Automated generation system of autonomous driving scenario library and method thereof

Abstract

The application discloses an automatic generation system and method of an automatic driving scene library, and relates to the technical field of automatic driving.The system comprises a data acquisition and preprocessing module, a knowledge graph construction module, a knowledge reasoning module, a scene generation module and a scene library management module.Through the construction of a knowledge graph, the application uses ontology modeling technology and a graph database to extract entities, attributes and relationships from the data of preprocessed traffic rules, accident cases and vehicle parameters, construct a unified and standardized knowledge framework and a domain ontology model, and calculate entity correlation strength to quantify the closeness between entities.The scattered data is associated into a structured knowledge network, different data is no longer isolated, data intercommunication is achieved, the value of data is deeply mined, the automatic driving scene library can more comprehensively cover complex relationships and potential risks in traffic scenes, and the quality and practicality of the scene library are improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically to an automated generation system and method for an autonomous driving scenario library. Background Technology

[0002] With the rapid development of autonomous driving technology, its commercial application has gradually moved from theoretical discussion to practical verification. As an important part of the future intelligent transportation system, the safety and reliability of autonomous vehicles have become the focus of public and industry attention. In order to ensure that autonomous driving systems can operate stably in various complex and ever-changing traffic environments, it is particularly important to conduct comprehensive and in-depth testing and verification. As a key tool for testing and verifying the performance of autonomous driving systems, the quality and efficiency of the construction of autonomous driving scenario libraries directly affect the maturity and market acceptance of autonomous driving technology.

[0003] However, traditional technologies for building autonomous driving scenario libraries have several shortcomings. In terms of data processing, their integration and correlation capabilities are insufficient. Faced with multi-source data, such as traffic rules, accident cases, and vehicle parameters, it is difficult to effectively integrate them and build a systematic knowledge system. This results in isolated data that cannot be interconnected, making it difficult to fully reveal the complex relationships and potential risks in traffic scenarios. In terms of relationship mining, there is a lack of hidden relationship discovery mechanisms. Many key safety risks in autonomous driving scenarios are hidden among seemingly unrelated scene elements. Traditional methods cannot capture these potential relationships, making autonomous driving systems prone to misjudgment or loss of control in complex traffic environments. In terms of scenario construction, diversity and adaptability are limited. Scenario libraries built by traditional methods are difficult to cover edge scenarios and lack dynamic update mechanisms. They cannot reflect changes in traffic rules, vehicle technology, and traffic environment in a timely manner, resulting in a significant reduction in timeliness and adaptability. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an automated generation system and method for an autonomous driving scenario library. This invention constructs a knowledge graph, utilizes ontology modeling techniques and graph databases to extract entities, attributes, and relationships from preprocessed traffic rules, accident cases, and vehicle parameter data. It then builds a unified and standardized knowledge framework and domain ontology model, calculates entity association strength to quantify the closeness between entities, and links scattered data into a structured knowledge network. This prevents different data from being isolated and enables interconnection between data, achieving in-depth mining of data value. This allows the autonomous driving scenario library to more comprehensively cover complex relationships and potential risks in traffic scenarios, improving the quality and practicality of the scenario library and providing more accurate scenario support for autonomous driving systems.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On one hand, an automated generation system for an autonomous driving scenario library, the system comprising:

[0006] Data acquisition and preprocessing module: Collects data on traffic rules, accident cases, and vehicle parameters, preprocesses the data, and transmits the preprocessed data;

[0007] Knowledge graph construction module: Receives preprocessed data, extracts entities and defines attributes, determines the relationships between entities, uses ontology modeling technology to define the types, attributes and constraints of entities and relationships, calculates the entity association strength, and constructs a domain ontology. It then uses a graph database to construct a knowledge graph by combining entities, relationships and association strengths according to ontology model specifications.

[0008] Knowledge Reasoning Module: Acquire knowledge graphs, analyze the information in the knowledge graphs, calculate the probability of hidden relationships using the hidden relationship reasoning probability formula, and obtain the reasoning results;

[0009] Scene generation module: Based on the reasoning results, set scene parameters, lay out scene elements, design scene plots, assess scene risks, convert the generated scene into simulation software format, and transmit it.

[0010] Scene library management module: Receives generated scenes in simulation software format, classifies them according to application scene type, risk level, and weather conditions, stores them, creates an index, and regularly evaluates scene quality. Based on the evaluation results, it regenerates or optimizes unqualified scenes and updates and maintains qualified scenes.

[0011] Furthermore, in the knowledge graph construction module, entities such as vehicles, roads, traffic signs, traffic lights, weather, and pedestrians are extracted and labeled from the preprocessed data. Attributes of each entity are extracted, and based on entity identification, the interactions and connections between entities are analyzed. Logical relationships between entities are determined through rule matching. A unified and standardized knowledge framework is constructed using ontology modeling technology, defining the types, attributes, and constraints of entities and relationships. Entity association strength is calculated using the entity association strength calculation formula, and a domain ontology model is constructed. The identified entities, determined relationships, and entity association strengths are organized and stored using a graph database. Entities are represented by nodes, with node attributes corresponding to entity attributes. Relationships between entities are represented by edges, with edge attributes containing information about relationship type and association strength, thus constructing a knowledge graph.

[0012] Furthermore, in the knowledge graph construction module, the entity association strength is calculated using the entity association strength calculation formula, which is as follows: Among them, R ijLet N be the association strength between entity i and entity j. ij N represents the number of times entity i and entity j co-occur. i To represent the total number of times entity i appears, N j To represent the total number of times entity j appears, S ij To represent the semantic similarity between entity i and entity j, α and β are weight coefficients, and α+β=1, which are set according to the actual application scenario.

[0013] Furthermore, in the knowledge reasoning module, a temporary index is established from the received knowledge graph. Information such as entity type, attribute parameters, and relationship type and strength between entities is extracted and recorded. For each entity, an adjacency list containing its directly related entities, relationship type, and strength is constructed. Entities are categorized and labeled, and the types of hidden relationships to be mined are identified. Relevant scene element entities are filtered, and key attributes are extracted. Based on the reasoning target and scene element entities, intermediate entities directly related to them are retrieved from the knowledge graph. Commonly occurring related entities are found by querying the entity's adjacency list. The retrieved intermediate entities are filtered for relevance, and the number of remaining intermediate entities after filtering is counted. The probability of hidden relationships between scene element entities is calculated using the hidden relationship reasoning probability formula, and the hidden relationship probability is compared with a preset confidence threshold P. max The comparison is performed. If the value exceeds the threshold, the hidden relationship is established. For established hidden relationships, logical verification is performed by combining entity attributes in the knowledge graph and traffic rules to obtain the reasoning result. If there is a conflict, the intermediate entity screening process is re-examined and the probability of the hidden relationship is recalculated until the reasoning result is logically consistent.

[0014] Furthermore, in the knowledge reasoning module, the probability of hidden relationships between scene element entities is calculated using a hidden relationship reasoning probability formula, which is as follows: Among them, P h Let be the probability that scene elements i and j have a hidden relationship h, n be the number of intermediate entities in the knowledge graph that are related to both entities i and j, k be the number of intermediate entities, and R be the probability that scene elements i and j have a hidden relationship h. ik R represents the association strength between entity i and intermediate entity k. kj M represents the association strength between intermediate entity k and entity j. h The number of times the hidden relation h appears in historical cases is calculated from historical data such as preprocessed accident cases. M is the total number of historical cases, calculated from the preprocessed historical data. γ is the weighting coefficient, set according to historical data.

[0015] Furthermore, in the scene generation module, scene parameters are set according to the reasoning results, scene elements are laid out, scene plots are designed, and scene risks are assessed.

[0016] The setting of scene parameters involves determining various parameters of the scene based on the reasoning results, including weather parameters, target vehicle speed parameters, preceding vehicle speed parameters, distance parameters between the two vehicles, road parameters, and time parameters.

[0017] The layout scene elements are arranged in the virtual space according to scene parameters and actual road environment characteristics. The shape, length and width of the road are set, traffic signs and markings are placed, and the initial position and driving direction of traffic participants are determined.

[0018] The designed scenario is as follows: Based on the reasoning results, the dynamic development process of the scenario is designed. In the initial state, the vehicle involved is traveling at a set risk speed, and the vehicle in front is traveling at a set speed. As time goes by, the distance between the two vehicles gradually changes, the vehicle in front brakes, and the vehicle involved reacts accordingly.

[0019] During the scene generation process, the risk level of the generated scene is assessed using the scene risk coefficient calculation formula.

[0020] Furthermore, in the scene generation module, the risk level of the generated scene is assessed using a scene risk coefficient calculation formula, which is as follows: Where F is the scenario risk coefficient, P g The probability of hidden risk relationships existing in the scene is calculated using the hidden relationship inference probability formula, where V is the actual speed of the vehicle in the scene. max This represents the maximum speed limit for the road in this scenario, with δ being a weighting coefficient set based on the risk assessment priorities of the scenario.

[0021] Furthermore, the scene library management module periodically evaluates scene quality using a scene quality assessment formula, which is: Where Q is the scene quality assessment value, F is the scene risk coefficient, and C is the number of scene element types contained in the scene. total The total number of all scene element types included in this type of scene, where θ is a weighting coefficient set according to the focus of the scene quality assessment;

[0022] And set Q based on historical evaluation data statistics x =0.6, when Q x When Q > Q, it is judged as an unqualified scenario. x When a scenario is deemed acceptable, it is either regenerated or optimized for unacceptable scenarios, and the acceptable scenarios are updated and maintained.

[0023] On the other hand, the method for automatically generating an autonomous driving scenario library includes the following steps: ​

[0024] Data Acquisition and Preprocessing: Collect data on traffic rules, accident cases, and vehicle parameters; preprocess the data; and then transmit it.

[0025] Knowledge graph construction: Receive preprocessed data, extract entities and define attributes, determine the relationships between entities, use ontology modeling technology to define the types, attributes and constraints of entities and relationships, calculate the entity association strength, and construct a domain ontology. Then, construct a knowledge graph using a graph database by combining entities, relationships and association strengths according to ontology model specifications.

[0026] Knowledge reasoning: Obtain a knowledge graph, analyze the information in the knowledge graph, calculate the probability of hidden relationships using the hidden relationship reasoning probability formula, and obtain the reasoning result;

[0027] Scene generation: Based on the reasoning results, set scene parameters, lay out scene elements, design scene plots, assess scene risks, convert the generated scene into simulation software format, and transmit it.

[0028] Scene library management: Receive generated scenes in simulation software format, classify and store them, create an index, and regularly evaluate scene quality. Based on the evaluation results, regenerate or optimize unqualified scenes, and update and maintain qualified scenes.

[0029] Compared with existing technologies, the automated generation system and method for this autonomous driving scenario library have the following advantages:

[0030] I. This invention constructs a knowledge graph and utilizes ontology modeling technology and graph databases to extract entities, attributes, and relationships from preprocessed traffic rules, accident cases, and vehicle parameter data. It builds a unified and standardized knowledge framework and domain ontology model, and calculates the entity association strength to quantify the closeness between entities. This connects scattered data into a structured knowledge network, preventing different data from being isolated and enabling interconnection between data. This achieves in-depth mining of data value, allowing the autonomous driving scenario library to more comprehensively cover the complex relationships and potential risks in traffic scenarios, improving the quality and practicality of the scenario library, and providing more accurate scenario support for autonomous driving systems.

[0031] Second, this invention automatically sets scene parameters, lays out scene elements, and designs scene plots based on knowledge reasoning results through a scene generation module. It also assesses the risk level of the generated scene through a scene risk coefficient calculation formula, enabling the generation of comprehensive and diverse autonomous driving scenarios. At the same time, the scene library management module regularly evaluates the quality of the generated scenarios. Based on the evaluation results, unqualified scenarios are regenerated or optimized, while qualified scenarios are updated and maintained. This dynamic update and optimization mechanism ensures that the scene library always keeps up with changes in traffic rules, vehicle technology, and traffic environment, maintaining timeliness and accuracy, and enhancing the adaptability of the autonomous driving system to complex and ever-changing traffic environments.

[0032] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0034] Figure 1 A framework diagram of an automated generation system for an autonomous driving scenario library;

[0035] Figure 2 A flowchart of a method for automatically generating an autonomous driving scenario library. Detailed Implementation

[0036] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0037] Example 1:

[0038] Data Acquisition and Preprocessing Module: Collects rules from the "Urban Road Traffic Safety Law" regarding motor vehicles yielding to pedestrians at intersections and turning vehicles yielding to vehicles going straight, accident cases at urban intersections such as "right-turning vehicles failing to yield to pedestrians on zebra crossings" and "straight-going vehicles running red lights and colliding with left-turning vehicles," as well as parameters such as braking response time and maximum braking deceleration of family cars, and transmits the data after preprocessing.

[0039] The knowledge graph construction module receives preprocessed data and extracts entities including cars (attributes: speed range, braking distance), buses (attributes: length, turning radius), intersections (attributes: number of lanes, intersection width), traffic lights (attributes: red light duration, green light duration), and pedestrians (attributes: walking speed, location), etc., and determines the relationships between entities. For example, "traffic light is red" and "straight-going cars" have a mandatory relationship of "must stop," and "pedestrians at crosswalks" and "right-turning buses" have a "priority" relationship. It uses ontology modeling technology to define types such as "mandatory relationship" and "priority relationship," with the constraint condition "if a straight-going vehicle does not stop under red light conditions, it violates the rule." The module calculates the entity association strength using the formula: Among them, R ij Let N be the association strength between entity i and entity j. ij N represents the number of times entity i and entity j co-occur. i To represent the total number of times entity i appears, N j To represent the total number of times entity j appears, S ij To represent the semantic similarity between entity i and entity j, α and β are weight coefficients, and α+β=1. Based on the actual application scenario, the association strength between "red light" and "straight-going car" is calculated (based on the frequency of their co-occurrence in accident cases, and the semantic association between "red light" and "vehicle stopped"). The data is organized and arranged according to the ontology model, stored in a graph database, and entities are represented by nodes, with node attributes corresponding to entity attributes. Edges represent the relationships between entities, and the attributes of the edges contain information about the relationship type and association strength, thus constructing a knowledge graph.

[0040] Knowledge Reasoning Module: This module acquires a knowledge graph and establishes a temporary index. It constructs adjacency lists for "right-turning cars," "east-to-west zebra crossing pedestrians," and "traffic light - green light." These lists include the "permission" relationship and association strength between "right-turning cars" and "traffic light - green light," and the "location" relationship and association strength between "pedestrians" and "zebra crossing." The hidden relationship to be mined is identified as "collision risk relationship between right-turning cars and pedestrians on the zebra crossing during a green light." The module retrieves intermediate entities "zebra crossing location," "car right-turn speed," and "pedestrian walking speed." After filtering out entities strongly associated with both, the module counts their number and calculates the probability of this hidden relationship using a hidden relationship reasoning probability formula. The hidden relationship reasoning probability formula is as follows: Among them, P h Let be the probability that scene elements i and j have a hidden relationship h, n be the number of intermediate entities in the knowledge graph that are related to both entities i and j, k be the number of intermediate entities, and R be the probability that scene elements i and j have a hidden relationship h. ik R represents the association strength between entity i and intermediate entity k. kj M represents the association strength between intermediate entity k and entity j.h The hidden relation h is the number of times it appears in historical cases. This is calculated from preprocessed historical data such as accident cases. M is the total number of historical cases. γ is the weighting coefficient. Based on historical data, the probability of the hidden relation h is compared with a pre-set confidence threshold P. max By comparing and verifying the traffic rule that "vehicles turning right on a green light must still yield to pedestrians", we arrive at the inference that there is a high risk of collision.

[0041] Scene generation module: Based on the reasoning results, set the scene parameters: the weather is cloudy, the car turning right maintains a high initial speed, the pedestrian walks at a normal pace, the lateral distance between the car and the pedestrian is relatively close, and the time is during the evening rush hour;

[0042] Scene elements: The crossroads has a zebra crossing from east to west, and the intersection has a green light signal. Cars turning right are located in the northbound lane near the stop line, and pedestrians are located at the western starting point of the zebra crossing.

[0043] The scenario is designed so that the green light is on at the beginning, the car turns right and begins to turn right, and at the same time the pedestrian steps onto the zebra crossing to walk east. The car does not slow down. As time goes by, the lateral distance between the two gradually decreases to a dangerous range. The pedestrian notices the danger and stops. At this time, the car driver begins to brake.

[0044] During the scene generation process, the risk level of the generated scene is assessed using a scene risk coefficient formula, which is as follows: Where F is the scenario risk coefficient, P g The probability of hidden risk relationships existing in the scene is calculated using the hidden relationship inference probability formula, where V is the actual speed of the vehicle in the scene. max The maximum speed limit for the road in this scenario is defined by δ, which is a weighting coefficient. Based on the risk assessment priorities of the scenario, the risk coefficient F for this scenario is 0.75, classifying it as a high-risk scenario. The generated scenario is then converted into simulation software format for transmission. Figure 1 As shown.

[0045] The scenario library management module receives scenarios in simulation format, categorizes them by application scenario type (e.g., "urban roads - intersections"), marks the risk level as "high risk," and labels the weather conditions as "cloudy." After storage, it creates an index with keywords including "right-turning vehicles," "pedestrians," and "green light." It also periodically evaluates scenario quality using a scenario quality assessment formula, which is: Where Q is the scene quality assessment value, F is the scene risk coefficient, and C is the number of scene element types contained in the scene. totalθ represents the total number of all scene element types included in this type of scene, and θ is a weighting coefficient set according to the focus of the scene quality assessment, and Q is set based on historical assessment data. x =0.6, compare the scene quality assessment value Q with the preset quality threshold Qx. If the scene quality assessment value Q is higher than the quality threshold Qx, it is judged as a qualified scene and included in the scene library for updating and maintenance.

[0046] In summary, focusing on urban road intersection scenarios, the system collects and processes relevant traffic rules, accident cases, and vehicle parameters through a data acquisition and preprocessing module. A knowledge graph construction module extracts entities, clarifies relationships and correlation strengths to build a knowledge graph. A knowledge reasoning module uncovers hidden risk relationships and derives results. A scenario generation module then generates specific scenarios based on these results. Finally, a scenario library management module stores scenarios by category, establishes indexes, and maintains or optimizes scenarios according to quality assessment standards, thus forming a complete automated generation process for urban road intersection scenario libraries.

[0047] Example 2:

[0048] Data Acquisition and Preprocessing Module: Collects relevant rules on vehicle driving in severe weather from the "Highway Traffic Management Regulations", accident cases such as rear-end collisions and skidding that occur on highways in severe weather, as well as braking performance and anti-skid parameters of trucks and buses, and transmits the data after preprocessing.

[0049] The knowledge graph construction module receives preprocessed data, extracts entities including trucks, buses, highway sections, severe weather, and emergency lanes, clarifies entity attributes (such as the load capacity of trucks, the passenger capacity of buses, and the impact range of severe weather), and determines the relationships between entities. For example, "severe weather" and "trucks" have a speed limit relationship, and "buses" and "trucks" have a relationship of maintaining a safe distance when traveling in the same lane. It uses ontology modeling technology to define relationship types and constraints (such as "severe weather" and "vehicle speed" having a strong association constraint relationship), and calculates the entity association strength using the formula: Entity association strength is calculated (e.g., the association strength between "severe weather" and "bus" based on their common occurrence frequency and semantic similarity), and then organized and arranged according to the ontology model. This is stored in a graph database, with nodes representing entities, node attributes corresponding to entity attributes, and edges representing relationships between entities. Edge attributes contain information about relationship type and association strength, thus constructing a knowledge graph. Figure 2 As shown.

[0050] Knowledge Reasoning Module: This module acquires a knowledge graph and establishes a temporary index. It constructs adjacency lists for "truck," "bus," and "severe weather," identifies hidden relationships to be mined (e.g., the risk of a rear-end collision between a truck and a bus traveling in the same lane under severe weather conditions), retrieves intermediate entities associated with both (e.g., the degree of weather impact, vehicle distance), filters and counts these entities, and infers probability formulas based on hidden relationships. Calculate the probability of the hidden relationship, and then compare the probability of the hidden relationship with a pre-set confidence threshold P. max By comparing and verifying relevant highway rules (such as insufficient following distance in severe weather increasing the risk of rear-end collisions), the inference result is obtained (there is a high risk of rear-end collisions).

[0051] Scene generation module: Based on the reasoning results, set the scene parameters: the weather is severe, the truck maintains a high initial speed, the bus in front travels at a slower speed, the two vehicles are initially close, and the time is during a period of relatively sparse traffic.

[0052] The layout of the scene elements shows a highway with one-way multiple lanes, trucks in the middle lane, buses in the front of the same lane, no obstacles in the emergency lane, and the road surface is slippery due to the weather.

[0053] The scenario is designed so that both vehicles are initially traveling in bad weather. The bus begins to brake slowly after noticing abnormal road conditions ahead, while the truck driver only notices the braking signal when he is close to the vehicle in front due to the weather affecting visibility. At this point, he begins to brake suddenly, but the braking effect is affected by the slippery road surface.

[0054] During the scene generation process, the scene risk coefficient formula is used: The risk level of the generated scenario was assessed, and the risk coefficient F of the scenario was found to be 0.82, which is considered a high-risk scenario. The generated scenario was then converted into simulation software format and transmitted.

[0055] Scene library management module: Receives simulation-formatted scenes, categorizes them by application scenario type (e.g., "Highway - Severe Weather"), marks the risk level as "High Risk," and labels the weather condition as "Dense Fog." After storage, it creates an index with keywords including "Heavy Truck," "Large Bus," "Dense Fog," and "Rear-end Collision," and uses a scene quality assessment formula. Regularly assess scenario quality and set Q based on historical assessment data. x =0.6, compare the scene quality assessment value Q with the preset quality threshold Qx. If the scene quality assessment value Q is lower than the quality threshold Qx, return to the scene generation module to regenerate or optimize (such as supplementing missing scene elements such as road surface conditions).

[0056] In summary, for severe weather scenarios on highways, the data acquisition and preprocessing module first collects and processes relevant rules, accident cases, and vehicle parameters; the knowledge graph construction module extracts entities, determines relationships and correlation strengths to construct a knowledge graph; the knowledge reasoning module mines hidden rear-end collision risk relationships; the scenario generation module generates corresponding scenarios; and the scenario library management module classifies, stores, and indexes the scenarios, and maintains or optimizes them based on quality assessment results, thus achieving automated generation of a severe weather scenario library for highways.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An automated generation system for an autonomous driving scenario library, characterized in that, The system includes: Data acquisition and preprocessing module: Collects data on traffic rules, accident cases, and vehicle parameters, preprocesses the data, and transmits the preprocessed data; Knowledge graph construction module: Receives preprocessed data, extracts entities and defines attributes, determines the relationships between entities, uses ontology modeling technology to define the types, attributes and constraints of entities and relationships, calculates the entity association strength, and constructs a domain ontology. It then uses a graph database to construct a knowledge graph by combining entities, relationships and association strengths according to ontology model specifications. Knowledge Reasoning Module: Acquire knowledge graphs, analyze the information in the knowledge graphs, calculate the probability of hidden relationships using the hidden relationship reasoning probability formula, and obtain the reasoning results; The probability formula for hidden relationship inference is: ,in, Scene elements and There is a hidden relationship The probability, For entities in knowledge graphs and The number of relevant intermediate entities intermediate entity, For entities and intermediate entities The strength of the association, intermediate entity and entity The strength of the association, Hiding relationships in historical cases Number of times it appears Total number of historical cases These are weighting coefficients, set based on historical data; Scene generation module: Based on the reasoning results, it sets scene parameters, lays out scene elements, designs scene plots, and assesses the risk level of the generated scene using a scene risk coefficient calculation formula. The scene risk coefficient calculation formula is as follows: ,in, For scenario risk coefficient, The probability of hidden risk relationships existing in the scenario is calculated using the probability formula for inferring hidden relationships. The actual speed of the vehicles in the scene. This represents the maximum speed limit for the road in this scenario. These are the weighting coefficients; the generated scene is converted into simulation software format for transmission; Scene library management module: Receives generated scenes in simulation software format, classifies them according to application scene type, risk level, and weather conditions, stores them, creates an index, and regularly evaluates scene quality. Based on the evaluation results, it regenerates or optimizes unqualified scenes and updates and maintains qualified scenes.

2. The automated generation system for the autonomous driving scenario library according to claim 1, characterized in that, In the knowledge graph construction module, entities such as vehicles, roads, traffic signs, traffic lights, weather, and pedestrians are extracted and labeled from the preprocessed data. Attributes of each entity are extracted. Based on entity identification, the interactions and connections between entities are analyzed. Logical relationships between entities are determined through rule matching. A unified and standardized knowledge framework is constructed using ontology modeling technology. The types, attributes, and constraints of entities and relationships are defined. Entity association strength is calculated using the entity association strength calculation formula. A domain ontology model is constructed. The identified entities, determined relationships, and entity association strengths are organized and stored in a graph database. Entities are represented by nodes, with node attributes corresponding to entity attributes. Relationships between entities are represented by edges, with edge attributes containing information about relationship type and association strength, thus constructing a knowledge graph.

3. The automated generation system for the autonomous driving scenario library according to claim 2, characterized in that, In the knowledge graph construction module, the entity association strength is calculated using the entity association strength calculation formula, which is as follows: ,in, For entities and entity The strength of the association, For entities and entity The number of times they appear together To represent entities Total number of occurrences To represent entities Total number of occurrences To represent entities and entity semantic similarity, and These are the weighting coefficients, and It is set according to the actual application scenario.

4. The automated generation system for the autonomous driving scenario library according to claim 1, characterized in that, In the knowledge reasoning module, a temporary index is established from the knowledge graph. Information such as entity type, attribute parameters, and relationship type and strength between entities is extracted and recorded. For each entity, an adjacency list containing its directly related entities, relationship type, and strength is constructed. Entities are categorized and labeled, and the types of hidden relationships to be mined are identified. Relevant scene element entities are filtered, and key attributes are extracted. Based on the reasoning target and scene element entities, intermediate entities directly related to them are retrieved from the knowledge graph. Commonly occurring related entities are found by querying the entity's adjacency list. The retrieved intermediate entities are filtered for relevance, and the number of remaining intermediate entities is counted. The probability of hidden relationships between scene element entities is calculated using a hidden relationship reasoning probability formula, and the hidden relationship probability is compared with a preset confidence threshold. The comparison is performed. If the value exceeds the threshold, the hidden relationship is considered valid. For valid hidden relationships, logical verification is performed by combining entity attributes and traffic rules in the knowledge graph to obtain the reasoning result. If there is a conflict, the intermediate entity screening process is re-examined and the probability of the hidden relationship is recalculated until the reasoning result is logically consistent.

5. The automated generation system for the autonomous driving scenario library according to claim 1, characterized in that, In the scene generation module, scene parameters are set according to the reasoning results, scene elements are laid out, scene plots are designed, and scene risks are assessed. The setting of scene parameters involves determining various parameters of the scene based on the reasoning results, including weather parameters, target vehicle speed parameters, preceding vehicle speed parameters, distance parameters between the two vehicles, road parameters, and time parameters. The layout scene elements are arranged in the virtual space according to scene parameters and actual road environment characteristics. The shape, length and width of the road are set, traffic signs and markings are placed, and the initial position and driving direction of traffic participants are determined. The designed scenario is as follows: Based on the reasoning results, the dynamic development process of the scenario is designed. In the initial state, the vehicle involved is traveling at a set risk speed, and the vehicle in front is traveling at a set speed. As time goes by, the distance between the two vehicles gradually changes, the vehicle in front brakes, and the vehicle involved reacts accordingly. During the scene generation process, the risk level of the generated scene is assessed using the scene risk coefficient calculation formula.

6. The automated generation system for the autonomous driving scenario library according to claim 1, characterized in that, In the scene library management module, scene quality is periodically evaluated using a scene quality assessment formula, which is: ,in, This is a scene quality assessment value. For scenario risk coefficient, This represents the number of different types of scene elements contained in the scene. This represents the total number of all scene element types included in this type of scene. These are the weighting coefficients; And set according to historical evaluation data statistics =0.6, when When, it is judged as an unqualified scenario, when When a scenario is deemed acceptable, it is either regenerated or optimized for unacceptable scenarios, and the acceptable scenarios are updated and maintained.

7. An automated generation method for an autonomous driving scene library, applicable to the automated generation system for an autonomous driving scene library as described in any one of claims 1-6, characterized in that, The method includes: Data Acquisition and Preprocessing: Collect data on traffic rules, accident cases, and vehicle parameters; preprocess the data; and then transmit it. Knowledge graph construction: Receive preprocessed data, extract entities and define attributes, determine the relationships between entities, use ontology modeling technology to define the types, attributes and constraints of entities and relationships, calculate the entity association strength, and construct a domain ontology. Then, construct a knowledge graph using a graph database by combining entities, relationships and association strengths according to ontology model specifications. Knowledge reasoning: Obtain a knowledge graph, analyze the information in the knowledge graph, calculate the probability of hidden relationships using the hidden relationship reasoning probability formula, and obtain the reasoning result; Scene generation: Based on the reasoning results, set scene parameters, lay out scene elements, design scene plots, assess scene risks, convert the generated scene into simulation software format, and transmit it. Scene library management: Receive generated scenes in simulation software format, classify and store them, create an index, and regularly evaluate scene quality. Based on the evaluation results, regenerate or optimize unqualified scenes, and update and maintain qualified scenes.

Images