OTA knowledge base interaction method for automobile

Abstract

The invention discloses an OTA knowledge base interaction method for automobiles, which comprises the following steps: setting a vector database in an OTA platform server, and storing input texts, pictures and videos according to vector data to form an OTA knowledge base; when a user asks a question, the question is converted into a vector, paragraphs possibly containing answers are retrieved through a vector database, and a retrieval context is formed; and setting a large language model LLM, merging the question and the context in a prompt, and submitting the question and the context to the large language model to generate a final reply. According to the method, the problems of information hysteresis and rigid response of the existing automobile OTA knowledge base can be solved, the accuracy and correlation of the knowledge base can be enhanced, the response efficiency of the knowledge base can be improved, and the flexibility and adaptability of the knowledge base can be optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automobile OTA platforms, and more particularly to an OTA knowledge base interaction method for automobiles. BACKGROUND

[0002] The current OTA interaction system of intelligent automobiles is a pre-constructed and stored knowledge base that covers relevant information about vehicle or software functions, common question answers, and fault error guidelines. This information usually exists in various forms such as text, pictures, and videos, making it convenient for users to consult. However, the update of static knowledge bases is not real-time and needs to be manually updated periodically. Once the information in the knowledge base is modified, all applications or systems based on the database need to be updated synchronously. The current OTA interaction knowledge base system is based on traditional static database technology, which follows a linear process and adopts a simple question-and-answer mode, such as how to add a VIN number, how to upload software, and how to issue a strategy. Relevant information can be easily obtained in the pre-set knowledge base. However, it is difficult to handle complex queries that require multi-step logical reasoning and information that does not exist in the OTA knowledge base. If new parts or new vehicle models are connected to the platform, the traditional solution cannot retrieve the information in time and needs to wait for the user to update the database. The resulting information lag and rigid response will seriously affect the operation of the OTA platform if an error occurs. Therefore, it is of great significance to improve the response capability and response efficiency of the knowledge base and provide more accurate responses. SUMMARY

[0003] The present application provides an OTA knowledge base interaction method for automobiles, which solves the problem of information lag and rigid response in existing automobile OTA knowledge bases, enhances the accuracy and relevance of the knowledge base, improves the response efficiency of the knowledge base, and optimizes the flexibility and adaptability of the knowledge base.

[0004] To achieve the above purpose, the present application provides the following technical solutions:

[0005] An OTA knowledge base interaction method for automobiles, comprising:

[0006] Setting a vector database on the OTA platform server, storing input text, pictures, and videos as vector data to form an OTA knowledge base;

[0007] When a user asks a question, converting the question into a vector, retrieving a paragraph that may contain an answer through the vector database, and forming a retrieval context;

[0008] Setting a large language model LLM, merging the question and the context in a prompt, and submitting it to the language model to generate a final reply.

[0009] Preferably, it further comprises:

[0010] By giving the vehicle VIN code and query requirements, automatically obtain the vehicle status, parts information, hardware records, software records from the OTA platform to monitor the real-time data of the vehicle.

[0011] Preferably, it also includes:

[0012] According to the user input vehicle VIN code and demand, verify the user identity and view permission, obtain the latest information from the current state of the vehicle, and retrieve the corresponding repair and upgrade history records from the vector database to form a visual report.

[0013] Preferably, it also includes:

[0014] Through the large language model LLM, the input vehicle information is preprocessed, and text, pictures and videos are actively integrated to perform multi-modal data integration, and then generate comprehensive answers.

[0015] Preferably, the multi-modal data integration includes:

[0016] Standardize the collection of multi-source data;

[0017] Extract multi-modal key features;

[0018] Unified timestamp and location reference;

[0019] Correlation analysis through knowledge graph reasoning and statistical modeling;

[0020] Generate a visual report.

[0021] Preferably, it also includes:

[0022] Automatically convert the effective scheme confirmed by artificial into a knowledge graph node, dynamically adjust the strategy priority according to the success rate of the scheme, and mark the knowledge node that has not been triggered for 6 months as a to-be-verified state, so as to eliminate outdated schemes.

[0023] Preferably, it also includes:

[0024] In fault diagnosis, through real-time fusion of vehicle logs, network status, software version features and software upgrade record data, combined with knowledge graph reasoning and large language model LLM, generate a multi-level root cause analysis report.

[0025] Preferably, the large language model includes:

[0026] In the problem input stage, set up intelligent semantic recognition to identify the key elements of natural language questions;

[0027] Build a multi-source search engine, and perform multi-dimensional information retrieval on user questions through the multi-source search engine;

[0028] Construct structured Prompt to use clear format, mark and template for intelligent prompt;

[0029] Adopt RAG framework in large language model processing stage to carry out retrieval enhancement, generation optimization and output structured result.

[0030] Preferably, it further comprises:

[0031] The verified scheme is stored in the OTA knowledge base, and the validity score of the scheme is scored by the technician.

[0032] Preferably, it further comprises:

[0033] The mapping relationship between fault handling and effect tracking is established, when the OTA upgrade failure rate suddenly increases, the upgrade package is automatically associated and checked.

[0034] The present application provides an OTA knowledge base interaction method for a vehicle, adopts a large language model, retrieves relevant information from an OTA internal database, and inputs the information as a prompt Prompt to the large language model LLM, the model combines the prompts with its own language skills and outputs AI-optimized content to enhance the response efficiency and capability of the traditional OTA static database. The present application solves the problems of information lag and rigid response of the existing automobile OTA knowledge base, can enhance the accuracy and relevance of the knowledge base, improve the response efficiency of the knowledge base, and optimize the flexibility and adaptability of the knowledge base. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced as follows.

[0036] Figure 1 is a schematic diagram of an OTA knowledge base interaction method for a vehicle provided by the present application.

[0037] Figure 2 is a schematic diagram of a RGA framework provided by an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the persons skilled in the art better understand the scheme of the embodiments of the present application, the embodiments of the present application will be further described in detail in combination with the drawings and embodiments.

[0039] In order to solve the problems of information lag and rigid response of the existing automobile OTA knowledge base, the present application provides an OTA knowledge base interaction method for a vehicle, which can enhance the accuracy and relevance of the knowledge base, improve the response efficiency of the knowledge base, and optimize the flexibility and adaptability of the knowledge base.

[0040] As Figure 1 shown, an OTA knowledge base interaction method for a vehicle includes:

[0041] S1: Set up a vector database on an OTA platform server, store input text, pictures and videos as vector data to form an OTA knowledge base.

[0042] S2: When a user asks a question, convert the question into a vector, retrieve a paragraph that may contain an answer through the vector database, and form a retrieval context.

[0043] S3: Set up a large language model LLM, combine the question and context in a prompt, and submit it to the language large model to generate a final reply.

[0044] Specifically, a vector database is set up, a large amount of text is disassembled into paragraphs (chunks), and converted into vector data stored in the vector database. When a user asks a question, the question is converted into a vector, and a paragraph that may contain an answer is retrieved through the vector database to form a retrieval context. The question and context are combined in a prompt (Prompt) and submitted to a language large model (LLMs) to generate a final reply. This method combines traditional information retrieval technology with a large language generation model OTA dynamic knowledge base. This technology retrieves relevant information from the OTA internal database and inputs it as a prompt (Prompt) to the large language model (LLMs), and the model combines these prompts with its own language skills and outputs AI-optimized content to enhance the response efficiency and ability of the traditional OTA static database.

[0045] The method further includes: by giving a vehicle VIN code and query requirements, automatically obtaining vehicle status, component information, hardware records, and software records from the OTA platform to monitor real-time vehicle data.

[0046] The method further includes: according to the user input vehicle VIN code and requirements, verifying the user's identity and viewing permissions, obtaining the latest information from the current vehicle state, and retrieving the corresponding repair and upgrade history records from the vector database to form a visual report.

[0047] The method further includes: pre-processing the input vehicle information through a large language model LLM, actively integrating text, pictures and videos to perform multi-modal data integration, and then generating a comprehensive answer.

[0048] Further, the multi-modal data integration includes:

[0049] Standardize the collection of multi-source data;

[0050] Extract key features from multiple modalities;

[0051] Unified timestamps and location references;

[0052] Association analysis is performed using knowledge graph reasoning and statistical modeling.

[0053] Generate visual reports.

[0054] Preferred options also include:

[0055] Valid solutions confirmed by humans are automatically converted into knowledge graph nodes. The strategy priority is dynamically adjusted according to the success rate of the solutions, and knowledge nodes that have not been triggered for 6 months are marked as pending verification in order to eliminate outdated solutions.

[0056] The method also includes generating a multi-level root cause analysis report during fault diagnosis by real-time fusion of vehicle logs, network status, software version characteristics, and software upgrade records, combined with knowledge graph reasoning and large language model LLM.

[0057] Furthermore, the setting of the large language model includes:

[0058] Intelligent semantic recognition is set up during the question input stage to identify key elements of natural language questions.

[0059] Build a multi-source search engine and use it to retrieve information from user queries from multiple dimensions.

[0060] Build structured Prompts for intelligent suggestions using explicit formatting, markup, and templates.

[0061] The RAG framework is used in the large language model processing stage to enhance retrieval, optimize generation, and output structured results.

[0062] In one embodiment, the question input phase includes:

[0063] Natural language questions: Technical language input, such as "How to handle the sudden failure of upgrading to version V2.3.5?".

[0064] Intelligent semantic recognition: The system identifies key elements (version number, fault type, time characteristics, etc.).

[0065] Multi-dimensional information retrieval, such as:

[0066] A [User Question] --> B [Multi-Source Search Engine];

[0067] B --> C [Version Release Documentation];

[0068] B --> D [Historical Fault Database];

[0069] B --> E [Real-time Device Log];

[0070] B --> F [Knowledge Graph];

[0071] B --> G [Third-party API].

[0072] Intelligent suggestions: Building structured Prompts, including:

[0073] Background: The current version upgrade failure rate has suddenly increased by 15%, affecting the X8 PHEV model.

[0074] Related log: Error code ECU_228, CAN bus timeout.

[0075] Historical solution: The same type of fault was resolved by restarting the gateway in June 2024.

[0076] Knowledge item: "OTA Upgrade Anomaly Handling Specifications".

[0077] Please generate: a step-by-step solution + a risk assessment + an emergency plan.

[0078] The large-scale model processing stage employs the RAG framework: retrieval enhancement, generation optimization, and output of structured results. The RAG framework is as follows: Figure 2 As shown.

[0079] The method also includes storing verified solutions in the OTA knowledge base and having technical personnel score the effectiveness of the solutions.

[0080] The method also includes: establishing a mapping relationship between fault handling and effect tracking, and automatically associating and verifying the upgrade package when a sudden increase in the OTA upgrade failure rate is detected.

[0081] Therefore, this invention provides an interactive method for an OTA (Over-The-Air) knowledge base in automobiles. It employs a large language model (LLM) and retrieves relevant information from the internal OTA database, inputting this information as prompts into the LLM. The model combines these prompts with its own language skills and outputs AI-optimized content, thereby enhancing the response efficiency and capability of traditional static OTA databases. This addresses the problems of information lag and rigid response in existing automotive OTA knowledge bases, improving the accuracy and relevance of the knowledge base, increasing its response efficiency, and optimizing its flexibility and adaptability.

[0082] The structure, features, and effects of the present invention have been described in detail above with reference to the embodiments shown in the figures. The above description is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the figures. Any changes made in accordance with the concept of the present invention, or equivalent embodiments modified to have equivalent changes, shall be within the protection scope of the present invention as long as they do not exceed the spirit covered by the specification and figures.

Claims

1. A knowledge base interaction method for automobiles under OTA (Over-The-Air) updates, characterized in that, include: A vector database is set up on the OTA platform server to store input text, images, and videos as vector data to form an OTA knowledge base; When a user asks a question, the question is converted into a vector, and the vector database is used to retrieve paragraphs that may contain the answer, thus forming a search context; Set up a large language model (LLM), combine the question and context in the prompt, and submit it to the large language model to generate the final response.

2. The OTA knowledge base interaction method for automobiles according to claim 1, characterized in that, Also includes: By providing the vehicle's VIN code and query requirements, the system automatically retrieves vehicle status, component information, hardware records, and software records from the OTA platform for real-time vehicle data monitoring.

3. The OTA knowledge base interaction method for automobiles according to claim 2, characterized in that, Also includes: Based on the user's input of the vehicle's VIN code and requirements, the system verifies the user's identity and viewing permissions, retrieves the latest information from the vehicle's current status, and retrieves the corresponding maintenance and upgrade history records from the vector database to generate a visual report.

4. The OTA knowledge base interaction method for automobiles according to claim 3, characterized in that, Also includes: By using a large language model (LLM) to preprocess the input vehicle information, text, images, and videos are actively integrated to perform multimodal data integration and generate a comprehensive answer.

5. The OTA knowledge base interaction method for automobiles according to claim 4, characterized in that, The multimodal data integration includes: Standardized collection of multi-source data; Extract key features from multiple modalities; Unified timestamps and location references; Association analysis is performed using knowledge graph reasoning and statistical modeling. Generate visual reports.

6. The OTA knowledge base interaction method for automobiles according to claim 5, characterized in that, Also includes: Valid solutions confirmed by humans are automatically converted into knowledge graph nodes. The strategy priority is dynamically adjusted according to the success rate of the solutions, and knowledge nodes that have not been triggered for 6 months are marked as pending verification in order to eliminate outdated solutions.

7. The OTA knowledge base interaction method for automobiles according to claim 6, characterized in that, Also includes: During fault diagnosis, multi-level root cause analysis reports are generated by real-time fusion of vehicle logs, network status, software version characteristics, and software upgrade records, combined with knowledge graph reasoning and large language model LLM.

8. The OTA knowledge base interaction method for automobiles according to claim 7, characterized in that, The setting of the large language model includes: Intelligent semantic recognition is set up during the question input stage to identify key elements of natural language questions; Build a multi-source search engine and use it to retrieve multi-dimensional information from user queries. Build structured Prompts for intelligent suggestions using explicit formatting, markup, and templates; The RAG framework is used in the large language model processing stage to enhance retrieval, optimize generation, and output structured results.

9. The OTA knowledge base interaction method for automobiles according to claim 8, characterized in that, Also includes: Validated solutions are stored in the OTA knowledge base, and technical personnel evaluate the effectiveness of the solutions.

10. The OTA knowledge base interaction method for automobiles according to claim 8, characterized in that, Also includes: Establish a mapping relationship between fault handling and effect tracking. When a sudden increase in the OTA upgrade failure rate is detected, automatically associate the upgrade package and perform verification.

Images