High-speed intelligent management and control method and system special for passenger car based on road safety

Abstract

The embodiment of the invention discloses a passenger car special-purpose high-speed intelligent management and control method and system based on road safety, and belongs to the technical field of intelligent traffic. According to the vehicle-road space-time diagram neural network processing in the embodiment of the invention, the vehicle-mounted sensing monitoring state vector is obtained, multi-source vehicle data can be integrated, and the vehicle state can be comprehensively reflected. The spatio-temporal trajectory trend mining accurately encodes behaviors of the vehicle under different spatio-temporal conditions, and considers a multi-feature dimension collaborative trend, so that the vehicle behavior description is more accurate. The inter-class attention strategy focuses on the feature dimension relation in spatial-temporal trajectory deduction, adapts to various traffic scenes, and improves the trajectory deduction accuracy. Traffic flow scheduling decision processing integrates various factors including vehicle tracks and road network conditions, and decision reasonability is improved. According to intelligent control of potential congestion and code updating based on dynamic adjustment, congestion can be effectively avoided, the operation efficiency and safety of the highway network can be improved, and the method can also cooperate with other traffic management systems to optimize overall traffic resource configuration.

Description

Technical Field

[0001] The embodiments of the present application belong to the field of intelligent transportation technology, and specifically relate to a method and system for intelligent highway control for passenger cars based on road safety. Background Art

[0002] In the field of modern highway network traffic management, with the increasing number of vehicles, the driving management of passenger cars on the highway network faces many challenges. Traditional traffic management methods often lack the ability to perceive and analyze the vehicle's detailed and comprehensive status. For example, it is difficult to integrate data from multiple sensors of the vehicle, and it is impossible to accurately grasp the vehicle's speed, spacing, acceleration and other status information. When analyzing the vehicle's behavior trajectory, the synergistic relationship of multiple feature dimensions is ignored, resulting in an inaccurate grasp of the vehicle's spatiotemporal trajectory trend. At the same time, when making traffic flow scheduling decisions, the complex relationship between different factors, such as vehicle behavior and road carrying capacity, is not fully considered. In addition, in the face of potential traffic congestion, there is a lack of effective intelligent control methods, and it is impossible to dynamically adjust the vehicle's driving status in a timely manner to avoid congestion. It is also difficult to work with other traffic management systems to optimize traffic resource allocation. Summary of the invention

[0003] The embodiments of the present application provide a method and system for intelligent high-speed control of passenger cars based on road safety, which can solve or partially solve the technical problems involved in the above-mentioned background technology.

[0004] The embodiment of the present application provides a method for intelligent high-speed control of passenger cars based on road safety, which is applied to an intelligent control system. The method comprises: obtaining a vehicle-mounted perception monitoring state description set of a cluster of vehicles to be controlled in a highway network area, wherein the vehicle-mounted perception monitoring state description set comprises a vehicle-mounted perception monitoring state vector of at least one feature dimension obtained after processing by a vehicle-road spatiotemporal graph neural network, and the cluster of vehicles to be controlled comprises at least two passenger cars; based on the feature dimension of the vehicle-mounted perception monitoring state vector, performing spatiotemporal trajectory trend mining on the vehicle-mounted perception monitoring state vector to obtain the spatiotemporal trajectory trend encoding corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension; calling the vehicle-mounted perception monitoring of each feature dimension through an inter-class attention strategy. The spatiotemporal trajectory trend code corresponding to the state vector performs spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain a spatiotemporal trajectory deduction code; in response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each feature dimension, traffic flow scheduling decision processing is performed on the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network; in response to the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network being a potential congestion label, the spatiotemporal trajectory deduction code is updated to obtain a spatiotemporal trajectory trend update code, and the vehicle cluster to be controlled is intelligently controlled based on the spatiotemporal trajectory trend update code.

[0005] An embodiment of the present application provides an intelligent management and control system, comprising at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the above method.

[0006] An embodiment of the present application provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the above method are implemented.

[0007] The vehicle-road spatiotemporal graph neural network processing in the embodiment of the present application obtains the on-board perception monitoring state vector, which can integrate multi-source vehicle data and comprehensively reflect the vehicle status. Spatiotemporal trajectory trend mining accurately encodes the behavior of the vehicle in different time and space, and considers the collaborative trend of multiple feature dimensions to make the vehicle behavior description more accurate. The inter-class attention strategy focuses on the relationship between feature dimensions in the spatiotemporal trajectory deduction, adapts to a variety of traffic scenarios, and improves the accuracy of trajectory deduction. The traffic flow scheduling decision-making process integrates multiple factors, including vehicle trajectory and road network conditions, to improve the rationality of decision-making. For intelligent control of potential congestion, based on dynamically adjusted coding updates, it can effectively avoid congestion, improve the efficiency and safety of highway network operation, and can also coordinate with other traffic management systems to optimize the overall traffic resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flow chart of a method for intelligent highway control for passenger cars based on road safety provided in an embodiment of the present application.

[0009] Figure 2 A schematic diagram of the structure of an intelligent management and control system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of them. Based on the embodiments in the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the embodiments of the present application.

[0011] The terms "first", "second", etc. in the embodiments of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are generally a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, in the embodiments of the present application, "and / or" represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0012] Figure 1 A special high-speed intelligent control method for passenger cars based on road safety is shown, which is applied to an intelligent control system. The method includes S100-S500.

[0013] S100: Obtain a vehicle-mounted perception monitoring status description set of a vehicle cluster to be controlled in a highway network area, wherein the vehicle-mounted perception monitoring status description set includes a vehicle-mounted perception monitoring status vector of at least one feature dimension obtained after processing by a vehicle-road spatiotemporal graph neural network, and the vehicle cluster to be controlled includes at least two small passenger cars.

[0014] S200: Based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, the vehicle-mounted perception monitoring state vector is mined for spatiotemporal trajectory trends to obtain the spatiotemporal trajectory trend coding corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension.

[0015] S300: Calling the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension through the inter-class attention strategy to perform spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction code.

[0016] S400: In response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, traffic flow scheduling decision processing is performed on the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network.

[0017] S500: In response to the traffic flow scheduling decision label of the on-board perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network being a potential congestion label, the spatiotemporal trajectory deduction code is updated to obtain the spatiotemporal trajectory trend update code, and the vehicle cluster to be controlled is intelligently controlled based on the spatiotemporal trajectory trend update code.

[0018] First, the intelligent management and control system executes S100 to obtain the vehicle-mounted perception monitoring state description set of the vehicle cluster to be controlled in the highway network area. The vehicle-mounted perception monitoring state description set in the embodiment of the present application is obtained after processing by the vehicle-road spatiotemporal graph neural network, and contains at least one vehicle-mounted perception monitoring state vector of a characteristic dimension. The vehicle-road spatiotemporal graph neural network is a neural network model specifically used to process the spatiotemporal relationship between vehicles and roads. For example, this vehicle-mounted perception monitoring state vector may have multiple characteristic dimensions such as speed, acceleration, vehicle spacing, lane position, etc. The speed characteristic dimension may be in the range of kilometers per hour, such as 0-200 kilometers per hour; the vehicle spacing may be in meters, ranging from 5 to 100 meters. These values ​​reflect the actual status of the vehicle in the highway network.

[0019] Then enter S200, based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, the spatiotemporal trajectory trend of the vehicle-mounted perception monitoring state vector is mined, so as to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension. Taking the characteristic dimension of speed as an example, the intelligent management and control system will analyze the speed change trend of the vehicle at different times and road positions (spaces). If the vehicle speed shows a trend of gradually decreasing within a section of road, this trend will be encoded into the corresponding spatiotemporal trajectory trend code. For example, in the dimension of vehicle spacing, if the vehicle spacing gradually decreases, it will also be accurately encoded to reflect this spatiotemporal trajectory trend. This spatiotemporal trajectory trend code can effectively capture the dynamic behavior characteristics of vehicles on the highway network.

[0020] Then, S300 is executed, and the space-time trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is called through the inter-class attention strategy to perform space-time trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road space-time graph neural network to obtain the space-time trajectory deduction code. The inter-class attention strategy is a strategy that can focus on the relationship between different categories (here refers to different feature dimension categories). For example, when considering the two feature dimensions of speed and vehicle spacing, the inter-class attention strategy will focus on how the change in speed affects the vehicle spacing, and the reaction of the change in vehicle spacing to speed. Using this strategy, space-time trajectory deduction is performed in the vehicle-road space-time graph neural network. For example, on a certain road section, due to the deceleration of the vehicle in front (the speed feature dimension changes), the vehicle behind also decelerates and increases the vehicle spacing (the vehicle spacing feature dimension changes). This complex change in space-time relationship will be encoded into the space-time trajectory deduction code through space-time trajectory deduction.

[0021] In S400, when the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each feature dimension is completed, the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is processed for traffic flow scheduling decisions. In this process, the intelligent management and control system will comprehensively consider the spatiotemporal trajectory deduction coding of each vehicle and the overall condition of the highway network. For example, if it is found that the speed of vehicles in a certain area is generally reduced and the distance between vehicles is gradually reduced, this may indicate that traffic congestion is about to occur. The intelligent management and control system will judge this situation and make corresponding traffic flow scheduling decisions based on pre-set rules and algorithms.

[0022] Finally, in S500, when the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction in the vehicle-road space-time graph neural network is a potential congestion label, the space-time trajectory deduction code is updated to obtain the space-time trajectory trend update code, and the controlled vehicle cluster is intelligently controlled based on the space-time trajectory trend update code. For example, when it is determined that there is potential congestion in a certain area, the intelligent control system may update the space-time trajectory trend code of the vehicle, consider changing the driving route of some vehicles or adjusting the speed limit. If the speed limit is adjusted, the original speed limit of 120 kilometers per hour may be reduced to 80 kilometers per hour to ease traffic pressure and avoid congestion. The intelligent control system can send new speed limit information to the vehicle through a variable information sign, or directly send instructions to the vehicle's control system through vehicle-road communication technology, thereby realizing intelligent control of the controlled vehicle cluster.

[0023] The above technical solution uses the vehicle-road spatiotemporal graph neural network to conduct in-depth analysis of the vehicle status, from the acquisition of the vehicle-mounted perception monitoring state vector, to the spatiotemporal trajectory trend mining, spatiotemporal trajectory deduction, to the traffic flow scheduling decision processing and finally the intelligent management and control, forming a complete and coherent system. Each step is closely linked, making full use of the various perception and monitoring information of the vehicle, effectively managing and controlling the passenger car clusters in the highway network, and improving the safety and operation efficiency of the highway network.

[0024] Combined with the above content, the above technical solution is further elaborated. In the process of S100 obtaining the vehicle-mounted perception monitoring state description set, the vehicle-road spatiotemporal graph neural network is a complex structure. It can integrate data from different vehicle sensors, such as the vehicle's speed sensor, distance sensor, gyroscope, etc. Take a real scene as an example. Near the entrance of a busy highway, multiple passenger cars are merging into the main road. The speed data provided by the speed sensor of each vehicle may be between 10-100 kilometers per hour, depending on the acceleration state of the vehicle. The distance sensor provides the distance information from the front and rear vehicles, which may be in the range of 10-50 meters. The vehicle-road spatiotemporal graph neural network fuses these multi-source data from different vehicles to form a vehicle-mounted perception monitoring state vector with multiple feature dimensions. This vector not only contains numerical information, but also contains the spatial position information of the vehicle in the highway network, such as which lane it is in, the distance from a specific road section sign (such as a toll station, service area, etc.), etc.

[0025] In the spatiotemporal trajectory trend mining of S200, the intelligent management and control system adopts complex data analysis methods. In addition to the analysis of the trend of a single feature dimension mentioned above, the collaborative trend between multiple feature dimensions will also be considered. For example, there may be a certain correlation between the acceleration and steering angle of the vehicle. When the vehicle accelerates, if the steering angle suddenly increases, this may indicate that the vehicle is overtaking or avoiding. The intelligent management and control system will mine this multi-dimensional collaborative trend and encode it into the spatiotemporal trajectory trend coding. This encoding method can more comprehensively describe the behavior pattern of the vehicle. For example, for an on-board perception monitoring state vector with four feature dimensions of speed, acceleration, vehicle spacing, and steering angle, the spatiotemporal trajectory trend coding may be a multi-dimensional coding structure, with each dimension corresponding to the trend information of a feature dimension.

[0026] In S300, the inter-class attention strategy plays a key role in the spatiotemporal trajectory deduction. The inter-class attention strategy is similar to an intelligent focusing mechanism that can identify important relationships between different feature dimensions. Take a case with three feature dimensions: vehicle speed, vehicle weight, and road friction coefficient. During normal driving, the vehicle speed and vehicle weight may have little effect on the road friction coefficient, but when braking or turning, the vehicle speed and vehicle weight will have a greater impact on the road friction coefficient, thereby affecting the driving safety of the vehicle. The inter-class attention strategy can accurately capture the changes in this relationship in the spatiotemporal trajectory deduction, so that the vehicle-road spatiotemporal graph neural network can generate more accurate spatiotemporal trajectory deduction coding. This coding can reflect the possible changes in the actual driving trajectory of the vehicle under complex road conditions and multiple factors.

[0027] When entering the traffic flow scheduling decision processing stage of S400, the intelligent management and control system will consider more factors. In addition to the spatiotemporal trajectory deduction coding of the vehicle, factors such as the topological structure of the highway network and the carrying capacity of the road will also be considered. For example, in a highway section with multiple lanes and multiple ramps, if the spatiotemporal trajectory deduction coding of the vehicle in a lane shows that the vehicle speed is reduced and the distance between vehicles is reduced, and the traffic flow in the lane is close to the carrying capacity of the road, the intelligent management and control system may make a decision to guide some vehicles to switch to other lanes or control the vehicle merging speed of the ramp. This decision-making process is based on the analysis results of a large amount of historical data and real-time data. The intelligent management and control system may establish a decision model, which is pre-trained according to different traffic flow patterns, road conditions and other factors, so that traffic flow scheduling decisions can be made quickly and accurately in actual situations.

[0028] In S500, when it is determined that there is a potential congestion tag, the update of the spatiotemporal trajectory deduction code is a delicate process. This update is not just a simple change of a certain value, but an adjustment based on the actual situation of the vehicle and the overall needs of the highway network. For example, if it is found that the potential congestion is caused by a vehicle failure in a certain lane, the intelligent control system may update the spatiotemporal trajectory trend update code of the vehicles around the affected vehicle to guide these vehicles to bypass the faulty vehicle. In this process, the intelligent control system may calculate the best detour path for each vehicle based on factors such as the vehicle's speed, the distance from the faulty vehicle, and the traffic conditions of the adjacent lanes, and encode this information into the spatiotemporal trajectory trend update code. Then, based on this update code, the controlled vehicle cluster is intelligently controlled in a variety of ways. For example, the duration of the signal light is adjusted through the intelligent traffic signal system to guide the vehicle to travel along the planned path; or instructions are sent directly to the vehicle through the vehicle-road communication system to inform the vehicle to adjust the speed and driving direction.

[0029] It can be seen that through a series of closely connected steps, the vehicle-road spatiotemporal graph neural network and various strategies and algorithms can be used to comprehensively and meticulously manage and control the clusters of passenger cars in the highway network, thereby improving the operating efficiency and safety of the highway network and effectively dealing with potential traffic congestion and other problems.

[0030] In another application scenario, each link of the above technical solution can be further elaborated in detail. In terms of S100 obtaining the vehicle-mounted perception monitoring state description set, the vehicle-road spatiotemporal graph neural network has high efficiency and accuracy in processing data. It is capable of processing large-scale vehicle perception data. For example, in a large highway network area, there may be perception data of hundreds or thousands of passenger cars that need to be processed. The vehicle-road spatiotemporal graph neural network can be partitioned according to the geographical location of the vehicle, dividing the entire highway network into multiple sub-areas, and the vehicle perception data of each sub-area is processed separately and then aggregated. This processing method can increase the speed of data processing and reduce the delay of data transmission and processing. At the same time, there are strict accuracy requirements for each feature dimension in the vehicle-mounted perception monitoring state vector. Taking vehicle speed as an example, the accuracy may reach ±1 km / h, which can accurately reflect the actual driving speed of the vehicle and provide a reliable data basis for subsequent analysis.

[0031] In the process of mining the spatiotemporal trajectory trend of S200, the data mining algorithm used by the intelligent control system is specially designed for the behavior of vehicles on the highway network. This algorithm can identify the trend changes of vehicles at different time scales. For example, in a short time scale (such as 1-5 minutes), the vehicle may experience speed fluctuations due to the temporary braking of the vehicle in front; in a long time scale (such as 10-30 minutes), the vehicle may adjust the speed and driving route due to the overall change of traffic flow. The spatiotemporal trajectory trend coding can accurately record the trend information at these different time scales. For different feature dimensions, the structure and representation of the spatiotemporal trajectory trend coding are also different. Taking the feature dimension of vehicle spacing as an example, its spatiotemporal trajectory trend coding may adopt a coding method based on the rate of change of distance, and express its spatiotemporal trajectory trend by calculating the rate of change of vehicle spacing at different time points.

[0032] In S300, the implementation of the inter-class attention strategy in the spatiotemporal trajectory deduction is based on a complex weight allocation mechanism. This mechanism can assign different weights according to the importance of different feature dimensions in specific traffic scenarios. For example, when driving on rainy days, the weight of the feature dimension of the road friction coefficient will be relatively high because it has an important impact on the driving safety and spatiotemporal trajectory of the vehicle. When driving on normal dry roads, the weights of feature dimensions such as vehicle speed and vehicle spacing may be relatively high. Through this weight allocation mechanism, the inter-class attention strategy can more accurately call the spatiotemporal trajectory trend code to perform spatiotemporal trajectory deduction on the on-board perception monitoring state vector, and the obtained spatiotemporal trajectory deduction code can more realistically reflect the changing trend of the vehicle's driving trajectory in actual traffic scenarios.

[0033] In the traffic flow scheduling decision-making process of S400, the decision model adopted by the intelligent control system is a complex model that integrates multiple factors. This model not only takes into account the vehicle behavior information reflected by the vehicle's spatiotemporal trajectory deduction coding, but also takes into account the macro information of the highway network, such as the traffic flow distribution of the entire highway network, the traffic flow rules during peak hours and off-peak hours, etc. For example, during peak hours, traffic flow is mainly concentrated on the highway sections around the city. The intelligent control system will focus on monitoring and scheduling vehicles on these sections based on this rule. At the same time, the decision model will also consider the type and grade of the road, and different strategies will be adopted for traffic flow scheduling of main roads and ramps. Traffic flow scheduling on main roads pays more attention to overall stability and smoothness, while traffic flow scheduling on ramps pays more attention to connection and coordination with main roads.

[0034] In S500, when the space-time trajectory deduction code is updated to obtain the space-time trajectory trend update code, the update process is based on a dynamic adjustment mechanism. This mechanism will continuously adjust the code content according to the real-time changes in the highway network. For example, when it is found that the traffic conditions in the potential congestion area have eased, the space-time trajectory trend update code will promptly adjust the vehicle guidance strategy, such as restoring the original speed limit or adjusting the vehicle's driving route, so that the vehicle can travel more efficiently on the highway network. At the same time, when the controlled vehicle cluster is intelligently controlled based on the space-time trajectory trend update code, the intelligent control system will use a variety of communication methods to ensure the accurate transmission of instructions. In addition to the variable information signs and vehicle-road communication technologies mentioned above, auxiliary means such as satellite communications may also be used to ensure that vehicles can receive correct control instructions in some areas with poor signal coverage.

[0035] In summary, the technical solution involved in this invention application has constructed a complete and efficient high-speed intelligent control system for passenger cars through the careful design and processing of each link, which can effectively cope with various traffic conditions in the highway network and improve traffic safety and operation efficiency.

[0036] In addition, some technical details of the above technical solution are introduced as follows: In S100, the vehicle-road spatiotemporal graph neural network will pre-process the vehicle perception data when constructing the vehicle-mounted perception monitoring state vector. For example, for vehicle speed data, filtering may be performed to remove some outliers caused by sensor errors or instantaneous interference. For example, due to the influence of factors such as road bumps, the vehicle speed sensor occasionally has some extremely high or extremely low values ​​that are beyond the normal range. These outliers can be excluded through filtering, making the obtained speed data more accurate and reliable. For vehicle spacing data, data correction may be performed because the distance sensors of different vehicles may have certain measurement errors. Through cross-validation with adjacent vehicles and other reference data (such as the location information of road signs), the vehicle spacing data can be corrected to improve its accuracy.

[0037] In the spatiotemporal trajectory trend mining of S200, in order to more accurately mine the spatiotemporal trajectory trend, the intelligent management and control system will adopt a hierarchical mining method. First, a separate trend mining is performed on each feature dimension, which is the mining of the basic layer. For example, for the acceleration feature dimension of the vehicle, its acceleration change trend in different sections and time periods will be analyzed, whether it is continuous acceleration, continuous deceleration, or alternating acceleration and deceleration. Then in the second layer of mining, the combination trend between different feature dimensions will be analyzed. For example, the combination trend between the acceleration of the vehicle and the distance between vehicles, when the vehicle accelerates, how the distance between vehicles changes, and whether this change complies with normal traffic rules and safety requirements. Through this hierarchical mining method, the spatiotemporal trajectory trend encoding can more comprehensively and meticulously reflect the behavior pattern of the vehicle.

[0038] In S300, the weight allocation mechanism of the inter-class attention strategy in the spatiotemporal trajectory deduction is dynamic. It will be adjusted according to the real-time traffic conditions in the highway network. For example, when a traffic accident occurs on the highway, the weights of the feature dimensions related to the accident (such as the speed of vehicles near the accident site, the distance between vehicles, etc.) will increase rapidly. This is because these feature dimensions are crucial to the spatiotemporal trajectory deduction of vehicles in accident scenarios. At the same time, the inter-class attention strategy will also take into account the differences between different vehicle types. Although the embodiments of the present application are mainly aimed at passenger cars, there may be a small number of other types of vehicles (such as emergency rescue vehicles) in the actual highway network. For these special vehicles, the inter-class attention strategy will give special weight processing to ensure that their impact on passenger car traffic can be correctly considered in the spatiotemporal trajectory deduction.

[0039] In the traffic flow scheduling decision-making process of S400, the intelligent management and control system will continuously update the parameters in the decision model. With the continuous development of the highway network and the changes in traffic flow, the original decision model parameters may no longer be applicable. For example, with the opening of new highway sections or the improvement of traffic facilities, the distribution pattern of traffic flow will change. The intelligent management and control system will regularly optimize the parameters in the decision model based on the new traffic flow data and road condition data. At the same time, in the decision-making process, the intelligent management and control system will take into account the driver's behavior and reaction ability. Different drivers have different reaction speeds and execution capabilities to traffic control instructions. The intelligent management and control system will appropriately consider this difference when making decisions to ensure the effectiveness and operability of the decision.

[0040] In S500, the update of the spatiotemporal trajectory trend update code is not only based on the current traffic conditions, but also takes into account future traffic forecasts. The intelligent control system will use historical traffic data and real-time traffic data to establish a traffic flow prediction model. For example, by analyzing the traffic flow data of the same time period and the same highway section in the past few years, combined with the current traffic flow trend, the traffic flow changes in the future (such as 10-30 minutes) are predicted. If the traffic conditions in the potential congestion area are predicted to deteriorate further, the spatiotemporal trajectory trend update code will take more stringent control measures, such as further reducing the speed limit or expanding the vehicle detour range. When performing intelligent control based on the spatiotemporal trajectory trend update code, the intelligent control system will coordinate with other traffic management systems (such as urban traffic management systems, parking management systems around highways, etc.). For example, if passenger cars in the highway network need to be diverted to surrounding parking lots, the intelligent control system will communicate with the parking lot management system to ensure that there are enough parking spaces in the parking lot to receive these vehicles.

[0041] In an exemplary application scenario, for example, the vehicle-road spatiotemporal graph neural network is constructed using the idea of ​​a graph convolutional neural network (GCN). In this network, nodes represent vehicles and key road facilities (such as toll booths, service areas, etc.) in the highway network, and edges represent the relationship between vehicles and vehicles, and between vehicles and road facilities (such as the distance relationship between vehicles, the accessibility relationship between vehicles and service areas, etc.). In the S100 acquisition of the vehicle perception monitoring state description set stage, the sensor data of each vehicle (such as speed sensor, distance sensor, gyroscope, etc.) is first preprocessed. For example, after the vehicle speed data is filtered to remove outliers and the vehicle spacing data is corrected to improve accuracy, these data constitute the initial feature vector of each vehicle node. These initial feature vectors contain information such as speed, acceleration, vehicle spacing, lane position, and distance relative to road facilities. When these node feature vectors are propagated in the network, the graph convolution layer aggregates the features of adjacent nodes according to the relationship represented by the edge. In the S200 spatiotemporal trajectory trend mining stage, since the network has learned the relationship between vehicles and between vehicles and road facilities, it can mine spatiotemporal trajectory trends by analyzing the changes in node feature vectors at different times. For example, by observing the changes in vehicle speed features in the graph over multiple time steps, as well as the changes in the correlation between vehicle spacing features and other vehicle node features, the spatiotemporal trajectory trends of vehicles can be identified. This mining is not just a simple numerical analysis, but is also based on the network's understanding of the entire highway network structure and vehicle relationships. For example, when the speed of a vehicle decreases, the network can consider whether the spacing between its adjacent vehicles also changes accordingly, and the propagation mode of this change in the entire highway network, thereby constructing a spatiotemporal trajectory trend encoding for each feature dimension. Entering the S300 spatiotemporal trajectory deduction stage, the inter-class attention strategy can be implemented by adjusting the weights corresponding to different feature dimensions in the graph convolution layer. In this example, for example, the inter-class attention mechanism is combined with the graph convolution layer through an additional weight matrix. In normal driving scenarios, feature dimensions such as speed and vehicle spacing have higher weights, which means that when performing spatiotemporal trajectory deduction, the network pays more attention to the propagation and influence of these features between nodes. When encountering special circumstances (such as bad weather), the network increases the weight of the characteristic dimension of the road friction coefficient according to pre-set rules or patterns learned from the data, so that the spatiotemporal trajectory deduction encoding can accurately reflect the trend of vehicle driving trajectory changes in this situation. This weight adjustment is dynamic and is adaptively adjusted according to real-time traffic conditions and different vehicle types (although it is mainly for passenger cars, the impact of special vehicles is taken into account). At the S400 traffic flow scheduling decision processing stage, the output of the entire vehicle-road spatiotemporal graph neural network (including the results of the node feature vector after spatiotemporal trajectory deduction) is input into a decision module.This decision module can be regarded as an additional layer of the network, which comprehensively considers the vehicle spatiotemporal trajectory deduction code output by the network, the topological structure of the highway network (represented by the nodes and edges of the graph), and the carrying capacity of the road. For example, based on the speed and spacing relationship of the vehicles between nodes, the lane structure of the entire road network, the traffic flow distribution and other information, it is judged whether there is a risk of traffic congestion or an imbalance in traffic flow. This decision process is based on the network's learning results of a large amount of historical data and real-time data, and can continuously update the parameters in the decision model as the highway network develops and the traffic flow changes, while considering the driver's behavior and reaction ability. Finally, in the S500 stage, when the decision module determines that the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, the spatiotemporal trajectory deduction code is updated. This update is based on the results of the traffic flow prediction model, which uses the historical traffic data learned by the network and the current real-time traffic data. For example, if it is predicted that traffic congestion in a certain area will worsen, the network will update the spatiotemporal trajectory trend code and adjust the vehicle's driving route or speed limit. When updating the coding based on the trend of space-time trajectory for intelligent control, the vehicle-road space-time graph neural network can also work in conjunction with other traffic management systems. For example, it communicates with the parking lot management system to coordinate the diversion of passenger cars to ensure the efficient operation of the entire highway network and surrounding traffic systems. Throughout the process, the various links of the vehicle-road space-time graph neural network are closely connected to form a complete intelligent control system.

[0042] In an exemplary technical solution, the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension is called through the inter-class attention strategy to perform spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction code, including: determining the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension based on the characteristic dimension of the vehicle-mounted perception monitoring state vector; calling the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension through the inter-class attention strategy to enhance the confidence of the corresponding spatiotemporal trajectory deduction module to obtain the spatiotemporal trajectory deduction enhancement module; calling the spatiotemporal trajectory deduction enhancement module corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension through the inter-class attention strategy to perform spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction code.

[0043] Furthermore, the confidence of the corresponding space-time trajectory deduction module is enhanced by calling the space-time trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension through the inter-class attention strategy to obtain the space-time trajectory deduction enhancement module, including: extracting the vehicle behavior element vector corresponding to the vehicle cluster to be controlled by calling the space-time trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension through the inter-class attention strategy; determining the deduction logic indication vector corresponding to each space-time trajectory deduction module through the inter-class attention strategy; adding the vehicle behavior element vector corresponding to the vehicle cluster to be controlled to the deduction logic indication vector of the corresponding space-time trajectory deduction module through the inter-class attention strategy to obtain the space-time trajectory deduction enhancement module.

[0044] Furthermore, the space-time trajectory deduction module corresponding to the vehicle-borne perception monitoring state vector of each characteristic dimension is determined based on the characteristic dimension of the vehicle-borne perception monitoring state vector, including: when the characteristic dimension of the vehicle-borne perception monitoring state vector is a dynamic characteristic dimension, the space-time trajectory deduction module corresponding to the vehicle-borne perception monitoring state vector of the dynamic characteristic dimension is determined to be a space-time trajectory deduction module based on environmental sensor monitoring; when the characteristic dimension of the vehicle-borne perception monitoring state vector is an event characteristic dimension, the space-time trajectory deduction module corresponding to the vehicle-borne perception monitoring state vector of the event characteristic dimension is determined to be a space-time trajectory deduction module based on congestion prediction and judgment.

[0045] In the technical solution of the embodiment of the present application, for the process of performing spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction coding by calling the spatiotemporal trajectory trend coding corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension through the inter-class attention strategy, firstly, the spatiotemporal trajectory deduction module corresponding to each feature dimension is determined based on the feature dimension of the vehicle-mounted perception monitoring state vector. Here, the division is made according to the different properties of the feature dimensions.

[0046] When the characteristic dimension of the vehicle-mounted perception monitoring state vector is a dynamic characteristic dimension, such as the vehicle's speed, acceleration, steering angle, etc., which are constantly changing during driving, the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector with dynamic characteristic dimensions is determined to be a spatiotemporal trajectory deduction module based on environmental sensor monitoring.

[0047] Taking the dynamic feature dimension of speed as an example, its value range may be between 0-200 km / h. For example, at a certain moment, the vehicle speed is 80 km / h, and the speed feature vector can be expressed as [80, 0], where 0 indicates that the rate of change of speed is 0 (temporarily in a uniform state). The value range of acceleration may be between -5 m / s² and 5 m / s². If the acceleration is 1 m / s² at a certain moment, its feature vector can be expressed as [1, 0], where 0 indicates that the change trend of acceleration is stable. The value range of steering angle is -180° to 180°. For example, if the current steering angle is 10°, the feature vector is [10, 0], indicating that the steering angle is 10° in the positive direction and there is no turning change trend. In this spatiotemporal trajectory deduction module based on environmental sensor monitoring, the spatiotemporal trajectory related to vehicle speed, acceleration and steering angle will be deduced based on the data of environmental sensor monitoring. The data of environmental sensor monitoring can include information such as the friction of the road surface, wind direction and speed, and the relative position of surrounding vehicles. For example, if the road friction is small (such as a slippery road), for example, the friction coefficient is 0.2 (the friction coefficient of a normal dry road may be around 0.8), the vehicle's acceleration will become smaller and the speed increase will slow down. When the wind direction is headwind and the wind speed is 5 meters per second, this environmental factor will affect the vehicle's speed and acceleration, and this relationship will be reflected in the spatiotemporal trajectory deduction module based on environmental sensor monitoring.

[0048] When the characteristic dimension of the vehicle-mounted perception monitoring state vector is an event characteristic dimension, such as a dimension related to events such as traffic accidents, road construction, special weather (heavy rain, heavy snow, etc.), the space-time trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the event characteristic dimension is determined to be a space-time trajectory deduction module based on congestion prediction and judgment.

[0049] Taking the event feature dimension of road construction as an example, if the construction section occupies part of the lane, for example, the construction section occupies one lane, and the section normally has four lanes in both directions. The current traffic flow is 1,000 vehicles per hour, and the feature vector of the traffic flow can be expressed as [1000, 1], where 1 indicates that the current traffic flow is in a stable growth trend. The location information of the construction section can also be encoded as a feature vector. For example, if the construction section is located 20 kilometers from the starting point, it can be expressed as [20, 0], where 0 indicates that the position is fixed. In this spatiotemporal trajectory deduction module based on congestion prediction and discrimination, the location of the construction section, the scale of construction, the current traffic flow and other factors will be comprehensively considered to deduce the spatiotemporal trajectory of the vehicle. If the construction section is located on a main road with a large traffic flow, and the construction scale is large and occupies more lanes, then the spatiotemporal trajectory deduction for passenger cars may be to guide the vehicle to change lanes in advance or leave the highway from the nearest exit to avoid the congested section.

[0050] Then, the space-time trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is called through the inter-class attention strategy to enhance the confidence of the corresponding space-time trajectory deduction module, and the space-time trajectory deduction enhancement module is obtained. This process also includes multiple sub-steps.

[0051] First, the inter-class attention strategy is used to call the spatiotemporal trajectory trend encoding corresponding to the on-board perception monitoring state vector of each feature dimension to extract the vehicle behavior element vector corresponding to the vehicle cluster to be controlled.

[0052] Taking the feature dimension of vehicle spacing as an example, if the vehicle spacing gradually decreases from 30 meters to 20 meters in a certain period of time, the value range of the vehicle spacing is, for example, 5-100 meters, and this trend in the spatiotemporal trajectory trend coding can be expressed as [30, 20, -1], where -1 indicates that the spacing is decreasing. Through the inter-class attention strategy, the vehicle behavior element vector of the vehicle spacing reduction can be extracted from this coding, which can be expressed as [-1, 0], and 0 indicates that the trend of the spacing reduction is stable. This vector can reflect a behavior state of the vehicle during driving, for example, it may indicate that the vehicle is approaching the vehicle in front, or it is a normal spacing adjustment during the following process. Another example is the braking behavior of the vehicle. For example, the travel range of the brake pedal is 0-10 cm, and the brake pedal travel is 3 cm at a certain moment. The spatiotemporal trajectory trend coding of its braking behavior may be [3, 0], 0 indicates that there is no change trend in the brake pedal travel, and the vehicle behavior element vector extracted from this coding is [0, 0], indicating that the current braking behavior is stable.

[0053] Then, the deduction logic indicator vector corresponding to each spatiotemporal trajectory deduction module is determined through the inter-class attention strategy.

[0054] For the spatiotemporal trajectory deduction module based on environmental sensor monitoring, its deduction logic indicator vector may include the setting of the weight of environmental factors, the evaluation of the credibility of different environmental sensor data, etc. For example, when the weather conditions are good, the data credibility of the road sensor may be set to 0.8, the data credibility of the wind direction and wind speed sensor to 0.6, and the data credibility of the surrounding vehicle distance sensor to 0.7. The corresponding deduction logic indicator vector can be expressed as [0.8, 0.6, 0.7]. For different environmental factors, for example, the weight of the road friction on the speed is 0.4, the weight of the wind direction and wind speed on the speed is 0.3, and the weight of the relative position of the surrounding vehicles on the speed is 0.3, then the partial vector representing the weight of the environmental factors is [0.4, 0.3, 0.3]. In summary, the deduction logic indicator vector of this spatiotemporal trajectory deduction module based on environmental sensor monitoring is [[0.8, 0.6, 0.7], [0.4, 0.3, 0.3]].

[0055] For the spatiotemporal trajectory deduction module based on congestion prediction and discrimination, its deduction logic indicator vector may include the evaluation factors of the impact of different events on the degree of congestion, the estimation of the congestion propagation range of different road sections, etc. For example, for road construction events near transportation hubs, the evaluation factor of the congestion degree is 0.8 (indicating a large impact), the estimated congestion propagation range is 5 kilometers, and the characteristic vector of the congestion propagation range is [5, 0], where 0 indicates a fixed range. For construction events in remote sections of the road, the congestion degree assessment factor may be 0.3, the congestion propagation range is estimated to be 2 kilometers, and the characteristic vector is [2, 0]. Then the deduction logic indicator vector of the spatiotemporal trajectory deduction module based on congestion prediction and discrimination corresponding to the construction events near transportation hubs is [[0.8, 0.3], [5, 0]].

[0056] Finally, the vehicle behavior element vector corresponding to the vehicle cluster to be controlled is added to the deduction logic indication vector of the corresponding space-time trajectory deduction module through the inter-class attention strategy to obtain the space-time trajectory deduction enhancement module.

[0057] For example, the vehicle behavior element vector [-1, 0] of the reduction in vehicle spacing is added to the deduction logic indicator vector [[0.8, 0.6, 0.7], [0.4, 0.3, 0.3]] of the spatiotemporal trajectory deduction module based on environmental sensor monitoring. For example, at this time, the environmental sensor detects a slight slope change (uphill) in the road ahead, and the characteristic vector of the slope change is [5°, 0], where 0 indicates no slope change trend. Then in this enhanced module, the spatiotemporal trajectory of the vehicle will be more accurately deduced based on factors such as the reduction in vehicle spacing and the slope change in the road ahead. It may be concluded that the vehicle will appropriately reduce its speed to maintain a safe distance due to the reduction in spacing and the uphill road ahead. For example, the original speed of the vehicle is 80 km / h, and according to the deduction of the enhanced module, the speed may be adjusted to 60 km / h.

[0058] Finally, the space-time trajectory deduction enhancement module corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is called through the inter-class attention strategy, and the space-time trajectory deduction of the vehicle-mounted perception monitoring state vector is performed in the vehicle-road space-time graph neural network to obtain the space-time trajectory deduction encoding.

[0059] In this process, the information of each reinforcement module is integrated to comprehensively deduce the spatiotemporal trajectory of the vehicle in the entire highway network. For example, for a situation with multiple feature dimensions such as speed, vehicle spacing, and the impact of road construction events, each corresponding reinforcement module will integrate its own information.

[0060] If the reinforcement module corresponding to the speed feature dimension shows that the vehicle is accelerating, for example, the acceleration is 2 meters per second squared, the speed feature vector is [80, 2], the reinforcement module corresponding to the vehicle spacing feature dimension shows that the vehicle spacing is stable at 25 meters, and the vehicle spacing feature vector is [25, 0], and the reinforcement module corresponding to the impact of road construction events shows that it is about to enter the construction section and needs to slow down, the feature vector of the impact of construction events is [1, -1], where 1 indicates the existence of a construction event and -1 indicates the need to slow down. Then, through the spatiotemporal trajectory deduction of the vehicle-road spatiotemporal graph neural network, a spatiotemporal trajectory deduction code that comprehensively considers these factors will be obtained, which can accurately reflect the spatiotemporal trajectory direction of the vehicle in complex situations.

[0061] It can be seen that by classifying the characteristic dimensions of the vehicle-mounted perception monitoring state vector and determining the spatiotemporal trajectory deduction modules corresponding to different types of dimensions, it is possible to process vehicle driving related information of different natures in a more targeted manner. The inter-class attention strategy is used to enhance confidence, and the vehicle behavior element vector is combined with the deduction logic indication vector to obtain the enhanced module, which further improves the accuracy and reliability of each module. Finally, the spatiotemporal trajectory deduction coding obtained by the spatiotemporal trajectory deduction through these enhanced modules can more comprehensively and accurately reflect the spatiotemporal trajectory of the vehicle in the highway network, thereby providing a more accurate basis for subsequent traffic flow scheduling decision-making and processing operations, which helps to improve the safety and operation efficiency of the highway network and realize more effective intelligent control of passenger car clusters.

[0062] In an optional embodiment, based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, the vehicle-mounted perception monitoring state vector is subjected to spatiotemporal trajectory trend mining to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension, including: based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, the original spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension is generated; behavior recognition is performed on the vehicle-mounted perception monitoring state vector of each characteristic dimension to obtain the vehicle behavior element vector corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension; the vehicle behavior element vector corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension is added to the antecedent spatiotemporal data chain corresponding to the corresponding original spatiotemporal trajectory trend code to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension.

[0063] In the technical solution of the embodiment of the present application, for the feature dimension based on the vehicle-mounted perception monitoring state vector, the vehicle-mounted perception monitoring state vector is subjected to spatiotemporal trajectory trend mining to obtain the spatiotemporal trajectory trend coding corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension. First, based on the feature dimension of the vehicle-mounted perception monitoring state vector, the original spatiotemporal trajectory trend coding corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is generated. Taking the feature dimension of vehicle speed as an example, the numerical range of vehicle speed may be between 0-200 km / h. If the vehicle speed gradually increases from 50 km / h to 80 km / h in a certain period of time, this speed change information will be encoded into the original spatiotemporal trajectory trend coding. The speed data is collected at a certain time interval, for example, this time interval is 1 second, then the speed change sequence within these seconds constitutes a part of the original spatiotemporal trajectory trend coding. For another example, the feature dimension of vehicle spacing, whose value range is, for example, 5-100 meters, if the vehicle spacing fluctuates from 30 meters to 25 meters during a period of driving, the change of this spacing and the time sequence of the change will be encoded into the original spatiotemporal trajectory trend coding. The acceleration of the vehicle may range from -5 m / s² to 5 m / s², and the change in acceleration will also be recorded in the original spatiotemporal trajectory trend code. For example, when a vehicle accelerates from a standstill, the acceleration gradually increases from 0 to 2 m / s², and the acceleration change value and corresponding time point in this process will be encoded.

[0064] Next, the vehicle-mounted sensing monitoring state vector of each feature dimension is subjected to behavior recognition to obtain the vehicle behavior element vector corresponding to the vehicle-mounted sensing monitoring state vector of each feature dimension. For the speed feature dimension, when the vehicle speed continues to increase and the acceleration remains a stable positive value, the vehicle behavior element vector that the vehicle is accelerating can be obtained through behavior recognition. This vector may be represented as [1, 0], where 1 represents the acceleration behavior and 0 represents the stable trend of this acceleration behavior. For the vehicle spacing, if the vehicle spacing gradually decreases, the vehicle behavior element vector may be [-1, 0], where -1 represents the behavior of the spacing decreasing and 0 represents the stable trend of the spacing decreasing. For the steering angle of the vehicle, for example, the steering angle ranges from -180° to 180°. When the vehicle steering angle gradually increases from 0° to 10°, the vehicle behavior element vector obtained through behavior recognition may be [10, 1], where 10 represents the value of the steering angle and 1 represents the steering angle is in an increasing trend.

[0065] Then, the vehicle behavior element vector corresponding to the vehicle perception monitoring state vector of each feature dimension is added to the antecedent spatiotemporal data chain corresponding to the corresponding original spatiotemporal trajectory trend code, and the spatiotemporal trajectory trend code corresponding to the vehicle perception monitoring state vector of each feature dimension is obtained. Taking vehicle speed as an example, the original spatiotemporal trajectory trend code records the change of speed value over time, such as [50, 52, 55, 58, 60, 62, 65, 70, 75, 80] (where each value represents the speed at a different time point), and the corresponding vehicle behavior element vector is [1, 0]. The antecedent spatiotemporal data chain contains not only these speed values, but also other time-related information related to speed, such as timestamps. After adding the vehicle behavior element vector, the spatiotemporal trajectory trend code becomes a code that contains the behavior element vector and richer spatiotemporal information, such as [[50, 0], [52, 0], [55, 0], [58, 0], [60, 0], [62, 0], [65, 0], [70, 0], [75, 0], [80, 0], [1, 0]], where the timestamp information corresponding to each speed value is 0 (for example only). For the vehicle spacing, the original spatiotemporal trajectory trend code is [30, 29, 28, 27, 26, 25], the vehicle behavior element vector is [-1, 0], and the spatiotemporal trajectory trend code after adding may be [[30, 0], [29, 0], [28, 0], [27, 0], [26, 0], [25, 0], [-1, 0]]. The spatiotemporal trajectory trend encoding obtained in this way can more comprehensively reflect the behavior status of the vehicle in each feature dimension and the changing trend of the spatiotemporal trajectory.

[0066] This method of adding the vehicle behavior element vector to the antecedent spatiotemporal data chain of the original spatiotemporal trajectory trend coding makes the spatiotemporal trajectory trend coding not just a simple record of the vehicle state value, but a more meaningful coding that integrates the behavior recognition results. For example, for the braking behavior of the vehicle, if the original spatiotemporal trajectory trend coding records the numerical changes of the brake pedal stroke, such as [0, 1, 2, 3, 4] (for example, the value range of the brake pedal stroke is 0-10 cm, which means that the brake pedal is gradually pressed down over time), the vehicle behavior element vector is [1, 1] (1 indicates that the braking behavior occurs, and 1 indicates that the degree of braking is deepening), and the added spatiotemporal trajectory trend coding is [[0, 0], [1, 0], [2, 0], [3, 0], [4, 0], [1, 1]], so that the vehicle's braking behavior and its relationship with the numerical changes of the brake pedal stroke can be more intuitively interpreted from the coding.

[0067] From a comprehensive perspective of multiple feature dimensions, each feature dimension has its corresponding spatiotemporal trajectory trend code. For example, the spatiotemporal trajectory trend codes of feature dimensions such as speed, vehicle spacing, and steering angle are combined together to fully describe the overall driving status of the vehicle within a certain period of time. If the spatiotemporal trajectory trend code of speed shows that the vehicle is accelerating, the spatiotemporal trajectory trend code of vehicle spacing shows that the spacing is stable, and the spatiotemporal trajectory trend code of steering angle shows that the steering angle has not changed, then it can be comprehensively judged that the vehicle is in a normal state of accelerating and maintaining a stable lane. This multi-dimensional spatiotemporal trajectory trend code provides a comprehensive and accurate data basis for subsequent operations, such as spatiotemporal trajectory deduction.

[0068] During the driving process of a vehicle, various feature dimensions are interrelated. For example, changes in vehicle speed will affect the distance between vehicles. When a vehicle accelerates, if appropriate operations are not performed, the distance between vehicles may gradually decrease. This correlation is also reflected in the spatiotemporal trajectory trend coding. By comprehensively analyzing the spatiotemporal trajectory trend coding of different feature dimensions, deeper vehicle driving patterns and trends can be discovered. For example, over a period of time, the vehicle speed continues to increase, the distance between vehicles gradually decreases, and the steering angle fluctuates slightly. This combination may indicate that the vehicle is performing an overtaking operation. Through a detailed analysis of the spatiotemporal trajectory trend coding of each feature dimension, this complex vehicle behavior can be accurately identified.

[0069] In this way, by generating the original spatiotemporal trajectory trend code, performing behavior recognition to obtain the vehicle behavior element vector, and adding it to the pre- and post-sequential spatiotemporal data chain to obtain the spatiotemporal trajectory trend code, this method can comprehensively describe the spatiotemporal trajectory changes of the vehicle in each feature dimension from multiple levels and angles. The spatiotemporal trajectory trend code of each feature dimension not only includes the change of state value, but also integrates the result of behavior recognition, making the code more informative. The combination of spatiotemporal trajectory trend codes of multiple feature dimensions can accurately reflect the overall driving state and complex behavior patterns of the vehicle, providing an accurate and comprehensive data foundation for subsequent operations based on the vehicle-road spatiotemporal graph neural network, such as spatiotemporal trajectory deduction, traffic flow scheduling decision processing, etc., which helps to improve the accuracy and efficiency of the control of passenger car clusters in the highway network, thereby improving the safety and operation efficiency of the highway network.

[0070] In some further optional embodiments, the spatiotemporal trajectory deduction task of the onboard perception monitoring state vector of each characteristic dimension is completed, and traffic flow scheduling decision processing is performed on the onboard perception monitoring state vector that has completed the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network, including: in response to the completion of the spatiotemporal trajectory deduction task of the onboard perception monitoring state vector of each characteristic dimension, obtaining the vehicle behavior element deduction vector corresponding to the onboard perception monitoring state vector that has completed the spatiotemporal trajectory deduction; and performing traffic flow scheduling decision processing on the vehicle behavior element deduction vector corresponding to the onboard perception monitoring state vector that has completed the spatiotemporal trajectory deduction.

[0071] In the next step, the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension is completed, and the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction is obtained, including: in response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, based on the state node information corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction, the basic trajectory feature data and spatiotemporal trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction are obtained; the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction is obtained. The behavior element deduction vector performs traffic flow scheduling decision processing, including: pairing the basic trajectory feature data and the space-time trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction; when the basic trajectory feature data and the space-time trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction are paired, the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction meets the traffic flow scheduling decision condition; when the basic trajectory feature data and the space-time trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector are not paired, the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction does not meet the traffic flow scheduling decision condition.

[0072] It can be understood that after the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector in response to each feature dimension is completed, the process of traffic flow scheduling decision processing is performed on the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network. When the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector in each feature dimension is completed, the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction is obtained. In this process, the basic trajectory feature data and spatiotemporal trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction are first obtained based on the state node information corresponding to the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction.

[0073] Take the characteristic dimension of vehicle speed as an example. For example, the range of vehicle speed is 0-200 km / h. In the process of spatiotemporal trajectory deduction, the value of vehicle speed in a certain section of road changes as follows: starting from 100 km / h, it gradually changes to 120 km / h within 10 seconds. This speed value and the related speed change rate (in this example, the speed change rate is (120-100) / 10=2 km / h²), the distribution of speed in different sections and other information constitute the basic trajectory feature data. The spatiotemporal trajectory derived data is data related to the traffic environment that is further derived from the basic trajectory feature data. For example, the speed limit of the section is 110 km / h. According to the speed and the speed limit regulations of the section, it is calculated that the vehicle exceeded the speed limit for 5 seconds in these 10 seconds (because the speeding stage from 110 km / h to 120 km / h lasted for 5 seconds), and the speeding ratio is (120-110) / 110≈9.1%. The characteristic vector of vehicle speed can be expressed as [100, 120, 2, 5, 9.1%], where each value represents the initial speed, final speed, speed change rate, speeding duration, and speeding ratio.

[0074] For the feature dimension of vehicle spacing, for example, the range of vehicle spacing is 5-100 meters. The basic trajectory feature data may include the numerical changes of vehicle spacing, for example, the vehicle spacing fluctuates from 30 meters to 25 meters over a period of time, and also include the comparison of vehicle spacing with the safety spacing standard (for example, the safety spacing standard is 30 meters). Spatiotemporal trajectory derivative data may be the potential collision risk value calculated based on the vehicle spacing and the speed change of the front and rear vehicles. For example, according to the relevant algorithm, the potential collision risk value is calculated to be 0.3 (here 0.3 is a relative value indicating the degree of risk, and the value range can be 0-1, and the larger the value, the higher the risk). The feature vector of vehicle spacing can be expressed as [30, 25, -5, 0.3], where each value represents the initial spacing, the final spacing, the spacing change, and the potential collision risk value.

[0075] After obtaining these data, the vehicle behavior element deduction vector corresponding to the vehicle perception monitoring state vector that has completed the space-time trajectory deduction is processed for traffic flow scheduling decision. This processing process includes pairing the basic trajectory feature data corresponding to the vehicle perception monitoring state vector that has completed the space-time trajectory deduction with the space-time trajectory derivative data.

[0076] Continuing with the example of vehicle speed, the speed change in the basic trajectory feature data is paired with the speeding duration or speeding ratio in the spatiotemporal trajectory derived data. There is a logical association between the basic trajectory feature data [100, 120, 2] of vehicle speed and the spatiotemporal trajectory derived data [5, 9.1%], and this association is a pairing relationship. When the basic trajectory feature data corresponding to the vehicle perception monitoring state vector is paired with the spatiotemporal trajectory derived data, the vehicle perception monitoring state vector that completes the spatiotemporal trajectory deduction meets the traffic flow scheduling decision conditions. For example, when the basic trajectory feature data of vehicle speed matches the speeding situation in the spatiotemporal trajectory derived data, it indicates that the vehicle's speed behavior has affected the traffic flow, such as possibly causing the vehicle spacing on the road section to become smaller or increasing the potential risk of collision. In this case, the traffic flow scheduling decision conditions are met. Possible decisions are to give speed limit reminders to vehicles or adjust the overall speed limit value of the road section.

[0077] On the contrary, when the basic trajectory feature data and the spatiotemporal trajectory derived data corresponding to the vehicle-mounted perception monitoring state vector are not matched, the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction does not meet the traffic flow scheduling decision conditions. For example, the basic trajectory feature data of the vehicle speed shows that the speed fluctuates within the normal range, such as from 90 km / h to 95 km / h, while the speeding-related data calculated in the spatiotemporal trajectory derived data is zero. There is no matching relationship between the two that indicates that the vehicle speed affects the traffic flow, so the traffic flow scheduling decision conditions are not met. At this time, there is no need to perform speed-related traffic flow scheduling operations on the vehicle.

[0078] From the perspective of multiple feature dimensions, the pairing relationship between the basic trajectory feature data of different feature dimensions such as vehicle speed, vehicle spacing, steering angle, and the spatiotemporal trajectory derived data jointly determines whether the vehicle meets the traffic flow scheduling decision conditions.

[0079] For the characteristic dimension of vehicle acceleration, for example, the value range of acceleration is -5 m / s² to 5 m / s². The basic trajectory feature data may include the numerical change of acceleration and the distribution of acceleration in different sections of the road. For example, under a certain operation, the vehicle acceleration starts from 0 m / s² and gradually increases to 2 m / s² within 5 seconds. The distribution of acceleration in different sections of the road is, for example, starting to accelerate before entering a curve. Spatiotemporal trajectory derivative data may be the time required for the vehicle to reach a certain speed calculated based on the acceleration and vehicle speed. For example, the initial speed of the vehicle is 80 km / h. According to relevant calculations, the time required for the vehicle to reach 100 km / h is 10 seconds (this is only an example calculation). The characteristic vector of acceleration can be expressed as [0, 2, 5, 10], where each value represents the initial acceleration, the final acceleration, the acceleration duration, and the time required to reach the target speed. If the basic trajectory feature data of acceleration shows that the vehicle acceleration suddenly increases, the spatiotemporal trajectory derived data calculates that the vehicle will reach a higher speed in a short time, and is successfully paired with data of other feature dimensions such as vehicle speed and vehicle spacing, it indicates that this change in acceleration has an impact on traffic flow and meets the traffic flow scheduling decision conditions. Possible decisions are to perform a rationality check on the vehicle acceleration or to limit the vehicle acceleration based on the overall traffic flow situation.

[0080] Let's look at the characteristic dimension of the vehicle's steering angle. For example, the range of the steering angle is -180° to 180°. The basic trajectory feature data includes the numerical changes of the steering angle, such as gradually turning from 0° to 10°, and the turning process lasts for 3 seconds. The spatiotemporal trajectory derived data may be data such as the vehicle turning radius calculated based on the steering angle and vehicle speed and the occupied space during the turning process. For example, based on the vehicle speed (80 km / h) and the steering angle (10°), the vehicle turning radius is calculated to be 50 meters, and the space occupied during the turning process is a rectangular area of ​​10 meters long and 8 meters wide (the values ​​here are only for example convenience). The characteristic vector of the steering angle can be expressed as [0, 10, 3, 50, 10, 8], where each value represents the initial steering angle, final steering angle, steering duration, turning radius, occupied space length, and occupied space width. When the basic trajectory feature data of the steering angle is paired with the spatiotemporal trajectory derived data, for example, the change in the steering angle causes the turning radius to become smaller and may affect the driving of vehicles in adjacent lanes, which meets the traffic flow scheduling decision conditions. Possible decisions at this time are to give the vehicle a steering prompt or adjust the traffic flow in the adjacent lane to avoid a collision.

[0081] This traffic flow scheduling decision-making process is a complex process that comprehensively considers multiple feature dimensions, basic trajectory feature data, and spatiotemporal trajectory derived data. Data from different feature dimensions influence and correlate with each other, and accurate pairing processing is used to determine whether the traffic flow scheduling decision conditions are met, thereby achieving effective scheduling of traffic flow. For example, on a road section containing multiple passenger cars, if the data of feature dimensions such as speed, spacing, acceleration, and steering angle of some vehicles meet the traffic flow scheduling decision conditions after processing, then through reasonable scheduling decisions, traffic congestion can be avoided, the traffic efficiency of the entire road section can be improved, and the safety of vehicle driving can be guaranteed.

[0082] At the same time, during the processing, there are different ways to deal with different types of traffic conditions. Under normal traffic flow conditions, the data of various characteristic dimensions of vehicles are relatively stable, and the pairing results of basic trajectory feature data and spatiotemporal trajectory derived data may only require slight traffic flow adjustments, such as fine-tuning speed limits or maintaining vehicle spacing.

[0083] For example, under normal traffic flow, the basic trajectory feature data of a vehicle speed is [80, 85, 1] ​​(initial speed 80 km / h, final speed 85 km / h, speed change rate 1 km / h²), the spatiotemporal trajectory derivative data is [0, 0] (speeding duration is 0, speeding ratio is 0), the basic trajectory feature data of vehicle spacing is [35, 35, 0] (initial spacing is 35 meters, final spacing is 35 meters, spacing change is 0), the spatiotemporal trajectory derivative data is [0.1] (potential collision risk value is 0.1), the basic trajectory feature data of the steering angle is [0, 0, 0] (initial steering angle, final steering angle, steering duration are all 0), and the spatiotemporal trajectory derivative data is [0] (no special derivative data indicates normal). In this case, after the data of each dimension is paired, it shows that the traffic situation is normal, and it may only need to perform simple traffic flow maintenance operations, such as reminding vehicles to maintain a safe distance.

[0084] During peak traffic hours or when special traffic events (such as road construction, traffic accidents, etc.) occur, the behavior characteristics of vehicles will change significantly. The pairing results of basic trajectory feature data and spatiotemporal trajectory derived data may lead to more stringent traffic flow scheduling decisions.

[0085] For example, near the scene of a traffic accident, the speed of the vehicle will decrease, the distance between vehicles will increase, and the steering angle may change irregularly. For example, the basic trajectory feature data of the speed of a vehicle near the scene of the accident is [60, 40, -2] (initial speed 60 km / h, final speed 40 km / h, speed change rate -2 km / h²), the derived data of the spatiotemporal trajectory is [0, 0] (speeding duration is 0, speeding ratio is 0), the basic trajectory feature data of the vehicle distance is [30, 50, 20] (initial distance 30 meters, final distance 50 meters, distance change 20 meters), the derived data of the spatiotemporal trajectory is [0.05] (the potential collision risk value is 0.05), the basic trajectory feature data of the steering angle is [0, 5, 2] (initial steering angle 0°, final steering angle 5°, steering duration 2 seconds), and the derived data of the spatiotemporal trajectory is

[20] (for example, indicating that the value of the additional space occupied when turning increases). After the spatiotemporal trajectory of these feature dimension data is deduced, the pairing results of the basic trajectory feature data and the spatiotemporal trajectory derived data will prompt the traffic flow scheduling system to take strict traffic flow scheduling decisions such as closing lanes and guiding vehicles to detour, so as to ensure the safety of the accident site and minimize the impact on the overall traffic flow.

[0086] In the process of traffic flow scheduling decision-making, the real-time and accuracy of the data must also be considered. Since vehicles are in a dynamic driving state in the highway network, the basic trajectory feature data and spatiotemporal trajectory derivative data need to be continuously updated. For example, the vehicle speed may change instantly due to the driver's operation or changes in road conditions, which requires the data acquisition and processing system to be able to capture these changes in time and update the relevant data. Only in this way can the paired processing results of the basic trajectory feature data and the spatiotemporal trajectory derivative data accurately reflect the impact of the vehicle's actual behavior on traffic flow, so as to make correct traffic flow scheduling decisions.

[0087] From the perspective of the entire highway network, traffic flow scheduling decisions on different sections are interrelated. Traffic flow scheduling decisions on one section may affect the traffic flow of adjacent sections or even the entire highway network. For example, if a vehicle flow restriction decision is made on a certain entrance section, this may lead to a reduction in traffic flow on subsequent sections, and the traffic flow scheduling decisions on subsequent sections need to be adjusted accordingly. Similarly, if a decision is made to divert vehicles on a congested section, such as directing vehicles to detour to an adjacent section, then the adjacent sections need to make reasonable traffic flow scheduling based on the increased traffic flow, such as adjusting lane usage rules or speed limits.

[0088] Therefore, by obtaining the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction, including the basic trajectory feature data and the spatiotemporal trajectory derivative data, and performing pairing processing to determine whether the traffic flow scheduling decision conditions are met, this method can comprehensively and accurately analyze the impact of vehicle behavior on traffic flow. Comprehensively considering the behavior of vehicles from multiple feature dimensions makes the traffic flow scheduling decision not only based on a single factor, but also integrates multiple factors such as vehicle speed, spacing, acceleration, steering angle, etc., which improves the accuracy of the decision. This precise traffic flow scheduling decision helps to optimize the traffic flow in the highway network and avoid traffic congestion. It can ensure the safety of vehicle driving under normal and special traffic conditions, and takes into account the mutual influence between different sections, realizing the effective control of the traffic flow of the entire highway network.

[0089] In some preferred technical ideas, the traffic flow scheduling decision label of the on-board perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, and the spatiotemporal trajectory deduction code is updated to obtain the spatiotemporal trajectory trend update code, including: in response to the traffic flow scheduling decision label of the on-board perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network, the traffic flow scheduling decision label is a potential congestion label, and based on the characteristic dimension of the on-board perception monitoring state vector, the trajectory trend update module corresponding to the on-board perception monitoring state vector is determined; the spatiotemporal trajectory deduction code is updated by calling the trajectory trend update module corresponding to the on-board perception monitoring state vector of each characteristic dimension through the inter-class attention strategy to obtain the spatiotemporal trajectory trend update code.

[0090] Furthermore, the traffic flow scheduling decision label of the on-board perception monitoring state vector that completes the spatiotemporal trajectory deduction in response to the vehicle-road spatiotemporal graph neural network is a potential congestion label, and based on the characteristic dimension of the on-board perception monitoring state vector, a trajectory trend update module corresponding to the on-board perception monitoring state vector is determined, including: when the characteristic dimension of the on-board perception monitoring state vector is a dynamic characteristic dimension, the spatiotemporal trajectory deduction module corresponding to the on-board perception monitoring state vector of the dynamic characteristic dimension is determined to be a temporal attention trajectory trend update module; when the characteristic dimension of the on-board perception monitoring state vector is an event characteristic dimension, the spatiotemporal trajectory deduction module corresponding to the on-board perception monitoring state vector of the event characteristic dimension is determined to be a spatial attention trajectory trend update module.

[0091] In the actual application process, for the technical solution of updating the spatiotemporal trajectory deduction code to obtain the spatiotemporal trajectory trend update code when the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, when the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, determine the trajectory trend update module corresponding to the vehicle-mounted perception monitoring state vector. Here, the distinction is made based on the different properties of the characteristic dimensions.

[0092] When the characteristic dimension of the vehicle-mounted perception monitoring state vector is a dynamic characteristic dimension, such as the speed and acceleration of the vehicle that continuously change during driving, the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the dynamic characteristic dimension is determined as the time attention trajectory trend update module. Taking the dynamic characteristic dimension of speed as an example, the value range of speed may be between 0-200 km / h. In the case of potential congestion, the impact of the change of speed factor on traffic flow has different importance at different time points. For example, in the period before entering the congested area, the speed of vehicle speed reduction (i.e., speed change rate) may have a key impact on subsequent traffic conditions. For example, at a distance of 5 kilometers from the possible congestion point, the vehicle speed gradually decreases from 100 km / h and drops to 80 km / h within 10 minutes. The change of this speed at different time points will be focused on in the time attention trajectory trend update module. For the dynamic characteristic dimension of acceleration, its value range may be between -5 m / s² and 5 m / s². In a potential congestion scenario, if the vehicle suddenly accelerates or decelerates (for example, the acceleration suddenly changes to -3 m / s² for emergency braking), this sudden change in acceleration and the time when it occurs are important considerations in the temporal attention trajectory trend update module.

[0093] When the feature dimension of the vehicle-mounted perception monitoring state vector is the event feature dimension, such as the dimension related to events such as traffic accidents and road construction, the space-time trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the event feature dimension is determined as the spatial attention trajectory trend update module. For example, under the event feature dimension of road construction, for example, the construction section occupies one lane, and the section is normally a two-way four-lane section. The location information of the construction section becomes the focus of the spatial attention trajectory trend update module. If the construction section is located 20 kilometers from the starting point of the road, this spatial location information and the relative position relationship of the vehicle on the highway network are crucial to alleviating potential congestion. For another example, in the event feature dimension of traffic accidents, spatial factors such as the location of the accident, the spatial range occupied by the accident vehicle, and the impact on the surrounding lanes will be analyzed in detail in the spatial attention trajectory trend update module. For example, if an accident occurs in a lane and the vehicle occupies part of the lane, causing the surrounding vehicles to need to avoid, this change in spatial layout and the spatial impact on traffic flow are the key analysis content of the spatial attention trajectory trend update module.

[0094] Then, the trajectory trend update module corresponding to the vehicle perception monitoring state vector of each feature dimension is called through the inter-class attention strategy to update the spatiotemporal trajectory deduction code and obtain the spatiotemporal trajectory trend update code.

[0095] In the process of updating the spatiotemporal trajectory deduction code of the dynamic feature dimension of speed by the time attention trajectory trend update module, since the time variation characteristics of speed are critical in potential congestion conditions, the module will adjust the original spatiotemporal trajectory deduction code according to the change of speed at different times. For example, the original spatiotemporal trajectory deduction code records the values ​​of vehicle speed at certain time points. When updating, the code will be supplemented and corrected according to the current speed change trend (such as continuous decrease and accelerated decrease speed). For example, the speed-related part of the original code is [100, 90, 85] (indicating the speed at different time points). According to the current speed dropping rapidly from 85 km / h to 80 km / h in a short period of time, the updated code may become [100, 90, 85, 80, -5], where -5 represents the value of the speed decrease trend. For the dynamic feature dimension of acceleration, the original space-time trajectory deduction code records the changes in acceleration, such as [0, 1, -2] (indicating acceleration at different time points). If the current vehicle has frequent acceleration fluctuations (such as rapid changes between -2 m / s² and 1 m / s²), the updated code may record this fluctuation in detail and become [0, 1, -2, -1, 1] to more accurately reflect the complex changes in vehicle acceleration under potential congestion conditions.

[0096] When the spatial attention trajectory trend update module updates the spatiotemporal trajectory deduction code of the event feature dimension of road construction, since the spatial position of the construction section is a key factor, the module will update the code according to the spatial relationship between the construction section and the vehicle. For example, the part of the original spatiotemporal trajectory deduction code about the construction section is [20, 1] (20 means that the construction section is located at 20 kilometers, and 1 means that it occupies one lane). If it is found that the construction section has a larger impact on the surrounding vehicles than previously estimated, the updated code may become [20, 1, 3], where 3 means that the construction section has an impact on the vehicles in the three surrounding lanes. For the event feature dimension of traffic accidents, the original spatiotemporal trajectory deduction code records the basic spatial information of the accident, such as [15, 2, 1] (15 means that the accident occurred at 15 kilometers, 2 means that the accident vehicle occupies two lanes, and 1 means that the accident has an impact on one lane). If the impact range of the accident changes (for example, it affects both lanes), the updated code may become [15, 2, 2].

[0097] From a comprehensive perspective of multiple feature dimensions, under the label of traffic flow scheduling decision with potential congestion, the temporal attention trajectory trend update module of dynamic feature dimensions such as speed and acceleration and the spatial attention trajectory trend update module of event feature dimensions such as road construction and traffic accidents work together to update the spatiotemporal trajectory deduction code. For example, when the speed reduction is due to road construction or traffic accidents ahead, the temporal attention trajectory trend update module and the spatial attention trajectory trend update module will work together. There is an intrinsic connection between the speed reduction trend (temporal dimension) and the spatial location (spatial dimension) of the construction or accident, which will be reflected when updating the spatiotemporal trajectory trend update code. If the speed continues to decrease and gets closer to the accident or construction section, the spatiotemporal trajectory trend update code will accurately reflect the change in this spatiotemporal relationship, so that more effective intelligent management and control can be carried out based on this update code in the future.

[0098] This method of determining different trajectory trend update modules based on different feature dimensions and calling these modules to update the spatiotemporal trajectory deduction code through the inter-class attention strategy can comprehensively and accurately adjust the spatiotemporal trajectory information of the vehicle according to the potential congestion situation. Information of different feature dimensions is processed in their respective modules, and the coordination between modules is achieved through the inter-class attention strategy, so that the updated spatiotemporal trajectory trend update code can better reflect the actual status and development trend of the vehicle in the potential congestion scenario.

[0099] With this design, by determining different trajectory trend update modules according to the feature dimensions of the vehicle-mounted perception monitoring state vector, it is possible to process the information of dynamic feature dimensions and event feature dimensions in potential congestion situations in a targeted manner. These modules are called using the inter-class attention strategy to update the spatiotemporal trajectory deduction coding, which improves the accuracy and comprehensiveness of the coding update. The resulting spatiotemporal trajectory trend update coding can more accurately reflect the spatiotemporal trajectory changes of vehicles under potential congestion conditions, thereby providing a more reliable basis for the intelligent control of the vehicle cluster to be controlled based on this coding, which helps to alleviate potential congestion and improve the operating efficiency and safety of the highway network.

[0100] In some other examples, the method of obtaining the on-board perception monitoring status description set of the vehicle cluster to be controlled in the highway network area includes: obtaining the original on-board perception monitoring state vectors of multiple different feature dimensions of the vehicle cluster to be controlled; performing classification operations on the original on-board perception monitoring state vectors of each feature dimension to obtain the on-board perception monitoring state vectors of multiple different feature dimensions; based on the statistical rules of the set monitoring state description set, mapping the on-board perception monitoring state vectors of multiple different feature dimensions to the set monitoring state description set to obtain the on-board perception monitoring state description set of the vehicle cluster to be controlled.

[0101] In this technical solution, the step of obtaining a vehicle-mounted perception monitoring state description set of a vehicle cluster to be controlled in a highway network area is to first obtain original vehicle-mounted perception monitoring state vectors of multiple different feature dimensions of the vehicle cluster to be controlled.

[0102] Take the characteristic dimension of vehicle speed as an example. For example, within a period of time (such as 1 minute, with a collection frequency of once every 5 seconds), the values ​​in the original vehicle-mounted perception monitoring state vector of vehicle speed are [30, 35, 40, 42, 45, 50] kilometers per hour (the value range here is between 0-200 kilometers per hour). This numerical sequence is collected by the vehicle's speed sensor, which reflects the change in the vehicle's speed within this 1 minute.

[0103] For the feature dimension of vehicle spacing, for example, the range of vehicle spacing is between 5-100 meters, and the values ​​in the original vehicle perception monitoring state vector (also with a time period of 1 minute and a collection frequency of once every 5 seconds) are [30, 32, 35, 33, 30, 28] meters. These values ​​are the measured values ​​of the distance between the vehicle and the vehicle in front, reflecting the dynamic changes in the distance between the vehicle and the vehicle in front during driving.

[0104] For the characteristic dimension of vehicle acceleration, the value range may be between -5 m / s² and 5 m / s². For example, under a specific operation (such as accelerating to overtake), the value in the original vehicle perception monitoring state vector (collected once every 3 seconds for 15 seconds) is [0.5, 1, 1.2, 1.5, 2] m / s². This numerical sequence shows the change of vehicle acceleration during this period of time, reflecting the change of vehicle power.

[0105] Let's look at the characteristic dimension of the vehicle's steering angle. Its value range is -180° to 180°. For example, during the vehicle's turning process (the acquisition frequency is once every 2 seconds for 10 seconds), the values ​​in the original vehicle-mounted perception monitoring state vector are [5, 8, 10, 12, 15]°. These values ​​are collected by the vehicle's steering sensor and reflect the gradual change in the vehicle's steering angle when turning.

[0106] Then, the original vehicle-mounted perception monitoring state vector of each feature dimension is classified to obtain vehicle-mounted perception monitoring state vectors of multiple different feature dimensions.

[0107] For vehicle speed, according to the classification method mentioned above (0-60 km / h is low speed, 60-120 km / h is medium speed, and 120-200 km / h is high speed), the values ​​in the above original speed vector [30, 35, 40, 42, 45, 50] km / h all belong to the low speed category, so the on-board perception monitoring state vector of the characteristic dimension of vehicle speed after classification can be expressed as [low speed, low speed, low speed, low speed, low speed, low speed].

[0108] For vehicle spacing, classification is performed based on the relationship between the spacing and the safety spacing standard (for example, the safety spacing is 30 meters) (a spacing greater than twice the safety spacing, i.e. 60 meters, is a sufficient safety spacing class; a spacing between 0.8 and 2 times the safety spacing, i.e. 24 and 60 meters, is a normal spacing class; a spacing less than 0.8 times the safety spacing, i.e. less than 24 meters, is a dangerous spacing class). Except for the last 28 meters, which belongs to the normal spacing class, the values ​​in the original vehicle spacing vector [30, 32, 35, 33, 30, 28] meters all belong to the normal spacing class. Therefore, the on-board perception monitoring state vector of the feature dimension of vehicle spacing after classification is [normal spacing class, normal spacing class, normal spacing class, normal spacing class, normal spacing class, normal spacing class].

[0109] For the vehicle's acceleration, according to the previous classification (acceleration greater than 0 and less than 3 m / s² is normal acceleration, acceleration less than 0 and greater than -3 m / s² is normal deceleration, acceleration greater than 3 m / s² is sudden acceleration, and acceleration less than -3 m / s² is sudden deceleration), the values ​​in the original acceleration vector [0.5, 1, 1.2, 1.5, 2] m / s² all belong to the normal acceleration class, so the classified on-board perception monitoring state vector of the acceleration feature dimension is [normal acceleration class, normal acceleration class, normal acceleration class, normal acceleration class, normal acceleration class].

[0110] For the vehicle steering angle, according to the previous classification (steering angles between -10° and 10° and low steering frequency are straight-line driving, steering angles between 10°-45° or -10°--45° are small-amplitude steering, and steering angles greater than 45° or less than -45° are large-amplitude steering), the values ​​in the original steering angle vector [5, 8, 10, 12, 15]° all belong to the small-amplitude steering class, so the classified on-board perception monitoring state vector of the feature dimension of the vehicle steering angle is [small-amplitude steering class, small-amplitude steering class, small-amplitude steering class, small-amplitude steering class].

[0111] Then, based on the statistical rules of the set monitoring state description set, the vehicle-borne perception monitoring state vectors of multiple different feature dimensions are mapped to the set monitoring state description set to obtain the vehicle-borne perception monitoring state description set of the vehicle cluster to be controlled.

[0112] The monitoring state description set may include descriptions such as "speed is stable in the low speed range", "speed changes from low speed to medium speed", and "speed is stable in the medium speed range" for vehicle speed. Since all values ​​in the vehicle perception monitoring state vector of the classified vehicle speed feature dimension belong to the low speed category, according to statistical rules (for example, when the value proportion in a certain category reaches 100%), the vehicle speed feature dimension is mapped to the description of "speed is stable in the low speed range".

[0113] For vehicle spacing, the monitoring status description set may include descriptions such as "vehicle spacing is in a sufficient state", "vehicle spacing is in a normal range", and "vehicle spacing is in a dangerous range". Since most of the values ​​in the vehicle perception monitoring state vector of the classified vehicle spacing feature dimension belong to the normal spacing category, according to statistical rules (for example, when the value in a certain category accounts for more than 80%), the vehicle spacing feature dimension is mapped to the description of "vehicle spacing is in a normal range".

[0114] For vehicle acceleration, the monitoring state description set may include descriptions such as "the vehicle is in stable driving (acceleration is zero)", "the vehicle is in normal acceleration state", and "the vehicle is in rapid acceleration state". Since the values ​​in the vehicle perception monitoring state vector of the classified acceleration feature dimension all belong to the normal acceleration category, according to statistical rules (for example, when the value in a certain category accounts for 100%), the acceleration feature dimension is mapped to the description of "the vehicle is in normal acceleration state".

[0115] For the vehicle steering angle, the monitoring state description set may include descriptions such as "the vehicle is in a straight-line driving state", "the vehicle is in a small-amplitude steering state", and "the vehicle is in a large-amplitude steering state". Since the values ​​in the vehicle perception monitoring state vector of the classified vehicle steering angle feature dimension all belong to the small-amplitude steering category, according to statistical rules (for example, when the value in a certain category accounts for 100%), the steering angle feature dimension is mapped to the description of "the vehicle is in a small-amplitude steering state".

[0116] In the actual highway network, the state vectors of various characteristic dimensions of different vehicles, after such mapping operations, together constitute the vehicle perception monitoring state description set of the vehicle cluster to be controlled. For example, in a vehicle cluster to be controlled that includes three passenger cars, the speed of the first car is mapped to "the speed is stable in the low speed range", the spacing is mapped to "the distance between vehicles is in the normal range", the acceleration is mapped to "the vehicle is in normal acceleration state", and the steering angle is mapped to "the vehicle is in a small steering state"; the speed of the second car is mapped to "the speed is stable in the medium speed range", the spacing is mapped to "the distance between vehicles is in the normal range", the acceleration is mapped to "the vehicle is in normal acceleration state", and the steering angle is mapped to "the vehicle is in a straight driving state"; the speed of the third car is mapped to "the speed is stable in the low speed range", the spacing is mapped to "the distance between vehicles is in the normal range", the acceleration is mapped to "the vehicle is in a normal acceleration state", and the steering angle is mapped to "the vehicle is in a small steering state". This vehicle perception monitoring state description set describes the driving state of this vehicle cluster as a whole. Most vehicles are in a normal driving state, but some vehicles are in a low speed range, and further attention may be needed to determine whether it will affect traffic flow.

[0117] From the perspective of different feature dimensions of multiple vehicles, the mapping results of each feature dimension are interrelated in the vehicle perception monitoring state description set. For example, if the speed of a vehicle is mapped to "the speed changes from low speed to medium speed", the acceleration is mapped to "the vehicle is in a state of rapid acceleration", and the distance between vehicles is mapped to "the distance between vehicles is within the normal range", this series of mapping results indicates that the driving state of the vehicle is changing, which may affect the surrounding vehicles and require appropriate control or early warning.

[0118] The same feature dimensions between different vehicles also have mutual relationships. For example, in a vehicle cluster to be controlled, the speed of one vehicle is in a low-speed range, while the speed of another vehicle is in a high-speed range. This speed difference will affect the traffic smoothness of the entire vehicle cluster, which is reflected in the vehicle perception monitoring state description set through different mapping results of the feature dimension of speed. Similarly, the mapping results of the feature dimension of vehicle spacing between different vehicles will also affect the driving safety of the entire vehicle cluster. If the spacing of one vehicle is in the dangerous spacing category, even if the spacing of other vehicles is normal, it may pose a threat to the safe driving of the entire vehicle cluster.

[0119] This method of obtaining the vehicle-mounted perception monitoring state description set converts the complex vehicle driving state data into a more semantic and overall descriptive state description set by classifying and mapping the original vehicle-mounted perception monitoring state vector. This not only helps to accurately summarize the driving state of a single vehicle, but also comprehensively grasps the driving conditions of vehicles in the entire highway network area from the perspective of vehicle clusters, providing rich and effective data support for subsequent traffic management and control, thereby improving the safety and operation efficiency of the highway network.

[0120] It can be seen that by obtaining the original vehicle-mounted perception monitoring state vectors of multiple different feature dimensions and performing classification operations, and then mapping according to the statistical rules of the set monitoring state description set, the complex vehicle driving state information can be integrated into a comprehensive and overall descriptive vehicle-mounted perception monitoring state description set. This method can effectively summarize the state of a single vehicle and grasp the overall driving status of the vehicle cluster. The correlation between different feature dimensions and between different vehicles can be reflected in the state description set, providing an accurate data basis for subsequent traffic management operations, helping to improve the ability to control vehicles in the highway network, and thus improve the safety and operation efficiency of the highway network.

[0121] Based on S100-S500, the method also includes: in response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each feature dimension, obtaining the highway road condition simulation result corresponding to the vehicle-road spatiotemporal graph neural network; labeling the vehicle cluster to be controlled based on the highway road condition simulation result to obtain the driving status labeling information of the vehicle cluster to be controlled; based on the driving status labeling information, performing road section diversion operations on the vehicle cluster to be controlled in the highway network area.

[0122] In the technical solution of the embodiment of the present application, based on the operation process of S100-S500, when the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each feature dimension is completed, the highway road condition simulation result corresponding to the vehicle-road spatiotemporal graph neural network will be obtained. This highway road condition simulation result is a simulation presentation of the highway network condition obtained through a series of previous operations on the vehicle-mounted perception monitoring state vector (such as spatiotemporal trajectory deduction, etc.) and the characteristics of the vehicle-road spatiotemporal graph neural network.

[0123] After obtaining the high-speed road condition simulation results, the vehicle cluster to be controlled is annotated based on this result, so as to obtain the driving status annotation information of the vehicle cluster to be controlled. Taking the vehicle speed as an example, if the high-speed road condition simulation results show that the speed of a vehicle in a specific section is continuously lower than the average speed of the section and lower than the normal speed range (for example, the normal speed range of the section is 80-120 kilometers / hour, and the vehicle speed is continuously below 60 kilometers / hour), then the vehicle may be marked as "low-speed driving state" when annotating. From the perspective of vehicle spacing, if the simulation results show that the spacing between a vehicle and the front and rear vehicles is longer than the normal spacing for a long time (for example, the normal spacing range is 30-50 meters, and the spacing between the vehicle and the front and rear vehicles is continuously more than 60 meters), it may be marked as "excessive spacing state". For vehicle acceleration, if a vehicle frequently accelerates or decelerates rapidly in the simulation results (for example, the absolute value of acceleration is frequently greater than 3 meters / second²), it may be marked as "unstable acceleration state".

[0124] For the steering angle of the vehicle, if the simulation results show that the steering angle of the vehicle changes frequently and with a large amplitude (for example, the steering angle changes between -45° and 45° multiple times in a short period of time), it may be marked as "frequent steering state". These annotations of different dimensions are combined to form the driving state annotation information of each vehicle in the vehicle cluster to be controlled.

[0125] Based on such driving status labeling information, road section guidance operations are performed on the controlled vehicle clusters in the highway network area. If some vehicles are marked as "low-speed driving state" and are on the main road, and there are vehicles behind them driving at normal speeds, in order to avoid traffic congestion or rear-end collision risks, the road section guidance operation may be to guide the low-speed vehicles to change lanes or remind them to speed up. For vehicles marked as "excessive spacing state", they may need to be reminded to maintain a reasonable spacing to prevent other vehicles from illegally inserting and causing danger, while also avoiding unnecessary waste of road space.

[0126] When there is a vehicle marked as "unstable acceleration state", the surrounding vehicles may be warned during the road section diversion operation to inform them that there are unstable vehicles around them, so that the surrounding vehicles can remain vigilant and adjust the distance or speed appropriately. For vehicles marked as "frequent turning state", if they are on a curve or a road with heavy traffic, they may be guided to drive smoothly or their speed may be temporarily limited to ensure traffic safety and traffic smoothness of the entire road section.

[0127] From the perspective of the entire vehicle cluster to be controlled, the driving status labeling information of different vehicles has a mutual influence relationship. For example, when multiple vehicles are marked as "low-speed driving state" and concentrated in the same area, this will have a greater impact on the traffic flow of the entire road section. The road section diversion operation needs to consider how to adjust the driving status of these vehicles as a whole, such as guiding this group of vehicles to change lanes or adjust to a suitable speed range through variable information signs.

[0128] At the same time, when conducting road section diversion operations, the overall structure of the expressway network and traffic rules also need to be considered. For example, in some special sections (such as tunnels, bridges, etc.), the driving status of vehicles is more stringent, and diversion operations need to follow the regulations of these special sections. If a vehicle is found near the entrance of a tunnel and is marked as "unstable acceleration state", due to the special conditions of light, ventilation, etc. in the tunnel, diversion operations may be more cautious, such as guiding the vehicle to adjust its driving status before entering the tunnel, or setting up a temporary speed limit sign at the entrance of the tunnel and reminding other vehicles to pay attention.

[0129] The highway simulation results may also include some information related to road facilities. For example, there may be road construction or road damage on a certain road section, which will affect the normal driving of vehicles. When vehicles marked with different driving states pass through these sections, the road section diversion operation needs to be adjusted according to the specific conditions of the road facilities. If a vehicle is marked as "too large spacing state" and is about to pass through a road construction section, the diversion operation may not only remind it to maintain a reasonable spacing, but also inform it of the road construction ahead and guide it to make preparations such as slowing down in advance.

[0130] In addition, the type of vehicle may also play a role in the driving status labeling information and road section diversion operations. Although the embodiments of the present application are mainly aimed at passenger cars, different types of passenger cars (such as sedans, SUVs, etc.) may have differences in driving performance. For example, SUVs have larger bodies and their performance in steering and braking is different from that of sedans. If an SUV is marked as "frequent turning state", it may be necessary to give more targeted guidance based on its vehicle characteristics during road section diversion operations, such as reminding it to pay attention to the body swing range.

[0131] In the process of conducting road section diversion operations on the controlled vehicle cluster, it is also necessary to continuously update the highway road condition simulation results and driving status labeling information. Because the vehicle status changes dynamically during driving, for example, a vehicle originally marked as "normal driving state" may become "unstable driving state" due to sudden operations by the driver or sudden road conditions (such as sudden obstacles). Therefore, it is necessary to adjust the road section diversion operations in a timely manner according to the new situation to ensure the safety and smooth traffic of the highway network.

[0132] The following design, by obtaining the highway road condition simulation results and marking the driving status of the controlled vehicle cluster, can fully and carefully understand the driving status of the vehicle, and provide an accurate basis for the road section diversion operation. The road section diversion operation based on the driving status marking information comprehensively considers multiple factors such as the individual status of the vehicle, the mutual influence between vehicles, the highway network structure, the road facility conditions, and the vehicle type, thereby improving the effectiveness and pertinence of the road section diversion operation, helping to optimize the traffic flow of the highway network, ensure traffic safety, and improve the operating efficiency of the entire highway network.

[0133] In summary, the vehicle-road spatiotemporal graph neural network processing in the embodiment of the present application obtains the on-board perception monitoring state vector, which can integrate multi-source vehicle data and comprehensively reflect the vehicle status. Spatiotemporal trajectory trend mining accurately encodes the behavior of the vehicle in different time and space, and considers the collaborative trend of multiple feature dimensions to make the vehicle behavior description more accurate. The inter-class attention strategy focuses on the relationship between feature dimensions in the spatiotemporal trajectory deduction, adapts to a variety of traffic scenarios, and improves the accuracy of trajectory deduction. Traffic flow scheduling decision processing integrates multiple factors, including vehicle trajectory and road network conditions, to improve the rationality of decision-making. For intelligent control of potential congestion, based on dynamically adjusted coding updates, it can effectively avoid congestion, improve the efficiency and safety of highway network operation, and can also coordinate with other traffic management systems to optimize the overall traffic resource allocation.

[0134] Further, Figure 2 This is a schematic diagram of the structure of an intelligent management and control system 200 provided in an embodiment of the present application. Figure 2 The intelligent management and control system 200 shown includes a processor 210, which can call and run a computer program from a memory to implement the method in the embodiment of the present application.

[0135] Alternatively, if Figure 2 As shown, the intelligent management and control system 200 may further include a memory 230. The processor 210 may call and run a computer program from the memory 230 to implement the method in the embodiment of the present application.

[0136] The memory 230 may be a separate device independent of the processor 210 , or may be integrated into the processor 210 .

[0137] Alternatively, if Figure 2 As shown, the intelligent management and control system 200 may further include a transceiver 220, and the processor 210 may control the transceiver 220 to interact with other devices. Specifically, it may send information or data to other devices, or receive information or data sent by other devices.

[0138] Optionally, the intelligent management and control system 200 can implement the corresponding processes corresponding to the storage engine or components in the storage engine (such as a processing module) or a device deployed with a storage engine in each method of the embodiments of the present application. For the sake of brevity, they are not repeated here.

[0139] It should be understood that the processor of the embodiment of the present application may be an integrated circuit chip with signal processing capabilities.

[0140] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. It should be noted that the memory of the systems and methods described herein is intended to include but is not limited to suitable types of memory.

[0141] Based on the above, a readable storage medium is provided, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the steps of the above method are implemented.

[0142] It should be noted that, in this article, the term "includes", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method, article or device including the element. In addition, it should be noted that the scope of the method and device in the embodiment of the present application is not limited to performing functions in the order shown or discussed, and may also include performing functions in a substantially simultaneous manner or in reverse order according to the functions involved, for example, the described method may be performed in an order different from that described, and various steps may also be added, omitted, or combined. In addition, the features described with reference to certain examples may be combined in other examples.

[0143] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a computer software product, which is stored in a storage medium (such as ROM / RAM, a disk, or an optical disk), including a number of instructions for a terminal (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods described in each embodiment of the embodiment of the present application.

[0144] The embodiments of the embodiments of the present application are described above in conjunction with the accompanying drawings, but the embodiments of the present application are not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the embodiments of the present application, ordinary technicians in this field can also make many forms without departing from the purpose of the embodiments of the present application and the scope of protection of the embodiments of the present application, all of which are within the protection of the embodiments of the present application.

Claims

1. A special high-speed intelligent control method for passenger cars based on road safety, characterized in that: The method is implemented by an intelligent management and control system, and the method includes: Obtaining a vehicle-mounted perception monitoring state description set of a vehicle cluster to be controlled in a highway network area, wherein the vehicle-mounted perception monitoring state description set includes a vehicle-mounted perception monitoring state vector of at least one feature dimension obtained after being processed by a vehicle-road spatiotemporal graph neural network, and the vehicle cluster to be controlled includes at least two passenger cars; Based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, the vehicle-mounted perception monitoring state vector is mined for spatiotemporal trajectory trends to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension; Calling the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension through the inter-class attention strategy to perform spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction code; In response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, traffic flow scheduling decision processing is performed on the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network; In response to the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the space-time trajectory deduction in the vehicle-road space-time graph neural network being a potential congestion label, the space-time trajectory deduction code is updated to obtain the space-time trajectory trend update code, and the vehicle cluster to be controlled is intelligently controlled based on the space-time trajectory trend update code.

2. The method according to claim 1, characterized in that The inter-class attention strategy is used to call the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension to perform spatiotemporal trajectory deduction on the vehicle-mounted perception monitoring state vector in the vehicle-road spatiotemporal graph neural network to obtain the spatiotemporal trajectory deduction code, including: Based on the characteristic dimensions of the vehicle-mounted sensing monitoring state vector, determine the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted sensing monitoring state vector of each characteristic dimension; The spatiotemporal trajectory trend encoding corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is called through the inter-class attention strategy to enhance the confidence of the corresponding spatiotemporal trajectory deduction module, thereby obtaining a spatiotemporal trajectory deduction enhancement module; The space-time trajectory deduction enhancement module corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is called through the inter-class attention strategy, and the space-time trajectory deduction of the vehicle-mounted perception monitoring state vector is performed in the vehicle-road space-time graph neural network to obtain the space-time trajectory deduction code.

3. The method according to claim 2, characterized in that The inter-class attention strategy is used to call the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension to enhance the confidence of the corresponding spatiotemporal trajectory deduction module, and obtain the spatiotemporal trajectory deduction enhancement module, including: The inter-class attention strategy is used to call the spatiotemporal trajectory trend encoding corresponding to the vehicle perception monitoring state vector of each feature dimension to extract the vehicle behavior element vector corresponding to the vehicle cluster to be controlled; Determine the deduction logic indicator vector corresponding to each spatiotemporal trajectory deduction module through the inter-class attention strategy; Through the inter-class attention strategy, the vehicle behavior element vector corresponding to the vehicle cluster to be controlled is added to the deduction logic indication vector of the corresponding space-time trajectory deduction module to obtain the space-time trajectory deduction enhancement module.

4. The method according to claim 2, characterized in that The method of determining the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted sensing monitoring state vector of each characteristic dimension based on the characteristic dimension of the vehicle-mounted sensing monitoring state vector includes: When the characteristic dimension of the vehicle-mounted perception monitoring state vector is a dynamic characteristic dimension, determining that the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the dynamic characteristic dimension is a spatiotemporal trajectory deduction module based on environmental sensor monitoring; When the characteristic dimension of the vehicle-mounted perception monitoring state vector is an event characteristic dimension, it is determined that the space-time trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the event characteristic dimension is a space-time trajectory deduction module based on congestion prediction and discrimination.

5. The method according to claim 1, characterized in that The method of performing spatiotemporal trajectory trend mining on the vehicle-mounted perception monitoring state vector based on the characteristic dimension of the vehicle-mounted perception monitoring state vector to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each characteristic dimension includes: Based on the characteristic dimensions of the vehicle-mounted sensing monitoring state vector, generating an original spatiotemporal trajectory trend code corresponding to the vehicle-mounted sensing monitoring state vector of each characteristic dimension; Performing behavior recognition on the vehicle-mounted sensing monitoring state vector of each feature dimension to obtain a vehicle behavior element vector corresponding to the vehicle-mounted sensing monitoring state vector of each feature dimension; The vehicle behavior element vector corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension is added to the antecedent spatiotemporal data chain corresponding to the corresponding original spatiotemporal trajectory trend code to obtain the spatiotemporal trajectory trend code corresponding to the vehicle-mounted perception monitoring state vector of each feature dimension.

6. The method according to claim 1, characterized in that The task of spatiotemporal trajectory deduction of the vehicle-mounted perception monitoring state vector in response to each characteristic dimension is completed, and traffic flow scheduling decision processing is performed on the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network, including: In response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, obtaining a vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector for which the spatiotemporal trajectory deduction is completed; Performing traffic flow scheduling decision processing on the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction; The step of obtaining a vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that has completed the space-time trajectory deduction in response to the completion of the space-time trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, including: obtaining basic trajectory feature data and space-time trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that has completed the space-time trajectory deduction based on the state node information corresponding to the vehicle-mounted perception monitoring state vector that has completed the space-time trajectory deduction in response to the completion of the space-time trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension; The traffic flow scheduling decision processing of the vehicle behavior element deduction vector corresponding to the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction includes: Pairing the basic trajectory feature data and the spatiotemporal trajectory derivative data corresponding to the vehicle-mounted perception monitoring state vector that has completed the spatiotemporal trajectory deduction; When the basic trajectory feature data and the spatiotemporal trajectory derived data corresponding to the vehicle-mounted perception monitoring state vector are matched, the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction meets the traffic flow scheduling decision condition; When the basic trajectory feature data and the spatiotemporal trajectory derived data corresponding to the vehicle-mounted perception monitoring state vector are not matched, the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction does not meet the traffic flow scheduling decision condition.

7. The method according to claim 1, characterized in that The traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, and the spatiotemporal trajectory deduction code is updated to obtain the spatiotemporal trajectory trend update code, including: In response to the traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network being a potential congestion label, a trajectory trend update module corresponding to the vehicle-mounted perception monitoring state vector is determined based on the characteristic dimension of the vehicle-mounted perception monitoring state vector; The trajectory trend update module corresponding to the vehicle perception monitoring state vector of each feature dimension is called through the inter-class attention strategy to update the spatiotemporal trajectory deduction code to obtain the spatiotemporal trajectory trend update code; The traffic flow scheduling decision label of the vehicle-mounted perception monitoring state vector that completes the spatiotemporal trajectory deduction in the vehicle-road spatiotemporal graph neural network is a potential congestion label, and based on the characteristic dimension of the vehicle-mounted perception monitoring state vector, a trajectory trend update module corresponding to the vehicle-mounted perception monitoring state vector is determined, including: When the characteristic dimension of the vehicle-mounted perception monitoring state vector is a dynamic characteristic dimension, determining that the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the dynamic characteristic dimension is a temporal attention trajectory trend updating module; When the characteristic dimension of the vehicle-mounted perception monitoring state vector is an event characteristic dimension, it is determined that the spatiotemporal trajectory deduction module corresponding to the vehicle-mounted perception monitoring state vector of the event characteristic dimension is a spatial attention trajectory trend updating module.

8. The method according to claim 1, characterized in that The step of obtaining a vehicle-mounted sensing monitoring status description set of a vehicle cluster to be controlled in a highway network area includes: Acquire original vehicle-mounted perception monitoring state vectors of multiple different feature dimensions of the vehicle cluster to be controlled; Classify the original vehicle-mounted perception monitoring state vector of each feature dimension to obtain a plurality of vehicle-mounted perception monitoring state vectors of different feature dimensions; Based on the statistical rules of the set monitoring state description set, multiple vehicle-borne perception monitoring state vectors of different feature dimensions are mapped to the set monitoring state description set to obtain the vehicle-borne perception monitoring state description set of the vehicle cluster to be controlled.

9. The method according to claim 1, characterized in that The method further comprises: In response to the completion of the spatiotemporal trajectory deduction task of the vehicle-mounted perception monitoring state vector of each characteristic dimension, obtaining a high-speed road condition simulation result corresponding to the vehicle-road spatiotemporal graph neural network; Based on the high-speed road condition simulation result, the vehicle cluster to be controlled is labeled to obtain driving state labeling information of the vehicle cluster to be controlled; Based on the driving status labeling information, a road section diversion operation is performed on the vehicle cluster to be controlled in the expressway network area.

10. An intelligent management and control system, characterized in that: The method comprises at least one processor and a memory; the memory stores computer-executable instructions; the at least one processor executes the computer-executable instructions stored in the memory, so that the at least one processor executes the method according to any one of claims 1 to 9.

Images