Signal control method for vehicle-road cooperation of different intelligent level vehicles based on abm and deep learning model

Abstract

The application provides a signal control method for vehicle-road cooperation of vehicles with different intelligent levels based on ABM and a deep learning model, which comprises the following steps: dividing a multi-source heterogeneous traffic flow data in a time-space unit, obtaining road section unit data and topological relations; processing the data by using an ABM model, generating complete trajectory agents of L3-L4 vehicles and boundary associated agents of L1-L2 vehicles, and determining an interaction type identifier according to a cooperation boundary; for a target agent, determining adjacent road section units and candidate agents based on the interaction identifier and the topological relations, matching associated agents that meet a traffic flow continuity threshold through cooperative perception coordinates, and assigning a cooperation identifier; counting the number of complete trajectory agents and the number of cooperation identifiers, constructing a density distribution matrix based on a weighted sum, predicting traffic conflict points by using a deep learning, and optimizing a signal control strategy. The application can optimize a signal strategy, improve traffic efficiency, and reduce conflict points.

Description

Technical Field

[0001] This invention relates to the field of intelligent traffic control technology, and more specifically, to a signal control method for vehicle-road cooperation of different intelligence levels based on ABM and deep learning models. Background Technology

[0002] With the continuous development of intelligent transportation systems, the level of vehicle intelligence is gradually improving, and the phenomenon of vehicles of different intelligence levels driving together in traffic flow is becoming increasingly common. Existing traffic signal control methods, mainly based on traditional traffic flow detection and fixed-phase control strategies, are ill-suited to this complex and ever-changing traffic environment. For intelligent vehicles, especially Level 3-4 vehicles with advanced autonomous driving capabilities, their response to traffic signals and their coordination needs differ significantly from those of ordinary vehicles. Level 3-4 intelligent vehicles can achieve more precise path planning and speed adjustment through vehicle-to-infrastructure (V2I) technology, while Level 1-2 ordinary vehicles rely more on traditional traffic signal instructions. However, current technologies lack a V2I-infrastructure (V2I) signal control method that can effectively integrate vehicles of different intelligence levels, failing to fully realize the potential of intelligent vehicles while also failing to effectively coordinate the driving needs of ordinary vehicles.

[0003] In implementing the embodiments of the present invention, the inventors discovered that the prior art has at least the following problems or defects: existing traffic signal control methods cannot accurately identify and process the cooperative needs of vehicles with different intelligence levels, which affects the continuity and safety of traffic flow; there is a lack of effective technical means for trajectory prediction and cooperative control of intelligent vehicles, making it impossible to achieve efficient vehicle-road cooperation; in the processing of multi-source heterogeneous traffic data, the existing technology is difficult to achieve spatiotemporal fusion of data and accurate intelligent agent interaction modeling, thus making it impossible to accurately predict traffic conflict points and optimize signal phase cooperative control strategies. Summary of the Invention

[0004] This invention provides a signal control method for vehicle-road cooperation of different intelligence levels based on ABM and deep learning models.

[0005] In a first aspect of the present invention, a signal control method for vehicle-road cooperation at different intelligence levels based on ABM and a deep learning model is provided, comprising:

[0006] Spatiotemporal units are divided into multiple heterogeneous traffic flow data from multiple sources to obtain the topological association between road segment unit data and the road segment units.

[0007] The road segment unit data is processed using the ABM agent interaction model to obtain the complete trajectory agents of L3-L4 level intelligent vehicles and the boundary association agents of L1-L2 level ordinary vehicles in the road segment unit, and the interaction type identifier is determined based on the collaborative boundary of the boundary association agents.

[0008] For each boundary-associated agent, perform the following operations:

[0009] For a target intelligent agent, adjacent road segment units are determined based on the interaction type identifier, the road segment unit to which it belongs, and the topological association relationship;

[0010] Based on the interaction type identifier, candidate agents are determined from all boundary-associated agents included in the adjacent road segment unit;

[0011] Based on the cooperative perception coordinates of the target intelligent agent's cooperative boundary and the cooperative perception coordinates of each candidate intelligent agent's cooperative boundary, an associated intelligent agent is determined among the candidate intelligent agents, and the associated intelligent agent is assigned the same cooperative identifier as the target intelligent agent; wherein the cooperative perception coordinates are spatiotemporal fusion coordinates in the vehicle-road cooperative perception coordinate system, and the associated intelligent agent is an intelligent agent that satisfies the traffic flow continuity threshold with the target intelligent agent.

[0012] Count the number of agents with complete trajectories and the number of cooperative identifiers;

[0013] A density distribution matrix is ​​generated based on the weighted sum of the complete trajectory agent and the cooperative identifier. Traffic conflict points are predicted through a deep learning model, and the signal phase cooperative control strategy is optimized.

[0014] Furthermore, when the interaction type identifier includes a left boundary identifier or a right boundary identifier for the traffic flow direction, the associated intelligent agent is determined based on the cooperative perception coordinates of the target intelligent agent's cooperative boundary and the cooperative perception coordinates of each candidate intelligent agent's cooperative boundary, including:

[0015] Calculate the continuity matching degree between the longitudinal perception coordinates of the target intelligent agent's traffic flow direction boundary and the longitudinal perception coordinates of the candidate intelligent agent's traffic flow direction boundary;

[0016] Candidate agents with a continuity matching degree greater than the traffic flow continuity threshold are identified as associated agents, and L4-level intelligent vehicles are triggered as collaborative agent nodes to perform trajectory verification.

[0017] Furthermore, the calculation of the continuity matching degree includes:

[0018] The spatiotemporal relationship is determined based on the longitudinal perception coordinates of the target agent and the candidate agents. The spatiotemporal relationship can be either a positive following relationship or a negative conflict relationship.

[0019] In a positive following relationship, calculate the phase difference between the longitudinal coordinates of the end point of the leading agent's boundary and the longitudinal coordinates of the starting point of the following agent's boundary;

[0020] In the reverse conflict relationship, calculate the phase difference between the longitudinal coordinates of the starting point of the conflict agent boundary and the longitudinal coordinates of the conflict avoidance boundary;

[0021] When the vertical axis of the vehicle-road cooperative perception coordinate system represents the time dimension, a positive phase difference is used as the continuity matching degree; when the vertical axis represents the spatial dimension, the absolute value of the negative phase difference is used as the continuity matching degree.

[0022] Furthermore, determining the spatiotemporal correlation includes:

[0023] Extract the spatiotemporal feature point coordinates of the target agent cooperative boundary and the candidate agent cooperative boundary;

[0024] Based on the projection of the spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, it is determined whether the two belong to the vehicle queuing following mode or the cross-collision mode, where:

[0025] If the spatiotemporal feature point of the left agent is located within the upstream time window of the feature point of the right agent, it is determined to be a positive following relationship;

[0026] If there is a spatiotemporal overlap between the spatiotemporal feature points of the left agent and the feature points of the right agent, it is determined to be a reverse conflict relationship.

[0027] Furthermore, it also includes:

[0028] The lateral offset is calculated based on the lateral sensing coordinates of the candidate agent's traffic flow direction boundary and the target agent's boundary.

[0029] If the lateral offset is less than the preset lane tolerance value, the longitudinal continuity matching degree calculation is triggered.

[0030] If the lateral offset exceeds the lane tolerance value, it is marked as a cross-lane interference source and excluded from association.

[0031] Furthermore, when the interaction type identifier includes a traffic flow direction boundary identifier or a lower boundary identifier, determining the associated agent includes:

[0032] Calculate the lane alignment between the lateral perception coordinates of the target agent's cooperative boundary and the lateral perception coordinates of the candidate agent's boundary.

[0033] Candidate agents whose lane alignment is greater than the lane keeping threshold are identified as associated agents, and the expected paths of each agent are synchronized through V2X communication.

[0034] Furthermore, the calculation of lane alignment includes:

[0035] In the vehicle-road cooperative perception coordinate system, the curvature features of key points at the boundary between the target intelligent agent and the candidate intelligent agents are extracted;

[0036] When the curvature feature conforms to the lane centerline fitting model, calculate the standard deviation of the lateral coordinates relative to the lane centerline.

[0037] The reciprocal of the standard deviation is used as the lane alignment degree, and it is corrected by incorporating lane keeping status data reported by L3-level intelligent vehicles.

[0038] Furthermore, it also includes:

[0039] Statistically analyze the spatiotemporal boundary coordinates of boundary-associated agents with the same cooperative identifier, and extract the minimum / maximum lateral coordinates and minimum / maximum longitudinal coordinates;

[0040] A multi-agent trajectory fusion canvas is constructed based on coordinate extreme values, and the local trajectories of related agents are spliced ​​into a complete trajectory according to spatiotemporal relationships.

[0041] The continuity of the trajectory is verified by a convolutional neural network. If a trajectory break is detected, a cooperative identifier splitting instruction is generated and the identifier is reassigned.

[0042] Furthermore, the method also includes:

[0043] When an L4-level intelligent vehicle acts as a collaborative agent node, its high-precision positioning data is used as the baseline trajectory.

[0044] Calculate the matching error between the trajectory of a normal vehicle and the baseline trajectory. If the error exceeds a preset threshold, trigger dynamic adjustment of the identifier.

[0045] When the matching error is ≤5%, the original cooperative identifier is retained;

[0046] When the matching error is 5%-15%, a new sub-identifier is added and associated with the baseline trajectory;

[0047] When the matching error is greater than 15%, create an independent collaborative identifier and mark it as an abnormal traffic flow.

[0048] In a second aspect of the present invention, a vehicle-road cooperative edge computing device is provided, comprising:

[0049] The multi-source heterogeneous data access module is used to acquire perception data from roadside units (RSUs) and collaborative information from vehicle on-board units (OBUs).

[0050] ABM agent management engine, used to perform agent interaction and identifier allocation in the method described in the first aspect;

[0051] The signal optimization decision module is used to generate phase control commands based on the density distribution matrix and collision point prediction results.

[0052] The V2X communication module is used to send collaborative control strategies to traffic lights and intelligent vehicles.

[0053] The embodiments of the present invention have at least the following beneficial effects: The method described in this invention can achieve efficient spatiotemporal unit division of multi-source heterogeneous traffic flow data, accurately process road segment unit data through the ABM agent interaction model, accurately distinguish the cooperative needs of L3-L4 level intelligent vehicles and L1-L2 level ordinary vehicles, and achieve effective association between agents through the allocation of cooperative identifiers. This method can generate a density distribution matrix based on the weighted sum of complete trajectory agents and cooperative identifiers, use a deep learning model to predict traffic conflict points, thereby optimizing the signal phase cooperative control strategy and improving the overall operating efficiency and safety of the traffic system.

[0054] Furthermore, this invention can further enhance the collaborative effect between vehicles of different intelligence levels by recognizing and processing various intelligent agent interaction types, such as calculating the continuity matching degree of traffic flow direction boundaries and lane alignment degree. By synchronizing the expected paths of each intelligent agent through V2X communication, and by verifying and dynamically adjusting the continuity of trajectories in real time, it can effectively cope with various interference factors in complex traffic environments, ensuring the continuity and stability of traffic flow. Simultaneously, for the application of L4 level intelligent vehicles as collaborative agent nodes, it can further improve the accuracy and reliability of collaborative control, providing strong support for the efficient operation of intelligent transportation systems. Attached Figure Description

[0055] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent from the following detailed description taken in conjunction with the accompanying drawings. Several embodiments of the invention are illustrated in the drawings by way of example and not limitation, wherein:

[0056] Figure 1 A flowchart illustrating a signal control method for vehicle-road cooperation at different intelligence levels based on ABM and deep learning models, provided in an embodiment of the present invention.

[0057] Figure 2 This is a schematic diagram of the structure of a vehicle-road cooperative edge computing device provided in an embodiment of the present invention. Detailed Implementation

[0058] The principles and spirit of the invention will now be described with reference to several exemplary embodiments. It should be understood that these embodiments are provided merely to enable those skilled in the art to better understand and implement the invention, and are not intended to limit the scope of the invention in any way. Rather, these embodiments are provided to make the invention more thorough and complete, and to fully convey the scope of the invention to those skilled in the art.

[0059] Those skilled in the art will understand that embodiments of the present invention can be implemented as a system, apparatus, device, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0060] It should be noted that the number of any elements in the accompanying drawings is for illustrative purposes only and not as a limitation, and any naming is for distinction only and has no limiting meaning.

[0061] The following is for reference. Figure 1 , Figure 1 This is a flowchart illustrating a signal control method for vehicle-road cooperation at different intelligence levels based on ABM and deep learning models, provided as an embodiment of the present invention. Figure 1 As shown, a signal control method for vehicle-road cooperation at different intelligence levels based on ABM and deep learning models includes:

[0062] S1 divides multi-source heterogeneous traffic flow data into spatiotemporal units to obtain the road segment unit data and the topological relationship between the road segment units;

[0063] S2 uses the ABM agent interaction model to process the road segment unit data, obtains the complete trajectory agents of L3-L4 level intelligent vehicles and the boundary association agents of L1-L2 level ordinary vehicles in the road segment unit, and determines the interaction type identifier based on the collaborative boundary of the boundary association agents.

[0064] For each boundary-associated agent, perform the following operations:

[0065] S3 determines adjacent road segment units for a target intelligent agent based on the interaction type identifier, the road segment unit to which it belongs, and the topological association relationship;

[0066] S4 determines the candidate agents among all boundary-associated agents included in the adjacent road segment unit based on the interaction type identifier;

[0067] S5 determines the associated intelligent agent among the candidate intelligent agents based on the cooperative perception coordinates of the target intelligent agent's cooperative boundary and the cooperative perception coordinates of each candidate intelligent agent's cooperative boundary, and assigns the associated intelligent agent the same cooperative identifier as the target intelligent agent; wherein the cooperative perception coordinates are spatiotemporal fusion coordinates in the vehicle-road cooperative perception coordinate system, and the associated intelligent agent is an intelligent agent that satisfies the traffic flow continuity threshold with the target intelligent agent.

[0068] S6 counts the number of agents with complete trajectories and the number of cooperative identifiers;

[0069] S7 generates a density distribution matrix based on the weighted sum of the complete trajectory agent and the cooperative identifier, predicts traffic conflict points through a deep learning model, and optimizes the signal phase cooperative control strategy.

[0070] It should be noted that this invention proposes a vehicle-road cooperative signal control method based on ABM (Agent-Based Model) and deep learning models for different levels of vehicle intelligence. The core of this method lies in dividing multi-source heterogeneous traffic flow data into spatiotemporal units to obtain road segment unit data and their topological relationships. The multi-source heterogeneous traffic flow data in this application refers to traffic data from different sources (such as Roadside Units (RSUs), Vehicle On-Board Units (OBUs), etc.) with different data formats and types. Spatiotemporal unit division involves dividing these data according to time and space dimensions for better processing and analysis. Topological relationships refer to the spatial connectivity between road segment units, such as which road segments are adjacent and which intersect. This approach provides fundamental data support for subsequent agent interaction and signal control strategy optimization.

[0071] Specifically, the road segment unit data mentioned in this invention refers to various data related to traffic flow within the divided spatiotemporal units, such as vehicle speed, position, and direction of travel. Topological relationships refer to the spatial connections between these road segment units, such as adjacency and intersection relationships. In this embodiment, a road segment unit can be a specific area on a road, such as an intersection, a straight section, or a curve. Topological relationships can be defined by the connection points between road segment units; for example, if two road segment units are connected through an intersection, then a topological relationship exists between them. Furthermore, the ABM (Automatic Agent Interaction Model) is a model that simulates the interaction between intelligent agents and is used to process road segment unit data. In this invention, intelligent agents can be vehicles, categorized into L3-L4 level intelligent vehicles and L1-L2 level ordinary vehicles based on their intelligence level. L3-L4 level intelligent vehicles possess higher autonomous driving capabilities and can communicate and collaborate with other vehicles and infrastructure through vehicle-to-everything (V2X) technology; while L1-L2 level ordinary vehicles primarily rely on traditional driver assistance systems. The ABM model can simulate the interaction behavior between vehicles with different intelligence levels, thus providing a basis for optimizing signal control strategies.

[0072] Preferably, the following specific method can be used for the step of dividing multi-source heterogeneous traffic flow data into spatiotemporal units: First, the road is divided into multiple road segment units according to the road's geometric features and traffic flow characteristics. For example, the road can be divided into different road segment units according to the location of intersections, and each road segment unit can be a straight-through segment, a left-turn segment, or a right-turn segment, etc. Then, the traffic flow data within each road segment unit is divided into multiple spatiotemporal units according to the time window. For example, a day can be divided into multiple time periods, such as morning peak, off-peak, and evening peak, and the traffic flow data within each time period is considered as a spatiotemporal unit. For the step of processing the road segment unit data using the ABM intelligent agent interaction model, the following specific method can be used: First, the state of the intelligent agent is initialized based on the vehicle data within the road segment unit, including position, speed, and driving direction. Then, the interaction between the intelligent agents is simulated according to the interaction rules between them. For example, when two vehicles are traveling in the same lane, they may have interaction behaviors such as following or overtaking; when two vehicles meet at an intersection, they may have interaction behaviors such as conflict or yielding. By simulating these interactive behaviors, the trajectory and state changes of the agent can be obtained, thus providing a basis for optimizing the signal control strategy.

[0073] In some embodiments, when the interaction type identifier includes a left boundary identifier or a right boundary identifier for the traffic flow direction, determining the associated agent based on the cooperative perception coordinates of the target agent's cooperative boundary and the cooperative perception coordinates of each candidate agent's cooperative boundary includes:

[0074] Calculate the continuity matching degree between the longitudinal perception coordinates of the target intelligent agent's traffic flow direction boundary and the longitudinal perception coordinates of the candidate intelligent agent's traffic flow direction boundary;

[0075] Candidate agents with a continuity matching degree greater than the traffic flow continuity threshold are identified as associated agents, and L4-level intelligent vehicles are triggered as collaborative agent nodes to perform trajectory verification.

[0076] It should be noted that this invention pays particular attention to cases where the interaction type is identified by the left or right boundary of the traffic flow direction when handling interactions between vehicles of different intelligence levels. In this case, the associated intelligent agent is determined by calculating the cooperative perception coordinates of the target intelligent agent and candidate intelligent agents. Here, cooperative perception coordinates refer to coordinates that fuse temporal and spatial information in the vehicle-road cooperative perception coordinate system, used to describe the vehicle's position and temporal state in the traffic flow. In this way, candidate intelligent agents with a continuous relationship with the target intelligent agent in the traffic flow direction can be effectively identified, and L4-level intelligent vehicles can be further triggered as cooperative agent nodes to perform trajectory verification, thereby ensuring the continuity and safety of the traffic flow.

[0077] Specifically, the interaction type identifiers mentioned in this invention refer to markers used to distinguish the interaction relationships between different agents, such as left or right boundary identifiers for traffic flow direction. These identifiers are used to indicate the relative position and direction of movement of agents in traffic flow. For example, a left boundary identifier indicates that the agent is located on the left boundary of the traffic flow direction, while a right boundary identifier indicates that the agent is located on the right boundary of the traffic flow direction. In this embodiment, the target agent refers to the agent currently being processed, while candidate agents refer to other agents in adjacent road segment units that may have an interaction relationship with the target agent. By calculating the cooperative perception coordinates of the target agent and candidate agents, their spatiotemporal correlation can be determined. Here, the spatiotemporal correlation refers to the relative positional relationship of agents in time and space, such as a forward following relationship or a reverse conflict relationship. A forward following relationship means that one agent follows another agent in time, while a reverse conflict relationship means that there is a possibility of conflict between two agents in time and space.

[0078] Preferably, the following specific steps can be used to calculate the cooperative perception coordinates of the target agent and the candidate agent: First, extract the spatiotemporal feature point coordinates of the target agent and the candidate agent. These coordinates include information in both time and space dimensions. For example, information such as the vehicle's current position, speed, acceleration, and timestamp can be extracted. Then, based on the projection of these spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, determine the spatiotemporal association between the target agent and the candidate agent. Specifically, for a forward following relationship, calculate the phase difference between the longitudinal coordinates of the leading agent's boundary endpoint and the longitudinal coordinates of the following agent's boundary starting point; for a reverse conflict relationship, calculate the phase difference between the longitudinal coordinates of the conflicting agent's boundary starting point and the longitudinal coordinates of the conflict avoidance boundary. When the vertical axis of the vehicle-road cooperative perception coordinate system represents the time dimension, a positive phase difference can be used as the continuity matching degree; when the vertical axis represents the spatial dimension, the absolute value of a negative phase difference can be used as the continuity matching degree. In this way, the continuity matching degree between the target agent and the candidate agent can be accurately calculated, thereby determining the associated agent.

[0079] In some embodiments, calculating the continuity matching degree includes:

[0080] The spatiotemporal relationship is determined based on the longitudinal perception coordinates of the target agent and the candidate agents. The spatiotemporal relationship can be either a positive following relationship or a negative conflict relationship.

[0081] In a positive following relationship, calculate the phase difference between the longitudinal coordinates of the end point of the leading agent's boundary and the longitudinal coordinates of the starting point of the following agent's boundary;

[0082] In the reverse conflict relationship, calculate the phase difference between the longitudinal coordinates of the starting point of the conflict agent boundary and the longitudinal coordinates of the conflict avoidance boundary;

[0083] When the vertical axis of the vehicle-road cooperative perception coordinate system represents the time dimension, a positive phase difference is used as the continuity matching degree; when the vertical axis represents the spatial dimension, the absolute value of the negative phase difference is used as the continuity matching degree.

[0084] It should be noted that, in determining spatiotemporal relationships, this invention extracts the spatiotemporal feature point coordinates of the target agent and candidate agents, and judges the relationship between them based on the projection of these coordinates into the vehicle-road cooperative perception coordinate system. Here, the spatiotemporal feature point coordinates refer to the key point coordinates describing the position and temporal state of the agent in the vehicle-road cooperative perception coordinate system. These coordinates integrate temporal and spatial information to accurately describe the agent's motion state in traffic flow. In this way, it is possible to determine whether there is a vehicle queuing following pattern or a cross-collision pattern between agents, thus providing a basis for subsequent continuous matching degree calculation.

[0085] Specifically, the spatiotemporal correlation mentioned in this invention refers to the relative positional relationship between the target agent and candidate agents in time and space. This relationship can be divided into a forward following relationship and a reverse conflict relationship. A forward following relationship means that one agent follows another agent in time. For example, when the spatiotemporal feature point of the leading agent is located within the upstream time window of the following agent, it can be determined as a forward following relationship. A reverse conflict relationship means that there is a possibility of conflict between two agents in time and space. For example, when there is a spatiotemporal overlap between the spatiotemporal feature points of the left agent and the right agent, it can be determined as a reverse conflict relationship. The coordinates of the spatiotemporal feature points here include the agent's position coordinates (such as longitude and latitude), timestamp, and speed information. With this information, projection analysis can be performed in the vehicle-road cooperative perception coordinate system to accurately determine the spatiotemporal correlation between agents.

[0086] Preferably, to more accurately calculate the continuity matching degree, the following specific steps can be adopted: First, extract the spatiotemporal feature point coordinates of the target agent and candidate agents. These coordinates include information such as the agent's position, velocity, acceleration, and timestamp. Then, based on the projection of these spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, determine the spatiotemporal association between the target agent and candidate agents. For a forward following relationship, calculate the phase difference between the longitudinal coordinates of the end point of the leading agent's boundary and the longitudinal coordinates of the starting point of the following agent's boundary; for a reverse conflict relationship, calculate the phase difference between the longitudinal coordinates of the starting point of the conflicting agent's boundary and the longitudinal coordinates of the conflict avoidance boundary. Specifically, when the vertical axis of the vehicle-road cooperative perception coordinate system represents the time dimension, a positive phase difference can be used as the continuity matching degree; when the vertical axis represents the spatial dimension, the absolute value of a negative phase difference can be used as the continuity matching degree. In this way, the continuity matching degree between the target agent and candidate agents can be accurately calculated, thereby determining the associated agents.

[0087] In some embodiments, determining the spatiotemporal correlation includes:

[0088] Extract the spatiotemporal feature point coordinates of the target agent cooperative boundary and the candidate agent cooperative boundary;

[0089] Based on the projection of the spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, it is determined whether the two belong to the vehicle queuing following mode or the cross-collision mode, where:

[0090] If the spatiotemporal feature point of the left agent is located within the upstream time window of the feature point of the right agent, it is determined to be a positive following relationship;

[0091] If there is a spatiotemporal overlap between the spatiotemporal feature points of the left agent and the feature points of the right agent, it is determined to be a reverse conflict relationship.

[0092] It should be noted that, in determining spatiotemporal relationships, this invention extracts the spatiotemporal feature point coordinates of the target agent and candidate agents, and judges the relationship between them based on the projection of these coordinates into the vehicle-road cooperative perception coordinate system. Here, the spatiotemporal feature point coordinates refer to the key point coordinates describing the position and temporal state of the agent in the vehicle-road cooperative perception coordinate system. These coordinates integrate temporal and spatial information to accurately describe the agent's motion state in traffic flow. In this way, it is possible to determine whether there is a vehicle queuing following pattern or a cross-collision pattern between agents, thus providing a basis for subsequent continuous matching degree calculation.

[0093] Specifically, the spatiotemporal feature point coordinates mentioned in this invention refer to the position and time information of the intelligent agent in the vehicle-road cooperative perception coordinate system, including but not limited to the intelligent agent's current position (such as longitude and latitude), speed, acceleration, and timestamp. This information is used to describe the movement state of the intelligent agent in traffic flow. The spatiotemporal correlation refers to the relative positional relationship between the target intelligent agent and candidate intelligent agents in time and space, which can be divided into forward following relationship and reverse conflict relationship. A forward following relationship means that one intelligent agent follows another intelligent agent in time. For example, when the spatiotemporal feature point of the leading intelligent agent is located within the upstream time window of the following intelligent agent, it can be determined as a forward following relationship. A reverse conflict relationship refers to the possibility of conflict between two intelligent agents in time and space. For example, when the spatiotemporal feature points of the left intelligent agent and the right intelligent agent have a spatiotemporal overlap, it can be determined as a reverse conflict relationship. Here, the upstream time window refers to the time interval in the time dimension where the position of the leading intelligent agent is before that of the following intelligent agent, and it is usually set according to the speed of traffic flow and vehicle spacing. For example, if the average speed of the traffic flow is 60 km / h and the vehicle spacing is 50 meters, the upstream time window can be set to the time required for a vehicle to travel that distance, i.e., 3 seconds.

[0094] Preferably, to more accurately determine the spatiotemporal correlation, the following specific steps can be adopted: First, extract the spatiotemporal feature point coordinates of the target agent and candidate agents. These coordinates include information such as the agent's position, velocity, acceleration, and timestamp. Then, based on the projection of these spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, determine the spatiotemporal correlation between the target agent and candidate agents. The specific determination method is as follows: if the spatiotemporal feature point of the left agent is located within the upstream time window of the feature point of the right agent, that is, the timestamp of the left agent is earlier than the timestamp of the right agent, and the distance between them is within the preset vehicle spacing range, then it is determined to be a positive following relationship; if the spatiotemporal feature points of the left agent and the right agent have a spatiotemporal overlap region, that is, they appear in the same spatial region within the same time period, then it is determined to be a negative conflict relationship. In this way, the spatiotemporal correlation between the target agent and candidate agents can be accurately determined, thus providing an accurate basis for subsequent continuous matching degree calculation.

[0095] In some embodiments, it also includes:

[0096] The lateral offset is calculated based on the lateral sensing coordinates of the candidate agent's traffic flow direction boundary and the target agent's boundary.

[0097] If the lateral offset is less than the preset lane tolerance value, the longitudinal continuity matching degree calculation is triggered.

[0098] If the lateral offset exceeds the lane tolerance value, it is marked as a cross-lane interference source and excluded from association.

[0099] It should be noted that, when processing lateral offsets between agents, this invention determines the presence of cross-lane interference by calculating the lateral sensing coordinates of the candidate agent's traffic flow direction boundary and the target agent's boundary. Here, the lateral sensing coordinates refer to the coordinate information describing the agent's lateral position in the vehicle-road cooperative sensing coordinate system, used to determine whether the agent is within the same lane or engaging in cross-lane behavior. In this way, cross-lane interference can be effectively identified, and based on this, a decision can be made on whether to trigger the calculation of longitudinal continuity matching, thereby improving the accuracy and reliability of agent interactions.

[0100] Specifically, the lateral perception coordinates mentioned in this invention refer to the lateral position information of the intelligent agent in the vehicle-road cooperative perception coordinate system, typically measured with the lane centerline as the reference. For example, the lateral offset of the vehicle can be used to represent its position relative to the lane centerline. Here, the lateral offset refers to the distance difference between the actual position of the intelligent agent and the lane centerline. If the lateral offset is less than a preset lane tolerance value, the intelligent agent is considered to be in the same lane, and the calculation of longitudinal continuity matching degree can be triggered; if the lateral offset exceeds the lane tolerance value, the intelligent agent is considered to be engaging in cross-lane behavior, and should be marked as a cross-lane interference source and excluded from association. The lane tolerance value is a preset parameter used to define the maximum allowable offset range of the intelligent agent within the same lane, and is usually set according to the lane width and vehicle driving characteristics. For example, for a road section with a standard lane width of 3.5 meters, the lane tolerance value can be set to 0.5 meters, indicating that a vehicle within 0.5 meters on either side of the lane centerline is considered to be driving normally within the same lane.

[0101] Preferably, to more accurately calculate lateral offset and determine cross-lane interference, the following specific steps can be adopted: First, extract the lateral perception coordinates of the target agent and candidate agents. These coordinates can be obtained through the vehicle's positioning system (such as GPS) and lane detection sensors (such as cameras or LiDAR). Then, calculate the difference between the lateral perception coordinates of the candidate agent and the target agent to obtain the lateral offset. The specific calculation method is: Lateral offset = Lateral perception coordinates of candidate agent - Lateral perception coordinates of target agent. Next, compare the calculated lateral offset with a preset lane tolerance value. If the lateral offset is less than the lane tolerance value, it is considered that the candidate agent and the target agent are in the same lane, and the calculation of longitudinal continuity matching degree can be triggered; if the lateral offset exceeds the lane tolerance value, it is considered that the candidate agent has cross-lane behavior, should be marked as a cross-lane interference source, and the association should be excluded. In this way, cross-lane interference can be effectively identified and handled, improving the accuracy and reliability of agent interaction.

[0102] In some embodiments, when the interaction type identifier includes a traffic flow direction boundary identifier or a lower boundary identifier, determining the associated agent includes:

[0103] Calculate the lane alignment between the lateral perception coordinates of the target agent's cooperative boundary and the lateral perception coordinates of the candidate agent's boundary.

[0104] Candidate agents whose lane alignment is greater than the lane keeping threshold are identified as associated agents, and the expected paths of each agent are synchronized through V2X communication.

[0105] It should be noted that, when processing the directional or lower boundary markers of traffic flow, this invention determines the associated intelligent agent by calculating the lane alignment between the lateral perception coordinates of the target intelligent agent's cooperative boundary and the lateral perception coordinates of the candidate intelligent agent's boundary. Here, lane alignment refers to the degree of alignment between the target intelligent agent and the candidate intelligent agent in the lateral position within the vehicle-road cooperative perception coordinate system, used to determine whether the two intelligent agents are traveling in the same or adjacent lanes. In this way, candidate intelligent agents traveling in the same or adjacent lanes as the target intelligent agent can be effectively identified, and the expected paths of each intelligent agent can be synchronized through V2X communication, thereby improving the coordination and safety of traffic flow.

[0106] Specifically, the lateral perception coordinates mentioned in this invention refer to the lateral position information of the intelligent agent in the vehicle-road cooperative perception coordinate system, typically measured with the lane centerline as a reference. For example, the lateral offset of the vehicle can be used to represent its position relative to the lane centerline. Lane alignment refers to the degree of alignment between the target intelligent agent and the candidate intelligent agent in the lateral position, measured by calculating the standard deviation of the lateral coordinates from the lane centerline. The higher the lane alignment, the better the alignment between the two intelligent agents in the lateral position, meaning a greater likelihood that they are traveling in the same or adjacent lanes. The lane keeping threshold here is a preset parameter used to define the minimum alignment degree for intelligent agents to travel in the same or adjacent lanes. For example, the lane keeping threshold can be set to 0.8, meaning that when the lane alignment is greater than 0.8, the candidate intelligent agent is considered to be traveling in the same or adjacent lane as the target intelligent agent.

[0107] Preferably, to more accurately calculate lane alignment and determine associated agents, the following specific steps can be taken: First, extract the curvature features of key points at the boundary between the target agent and candidate agents. These curvature features are used to describe the agent's trajectory on the lane. Then, in the vehicle-road cooperative perception coordinate system, calculate the standard deviation of the lateral coordinates from the lane centerline, and use the reciprocal of the standard deviation as the lane alignment. The specific calculation method is: Lane alignment = 1 / Standard deviation of lateral coordinates from the lane centerline. Next, compare the calculated lane alignment with a preset lane keeping threshold. If the lane alignment is greater than the lane keeping threshold, the candidate agent is considered to be traveling in the same lane or adjacent lane as the target agent. The candidate agent can be identified as an associated agent, and the expected paths of each agent can be synchronized through V2X communication. In this way, agents in the same lane or adjacent lanes can be effectively identified and processed, improving traffic flow coordination and safety.

[0108] In some embodiments, calculating lane alignment includes:

[0109] In the vehicle-road cooperative perception coordinate system, the curvature features of key points at the boundary between the target intelligent agent and the candidate intelligent agents are extracted;

[0110] When the curvature feature conforms to the lane centerline fitting model, calculate the standard deviation of the lateral coordinates relative to the lane centerline.

[0111] The reciprocal of the standard deviation is used as the lane alignment degree, and it is corrected by incorporating lane keeping status data reported by L3-level intelligent vehicles.

[0112] It should be noted that this invention further refines the curvature feature extraction process of key boundary points between the target agent and candidate agents when calculating lane alignment. By analyzing the curvature features of these key points in the vehicle-road cooperative perception coordinate system, the alignment degree between the lateral coordinate and the lane centerline can be more accurately assessed. Here, curvature features refer to parameters describing the curvature of the agent's driving trajectory; these features can be used to determine whether the agent is maintaining its position on the lane centerline. Furthermore, this invention incorporates lane-keeping status data reported by Level 3 intelligent vehicles for correction, further improving the accuracy and reliability of lane alignment calculation.

[0113] Specifically, the curvature feature mentioned in this invention refers to the degree of curvature of the intelligent agent's driving trajectory, usually represented by the curvature of key points on the trajectory. The curvature feature can reflect the intelligent agent's driving state in the lane, such as whether it deviates from the lane centerline. In the vehicle-road cooperative perception coordinate system, by extracting the curvature features of key points at the boundary between the target intelligent agent and candidate intelligent agents, the standard deviation of the lateral coordinates from the lane centerline can be calculated. Here, the standard deviation refers to the degree of deviation between the lateral coordinates and the lane centerline; the smaller the standard deviation, the closer the intelligent agent is to the lane centerline. Lane alignment is calculated using the reciprocal of the standard deviation, i.e., lane alignment = 1 / standard deviation. Furthermore, lane keeping status data reported by L3-level intelligent vehicles can be used to correct the lane alignment calculation results, improving its accuracy. Lane keeping status data typically includes the vehicle's lateral offset, lane departure warning information, etc., which can be acquired through the vehicle's sensors (such as cameras, LiDAR, etc.).

[0114] Preferably, to calculate lane alignment more accurately, the following specific steps can be taken: First, extract the curvature features of key points at the boundary between the target agent and candidate agents in the vehicle-road cooperative perception coordinate system. These key points can be obtained through the vehicle's positioning system (such as GPS) and sensors (such as cameras, LiDAR, etc.). Then, calculate the standard deviation of the lateral coordinates from the lane centerline, and use the reciprocal of the standard deviation as the preliminary calculation result of lane alignment. The specific calculation method is: Lane alignment = 1 / Standard deviation of lateral coordinates from the lane centerline. Next, the lane alignment is corrected by combining the lane keeping status data reported by L3-level intelligent vehicles. For example, if the lane keeping status data shows that the vehicle has a slight lateral deviation, the calculated result of lane alignment can be adjusted appropriately. In this way, the alignment degree of the agent on the lane can be more accurately evaluated, thereby improving the coordination and safety of traffic flow.

[0115] In some embodiments, it also includes:

[0116] Statistically analyze the spatiotemporal boundary coordinates of boundary-associated agents with the same cooperative identifier, and extract the minimum / maximum lateral coordinates and minimum / maximum longitudinal coordinates;

[0117] A multi-agent trajectory fusion canvas is constructed based on coordinate extreme values, and the local trajectories of related agents are spliced ​​into a complete trajectory according to spatiotemporal relationships.

[0118] The continuity of the trajectory is verified by a convolutional neural network. If a trajectory break is detected, a cooperative identifier splitting instruction is generated and the identifier is reassigned.

[0119] It should be noted that, when processing boundary-associated agents with the same cooperative identifier, this invention extracts the minimum and maximum lateral and longitudinal coordinates by statistically analyzing the spatiotemporal boundary coordinates of these agents, thereby constructing a multi-agent trajectory fusion canvas. This process aims to stitch together the local trajectories of associated agents into a complete trajectory according to spatiotemporal relationships, thus achieving overall cooperative control of traffic flow. Here, the spatiotemporal boundary coordinates refer to the positional information of the agent in the vehicle-road cooperative perception coordinate system, including both time and space dimensions. Through these coordinates, the specific position and motion state of the agent in the traffic flow can be determined. Furthermore, this invention also verifies trajectory continuity using a convolutional neural network (CNN) to ensure the integrity and accuracy of the trajectory. If a trajectory break is detected, a cooperative identifier splitting instruction is generated, and the identifier is reassigned to maintain the continuity and cooperativeness of the traffic flow.

[0120] Specifically, the spatiotemporal boundary coordinates mentioned in this invention refer to the positional information of an agent in the vehicle-road cooperative perception coordinate system, including lateral coordinates (such as lane position) and longitudinal coordinates (such as the vehicle's front and rear position on the road). These coordinates are used to describe the specific position and motion state of the agent in the traffic flow. By statistically analyzing the spatiotemporal boundary coordinates of boundary-associated agents with the same cooperative identifier, the minimum and maximum lateral coordinates and minimum and maximum longitudinal coordinates of these agents can be extracted. These extreme coordinates are used to construct a multi-agent trajectory fusion canvas, that is, to stitch together the local trajectories of multiple agents into a complete trajectory according to spatiotemporal relationships. In addition, a convolutional neural network (CNN) is a deep learning model used to verify trajectory continuity. CNN detects whether there are breaks or anomalies in the trajectory by analyzing the spatiotemporal features in the trajectory data. If a trajectory break is detected, the system will generate a cooperative identifier splitting instruction and reassign the identifier to ensure the continuity and cooperativeness of the traffic flow.

[0121] Preferably, to more accurately achieve multi-agent trajectory fusion and continuity verification, the following specific steps can be adopted: First, the spatiotemporal boundary coordinates of boundary-associated agents with the same cooperative identifier are acquired through sensors (such as GPS, cameras, LiDAR, etc.). Then, the minimum and maximum values ​​of these coordinates are extracted, including the minimum lateral coordinate, maximum lateral coordinate, minimum vertical coordinate, and maximum vertical coordinate. These extreme coordinates are used to construct a multi-agent trajectory fusion canvas, splicing the local trajectories of associated agents into a complete trajectory according to spatiotemporal relationships. Next, a convolutional neural network (CNN) is used to verify the continuity of the spliced ​​trajectory. The CNN model can be constructed through the following steps: 1) Data preprocessing, including normalization and standardization of trajectory data; 2) Model construction, using multiple convolutional layers and pooling layers to extract spatiotemporal features from the trajectory data; 3) Model training, training the CNN model using labeled trajectory data; 4) Model verification, evaluating the model's performance using a verification dataset. During the verification process, if the CNN detects trajectory breaks or anomalies, the system will generate a cooperative identifier splitting instruction and reassign identifiers to maintain the continuity and cooperativeness of traffic flow. This approach can effectively improve the coordination and safety of traffic flow.

[0122] In some embodiments, the method further includes:

[0123] When an L4-level intelligent vehicle acts as a collaborative agent node, its high-precision positioning data is used as the baseline trajectory.

[0124] Calculate the matching error between the trajectory of a normal vehicle and the baseline trajectory. If the error exceeds a preset threshold, trigger dynamic adjustment of the identifier.

[0125] When the matching error is ≤5%, the original cooperative identifier is retained;

[0126] When the matching error is 5%-15%, a new sub-identifier is added and associated with the baseline trajectory;

[0127] When the matching error is greater than 15%, create an independent collaborative identifier and mark it as an abnormal traffic flow.

[0128] It should be noted that when processing Level 4 intelligent vehicles as cooperative agent nodes, this invention utilizes their high-precision positioning data as a reference trajectory to calculate the matching error between the trajectory of ordinary vehicles and the reference trajectory. In this way, the cooperative identifier can be dynamically adjusted according to the magnitude of the matching error, thereby ensuring the continuity and coordination of traffic flow. Here, the matching error refers to the degree of deviation between the trajectory of ordinary vehicles and the reference trajectory of Level 4 intelligent vehicles, used to assess whether ordinary vehicles can effectively follow the reference trajectory. By setting different error thresholds, the cooperative identifier can be dynamically adjusted to adapt to different traffic scenarios and vehicle behaviors.

[0129] Specifically, the matching error mentioned in this invention refers to the degree of deviation between the trajectory of a conventional vehicle and the reference trajectory of an L4-level intelligent vehicle. The reference trajectory is generated using high-precision positioning data from the L4-level intelligent vehicle, typically sourced from sensors such as the vehicle's high-precision GPS and IMU (Inertial Measurement Unit). The matching error can be evaluated by calculating parameters such as the distance difference and speed difference between the conventional vehicle's trajectory and the reference trajectory. For example, the Euclidean distance between the conventional vehicle's position at each time point and the corresponding position on the reference trajectory can be calculated to obtain the position matching error. Furthermore, the speed difference between the conventional vehicle's speed and the speed of the corresponding point on the reference trajectory can be calculated to obtain the speed matching error. By combining these errors, an overall matching error value can be obtained. The error threshold here is a preset parameter used to define the maximum permissible deviation range between the conventional vehicle's trajectory and the reference trajectory. For example, the matching error threshold can be set to 5%, indicating that when the matching error is less than or equal to 5%, the deviation between the conventional vehicle's trajectory and the reference trajectory is within an acceptable range.

[0130] Preferably, to more accurately calculate the matching error and dynamically adjust the cooperative identifier, the following specific steps can be adopted: First, acquire high-precision positioning data of L4-level intelligent vehicles and use it as the reference trajectory. Then, calculate the matching error between the trajectory of ordinary vehicles and the reference trajectory. Specifically, the calculation method can be: for each time point, calculate the Euclidean distance between the position of the ordinary vehicle and the corresponding position on the reference trajectory to obtain the position matching error; calculate the speed difference between the speed of the ordinary vehicle and the speed of the corresponding point on the reference trajectory to obtain the speed matching error. Combine these errors to obtain the overall matching error value. Next, dynamically adjust the cooperative identifier based on a comparison between the matching error value and a preset error threshold. For example, when the matching error is less than or equal to 5%, maintain the original cooperative identifier; when the matching error is between 5% and 15%, add a sub-identifier and associate it with the reference trajectory; when the matching error is greater than 15%, create an independent cooperative identifier and mark it as abnormal traffic flow. In this way, changes in vehicle behavior under different traffic scenarios can be effectively addressed, ensuring the continuity and coordination of traffic flow.

[0131] The various embodiments of the present invention have the following beneficial effects: The present invention can improve the accuracy of vehicle-road cooperative control in mixed traffic flow environments. Through agent modeling and spatiotemporal data fusion, it can accurately identify the complete trajectories of L3-L4 level intelligent vehicles and the cooperative boundaries of L1-L2 level vehicles, and establish correlations based on traffic flow continuity thresholds. This multi-agent cooperative mechanism can dynamically generate a density distribution matrix and combine deep learning to predict conflict points, thereby optimizing signal timing strategies and alleviating traffic congestion. In addition, through dual verification of longitudinal matching degree and lateral alignment degree, it can accurately determine the following or conflict relationship between vehicles, avoid misjudgment interference, and improve the reliability of cooperative control.

[0132] This invention can also achieve trajectory collaborative verification across vehicle levels. Utilizing high-precision positioning data from L4-level vehicles as the baseline trajectory, the collaborative identifier allocation for ordinary vehicles can be dynamically corrected, ensuring the accuracy of traffic flow grouping. By constructing a multi-agent trajectory fusion canvas, trajectory breakage issues and flow continuity can be automatically detected and repaired. Simultaneously, the expected path synchronization mechanism based on V2X communication can further improve the collaborative efficiency of lane keeping and conflict avoidance, ultimately achieving efficient collaborative passage for vehicles of different intelligence levels.

[0133] like Figure 2 As shown in some embodiments, a vehicle-road cooperative edge computing device includes:

[0134] The multi-source heterogeneous data access module 201 is used to acquire perception data from roadside units (RSUs) and collaborative information from vehicle on-board units (OBUs).

[0135] ABM agent management engine 202 is used to perform agent interaction and identifier allocation in the method described in the first aspect;

[0136] The signal optimization decision module 203 is used to generate phase control commands based on the density distribution matrix and the collision point prediction results.

[0137] V2X communication module 204 is used to send collaborative control strategies to traffic lights and intelligent vehicles.

[0138] It should be noted that this invention proposes a vehicle-road cooperative edge computing device. This device integrates multiple functional modules to achieve efficient processing of traffic flow data and optimization of signal control strategies. The core modules of this device include a multi-source heterogeneous data access module, an ABM (Automatic Vehicle Management) intelligent agent management engine, a signal optimization decision module, and a V2X communication module. These modules work together to acquire traffic data in real time, handle the interaction needs of vehicles with different intelligence levels, and distribute the optimized signal control strategies to traffic signal controllers and intelligent vehicles via V2X communication, thereby improving the overall operational efficiency and safety of the traffic system. Here, the vehicle-road cooperative edge computing device refers to a computing device deployed at the edge of the road for real-time processing of traffic data and generation of control commands.

[0139] Specifically, the multi-source heterogeneous data access module mentioned in this invention is used to acquire traffic data from different sources, including perception data from roadside units (RSUs) and collaborative information from vehicle on-board units (OBUs). This data may include vehicle location, speed, and direction of travel. The ABM (Automatic Vehicle Management) engine is responsible for executing agent interaction and identifier allocation, generating collaborative control strategies by simulating the interactive behaviors of vehicles at different intelligence levels. The signal optimization decision module generates phase control commands based on the density distribution matrix and conflict point prediction results. These commands are used to optimize the phase of traffic signals to reduce traffic conflicts and improve traffic efficiency. Finally, the V2X communication module is responsible for sending these control commands to traffic lights and intelligent vehicles to ensure traffic flow coordination. The collaborative work of these modules enables the vehicle-road cooperative edge computing device to respond to traffic changes in real time and optimize traffic signal control.

[0140] Preferably, to more efficiently realize the functions of the vehicle-road cooperative edge computing device, each module can be further refined. For example, the multi-source heterogeneous data access module can access data from the RSU and OBU through various sensor interfaces, supporting multiple data formats and communication protocols. In the ABM agent management engine, an agent-based model can be built, with input parameters including the vehicle's initial position, speed, and direction of travel. By simulating vehicle-to-vehicle interaction behaviors, such as overtaking and yielding, a cooperative control strategy is generated. The signal optimization decision module can optimize the signal phase based on the output of the agent management engine and the traffic conflict points predicted by the deep learning model. For example, if a traffic conflict is predicted at an intersection, the signal phase can be adjusted to extend the green light time and reduce the conflict. Finally, the V2X communication module can use wireless communication technologies, such as 5G or DSRC (Dedicated Short Range Communication), to send control commands to the traffic signal controller and intelligent vehicles in real time. In this way, the vehicle-road cooperative edge computing device can efficiently process traffic data, optimize signal control strategies, and improve the overall operating efficiency and safety of the traffic system.

[0141] Furthermore, the storage medium in the embodiments of this application stores program instructions capable of implementing all the above methods. These program instructions can be stored in the storage medium in the form of a software product, including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks, or terminal devices such as computers, servers, mobile phones, and tablets.

[0142] The above description is merely a selection of preferred embodiments of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention as described in the embodiments is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention.

Claims

1. A signal control method for vehicle-road cooperation of different intelligence levels based on ABM and deep learning models, characterized in that, include: Spatiotemporal units are divided into multiple heterogeneous traffic flow data from multiple sources to obtain the topological association between road segment unit data and the road segment units. The road segment unit data is processed using the ABM agent interaction model to obtain the complete trajectory agents of L3-L4 level intelligent vehicles and the boundary association agents of L1-L2 level ordinary vehicles in the road segment unit, and the interaction type identifier is determined based on the collaborative boundary of the boundary association agents. For each boundary-associated agent, perform the following operations: For a target intelligent agent, adjacent road segment units are determined based on the interaction type identifier, the road segment unit to which it belongs, and the topological association relationship; Based on the interaction type identifier, candidate agents are determined from all boundary-associated agents included in the adjacent road segment unit; Based on the cooperative perception coordinates of the target agent's cooperative boundary and the cooperative perception coordinates of each candidate agent's cooperative boundary, an associated agent is determined among the candidate agents, and the associated agent is assigned the same cooperative identifier as the target agent. The collaborative perception coordinates are spatiotemporal fusion coordinates in the vehicle-road collaborative perception coordinate system, and the associated intelligent agent is an intelligent agent that satisfies the traffic flow continuity threshold with the target intelligent agent; Count the number of agents with complete trajectories and the number of cooperative identifiers; A density distribution matrix is ​​generated based on the weighted sum of the complete trajectory agent and the cooperative identifier. Traffic conflict points are predicted through a deep learning model, and the signal phase cooperative control strategy is optimized. When the interaction type identifier includes a left boundary identifier or a right boundary identifier for the traffic flow direction, the associated intelligent agent is determined based on the cooperative perception coordinates of the target intelligent agent's cooperative boundary and the cooperative perception coordinates of each candidate intelligent agent's cooperative boundary, including: Calculate the continuity matching degree between the longitudinal perception coordinates of the target intelligent agent's traffic flow direction boundary and the longitudinal perception coordinates of the candidate intelligent agent's traffic flow direction boundary; Candidate agents with a continuity matching degree greater than the traffic flow continuity threshold are identified as associated agents, and L4-level intelligent vehicles are triggered as collaborative agent nodes to perform trajectory verification. The calculation of the continuity matching degree includes: The spatiotemporal relationship is determined based on the longitudinal perception coordinates of the target agent and the candidate agents. The spatiotemporal relationship can be either a positive following relationship or a negative conflict relationship. In a positive following relationship, calculate the phase difference between the longitudinal coordinates of the end point of the leading agent's boundary and the longitudinal coordinates of the starting point of the following agent's boundary; In the reverse conflict relationship, calculate the phase difference between the longitudinal coordinates of the starting point of the conflict agent boundary and the longitudinal coordinates of the conflict avoidance boundary; When the vertical axis of the vehicle-road cooperative perception coordinate system represents the time dimension, a positive phase difference is used as the continuity matching degree; when the vertical axis represents the spatial dimension, the absolute value of the negative phase difference is used as the continuity matching degree. When the interaction type identifier includes a traffic flow direction boundary identifier or a lower boundary identifier, determining the associated agent includes: Calculate the lane alignment between the lateral perception coordinates of the target agent's cooperative boundary and the lateral perception coordinates of the candidate agent's boundary. Candidate agents whose lane alignment is greater than the lane keeping threshold are identified as associated agents, and the expected paths of each agent are synchronized through V2X communication.

2. The method according to claim 1, characterized in that, The determination of spatiotemporal correlation includes: Extract the spatiotemporal feature point coordinates of the target agent cooperative boundary and the candidate agent cooperative boundary; Based on the projection of the spatiotemporal feature point coordinates into the vehicle-road cooperative perception coordinate system, it is determined whether the two belong to the vehicle queuing following mode or the cross-collision mode, where: If the spatiotemporal feature point of the left agent is located within the upstream time window of the feature point of the right agent, it is determined to be a positive following relationship; If there is a spatiotemporal overlap between the spatiotemporal feature points of the left agent and the feature points of the right agent, it is determined to be a reverse conflict relationship.

3. The method according to claim 1, characterized in that, Also includes: The lateral offset is calculated based on the lateral sensing coordinates of the candidate agent's traffic flow direction boundary and the target agent's boundary. If the lateral offset is less than the preset lane tolerance value, the longitudinal continuity matching degree calculation is triggered. If the lateral offset exceeds the lane tolerance value, it is marked as a cross-lane interference source and excluded from association.

4. The method according to claim 1, characterized in that, The calculation of lane alignment includes: In the vehicle-road cooperative perception coordinate system, the curvature features of key points at the boundary between the target intelligent agent and the candidate intelligent agents are extracted; When the curvature feature conforms to the lane centerline fitting model, calculate the standard deviation of the lateral coordinates relative to the lane centerline. The reciprocal of the standard deviation is used as the lane alignment degree, and it is corrected by incorporating lane keeping status data reported by L3-level intelligent vehicles.

5. The method according to any one of claims 1-4, characterized in that, Also includes: Statistically analyze the spatiotemporal boundary coordinates of boundary-associated agents with the same cooperative identifier, and extract the minimum / maximum lateral coordinates and minimum / maximum longitudinal coordinates; A multi-agent trajectory fusion canvas is constructed based on coordinate extreme values, and the local trajectories of related agents are spliced ​​into a complete trajectory according to spatiotemporal relationships. The continuity of the trajectory is verified by a convolutional neural network. If a trajectory break is detected, a cooperative identifier splitting instruction is generated and the identifier is reassigned.

6. The method according to claim 5, characterized in that, The method further includes: When an L4-level intelligent vehicle acts as a collaborative agent node, its high-precision positioning data is used as the baseline trajectory. Calculate the matching error between the trajectory of a normal vehicle and the baseline trajectory. If the error exceeds a preset threshold, trigger dynamic adjustment of the identifier. When the matching error is ≤5%, the original cooperative identifier is retained; When the matching error is 5%-15%, a new sub-identifier is added and associated with the baseline trajectory; When the matching error is greater than 15%, create an independent collaborative identifier and mark it as an abnormal traffic flow.

7. A vehicle-road cooperative edge computing device, characterized in that, include: The multi-source heterogeneous data access module is used to acquire perception data from roadside units (RSUs) and collaborative information from vehicle on-board units (OBUs). ABM agent management engine, used to perform agent interaction and identifier allocation in the method of any one of claims 1-4; The signal optimization decision module is used to generate phase control commands based on the density distribution matrix and collision point prediction results. The V2X communication module is used to send collaborative control strategies to traffic lights and intelligent vehicles.

Images