

[0096] Based on the recorded window information and data point information in the window, taking the ith window as an example, assuming that it contains n data points, the slope between two points is calculated point by point in the window, and n-1 slope values {m1, m2, …, mn-1} are obtained. n-1}.

[0097] Based on the n-1 slope values, the average slope of the window is calculated, and the average slope of all windows is iteratively calculated based on the average slope; wherein the average slope M i of the ith window is defined as:

[0098]

[0099] 3. Based on the average slope of each window, all average slopes are summarized and the minimum average slope and the corresponding window information are recorded.

[0100] Based on the average slope of each window, all average slopes {M1, M2, …} are summarized.

[0101] The minimum average slope M min and the corresponding sliding window information, including the voltage value and the charge value corresponding to each data point in the window, are extracted.

[0102] 4. Based on the window information, the aging characteristics corresponding to the cycle are extracted, including the window extreme voltage and the cumulative charge.

[0103] The window extreme voltage is defined as the maximum voltage value V max and the minimum voltage value V min corresponding to the upper and lower boundaries of the window, and the cumulative charge Qc defined as the difference between the charge value corresponding to the last data point in the window and the charge value corresponding to the first data point in the window.

[0104] Embodiment Two

[0105] As shown in Figure 4 , the embodiment provides a method for multi-state collaborative estimation of power batteries. In the embodiment, the method is realized based on a large language model, and specifically includes the following steps:

[0106] The 3B parameter scale Qwen2.5 large language model disclosed by Aliyun is adopted as the large language model in the embodiment, and the model architecture, parameters and tokenizer are initialized; specifically including:

[0107] Based on the 3B parameter scale large language model, the model architecture is loaded, including an embedding layer and 36 Qwen layers based on improved Transformer decoders;

[0108] Based on the model architecture, the pre-trained model parameters and tokenizer are loaded for model initialization.

[0109] Based on the 3B parameter scale large language model, the model input and output are determined according to the multi-state collaborative estimation task, and the input type is adjusted; specifically including:

[0110] Based on the multi-state collaborative estimation task and the power battery aging feature extraction method provided in Embodiment One, the model input is set to the maximum voltage value V max , the minimum voltage value V min , the cumulative charge Q c , the current time voltage value V t , the current time current value I t , and the last time health state estimation value SOH t-1 , and the model output is set to the health state, the state of charge and the remaining service life;

[0111] Based on the model input and output, the input type is adjusted to a string type, and the output type is adjusted to a floating point type.

[0112] Based on the model input and output, an output layer module is added to the large language model to adapt to the multi-state collaborative estimation task; specifically including:

[0113] Based on the model output type and dimension, an output layer module is added to the large language model, including multiple linear layers, their activation functions and normalization layers, to realize the conversion of the dimension from 2048 to 2, i.e. the health state and the state of charge;

[0114] Based on the output layer module, a lightweight model (such as XGBoost) is added to realize the remaining useful life prediction when the health state estimation accumulates k values; wherein k is 20.

[0115] Based on the improved large language model and the actual data set, the model is fine-tuned by using the low-rank adaptive technology (i.e. LoRA);

[0116] The actual data set is the historical collection data of the actual power battery.

[0117] Specifically, it includes:

[0118] The data set is adjusted to simulate the actual working condition, and is divided into a training set and a test set, the training set is used for model fine-tuning, and the test set is used for verifying the performance of the model;

[0119] The LoRA technology is used to fine-tune the model, which is applied to the query matrix W Q and the value matrix W V of the multi-head attention mechanism in the Qwen layer.

[0120] Based on the fine-tuning technology, the required hyperparameters are set, including the learning rate of 1e-4, the batch size of 128, and the number of training rounds of 50.

[0121] Based on the fine-tuned model and the fragmented data, multi-state collaborative estimation is realized.

[0122] The fragmented data here is the fragmented data collected in a certain cycle;

[0123] Specifically, it includes:

[0124] The root mean square error (Root Mean Square Error, RMSE) is used as an index to evaluate the performance of health state and state of charge estimation, and the difference d R between the estimated value and the real value of the cycle number corresponding to the preset health threshold is used as an index to evaluate the performance of the remaining useful life prediction; wherein the preset health threshold is 80%.

[0125] The method and effect of the second embodiment will be described below with specific examples:

[0126] 1. The 3B parameter scale Qwen2.5 large language model disclosed by Aliyun is used as the large language model in the present application, and the model architecture, parameters and tokenizer are initialized:

[0127] The 3B parameter scale Qwen2.5 large language model disclosed by Aliyun is used as the large language model in the present case, which includes a Qwen layer based on an improved Transformer decoder, and the total parameter quantity reaches 3.09 billion.

[0128] Based on the 3B parameter scale large language model, load the model architecture, including the embedding layer, 36 Qwen layers based on the improved Transformer decoder. The embedding layer includes word embedding and rotated position encoding (RoPE), and the Qwen layer includes multi-head attention mechanism and feedforward network. This model architecture serves as the model architecture of the large language model in this case, and is loaded in the pre-trained model parameters and tokenizer for model initialization.

[0129] 2. Based on the 3B parameter scale large language model, determine the model input and output according to the multi-state collaborative estimation task, and adjust the input type:

[0130] Based on the multi-state collaborative estimation task and the aging feature extraction method, the model input is set to the maximum voltage value V max , the minimum voltage value V min , the cumulative charge Q c , the current voltage value V t , the current current value I t , and the last time health state estimation value SOH t-1 , and the model output is set to the health state, state of charge and remaining useful life.

[0131] Based on the model input and output, the input type is adjusted to string type, i.e. {str(V max ), str(V min ), str(Q c ), str(V t ), str(I t ), str(SOH t-1 )} and the output type is adjusted to float type, i.e. {SOH t , SOC t , RUL t}.

[0132] 3. Based on the model input and output, add an output layer module to the large language model to adapt to the multi-state collaborative estimation task:

[0133] Based on the model output type and dimension, add an output layer module to the large language model, including multi-layer linear layer and its activation function and normalization layer, to realize the conversion of dimension from 2048 to 2, i.e. health state and state of charge. Specifically, the multi-layer linear layer contains four layers, i.e. {2048, 1024}, {1024, 256}, {256, 64} and {64, 2}, and the activation function is uniform ReLU.

[0134] Based on the output layer module, add a lightweight model, i.e. XGBoost, to realize the remaining useful life prediction when the health state estimation accumulates k values; where k is 20.

[0135] 4. Based on the improved large language model and the actual data set, the model is fine-tuned by using the LoRA technology:

[0136] In this embodiment, the data set comes from the public data set in the journal article (https: / / doi.org / 10.1038 / s41467-022-29837-w). The data set is divided into a training set and a test set, and the data of Cell#1, Cell#2, Cell#4, Cell#5, Cell#7 and Cell#8 are set as the training set, and the data of Cell#3, Cell#6 and Cell#9 are set as the test set. The training set is used for model fine-tuning, and the test set is used for verifying the performance of the model.

[0137] The model is fine-tuned by using the LoRA technology, which is applied to the query matrix W of the multi-head attention mechanism in the Qwen layer Q and the value matrix W V . The required hyperparameters are set, including the learning rate of 1e-4, the batch size of 128, and the number of training rounds of 50.

[0138] 5. Based on the fine-tuned model and the fragmented data, multi-state collaborative estimation is realized:

[0139] The root mean square error (Root Mean Square Error, RMSE) is used as an indicator to evaluate the performance of state of health and state of charge estimation, and the difference d R between the estimated value and the true value of the cycle number corresponding to the preset health threshold is used as an indicator to evaluate the performance of the remaining useful life prediction; wherein the preset health threshold is 80%.

[0140] In this case, Figure 5 and Figure 6 respectively show the estimation results of the battery state of health and state of charge, and the corresponding RMSE is 0.0046 and 0.0087, respectively. Figure 7 shows the prediction result of the remaining useful life of Cell#9, and d R is only 3.

[0141] The above results show that in the estimation of state of health and state of charge, the estimated value can effectively and reliably follow the actual value, and the estimation accuracy is high. As Figure 5 shown, there are no abnormal values in the estimated value of the battery state of health, and the estimation stability is good. In Figure 6 , the change trend of the state of charge estimation value is consistent with the true value, and there is no frequent fluctuation caused by large model parameters. At the same time, relying on accurate state of health estimation, the invention performs well in the prediction of the remaining useful life, as Figure 7As shown, the remaining useful life can be accurately predicted with only 20 health state estimates.

[0142] Embodiment three

[0143] A power battery multi-state collaborative estimation system based on a large language model, comprising:

[0144] A model initialization module configured to initialize the large language model;

[0145] A feature extraction module configured to extract aging features under random charging behavior, and extract window extreme voltage and cumulative charge as aging features;

[0146] An input feature determination module configured to determine the input and output of the model based on the initialized large language model according to the multi-state collaborative estimation task, to fuse the aging features extracted by the feature extraction module and the battery state parameters as input features, and to adjust the input type;

[0147] A model improvement module configured to improve the large language model based on the input and output of the model, and add an output layer module to the large language model to adapt to the multi-state collaborative estimation task;

[0148] A model fine-tuning module configured to fine-tune the model using a low-rank adaptation method based on the improved large language model and actual data sets;

[0149] A multi-state collaborative estimation module configured to process target fragmented charging data based on the fine-tuned model and perform multi-state collaborative estimation.

[0150] The feature extraction module comprises:

[0151] A data preprocessing module configured to obtain fragmented charging data of a power battery within a cycle, and to filter and combine the fragmented charging data to form voltage-charge data;

[0152] An iterative calculation module configured to calculate the average slope corresponding to each window based on the voltage-charge data using a two-dimensional sliding window method;

[0153] A summary module configured to summarize the average slopes corresponding to all windows and record the minimum average slope and its corresponding window information;

[0154] The feature extraction module is configured to extract the corresponding aging features within the cycle based on the window information, including window extreme voltage and cumulative charge.

[0155] Those skilled in the art will appreciate that embodiments of the present application can be devised for a variety of applications. FIG. 1 illustrates an example of a system 100 that can implement an embodiment of the present application. As shown in FIG. 1, system 100 includes a host computer 110 that is configured to communicate via one or more wired or wireless communication links 120 with one or more remote devices, such as a user terminal 130. Host computer 110 can be configured to communicate with user terminal 130 directly or via one or more intermediary devices. For example, host computer 110 can be configured to communicate with user terminal 130 via one or more core networks 140, one or more core network 140, and / or one or more networks 150. By way of example, network 150 can include a local wired or wireless network, and / or a network 150 can include a cellular network. By way of further example, user terminal 130 can communicate with host computer 110 directly or via one or more intermediary devices over network 150 and / or core network 140.

[0156] The present application is described in reference to the drawings, which are as follows: Figure 1 Figure 1

[0157] Figure 1 Figure 1

[0158] Figure 1 Figure 1

[0159] The foregoing is merely illustrative of the principles of this application and various modifications can be made by those skilled in the art without departing from the scope and spirit of the application. For example, the sequence of operations can be varied and / or other steps can be added or made optional. Thus, the foregoing description is not intended to be limiting. The scope of the application is ported to the following claims.​​​​​​​​

Claims

1. A method for collaborative estimation of multi-states of power batteries based on a large language model, characterized by: The following steps are involved: Initialize the large language model; Aging features are extracted under random charging behavior, and the window extreme voltage and cumulative charge capacity are extracted as aging features; The process of extracting aging characteristics under random charging behavior includes: Obtaining fragmented charging data of the power battery within a certain cycle, and screening and combining the fragmented charging data to form voltage-charge capacity data; Based on the voltage-charge data, a two-dimensional sliding window method is used to iteratively calculate the average slope corresponding to each window; Summarize the average slopes corresponding to all windows, and record the minimum average slope and its corresponding window information; Based on the window information, extracting corresponding aging characteristics within the cycle, including window extreme voltage and cumulative charge capacity; Based on the initialized large language model, the model input and output are determined according to the multi-state collaborative estimation task. The extracted aging features are fused with the battery state parameters as input features, and the input type is adjusted. Based on the input and output of the model, the large language model is improved by adding an output layer module to the large language model to adapt to the multi-state collaborative estimation task; The process of improving the large language model includes: adding an output layer module to the large language model based on the model output type and dimension, wherein the output layer module includes multiple linear layers and their activation functions and normalization layers to achieve the conversion of the dimension from 2048 to 2, namely the health state and the charge state; Based on the output layer module, a lightweight model is added to realize the remaining service life prediction when the health state estimation accumulates to the set value; Based on the improved large language model and real-world datasets, the low-rank adaptation method is used to fine-tune the model; The target fragmented charging data is processed based on the fine-tuned model to perform multi-state collaborative estimation.

2. The method for collaboratively estimating multiple states of a power battery based on a large language model according to claim 1, characterized in that: The process of obtaining fragmented charging data of a power battery in a certain cycle, screening and combining the fragmented charging data to form voltage-charge capacity data includes: Obtain the voltage, current and charge capacity data of the power battery, remove irrelevant parameter data, and retain the voltage and charge capacity data; Based on the voltage and charge data, the charge is x Direction, voltage is y direction, combined into voltage-charge two-dimensional data; Alternatively, based on the voltage-charge capacity data, a process of iteratively calculating the average slope corresponding to each window using a two-dimensional sliding window method includes: Based on the voltage-charge data, a two-dimensional sliding window method is used to set the voltage length ∆ V is the sliding window size, and the direction of voltage increase is the window sliding direction; According to the set sliding window size and sliding direction, the window slides point by point, and the window information corresponding to each slide and the data point information in the window are recorded until all data points are covered and the sliding ends; Based on the recorded window information and the data point information in the window, if a window contains multiple data points, the slope between two points is calculated point by point in the window and the total is summarized. n -1 slope value , n is the total number of data points in the window; Aggregate-based n -1 slope value, calculate the average slope corresponding to the window, and iteratively calculate the average slope corresponding to all windows based on this; i The average slope corresponding to the window M i Defined as: ； Alternatively, the process of summarizing the average slopes corresponding to all windows and recording the minimum average slope and its corresponding window information includes: summarizing all the average slopes based on the average slopes of the windows ; Extract the minimum average slope M min and its corresponding sliding window information, including the voltage value and charge capacity value corresponding to each data point in the window.

3. The method for collaboratively estimating multiple states of a power battery based on a large language model according to claim 1, wherein: The window extreme voltage is the maximum voltage value corresponding to the upper and lower boundaries of the window V max and minimum voltage V min , cumulative charge Q c It is the difference between the charge value corresponding to the last data point in the window and the charge value corresponding to the first data point.

4. The method for collaboratively estimating multiple states of a power battery based on a large language model according to claim 1, wherein: The input and output of the model are determined according to the multi-state collaborative estimation task, and the process of adjusting the input includes: setting the model input to the maximum voltage value V max , minimum voltage value V min , cumulative charge Q c , and the current voltage value V t , Current value at the current moment I t and the estimated health status at the previous moment SOH t-1 , set the model output to health state, state of charge and remaining service life; Based on the model input and output, the input type is adjusted to a string type, and the output type is adjusted to a floating-point number type.

5. The method for collaboratively estimating multiple states of a power battery based on a large language model according to claim 1, wherein: The actual data set is based on historical data from actual power batteries. This data set is divided into a training set and a test set. The training set is used to fine-tune the large language model, and the test set is used to verify the performance of the large language model. Apply the low-rank adaptation method to the query matrix W of the multi-head attention mechanism in the Qwen layer of the large language model Q Sum matrix W V superior; Based on the low-rank adaptation method, the required hyperparameters were set, including a learning rate of 1e-4, a batch size of 128, and a number of training rounds of 50.

6. The method for collaboratively estimating multiple states of a power battery based on a large language model according to claim 1, wherein: In the process of multi-state collaborative estimation, the root mean square error is used as an indicator to evaluate the health state and state of charge estimation performance, and the difference between the estimated value and the true value of the number of cycles corresponding to the preset health threshold is used. d R As an indicator to evaluate the performance of remaining useful life prediction.

7. A power battery multi-state collaborative estimation system based on a large language model using the method according to claim 1, characterized in that: include: A model initialization module is configured to initialize a large language model; A feature extraction module is configured to extract aging features under random charging behavior, and extract window extreme voltage and cumulative charge amount as aging features; An input feature determination module is configured to determine the input and output of the model based on the initialized large language model and the multi-state collaborative estimation task, fuse the aging features extracted by the feature extraction module with the battery state parameters as input features, and adjust the input type; A model improvement module is configured to improve the large language model based on the input and output of the model, and add an output layer module to the large language model to adapt to the multi-state collaborative estimation task; The model fine-tuning module is configured to fine-tune the model using a low-rank adaptation method based on the improved large language model and the actual dataset; The multi-state collaborative estimation module is configured to process the target fragmented charging data based on the fine-tuned model and perform multi-state collaborative estimation.

8. The power battery multi-state collaborative estimation system based on a large language model as claimed in claim 7 is characterized in that: The feature extraction module includes: a data preprocessing module configured to obtain fragmented charging data of the power battery within a certain cycle, and to filter and combine the fragmented charging data to form voltage-charge capacity data; an iterative calculation module configured to iteratively calculate an average slope corresponding to each window using a two-dimensional sliding window method based on the voltage-charge capacity data; A summarization module is configured to summarize the average slopes corresponding to all windows and record the minimum average slope and its corresponding window information; The feature extraction module is configured to extract the corresponding aging features within the cycle based on the window information, including the window extreme voltage and the accumulated charge capacity.

Images