A vehicle-road cooperation control method and device of an intelligent vehicle

Abstract

The application discloses a vehicle-road cooperation control method and device of an intelligent vehicle, and relates to the field of artificial intelligence. The method comprises the following steps: acquiring attribute data of each intelligent vehicle on a current road and environment data of the current road; then obtaining initial global cooperation strategies and vehicle state update data in an ABM modeling manner; determining a traffic flow prediction result by a traffic prediction model according to the vehicle state update data and the environment data, adjusting the initial global cooperation strategies based on the traffic flow prediction result, and obtaining target global cooperation strategies; and performing coordination control on the intelligent vehicles on the current road based on the target global cooperation strategies. In the application, the global cooperation strategies are determined in the ABM and ST-GCN combined manner, the complex interaction of different intelligent level vehicles in the mixed traffic flow can be processed, accurate control instructions and prediction results can be output, and therefore the traffic efficiency and safety are improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and in particular to a vehicle-road cooperative control method and device for intelligent vehicles. Background Technology

[0002] With the rapid development of autonomous driving technology, intelligent connected vehicles, as a core component of future intelligent transportation systems, are gradually reshaping transportation patterns. According to autonomous driving classification standards, from Level 1 assisted driving to Level 5 fully autonomous driving, vehicles of different intelligence levels exhibit significant performance differences in environmental perception, decision-making and planning, and execution control. However, in mixed traffic environments, the differences in perception capabilities, decision-making speed, and execution accuracy among vehicles of different intelligence levels lead to a decrease in the overall efficiency of cooperative control in mixed traffic scenarios. Furthermore, when dealing with sudden traffic events, complex road topologies, and highly dynamic traffic flows, they often exhibit lower safety and insufficient adaptability.

[0003] Therefore, there is an urgent need for a vehicle-road cooperative control method that can adapt to mixed-traffic scenarios with multiple levels of intelligence, so as to promote the large-scale application of intelligent connected vehicles and the further development of intelligent transportation systems. Summary of the Invention

[0004] The purpose of this application is to address at least one of the aforementioned technical deficiencies.

[0005] On one hand, embodiments of this application provide a vehicle-road cooperative control method for intelligent vehicles, the method comprising:

[0006] Acquire attribute data for each intelligent vehicle on the current road and environmental data for the current road;

[0007] Based on the attribute data of each intelligent vehicle and the current road environment data, the initial global coordination strategy and vehicle status update data are obtained through ABM modeling.

[0008] Based on vehicle status update data and environmental data, traffic flow prediction results are determined through a traffic prediction model, and the initial global coordination strategy is adjusted based on the traffic flow prediction results to obtain the target global coordination strategy.

[0009] The system coordinates and controls intelligent vehicles on the current road based on a global collaborative strategy.

[0010] Optionally, vehicle status update data includes vehicle update trajectory and interaction event data. Based on the attribute data of each intelligent vehicle and the current road environment data, an initial global collaborative strategy and vehicle status update data are obtained through ABM modeling, including:

[0011] Acquire behavioral decision-making models, kinematic models, and interaction rules;

[0012] Based on the attribute data, environmental data, behavioral decision-making model, and kinematic model of each intelligent vehicle, the vehicle update trajectory corresponding to each intelligent vehicle is obtained.

[0013] Neighborhood detection is performed based on the attribute data and environmental data of each intelligent vehicle to obtain neighborhood detection results;

[0014] Based on the neighborhood detection results and interaction rules, interaction event data is determined, and an initial global coordination strategy is obtained based on the vehicle update status and interaction event data.

[0015] Optionally, based on the attribute data, environmental data, behavioral decision-making model, and kinematic model of each intelligent vehicle, the vehicle update trajectory corresponding to each intelligent vehicle is obtained, including:

[0016] The attribute data and environmental data of each intelligent vehicle are input into the behavior decision model to obtain the vehicle behavior decision results for each intelligent vehicle.

[0017] Based on the attribute data and kinematic model of each intelligent vehicle, the initial vehicle state corresponding to each intelligent vehicle is obtained;

[0018] For each intelligent vehicle, the vehicle update trajectory is obtained based on the vehicle behavior decision results and the initial vehicle state.

[0019] Optionally, attribute data includes vehicle intelligence level, performance parameters, and target route; environmental data includes road topology, traffic signal status, obstacle location, and weather conditions; vehicle update trajectory includes location, speed, acceleration, and driving direction; and interaction event data includes vehicle interaction events and environmental response data. Vehicle interaction events include lane change conflict counts, following distance warnings, and collision risk predictions. Environmental response data includes feedback on traffic light changes and obstacle avoidance trajectories.

[0020] Optionally, the traffic prediction model is an ST-GCN network model. The ST-GCN network model includes an input layer, a spatiotemporal convolutional layer, a global pooling layer, and an output layer. Based on vehicle state update data and environmental data, the traffic prediction model determines the traffic flow prediction results, including:

[0021] Vehicle status update data and environmental data are input into the input layer to obtain spatiotemporal graph data. The spatiotemporal graph data includes an adaptive adjacency matrix and a node feature matrix. The adaptive adjacency matrix is ​​determined based on the vehicle status update data and the road topology, and the node feature matrix is ​​determined based on the vehicle status update data.

[0022] The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, and the high-level spatiotemporal node features are then input into the global pooling layer for mean pooling to obtain pooled high-level spatiotemporal node features.

[0023] The pooled high-level spatiotemporal node features are input into the output layer to obtain traffic flow prediction results.

[0024] Optionally, the spatiotemporal convolutional layer includes at least one sequentially connected spatiotemporal convolutional module. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, including:

[0025] The spatiotemporal graph data is input into the spatial graph convolution module for spatial information aggregation processing to obtain the updated node feature matrix.

[0026] The updated node feature matrix is ​​input into the temporal convolution module for temporal convolution processing to obtain the temporally convolutioned node feature matrix.

[0027] The node feature matrix after temporal convolution is input into the residual normalization module for residual and normalization processing to obtain high-level spatiotemporal node features.

[0028] Optionally, the traffic prediction model is trained in the following ways:

[0029] Obtain at least one training sample, the labeled prediction results corresponding to each training sample, and the initial ST-GCN network model. The labeled prediction results serve as the traffic flow prediction result identifiers.

[0030] The initial ST-GCN network model is trained based on each training sample until the loss function corresponding to the initial ST-GCN network model converges. The ST-GCN network model at the end of training is then used as the traffic prediction model.

[0031] The traffic prediction model takes training samples as input and outputs the prediction results of the training samples. The value of the loss function represents the difference between the prediction results of each training sample output by the model and the labeled prediction results corresponding to each training sample.

[0032] Optionally, the loss function can be expressed by the following formula:

[0033]

[0034] in, Indicates regression loss, Indicates classification loss, Indicates time constraints, Indicates spatial constraints. Indicates traffic rule penalties, λ represents the physical feasibility constraint. reg λ represents the weights of the regression loss. cls The weights, λ, represent the classification loss. time The weights representing time constraints, λ space The weights representing spatial constraints, λ rule Indicates the weight of traffic rule penalties, λ physics The weights represent the physical feasibility constraints.

[0035] Optionally, the vehicle intelligence level includes a first level and a second level. The target global cooperative strategy includes control commands corresponding to each intelligent vehicle on the current road. The control commands include at least one of acceleration commands, lane change commands, driving direction commands, vehicle avoidance commands, and traffic signal indication commands. Based on the target global cooperative strategy, coordinated control is performed on the intelligent vehicles on the current road, including:

[0036] For Level 1 intelligent vehicles, corresponding control commands are sent to each Level 1 intelligent vehicle through conventional methods, including at least one of in-vehicle navigation systems and ADAS.

[0037] For Level 2 intelligent vehicles, corresponding control commands are sent to each Level 2 intelligent vehicle via the vehicle-to-everything (V2X) network.

[0038] On the other hand, embodiments of this application provide a vehicle-road cooperative control device for intelligent vehicles, including:

[0039] The data acquisition module is used to acquire attribute data of each intelligent vehicle on the current road and environmental data of the current road.

[0040] The update data determination module is used to obtain the initial global collaborative strategy and vehicle status update data based on the attribute data of each intelligent vehicle and the current road environment data through ABM modeling.

[0041] The collaborative strategy determination module is used to determine the traffic flow prediction results through a traffic prediction model based on vehicle status update data and environmental data, and to adjust the initial global collaborative strategy based on the traffic flow prediction results to obtain the target global collaborative strategy.

[0042] The intelligent vehicle control module is used to coordinate and control intelligent vehicles on the current road based on a target global collaborative strategy.

[0043] In another aspect, embodiments of this application provide an electronic device, including a processor and a memory:

[0044] The memory is configured to store machine-readable instructions that, when executed by the processor, cause the processor to perform any of the methods in a vehicle-to-infrastructure (V2I) control method for an intelligent vehicle.

[0045] In another aspect, embodiments of this application provide a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement any of the methods in the vehicle-road cooperative control method for intelligent vehicles.

[0046] The beneficial effects of the technical solutions provided in this application include at least the following:

[0047] In this application, after obtaining the attribute data and environmental data of each intelligent vehicle on the current road, the updated vehicle trajectory and initial global coordination strategy can be obtained based on agent modeling. The updated vehicle trajectory and initial global coordination strategy obtained at this time can dynamically capture the complex interactions between vehicles and between vehicles and the environment, and can adapt to complex behaviors. It can ensure that the obtained global coordination strategy reflects the real driving behavior and adapt to the dynamic traffic environment.

[0048] Furthermore, traffic flow prediction results can be obtained through the ST-GCN network model based on vehicle status update data and environmental data. Compared with the traffic flow prediction results based on the traditional GCN+LSTM model, this model can overcome the shortcomings of spatiotemporal fragmentation, computational redundancy, and long-term dependence in GCN+LSTM. It is also significantly better than the separate model in terms of prediction accuracy, real-time performance, and resource efficiency. Especially in mixed traffic flow scenarios, it can more accurately capture the complexity of spatiotemporal interaction between vehicles, further improving the accuracy of prediction results.

[0049] Furthermore, the initial global coordination strategy is adjusted based on the obtained traffic flow prediction results to obtain a target global coordination strategy. Then, the intelligent vehicles on the current road are coordinated and controlled based on the target global coordination strategy. In other words, this application achieves closed-loop optimization from micro-behavior to macro-control through the collaboration of ABM and ST-GCN. This enables the handling of complex interactions between vehicles of different intelligence levels in mixed traffic flows, outputting accurate control commands and prediction results, thereby improving traffic efficiency and safety. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart illustrating the vehicle-road cooperative control method for intelligent vehicles provided in this application embodiment;

[0052] Figure 2 A schematic diagram of the ST-GCN network model provided in the embodiments of this application;

[0053] Figure 3 A schematic diagram of the structure of the vehicle-road cooperative control device for intelligent vehicles provided in this application embodiment;

[0054] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0055] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting the invention.

[0056] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0058] It is understood that the executing entity of the method provided in this application embodiment can be set according to actual needs, and this application embodiment does not limit it. For example, the central management platform, roadside unit, cloud computing service provider, third-party collaborative platform and vehicle control terminal in intelligent vehicle network can all execute the method provided in this application embodiment.

[0059] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0060] Specifically, such as Figure 1 As shown, the method may include:

[0061] Step S101: Obtain the attribute data of each intelligent vehicle on the current road and the environmental data of the current road.

[0062] In optional embodiments of this application, the attribute data of each intelligent vehicle includes the vehicle intelligence level, performance parameters and target route, and the environmental data includes road topology, traffic signal status, obstacle location and weather conditions.

[0063] Among them, performance parameters may include vehicle position coordinates, real-time speed, etc.; road topology refers to the way the road is constructed, specifically the lanes, intersections, etc. included in the road; traffic signal status refers to the status and change time of traffic lights, etc.; and weather conditions refer to the current real-time weather conditions of the road, such as rain, snow, fog, etc.

[0064] Optionally, the method for obtaining attribute data of each intelligent vehicle and environmental data of the current road can be set according to the actual situation. This application embodiment does not limit this. For example, it can be obtained through vehicle sensors (such as cameras, radar, etc.), roadside units (RSU), and traffic lights.

[0065] Step S102: Based on the attribute data of each intelligent vehicle and the current road environment data, the initial global collaborative strategy and vehicle state update data are obtained through ABM modeling.

[0066] Among them, ABM modeling refers to the precise modeling of micro-level individual behaviors (such as the optimization decisions of autonomous vehicles and the random disturbances of traditional vehicles), dynamically capturing complex interactions between vehicles and between vehicles and the environment, supporting hierarchical collaborative decision-making among multiple agents, and flexibly adapting to heterogeneous traffic scenarios. At the same time, it supports low-risk verification of traffic policies and achieves high interpretability, becoming a core tool connecting micro-level behaviors and macro-level control, and providing a theoretical basis and technical support for the design and optimization of intelligent connected and human-driven hybrid systems.

[0067] Optionally, for the acquired attribute information and environmental data of each intelligent vehicle, road traffic simulation can be performed based on ABM modeling to obtain the initial global collaborative strategy and vehicle state update data.

[0068] In optional embodiments of this application, the vehicle state update data includes vehicle update trajectory and interaction event data. Based on the attribute data of each intelligent vehicle and the current road environment data, an initial global collaborative strategy and vehicle state update data are obtained through ABM modeling, including:

[0069] Acquire behavioral decision-making models, kinematic models, and interaction rules;

[0070] Based on the attribute data, environmental data, behavioral decision-making model, and kinematic model of each intelligent vehicle, the vehicle update trajectory corresponding to each intelligent vehicle is obtained.

[0071] Neighborhood detection is performed based on the attribute data and environmental data of each intelligent vehicle to obtain neighborhood detection results;

[0072] Based on the neighborhood detection results and interaction rules, interaction event data is determined, and an initial global coordination strategy is obtained based on the vehicle update status and interaction event data.

[0073] Optionally, in this embodiment of the application, the vehicle status update data may specifically include two types: vehicle update trajectory and interaction event data. The vehicle update trajectory may include position, speed, acceleration, driving direction, etc., while the interaction event data may include vehicle interaction events and environmental response data. Among them, vehicle interaction events include lane change conflict times, following distance warnings, collision risk predictions, etc., and environmental response data may include feedback on traffic light changes, obstacle avoidance trajectories, etc.

[0074] In this application, a behavior decision model and a kinematic model can be pre-set, and then the attribute data and environmental data of each intelligent vehicle can be input into the behavior decision model and the kinematic model to obtain the vehicle update trajectory corresponding to each intelligent vehicle, that is, to obtain the updated position, speed, acceleration and driving direction of each intelligent vehicle in the current road.

[0075] Furthermore, neighborhood detection can be performed based on the attribute data and environmental data of each intelligent vehicle to obtain neighborhood detection results. These results specifically refer to the other traffic participants (such as vehicles, pedestrians, etc.) and environmental elements around each intelligent vehicle that may influence its behavior, obtained after dynamic identification. These neighborhood detection results can be represented as a list of nearby vehicles and a list of pedestrians / obstacles. The list of nearby vehicles can include ID, location, speed, acceleration, vehicle type, etc., while the list of pedestrians / obstacles can include location, direction of movement, and speed.

[0076] For example, dynamic recognition and detection are performed centered on the current intelligent vehicle, with all traffic participants within a 50-meter range, as well as vehicles in the same or adjacent lanes, considered as neighborhood detection results. Quadtrees or grid partitioning can be used to accelerate neighborhood queries during detection. For autonomous vehicles, LiDAR point clouds can be used to simulate obstacle detection, and camera views can be used to simulate pedestrian recognition. It is understood that the neighborhood detection method provided in this application can be selected according to actual circumstances, and this application does not limit this approach.

[0077] Correspondingly, based on the neighborhood detection results and pre-set interaction rules, interaction rules can be applied to obtain conflict events in the interaction event data. Then, based on the obtained vehicle update status, conflict events can be resolved to obtain an initial global coordination strategy (such as vehicle H changing lanes and updating its position to the left lane). The pre-set interaction rules can be set according to actual needs, and may include speed limits, yield rules (such as stop signs, pedestrian priority, vehicle priority (such as ambulances having higher priority than ordinary vehicles), and intersection passage order (such as turning vehicles yielding to straight-going vehicles).

[0078] In optional embodiments of this application, based on the attribute data, environmental data, behavioral decision model, and kinematic model of each intelligent vehicle, the vehicle update trajectory corresponding to each intelligent vehicle is obtained, including:

[0079] The attribute data and environmental data of each intelligent vehicle are input into the behavior decision model to obtain the vehicle behavior decision results for each intelligent vehicle.

[0080] Based on the attribute data and kinematic model of each intelligent vehicle, the initial vehicle state corresponding to each intelligent vehicle is obtained;

[0081] For each intelligent vehicle, the vehicle update trajectory is obtained based on the vehicle behavior decision results and the initial vehicle state.

[0082] Optionally, after acquiring the attribute data and environmental data of each intelligent vehicle, the kinematic model and behavioral decision model can be triggered. At this point, based on the attribute data (such as position, speed, and acceleration) of each intelligent vehicle, real-time state calculations can be performed using the kinematic model to obtain the initial vehicle state. Simultaneously, based on the behavioral decision model, the vehicle behavior decision results (such as adjusting acceleration and steering angle) for each intelligent vehicle are obtained using the attribute data and environmental data. Furthermore, based on the vehicle behavior decision results corresponding to each intelligent vehicle, the initial vehicle state of that intelligent vehicle is adjusted to obtain the updated vehicle trajectory.

[0083] The kinematic model described above supports multi-time-step iterative calculations to ensure the continuity of state updates. Real-time state calculations include vehicle position updates and vehicle speed updates, where the vehicle position update is determined by the following formula:

[0084]

[0085] Where, x t+1 Let x be the position at time t+1. t Let v be the position at time t. t Let be the velocity at time t, a be the acceleration, and Δt be the time difference between time t+1 and time t.

[0086] The vehicle speed update is determined by the following formula:

[0087] v t+1 =v t +a*Δt

[0088] Among them, v t+1 Let v be the velocity at time t+1. t Let be the velocity at time t, a be the acceleration, and Δt be the time difference between time t+1 and time t.

[0089] Optionally, the behavior decision model can update the steering angle based on the path tracking algorithm, define the vehicle following behavior of the autonomous vehicle through the intelligent driving model (IDM), and introduce random disturbance factors to simulate the uncertainty of driving behavior (equivalent to vehicle following behavior) for human-driven vehicles. Then, lane-changing decisions are triggered based on the MOBIL model and the obtained vehicle following behavior.

[0090] In this application, the updated vehicle trajectory and initial global coordination strategy obtained by agent-based modeling can dynamically capture complex interactions between vehicles and between vehicles and the environment (such as conflict resolution, cooperative formation, etc.), and specifically balances efficiency and flexibility, can adapt to complex behaviors, and flexibly adapt to heterogeneous traffic scenarios, thereby ensuring that the obtained global coordination strategy can reflect real driving behavior and adapt to dynamic traffic environment.

[0091] Step S103: Based on vehicle status update data and environmental data, determine the traffic flow prediction result through the traffic prediction model, and adjust the initial global collaborative strategy based on the traffic flow prediction result to obtain the target global collaborative strategy.

[0092] Optionally, for the obtained vehicle status update data, the vehicle status update data and the previously acquired environmental data can be used as inputs to the traffic prediction model to obtain traffic flow prediction results. Then, the initial global coordination strategy is adjusted based on the traffic flow prediction results to obtain the target global coordination strategy, thereby ensuring that the final target global coordination strategy is more adapted to the current road traffic environment and further improves traffic safety.

[0093] In an optional embodiment of this application, the traffic prediction model is an ST-GCN (Spatio-Temporal Graph Neural Network) network model. The ST-GCN network model includes an input layer, a spatio-temporal convolutional layer, a global pooling layer, and an output layer. Based on vehicle state update data and environmental data, the traffic prediction model determines the traffic flow prediction result, including:

[0094] Vehicle status update data and environmental data are input into the input layer to obtain spatiotemporal graph data. The spatiotemporal graph data includes an adaptive adjacency matrix and a node feature matrix. The adaptive adjacency matrix is ​​determined based on the vehicle status update data and environmental data, and the node feature matrix is ​​determined based on the vehicle status update data.

[0095] The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, and the high-level spatiotemporal node features are then input into the global pooling layer for mean pooling to obtain pooled high-level spatiotemporal node features.

[0096] The pooled high-level spatiotemporal node features are input into the output layer to obtain traffic flow prediction results.

[0097] Optionally, the traffic prediction model can be an ST-GCN network model, which may include an input layer, a spatiotemporal convolutional layer, a global pooling layer, and an output layer connected in sequence. Accordingly, vehicle state update data and environmental data can be input into the input layer to obtain spatiotemporal graph data. This spatiotemporal graph data is a constructed graph structure, which may include an adaptive adjacency matrix (equivalent to edges in the graph structure) and a node feature matrix (equivalent to nodes in the graph structure).

[0098] The node feature matrix is ​​determined based on vehicle state update data and is used to characterize vehicles; it is typically represented as X∈R. T*N*D(T represents the time step, N represents the number of nodes (e.g., vehicles, road key points, etc.), and D represents the feature dimension (e.g., position, speed, acceleration, etc.)). The adaptive adjacency matrix represents the distance or interaction strength between vehicles (i.e., the relationship between vehicles). It can be dynamically adjusted by learning parameters to change the node relationship weights, thus better matching the current road traffic environment. In this embodiment, the adaptive adjacency matrix is ​​obtained by fusing the global graph structure attributes obtained from vehicle state update data and the graph structure attributes obtained from road topology relationships in environmental data. Specifically, it is used to represent the spatial relationships between nodes (e.g., distance between vehicles, road connectivity), and can typically be represented as A∈R. N*D (N represents the number of nodes (such as vehicles, key points on roads, etc.), and D represents the feature dimension (such as position, speed, acceleration, etc.)).

[0099] In this application, since the graph structure attributes obtained based on road topology are incorporated when determining the adaptive adjacency matrix, the resulting graph structure can not only capture the dynamic interactions between vehicles, but also embed the structured knowledge of the road network, thereby achieving higher accuracy and reliability in traffic prediction, path planning and cooperative control tasks.

[0100] Furthermore, the obtained spatiotemporal map data is input into a spatiotemporal convolutional layer to obtain high-level spatiotemporal node features. These features explicitly model the spatial relationships between nodes and capture dynamic evolution (such as speed change trends). Correspondingly, these high-level spatiotemporal node features are input into a global pooling layer for feature aggregation along the time or node dimension, ensuring feature accuracy. Finally, the pooled high-level spatiotemporal node features are input into the output layer to obtain the traffic flow prediction results.

[0101] In this embodiment, traffic flow is predicted using the ST-GCN network model. Compared to the traditional GCN (Graph Convolutional Network) + LSTM (Long Short-Term Memory) model, it overcomes the shortcomings of spatiotemporal fragmentation, computational redundancy, and long-term dependence in GCN + LSTM. It is also significantly better than the separate model in terms of prediction accuracy, real-time performance, and resource efficiency. Especially in mixed traffic flow scenarios, it can more accurately capture the complexity of spatiotemporal interactions between vehicles, further improving the accuracy of traffic flow prediction results.

[0102] In an optional embodiment of this application, the spatiotemporal convolutional layer includes at least one spatiotemporal convolutional module connected in sequence. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, including:

[0103] The spatiotemporal graph data is input into the spatial graph convolution module for spatial information aggregation processing to obtain the updated node feature matrix.

[0104] The updated node feature matrix is ​​input into the temporal convolution module for temporal convolution processing to obtain the temporally convolutioned node feature matrix.

[0105] The node feature matrix after temporal convolution is input into the residual normalization module for residual and normalization processing to obtain high-level spatiotemporal node features.

[0106] Optionally, the spatiotemporal convolutional layer may include at least one spatiotemporal convolutional module connected and stacked sequentially. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. Accordingly, the obtained spatiotemporal graph data is input into the spatial graph convolutional module for spatial information aggregation processing to obtain an updated node feature matrix. Then, the updated node feature matrix is ​​input into the temporal convolutional module for temporal convolution processing. The node feature matrix obtained at this time can capture local patterns and long-term dependencies in the temporal dimension. The node feature matrix after temporal convolution is input into the residual normalization module for residual and normalization processing, thereby avoiding gradient vanishing and improving the training stability of deep networks.

[0107] In an optional embodiment of this application, the traffic prediction model is trained in the following manner:

[0108] Obtain at least one training sample, the labeled prediction results corresponding to each training sample, and the initial ST-GCN network model. The labeled prediction results serve as the traffic flow prediction result identifiers.

[0109] The initial ST-GCN network model is trained based on each training sample until the loss function corresponding to the initial ST-GCN network model converges. The ST-GCN network model at the end of training is then used as the traffic prediction model.

[0110] The traffic prediction model takes training samples as input and outputs the prediction results of the training samples. The value of the loss function represents the difference between the prediction results of each training sample output by the model and the labeled prediction results corresponding to each training sample.

[0111] Optionally, this application may obtain at least one training sample and the labeled prediction result corresponding to each training sample, wherein the training sample is vehicle state update data and environmental data, and the corresponding labeled prediction result is the traffic flow prediction result corresponding to the input training sample.

[0112] Furthermore, each training sample can be input into the initial ST-GCN network model. At this time, the initial ST-GCN network model can output the traffic flow prediction result corresponding to each training sample. Then, based on the traffic flow prediction result of each training sample and the labeled prediction result corresponding to each training sample, the value of the loss function corresponding to the model can be determined. If the determined loss function value does not converge, the network parameters of the initial ST-GCN network model can be adjusted. Then, each training sample can be input into the adjusted initial ST-GCN network model again, and the value of the loss function can be determined again based on the output traffic flow prediction result and the labeled prediction result corresponding to the training sample. If it still does not converge, the network parameters of the initial ST-GCN network model can be adjusted again until the value of the corresponding loss function converges. The loss function of the model represents the difference between the traffic flow prediction results of each training sample and the labeled prediction results of each training sample. When the loss function converges, it means that the difference between the traffic flow prediction results of each training sample and the labeled prediction results of each training sample meets the requirements, that is, the traffic flow prediction results of the training samples output by the model are close to the labeled prediction results of the training samples.

[0113] Furthermore, when the loss function converges, a test sample set can be obtained to further test and verify the initial fault monitoring model. If the test and verification results corresponding to the trained ST-GCN network model do not meet the set verification requirements, the step of training the initial ST-GCN network model based on the training sample set is repeated until the test and verification results corresponding to the initial ST-GCN network model meet the set verification requirements.

[0114] In optional embodiments of this application, the loss function can be represented as follows:

[0115]

[0116] in, Indicates regression loss, Indicates classification loss, Indicates time constraints, Indicates spatial constraints. Indicates traffic rule penalties, λ represents the physical feasibility constraint. reg λ represents the weights of the regression loss. cls The weights, λ, represent the classification loss. time The weights representing time constraints, λ space The weights representing spatial constraints, λ rule Indicates the weight of traffic rule penalties, λ physics The weights represent the physical feasibility constraints.

[0117] Optionally, the ST-GCN network model used for training in this application consists of regression loss, classification loss, time constraints, traffic rule penalties, and physical feasibility constraints. The regression loss is used to constrain trajectory prediction and velocity regression, and can take the following form:

[0118]

[0119] Classification loss is used to address class imbalance in behavior classification, and can take the following forms:

[0120]

[0121] Time constraints are used to ensure time smoothness, and the possible forms are:

[0122]

[0123] Spatial constraints are used to ensure spatial consistency, and the possible forms are:

[0124]

[0125] Traffic rule penalties and physical feasibility constraints are used to strictly enforce physical constraints to avoid exceeding limits on acceleration, curvature, etc. The forms of traffic rule penalties that can be selected are:

[0126]

[0127] Π violate Penalty is a violation indicator function, and Penalty is a high constant.

[0128] The physical feasibility constraints can take the following forms:

[0129]

[0130] The loss function in this application is modularly designed and can be flexibly adapted to various tasks such as traffic prediction, behavior recognition, and path planning, while meeting the requirements of stability, security, and interpretability.

[0131] To better understand the ST-GCN network model provided in this application, such as Figure 2 As shown in the diagram, this application provides a schematic diagram of the ST-GCN network model. The ST-GCN network model includes an input layer, a spatiotemporal convolutional layer, a global pooling layer, and an output layer connected in sequence. The spatiotemporal convolutional layer includes at least one spatiotemporal convolutional module connected and stacked in sequence. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. Figure 2 The example shown is a spatiotemporal convolution module.

[0132] Step S104: Coordinate and control the intelligent vehicles on the current road based on the target global collaborative strategy.

[0133] In optional embodiments of this application, the vehicle intelligence level includes a first level and a second level, and the target global cooperative strategy includes control instructions corresponding to each intelligent vehicle on the current road. The control instructions include at least one of acceleration instructions, lane change instructions, driving direction instructions, vehicle avoidance instructions, and traffic signal indication instructions. Coordinated control of the intelligent vehicles on the current road based on the target global cooperative strategy includes:

[0134] For Level 1 intelligent vehicles, corresponding control commands are sent to each Level 1 intelligent vehicle through conventional methods, including at least one of in-vehicle navigation systems and ADAS.

[0135] For Level 2 intelligent vehicles, corresponding control commands are sent to each Level 2 intelligent vehicle via the vehicle-to-everything (V2X) network.

[0136] Optionally, based on the level of vehicle intelligence, intelligent vehicles can be divided into low-intelligence vehicles (i.e., Level 1 in this application) and high-intelligence vehicles (i.e., Level 2 in this application). Low-intelligence vehicles (L0-L2) typically refer to vehicles with relatively low autonomous driving capabilities, whose behavior is mainly driven by traditional car-following models or rules. High-intelligence vehicles refer to vehicles with relatively high autonomous driving capabilities, typically possessing an intelligence level from Level 3 (conditional autonomous driving) to Level 5 (fully autonomous driving).

[0137] Correspondingly, the coordination and control methods differ depending on the level of vehicle intelligence. For Level 1 intelligent vehicles, corresponding control commands can be sent to them through traditional methods, such as using traditional methods including in-vehicle navigation systems and ADAS for communication. For Level 2 intelligent vehicles, control commands can be sent through vehicle-to-everything (V2V) or vehicle-to-infrastructure (V2I) communication.

[0138] It is understandable that, in practical applications, to better implement the methods provided in the embodiments of this application, the vehicle-mounted terminal needs to integrate a communication module (5G / V2X), sensors, and edge computing chips, while the roadside unit (RSU) needs to be equipped with an AI processor, traffic signal controller, and environmental perception equipment, and global model training and large-scale simulation should be performed through a central server. In addition, a data repository needs to be configured to store the parameters of the trained traffic prediction model, the ABM vehicle behavior rule library, and the historical traffic event database, and distributed cloud storage + edge caching should be used to support rapid access and updates.

[0139] This application provides a vehicle-road cooperative control device for intelligent vehicles, such as... Figure 3 As shown, the device may include: a data acquisition module 301, an update data determination module 302, a cooperative strategy determination module 303, and an intelligent vehicle control module 304, wherein...

[0140] The data acquisition module is used to acquire attribute data of each intelligent vehicle on the current road and environmental data of the current road.

[0141] The update data determination module is used to obtain the initial global collaborative strategy and vehicle status update data based on the attribute data of each intelligent vehicle and the current road environment data through ABM modeling.

[0142] The collaborative strategy determination module is used to determine the traffic flow prediction results through a traffic prediction model based on vehicle status update data and environmental data, and to adjust the initial global collaborative strategy based on the traffic flow prediction results to obtain the target global collaborative strategy.

[0143] The intelligent vehicle control module is used to coordinate and control intelligent vehicles on the current road based on a target global collaborative strategy.

[0144] Optionally, vehicle status update data includes vehicle update trajectory and interaction event data. When the update data determination module obtains the initial global collaborative strategy and vehicle status update data based on the attribute data of each intelligent vehicle and the current road environment data through ABM modeling, it is specifically used for:

[0145] Acquire behavioral decision-making models, kinematic models, and interaction rules;

[0146] Based on the attribute data, environmental data, behavioral decision-making model, and kinematic model of each intelligent vehicle, the vehicle update trajectory corresponding to each intelligent vehicle is obtained.

[0147] Neighborhood detection is performed based on the attribute data and environmental data of each intelligent vehicle to obtain neighborhood detection results;

[0148] Based on the neighborhood detection results and interaction rules, interaction event data is determined, and an initial global coordination strategy is obtained based on the vehicle update status and interaction event data.

[0149] Optionally, when the update data determination module obtains the vehicle update trajectory for each intelligent vehicle based on its attribute data, environmental data, behavioral decision model, and kinematic model, it is specifically used for:

[0150] The attribute data and environmental data of each intelligent vehicle are input into the behavior decision model to obtain the vehicle behavior decision results for each intelligent vehicle.

[0151] Based on the attribute data and kinematic model of each intelligent vehicle, the initial vehicle state corresponding to each intelligent vehicle is obtained;

[0152] For each intelligent vehicle, the vehicle update trajectory is obtained based on the vehicle behavior decision results and the initial vehicle state.

[0153] Optionally, attribute data includes vehicle intelligence level, performance parameters, and target route; environmental data includes road topology, traffic signal status, obstacle location, and weather conditions; vehicle update trajectory includes location, speed, acceleration, and driving direction; and interaction event data includes vehicle interaction events and environmental response data. Vehicle interaction events include lane change conflict counts, following distance warnings, and collision risk predictions. Environmental response data includes feedback on traffic light changes and obstacle avoidance trajectories.

[0154] Optionally, the traffic prediction model is an ST-GCN network model, which includes an input layer, a spatiotemporal convolutional layer, a global pooling layer, and an output layer. The collaborative strategy determination module, when determining traffic flow prediction results based on vehicle state update data and environmental data using the traffic prediction model, is specifically used for:

[0155] Vehicle status update data and environmental data are input into the input layer to obtain spatiotemporal graph data. The spatiotemporal graph data includes an adaptive adjacency matrix and a node feature matrix. The adaptive adjacency matrix is ​​determined based on the vehicle status update data and the road topology, and the node feature matrix is ​​determined based on the vehicle status update data.

[0156] The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, and the high-level spatiotemporal node features are then input into the global pooling layer for mean pooling to obtain pooled high-level spatiotemporal node features.

[0157] The pooled high-level spatiotemporal node features are input into the output layer to obtain traffic flow prediction results.

[0158] Optionally, the spatiotemporal convolutional layer includes at least one sequentially connected spatiotemporal convolutional module. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. The collaborative policy determination module, when inputting spatiotemporal graph data into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, is specifically used for:

[0159] The spatiotemporal graph data is input into the spatial graph convolution module for spatial information aggregation processing to obtain the updated node feature matrix.

[0160] The updated node feature matrix is ​​input into the temporal convolution module for temporal convolution processing to obtain the temporally convolutioned node feature matrix.

[0161] The node feature matrix after temporal convolution is input into the residual normalization module for residual and normalization processing to obtain high-level spatiotemporal node features.

[0162] Optionally, the traffic prediction model is trained in the following ways:

[0163] Obtain at least one training sample, the labeled prediction results corresponding to each training sample, and the initial ST-GCN network model. The labeled prediction results serve as the traffic flow prediction result identifiers.

[0164] The initial ST-GCN network model is trained based on each training sample until the loss function corresponding to the initial ST-GCN network model converges. The ST-GCN network model at the end of training is then used as the traffic prediction model.

[0165] The traffic prediction model takes training samples as input and outputs the prediction results of the training samples. The value of the loss function represents the difference between the prediction results of each training sample output by the model and the labeled prediction results corresponding to each training sample.

[0166] Optionally, the loss function can be expressed as follows:

[0167]

[0168] in, Indicates regression loss, Indicates classification loss, Indicates time constraints, Indicates spatial constraints. Indicates traffic rule penalties, λ represents the physical feasibility constraint. reg λ represents the weights of the regression loss. cls The weights, λ, represent the classification loss. time The weights representing time constraints, λ space The weights representing spatial constraints, λ rule Indicates the weight of traffic rule penalties, λ physics The weights represent the physical feasibility constraints.

[0169] Optionally, the vehicle intelligence level includes a first level and a second level. The target global coordination strategy includes control commands corresponding to each intelligent vehicle on the current road. The control commands include at least one of acceleration commands, lane change commands, driving direction commands, vehicle avoidance commands, and traffic signal indication commands. When the intelligent vehicle control module coordinates and controls the intelligent vehicles on the current road based on the target global coordination strategy, it is specifically used for:

[0170] For Level 1 intelligent vehicles, corresponding control commands are sent to each Level 1 intelligent vehicle through conventional methods, including at least one of in-vehicle navigation systems and ADAS.

[0171] For Level 2 intelligent vehicles, corresponding control commands are sent to each Level 2 intelligent vehicle via the vehicle-to-everything (V2X) network.

[0172] The vehicle-road cooperative control device for an intelligent vehicle in this embodiment can execute the vehicle-road cooperative control method for an intelligent vehicle shown in the embodiment of this application. The implementation principle is similar and will not be described again here.

[0173] This application provides a computer-readable storage medium storing at least one instruction, at least one program, code set, or instruction set, wherein the at least one instruction, at least one program, code set, or instruction set is loaded and executed by a processor to implement any of the methods in the vehicle-road cooperative control method for intelligent vehicles.

[0174] This application provides an electronic device, which includes a processor and a memory configured to store machine-readable instructions that, when executed by the processor, cause the processor to perform a vehicle-to-everything (V2X) control method for an intelligent vehicle.

[0175] This application provides an electronic device, such as... Figure 4 As shown, Figure 4 The illustrated electronic device 2000 includes a processor 2001 and a memory 2003. The processor 2001 and the memory 2003 are connected, for example, via a bus 2002. Optionally, the electronic device 2000 may also include a transceiver 2004. It should be noted that in practical applications, the transceiver 2004 is not limited to one type, and the structure of this electronic device 2000 does not constitute a limitation on the embodiments of this application.

[0176] Processor 2001 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 2001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0177] Bus 2002 may include a pathway for transmitting information between the aforementioned components. Bus 2002 may be a PCI bus or an EISA bus, etc. Bus 2002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0178] The memory 2003 may be ROM or other type of static storage device capable of storing static information and instructions, RAM or other type of dynamic storage device capable of storing information and instructions, or EEPROM, CD-ROM or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0179] The memory 2003 stores the application code that executes the scheme of this application, and its execution is controlled by the processor 2001. The processor 2001 executes the application code stored in the memory 2003 to implement... Figure 3 The illustrated embodiment provides the operation of a vehicle-road cooperative control device for an intelligent vehicle.

[0180] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0181] The above description is only a partial embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A vehicle-road cooperative control method for intelligent vehicles, characterized in that, include: Acquire the attribute data of each intelligent vehicle on the current road and the environmental data of the current road; Based on the attribute data of each intelligent vehicle and the environmental data of the current road, the initial global coordination strategy and vehicle status update data are obtained through ABM modeling. Based on the vehicle status update data and the environmental data, the traffic flow prediction result is determined by the traffic prediction model, and the initial global coordination strategy is adjusted based on the traffic flow prediction result to obtain the target global coordination strategy. Based on the target global collaborative strategy, coordinate and control the intelligent vehicles on the current road; The vehicle status update data includes vehicle update trajectory and interaction event data. The initial global coordination strategy and vehicle status update data are obtained through ABM modeling based on the attribute data of each intelligent vehicle and the current road environment data, including: Acquire behavioral decision-making models, kinematic models, and interaction rules; Based on the attribute data of each intelligent vehicle, the environmental data, the behavioral decision model, and the kinematic model, the vehicle update trajectory corresponding to each intelligent vehicle is obtained; Neighborhood detection is performed based on the attribute data of each intelligent vehicle and the environmental data to obtain neighborhood detection results; Based on the neighborhood detection results and the interaction rules, the interaction event data is determined, and the initial global coordination strategy is obtained according to the vehicle update status and the interaction event data. The attribute data includes vehicle intelligence level, performance parameters, and target route; the environmental data includes road topology, traffic signal status, obstacle location, and weather conditions; the vehicle updated trajectory includes location, speed, acceleration, and driving direction; the interaction event data includes vehicle interaction events and environmental response data; the vehicle interaction events include lane change conflict count, following distance warning, and collision risk prediction; and the environmental response data includes feedback on traffic light changes and obstacle avoidance trajectory.

2. The method according to claim 1, characterized in that, The step of obtaining the vehicle update trajectory corresponding to each intelligent vehicle based on the attribute data of each intelligent vehicle, the environmental data, the behavioral decision model, and the kinematic model includes: The attribute data and environmental data of each intelligent vehicle are input into the behavior decision model to obtain the vehicle behavior decision results of each intelligent vehicle. Based on the attribute data of each intelligent vehicle and the kinematic model, the initial vehicle state corresponding to each intelligent vehicle is obtained; For each intelligent vehicle, the vehicle update trajectory is obtained based on the vehicle behavior decision result and the initial vehicle state corresponding to the intelligent vehicle.

3. The method according to claim 2, characterized in that, The traffic prediction model is an ST-GCN network model, which includes an input layer, a spatiotemporal convolutional layer, a global pooling layer, and an output layer. The step of determining the traffic flow prediction result based on the vehicle state update data and the environmental data using the traffic prediction model includes: The vehicle state update data and the environmental data are input into the input layer to obtain spatiotemporal graph data. The spatiotemporal graph data includes an adaptive adjacency matrix and a node feature matrix. The adaptive adjacency matrix is ​​determined based on the vehicle state update data and the road topology, and the node feature matrix is ​​determined based on the vehicle state update data. The spatiotemporal graph data is input into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features, and the high-level spatiotemporal node features are input into the global pooling layer for mean pooling to obtain pooled high-level spatiotemporal node features. The pooled high-level spatiotemporal node features are input into the output layer to obtain traffic flow prediction results.

4. The method according to claim 3, characterized in that, The spatiotemporal convolutional layer includes at least one spatiotemporal convolutional module connected in sequence. Each spatiotemporal convolutional module includes a spatial graph convolutional module, a temporal convolutional module, and a residual normalization module. The step of inputting the spatiotemporal graph data into the spatiotemporal convolutional layer to obtain high-level spatiotemporal node features includes: The spatiotemporal graph data is input into the spatial graph convolution module for spatial information aggregation processing to obtain an updated node feature matrix. The updated node feature matrix is ​​input into the temporal convolution module for temporal convolution processing to obtain the temporally convolutioned node feature matrix. The node feature matrix after temporal convolution is input into the residual normalization module for residual and normalization processing to obtain the high-level spatiotemporal node features.

5. The method according to claim 4, characterized in that, The traffic prediction model was trained in the following way: Obtain at least one training sample, the labeled prediction result corresponding to each training sample, and an initial ST-GCN network model, wherein the labeled prediction result is a traffic flow prediction result identifier; The initial ST-GCN network model is trained based on each of the training samples until the loss function corresponding to the initial ST-GCN network model converges. The ST-GCN network model at the end of training is then used as the traffic prediction model. The traffic prediction model takes training samples as input and outputs the prediction results of the training samples as output. The value of the loss function represents the difference between the prediction results of each training sample output by the model and the labeled prediction results corresponding to each training sample.

6. The method according to claim 5, wherein the loss function is expressed by the following formula: = + + + + + in, Indicates regression loss, Indicates classification loss, Indicates time constraints, Indicates spatial constraints. Indicates traffic rule penalties, This represents a physical feasibility constraint. The weights represent the regression loss. The weights representing the classification loss, The weights of the time constraints are represented. The weights representing the spatial constraints, This indicates the weight of the traffic rule penalty. This represents the weight of the physical feasibility constraint.

7. The method according to claim 1, characterized in that, The vehicle intelligence level includes a first level and a second level. The target global coordination strategy includes control instructions corresponding to each intelligent vehicle on the current road. The control instructions include at least one of acceleration instructions, lane change instructions, driving direction instructions, vehicle avoidance instructions, and traffic signal indication instructions. The coordinated control of the intelligent vehicles on the current road based on the target global coordination strategy includes: For intelligent vehicles belonging to the first level, corresponding control commands are sent to each intelligent vehicle belonging to the first level through conventional methods, wherein the conventional methods include at least one of in-vehicle navigation systems and ADAS. For intelligent vehicles belonging to the second level, corresponding control commands are sent to each intelligent vehicle belonging to the second level through the vehicle network.

8. A vehicle-road cooperative control device for intelligent vehicles, used to implement the method according to any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to acquire attribute data of each intelligent vehicle on the current road and environmental data of the current road; The update data determination module is used to obtain the initial global coordination strategy and vehicle status update data by means of ABM modeling based on the attribute data of each intelligent vehicle and the environmental data of the current road. The collaborative strategy determination module is used to determine the traffic flow prediction result through a traffic prediction model based on the vehicle status update data and the environmental data, and to adjust the initial global collaborative strategy based on the traffic flow prediction result to obtain the target global collaborative strategy. The intelligent vehicle control module is used to coordinate and control the intelligent vehicles on the current road based on the target global coordination strategy.

Images