Intelligent battery diagnosis system and method based on vehicle operation data and large model of hybrid electric vehicle

Abstract

The invention provides an intelligent battery diagnosis system and method based on vehicle operation data and a large model of a hybrid electric vehicle. The method comprises the following steps: collecting charging and discharging data of operating vehicles of various types in different working modes; fault data, fault types and fault handling suggestions of all vehicle types are collected in a classified mode according to the fault types; constructing a fault prediction model based on multi-model fusion based on the collected data; training the large language model according to a mapping relation generated by the fault prediction model, and constructing an intelligent fault knowledge base; and inputting the charging and discharging data and the fault data output by the vehicle-mounted BMS into the intelligent fault knowledge base to obtain a disposal scheme corresponding to the fault data. According to the method, charging, discharging and fault data of the series plug-in hybrid electric vehicle are considered, the fault prediction model is constructed by adopting an intelligent algorithm, and the fault knowledge base is constructed through the fault data and the fault disposal data, so that the fault recognition accuracy and efficiency of the power battery and the fault disposal efficiency of the power battery can be greatly improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of new energy vehicles, and in particular relates to an intelligent battery diagnosis system and method based on hybrid electric vehicle operation data and a large model. Background Art

[0002] Driven by global carbon neutrality goals and the rapid evolution of new energy vehicle technologies, hybrid vehicles (HEVs), with their fuel economy and low emissions, have become a key solution for the transition from traditional fuel vehicles to pure electric vehicles. As the core energy carrier of HEVs, the performance of the power battery system directly determines the vehicle's power output efficiency, range, and service life. However, issues such as degradation of the battery's electrochemical properties under complex operating conditions, increased cell inconsistencies, and the risk of thermal runaway have become key bottlenecks hindering the reliability and safety of HEVs.

[0003] Currently, power battery fault diagnosis faces numerous technical challenges: 1. Battery aging is influenced by multiple factors, including temperature, charge / discharge rate, and cycle count. Its nonlinear degradation characteristics are difficult to accurately model using single sensor data. 2. Traditional diagnostic methods (such as open-circuit voltage and internal resistance measurement) suffer from limitations such as poor real-time performance and strong dependence on operating conditions, making them difficult to meet the online monitoring requirements of the vehicle's battery management system (BMS). Furthermore, the differential degradation of hundreds of cells within a battery pack can trigger a "short board effect," leading to a sharp decrease in system usable capacity and even safety hazards. According to industry statistics, approximately 40% of vehicle recalls due to battery failure are directly related to deviations in SOH assessments or ineffective fault warnings. Summary of the Invention

[0004] In view of the technical challenges faced by the above-mentioned power battery fault diagnosis, the present invention proposes an intelligent battery diagnosis system and method based on hybrid electric vehicle vehicle operation data and a large model.

[0005] An intelligent battery diagnostic system based on hybrid electric vehicle operating data and a large model to achieve one of the objectives of the present invention includes: Charge and discharge data statistics module: used to collect charge and discharge data of various vehicle models in different working modes (including electric mode and series mode); The discharge data includes the duration of the vehicle in different temperature ranges and different discharge power segments, as well as the transient discharge characteristic parameters before and after mode switching. The transient discharge characteristic parameters before and after mode switching include: the average discharge power for a set time (such as 5 seconds) before switching from electric mode to series mode, and the power fluctuation amplitude within a set time (such as 10 seconds) after switching. The temperature range is divided into -30℃~-25℃, -25℃~-20℃...25℃~30℃, and the discharge power segment is divided into 0-5kW, 5-10kW, and 10-15kW. The duration is the cumulative operating time within the corresponding range. The charging data includes the duration of the power segment and the number of charges for external charging pile charging and on-board charging in series mode. The data structure of external charging pile charging and on-board charging is set as a three-dimensional matrix of power segment-temperature segment-duration. The statistical rules for the number of charging times include: a single charging cycle from the establishment to the disconnection of the power supply connection is counted as 1 time. If the charging is interrupted due to mode switching and resumed within a set time (such as 2 minutes), it is counted as 1 time to avoid counting distortion under high-frequency mode switching.

[0006] Fault statistics module: used to collect fault data, fault types and fault handling suggestions for each vehicle type according to fault type classification, record key parameters when the fault occurs (including temperature, pressure difference, current, voltage), corresponding working mode and mode switching status; Fault data includes: vehicle's current operating mode (electric / series), mode switching status (before / during / after switching), power battery temperature, battery voltage difference, battery current, battery voltage, historical charging times (including external and on-board charging classification statistics), and charging and discharging data statistics; the number of fault samples must be no less than n (for example, 1,000), and each fault must be associated with a corresponding fault handling suggestion and a fault signal value.

[0007] Mode switching states refer to the different stages of the vehicle's switching between electric mode and series mode, including before switching, during switching, and after switching: Before switching: The vehicle is in a critical state before switching from the current mode (such as electric mode) to another mode (such as series mode). At this time, the battery power is usually close to the switching threshold (such as 20%), and the system begins to prepare for the switching process (such as pre-starting the engine); Switching: The vehicle is in the dynamic process of mode conversion, such as switching from electric mode to series mode. The engine starts, the generator starts to work and charges the battery, and the battery's charge and discharge status and power output are in a stage of drastic changes; After switching: The mode switching is completed, the vehicle runs stably in the new mode (such as series mode), and the working conditions of components such as the engine, generator, and drive motor tend to be stable.

[0008] Because faults in series plug-in hybrid vehicles (such as abnormal pressure differences and temperatures) are often strongly correlated with power fluctuations and stress changes during mode switching, the technical benefits of recording the mode switching state include: capturing the transient characteristics of the battery during this process (such as sudden changes in voltage, current, and pressure difference), and targeted statistics of fault data in this state, thereby improving the accuracy of the diagnostic model.

[0009] Fusion model construction and training module: This module is used to construct a multi-model fusion fault prediction model based on the collected charge and discharge data, fault data, fault type, and fault handling suggestions. The fault prediction model is used to obtain a mapping relationship between vehicle charge and discharge data, fault data, fault handling suggestions, and fault types. The input features of the fault prediction model include mode switching transient parameters, and the final fault type is output through the mode of the multi-model prediction results. Intelligent fault knowledge base construction module: used to train the large language model based on the mapping relationship generated by the fault prediction model to build an intelligent fault knowledge base; Fault diagnosis module: used to input the charging and discharging data and fault data output by the on-board BMS into the intelligent fault knowledge base, and the intelligent fault knowledge base outputs the corresponding treatment plan for the fault data.

[0010] Furthermore, in the fusion model construction and training module, a neural network model and a machine learning model are used to build a fault prediction model based on multi-model fusion. The final prediction results include: y = mode ( y neural ,y xgboost ,y random ) in y is the final prediction result; y neural 、 y xgboost 、 y random They represent neural networks, xgboost And the results of random forest prediction, mode (.) means taking the mode.

[0011] Furthermore, the F1-SCORE evaluation formula is used to evaluate the accuracy of the fault prediction model. The calculation formula includes: P Indicates accuracy; R Represents the recall rate.

[0012] Furthermore, the method of constructing an intelligent fault knowledge base includes: Data preprocessing: Break down fault type, fault occurrence mode, fault triggering conditions, associated operating data, and handling suggestions into structured data of "fault occurrence mode - fault characteristics - handling logic"; Model training and optimization: Using a large language model based on the Transformer architecture, the preprocessed structured data is divided into training and test sets according to a set ratio. During training, prompt engineering is used to design fault diagnosis scenario instructions of "fault occurrence mode-fault characteristics-handling logic" to enable the model to learn the mapping relationship between fault characteristics and handling methods. The test set is used to evaluate whether the model training meets the requirements. When the matching degree of the fault handling suggestions in the test set is greater than the set ratio, the training is considered to have met the requirements.

[0013] Furthermore, the method for constructing an intelligent fault knowledge base also includes: Knowledge base application and iteration: When new fault cases are added or the handling method is optimized, the parameters of the large language model and fault prediction model are updated through incremental training. Separate training is performed on the fault data of the mode switching scenario, and the fault type, data volume, and accuracy changes of each update are recorded to ensure that the intelligent fault knowledge base continues to adapt to new scenarios of power battery failures.

[0014] Furthermore, the fault diagnosis module includes: The sensing unit is used to collect the operating data of the vehicle's power battery in real time, and transmit the operating data to the decision engine unit after pre-processing. It is also used to identify the vehicle's operating mode and switching status, collect transient battery parameters within a set time period before and after mode switching, and transmit the transient battery parameters to the decision engine unit after pre-processing. Knowledge base: used to store fault types, pattern-related feature parameters, handling suggestions, and fault signal values ​​after training with a large language model; Decision engine unit: This unit receives data from the perception unit, matches fault features in the built intelligent fault knowledge base, and combines the reasoning capabilities of the large language model unit to determine the fault type and severity level. Execution unit: used to execute corresponding handling actions according to the fault type and severity level output by the decision engine unit; Large language model unit: used to train fault data and handling suggestions, and learn the mapping relationship between "fault characteristics, mode status, and handling logic."

[0015] Furthermore, in the associated fault statistics module, a method for determining a fault signal value in the fault data corresponding to the abnormal battery temperature fault type includes: Filter the fault data of the vehicle in series mode, count the temperatures of all temperature abnormality faults in the fault data, take the set percentile value of all temperatures as the fault signal value, and use the fault signal value as the battery temperature that triggers the temperature abnormality warning of the vehicle in series mode.

[0016] Furthermore, in the associated fault statistics module, the method for determining the fault signal value P0 in the fault data corresponding to the abnormal pressure difference fault type includes: counting the average maximum pressure difference at the moment when the vehicle of this model switches from electric mode to series mode in history, and multiplying it by a set safety factor as the fault signal value p0.

[0017] A smart battery diagnostic method based on hybrid electric vehicle operating data and a large model to achieve the second objective of the present invention includes: Collect charging and discharging data of various types of vehicles in different working modes; Collect fault data, fault types and fault handling suggestions for each vehicle type based on fault type classification; Based on the collected charging and discharging data, fault data, fault types, and fault handling suggestions, a fault prediction model based on multi-model fusion is constructed. The fault prediction model is used to obtain a mapping relationship between vehicle charging and discharging data, fault data, fault handling suggestions, and fault types; The large language model is trained based on the mapping relationships generated by the fault prediction model to build an intelligent fault knowledge base; The charging and discharging data and fault data output by the on-board BMS are input into the intelligent fault knowledge base, and the intelligent fault knowledge base outputs the corresponding treatment plan for the fault data.

[0018] A non-transitory computer-readable storage medium is provided to achieve the third objective of the present invention, on which a computer program is stored. When the computer program is executed by a processor, the steps of the intelligent battery diagnosis method based on hybrid vehicle operation data and a large model are implemented.

[0019] A computer program product for achieving the fourth objective of the present invention includes a computer program / instruction, which, when executed by a processor, implements the steps of the intelligent battery diagnosis method based on hybrid vehicle operation data and a large model.

[0020] The beneficial effects of the present invention include: The present invention targets series plug-in hybrid electric vehicles, considers the charging, discharging and fault data of series plug-in hybrid electric vehicles, and adopts an intelligent algorithm to build a fault prediction model, especially a power battery fault prediction model; and builds a fault knowledge base based on fault data and fault handling data, which can greatly improve the accuracy and efficiency of power battery fault identification and the efficiency of power battery fault handling. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic flow chart of an embodiment of the method of the present invention; Figure 2 This is a structural diagram of a series plug-in car; Figure 3 It is an intelligent fault knowledge base. DETAILED DESCRIPTION

[0022] The following specific embodiments are provided to explain the technical solutions of the present invention so that those skilled in the art can understand the present invention. The scope of protection of the present invention is not limited to the specific implementation structures described below. Any implementation schemes created by those skilled in the art that include the technical solutions of the present invention but differ from the following specific implementation schemes are also within the scope of protection of the present invention.

[0023] The series plug-in hybrid electric vehicle of the present invention is as follows Figure 2 As shown, the engine is a gasoline engine. The black lines in the figure represent mechanical connections, and the red lines represent electrical connections. There are two drive modes: electric mode and series mode. In electric mode, the vehicle relies on the drive motor, which is powered by the high-voltage battery, and the engine is inoperative. When the battery charge reaches a certain threshold (for example, 20%), the vehicle enters series mode, relying on the drive motor. The engine is active, driving the generator to charge the high-voltage battery and provide power to the drive motor. Unlike traditional electric vehicles, the engine continuously charges the high-voltage battery, so the high-voltage battery is charged and discharged more frequently, making it more susceptible to battery degradation and failure.

[0024] Based on the above-mentioned plug-in hybrid electric vehicle, an embodiment of the present invention provides an intelligent battery diagnosis method based on hybrid electric vehicle operation data and a large model, and the specific steps are as follows: 1. Discharge data statistics: Discharge data is collected for each vehicle under different vehicle models (differentiated by hybrid battery type) under discharge conditions. This includes the duration of discharge at different power levels for each vehicle in different battery temperature ranges (categorized by temperature: -30°C to -25°C, -25°C to -20°C, -20°C to -15°C, etc.). Discharge data is also differentiated between electric and series modes, with a focus on collecting transient discharge characteristic parameters before and after mode switching. These include the average discharge power in the 5 seconds before the switch, the power fluctuation amplitude within 10 seconds after the switch (a fluctuation >15% is considered significant), and the duration of the switch. For example, for the battery temperature range T1 (20-25°C), the discharge duration for 10-5 kW in electric mode is t1, the discharge duration for 5-10 kW in series mode is t2, and the discharge duration for 10-15 kW during the mode switch transition is t3.

[0025] 2. Charging Data Statistics: Data is collected for all vehicles of the same model under different charging conditions, distinguished by vehicle type (and hybrid battery type). This includes the duration of each vehicle's charging at different battery temperature ranges (categorized by temperature range: -30°C to -25°C, -25°C to -20°C, -20°C to -15°C, etc.) and at different charging power levels. Charging data includes both external charging station charging and on-board charging (using the engine-driven generator in series mode). The power ranges and duration calculations for the two types of charging are kept consistent. For example, for battery temperature range T1 (20-25°C), the duration of external charging at 0-5 kW is c1, the duration of on-board charging at 5-10 kW is c2, and the duration of external charging at 10-15 kW is c3. The number of historical charging times for each vehicle is also counted. The number of on-board charging times is linked to the number of times the series mode is activated. Any charging interruption due to a mode switch and resumption within 2 minutes is counted as one to avoid counting errors caused by mode switching within a short period of time.

[0026] 3. Power battery fault statistics: Differentiate vehicle models (by hybrid battery type), and at the same time distinguish fault types S (such as thermal runaway S1, voltage difference abnormal S2, battery temperature abnormal S3, etc.), and record key data when the fault occurs, including: vehicle current working mode (electric / series), mode switching status (before / during / after switching), power battery temperature, battery voltage difference, battery current, battery voltage, historical charging times (including external and on-board charging classification statistics), discharge data statistics recorded in the above 1, and charging data statistics recorded in the above 2; the number of fault samples should be no less than n (for example, 1000), and each fault must be associated with the corresponding fault handling suggestions and fault signal value (for example, when the battery temperature is abnormal S3, the battery temperature T w .

[0027] In series mode, the engine generates additional heat when operating, and the battery heat dissipation environment is more complex. Therefore, for the series mode, in some embodiments, the fault signal value corresponding to the abnormal battery temperature fault type (battery temperature T w The method for determining the temperature threshold of the vehicle includes: when the vehicle is in series mode (the engine is working, charging the battery and driving the motor at the same time), counting the temperature thresholds of all temperature abnormality failure cases in the history of the vehicle model, and taking the 95% percentile value as the T w For example, a certain vehicle model has 100 battery temperature abnormality faults S3 in series mode. The temperature values ​​that trigger the battery temperature abnormality fault S3 alarm are 55℃, 56℃...65℃. After sorting these values, the 95th value is T w When the vehicle is in series mode, the battery temperature reaches T wThe technical effects include: avoiding missed alarms caused by a single fixed threshold, and adapting to complex and changing working conditions in series mode.

[0028] Because series plug-in hybrid vehicles frequently switch between electric and series modes, the instantaneous switching causes a sudden surge in the battery system's input power, which, combined with the existing discharge power (driving the motor), triggers dramatic instantaneous voltage and current fluctuations. This fluctuation can cause the voltage differential between battery cells to increase dramatically over a short period of time, far exceeding the voltage differential fluctuations of pure electric vehicles. Therefore, to avoid missed detections due to excessively low thresholds, which could lead to potential failures under such transient high stress, in some embodiments, the fault signal value p0 for the voltage differential abnormality S2 is determined by calculating the average maximum voltage differential at the moment of switching from electric to series mode for that vehicle model and multiplying this average by a set safety factor (e.g., 1.5) to determine p0. For example, if the average maximum voltage differential at the moment of switching from electric to series mode for a particular vehicle model is 50mV, then p0 = 50mV × 1.5 = 75mV. A voltage differential abnormality is determined when the voltage differential at the moment of switching exceeds 75mV. Its technical effects include: the power fluctuates violently when switching between series plug-in hybrid modes, and the cell pressure difference is more likely to change suddenly. Therefore, a higher safety factor is adopted to deal with instantaneous stress failures during mode switching, and to issue early warnings for abnormal pressure differences that exceed normal fluctuations, which can significantly improve the accuracy and reliability of abnormal pressure difference failure warnings.

[0029] 4. Construction of an Intelligent Fault Knowledge Base: Based on the power battery fault data collected in step 3 above, 80% of the fault samples were used as the training set. A neural network model and machine learning models (XGBOOST and random forest algorithms) were used to construct a mapping relationship between fault types and the following characteristic parameters: vehicle current mode, mode switching transient parameters, power battery temperature, vehicle model, power battery temperature, historical charge times, discharge data recorded in step 1 above, and charging data statistics recorded in step 2 above.

[0030] The neural network includes 2 hidden layers and 1 feature enhancement layer. The input layer dimension includes the total dimension of the above feature parameters + 3 mode switching features: power before switching, fluctuation amplitude after switching, and switching time. The number of neurons in the hidden layer is 1.8 times that of the input layer. The output layer is the number of categories of fault type S, that is, the number of neurons in the output layer is consistent with the number of preset fault types. Each neuron corresponds to the prediction result of a fault type. For example: if the preset fault type S includes three categories: thermal runaway S1, voltage difference abnormality S2, and battery temperature abnormality S3, then the output layer of the neural network sets 3 neurons, corresponding to S1, S2, and S3 respectively; when the model predicts a battery abnormality data, the 3 neurons in the output layer output the probability values ​​of the three types of faults: S1 is 0.1, S2 is 0.8, and S3 is 0.1. At this time, the prediction result of the model is yneural The pressure difference anomaly S2 is determined to have the highest probability. This setting allows the neural network to directly output predictions for various fault types, matching the classification results of the XGBoost and random forest models. Ultimately, a unified fault prediction conclusion is obtained by "taking the mode." The activation function is ReLU, and the cross-entropy loss function is used during training. The number of iterations is set to 500 until the loss function converges.

[0031] The number of XGBoost trees was set to 100-200, the maximum tree depth was limited to 5-8, the learning rate was initialized to 0.1, and the learning of pattern-related features was enhanced by setting the “mode switch” to a high-order feature interaction term (gamma = 0.3). The parameters were adjusted through cross-validation of the training set to minimize the prediction error.

[0032] The number of decision trees in the random forest is 200, the maximum number of features is set to 2 / 3 of the total number of features, the Gini coefficient is used for node splitting, and the stability of the model in the mode switching scenario is evaluated by out-of-bag data (OOB).

[0033] The final model prediction results are: y = mode ( y neural ,y xgboost ,y random ) (1) The final prediction result is y , y neural 、 y xgboost 、 y random They represent neural networks, xgboost And the results of random forest prediction, mode (.) means taking the majority, that is y For the neural network model, xgboost The result that appears most frequently among the classification results predicted by the three models, the neural network model and the random forest model, is used as the final fault prediction result; for example, if a power battery of a series plug-in hybrid electric vehicle is abnormal, the fault type is predicted by the three models respectively: neural network ( y neural ) The prediction result is "pressure difference abnormal (S2)"; XGBoost ( y xgboost ) The prediction result is "pressure difference abnormality (S2)"; Random Forest ( y random) The prediction result is "Battery temperature abnormality (S3)". Among these three results, "Voltage difference abnormality (S2)" appears 2 times, and "Battery temperature abnormality (S3)" appears once. According to the "take the majority" rule, the final prediction result is y It is "Abnormal pressure difference (S2)".

[0034] 20% of the faulty samples were used as the test set. The F1-SCORE evaluation formula was used to evaluate the model's balance index F1. The F1 balance index was compared with the set value to determine whether the model met the standard. The balance index was calculated by calculating the harmonic mean of precision and recall, comprehensively measuring the model's ability to strike a balance between accuracy (the proportion of predicted positive results that are actually positive) and completeness (the proportion of all actually positive samples that are successfully predicted): (2) in P The accuracy rate refers to the proportion of samples predicted by the model as a certain type of fault that are actually of that type of fault, reflecting the accuracy of the prediction results; P = the number of samples predicted to be faults correctly / (the number of samples predicted to be faults correctly + the number of samples predicted to be faults but not actually faults). R Recall rate refers to the proportion of samples that are actually a certain type of fault that are successfully predicted by the model, reflecting the completeness of the prediction results. R = Number of correctly predicted fault samples / (number of correctly predicted fault samples + number of samples that are actually faults but not predicted); F1 > α When , it means that the above fault prediction model is available, 0< α <1, in this embodiment α The value is 0.9.

[0035] For example, suppose there are 100 actual samples with compression abnormality S2 in the test set, and the model prediction results are as follows: Number of correctly predicted fault samples: 80; Number of samples predicted to be fault S2 but not actually fault S2: 20; The number of samples that are actually fault S2 but not predicted as fault S2 is 10.

[0036] Then: precision P = 80 / (80+20) = 0.8; recall R = 80 / (80+10) ≈ 0.89; F 1=(2×0.8×0.89)(0.8+0.89}≈0.842.

[0037] When F1 > 0.9, the model's comprehensive predictive performance meets acceptable standards and can be used for power battery failure prediction. This formula balances accuracy and recall, avoiding bias in model evaluation caused by a single metric (such as accuracy alone), and more comprehensively reflects model reliability.

[0038] 5. Such as Figure 3 As shown in the figure, faults and corresponding treatment suggestions are analyzed using a large language model to build an intelligent fault knowledge base. Different faults correspond to different treatment methods. The specific construction process includes three stages: data preprocessing, model training and optimization, and knowledge base iteration: 5.1 Data Preprocessing The statistical fault data is structured, breaking down the fault type, fault mode (electric / series / switching process), fault triggering conditions (such as battery temperature range, pressure difference threshold, duration), associated operating data (such as discharge power range, number of charges, mode switching parameters), and treatment recommendations into key-value pairs of "fault mode - fault characteristics - treatment logic." For example, for pressure difference abnormality S2, its characteristics are clearly defined as "pressure difference > p0 within 3 seconds after switching from electric to series mode and duration > T0 (e.g., 3 minutes)." The corresponding treatment recommendation mp1 is "limit engine output power to 70% of rated power, reduce mode switching frequency, and prompt the user to visit a service station for repair."

[0039] 5.2 Model Training and Optimization A large language model based on the Transformer architecture is used. The training data includes 15% mode switching fault cases (e.g., voltage differential fluctuations during low-temperature and high-power switching). The preprocessed structured data is divided into a training set and a validation set at a set ratio (e.g., 7:3). During training, prompt engineering is used to design fault diagnosis scenarios, including "fault occurrence mode - fault characteristics - action logic" (e.g., "When, in a -10°C environment, the vehicle detects a battery voltage differential >50mV within 3 seconds after switching from series mode to electric mode and persists for 5 minutes, while the current temperature is 28°C and the discharge power is 12kW, what action should be taken?"). This guides the model to learn the mapping between fault characteristics and action methods. During training, LoRa lightweight fine-tuning technology is used to freeze the basic model parameters and train only the low-rank matrix parameters to reduce computational cost. The accuracy of the fault action recommendations output by the model is monitored in real time on the validation set. The accuracy of the recommended fault action recommendations is required to be ≥98%. If the accuracy falls below this standard, the data preprocessing rules are retroactively adjusted, such as by supplementing edge case data.

[0040] 5.3 Knowledge Base Application and Iteration The constructed intelligent fault knowledge base communicates in real time with the vehicle's BMS and mode perception module via an API interface. When the BMS detects a fault, it automatically identifies the current vehicle mode and switching status, automatically extracts fault characteristics (such as real-time pressure difference, temperature, power, etc.), and enters them into the intelligent fault knowledge base. The model then generates a solution within a set time. For example, a voice or text notification can be sent to the user: "Abnormal power battery temperature detected in series mode. Engine power has been reduced. Immediate inspection at the nearest service station is recommended." A text message or email can be sent to the vehicle's cloud-based after-sales system, containing the faulty vehicle's VIN number, real-time location, current operating mode and switching history, fault time, core fault parameters (such as a pressure difference of 52mV or a temperature of 28°C), and a recommended priority level (such as "urgent"), along with detailed steps for the recommended solution.

[0041] At the same time, the knowledge base has a dynamic update mechanism for mode adaptability. When new fault cases are added (such as failures of new model vehicles under specific mode combinations) or the handling methods are optimized, the model parameters are updated through incremental training. Every quarter, a separate intensive training is conducted on the fault data of the mode switching scenario, and the fault type, data volume and accuracy changes of each update are recorded to form a traceable knowledge base iteration log to ensure its continuous adaptation to new scenarios of power battery failures.

[0042] The completed intelligent battery diagnostic system includes the following modules, and the functions and coordination logic of each module are as follows: The sensing unit is used to collect real-time operating data of the vehicle's power battery, including charging and discharging power and time in different temperature ranges, historical charging times, and parameters such as temperature / pressure difference / current / voltage at the time of the fault (corresponding to the data statistics in steps 1-3 above). The sensing unit then pre-processes the data and transmits it to the decision engine unit. The sensing unit is also used to identify the vehicle's operating mode (electric / series) and switching status, and collect transient battery parameters (voltage, current, power) within a set time period (e.g., 5 seconds) before and after mode switching, and transmits the pre-processed data to the decision engine. Knowledge base: used to store fault types, pattern-related feature parameters, handling suggestions, and fault signal values ​​(such as threshold conditions for S1-S3 faults and corresponding handling methods such as mp1) trained by the large language model, providing data support for decision-making; Decision engine unit: Receives real-time data from the sensing unit, matches fault features in the built intelligent fault knowledge base, and uses the reasoning capabilities of a large language model to determine the fault type and severity level (e.g., emergency / warning / prompt). It automatically increases the response priority to the first level for faults encountered during mode switching. Execution unit: Used to execute specific disposal actions based on the output of the decision engine, including notifying the user by voice, notifying after-sales personnel by SMS / email, and linking the vehicle system to limit battery power; Large language model unit: as the core algorithm support, train the fault data and treatment suggestions in step 3, learn the mapping relationship of "fault feature-mode state-treatment logic", and provide reasoning basis for decision engine; Tool: including data cleaning tool for cleaning raw data collected by processing perception module, model evaluation tool for calculating F1-SCORE, interface tool for realizing data transmission between modules, and guaranteeing stable operation of system.

[0043] For example: when the power battery appears a pressure difference fault, the pressure difference is greater than a certain threshold p0, and the duration is greater than a certain time range T0, the perception module collects the pressure difference data and corresponding temperature / power parameters, the decision engine calls the features of the pressure difference anomaly S2 in the knowledge base for matching, determines the treatment method as mp1 after reasoning by the large language model, and then the execution module completes user notification and after-sales linkage, the whole process realizes data flow and model optimization through the tool. All faults, corresponding fault data and treatment methods are input to the large model training to build a power battery fault knowledge base, automatically give fault treatment methods, and notify users of the treatment method in the form of voice, and send the treatment method to after-sales personnel in the form of short message or email.

[0044] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present application.

[0045] The embodiment of the present application also provides a non-transitory computer readable storage medium, which stores a computer program, the computer program includes program instructions, and the program instructions are executed by a processor to realize each step of the method of the present application, which will not be repeated here.

[0046] The computer readable storage medium can be an internal storage unit of the data transmission device or the computer device provided in any of the preceding embodiments, such as the hard disk or the memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0047] Further, the computer readable storage medium can include both the internal storage unit and the external storage device of the computer device. The computer readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer readable storage medium can also be used to temporarily store data to be output or data that has been output.

[0048] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0049] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0050] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0051] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0052] The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

Claims

1. An intelligent battery diagnostic system based on hybrid electric vehicle operating data and a large model, characterized in that: include: Charging and discharging data statistics module: used to collect charging and discharging data of various vehicle models under different working modes; Fault statistics module: used to collect fault data, fault types and fault handling suggestions for each vehicle type according to fault type classification; Fusion model construction and training module: used to build a multi-model fusion fault prediction model based on the collected charge and discharge data, fault data, fault type and fault treatment suggestions. The fault prediction model is used to obtain the mapping relationship between vehicle charge and discharge data, fault data, fault treatment suggestions and fault type; Intelligent fault knowledge base construction module: used to train the large language model based on the mapping relationship generated by the fault prediction model to build an intelligent fault knowledge base; Fault diagnosis module: used to input the charging and discharging data and fault data output by the on-board BMS into the intelligent fault knowledge base, and the intelligent fault knowledge base outputs the corresponding treatment plan for the fault data.

2. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 1, characterized in that: In the fusion model construction and training module, neural network models and machine learning models are used to build a fault prediction model based on multi-model fusion. The final prediction results include: y = mode ( y neural ,y xgboost ,y random ) in y is the final prediction result; y neural 、 y xgboost 、 y random They represent neural networks, xgboost And the results of random forest prediction, mode (.) means taking the mode.

3. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 2, characterized in that: The F1-SCORE evaluation formula is used to evaluate the accuracy of the fault prediction model. The calculation formula includes: P Indicates accuracy; R Represents the recall rate.

4. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 1, characterized in that: Methods for building an intelligent fault knowledge base include: Data preprocessing: Break down fault type, fault occurrence mode, fault triggering conditions, associated operating data, and handling suggestions into structured data of "fault occurrence mode - fault characteristics - handling logic"; Model training and optimization: Using a large language model based on the Transformer architecture, the preprocessed structured data is divided into training and validation sets according to a set ratio. During training, prompt engineering is used to design fault diagnosis scenario instructions (fault occurrence mode - fault characteristics - handling logic) to enable the model to learn the mapping relationship between fault characteristics and handling methods. The validation set is used to evaluate whether the model training meets the requirements.

5. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 4, characterized in that: The method of building an intelligent fault knowledge base also includes: When new fault cases or handling methods are added, the parameters of the large language model and fault prediction model are updated through incremental training, and separate training is performed on the fault data of the mode switching scenario.

6. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 1, characterized in that: The fault diagnosis module includes: The sensing unit is used to collect the operating data of the vehicle's power battery in real time, and transmit the operating data to the decision engine unit after pre-processing. It is also used to identify the vehicle's operating mode and switching status, collect transient battery parameters within a set time period before and after mode switching, and transmit the transient battery parameters to the decision engine unit after pre-processing. Knowledge base: used to store fault types, pattern-related feature parameters, handling suggestions, and fault signal values ​​after training with a large language model; Decision engine unit: This unit receives data from the perception unit, matches fault features in the built intelligent fault knowledge base, and combines the reasoning capabilities of the large language model unit to determine the fault type and severity level. Execution unit: used to execute corresponding handling actions according to the fault type and severity level output by the decision engine unit; Large language model unit: used to train fault data and handling suggestions, and learn the mapping relationship between fault characteristics, mode status and handling logic.

7. The intelligent battery diagnostic system based on hybrid electric vehicle operation data and a large model as claimed in claim 1, characterized in that: In the associated fault statistics module, a method for determining the fault signal value in the fault data corresponding to the abnormal battery temperature fault type includes: Filter the fault data of the vehicle in series mode, count the temperatures of all temperature abnormality faults in the fault data, take the set percentile value of all temperatures as the fault signal value, and use the fault signal value as the battery temperature that triggers the temperature abnormality warning of the vehicle in series mode.

8. An intelligent battery diagnosis method based on hybrid electric vehicle operation data and a large model using the system of claim 1, characterized in that: include: Collect charging and discharging data of various types of vehicles in different working modes; Collect fault data, fault types and fault handling suggestions for each vehicle type based on fault type classification; Based on the collected charging and discharging data, fault data, fault types, and fault handling suggestions, a fault prediction model based on multi-model fusion is constructed. The fault prediction model is used to obtain a mapping relationship between vehicle charging and discharging data, fault data, fault handling suggestions, and fault types; The large language model is trained based on the mapping relationships generated by the fault prediction model to build an intelligent fault knowledge base; The charging and discharging data and fault data output by the on-board BMS are input into the intelligent fault knowledge base, and the intelligent fault knowledge base outputs the corresponding treatment plan for the fault data.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the intelligent battery diagnosis method based on hybrid vehicle operation data and a large model as claimed in claim 8 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the intelligent battery diagnosis method based on hybrid vehicle operation data and a large model as claimed in claim 8 are implemented.

Images