Vehicle abnormality processing method, device, new energy vehicle and readable storage medium

Abstract

The present application relates to a vehicle abnormality processing method, device, new energy vehicle and readable storage medium. The method includes: receiving user question information containing a description of the abnormal performance of the vehicle, first searching for reference processing information based on a pre-built knowledge base, and obtaining the health status information of the vehicle from a target platform associated with the vehicle, then generating corresponding prompt words based on the aforementioned information, and finally inputting the prompt words into a pre-trained generative question-answering model, and obtaining the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan suggestion based on the reasoning ability of the generative question-answering model. By introducing the health status information of the vehicle and combining it with the reference processing information searched in the knowledge base, the corresponding prompt words are generated, providing the generative question-answering model with richer context information and more accurate guidance, which is conducive to the effective reasoning of the generative question-answering model, so that the obtained judgment results and corresponding treatment plans are more efficient and accurate.

Description

Technical Field

[0001] The present application relates to the field of vehicle technology, and in particular to a vehicle abnormality handling method, device, new energy vehicle, computer-readable storage medium and computer program product. Background Art

[0002] At present, automobile fault handling mainly involves users observing abnormal vehicle performance and consulting after-sales service for problem troubleshooting and repair. The solution to the problem mainly depends on the level of technicians, maintenance manuals, and manufacturer guidance. There are also some solutions that use knowledge spectra to query fault information and repair plans.

[0003] However, the above solutions all have the problem of low fault handling efficiency. Summary of the invention

[0004] In view of this, a vehicle abnormality handling method, device, new energy vehicle, computer-readable storage medium and computer program product are provided, which can improve the efficiency of vehicle fault handling.

[0005] In a first aspect, the present application provides a vehicle abnormality handling method, comprising:

[0006] Receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0007] Based on the pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information;

[0008] Acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0009] Generate corresponding prompt words based on user question information, reference processing information and health status information;

[0010] The prompt words are input into a pre-trained generative question-answering model, and the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan are obtained based on the generative question-answering model.

[0011] In one embodiment, the method further comprises:

[0012] Output the determination result and processing plan through the vehicle terminal; and,

[0013] In the case where the determination result indicates that the abnormal performance is a vehicle failure, the vehicle failure information and the processing solution are sent to the target platform, so as to send the vehicle failure information and the processing solution to the maintenance service provider through the target platform.

[0014] In one embodiment, the knowledge base is a vector knowledge base; based on the pre-built knowledge base, the reference processing information corresponding to the user's question information is searched and obtained, including:

[0015] Vectorize the user's question information into text to obtain the question vector;

[0016] Based on the question vector and the user question information, a vector search and a text search are performed in the vector knowledge base to obtain reference processing information corresponding to the user question information.

[0017] In one embodiment, the vector knowledge base is constructed by the following steps:

[0018] Acquire reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information;

[0019] Extract metadata of reference data; metadata is used to distinguish different reference data;

[0020] Divide the reference data into text blocks, and perform vectorization processing on each obtained text block data to obtain a block vector;

[0021] A vector knowledge base is constructed based on metadata, text block data and block vectors.

[0022] In an exemplary embodiment, when the determination result indicates that the abnormal performance is a vehicle failure, an actual solution to the vehicle failure is obtained, and the vector knowledge base is updated based on the actual solution.

[0023] In one embodiment, the generative question answering model is trained by:

[0024] Acquire reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information;

[0025] The reference data is annotated to obtain a question-and-answer data sample; each question-and-answer data sample at least includes a question-and-answer round, a first question-and-answer object, a first question-and-answer content, a second question-and-answer object, and a second question-and-answer content; wherein the second question-and-answer content is a response to the first question-and-answer content;

[0026] Based on the question and answer data samples, the generative model is fine-tuned to obtain a generative question and answer model.

[0027] In one embodiment, the corresponding prompt words are generated based on the user question information, the reference processing information and the health status information, including:

[0028] Generate context information based on reference processing information and health status information, and generate prompt information based on user question information;

[0029] Prompt words are generated using user role information, context information, and prompt information of a pre-set generative question-answering model.

[0030] In an exemplary embodiment, receiving user question information includes:

[0031] Receive the user's voice question and convert the voice question into corresponding question text;

[0032] Identify the problem domain based on the question text;

[0033] When the question domain belongs to the vehicle maintenance domain, the question text is determined as the user question information.

[0034] In a second aspect, the present application also provides a vehicle abnormality handling device, comprising:

[0035] A question receiving module is used to receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0036] A search module, used to search for reference processing information corresponding to the user's question information based on a pre-built knowledge base;

[0037] A status information acquisition module is used to acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0038] A prompt word generation module, used to generate corresponding prompt words based on user question information, reference processing information and health status information;

[0039] The question-answering module is used to input prompt words into a pre-trained generative question-answering model, and obtain a determination result of the abnormal performance of the vehicle and a corresponding treatment plan based on the generative question-answering model.

[0040] In a third aspect, the present application further provides a new energy vehicle, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0041] Receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0042] Based on the pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information;

[0043] Acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0044] Generate corresponding prompt words based on user question information, reference processing information and health status information;

[0045] The prompt words are input into a pre-trained generative question-answering model, and the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan are obtained based on the generative question-answering model.

[0046] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the following steps are implemented:

[0047] Receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0048] Based on the pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information;

[0049] Acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0050] Generate corresponding prompt words based on user question information, reference processing information and health status information;

[0051] The prompt words are input into a pre-trained generative question-answering model, and the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan are obtained based on the generative question-answering model.

[0052] In a fifth aspect, the present application further provides a computer program product, including a computer program, which implements the following steps when executed by a processor:

[0053] Receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0054] Based on the pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information;

[0055] Acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0056] Generate corresponding prompt words based on user question information, reference processing information and health status information;

[0057] The prompt words are input into a pre-trained generative question-answering model, and the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan are obtained based on the generative question-answering model.

[0058] The above-mentioned vehicle abnormality processing method, device, new energy vehicle, computer-readable storage medium and computer program product receive user question information containing a description of the abnormal performance of the vehicle, first search for reference processing information based on a pre-built knowledge base, and obtain the health status information of the vehicle from the target platform associated with the vehicle, then generate corresponding prompt words based on the user question information, reference information and health status information, and finally input the prompt words into the pre-trained generative question-answering model, and obtain the judgment result of the abnormal performance of the vehicle and the corresponding processing solution suggestion based on the reasoning ability of the generative question-answering model. By introducing the health status information of the vehicle and combining the reference processing information searched in the pre-built knowledge base to generate corresponding prompt words, the generative question-answering model is provided with richer context information and more accurate guidance, which is conducive to the effective reasoning of the generative question-answering model, so that the obtained judgment results and corresponding processing solutions are more efficient and accurate, thereby accelerating the processing efficiency of vehicle failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0060] Figure 1 A diagram showing an application environment of a vehicle abnormality handling method in an embodiment;

[0061] Figure 2 A schematic diagram of a process flow of a vehicle abnormality handling method in one embodiment;

[0062] Figure 3 A schematic diagram of a process of building a vector knowledge base in one embodiment;

[0063] Figure 4 is a flowchart of a method for training a generative question answering method in another embodiment;

[0064] Figure 5 A schematic diagram of constructing a RAG knowledge base in another embodiment;

[0065] Figure 6 A detailed flowchart of a vehicle abnormality handling method in another embodiment;

[0066] Figure 7 is a structural block diagram of a vehicle abnormality processing device in one embodiment;

[0067] Figure 8 Schematic diagram of the internal structure of a new energy vehicle in one embodiment. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0069] The vehicle abnormality handling method provided in the embodiment of the present application can be applied to Figure 1 In the application environment shown, a user can have an interactive conversation with the vehicle 102. The vehicle 102 communicates with the server 104 via a network, and the target platform associated with the vehicle 102 can be integrated inside the server 104. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on the cloud or other network servers.

[0070] Among them, vehicle 102 can receive user question information describing abnormal performance of the vehicle, and search for reference processing information corresponding to the user question information based on a pre-built knowledge base. Thereafter, vehicle 102 obtains recorded health status information of vehicle 102 from the target platform through interaction with server 104, and generates prompt words containing more contextual information based on the user question information, reference processing information and health status information, and inputs the prompt words into a pre-trained large model. Based on the reasoning ability of the large model, a determination result of the abnormal performance of the vehicle and corresponding processing solution recommendations are obtained.

[0071] Alternatively, in other cases, vehicle 102 receives user question information describing abnormal performance of the vehicle, and sends the user question information to server 104. Server 104 searches for reference processing information corresponding to the user question information based on a pre-built knowledge base. Server 104 then obtains recorded health status information of vehicle 102 from the target platform, and generates prompt words containing more context information based on the user question information, reference processing information, and health status information. The prompt words are input into a pre-trained large model, and based on the reasoning ability of the large model, a determination result of the abnormal performance of the vehicle and a corresponding processing solution suggestion are obtained. Finally, server 104 sends the determination result and the corresponding processing solution suggestion to vehicle 102.

[0072] The server 104 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services.

[0073] In an exemplary embodiment, Figure 2 As shown, a vehicle abnormality method is provided, which is applied to Figure 1 Taking the vehicle 102 in the example as an example, the method includes the following steps S201 to S205:

[0074] Step S201, receiving user question information; the user question information includes a description of the abnormal performance of the vehicle.

[0075] The abnormal performance of the vehicle may be understood as noise, vibration, steering, difficulty in starting, brake failure, etc. The received user question information may be a text message or a voice message.

[0076] For example, the vehicle 102 can interact with the user through voice. For example, when the vehicle 102 cannot start normally, the user can call the smart assistant preset on the vehicle side and directly tell the smart assistant his or her question. The smart assistant receives the voice question, converts and recognizes the voice question into text, and obtains the user's question information. Alternatively, the user interacts with text information through the vehicle terminal, and the vehicle 102 recognizes the obtained text information to obtain the user's question information.

[0077] This application improves the user's interactive experience through a variety of user question information receiving methods, and can quickly receive relevant questions to ensure timely response to user needs.

[0078] Step S202: Based on the pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information.

[0079] The knowledge base may be understood as a database storing multiple question-answer pairs, and the reference processing information may be understood as answer data in one or more question-answer pairs. In some embodiments, the knowledge base may be constructed based on manuals or cases related to vehicle maintenance.

[0080] The knowledge base may be a vector knowledge base, including vector indexes, vector data, text indexes, text data, and metadata for distinguishing the source of text data. The vector knowledge base supports the search of vectors and texts, thereby improving the accuracy and completeness of search results.

[0081] Exemplarily, after receiving the user's question information, the vehicle controller searches for multiple reference processing information corresponding to the user's question information based on the knowledge base pre-loaded in the vehicle or deployed in the cloud / server. Since the knowledge base is constructed based on verified vehicle abnormality or fault processing solution information, the reference processing information obtained based on the knowledge base search meets the accuracy requirements. The present application first obtains these reference processing information, laying a data foundation for the subsequent generation of accurate prompt words.

[0082] Step S203, obtaining the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle.

[0083] The target platform can be understood as a vehicle health reporting platform, whose main function is to monitor and manage the health status of the vehicle and provide users with information about vehicle performance and maintenance through data analysis. The health status information can be understood as information about the vehicle in various aspects and performance indicators.

[0084] As an example, the vehicle 102 can achieve information exchange with the target platform through a wireless network. For example, the vehicle 102 can upload some status data to the target platform in real time or periodically through the network, and the target platform can save the data uploaded by the vehicle; in addition, the target platform can also obtain the specific health status of the vehicle based on data analysis and return it to the vehicle 102 through the wireless network. It can be understood that the vehicle 102 can also achieve information exchange with the target platform in other ways, and this application does not limit this.

[0085] Step S204: Generate corresponding prompt words based on the user question information, reference processing information and health status information.

[0086] Among them, the prompt words can be understood as the input provided to the model to guide the model to reason. For example, the prompt words can clearly indicate the type or topic of the expected model output. By constructing high-quality prompt words, the accuracy of the model output results can be improved.

[0087] In some embodiments, the prompt word may be generated based on a pre-set prompt word template. For example, the prompt word template may include a role indication portion, a requirement expression portion, a task instruction portion, and related context information.

[0088] In an exemplary embodiment, the vehicle controller integrates vehicle health status information recorded based on the target platform, user question information, and reference processing information obtained based on a knowledge base search, and generates prompt words with clear instructions and complete context information based on a preset prompt word template, thereby improving the accuracy of the model output results.

[0089] Step S205: input the prompt word into a pre-trained generative question-answering model, and obtain a determination result of the abnormal performance of the vehicle and a corresponding treatment plan based on the generative question-answering model.

[0090] Among them, the generative question-answering model can be understood as a trained large language model with reasoning and question-answering capabilities.

[0091] For example, the vehicle controller inputs the prompt words obtained in the above manner into the pre-trained generative question-answering model, and obtains the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan based on the reasoning ability and question-answering ability of the generative question-answering model. Since the prompt words carry reference information and clear instruction information searched in combination with the knowledge base, the model hallucination problem that may exist in the generative question-answering model is effectively avoided, thereby improving the judgment accuracy and processing efficiency of the generative question-answering model, and providing users with more accurate question-answering content.

[0092] In the above vehicle abnormality processing method, user question information containing a description of the abnormal performance of the vehicle is received, reference processing information is first searched based on a pre-built knowledge base, and the health status information of the vehicle is obtained from a target platform associated with the vehicle, and then corresponding prompt words are generated based on the user question information, reference information and health status information, and finally the prompt words are input into a pre-trained generative question-answering model, and based on the reasoning ability of the generative question-answering model, the judgment result of the abnormal performance of the vehicle and the corresponding treatment plan suggestion are obtained. By introducing the health status information of the vehicle and combining it with the reference processing information searched in the pre-built knowledge base, the corresponding prompt words are generated, giving the generative question-answering model richer context information and more accurate guidance, which is conducive to the effective reasoning of the generative question-answering model, so that the obtained judgment results and corresponding treatment plans are more efficient and accurate, thereby accelerating the efficiency of vehicle fault processing.

[0093] In one embodiment, the aforementioned knowledge base is a vector knowledge base; based on the knowledge base, reference processing information corresponding to the user's question information is searched, including: text vectorizing the user's question information to obtain a question vector; based on the question vector and the user's question information, vector search and text search are performed in the vector knowledge base to obtain reference processing information corresponding to the user's question information.

[0094] Among them, the vector knowledge base can be understood as a knowledge base that stores vector information, and fast vector search can be performed by calculating the similarity between vectors; vector search can be understood as the process of retrieval and matching using vector representation, which mainly relies on converting data (such as text, images, audio, etc.) into high-dimensional vectors and finding matches by calculating the similarity between vectors; text search can be understood as the process of finding relevant information in text data through specific query words or phrases.

[0095] In an exemplary embodiment, the vehicle controller inputs the user's question information into an open source vectorization model to perform text vectorization, obtain the corresponding question vector, and perform vector search and text search in the vector knowledge base based on the question vector and the user's question information. For example, the specific search method can be: respectively calculate the distance between the question vector and each vector in the vector knowledge base, select the vector with the shortest distance or the distance that meets the set conditions as the target vector most similar to the question vector, and at the same time, perform keyword matching retrieval on the text information stored in the vector knowledge base according to the keywords in the user's question information, determine the matching target text information, and determine the processing information that associates the target vector with the target text information in the vector knowledge base as the reference processing information corresponding to the user's question information. Using two search methods to search for reference processing information at the same time can increase the search speed, and at the same time increase the comprehensiveness of the search, improve the accuracy of the reference processing information, and thus improve the efficiency of handling vehicle faults.

[0096] In an exemplary embodiment, Figure 3 As shown, the vector knowledge base is constructed through steps S301 to S304:

[0097] Step S301, obtaining reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information.

[0098] Step S302: extract metadata of the reference data; the metadata is used to distinguish different reference data.

[0099] Among them, vehicle maintenance manual information can be understood as detailed knowledge and guidance related to the maintenance, repair and care of various types of vehicles, and vehicle user manual information can be understood as guidance and instructions provided to vehicle owners and drivers on the use, maintenance and safe operation of specific models; the difference in different reference data can be understood as different data sources or reference data with different data contents;

[0100] Metadata can be understood as leading descriptive information or characteristic information fields used to describe data content. Different data contents correspond to different leading descriptive information. Therefore, metadata can be used to distinguish different reference data, such as file names, chapters, rounds, conversation content, question and answer objects, question and answer content and other characteristic information fields. For example, the file name characteristic information field can be used to distinguish reference data from different sources, and the chapter, round and other characteristic information fields can be used to distinguish reference data of data content from different sources.

[0101] Exemplarily, the server obtains reference data related to vehicle maintenance, the reference data including at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information, and then extracts metadata from the reference data for distinguishing different reference data, thereby realizing categorized storage of the reference data, ensuring the regularity of the data, and laying a solid data foundation for the subsequent construction of a vector knowledge base.

[0102] Step S303, dividing the reference data into text blocks, and performing vectorization processing on each obtained text block data to obtain a block vector.

[0103] Step S304: construct a vector knowledge base based on the metadata, text block data and block vectors.

[0104] In an exemplary embodiment, the vehicle controller performs text segmentation on the reference data according to sentence segmentation to obtain multiple text segmentation data, and the data volume of the text segmentation data must meet the set threshold and the text segmentation data can also retain the complete semantics of a sentence. After that, each text segmentation data is input into an open source vectorization model for vectorization processing to obtain a segmentation vector. The text index of each text segmentation data and the vector index of the segmentation vector are constructed respectively. Finally, a vector knowledge base is constructed based on metadata, text segmentation data, text index, segmentation vector and vector index. The vector knowledge base with index can quickly find the corresponding search results when used, laying the foundation for accelerating the efficiency of vehicle fault handling.

[0105] In one of the embodiments, the method also includes: outputting a determination result and a processing plan through a vehicle-mounted terminal; and, when the determination result characterizes that the abnormal performance is a vehicle failure, sending the vehicle failure information and the processing plan to a target platform, so as to send the vehicle failure information and the processing plan to a maintenance service provider through the target platform.

[0106] For example, the vehicle controller outputs the determination result and the processing scheme through the vehicle terminal to notify the user, and, when the determination result indicates that the abnormal performance is a vehicle fault, the vehicle fault information and the processing scheme are sent to the target platform, so that the vehicle fault information and the processing scheme are sent to the maintenance service provider through the target platform. At the same time, the contact information of the maintenance service provider can also be sent to the user to facilitate the user to contact the maintenance service provider to handle the vehicle fault. Alternatively, the target platform can also send relevant information of multiple maintenance service providers to the vehicle terminal for the user to select. The above method ensures timely feedback on the corresponding vehicle status. Secondly, through the dual-end information notification of the maintenance service provider and the user, it is conducive to the timely handling of vehicle faults and improves the efficiency of handling vehicle faults.

[0107] In one embodiment, the method further includes: when the determination result indicates that the abnormal performance is a vehicle failure, obtaining an actual solution to the vehicle failure, and updating the vector knowledge base based on the actual solution.

[0108] In an exemplary embodiment, when the judgment result of the large model indicates that the abnormal performance of the vehicle is a vehicle failure, the actual solution of the maintenance service provider to the vehicle failure is obtained through the above target platform, and the actual solution is returned to the vector knowledge base, and the corresponding question and answer data pairs are dynamically updated. The accuracy of the vector knowledge base is enhanced in the above manner.

[0109] In an exemplary embodiment, the generative question answering model is trained through steps S401 to S403, wherein:

[0110] Step S401, obtaining reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information.

[0111] Step S402, annotate the reference data to obtain question and answer data samples; each question and answer data sample at least includes a question and answer round, a first question and answer object, a first question and answer content, a second question and answer object and a second question and answer content; wherein the second question and answer content is a response content to the first question and answer content.

[0112] Among them, the first question and answer object can be understood as the object that initiates questions in the first round, the first questioning object can be understood as the user, and the first question and answer content can be understood as the content of questions in the first round. For example, when it is known that a component is faulty but the specific location of the specific component is unclear, the user can initiate a question about what specific components a component has.

[0113] Similarly, the second question and answer object can be understood as the object that answers the questions in the first round, the second question object can be understood as a large model or an intelligent assistant, and the second question and answer content can be understood as a reference answer to the question content in the first round, for example, it can be the fault handling method or fault handling steps that should be performed when a certain component fails.

[0114] For example, vehicle 102 obtains reference data related to vehicle maintenance, the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information, and then annotates the reference data to obtain question-answer data samples, each of which includes at least question-answer rounds, first question-answer objects, second question-answer content, second question-answer objects, and second question-answer content. By annotating the data and using a large amount of annotated data for fine-tuning the large model, the trained large model (i.e., the generative question-answer model) has stronger understanding and reasoning capabilities in the field of vehicle maintenance technology.

[0115] Step S403: fine-tune the generative model based on the question and answer data sample to obtain a generative question and answer model.

[0116] In an exemplary embodiment, the generative model is trained by using question and answer data samples in small batches, and the generative model is gradually fine-tuned to avoid the model falling into a local optimal state in the early stage of fine-tuning. Finally, when the maximum number of iterations is reached or the model error meets the error threshold, a generative question and answer model is obtained. The above method further improves the understanding and reasoning capabilities of the generative question and answer model in the field of vehicle maintenance technology.

[0117] In one embodiment, context information in a prompt word template may be generated based on reference processing information and health status information, and task prompt information in a prompt word template may be generated based on user question information, and corresponding prompt words may be generated in combination with preset role information in the prompt word template.

[0118] Among them, the role information preset in the prompt word template can be understood as professional technicians, professional maintenance personnel, etc.

[0119] For example, the vehicle controller generates context information based on reference processing information and health status information, and generates prompt information based on user question information, and finally generates prompt words using the user role information, context information and prompt information of the pre-set generative question-answering model. The prompt words generated in the above manner can enhance context awareness, and at the same time, combined with the reference processing information searched in the knowledge base, can effectively avoid the model hallucination problem that may occur in the subsequent use of the generative question-answering model, thereby improving the accuracy of the model output.

[0120] In one embodiment, receiving user question information includes: receiving the user's question voice, converting the question voice into corresponding question text; identifying the problem domain based on the question text; and when the problem domain belongs to the vehicle maintenance domain, determining the question text as the user's question information.

[0121] For example, the vehicle receives the user's voice question and converts the voice question into corresponding question text, and then identifies the corresponding problem field based on the question text. If the problem field belongs to the vehicle maintenance field, the question text is determined as the user's question information, and the subsequent vehicle abnormality processing continues; if the problem field does not belong to the vehicle maintenance field, other corresponding processing is performed.

[0122] In an exemplary embodiment, a specific implementation of a vehicle abnormality handling method is provided. It can be understood that the definitions and code examples given below are specific descriptions of a certain situation, and do not limit the vehicle abnormality handling method to be implemented only in this situation. Among them:

[0123] 1. Data annotation:

[0124] By annotating data from vehicle maintenance manuals, maintenance cases, user manuals, general motors, etc., it is used for fine-tuning training of large vehicle maintenance models.

[0125] For example, based on the maintenance manual data of a specific vehicle model, equipment fault diagnosis question and answer data is constructed. The characteristic information fields (constituting the aforementioned metadata) mainly include reference manual (referenceTo), chapter (chapter), category (category), subcategory (subCategory), rounds (rounds), conversation content (conversations), question and answer object (from), and question and answer content (value). The following is a Json example, which is a specific example given based on the above definition:

[0126] {

[0127] "category": "Electrical Appliances",

[0128] "subCategory": "Body Electronic Control System-Window and Sunroof Control Problems",

[0129] "chapter": "Diagnosis of sunroof control system faults - main component structure",

[0130] "referenceTo": "Detailed Analysis of Failures of Automotive Electrical Equipment_Volume 2",

[0131] "rounds": "1",

[0132] "conversations": [

[0133] {

[0134] "from": "user", / / user represents the user who asked the question

[0135] "value": "What components does the sliding sunroof motor assembly include?"

[0136] },

[0137] {

[0138] "from": "assistant", / / where assistant represents the machine assistant who answers questions

[0139] "value": "Hello! The sliding sunroof motor assembly includes the sliding sunroof motor, drive gear (worm gear assembly) and sliding sunroof control ECU."

[0140] } ]

[0142] }

[0143] According to the maintenance manual data of specific models, equipment maintenance question and answer data is constructed. The characteristic information fields mainly include the reference manual name (referenceTo), chapter (chapter), rounds (rounds), conversation content (conversations), question and answer object (from), and question and answer content (value). The marked Json example is as follows, which is a specific example given according to the above definition:

[0144] {

[0145] "chapter": "2.5.5.8-DTC: P9B03063",

[0146] "referenceTo": "CAR-001 Model Maintenance Manual 2024 Edition",

[0147] "rounds":"1",

[0148] "conversations": [

[0149] {

[0150] "from": "user",

[0151] "value": "What are the steps to handle the CAN BusOff fault in the xxx motor control system?"

[0152] },

[0153] {

[0154] "from": "assistant",

[0155] "value": "The steps to handle the CAN BusOff fault in the xxx motor control system are: Step 1 Check xxxxx. If normal, proceed to step 2 Step 2 Check xxxxx. If normal, proceed to step 3 Step 3: Check whether the power supply and grounding wires of the drive motor and controller are normal. If normal, proceed to step 4. Otherwise, repair or replace the wiring harness. Step 4 Replace the drive motor and controller two-in-one. A Replace the drive motor and controller two-in-one. B After replacing the drive motor and controller two-in-one, check whether the fault is eliminated. If it is eliminated, the system is normal. Step 5 The diagnosis is complete. If the fault is eliminated, the system is normal and the diagnosis is completed.

[0156] } ]

[0158] }

[0159] 2. Fine-tune the large vehicle maintenance model (i.e. the aforementioned generative question-answering model):

[0160] We selected the pre-trained version of the open source large model llama3 (the aforementioned generative question-answering model) and the labeled data for model fine-tuning training. The trained maintenance large model already has certain vehicle maintenance question-answering capabilities, but there may be problems such as model hallucinations, which have a certain impact on the accuracy of the questions.

[0161] 3. Establish a vehicle maintenance RAG knowledge base (Retrieval-Augmented Generation), which is the aforementioned vector knowledge base:

[0162] like Figure 5 As shown in the figure, a RAG knowledge base construction process is provided. The vehicle maintenance manual, maintenance case, user manual, vehicle status and other data are vectorized and stored. The specific steps are as follows:

[0163] (1) Data extraction:

[0164] Obtain data such as vehicle maintenance manuals, maintenance cases, user manuals, vehicle status, etc., perform normalization processing such as cleaning, deduplication, and classification, and extract metadata such as file names and chapters.

[0165] (2) Text segmentation:

[0166] The text is divided into chunks using sentence segmentation, and the length of tokens is less than 2k to retain the complete semantics of a sentence.

[0167] (3) Vectorization (embedding):

[0168] The BGE-M3 open source vectorization model is used to convert the segmented text and other data into a vector matrix, where the model output dims dimension is 1024.

[0169] (4) Data storage:

[0170] Select ElasticSearch8 as the vectorized data repository, create ElasticSearch text and vectorized indexes, and construct the Json example as follows:

[0171] {

[0172] "mappings": { / / Define the mapping rules in the index

[0173] "properties": { / / Define the properties in the mapping

[0174] "content_vector": { / / Define content vector properties

[0175] "type": "dense_vector", / / Data type is dense vector

[0176] "dims": 1024, / / The dimension of the vector is 1024

[0177] "index": true, / / Enable index to support similarity search

[0178] "similarity": "l2_norm" / / Use L2 norm as similarity measure

[0179] },

[0180] "content": { / / Define text content attributes

[0181] "type" : "text" / / Data type is text

[0182] },

[0183] "reference" : { / / Define reference manual properties

[0184] "type" : "text" / / Data type is text

[0185] },

[0186] "chapter": { / / Define chapter attributes

[0187] "type" : "text" / / Data type is text

[0188] }

[0189] }

[0190] }

[0191] }

[0192] Among them, content_vector is the vectorized data storage field, type = dense_vector means to establish a vectorized index, the dims value is consistent with the dimension of BGE-M3 output, index is set to true to indicate that it can be indexed, and similarity uses l2_norm (Euclidean distance) as the similarity algorithm between documents. content is the text block content field, reference and chapter are metadata content fields, which are used to identify which chapter of which reference manual the relevant content belongs to. Storing text and vectorized data in the index library completes data storage.

[0193] Through the above operations, the vehicle maintenance RAG knowledge base is constructed.

[0194] 4. Build a vehicle maintenance RAG question-answering system:

[0195] like Figure 6 As shown, a detailed flow chart of a vehicle abnormality handling method is provided. Through the previously constructed vehicle maintenance model and knowledge base, a vehicle maintenance question-and-answer system can be constructed. The specific steps are as follows:

[0196] (1) Knowledge search:

[0197] Construct a user question search request. The parameters include: userId (user ID), requestType (request type, including vehicle model parameters), data (request content), role (role corresponding to the question), rounds (rounds), and conversation content (content). The content in the request is embedded and vectorized using the BGE vector model to obtain content_vector. The request parameters are as follows:

[0198] {

[0199] "userId": "xxxxxxxxx",

[0200] "requestType": "CAR-001",

[0201] "data": {

[0202] "role": "user",

[0203] "rounds":"1",

[0204] "content": "What should I do if my car has a CAN BusOff alarm problem?"

[0205] "content_vector": [

[0206] 0.008829478174448013,

[0207] -0.029355255886912346,

[0208] -0.025615187361836433,

[0209] xxxxx ]

[0211] }

[0212] }

[0213] Among them, userId (user ID) and requestType (request type, including vehicle model parameters) can be obtained through the vehicle system. Users need to log in to the system before using the vehicle, so user information is stored in the system. When the system is installed when the vehicle leaves the factory, the vehicle model parameters can be stored in the system.

[0214] Use the request parameters to search the knowledge base, where ElasticSearch uses a hybrid search of the kNN-search algorithm vector search and full-text search to recall search results and obtain the corresponding topN (first N) knowledge data results:

[0215] [{

[0216] "chapter": "2.5.5.8-DTC: P9B03063",

[0217] "referenceTo": " "CAR-001 Model Maintenance Manual 2024 Edition",

[0218] "content": "The steps to handle the CAN BusOff fault in the motor control system are: Step 1 Check xxxxx. If normal, proceed to step 2 Step 2 Check xxxxx. If normal, proceed to step 3 Step 3 Check whether the power supply and grounding wires of the drive motor and controller are normal. If normal, proceed to step 4. Otherwise, repair or replace the wiring harness. Step 4 Replace the drive motor and controller two-in-one. A Replace the drive motor and controller two-in-one. B After replacing the drive motor and controller two-in-one, check whether the fault is eliminated. If it is eliminated, the system is normal. Step 5 The diagnosis is complete. If the fault is eliminated, the system is normal and the diagnosis is completed.

[0219] }]

[0220] (2) VHR platform accesses vehicle health data:

[0221] As a vehicle health data recorder, the VHR platform records the health status of the vehicle, including historical driving data, faults, repairs, etc. The health information of the vehicle can be obtained by obtaining the historical status records of the vehicle.

[0222] (3) Prompt construction:

[0223] The user request, vehicle health information and knowledge base results are constructed into a prompt to obtain the following request result:

[0224] "If you are an auto repair expert, and there is a problem with a car model called CAR-001, its user asks: What should I do if my car has a CAN BusOff alarm problem? The vehicle's health status information: xxx. According to the query results of the 2024 edition of the Maintenance Manual of the CAR-001 model: The steps to deal with the CANBusOff fault in the motor control system of the model with the CAN BusOff alarm problem are: Step 1 Check xxxxx. If normal, proceed to step 2 Step 2 Check xxxxx. If normal, proceed to step 3 Step 3 Check whether the power supply and grounding wires of the drive motor and controller are normal. If normal, proceed to step 4. Otherwise, repair or replace the wiring harness. Step 4 Replace the drive motor and controller two-in-one. A Replace the drive motor and controller two-in-one. B After replacing the drive motor and controller two-in-one, check whether the fault is eliminated. If it is eliminated, the system is normal. Step 5 The diagnosis is complete. If the fault is eliminated, the system is normal and the diagnosis is completed. Please provide this user with a repair suggestion based on the above information."

[0225] It can be seen that the above prompt words include: user role information of the set automobile maintenance model, more context information, and the final task prompt information.

[0226] (4) Large model result generation:

[0227] Input the prompt question information and the user's context information into the big model for question and answer. You can use frameworks such as langchain or LlamaIndex to build the process. Output result:

[0228] "Of course! Here are some suggestions for step-by-step handling of the CAN BusOff warning problem in your vehicle: ###Step 1: Check the fault code status 1. **Connect the diagnostic tool to the OBD interface**: - Find the vehicle's OBD interface and insert the diagnostic tool. 2. **Perform the vehicle's ON power-on procedure**: - Turn the vehicle's ignition switch to the ON position, but do not start the engine. 3. **Read the fault code and record it**: - Use the diagnostic tool to read the vehicle's fault code and record these codes. 4. **Clear the fault code**: - Use the diagnostic tool to clear the fault code. 5. **Read the fault code again**: - After clearing the fault code, read the fault code again. 6. **Check whether the same fault code is displayed**: - If the same fault code is still displayed, continue to step 2, otherwise it may be an intermittent fault and further check the cause of the intermittent fault. ###Step 2: Check the CAN data communication wires of the drive motor and controller 2-in-1 1. **Perform the vehicle power-off procedure**: - Disconnect the power of the vehicle, including removing the key and disconnecting the power. 2. **Disconnect the FB10 harness connector of the drive motor and controller two-in-one**: - Find the FB10 harness connector of the drive motor and controller two-in-one and disconnect it. 3. **Measure the resistance value**: - Use a multimeter to measure the resistance value between terminal 18 and terminal 19 of the FB10 harness connector of the drive motor and controller two-in-one. - Standard resistance value: about 120Ω. 4. **Confirm whether the resistance meets the standard value**: - If the resistance value meets the standard value, proceed to step 3; - If it does not meet the standard value, it means that the EVCAN communication line may be short-circuited and the wiring harness needs to be repaired or replaced. ### Step 3: Check the power supply and grounding wire of the drive motor and controller two-in-one 1. **Check whether the power supply and grounding wire are normal**: - Make sure that the power supply and grounding wire of the drive motor and controller two-in-one is connected normally. - If the power supply and grounding wire are normal, proceed to step 4; - If the power supply or grounding wire is abnormal, the wiring harness needs to be repaired or replaced. ### Step 4: Replace the drive motor and controller two-in-one 1. **Replace the drive motor and controller two-in-one**: -Replace the drive motor and controller 2-in-1, as it may be defective. 2. **Is the fault eliminated**: -After replacement, check whether the fault has been eliminated; -If the fault is eliminated, the system returns to normal. ### Step 5: End of diagnosis 1. **If the fault is eliminated**: -After the fault is eliminated, the system returns to normal and the diagnosis ends. If you encounter any difficulties while performing these steps, it is recommended that you seek professional car repair services to ensure the accuracy and safety of the repair. "

[0229] At this point, the vehicle maintenance RAG question-and-answer system has been completed.

[0230] (5) User Q&A dialogue:

[0231] The user interacts with the vehicle computer through voice, the audio is converted into text through voice recognition, and the recognition results of the vehicle maintenance field are forwarded to the vehicle maintenance question and answer system through field recognition. Finally, the vehicle maintenance answers are returned to the user.

[0232] (6) Question push:

[0233] The large model combines vehicle health status information and user questions for reasoning, pushes questions and answers to users, and determines whether there is a problem with the current vehicle. If so, the problem data will be pushed to the VHR platform. After the VHR platform confirms the problem, it will push the vehicle fault information and maintenance plan directly to the official store near the vehicle, and the official will directly contact the user for follow-up processing. Subsequent maintenance data will be fed back to the platform for subsequent improvement of the RAG knowledge base.

[0234] Compared with the prior art, this application has the following advantages:

[0235] 1. Use large models to increase knowledge judgment and reasoning, improve the accuracy of the output processing plan, and thus improve the efficiency of vehicle fault processing.

[0236] 2. In the event of a vehicle failure, the actual solution to the failure will be returned to the knowledge base to resolve the problem of incomplete knowledge base, providing a more user-friendly experience.

[0237] 3. Use full-text search + vector search to build a knowledge base to improve search accuracy.

[0238] 4. Combined with the vehicle health status information, it can accurately restore user problems and promptly determine abnormal vehicle performance, ensuring timely discovery of faults and efficient handling, thereby improving user experience and safety.

[0239] 5. Combining the vehicle maintenance large model and knowledge base, the model hallucination problem that may exist in the reasoning of the vehicle maintenance large model is solved, thereby improving the accuracy of the model output.

[0240] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0241] Based on the same inventive concept, the embodiment of the present application also provides a vehicle abnormality handling device for implementing the vehicle abnormality handling method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more vehicle abnormality handling device embodiments provided below can refer to the limitations of the vehicle abnormality handling method above, and will not be repeated here.

[0242] In an exemplary embodiment, Figure 7 As shown, a vehicle abnormality processing device is provided, including: a question receiving module 701, a search module 702, a state information acquisition module 703, a prompt word generation module 704 and a question and answer module 705, wherein:

[0243] The question receiving module 701 is used to receive user question information; the user question information includes a description of the abnormal performance of the vehicle;

[0244] A search module 702, for searching for reference processing information corresponding to the user's question information based on a pre-built knowledge base;

[0245] The status information acquisition module 703 is used to acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle;

[0246] A prompt word generation module 704, used to generate corresponding prompt words based on user question information, reference processing information and health status information;

[0247] The question-answering module 705 is used to input the prompt words into the pre-trained generative question-answering model, and obtain the determination result of the abnormal performance of the vehicle and the corresponding processing plan based on the generative question-answering model.

[0248] Based on the vehicle abnormality handling device of the above embodiment, the question receiving module receives user question information containing a description of the abnormal performance of the vehicle, first searches for reference processing information corresponding to the user question information from a pre-built knowledge base based on the search module, obtains the health status information of the vehicle from a target platform associated with the vehicle based on the status information acquisition module, and then sends the health status information to the prompt word generation module, which generates corresponding prompt words based on the user question information, the reference processing information and the health status information, and then inputs the prompt words into the pre-trained generative question-answering model through the question-answering module, and obtains the judgment result and corresponding processing solution of the abnormal performance of the vehicle based on the generative question-answering model. Based on the prompt words of the present application, the generative question-answering model is provided with richer context information and more accurate guidance, which is conducive to the effective reasoning of the generative question-answering model, so that the obtained judgment results and corresponding processing solutions are more efficient and accurate, thereby accelerating the processing efficiency of vehicle failures.

[0249] In one embodiment, the knowledge base is a vector knowledge base; the search module 702 is also used to vectorize the user's question information to obtain a question vector; based on the question vector and the user's question information, a vector search and a text search are performed in the vector knowledge base to obtain reference processing information corresponding to the user's question information.

[0250] In an exemplary embodiment, the vehicle abnormality handling device also includes: a knowledge base construction module, which is used to obtain reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information; metadata of the reference data is extracted; the metadata is used to distinguish different reference data; text segmentation of the reference data, and vectorization of each obtained text segmentation data to obtain a segmentation vector; and constructing a vector knowledge base based on the metadata, text segmentation data and segmentation vector.

[0251] In one embodiment, the vehicle abnormality handling device also includes: a sending module, which is used to output the judgment result and the processing plan through the vehicle-mounted terminal; and, when the judgment result characterizes that the abnormality is manifested as a vehicle failure, the vehicle fault information and the processing plan are sent to the target platform, so as to send the vehicle fault information and the processing plan to the maintenance service provider through the target platform.

[0252] In an exemplary embodiment, the sending module is further used to obtain a practical solution to the vehicle failure when the determination result indicates that the abnormal performance is a vehicle failure, and to update the vector knowledge base based on the practical solution.

[0253] In one embodiment, the vehicle abnormality handling device also includes: a model training module, which is used to obtain reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information; the reference data is labeled to obtain a question and answer data sample; each question and answer data sample includes at least a question and answer round, a first question and answer object, a first question and answer content, a second question and answer object, and a second question and answer content, wherein the second question and answer content is a response to the first question and answer content; based on the question and answer data sample, the generative model is fine-tuned to obtain a generative question and answer model.

[0254] In one embodiment, the prompt word generation module 704 is also used to generate context information based on reference processing information and health status information, and to generate prompt information based on user question information; and to generate prompt words using user role information, context information and prompt information of a pre-set generative question-answering model.

[0255] In an exemplary embodiment, the question receiving module 701 is also used to receive the user's question voice, convert the question voice into corresponding question text; identify the problem field based on the question text; when the problem field belongs to the vehicle maintenance field, determine the question text as user question information.

[0256] Each module in the above-mentioned vehicle abnormality handling device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the new energy vehicle in the form of hardware, or can be stored in the memory in the new energy vehicle in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0257] In an exemplary embodiment, a new energy vehicle is provided, and its internal structure diagram can be as follows: Figure 8 As shown. The new energy vehicle includes a processor and a controller. The processor and the memory are connected via a system bus. The processor of the new energy vehicle is used to provide computing and control capabilities. The memory of the new energy vehicle includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the new energy vehicle is used to store user question information, reference processing information, health status information, prompt words, vehicle fault information and processing solution data. The input / output interface of the new energy vehicle is used to exchange information between the processor and an external device. When the computer program is executed by the processor, a vehicle abnormality processing method is implemented.

[0258] Those skilled in the art will understand that Figure 8The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present application, and does not constitute a limitation on the new energy vehicle to which the scheme of the present application is applied. A specific new energy vehicle may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0259] In an exemplary embodiment, a new energy vehicle is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the vehicle abnormality handling method of the above embodiment is implemented when the processor executes the computer program.

[0260] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the vehicle abnormality handling method of the above embodiment is implemented.

[0261] In one embodiment, a computer program product is provided, including a computer program, which implements the vehicle abnormality handling method of the above embodiment when executed by a processor.

[0262] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0263] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0264] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0265] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A vehicle abnormality handling method, characterized in that: The method comprises: Receiving user question information; the user question information includes a description of the abnormal performance of the vehicle; Based on a pre-built knowledge base, search and obtain reference processing information corresponding to the user's question information; Acquire the health status information of the vehicle from the target platform; the target platform is associated with the vehicle and records the health status information of the vehicle; the health status information includes historical driving data, fault information and maintenance information; Generate a corresponding prompt word based on the user question information, the reference processing information and the health status information; the prompt word includes the user role information, context information and task prompt information of the set generative question-answering model, the role information is generated according to the role information preset in the prompt word template, the context information is generated based on the health status information and the reference processing information, and the task prompt information is generated according to the user question information; The prompt word is input into the generative question-answering model, and a determination result of the abnormal performance of the vehicle and a corresponding treatment plan are obtained based on the generative question-answering model.

2. The method according to claim 1, characterized in that The method further comprises: Outputting the determination result and the processing scheme through the vehicle-mounted terminal; and, In the case where the determination result indicates that the abnormal performance is a vehicle fault, the vehicle fault information and the processing solution are sent to the target platform, so as to send the vehicle fault information and the processing solution to a maintenance service provider through the target platform.

3. The method according to claim 1, characterized in that The knowledge base is a vector knowledge base; the search based on the pre-built knowledge base to obtain reference processing information corresponding to the user question information includes: Vectorize the user's question information to obtain a question vector; Based on the question vector and the user question information, a vector search and a text search are performed in the vector knowledge base to obtain reference processing information corresponding to the user question information.

4. The method according to claim 3, characterized in that The vector knowledge base is constructed by the following steps: Acquire reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information; Extracting metadata of the reference data; the metadata is used to distinguish different reference data; Performing text segmentation on the reference data, and performing vectorization processing on each obtained text segmentation data to obtain a segmentation vector; A vector knowledge base is constructed according to the metadata, the text block data and the block vectors.

5. The method according to claim 4, characterized in that The method further comprises: In the case where the determination result indicates that the abnormal performance is a vehicle fault, an actual solution to the vehicle fault is obtained, and the vector knowledge base is updated based on the actual solution.

6. The method according to claim 1, characterized in that The generative question answering model is trained in the following way: Acquire reference data related to vehicle maintenance; the reference data includes at least one of vehicle maintenance manual information, vehicle user manual information, and maintenance record information; Annotate the reference data to obtain question-and-answer data samples; each of the question-and-answer data samples at least includes a question-and-answer round, a first question-and-answer object, a first question-and-answer content, a second question-and-answer object, and a second question-and-answer content; wherein the second question-and-answer content is a response content to the first question-and-answer content; Based on the question and answer data sample, the generative model is fine-tuned to obtain the generative question and answer model.

7. The method according to any one of claims 1 to 6, characterized in that: The receiving of user question information includes: Receive the user's voice question and convert the voice question into a corresponding question text; identifying a problem domain based on the problem text; When the question field belongs to the field of vehicle maintenance, the question text is determined as the user question information.

8. A vehicle abnormality handling device, characterized in that: The device comprises: A question receiving module, used to receive user question information; the user question information includes a description of the abnormal performance of the vehicle; A search module, used to search for reference processing information corresponding to the user's question information based on a pre-built knowledge base; A status information acquisition module, used to acquire the health status information of the vehicle from a target platform; the target platform is associated with the vehicle and records the health status information of the vehicle; the health status information includes historical driving data, fault information and maintenance information; A prompt word generation module, used to generate corresponding prompt words based on the user question information, the reference processing information and the health status information; the prompt words include user role information, context information and task prompt information of the set generative question-answering model, the role information is generated according to the role information preset in the prompt word template, the context information is generated based on the health status information and the reference processing information, and the task prompt information is generated according to the user question information; The question-answering module is used to input the prompt word into the generative question-answering model, and obtain a determination result and a corresponding treatment plan for the abnormal performance of the vehicle based on the generative question-answering model.

9. A new energy vehicle, comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A 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 method according to any one of claims 1 to 7 are implemented.

Images