Internet-of-vehicles multi-mode intelligent interaction system based on DeepSeek large model and SpringCloud architecture and implementation method of Internet-of-vehicles multi-mode intelligent interaction system based on DeepSeek large model and SpringCloud architecture

Abstract

The invention relates to an Internet of Vehicles intelligent query system based on a DeepSeek large model and an implementation method, and solves the problems of low efficiency and high threshold of a traditional query mode. The system architecture comprises an interaction layer, a computing layer and a data layer. The implementation method comprises the following steps: S1, data fusion and knowledge graph construction; S2, natural language intention conversion; S3, Spring-based dynamic service calling; and S4, multi-modal result feedback and display. The method has the following beneficial effects: 1, the query threshold is reduced: a complex analysis result can be obtained through dialogue without professional training; 2, the efficiency is improved: a manual report which needs 1 hour traditionally can be generated within 1 minute; 3, enhancing decision support: supporting multi-condition correlation analysis; and 4, the expansibility is high, and a newly-added data source can be adapted.

Description

Technical Field

[0001] This invention relates to a vehicle-to-everything (V2X) multimodal intelligent interaction system and its implementation method based on the DeepSeek large model and Spring Cloud architecture. Background Technology

[0002] With the rapid development of vehicle-to-everything (V2X) technology, vehicle sensors, operating platforms, and organizational management systems generate massive amounts of data (such as real-time vehicle location, mileage data, online status, and operational analysis). Currently, V2X data retrieval mainly relies on the following methods:

[0003] 1. Fixed format report generation: Static reports generated by preset templates cannot meet dynamic and personalized query needs (such as "comparing the fuel consumption per 100 kilometers of Team A and Team B over the past 3 months").

[0004] 2. Traditional database queries: rely on specialized query languages ​​such as SQL, require technical personnel, and are difficult for ordinary users to obtain information directly.

[0005] 3. Visual Dashboard Limitations: Existing dashboards only display preset indicators and cannot flexibly respond to complex query scenarios (such as "list of organizations with a vehicle online rate of less than 60% in a certain area").

[0006] The above method has the following drawbacks:

[0007] 1. Complex operation: requires switching between multiple pages and professional search skills, resulting in a poor user experience.

[0008] 2. Insufficient real-time performance: Static reports cannot dynamically respond to real-time data (such as the instantaneous online status of vehicles).

[0009] 3. Lack of natural language interaction: Results cannot be obtained directly through dialogue, resulting in high learning costs.

[0010] 4. Insufficient support for complex queries: It is difficult to handle multi-condition association analysis (such as "find vehicles with a mileage of more than 2,000 kilometers in the past week and an online rate of less than 50%"). Summary of the Invention

[0011] This invention proposes a vehicle-to-everything (V2X) intelligent query system and its implementation method based on the DeepSeek large model. It achieves multi-dimensional data fusion query through natural language dialogue, solving the problems of low efficiency and high threshold of traditional query methods.

[0012] This invention is achieved through the following technical solutions:

[0013] A vehicle-to-everything (V2X) multimodal intelligent interaction system based on DeepSeek large model and Spring Cloud architecture, the system architecture includes:

[0014] Interaction layer: Integrates DeepSeek multimodal interface, supporting voice noise reduction and intent disambiguation;

[0015] Computation layer: Deploy the Spring Cloud microservice framework to dynamically schedule real-time data streams and historical data warehouses;

[0016] Data layer: Constructs a knowledge graph for the Internet of Vehicles, including relation types.

[0017] The real-time data stream comes from Redis, and the historical data warehouse is HBase or pgsql.

[0018] The relationship types include multi-dimensional associations such as "vehicle-owning company-operating region".

[0019] A method for implementing a multimodal intelligent interaction system for connected vehicles based on the DeepSeek large model and Spring Cloud architecture includes the following steps:

[0020] S1: Data Fusion and Knowledge Graph Construction

[0021] 1) Access real-time vehicle sensor data, organizational hierarchy information, and historical operation records;

[0022] 2) Construct a knowledge graph of vehicle network operations: Store the vehicle network query operations that DeepSeek can execute in the database to assist in judgment during intent analysis;

[0023] S2: Natural Language Intent Transformation

[0024] The user enters a query, which is then processed by the DeepSeek engine.

[0025] Entity recognition: keyword extraction;

[0026] Intent Classification: Determine the operation chain based on all executable operations in the vehicle-to-everything (V2X) operation knowledge graph;

[0027] Search for all vehicles under company XX;

[0028] Query the total mileage of vehicles registered under company XX for the previous month;

[0029] Logical breakdown: Generate query conditions based on JSON examples of parameters from all executable operations in the vehicle network operation knowledge graph;

[0030] S3: Dynamic service invocation based on Spring

[0031] Based on the operation chain transformed by S2 and the parameters of each operation, Java reflection is used for execution, combining multiple data source query strategies:

[0032] Real-time data: Retrieves the current online status of vehicles and users from Redis;

[0033] Historical data: Extracting various aggregated vehicle data from the data warehouse using PostgreSQL;

[0034] Spatial data: Combined with a GIS engine to process geographical location conditions;

[0035] The query results will be fed back to DeepSeek for multimodal result display;

[0036] S4: Multimodal Result Feedback Display

[0037] Generate responses using operation descriptions from the vehicle-to-everything (V2X) operation knowledge graph:

[0038] Visual charts: Automatically select the optimal display format;

[0039] Operational suggestion: Utilize TTS technology to broadcast key performance indicators;

[0040] Source tracing report: Provides explanation of data sources and confidence level assessment.

[0041] In step S1, the real-time sensor data includes online rate, mileage, and energy consumption.

[0042] In step S2, the JSON example of the parameter specifically includes:

[0043] The parameter for querying all vehicles under the organization is: {"orgId":"1"}

[0044] The parameters for querying the mileage of the selected vehicle last month are:

[0045] {"vins":"LZ91DD210XXXXX321,LZ91DD210XXXXXXX","startTime":"2025-01-0200:00:00","endTime":"2025-02-0200:00:00"}.

[0046] In step S4, the optimal display format includes line charts and heatmaps. Line charts display trends, and heatmaps display regional distribution.

[0047] In step S4, the TTS technology broadcasts key indicators including "XX company's average online rate last month".

[0048] In step S4, the confidence assessment includes "data coverage of n% of active vehicles".

[0049] The beneficial effects of this invention are:

[0050] 1. Lower the barrier to entry: complex analysis results can be obtained through dialogue without professional training.

[0051] 2. Improve efficiency: Manual reports that traditionally take 1 hour to generate can be generated in 1 minute.

[0052] 3. Enhance decision support: Support multi-condition correlation analysis (e.g., "What was the total mileage of all vehicles of Company XX last month?").

[0053] 4. High scalability: It can be adapted to new data sources (such as new energy vehicle battery health query). Attached Figure Description

[0054] Figure 1 Flowchart of the method of this invention;

[0055] Figure 2 Vehicle-to-Everything (V2X) Operation Knowledge Graph Database;

[0056] Figure 3 Vehicle-to-Everything (V2X) Operation Knowledge Graph Database - Parameter Examples;

[0057] Figure 4 Java reflection mechanism code;

[0058] Figure 5 Vehicle-to-Everything (V2X) Operation Knowledge Graph Database - Operation Description;

[0059] Figure 6 DeepSeek multimodal results feedback display. Detailed Implementation

[0060] This invention relates to a multimodal intelligent interaction system for vehicle networking based on the DeepSeek large model and Spring Cloud architecture, and its implementation method. Specific objectives include:

[0061] 1. Provide a natural language interaction interface, allowing users to ask questions directly via voice or text (e.g., "Query the online time of vehicles in XX fleet yesterday").

[0062] 2. Supports multimodal data association queries, integrating multi-source data such as vehicle sensor data, organizational information, and operational records.

[0063] 3. Enables complex logic parsing, automatically converts user intent into structured query commands, and generates visual reports.

[0064] System Architecture:

[0065] Interaction layer: Integrates DeepSeek multimodal interface, supports voice noise reduction and intent disambiguation.

[0066] Computation layer: Deploy the Spring Cloud microservice framework to dynamically schedule real-time data streams (Redis) and historical data warehouses (HBase, pgsql).

[0067] Data layer: Construct a knowledge graph for the Internet of Vehicles, including relationship types (such as multi-dimensional associations such as "vehicle-owner company-operating area").

[0068] The specific method is as follows:

[0069] Step 1: Data Fusion and Knowledge Graph Construction (e.g.) Figure 2 (As shown)

[0070] 1. Access real-time vehicle sensor data (online rate, mileage, energy consumption, etc.), organizational hierarchy information, and historical operation records.

[0071] 2. Construct a knowledge graph of vehicle network operations: Store the vehicle network query operations that DeepSeek can execute in the database to assist in judgment during intent analysis.

[0072] Step 2: Natural Language Intent Conversion

[0073] The user inputs a query (e.g., "What was the total mileage of all vehicles belonging to Company XX last month?"), which is then processed by the DeepSeek engine.

[0074] Entity recognition: Extract keywords ("XX company", "mileage", "last month").

[0075] Intent Classification: Based on all executable operations in the vehicle-to-everything (V2X) operation knowledge graph, determine the operation chain:

[0076] Search all vehicles under company XX

[0077] Query all mileage of vehicles under company XX for the previous month

[0078] Logical breakdown: Generate query conditions based on JSON examples of parameters from all executable operations in the vehicle network operation knowledge graph (e.g., Figure 3 (As shown).

[0079] but:

[0080] The parameter for querying all vehicles under the organization is: {"orgId":"1"}

[0081] The parameters for querying the mileage of the selected vehicle last month are:

[0082] {"vins":"LZ91DD210XXXXX321,LZ91DD210XXXXXXX","startTime":"2025-01-0200:00:00","endTime":"2025-02-0200:00:00"}.

[0083] Step 3: Dynamic service invocation based on Spring

[0084] Based on the operation chain transformed in the second step and the parameters of each operation, Java reflection is used for execution, combining multiple data source query strategies (such as...). Figure 4 As shown):

[0085] Real-time data: Retrieves the current online status of vehicles and users (Redis).

[0086] Historical data: Extract various types of aggregated vehicle data from the data warehouse (pgsql).

[0087] Spatial data: Combined with GIS engine processing of geographical location conditions (such as "vehicles in a certain area").

[0088] The query results will be fed back to DeepSeek for multimodal result display.

[0089] Step 4: Multimodal result feedback display (e.g.) Figure 5 , 6 (As shown)

[0090] Generate responses using operation descriptions from the vehicle-to-everything (V2X) operation knowledge graph:

[0091] Visual charts: Automatically select the optimal display format (line charts show trends, heatmaps show regional distribution).

[0092] Operational suggestion: Announce key metrics ("XX Company's average online rate last month was 82.3%) using TTS technology.

[0093] Source tracing report: Provides a description of the data source and a confidence assessment (e.g., "Data covers 95% of active vehicles").

[0094] This invention uses Java reflection to execute instructions parsed from natural language, and achieves multi-dimensional data fusion query through natural language dialogue, solving the problems of low efficiency and high threshold of traditional query methods.

Claims

1. A multimodal intelligent interaction system for vehicle networking based on DeepSeek large model and Spring Cloud architecture, characterized by: The system architecture includes: Interaction layer: Integrates DeepSeek multimodal interface, supporting voice noise reduction and intent disambiguation; Computation layer: Deploy the Spring Cloud microservice framework to dynamically schedule real-time data streams and historical data warehouses; Data layer: Constructs a knowledge graph for the Internet of Vehicles, including relation types.

2. The vehicle-to-everything (V2X) multimodal intelligent interaction system based on DeepSeek large model and Spring Cloud architecture as described in claim 1, characterized in that: The real-time data stream comes from Redis, and the historical data warehouse is HBase or pgsql.

3. The vehicle-to-everything (V2X) multimodal intelligent interaction system based on DeepSeek large model and Spring Cloud architecture as described in claim 1, characterized in that: The relationship types include multi-dimensional associations such as "vehicle-owning company-operating region".

4. A method for implementing a multimodal intelligent interaction system for vehicle networking based on the DeepSeek large model and Spring Cloud architecture, characterized by: Includes the following steps: S1: Data Fusion and Knowledge Graph Construction 1) Access real-time vehicle sensor data, organizational hierarchy information, and historical operation records; 2) Construct a knowledge graph of vehicle network operations: Store the vehicle network query operations that DeepSeek can execute in the database to assist in judgment during intent analysis; S2: Natural Language Intent Transformation The user enters a query, which is then processed by the DeepSeek engine. Entity recognition: keyword extraction; Intent Classification: Determine the operation chain based on all executable operations in the vehicle-to-everything (V2X) operation knowledge graph; Search for all vehicles under company XX; Query the total mileage of vehicles registered under company XX for the previous month; Logical breakdown: Generate query conditions based on JSON examples of parameters from all executable operations in the vehicle network operation knowledge graph; S3: Dynamic service invocation based on Spring Based on the operation chain transformed by S2 and the parameters of each operation, Java reflection is used for execution, combining multiple data source query strategies: Real-time data: Retrieves the current online status of vehicles and users from Redis; Historical data: Extracting various aggregated vehicle data from the data warehouse using PostgreSQL; Spatial data: Combined with a GIS engine to process geographical location conditions; The query results will be fed back to DeepSeek for multimodal result display; S4: Multimodal Result Feedback Display Generate responses using operation descriptions from the vehicle-to-everything (V2X) operation knowledge graph: Visual charts: Automatically select the optimal display format; Operational suggestion: Utilize TTS technology to broadcast key performance indicators; Source tracing report: Provides explanation of data sources and confidence level assessment.

5. The method for implementing a multimodal intelligent interaction system for vehicle networking based on a DeepSeek large model and Spring Cloud architecture as described in claim 4, characterized in that: In step S1, the real-time sensor data includes online rate, mileage, and energy consumption.

6. The method for implementing a multimodal intelligent interaction system for vehicle networking based on a DeepSeek large model and Spring Cloud architecture as described in claim 4, characterized in that: In step S2, the JSON example of the parameter specifically includes: The parameter for querying all vehicles under the organization is: {"orgId":"1"} The parameters for querying the mileage of the selected vehicle last month are: {"vins":"LZ91DD210XXXXX321,LZ91DD210XXXXXXX","startTime":"2025-01-0200:00:00","endTime":"2025-02-0200:00:00"}.

7. The method for implementing a multimodal intelligent interaction system for vehicle networking based on a DeepSeek large model and Spring Cloud architecture as described in claim 4, characterized in that: In step S4, the optimal display format includes line charts and heatmaps. Line charts display trends, and heatmaps display regional distribution.

8. The method for implementing a multimodal intelligent interaction system for vehicle networking based on a DeepSeek large model and Spring Cloud architecture as described in claim 4, characterized in that: In step S4, the TTS technology broadcasts key indicators including "XX company's average online rate last month".

9. The method for implementing a multimodal intelligent interaction system for vehicle networking based on a DeepSeek large model and Spring Cloud architecture as described in claim 4, characterized in that: In step S4, the confidence assessment includes "data coverage of n% of active vehicles".

Images