CN118254640A - End-to-end vehicle battery on-line state evaluation system and method based on large language model - Google Patents

Abstract

The invention relates to an end-to-end vehicle battery on-line state evaluation system and method based on a large language model, and belongs to the field of electric automobiles. According to the invention, an end-to-end processing mode based on a large language model is introduced into a vehicle-mounted battery management system, and in vehicle-mounted battery state estimation and management, complicated and changeable heterogeneous data are dealt with by means of the large language model with strong generalization capability, so that less research on personalized battery management is realized. The general vehicle-mounted battery management system is designed by means of the open-source large language model, monitoring of abnormal states of the battery is achieved, potential faults of the battery are diagnosed and predicted, and the charging and discharging strategies of the battery can be intelligently adjusted according to factors such as the running state, the driving mode and the external environment conditions of the vehicle, so that efficient utilization of energy and maximization of endurance mileage are achieved.

Description

End-to-end vehicle battery on-line state evaluation system and method based on large language model

Technical Field

The invention belongs to the field of electric automobiles, and relates to an end-to-end vehicle battery on-line state evaluation system and method based on a large language model.

Background

A Vehicle Battery Management System (BMS) is a comprehensive system that is specifically used to monitor, control, protect, and optimize the performance of a Vehicle Battery. The method accurately evaluates the battery state by monitoring key parameters (such as voltage, current, temperature and the like) of each battery in the battery pack in real time and applying advanced algorithms and technical means, thereby ensuring safe and efficient operation of the battery, prolonging the service life of the battery and being very critical for realizing the optimal performance of the electric vehicle. In a real-time online battery state assessment scenario, on the one hand, battery state estimation can involve a large amount of complex data and varying environmental factors, which require a powerful generalization capability of the model to capture the inherent changes and laws of battery behavior; on the other hand, personalized factors such as vehicle type, driving habit and the like need to be considered for evaluation, maintenance and the like of the battery. How to perform efficient fusion on multi-source heterogeneous data in a real-time online battery evaluation scene and how to perform personalized battery evaluation, maintenance and maintenance according to data characteristics is a problem to be solved urgently.

In conventional scenarios, battery state estimation relies primarily on extracting features from basic information and state monitoring data (e.g., voltage, current, temperature, and state charge, SOC) of the battery. Because of the influence of factors such as vehicle types, sensor types, driving habits, environmental variability and the like, the model needs to be built through the processes of data acquisition, SOC calculation, data cleaning, data characterization, model building and training, and the traditional method is mainly limited in that the early data preparation process is complex, the data processing and model training need to consume a large amount of time, the delay is too high, the generalization, the universality and the mobility of the model are insufficient, and the real-time online calculation of the battery health state is difficult to realize.

In a real-time online scene, the system needs to timely give a battery state conclusion according to current data. In this scenario, involving a large amount of complex data and varying environmental factors, a powerful generalization capability of the model is required to extract features, capturing the rules of battery behavior. Meanwhile, battery state models of different vehicle types and vehicle conditions can only have tiny differences, most models have similar structures and parameters, reuse and migration of the models are needed to be realized in an upstream task, and repeated training is not needed each time. Therefore, a method which can efficiently process complex data, has strong generalization capability and task universality and can realize personalized requirements is needed to process and fuse multi-source heterogeneous data so as to realize on-line estimation and management of the state of a vehicle-mounted battery and improve the performance and safety of the whole vehicle.

In the application of the conventional on-vehicle battery on-line state estimation method, although various methods such as an equivalent circuit model, an electrochemical model, an empirical model-based, a kalman filtering method, variations thereof, and the like have been proposed and implemented to monitor and manage the state of health and performance of the battery, these methods still have drawbacks and disadvantages, which limit the application efficiency and accuracy thereof in a real-time scenario.

Such as an equivalent circuit model, simulates the electrical behavior of a battery by simplifying the battery into a network of basic circuit elements (e.g., resistors and capacitors) to reflect its charge-discharge characteristics. Although the calculation is simple and convenient, and the method is suitable for real-time application, the simplification degree often ignores complex electrochemical reactions inside the battery, and all battery behaviors cannot be accurately captured, so that the accuracy of state estimation is limited. Electrochemical-based models utilize the theory of chemical reactions inside the cell to describe the electrochemical behavior of the cell by reducing the complexity of the model, aimed at balancing the accuracy and computational efficiency of the model. Although higher accuracy can be provided, the calculation complexity is high, the real-time online requirement cannot be met, and the contradiction between the calculation complexity and the real-time performance is difficult to balance. Meanwhile, these two methods often perform model design and parameter calibration based on specific types or batches of batteries, and when facing batteries of different manufacturers and different aging states, the adaptability and accuracy of these models are significantly reduced. The empirical model is based on a historical data analysis method, and is used for quickly estimating the current state of the battery by establishing a statistical relationship between the measurable parameters of the battery and the state of the battery. Although the accuracy of state estimation can be improved, the methods have higher dependence on data quantity and data quality and have poor robustness. In practical applications, the use environment of the battery is changeable, and the acquired data may have noise, which poses challenges to the accuracy and reliability of the state estimation. While Kalman filtering and its variants are effective in processing system noise, providing optimal state estimates, their performance is highly dependent on accurate process models and assumptions of noise characteristics. In battery state estimation, these assumptions are often difficult to meet accurately, limiting the practical application performance of the filtering algorithm.

The patent application with publication number CN 114171809A proposes a vehicle-mounted battery management system, a vehicle-mounted battery management method, a vehicle-mounted battery management server and a vehicle-mounted storage medium based on a cloud platform for solving the problems that a traditional Battery Management System (BMS) data preparation process is complex and on-line battery health state calculation cannot be applied. The system realizes the real-time acquisition, pretreatment, storage, analysis and monitoring of various electric power data of the vehicle-mounted battery through the terminal module, the data collection pretreatment module, the big data storage module, the data real-time analysis module and the operation management module, analyzes and alarms the battery performance through comparison with standard electric power data, and realizes the capability of real-time perception of the health state of the battery monomer and quick maintenance decision-making of system operation management personnel. However, too many processing flows of the system module are complicated, the complexity is high, the difficulty of maintenance and upgrading is high, the cloud platform and the network connection are highly dependent, if the network connection is unstable or the cloud platform has a problem, the normal operation of the battery management system and the real-time processing of data can be influenced, and the system needs to adapt and adjust for batteries of different types and brands, so that the universality and the adaptability of the system can be influenced.

In order to solve the problem that the degradation degree of the vehicle-mounted battery is difficult to be calculated with high accuracy by the vehicle-mounted battery management device because the degradation degree of the vehicle-mounted battery is sometimes different depending on the difference of the calculation methods, the patent application with publication number CN 115742862A proposes a vehicle-mounted battery management device and a method. The device can calculate the degradation development degree of the vehicle-mounted battery according to the charge and discharge information, and inform a predetermined destination of the calculation result through a communication part. However, the method excessively depends on accurate charge and discharge information, and if errors exist, the calculation accuracy of the degradation development degree can be affected, so that the algorithm robustness is poor.

The patent application with publication number CN 115782689A proposes a vehicle-mounted battery maintenance method, device and management system for solving the problem that the battery SOC cannot be accurately estimated when the SOC correction is not performed for a long time. According to the method, by receiving historical behavior data of the vehicle-mounted battery, whether the battery is in a to-be-maintained state or not is judged, and a corresponding maintenance strategy is generated. The system can remind the user who does not carry out the correction of the vehicle-mounted battery SOC, and ensures that a correction mechanism is triggered in time. However, the method is mainly used for solving the problem that the SOC correction is not performed in a short period, the long-term effect is unknown, the maintenance strategy needs to be adjusted according to different types and brands of vehicle-mounted batteries, the method has limitation, and meanwhile, the method is highly dependent on the accuracy of historical data and is insufficient in robustness.

The patent application with the publication number CN 117341498A provides a vehicle-mounted battery management system, which comprises a battery pack, a charger, a control module and the like, in order to solve the problems of low matching degree of voltage and battery working condition and low charging efficiency in the charging process of an electric vehicle. The system controls the output voltage of the charger through the control module, so that the voltage of the battery pack is adapted to the output voltage, and the battery pack is charged. The method can improve the matching degree of the output voltage and the battery, and further improve the charging efficiency and the performance of the vehicle-mounted battery management system. However, this system has safety problems in that it is necessary to adapt and adjust for different types and brands of batteries, to ensure the matching of the voltage of the battery pack with the output voltage, and to control the connection relationship of the batteries.

Disclosure of Invention

In view of the above, the present invention is directed to an end-to-end vehicle battery on-line state evaluation system and method based on a large language model.

In order to achieve the above purpose, the present invention provides the following technical solutions:

The system comprises a data acquisition module, a data preprocessing module, a large language model reasoning module and a cloud platform integration module;

the data acquisition module collects the running data of the vehicle-mounted battery in real time and provides basic data support for downstream tasks; the data acquisition module consists of a sensor assembly, a data acquisition unit, an interface and a connector; the sensor component is used for monitoring various parameters of the battery in real time; the data acquisition unit DCU is responsible for collecting data transmitted by the sensor and performing analog-digital conversion and filtering processing on the data; the interface and the connector provide physical connection with the battery module, the data preprocessing module and the external component;

The data acquisition module interacts by:

(1) And (3) data transmission: the data acquisition module transmits acquired data to the cloud platform through the wireless communication module, or transmits the data to a local model in an end-to-end mode through an interface and a connector for further analysis and processing by other modules;

(2) And (3) receiving an instruction: the data acquisition module receives control instructions from other modules;

(3) Data synchronization: under the condition that a plurality of data acquisition modules work cooperatively, other modules coordinate data synchronization among the modules, so that the consistency and the integrity of data are ensured;

(4) And (5) updating the state: the data acquisition module reports the state of the data acquisition module to other modules regularly;

The data preprocessing module processes and analyzes the data transmitted by the data acquisition module for downstream tasks; the data required by local end-to-end calculation is input into a local large language model, and the data comprises real-time battery state estimation and health monitoring data, fault diagnosis and early warning data, emergency protection control data, performance optimization and real-time management data, and the data are preprocessed, wherein the data comprise the following operations:

and (3) filtering: in order to remove noise and abnormal values in the voltage, current and temperature data, a low-pass filter is adopted for filtering treatment so as to remove sensor noise and instantaneous fluctuation;

normalization: normalizing the data of different dimensions to within the same magnitude and range;

Threshold detection: for the basic protection function, safety thresholds of voltage, current and temperature are required to be set, and whether the condition exceeding the threshold occurs is detected in real time so as to trigger corresponding protection measures;

the data preprocessing module interacts by:

(1) And (3) a data interface: defining a standard data interface, and defining a data format and a transmission protocol to ensure that the data is transmitted to other modules;

(2) Real-time data flow: the preprocessed data is transmitted to other modules in real time in a real-time data stream mode so as to support real-time calculation and decision making;

(3) Format conversion: converting the data into a desired format for reading and processing;

(4) Feedback mechanism: other modules can provide feedback to the data preprocessing module according to the analysis result to guide the data preprocessing module to optimize the preprocessing strategy and parameter setting;

(5) Exception handling: when other modules detect data abnormality or analysis failure, notifying a data preprocessing module to reprocess or collect new data;

the large language model reasoning module is deployed locally and consists of the following parts:

Model loading part: automatically downloading the latest pre-trained large language model from the cloud to be loaded into the memory so as to perform real-time reasoning;

a data input section: the data preprocessing module is responsible for receiving data input by the data preprocessing module, extracting the data according to a specified data format, combining the data with the promt, and converting the data into a format suitable for model input for use by a large language model;

an inference calculation section: carrying out reasoning calculation on the input data by using the loaded large language model to obtain a battery state reasoning result;

a result output section: analyzing the result, and analyzing the reasoning result of the model into information with actual physical meaning; the result is packaged, and the analyzed result is packaged into a standard format, so that the use and analysis are convenient; the result transmission or display, the reasoning result is transmitted to the downstream module in time through the message queue or the event triggering mechanism;

the large language model reasoning module interacts by:

(1) Result transfer: transmitting the battery state prediction result obtained by reasoning to other modules;

(2) Feedback reception: receiving feedback information from other modules;

(3) Command response: responding to commands from other modules;

The cloud platform integration module is responsible for connecting and interacting the vehicle-mounted battery management system with a cloud platform of a vehicle provider, receiving an output result of the large language model reasoning module, realizing data backup and synchronization, remote diagnosis and management, model updating and other vehicle-mounted software updating, and receiving feedback information of a user on maintenance and maintenance advice; the cloud platform integration module consists of the following parts:

Communication interface: the system is responsible for data transmission and communication between the BMS and the cloud platform;

Data buffer: the cloud platform is used for temporarily storing data to be uploaded to the cloud platform, and receiving instructions and updates from the cloud platform;

data encryption and security: the encryption transmission and the security authentication of the data are realized, so that the security and the privacy of the data in the transmission process are ensured;

model update management: the method is responsible for downloading the latest large language model from the cloud platform and carrying out replacement and update locally;

system feedback and diagnostics: collecting the running state and performance index of the system, and uploading the running state and performance index to a cloud platform for monitoring and analysis;

The system self-diagnosis function is realized, and diagnosis results and abnormal information are fed back to the cloud platform so as to facilitate remote maintenance and fault investigation;

the cloud platform integration module interacts by:

(1) And (3) data exchange: uploading the result of the large language model reasoning module to a cloud platform for backup and analysis, and issuing suggestions and schemes of the cloud platform to a vehicle-mounted management system;

(2) Model update notification: when the cloud platform has new large language model update, the cloud platform integration module is responsible for downloading and notifying a downstream large language model reasoning module to perform model replacement and update;

(3) Configuration update and synchronization: the cloud platform integration module receives a configuration update instruction from the cloud platform and synchronizes the configuration update instruction to a related downstream module;

(4) Fault diagnosis and feedback: the fault or abnormal information found by the downstream module is uploaded to a cloud platform through a cloud platform integration module, the cloud platform provides corresponding processing suggestions or repairing schemes according to analysis results, and the cloud platform integration module issues the corresponding processing suggestions or repairing schemes to the downstream module for execution;

(5) Remote control and management: the cloud platform integration module receives a remote control instruction from the cloud platform and transmits the instruction to the corresponding downstream module for execution.

Further, the sensor assembly includes a voltage sensor, a current sensor, and a temperature sensor;

the data acquisition module acquires the following six types of data:

Voltage data: the voltage value of each battery unit is used for monitoring the charge state and the health condition of the battery and is acquired by a voltage sensor;

Current data: the current value of the battery during charging and discharging is used for calculating the energy inflow and outflow of the battery and is collected by a current sensor;

temperature data: the temperature of the battery unit and the module is used for preventing overheat and evaluating the thermal management effect of the battery, and is acquired by a temperature sensor;

Charge-discharge state data SOC: the charging level of the battery is used for estimating the remaining driving mileage and managing the charging and discharging process, and is collected by connecting a data line with the BMS system;

health status data SOH: the health condition of the battery is used for predicting the service life and performance degradation of the battery and is collected by connecting a data line with a BMS system;

Charge-discharge cycle times: the charge and discharge times of the battery are used for evaluating the service condition and service life of the battery, and the battery is collected by connecting a data line with the BMS system.

The end-to-end vehicle battery online state evaluation method based on the online state evaluation system comprises the following steps:

s1: monitoring the vehicle-mounted battery in real time, and collecting data of the vehicle-mounted battery;

Calibrating the sensor;

Setting sampling frequency, and setting data acquisition frequency according to the working state of the battery and data analysis requirements;

The format and the structure of the data are defined, so that the data transmission and the processing are convenient;

collecting data in a local storage mode, and regularly backing up the data to a cloud platform;

setting an abnormal threshold value, and immediately executing alarming and protecting measures if abnormal data are detected;

s2: preprocessing and converting the acquired vehicle-mounted battery data;

Initializing a module when starting;

Removing abnormal data by using an abnormal value detection method, and filling the missing data;

smoothing the data by applying a low-pass filter to reduce instantaneous fluctuation;

normalizing the data to between 0 and 1 using a min-max normalization method;

extracting statistical characteristics, and carrying out frequency domain characteristic extraction signs;

packaging the preprocessed data into JSON or CSV format, so as to facilitate analysis and processing of downstream modules;

the preprocessed data is sent to a downstream module in real time through a data bus;

S3: selecting a pre-trained model for loading, and automatically initializing model configuration; the method comprises the steps of receiving battery data transmitted by an upstream data preprocessing module through a data interface, and combining the received data with a preset promt, wherein the flow is as follows:

S31: formatting the received battery data into a form suitable for combination with a promt;

S32: constructing a Prompt template for embedding the formatted data into the Prompt; the template is designed according to the requirements of a large language model and a specific application scene;

S33: before embedding the data into the Prompt, processing special characters possibly existing in the data to avoid interference to the input of the large language model; mapping the numerical value and the characteristics of the data to vocabulary with corresponding semantics;

S34: embedding the formatted data into a Prompt template, and filling sensor data into corresponding placeholders to form a complete input text;

s35: taking the constructed promt as the input of a large language model to predict or analyze; the large language model will generate a corresponding output from the battery data provided in the Prompt;

Inputting the data embedded into the Prompt into a pre-trained large language model for reasoning and analysis;

Analyzing and processing the original result obtained by reasoning, and extracting the required information; post-processing the results according to the requirements, outputting the processed results in a standard format, displaying the processed results in a user interaction interface, and uploading the processed results to a cloud platform regularly through a communication interface;

s4: the communication and cooperative work between the vehicle-mounted battery management system and the cloud platform are realized, and the functions of instruction receiving, model updating, system updating, feedback reporting, maintenance and monitoring are supported;

Receiving an instruction and configuration update from a cloud platform, and monitoring the real-time instruction and configuration update from the cloud platform by using a WebSocket or long polling mechanism;

according to the collected real-time data and feedback information, a software provider learns the Llama model in a cloud platform; when the performance of the model is reduced or new data appear, performing thermal updating of the model;

After receiving the system updating instruction, downloading the latest model file from the cloud platform through HTTP or FTP protocol, executing the system updating flow, replacing the local old model file, and simultaneously notifying the downstream module to reload the model; updating the model and the driving software of the system regularly;

The method comprises the steps of regularly collecting running logs, performance indexes and fault diagnosis results of a vehicle-mounted battery management system, packaging the running logs, the performance indexes and the fault diagnosis results into a JSON format, and uploading the JSON format to a log analysis endpoint of a cloud platform for monitoring and analysis by operation and maintenance personnel; collecting and collecting execution conditions and feedback information of maintenance and maintenance suggestions of a user;

periodically checking the stability of network connection and the safety of data transmission; periodically checking the connection state with the cloud platform, monitoring the stability and safety of data transmission, and executing necessary maintenance operation;

and generating corresponding maintenance suggestions and maintenance plans according to the data uploaded by the vehicle-mounted battery management system.

Further, the sensor includes a voltage sensor, a current sensor, and a temperature sensor.

The invention has the beneficial effects that:

1. Different from CN 113884927A

1) Large language model applications. According to the invention, a local deployment large language model (for example Grok-1) is adopted to directly complete real-time assessment and intelligent management of the battery state in the vehicle-mounted system, and an external cloud platform is not required to be relied on. And CN113884927A focuses on cloud big data processing, realizes cloud and vehicle-mounted data interaction by establishing a battery life model and combining OTA and other means, and focuses on cloud data processing capacity.

2) Real-time and independence. The invention emphasizes the real-time performance and reliability of the system, reduces the dependence on network connection, and directly completes various tasks in the vehicle-mounted system in an end-to-end processing mode. CN113884927a relies on cloud computing and data processing in battery life management, subject to network condition limitations.

3) Generalization ability of large language models. According to the invention, by introducing a large language model, the processing capacity of the model on complex and changeable data and the accuracy of personalized battery management are improved. The CN113884927a processes mass data mainly through a cloud battery management system, and it is not explicitly proposed to use a model with strong generalization capability to perform data processing and state evaluation.

2. Different from CN110416636A

1) The system architecture and technical routes are different. The invention introduces a large language model technology (such as Grok-1), adopts an end-to-end system architecture, all modules (data acquisition, data preprocessing, large language model reasoning and the like) are completed in the vehicle-mounted system, and the battery state is directly evaluated and managed in real time locally to form a closed loop. By the end-to-end processing mode, no external cloud platform is needed for participation from data acquisition to state evaluation, so that the time of data transmission and processing is reduced, delay caused by network fluctuation is also reduced, the high efficiency of a processing flow and the safety of data are ensured, and the instant evaluation and management of the battery state are realized. CN110416636a relies more on traditional battery management strategies and algorithms, such as equivalent circuit models, rule-based algorithms, etc., than complex data processing capabilities that make use of large language models; in terms of data processing, the cloud end and the central server process and analyze data, the data is required to be sent to the cloud end for processing through a network, but the cloud end is not required for evaluating and managing the battery state in an end-to-end mode, so that the time delay of data processing is increased, and when the network is unstable or is interrupted, the instantaneity and the reliability of the system are affected.

2) And (5) personalized management. The invention utilizes the powerful generalization capability of the large language model, can process and analyze complex battery data more accurately, and performs personalized battery management according to factors such as vehicle running state, driving mode, external environment condition and the like, thereby improving the generalization capability of the system and the precision of personalized management. CN110416636a fails to describe in detail how to manage and optimize battery status for individual differences, data processing capability may be limited by the performance of traditional algorithms and models, it is difficult to efficiently process large-scale, multi-source heterogeneous battery data, and capturing and analyzing complex battery behavior may not be as accurate as large language model-based methods.

The real-time online battery state monitoring and processing is realized by an end-to-end large language model, and the cloud platform does not participate in the work of the real-time online battery state analysis emphasized by the invention. The cloud platform is used for periodically receiving operation data of the vehicle-mounted system, such as operation results of the large language model are uploaded to the cloud platform every other day, and only a manufacturer knows operation conditions of a customer, and continuous services, such as vehicle maintenance suggestions, feedback of the customer to the system and the like, are provided for the customer.

The real-time performance and independence are improved: according to the invention, through localization large language model reasoning, the real-time performance of battery state evaluation and management is greatly improved, meanwhile, the dependence on external networks and cloud services is reduced, and the reliability and independence of the system are improved.

Enhancing generalization and individualization management capabilities: by utilizing the high generalization capability of the large language model, the multi-source heterogeneous data can be effectively processed, accurate battery state evaluation and personalized management are realized, the service efficiency and service life of the battery are further improved, and the comprehensive performance of the electric automobile is optimized.

Complexity and cost reduction: compared with a management system relying on cloud computing, the method reduces the complexity of data transmission and processing through an end-to-end processing mode, reduces the requirements on high-performance cloud computing resources, and simplifies the system architecture, thereby reducing the operation and maintenance cost.

In summary, the invention introduces an end-to-end processing mode based on a large language model in the vehicle-mounted battery management system, and in the vehicle-mounted battery state estimation and management, the large language model with strong generalization capability is used for coping with complex and changeable heterogeneous data, so that the research on realizing personalized battery management is less. Therefore, we propose a vehicle-mounted battery on-line evaluation and management system based on a large language model, through the design of a general vehicle-mounted battery management system by means of an open source large language model, monitoring of abnormal states of the battery, such as over-temperature, over-voltage, over-current and the like, diagnosis and prediction of potential faults of the battery are realized, and charge and discharge strategies of the battery can be intelligently adjusted according to factors such as running states, driving modes and external environment conditions of the vehicle, so that efficient utilization of energy and maximization of endurance mileage are realized.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention. The objects and other advantages of the invention may be realized and obtained by means of the instrumentalities and combinations particularly pointed out in the specification.

Drawings

For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in the following preferred detail with reference to the accompanying drawings, in which:

FIG. 1 is a diagram of a system architecture of the present invention;

FIG. 2 is a flow chart of the method of the present invention.

Detailed Description

Other advantages and effects of the present invention will become apparent to those skilled in the art from the following disclosure, which describes the embodiments of the present invention with reference to specific examples. The invention may be practiced or carried out in other embodiments that depart from the specific details, and the details of the present description may be modified or varied from the spirit and scope of the present invention. It should be noted that the illustrations provided in the following embodiments merely illustrate the basic idea of the present invention by way of illustration, and the following embodiments and features in the embodiments may be combined with each other without conflict.

Wherein the drawings are for illustrative purposes only and are shown in schematic, non-physical, and not intended to limit the invention; for the purpose of better illustrating embodiments of the invention, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the size of the actual product; it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.

The same or similar reference numbers in the drawings of embodiments of the invention correspond to the same or similar components; in the description of the present invention, it should be understood that, if there are terms such as "upper", "lower", "left", "right", "front", "rear", etc., that indicate an azimuth or a positional relationship based on the azimuth or the positional relationship shown in the drawings, it is only for convenience of describing the present invention and simplifying the description, but not for indicating or suggesting that the referred device or element must have a specific azimuth, be constructed and operated in a specific azimuth, so that the terms describing the positional relationship in the drawings are merely for exemplary illustration and should not be construed as limiting the present invention, and that the specific meaning of the above terms may be understood by those of ordinary skill in the art according to the specific circumstances.

As shown in fig. 1, the invention provides an end-to-end vehicle battery on-line state evaluation system and method based on a large language model, which realizes real-time state evaluation and intelligent management of a vehicle battery through a locally deployed large language model (such as Grok-1) without depending on a cloud platform, and solves the problems of accurate evaluation of battery state, diagnosis of potential faults, adjustment of charge and discharge strategies and the like in a real-time on-line scene, thereby improving the service efficiency and service life of the battery. The system directly completes tasks such as data acquisition, preprocessing, state estimation, protection measures, balanced control and the like in the vehicle-mounted system in an end-to-end processing mode so as to improve instantaneity and reliability and reduce dependence on network connection. The system comprises a data acquisition module, a data preprocessing module, a large language model reasoning module, a cloud platform integration module and the like.

1. System composition

Data acquisition module

The data acquisition module is used for collecting the running data of the vehicle-mounted battery in real time and providing basic data support for downstream tasks. The module mainly comprises three parts, namely a sensor assembly, a data acquisition unit, an interface, a connector and the like. Wherein the sensor assembly: the system comprises a voltage sensor, a current sensor, a temperature sensor and the like, and is used for monitoring various parameters of the battery in real time; data acquisition unit (DCU): and is responsible for collecting data transmitted by the sensor and performing primary processing, such as analog-to-digital conversion, filtering and the like, on the data. Interface and connector: physical connections with the battery module, the data preprocessing module, and other system components are provided to ensure smooth transmission of data.

The data acquisition module mainly acquires the following six types of data:

Voltage data: the voltage value of each battery unit is used for monitoring the charge state and the health condition of the battery and is acquired by a voltage sensor; current data: the current value of the battery during charging and discharging is used for calculating the energy inflow and outflow of the battery and is collected by a current sensor; temperature data: the temperature of the battery unit and the module is used for preventing overheat and evaluating the thermal management effect of the battery, and is acquired by a temperature sensor; charge-discharge state data (SOC): the charging level of the battery is used for estimating the remaining driving mileage and managing the charging and discharging process, and is collected by connecting a data line with the BMS system; health status data (SOH): the health condition of the battery is used for predicting the service life and performance degradation of the battery and is collected by connecting a data line with a BMS system; charge-discharge cycle times: the charge and discharge times of the battery are used for evaluating the service condition and service life of the battery, and the battery is collected by connecting a data line with the BMS system.

The interaction between the data acquisition module and other modules is mainly performed in the following four modes:

1) And (3) data transmission: the data acquisition module transmits the acquired data to the cloud platform through the wireless communication module, or transmits the data to a locally deployed model through an interface and a connector in an end-to-end mode for further analysis and processing by other modules.

2) And (3) receiving an instruction: the data acquisition module may receive control instructions from other modules, such as adjusting acquisition frequency, activating specific sensors, etc.

3) Data synchronization: under the condition that a plurality of data acquisition modules work cooperatively, other modules can coordinate the data synchronization among the modules, and the consistency and the integrity of the data are ensured.

4) And (5) updating the state: the data acquisition module can periodically report its own status, such as battery power, communication signal strength, etc., to other modules so that the other modules monitor the working condition of the data acquisition module and perform necessary maintenance.

(II) data preprocessing module

The data preprocessing module is used for processing and analyzing the data transmitted by the data acquisition module for downstream tasks. The data required by local end-to-end calculation is mainly input into a local large language model, and relates to real-time battery state estimation and health monitoring, fault diagnosis and early warning, emergency protection control, performance optimization, real-time management and the like, so that low delay, high real-time performance and accuracy are required for preprocessing the data. The data preprocessing of this part involves several operations:

And (3) filtering: in order to remove noise and abnormal values in voltage, current and temperature data, the invention adopts a low-pass filter (such as a Kalman filter or a moving average filter) to carry out filtering treatment so as to remove sensor noise and instantaneous fluctuation. For example, a moving average filter may be used to smooth continuous temperature data, reducing the impact of abnormal fluctuations on system decisions.

Normalization: the data of the different dimensions are normalized to the same magnitude and range to facilitate subsequent comparison and calculation. Normalization may be performed using min-max normalization, Z-score normalization, etc.

Threshold detection: for the basic protection function, safety thresholds of voltage, current and temperature need to be set, and whether the condition exceeding the threshold occurs is detected in real time so as to trigger corresponding protection measures.

Interaction with other modules

Interaction of the data preprocessing module with other modules (such as a large language model evaluation module and the like) is mainly performed by the following modes:

and (3) a data interface: a standard data interface is defined, specifying the format and transmission protocol of the data to ensure that the data can be successfully transmitted to other modules.

Real-time data flow: and transmitting the preprocessed data to other modules in real time in a real-time data stream mode so as to support real-time calculation and decision.

Format conversion: the data is converted into a corresponding format (such as CSV, JSON, etc.) according to the requirements of other modules, so that the data can be read and processed conveniently.

Feedback mechanism: other modules can provide feedback to the data preprocessing module according to the analysis result, and guide the data preprocessing module to optimize the preprocessing strategy and parameter setting.

Exception handling: when other modules detect data anomalies or failed analysis, the data preprocessing module can be informed to reprocess or collect new data.

(III) big language model reasoning module

The large language model reasoning module is a core part responsible for intelligent state analysis and management of the battery in the vehicle-mounted battery management system, the large language model is deployed locally, so that the time of data transmission and processing can be reduced, the real-time response capability of a protection control function is enhanced, and the real-time dynamic balanced management, the timely adjustment of a charging and discharging strategy, the real-time battery state estimation, the emergency protection control and the like of the battery pack are realized. The module mainly comprises the following parts:

Model loading section

And automatically downloading the latest pre-trained large language model from the cloud to load the latest pre-trained large language model into the memory so as to perform real-time reasoning.

Data input section

And the data preprocessing module is responsible for receiving the data input by the data preprocessing module, extracting the data according to a specified data format, combining the data with the promt, and converting the data into a format suitable for model input for use by a large language model.

Inference calculation section

And carrying out inference calculation on the input data by using the loaded large language model to obtain an inference result of the battery state.

Result output section

First, the result is analyzed, and the reasoning result of the model is analyzed into information with practical meaning, such as the judgment of converting the probability value output by the model into the battery health state. And secondly, encapsulating the result, namely encapsulating the parsed result into a standard format, so that the use and the parsing of other modules are facilitated, for example, the result is encapsulated into a JSON or XML format. And finally, transmitting or displaying the result, and timely transmitting the reasoning result to a downstream module, such as a protection control module, an equalization management module and the like, through a message queue or an event triggering mechanism.

Interaction with other modules

The interaction of the large language model reasoning module and other modules is mainly performed in the following way:

result transfer: and transmitting the battery state prediction result obtained by reasoning to other modules, such as a protection control module, an equalization management module and the like, so as to carry out corresponding processing and decision.

Feedback reception: feedback information from other modules is received, such as actual execution, system state changes, etc., for optimizing and adjusting the model.

Command response: in response to commands from other modules, such as model updates, parameter adjustments, etc., to maintain model accuracy and adaptability.

(IV) cloud platform integration module

The cloud platform integration module is mainly responsible for connecting and interacting the vehicle-mounted battery management system with a cloud platform of a vehicle provider, is a communication bridge between a user and the vehicle provider, receives an output result of the large language model reasoning module, so as to realize functions of data backup and synchronization, remote diagnosis and management, model updating, other vehicle-mounted software updating and the like, provides data basis, suggestion strategy and the like for battery maintenance and maintenance for vehicle owners and maintenance personnel, and also receives feedback information of maintenance and maintenance suggestions of the user. The module mainly comprises the following parts:

Communication interface: and the vehicle-mounted BMS is responsible for data transmission and communication between the vehicle-mounted BMS and the cloud platform, and comprises data uploading, instruction receiving and the like. Multiple communication protocols, such as HTTP, MQTT, webSocket, etc., are supported to accommodate different cloud platforms and network environments.

Data buffer: the cloud platform is used for temporarily storing data to be uploaded to the cloud platform and receiving instructions and updates from the cloud platform. A data caching mechanism is provided to ensure that data integrity and consistency is maintained in the event of network instability.

Data encryption and security: the encryption transmission and the security authentication of the data are realized, so that the security and the privacy of the data in the transmission process are ensured. And supporting security mechanisms such as SSL/TLS encryption, OAuth authentication and the like.

Model update management: is responsible for downloading the latest large language model from the cloud platform and carrying out replacement and update locally. Model version management and verification mechanisms are provided to ensure the correctness and validity of model updates.

System feedback and diagnostics: and collecting the running state and performance index of the system, and uploading the running state and performance index to a cloud platform for monitoring and analysis.

The system self-diagnosis function is realized, and the diagnosis result and the abnormal information are fed back to the cloud platform, so that remote maintenance and fault investigation are facilitated.

Interaction with other modules

Interaction of the cloud platform integration module with other modules (such as a large language model reasoning module, a protection control module and the like) is mainly performed in the following manner

And (3) data exchange: and uploading the result of the large language model reasoning module to a cloud platform for backup and analysis, and issuing suggestions and schemes of the cloud platform to a vehicle-mounted management system.

Model update notification: when the cloud platform has new large language model update, the cloud platform integration module is responsible for downloading and informing the downstream large language model reasoning module to perform model replacement and update.

Configuration update and synchronization: the cloud platform integration module receives the configuration update instruction from the cloud platform and synchronizes the configuration update instruction to the related downstream module, such as updating protection control parameters, adjusting balance management strategies, and the like.

Fault diagnosis and feedback: the fault or abnormal information found by the downstream module is uploaded to the cloud platform through the cloud platform integration module, the cloud platform provides corresponding processing suggestions or repairing schemes according to analysis results, and the cloud platform integration module issues the corresponding processing suggestions or repairing schemes to the downstream module to execute the processing suggestions or repairing schemes.

Remote control and management: the cloud platform integration module receives remote control instructions from the cloud platform, such as start/stop detection, adjustment of working modes and the like, and transmits the instructions to the corresponding downstream modules for execution.

2. Workflow process

As shown in fig. 2, the specific method of the present invention is as follows:

Data acquisition module

Sensor calibration: and calibrating the voltage, current, temperature and other sensors to ensure the accuracy of the data.

Sampling frequency setting: according to the working state of the battery and the data analysis requirement, reasonable data acquisition frequency is set, such as acquisition is carried out every 2 seconds in the normal working state, acquisition is carried out every second in the charging and discharging process, and the like.

Data format definition: the format and structure of the data, such as acquisition time stamps, voltage values, current values, temperature values, etc., are defined to facilitate data transmission and processing.

Data storage and backup: the data is stored and collected locally for the need of time and time, and the data is backed up to the cloud platform periodically, so that the data is prevented from being lost.

The exception handling flow is as follows: an abnormality threshold such as an upper voltage limit, an upper temperature limit, etc. is set, and an abnormality processing flow is written, and once abnormal data is detected, alarm and protection measures are immediately performed.

Through the specific operation, the data acquisition module can effectively realize real-time monitoring and data acquisition of the vehicle-mounted battery, and support is provided for efficient operation and intelligent decision of the battery management system.

(II) data preprocessing module

Initializing: and when the system is started, the module is initialized, including loading configuration parameters, initializing a communication interface with a sensor, receiving data in real time and the like.

Data cleaning: abnormal data is removed using an abnormal value detection method (such as a box graph method). The missing data is filled in, e.g., using an average of neighboring data.

And (3) filtering: a low pass filter (e.g., a moving average filter) is applied to smooth the data to reduce transient fluctuations.

Normalization: data were normalized to between 0 and 1 using a min-max normalization method.

Feature extraction: statistical features such as mean, variance, maximum, minimum, etc. are extracted. Frequency domain feature extraction is performed as needed, such as by applying fourier transforms to extract frequency features.

And (3) data packaging: and packaging the preprocessed data into a JSON or CSV format, so that the downstream module can analyze and process conveniently.

And (3) data transmission: and sending the preprocessed data to the downstream module in real time through a data bus.

Through the specific operation, the data preprocessing module can effectively process and convert the original data, and high-quality data support is provided for accurate estimation and intelligent management of the battery state.

(III) big language model reasoning module

Model loading and initializing: and selecting a pre-trained model to load, such as a Grok-1 large language model after distillation. At the same time, the system automatically initializes the model-related configuration.

Data reception and conversion: and the battery data such as voltage, current, temperature and the like transmitted by the upstream data preprocessing module are received through the data interface. Combining the received data with a preset promt, wherein the flow is as follows:

1) Data formatting

First, the received battery data (e.g., voltage, current, temperature, etc.) is formatted into a form suitable for combination with the promt. One common approach is to convert the data into the form of key-value pairs, such as:

2) Construction of Prompt templates

Next, a template of promt is constructed for embedding the formatted data into promt. This template should be designed according to the requirements of the large language model and the specific application scenario, the promt of the present invention is given below:

Battery state estimation promt:

″Given the battery data：voltage：{voltage}V,current：{current}A,temperature：{temperature}℃,predict the health status of the battery.″

Real-time emphasis: the need for "real-time monitoring" and "real-time reporting" is explicitly indicated in Prompt to emphasize the real-time processing requirements for data.

″Monitor the battery data in real-time：voltage：{voltage}V,current：{current}A,temperature：{temperature}℃.Report any abnormal readings.″

Safety threshold setting: safety thresholds (such as voltage threshold, current threshold, temperature threshold) are introduced in the basic protection promt, and a protection mechanism is required to be triggered when the threshold is exceeded so as to ensure the safe operation of the battery.

″Check the battery data for safety：voltage：{voltage}V,current：{current}A,temperature：{temperature}℃.Trigger protection mechanisms if voltage exceeds{voltage_threshold}V,current exceeds{current_threshold}A,or temperature exceeds{temperature_threshold}℃.″

Equalization control refinement: in the balance control Prompt, the voltage of each battery unit is monitored, and a voltage balance threshold is set to guide the adjustment of balanced charge and discharge, so as to maintain the voltage consistency among the battery units.

″Balance the battery cells based on the data：cell_1_voltage：{cell_1_voltage}V,cell_2_voltage：{cell_2_voltage}V,…,cell_n_voltage：{cell_n_voltage}V.Adjust charging/discharging to maintain voltage balance within{balance_threshold}V.″

And (3) intelligent protection: in Prompt, a large language model is explicitly required to provide specific action suggestions such as "protection measures", "balance actions", "optimization adjustment" and the like so as to guide actual operations.

″Protection control：Given the battery data of voltage：{voltage}V,current：{current}A,temperature：{temperature}℃,determine if any protective actions are required.″

Fault diagnosis and early warning: in the Prompt of fault diagnosis, the identification and early warning of potential faults are emphasized so as to improve the safety and reliability of the system.

″Fault diagnosis：Analyze the real-time data：voltage：{voltage}V,current：{current}A,temperature：{temperature}℃.Identify any potential faults and provide early warnings.″

3) Semantic mapping and special character processing

Special characters that may be present in the data need to be processed before embedding the data in the Prompt to avoid interfering with the input of the large language model. For example, characters such as line breaks, quotation marks, etc. need to be escape or deleted. While to enhance comprehensibility of the Prompt, the values and features of the data may be mapped to words with corresponding semantics. For example, the voltage values are mapped to the levels "low", "normal", "high", etc.

4) Data embedding

The formatted data is embedded into a template of the Prompt, and the voltage, current and temperature data are filled into the corresponding placeholders to form a complete input text. For example, using the data and templates of the battery state estimates in the above example, the resulting input text is:

″Given the battery data：voltage：3.7V,current：1.5A,temperature：25℃,predict the health status of the battery.″

5) Inputting large language model

And finally, taking the constructed promt as the input of a large language model to predict or analyze. The large language model will generate a corresponding output from the battery data provided in the Prompt.

Inference calculations

And inputting the data embedded into the Prompt into a pre-trained large language model for reasoning and analysis.

Result processing and output: and analyzing and processing the original result obtained by reasoning, and extracting required information such as battery health state, battery balance strategy, intelligent protection, diagnosis, early warning and the like. Post-processing results, such as threshold judgment, probability conversion, etc., is performed as needed. And outputting the processed result in a standard format, for example, packaging the processed result into a JSON format, and then displaying the processed result in a user interaction interface and uploading the processed result to a cloud platform periodically through a communication interface.

(IV) cloud platform integration module

And (3) receiving an instruction: instructions and configuration updates, such as model update instructions, parameter adjustment instructions, etc., are received from the cloud platform. And monitoring real-time instructions and configuration updates from the cloud platform by using a WebSocket or long polling mechanism, so as to ensure timely response.

Model updating: and according to the collected real-time data and feedback information, the software provider learns and fine-tunes the Llama model in the cloud platform. When the performance of the model is reduced or new data appear, the thermal update of the model is carried out, and the accuracy and the effectiveness of the model are ensured.

Updating a system: after receiving the system updating instruction, downloading the latest model file from the cloud platform through HTTP or FTP protocol, and executing the system upgrading flow, including decompression, verification, installation, restarting and other steps, replacing the local old model file, and simultaneously notifying the downstream module to reload the model, so as to ensure the integrity and safety of the system function. And periodically updating software such as a model, a driver and the like of the system to adapt to new battery technology and management requirements.

Feedback reporting: and (3) regularly collecting running logs, performance indexes and fault diagnosis results of the vehicle-mounted battery management system, packaging the running logs, the performance indexes and the fault diagnosis results into a JSON format, and uploading the JSON format to a log analysis endpoint of the cloud platform for monitoring and analysis by operation and maintenance personnel. And meanwhile, collecting and collecting the execution condition and feedback information of the user on maintenance and maintenance advice.

Diagnosis and monitoring: the stability of network connection and the security of data transmission are checked regularly, for example, the connection state is monitored by using a heartbeat mechanism, and the security of data transmission is ensured by using SSL/TLS encryption. And periodically checking the connection state with the cloud platform, monitoring the stability and safety of data transmission, and executing necessary maintenance operation.

Maintenance recommendation generation: corresponding maintenance suggestions and maintenance plans, such as charging strategy adjustment, temperature control and the like, are generated according to relevant data, such as battery health status and the like, uploaded by the vehicle-mounted battery management system.

Through the specific operation, the cloud platform integrated module can realize high-efficiency communication and cooperative work between the vehicle-mounted battery management system and the cloud platform, support functions of instruction receiving, model updating, system updating, feedback reporting, maintenance and monitoring and the like, and improve the intelligent level and operation and maintenance efficiency of the system.

3. The specific application scene is as follows: personalized battery management in new energy automobile city daily commute

1) Scene background

The invention relates to a new energy automobile driven by Mr. Li, which is provided with an end-to-end vehicle battery on-line state evaluation and management system based on a large language model. He is mainly performing daily commute in cities, and challenges faced include frequent start-stop, changeable road conditions and the influence of seasonal temperature changes on battery performance. The system aims to optimize the driving experience of Mr. prune, prolong the service life of the battery, improve the energy efficiency and ensure the high-efficiency operation of the battery and the safety of the vehicle in the journey through a real-time and personalized battery management strategy.

2) System operation process

And a data acquisition module: as the car starts, the system automatically activates. The data acquisition module immediately starts working and collects various key parameters, such as voltage, current, temperature and the like, of the battery in real time. Meanwhile, a charge-discharge State (SOC), a health State (SOH), and the number of charge-discharge cycles were recorded.

And a data preprocessing module: the collected data is subjected to filtering, normalization, threshold detection and other processes through a data preprocessing module, noise and abnormal values are removed, and the data is converted into a format acceptable by a model.

Large language model reasoning module: the preprocessed data is combined with a preset promt to generate a large language model (such as Grok-1) with a final input format to be input into the vehicle-mounted system. The large language model comprehensively considers the driving habit of mr. Prune, the historical battery usage mode and the current environmental condition and analyzes the battery state. If the large language model identifies the possible negative influence of the low temperature in the current city on the battery performance, the charging and discharging strategy is adjusted according to the driving habit (such as frequent start and stop) of Mr. prune, and the battery heating program is started to maintain the battery in an optimal state, so that the efficient utilization of energy and the maximization of the endurance mileage are ensured. When the system detects that the preset SOC lower limit is about to be reached, the system reminds the Mr. prune to charge in time, recommends a nearby charging station, and simultaneously provides a charging suggestion according to the current state of the battery and the predicted use requirement so as to maximize the charging efficiency and the service life of the battery.

Cloud platform integration module: the system periodically uploads the reasoning result and the system operation data of the large language model to the cloud platform so as to carry out deep analysis, historical data comparison or receive suggestions of remote experts. Through OTA updating, the large language model timely receives the latest model version and strategy updating from the cloud platform, and continuous optimization of the battery management system is ensured.

Through the real-time state monitoring and intelligent management of the system, the new energy automobile of Mr. prune is excellent in daily commute of cities, the battery efficiency is remarkably improved, and meanwhile, the service life of the battery is prolonged due to fine management. Even under adverse conditions such as low temperature in winter, the system can automatically adjust strategies, so that the battery performance is not influenced, and the driving safety and the driving comfort are ensured. And the cloud platform can provide suggestions for future use and maintenance for the vehicle owners by regularly receiving operation results from the system.

The embodiment shows the application scene of the end-to-end vehicle-mounted battery on-line state evaluation and management system based on the large language model in personalized battery management in daily commute of new energy automobiles, and realizes high-efficiency management and fault prevention of the battery through real-time data processing and intelligent reasoning of the vehicle-mounted large language model, improves the driving safety and convenience, reduces the dependence on an external network, and ensures the stability and reliability of the system.

Finally, it is noted that the above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the present invention, which is intended to be covered by the claims of the present invention.

Claims (4)

1. The end-to-end vehicle battery on-line state evaluation system based on the large language model is characterized in that: the system comprises a data acquisition module, a data preprocessing module, a large language model reasoning module and a cloud platform integration module;

the data acquisition module collects the running data of the vehicle-mounted battery in real time and provides basic data support for downstream tasks; the data acquisition module consists of a sensor assembly, a data acquisition unit, an interface and a connector; the sensor component is used for monitoring various parameters of the battery in real time; the data acquisition unit DCU is responsible for collecting data transmitted by the sensor and performing analog-digital conversion and filtering processing on the data; the interface and the connector provide physical connection with the battery module, the data preprocessing module and the external component;

The data acquisition module interacts by:

(1) And (3) data transmission: the data acquisition module transmits acquired data to the cloud platform through the wireless communication module, or transmits the data to a local model in an end-to-end mode through an interface and a connector for further analysis and processing by other modules;

(2) And (3) receiving an instruction: the data acquisition module receives control instructions from other modules;

(3) Data synchronization: under the condition that a plurality of data acquisition modules work cooperatively, other modules coordinate data synchronization among the modules, so that the consistency and the integrity of data are ensured;

(4) And (5) updating the state: the data acquisition module reports the state of the data acquisition module to other modules regularly;

The data preprocessing module processes and analyzes the data transmitted by the data acquisition module for downstream tasks; the data required by local end-to-end calculation is input into a local large language model, and the data comprises real-time battery state estimation and health monitoring data, fault diagnosis and early warning data, emergency protection control data, performance optimization and real-time management data, and the data are preprocessed, wherein the data comprise the following operations:

and (3) filtering: in order to remove noise and abnormal values in the voltage, current and temperature data, a low-pass filter is adopted for filtering treatment so as to remove sensor noise and instantaneous fluctuation;

normalization: normalizing the data of different dimensions to within the same magnitude and range;

Threshold detection: for the basic protection function, safety thresholds of voltage, current and temperature are required to be set, and whether the condition exceeding the threshold occurs is detected in real time so as to trigger corresponding protection measures;

the data preprocessing module interacts by:

(1) And (3) a data interface: defining a standard data interface, and defining a data format and a transmission protocol to ensure that the data is transmitted to other modules;

(2) Real-time data flow: the preprocessed data is transmitted to other modules in real time in a real-time data stream mode so as to support real-time calculation and decision making;

(3) Format conversion: converting the data into a desired format for reading and processing;

(4) Feedback mechanism: other modules can provide feedback to the data preprocessing module according to the analysis result to guide the data preprocessing module to optimize the preprocessing strategy and parameter setting;

(5) Exception handling: when other modules detect data abnormality or analysis failure, notifying a data preprocessing module to reprocess or collect new data;

the large language model reasoning module is deployed locally and consists of the following parts:

Model loading part: automatically downloading the latest pre-trained large language model from the cloud to be loaded into the memory so as to perform real-time reasoning;

a data input section: the data preprocessing module is responsible for receiving data input by the data preprocessing module, extracting the data according to a specified data format, combining the data with the promt, and converting the data into a format suitable for model input for use by a large language model;

an inference calculation section: carrying out reasoning calculation on the input data by using the loaded large language model to obtain a battery state reasoning result;

a result output section: analyzing the result, and analyzing the reasoning result of the model into information with actual physical meaning; the result is packaged, and the analyzed result is packaged into a standard format, so that the use and analysis are convenient; the result transmission or display, the reasoning result is transmitted to the downstream module in time through the message queue or the event triggering mechanism;

the large language model reasoning module interacts by:

(1) Result transfer: transmitting the battery state prediction result obtained by reasoning to other modules;

(2) Feedback reception: receiving feedback information from other modules;

(3) Command response: responding to commands from other modules;

The cloud platform integration module is responsible for connecting and interacting the vehicle-mounted battery management system with a cloud platform of a vehicle provider, receiving an output result of the large language model reasoning module, realizing data backup and synchronization, remote diagnosis and management, model updating and other vehicle-mounted software updating, and receiving feedback information of a user on maintenance and maintenance advice; the cloud platform integration module consists of the following parts:

Communication interface: the system is responsible for data transmission and communication between the BMS and the cloud platform;

Data buffer: the cloud platform is used for temporarily storing data to be uploaded to the cloud platform, and receiving instructions and updates from the cloud platform;

data encryption and security: the encryption transmission and the security authentication of the data are realized, so that the security and the privacy of the data in the transmission process are ensured;

model update management: the method is responsible for downloading the latest large language model from the cloud platform and carrying out replacement and update locally;

system feedback and diagnostics: collecting the running state and performance index of the system, and uploading the running state and performance index to a cloud platform for monitoring and analysis;

The system self-diagnosis function is realized, and diagnosis results and abnormal information are fed back to the cloud platform so as to facilitate remote maintenance and fault investigation;

the cloud platform integration module interacts by:

(1) And (3) data exchange: uploading the result of the large language model reasoning module to a cloud platform for backup and analysis, and issuing suggestions and schemes of the cloud platform to a vehicle-mounted management system;

(2) Model update notification: when the cloud platform has new large language model update, the cloud platform integration module is responsible for downloading and notifying a downstream large language model reasoning module to perform model replacement and update;

(3) Configuration update and synchronization: the cloud platform integration module receives a configuration update instruction from the cloud platform and synchronizes the configuration update instruction to a related downstream module;

(4) Fault diagnosis and feedback: the fault or abnormal information found by the downstream module is uploaded to a cloud platform through a cloud platform integration module, the cloud platform provides corresponding processing suggestions or repairing schemes according to analysis results, and the cloud platform integration module issues the corresponding processing suggestions or repairing schemes to the downstream module for execution;

(5) Remote control and management: the cloud platform integration module receives a remote control instruction from the cloud platform and transmits the instruction to the corresponding downstream module for execution.

2. The end-to-end vehicle battery online state assessment system based on a large language model of claim 1, wherein: the sensor assembly comprises a voltage sensor, a current sensor and a temperature sensor;

the data acquisition module acquires the following six types of data:

Voltage data: the voltage value of each battery unit is used for monitoring the charge state and the health condition of the battery and is acquired by a voltage sensor;

Current data: the current value of the battery during charging and discharging is used for calculating the energy inflow and outflow of the battery and is collected by a current sensor;

temperature data: the temperature of the battery unit and the module is used for preventing overheat and evaluating the thermal management effect of the battery, and is acquired by a temperature sensor;

Charge-discharge state data SOC: the charging level of the battery is used for estimating the remaining driving mileage and managing the charging and discharging process, and is collected by connecting a data line with the BMS system;

health status data SOH: the health condition of the battery is used for predicting the service life and performance degradation of the battery and is collected by connecting a data line with a BMS system;

Charge-discharge cycle times: the charge and discharge times of the battery are used for evaluating the service condition and service life of the battery, and the battery is collected by connecting a data line with the BMS system.

3. The end-to-end vehicle battery online state evaluation method based on the online state evaluation system of claim 1 or 2, wherein the method comprises the following steps: the method comprises the following steps:

s1: monitoring the vehicle-mounted battery in real time, and collecting data of the vehicle-mounted battery;

Calibrating the sensor;

Setting sampling frequency, and setting data acquisition frequency according to the working state of the battery and data analysis requirements;

The format and the structure of the data are defined, so that the data transmission and the processing are convenient;

collecting data in a local storage mode, and regularly backing up the data to a cloud platform;

setting an abnormal threshold value, and immediately executing alarming and protecting measures if abnormal data are detected;

s2: preprocessing and converting the acquired vehicle-mounted battery data;

Initializing a module when starting;

Removing abnormal data by using an abnormal value detection method, and filling the missing data;

smoothing the data by applying a low-pass filter to reduce instantaneous fluctuation;

normalizing the data to between 0 and 1 using a min-max normalization method;

extracting statistical characteristics, and carrying out frequency domain characteristic extraction signs;

packaging the preprocessed data into JSON or CSV format, so as to facilitate analysis and processing of downstream modules;

the preprocessed data is sent to a downstream module in real time through a data bus;

S3: selecting a pre-trained model for loading, and automatically initializing model configuration; the method comprises the steps of receiving battery data transmitted by an upstream data preprocessing module through a data interface, and combining the received data with a preset promt, wherein the flow is as follows:

S31: formatting the received battery data into a form suitable for combination with a promt;

S32: constructing a Prompt template for embedding the formatted data into the Prompt; the template is designed according to the requirements of a large language model and a specific application scene;

S33: before embedding the data into the Prompt, processing special characters possibly existing in the data to avoid interference to the input of the large language model; mapping the numerical value and the characteristics of the data to vocabulary with corresponding semantics;

S34: embedding the formatted data into a Prompt template, and filling sensor data into corresponding placeholders to form a complete input text;

s35: taking the constructed promt as the input of a large language model to predict or analyze; the large language model will generate a corresponding output from the battery data provided in the Prompt;

Inputting the data embedded into the Prompt into a pre-trained large language model for reasoning and analysis;

Analyzing and processing the original result obtained by reasoning, and extracting the required information; post-processing the results according to the requirements, outputting the processed results in a standard format, displaying the processed results in a user interaction interface, and uploading the processed results to a cloud platform regularly through a communication interface;

s4: the communication and cooperative work between the vehicle-mounted battery management system and the cloud platform are realized, and the functions of instruction receiving, model updating, system updating, feedback reporting, maintenance and monitoring are supported;

Receiving an instruction and configuration update from a cloud platform, and monitoring the real-time instruction and configuration update from the cloud platform by using a WebSocket or long polling mechanism;

according to the collected real-time data and feedback information, a software provider learns the Llama model in a cloud platform; when the performance of the model is reduced or new data appear, performing thermal updating of the model;

After receiving the system updating instruction, downloading the latest model file from the cloud platform through HTTP or FTP protocol, executing the system updating flow, replacing the local old model file, and simultaneously notifying the downstream module to reload the model; updating the model and the driving software of the system regularly;

The method comprises the steps of regularly collecting running logs, performance indexes and fault diagnosis results of a vehicle-mounted battery management system, packaging the running logs, the performance indexes and the fault diagnosis results into a JSON format, and uploading the JSON format to a log analysis endpoint of a cloud platform for monitoring and analysis by operation and maintenance personnel; collecting and collecting execution conditions and feedback information of maintenance and maintenance suggestions of a user;

periodically checking the stability of network connection and the safety of data transmission; periodically checking the connection state with the cloud platform, monitoring the stability and safety of data transmission, and executing necessary maintenance operation;

and generating corresponding maintenance suggestions and maintenance plans according to the data uploaded by the vehicle-mounted battery management system.

4. The end-to-end on-vehicle battery presence assessment method of claim 3, wherein: the sensor includes a voltage sensor, a current sensor, and a temperature sensor.