A vehicle cooperative traffic control method and system based on cloud collaboration and road potential field

Abstract

This invention provides a vehicle cooperative traffic control method and system based on cloud collaboration and road potential field, belonging to the field of intelligent transportation technology. The method includes: discretizing the target road network into a directed graph structure; using continuous recursion within a preset time window as a staged prediction mode to obtain a dynamic potential energy map of the road network; performing global path coarse screening and pruning for each vehicle using breadth-first search to generate candidate path maps; evaluating potential energy using depth-first search; and performing local dynamic route replanning under spatiotemporal conflict localization based on the target travel paths of each vehicle to determine a cooperative traffic scheme for targeted deployment and traffic guidance. This solves the technical problems of insufficient utilization of real-time dynamic information of the road network in existing technologies, leading to frequent multi-vehicle traffic conflicts and low path planning efficiency. Through cloud collaboration and road potential field analysis, it achieves dynamic perception of the road network and multi-vehicle path cooperative optimization, improving the overall traffic efficiency.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation technology, specifically to a vehicle cooperative traffic control method and system based on cloud collaboration and road potential field. Background Technology

[0002] In existing traffic route planning and vehicle scheduling technologies, road network information typically relies on static map data or periodically updated historical traffic data for modeling. This makes it difficult to reflect the rapid changes in the actual road operating conditions in a timely manner. Furthermore, the utilization of data from roadside sensing devices and vehicle-to-everything (V2X) networks is fragmented, lacking a unified dynamic modeling and prediction mechanism. This results in an inability to effectively extrapolate and predict traffic conditions within short-term time windows. In addition, existing route planning often calculates based only on the local optimum at the current moment, ignoring the spatiotemporal propagation characteristics of traffic conditions. This makes the planning results difficult to adapt to rapidly changing road network environments, increasing the risk of local congestion and reducing overall traffic efficiency.

[0003] Existing technologies suffer from insufficient utilization of real-time dynamic information of the road network, leading to frequent conflicts among multiple vehicles and low efficiency in route planning. Summary of the Invention

[0004] The purpose of this application is to provide a vehicle cooperative traffic control method and system based on cloud collaboration and road potential field, in order to solve the technical problems of insufficient utilization of real-time dynamic information of road network in the existing technology, which leads to frequent conflicts of multiple vehicles and low efficiency of route planning.

[0005] In view of the above problems, this application provides a vehicle cooperative traffic control method and system based on cloud collaboration and road potential field.

[0006] The first aspect of this application provides a vehicle cooperative traffic control method based on cloud collaboration and road potential field. The method includes: discretizing the target road network into a directed graph structure; receiving real-time reported data through the cloud; using a continuous recursion within a preset time window as a staged prediction mode in the directed graph structure; employing first-order partial differential equations based on edge directions as the prediction method; obtaining a dynamic potential energy map of the road network through potential energy prediction; constructing a path search module based on the dynamic potential energy map of the road network; generating candidate path maps for each vehicle by performing global path coarse screening and pruning using breadth-first search based on the current location and destination of the vehicles; evaluating the potential energy of the candidate path maps using depth-first search; converging the optimal path through alternating iterations to solve for the target traffic path of each vehicle; and performing local dynamic route replanning under spatiotemporal conflict localization based on the target traffic path of each vehicle to determine a cooperative traffic scheme, and providing vehicle-side targeted deployment and traffic guidance through the vehicle network.

[0007] Optionally, in the directed graph structure, nodes represent intersections or key road segments, edges represent connecting roads, and each node is identified by a state feature vector; the cloud accesses roadside radar data and data reported by connected vehicles in real time, initializes the state features of the directed graph structure, performs recursive potential energy trend prediction based on a preset time window, and obtains the predicted potential energy value of the preset time window, wherein the predicted potential energy value is identified by a confidence interval; the predicted potential energy value is superimposed on the directed graph structure to obtain a dynamic potential energy map of the road network.

[0008] Optionally, based on the preset time window, multiple short-term continuous windows are divided; within the first short-term window, the first node is initialized and a first-order partial differential equation is performed along each directed edge to calculate the potential energy propagation rate from the high-density node to the adjacent low-density node, which is used as the first window predicted potential energy. Here, the first node is any node in the directed graph structure, and the first short-term window is the first window segment among the multiple short-term continuous windows; using the first window predicted potential energy as the starting condition, multiple rounds of iterative solutions are performed until the preset time window is covered to obtain the potential energy curve of the first node; the potential energy curve of the first node is used as the first predicted potential energy value and associated with the first node in the directed graph structure.

[0009] Optionally, the state feature vector includes instantaneous features and propagation features, wherein the instantaneous features include the node's real-time vehicle density, average vehicle speed, road length, and risk potential energy value, and the propagation features include the estimated arrival time of the congestion wave, the potential energy along the main propagation direction of each outgoing edge, and the attenuation coefficient of the potential energy with distance.

[0010] Optionally, a first search layer is deployed by using breadth-first search for coarse screening of road network accessibility and hierarchical pruning; a second search layer is deployed by using depth-first search to recursively refine candidate paths; a third search layer is deployed by using collaborative fusion and iteration of depth search and breadth search; the first search layer, the second search layer and the third search layer are cascaded, and the dynamic potential energy map of the road network is embedded in the first search layer to supervise the training of the path search module.

[0011] Optionally, based on the road network dynamic potential energy map, a first node layer is determined with the vehicle's current location as the starting point and the destination as the ending point. The first node layer consists of multiple reachable road network nodes extending outwards from the road network node corresponding to the starting point. Path pruning is performed on each reachable road network node from the starting point to the first node layer by calculating the minimum cost lower bound, resulting in a first-level candidate path backbone. The minimum cost lower bound is calculated using Euclidean distance and free-flow time. Path pruning decisions are then performed on the first-level candidate path backbone to the second node layer, resulting in a second-level candidate backbone node. The second node layer consists of multiple reachable road network nodes extending outwards from each road network node in the first-level candidate path backbone. Through iterative recursion until the destination is reached, hierarchical associations are performed on the obtained first-level candidate path backbone, second-level candidate backbone nodes, and up to N-level candidate backbone nodes to obtain a first candidate path graph.

[0012] Optionally, for the first candidate path map, a local evaluation based on the predicted potential value is performed from the starting point to any road network node in the first-level candidate path backbone to obtain a first evaluation result, wherein the evaluation type is lane change, detour, or waiting; based on the first evaluation result, a deep search instruction or a return instruction is generated, and the search of the first candidate path map is completed through multiple rounds of iteration to construct a first evaluation map; wherein, if it is a deep search instruction, the road network node is updated to the starting point, and any road network node is selected from the associated second-level candidate backbone nodes for evaluation; if it is a return instruction, the process returns to the starting point, and road network nodes are reselected and evaluated from the first-level candidate path backbone.

[0013] Optionally, the first evaluation graph is written into the open set of the first candidate path graph and updated with an expanded priority coefficient; the first candidate path graph is pruned using the expanded priority coefficient to obtain a second candidate path graph; a deep search evaluation based on the predicted potential value is performed on the second candidate path graph to obtain a second evaluation graph; the second evaluation graph is written into the second candidate path graph, and the target passage path is obtained by performing multiple rounds of search alternating iterations of breadth search pruning and depth search evaluation.

[0014] Optionally, the target travel paths of each vehicle are obtained, and a spatiotemporal conflict map is constructed. The spatiotemporal conflict map defines conflicts based on the spatiotemporal overlap of the same node and the same side. Based on the spatiotemporal conflict map, local dynamic route replanning is performed to obtain a cooperative travel scheme, wherein the replanning route is guided by the road network potential field gradient.

[0015] The second aspect of this application provides a vehicle cooperative traffic control system based on cloud collaboration and road potential field. The system includes: a potential energy map acquisition component, used to discretize the target road network into a directed graph structure, receive real-time reported data via the cloud, use a continuous recursive prediction mode with a preset time window in the directed graph structure as a stage prediction mode, and adopt first-order partial differential equations based on edge directions as the prediction method to obtain a dynamic potential energy map of the road network through potential energy prediction; a search module construction component, used to construct a path search module based on the dynamic potential energy map of the road network; a traffic path solving component, used to generate candidate path maps for each vehicle by performing global path coarse screening and pruning using breadth-first search, and then evaluating the potential energy of the candidate path maps using depth-first search, and solving for the target traffic path of each vehicle through alternating iterations; and a traffic scheme determination component, used to perform local dynamic route replanning under spatiotemporal conflict localization based on the target paths of each vehicle, determine a cooperative traffic scheme, and provide vehicle-side targeted delivery and traffic guidance through the vehicle network.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0017] By discretizing the target road network into a directed graph structure and combining real-time cloud data with short-term window potential energy recursive prediction, a dynamic potential energy map of the road network is generated, enabling the early quantification and prediction of congestion and risk status of each node and road in the road network. By constructing a multi-level path search module, combining breadth-first search for global path screening with depth-first search for local potential energy evaluation, and achieving optimal path convergence through alternating iterations, efficient and dynamically adjustable target travel paths can be generated for each vehicle, improving traffic efficiency and path planning accuracy. Furthermore, local dynamic route replanning based on spatiotemporal conflict localization, supplemented by road network potential field gradient guidance and vehicle-to-everything (V2X) execution, enables collaborative travel of multiple vehicles in complex road network environments, effectively reducing conflict risks and improving traffic safety and overall road network efficiency.

[0018] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the vehicle cooperative traffic control method based on cloud collaboration and road potential field provided in this application.

[0021] Figure 2 Potential energy prediction curve provided for this application.

[0022] Figure 3 This is a schematic diagram of the vehicle cooperative traffic control system based on cloud collaboration and road potential field provided in this application.

[0023] Figure labeling: Potential energy map acquisition component 11, search module construction component 12, path solving component 13, and path determination component 14. Detailed Implementation

[0024] This application provides a vehicle cooperative traffic control method and system based on cloud collaboration and road potential field analysis. It addresses the technical problems of insufficient utilization of real-time dynamic information of the road network in existing technologies, leading to frequent multi-vehicle traffic conflicts and low path planning efficiency. It achieves the technical effect of improving overall traffic efficiency by realizing dynamic perception of the road network and collaborative optimization of multi-vehicle paths through cloud collaboration and road potential field analysis.

[0025] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0026] Example 1, as Figure 1 As shown, this application provides a vehicle cooperative traffic control method based on cloud collaboration and road potential field, which includes: The target road network is discretized into a directed graph structure. Real-time data is received from the cloud. In the directed graph structure, a staged prediction mode is used by continuous recursion within a preset time window. The prediction method is based on the first-order partial differential equation solution of the edge direction. Through potential energy prediction, the dynamic potential energy map of the road network is obtained.

[0027] Furthermore, in the directed graph structure, nodes represent intersections or key road segments, edges represent connecting roads, and each node is identified by a state feature vector; the cloud accesses roadside radar data and data reported by connected vehicles in real time, initializes the state features of the directed graph structure, performs recursive potential energy trend prediction based on a preset time window, and obtains the predicted potential energy value for the preset time window, wherein the predicted potential energy value is identified by a confidence interval; the predicted potential energy value is superimposed on the directed graph structure to obtain a dynamic potential energy map of the road network.

[0028] Furthermore, the state feature vector includes instantaneous features and propagation features. The instantaneous features include the node's real-time vehicle density, average vehicle speed, road length, and risk potential energy value. The propagation features include the estimated arrival time of the congestion wave, the potential energy along the main propagation direction of each outgoing edge, and the attenuation coefficient of the potential energy with distance.

[0029] Specifically, basic data of the target road network is obtained from electronic map data, high-precision map data, or traffic management platforms, including road centerlines, intersection locations, lane information, traffic signs, and road lengths. The target road network is discretized according to road intersection relationships and traffic control characteristics. Intersection center points, ramp entrances and exits, road merging and diverging points, and road segmentation points exceeding a preset threshold (e.g., 300m-500m) are defined as nodes. Road segments between any two adjacent nodes that allow continuous vehicle traffic are defined as directed edges, forming a directed graph G=(V,E). Each node in the node set V represents an intersection or key road segment, and each directed edge in the edge set E represents a connecting road between two adjacent nodes, with the direction consistent with the actual traffic direction. For two-way roads, two directed edges with opposite directions are constructed for representation.

[0030] Furthermore, each node in the directed graph G is identified by a state feature vector, which includes instantaneous features and propagation features. The instantaneous features include the node's real-time vehicle density, average vehicle speed, road length, and risk potential value. The real-time vehicle density and average vehicle speed are calculated based on roadside sensing devices, such as millimeter-wave radar, geomagnetic detectors, and GPS trajectory points reported by V2X connected vehicles, by statistically analyzing the average number of vehicles and their speeds per unit road length within a fixed time window, such as 1-5 seconds. The road length is obtained based on electronic map data, high-precision map data, or a traffic management platform. The risk potential value is defined based on actual congestion and abnormal events. For example, it is normalized using the ratio of real-time vehicle density to maximum carrying capacity at a node to obtain a congestion index. For abnormal events, such as accidents, sudden braking, or construction, an event index is calculated based on the severity of the event and its historical frequency over a period of time, such as 10 minutes or half an hour. The event weight is determined based on the severity of the event and expert experience, and normalized using the historical maximum abnormal value. For example, the abnormal event index = (event weight × frequency of occurrence) / historical maximum abnormal value. Then, the normalized congestion index and the abnormal event index are weighted and summed according to preset weights. For example, the risk potential value P... risk =w1 × congestion index + w2 × abnormal event index, where w1 and w2 are set according to traffic management objectives, and their sum is 1. For example, if a node has a real-time vehicle density of 40 vehicles / 100 meters, a maximum carrying capacity of 50 vehicles / 100 meters, a normalized congestion index of 0.8, a minor accident with a weight of 0.5 occurred at this node in the last 10 minutes, the historical maximum abnormal value is 1, the abnormal event normalized index is 0.25, and the weights of w1 and w2 are set to 0.6 and 0.4 respectively, the risk potential value P corresponding to this node is... risk =0.6×0.8+0.4×0.25=0.56.

[0031] The propagation characteristics include the estimated arrival time of the congestion wave, the potential energy along the main propagation direction of each outgoing edge, and the attenuation coefficient of the potential energy with distance. The estimated arrival time of the congestion wave represents the time required for congestion generated from an upstream node to propagate to the current node, obtained by acquiring the road length L between nodes and the current vehicle density ρ of the upstream node. u and average vehicle speed v u The congestion wave velocity v can be calculated using linear traffic flow models, such as the LWR model. w = q / ρ, where q is the traffic flow, q = ρ u ×v u According to formula T arrival =L / |v w Calculate the estimated arrival time. If there are multiple upstream nodes, take the shortest time as the estimated arrival time of the congestion wave for that node.

[0032] The potential energy propagation direction along the outgoing edges is used to indicate which road the traffic risk or congestion of a node is most likely to extend to. For each outgoing edge from the current node i to the neighboring node j, the rate of change of the risk potential energy value ΔP is calculated. risk，ij =P j -P i The potential gradient along the edge is obtained by multiplying the edge's direction vector by the potential energy gradient of the edge. The direction vector of the edge is obtained by subtracting the coordinates of the current node i from the coordinates of the neighbor node j and normalizing the result. Then, the edge with the largest gradient among all outgoing edges is selected as the main propagation direction, which represents the potential energy, such as congestion or risk, that is most likely to spread along this direction.

[0033] The attenuation coefficient reflects the rate at which the potential energy of a node weakens as it propagates through space. A curve showing the change in potential energy value propagating downstream from the source node as a function of distance is collected, and an exponential decay model is fitted. Where P0 is the current node's risk potential energy value, L is the road length between the two nodes, and λ is the attenuation coefficient, obtained through least squares fitting or regression learning. λ represents the rate at which the potential energy decays with distance; a larger value indicates faster attenuation and less impact on distant nodes. For example, node A has an initial risk potential energy value of 0.8, and neighboring nodes B, C, and D are 50m, 100m, and 200m away, respectively, with observed risk potential energy values ​​of 0.7, 0.6, and 0.45. An exponential decay curve is fitted. .

[0034] The cloud-based system receives real-time data from roadside radar and vehicle-to-everything (V2X) vehicles to initialize the state characteristics of each node in the directed graph. Then, within each preset time window T, a recursive first-order partial differential equation method is used for phased potential energy prediction. This preset time window is divided into multiple short-term continuous windows Δt; for example, a 1-minute preset time window is divided into 6 short-term windows Δt, where each Δt = 10 seconds. For each node i, the first-order partial differential of the potential energy is calculated along its outgoing edge within the short-term window Δt. This process is iterated continuously over the entire preset window T to obtain the curve of the node's potential energy changing over time. At each time point, the mean μ(t) and standard deviation σ(t) of the predicted potential energy value of the node are calculated to obtain the confidence interval: CI(t) = μ(t) ± z × σ(t), where z is the confidence coefficient, such as 1.96 for a 95% confidence level. This confidence interval represents the credible range of the predicted potential energy value at a given time point, thus reflecting the uncertainty of traffic state prediction.

[0035] By dynamically modeling and predicting the traffic conditions of the target road network, basic data is provided for subsequent global path analysis and multi-vehicle cooperative traffic. This enables path planning to not only consider the current road conditions but also predict potential future congestion and risks, thus achieving forward-looking traffic optimization.

[0036] Furthermore, a recursive potential energy trend prediction based on a preset time window is performed to obtain the predicted potential energy value for the preset time window, including: dividing the preset time window into multiple short-term continuous windows; within the first short-term window, with the initial state as the starting condition for the first node, performing a first-order partial differential equation along each directed edge direction to calculate the potential energy propagation rate from the high-density node to the adjacent low-density node, which is used as the predicted potential energy for the first window, wherein the first node is any node in the directed graph structure, and the first short-term window is the first window segment among the multiple short-term continuous windows; using the predicted potential energy for the first window as the starting condition, performing multiple rounds of iterative solutions until the preset time window is covered to obtain the potential energy curve of the first node; and establishing a connection between the potential energy curve of the first node as the first predicted potential energy value and the first node in the directed graph structure.

[0037] Specifically, a preset time window T, such as 1 minute or 5 minutes, is divided into several short, continuous windows Δt of equal length, such as 10 seconds or 30 seconds. The specific division can be dynamically set according to the frequency of traffic fluctuations and computing resources, in order to improve prediction stability while ensuring real-time performance. Any node in the directed graph structure G is denoted as the first node i.

[0038] Then, within the first short-time window, starting with the initial state, the first-order partial differential equation is solved along all outgoing edges of the first node i. The first short-time window is the first and second window segment among multiple short-time consecutive windows. The initial state refers to the process of quantifying the basic traffic state of the first node i at the current time t0 into the node potential energy value, including the vehicle density ρ. i (t0), average velocity v i (t0), Risk potential value P risk,i After normalization (t0), the base value of the current potential energy is formed by weighted summation. The weights of vehicle density, average speed, and risk potential energy value can be set based on actual needs. The weighted summation process is similar to the aforementioned risk potential energy value calculation process and will not be repeated here. Then, the influence of the potential energy of adjacent nodes on this node is adjusted based on the propagation characteristics, taking into account the arrival time T of the congestion wave. arrival Given the potential energy propagation direction along the outgoing edge and the attenuation coefficient αij, according to the principle of potential energy transfer, the potential energy of adjacent nodes is mapped to the current node i according to the propagation direction and attenuation effect, forming a correction term: , of which E i This is a correction term, where θij is the angle between the edge direction and the direction of the node's potential gradient. The weights are used to determine the decay of the congestion wave based on the estimated arrival time. Then, the base potential energy value is superimposed with the correction term to obtain the initial state potential energy value of the node at time t, which serves as the starting condition for subsequent short-term window potential energy prediction.

[0039] For each directed outgoing edge of the first node i, perform a first-order partial differential equation based on the edge direction, first obtaining the potential energy P of the neighboring nodes. j (t0), then combined with the potential energy P of the first node i i (t0) Calculate the spatial potential gradient along the edge: P / x=(P j (t0)-P i (t0)) / L ij L ij Let be the road length corresponding to the edge between node i and node j. Then, calculate the potential gradient along each outgoing edge. P / x and the propagation speed between nodes v w Multiplying them gives the potential energy propagation rate in that direction. P i / Within the first short-time window Δt, the nodal potential energy is updated to: P i (t0+Δt)=P i (t0)+( P i / The updated result (t)×Δt is used as the first window to predict the potential energy, which is then used for the recursive calculation of the next short window.

[0040] The process iterates through multiple short time windows until it covers the entire preset time window T, thus forming the first node potential energy curve Pi(t) of node i. Finally, the first node potential energy curve is used as the predicted potential energy value of that node and associated with the node attributes in the directed graph structure G for storage, resulting in a dynamic potential energy graph of the road network. This graph contains not only current state information but also trend evolution information in the future time dimension, providing a predictive basis for subsequent path search and conflict resolution.

[0041] For example, such as Figure 2 Node A is a road intersection node in the target road network. At initial time t0, the vehicle density is 30 vehicles / 100 meters, the average speed is 25 km / h, and the risk potential energy is 0.3. Through weighted calculation, the initial potential energy is obtained as 0.4. The preset time window is 60 seconds, divided into 6 short-time windows of Δt = 10 seconds each. In the first window, the neighboring node B has a density of 45 vehicles / 100 meters, a speed of 15 km / h, a potential energy of 0.6, and a road length of 100 meters. This is calculated through spatial difference. P / x=(0.6 0.4) / 100=0.002, calculated by considering the time variation. P / If t=0.015, then PA(t0+10s)=0.4+0.015×10=0.55. Then, the iterative update continues. In subsequent windows, due to upstream congestion propagation, the potential energy gradually evolves to 0.55, 0.63, 0.68, 0.70, 0.66, and 0.60, eventually forming the complete potential energy prediction curve for node A.

[0042] By extending static or instantaneous traffic conditions into dynamic potential energy representations with time-evolution capabilities, each node in the target road network can not only reflect the current level of congestion or risk, but also the trend of change over a future period. This improves the ability to predict congestion propagation, thereby enhancing the reliability of route planning and multi-vehicle collaborative decision-making, as well as overall traffic efficiency.

[0043] A path search module is constructed based on the dynamic potential energy diagram of the road network.

[0044] Specifically, based on the road network dynamic potential energy map, a path search module is constructed, including: deploying a first search layer by coarse screening and hierarchical pruning of road network accessibility under breadth-first search; deploying a second search layer by recursively refining candidate paths using depth-first search; deploying a third search layer by synergistic fusion and iteration of depth search and breadth search; cascading the first search layer, the second search layer, and the third search layer, embedding the road network dynamic potential energy map within the first search layer, and supervising the training of the path search module.

[0045] Specifically, based on the dynamic potential energy graph of the road network, a three-layer collaborative path search module is constructed to provide vehicles with globally optimal travel paths. First, breadth-first search (BFS) is used for coarse screening and hierarchical pruning of road network accessibility, deploying the first search layer: starting from the vehicle's current position, based on the nodes of the dynamic potential energy graph, the reachable nodes are expanded layer by layer along each outgoing edge. The minimum cost lower bound of each path is calculated and hierarchical pruning is performed, discarding candidate nodes with high potential energy or long paths to form the first candidate path graph. This quickly eliminates impossible or overly costly paths, reduces the search space, and improves computational efficiency.

[0046] Then, a depth-first search (DFS) is used to recursively refine the first candidate path graph, deploying a second search layer: for each candidate path, nodes are recursively visited from the starting point along the path depth, and the local potential energy of each node on the path is evaluated based on the predicted potential energy value, generating depth search instructions or return instructions to iteratively update the candidate path graph for refined evaluation and multi-branch exploration, ensuring that locally optimal or even globally optimal paths can be found. By synergistically fusing breadth-first search and depth-first search, a third search layer is deployed to further optimize the candidate path graph obtained from the first and second search layers.

[0047] The entire path search module is deployed in a cascaded manner: the first search layer provides global coarse screening, the second search layer relies on the candidate graph output by the first layer for local refinement, and the third search layer performs iterative convergence. Furthermore, the dynamic potential energy graph of the road network is embedded in the first search layer as a pruning basis to supervise the training of the search strategy, so that the search can take into account both real-time performance and optimality.

[0048] By constructing a path search module, the vehicle traffic problem is transformed into a potential energy optimization path search problem. By utilizing predicted potential energy and spatial topology information, a multi-layer search is achieved, from global coarse screening to local refinement and then to iterative convergence. This enables each vehicle to obtain the optimal and safest driving path under future traffic conditions, thereby improving road network traffic efficiency and multi-vehicle collaboration capabilities.

[0049] Based on the vehicle's current location and destination, the path search module uses breadth-first search to perform global path screening and pruning to generate candidate path maps for each vehicle. Depth-first search is then used to evaluate the potential energy of the candidate path maps. Through alternating iterations, the optimal path is converged to solve for the target travel path of each vehicle.

[0050] Furthermore, a breadth-first search is employed for global path coarse screening and pruning to generate a candidate path graph. This includes: determining a first node layer based on the road network dynamic potential energy graph, with the vehicle's current location as the starting point and the destination as the ending point. The first node layer consists of multiple reachable road network nodes extending outward from the road network node corresponding to the starting point. Path pruning is performed on each reachable road network node from the starting point to the first node layer by calculating the minimum cost lower bound, resulting in a first-level candidate path backbone. The minimum cost lower bound is calculated using Euclidean distance and free-flow time. Path pruning decisions are executed from the first-level candidate path backbone to the second node layer, resulting in a second-level candidate backbone node. The second node layer consists of multiple reachable road network nodes extending outward from each road network node in the first-level candidate path backbone. Through iterative recursion until the destination is reached, hierarchical associations are performed on the obtained first-level candidate path backbone, second-level candidate backbone nodes, and up to N-level candidate backbone nodes to obtain the first candidate path graph.

[0051] Specifically, based on the vehicle's current location and destination, the first search layer in the path search module first uses breadth-first search to perform a global path coarse screening for each vehicle, taking the vehicle's current location as the starting node and the destination as the ending node. Based on the starting node, the path expands to the neighborhood along the outgoing edges of the directed graph in the road network dynamic potential energy graph to form the first node layer. The first node layer consists of multiple reachable road network nodes that expand to the neighborhood with the road network node corresponding to the starting point as the center, including all neighborhood nodes that can be directly reached within a short distance.

[0052] Then, for each reachable road network node from the starting point to the first node layer, the lower bound of the minimum cost is calculated. This lower bound is calculated by dividing the Euclidean distance by the free-flow time. The Euclidean distance is the straight-line distance between the coordinates of the starting node and the coordinates of the reachable road network node in the first node layer; the geographic coordinates can be obtained from GIS road network data. The free-flow time is determined by taking the highest speed limit of all roads in the entire road network or by averaging the speeds of different segments, representing the maximum speed a vehicle can achieve under ideal, completely unobstructed conditions. Then, path pruning is performed based on the lower bound of the minimum cost. Reachable road network nodes whose lower bound of the minimum cost is greater than the path cost threshold are removed from the first node layer, forming the first layer of candidate path backbones. The path cost threshold is determined based on historical traffic data and path length statistics, taking the percentile of the lower bound of the minimum cost, for example, retaining the top 80% of low-cost paths. To avoid prematurely removing potentially low-potential paths, a safety margin, such as 0.1 to 0.3, can be added to the lower bound of the minimum cost.

[0053] Then, each road network node in the first-layer candidate path backbone is used as a starting point to expand multiple reachable road network nodes to its neighborhood to form a second node layer, and the minimum cost lower bound of each path is calculated. Each minimum cost lower bound is compared with the path cost threshold, and pruning is performed again on path nodes that are greater than the path cost threshold to obtain the second-layer candidate backbone nodes.

[0054] After generating the first and second layer candidate backbone nodes, starting from the second layer candidate backbone nodes, expand to the neighborhood along the outgoing edges of each node, calculate the minimum cost lower bound of each path, prune nodes that exceed the path cost threshold, and form the third layer candidate backbone nodes. Then, starting from the third layer nodes, repeat the same operation to form the fourth layer candidate backbone nodes, and so on, through recursive iteration until the layer where the endpoint node is located is covered, forming N layers of candidate backbone nodes, where N is a positive integer. In each iteration, by recording the parent-child connection relationship between the nodes of the previous layer and the nodes of the current layer, the candidate path backbone nodes of the first layer, the candidate backbone nodes of the second layer, and so on up to the candidate backbone nodes of the Nth layer are hierarchically associated according to the hierarchical relationship, so that each candidate path from the starting point to the ending point can trace its node order, forming the first candidate path graph. The first candidate path graph is a directed acyclic graph, which contains all low-cost candidate paths from the starting point node to the ending point node. The nodes are all unpruned road network nodes, and the edges are the original directed edges connecting the road network nodes. The structure of the first candidate path graph is stored in the form of an adjacency list so that subsequent DFS can traverse it quickly.

[0055] For example, the vehicle's current starting point is network node A, coordinates (0,0), and the destination point G is coordinates (700,0). The free-flow speed is 20 m / s. First, expand from A along the outgoing edges into the neighborhood to obtain the first node layer, which includes node B (minimum cost lower bound 38.0 seconds) and node C (minimum cost lower bound 49.5 seconds). Assuming the path cost threshold is the first 80% of the minimum cost lower bound, approximately 44 seconds, then node B is retained, and node C is removed, making B the backbone of the first candidate path layer. Then, expand from B to obtain the second node layer: D (minimum cost lower bound 41.5 seconds) and E (minimum cost lower bound 42.6 seconds), both within the 44-second threshold. Continue expanding: from D to G, the minimum cost lower bound is 39.3 seconds; from E to G, the minimum cost lower bound is 43.5 seconds, both reaching the destination. Finally, the first candidate path graph is formed, containing two candidate paths: A→B→D→G and A→B→E→G. Nodes are stored hierarchically as an adjacency list.

[0056] By using breadth-first search for global path coarse screening, the feasible path space from the origin to the destination is rapidly reduced, effectively eliminating impossible or overly costly paths. This achieves a balance between global reachability and computational efficiency, ensuring that a preliminary path candidate set can be quickly generated in large-scale road networks during multi-vehicle collaborative planning.

[0057] Furthermore, a potential energy assessment of the candidate path graph is performed using depth-first search, including: for the first candidate path graph, performing a local assessment based on the predicted potential energy value from the starting point to any road network node in the first-level candidate path backbone to obtain a first assessment result, wherein the assessment type is lane change, detour, or waiting; based on the first assessment result, generating a depth search instruction or a return instruction, and iterating through multiple rounds until the search of the first candidate path graph is completed, constructing a first assessment graph; wherein, if it is a depth search instruction, updating the road network node to the starting point, and selecting any road network node from the associated second-level candidate backbone nodes for assessment; if it is a return instruction, returning to the starting point, and reselecting and assessing road network nodes from the first-level candidate path backbone.

[0058] Specifically, after generating the first candidate path map using breadth-first search, the path search module further employs depth-first search to assess the potential energy of the first candidate path map. First, starting from the origin node corresponding to the vehicle's current position, an associated node is randomly selected from the first-layer candidate path backbone of the first candidate path map as the current evaluation node. The cloud calculates the estimated arrival time of the vehicle to the current evaluation node based on the vehicle's current position, current speed, and the length of the searched path. Then, the predicted potential energy value of the node at the corresponding time is extracted from the road network dynamic potential energy map. Based on the predicted potential energy value, combined with the lane attributes of the road where the node is located, the accessibility of adjacent roads, and the potential energy gradient, a local evaluation of the current evaluation node is performed to obtain the first evaluation result. The evaluation result includes three categories: lane changing, using another lane, or waiting. The potential energy gradient refers to the potential energy difference between the current node and its successor nodes, used to reflect future congestion trends. A low potential energy threshold of 0.4 and a high potential energy threshold of 0.7 are set. If the predicted potential energy value of the current node is less than the low potential energy threshold, and there is a passable lane adjacent to the current lane with even lower potential energy, the future traffic environment is considered favorable, and a lane-changing assessment result is output. If the predicted potential energy value of the current node is between the low and high potential energy thresholds, and the potential energy of adjacent lanes or adjacent road nodes is significantly lower than that of the current node (e.g., the potential energy difference exceeds 0.15), it indicates an opportunity to bypass local congestion, and a lane-borrowing assessment result is output. If the predicted potential energy value of the current node is greater than the high potential energy threshold, it indicates that a congestion cluster or risk zone will form in this area within the future time window, and a pending assessment result is output.

[0059] Then, based on the first evaluation result, a depth search instruction or a return instruction is generated. When the evaluation type corresponding to the current node is lane change or detour, and the condition for continuing to expand forward is met, a depth search instruction is generated. The current node is updated as the new search starting point, and the potential energy evaluation continues along the next layer of candidate backbone nodes associated with it in the first candidate path graph, thereby deepening the search layer by layer along the candidate path towards the destination. When the evaluation result is "wait," a return instruction is generated, backtracking to the previous layer node. If there are no selectable nodes in the previous layer, it further returns to the initial starting point node and reselects other nodes in the first layer of candidate backbone as new search entry points, re-executing the potential energy evaluation.

[0060] Through multiple rounds of recursive iteration, all reachable branches in the first candidate path graph are traversed, and the potential energy value and evaluation type of each path node are recorded. This information is then integrated to form the first evaluation graph, which is a path graph structure containing all complete paths and their evaluation information obtained by DFS from the starting point to the end point. Nodes store local evaluation results, while edges store hierarchical connections and potential energy values, which can be directly called in subsequent breadth-first search pruning or alternating iterations to achieve optimal path convergence.

[0061] By employing depth-first search to perform a refined potential energy assessment on the candidate paths retained from breadth-first search, path branches with lower future traffic risk, less congestion, and higher traffic efficiency are further filtered from the global candidate paths. Simultaneously, by introducing a return mechanism and local traffic strategy evaluation, potential congestion propagation areas and high-risk areas can be identified in advance, preventing vehicles from entering potentially suboptimal paths, thereby improving the convergence speed of optimal paths and the quality of collaborative traffic decisions.

[0062] Furthermore, the optimal path convergence is achieved through alternating iterations, including: writing the first evaluation graph into the open set of the first candidate path graph and updating it with an expanded priority coefficient; performing breadth-first search pruning on the first candidate path graph using the expanded priority coefficient to obtain a second candidate path graph; performing a depth-first search evaluation based on the predicted potential value on the second candidate path graph to obtain a second evaluation graph; writing the second evaluation graph into the second candidate path graph, and obtaining the target path by performing breadth-first search pruning and depth-first search evaluation under multiple rounds of alternating search iterations.

[0063] Specifically, after evaluating the potential energy of the first candidate path graph's depth-first search, a continuous optimization of the candidate paths is achieved using an alternating breadth-first search and depth-first search approach. This process gradually converges the path selection process from initially feasible paths to the globally optimal path. The first evaluation graph is written into the open set corresponding to the first candidate path graph. The open set refers to the set of candidate path nodes that have not yet been eliminated and are allowed to continue expanding during the current search process. Simultaneously, a priority evaluation function is constructed based on the predicted potential energy value and remaining distance. An expansion priority coefficient is calculated for each candidate node in the open set, where the priority evaluation function y... k =a×(1-P k )+b×(1-D k / D max ), where a and b are weighting coefficients, which can be set based on actual needs, P k D represents the predicted potential energy value corresponding to the expected arrival node of the vehicle. k D represents the remaining distance from the node to the destination. max This represents the maximum remaining distance in the current candidate path graph. The lower the predicted potential and the closer to the destination, the higher the expansion priority.

[0064] Then, a second round of breadth-first search pruning is performed using the expansion priority coefficient. The open set of the first candidate path graph is sorted from highest to lowest according to the expansion priority coefficient. Breadth-first search pruning is performed on path branches with expansion priority coefficients lower than a preset priority threshold, retaining only path nodes with higher expansion priority and their associated edges to generate a second candidate path graph. The preset priority threshold is calculated by statistically analyzing the expansion priority coefficients of all paths in the current open set, calculating the mean and standard deviation, and subtracting the product of the adjustment coefficient and the standard deviation from the mean priority coefficient. The adjustment coefficient is between 0.3 and 1, automatically adjusted based on the overall quality of the current candidate paths: when the overall path quality is high and the distribution is concentrated, the standard deviation is small, and the threshold is increased accordingly, thus retaining only high-quality paths; when the overall path quality is low or the distribution is scattered, the standard deviation increases, and the threshold decreases, thus avoiding over-pruning that leads to insufficient feasible paths.

[0065] For the second candidate path map, a deep search evaluation based on predicted potential energy values ​​is performed again. Using the predicted potential energy curves in the road network dynamic potential energy map, the predicted potential energy values ​​for each node at the corresponding time are recalculated, and local evaluation results such as lane changing, detour, or waiting are regenerated to obtain the second evaluation map. This second evaluation map is then written back into the open set of the second candidate path map, the expansion priority coefficient is recalculated, and the next round of breadth-first search pruning is initiated. This process is iterated and repeated until the number of candidate paths is reduced to a preset number, such as only one or two paths remaining. At this point, the search results are considered converged, and the final target path is output.

[0066] By using the local potential energy assessment results obtained from depth search to guide the path expansion direction of breadth search, and then re-evaluating the deep potential energy using the candidate paths filtered by breadth search, the breadth search no longer relies solely on static topology for filtering, but dynamically adjusts its search direction based on the deep potential energy assessment results. Simultaneously, depth search no longer exhaustively analyzes all paths, but focuses on high-value candidate paths retained by breadth search, significantly reducing computational load. Through the alternating iteration of these two search strategies, a gradual convergence from passable paths to optimal passable paths is achieved, further improving path planning accuracy and real-time performance, and enhancing the effectiveness and efficiency of vehicle cooperative traffic control.

[0067] Based on the target travel routes of each vehicle, local dynamic route replanning is performed under the spatiotemporal conflict localization to determine a collaborative travel scheme, and vehicle-side targeted deployment and travel guidance are carried out through the vehicle network.

[0068] Furthermore, the local dynamic route replanning under spatiotemporal conflict localization includes: obtaining the target travel path of each vehicle and constructing a spatiotemporal conflict map, wherein the spatiotemporal conflict map defines conflict based on the spatiotemporal occupancy overlap of the same node and the same side; and performing local dynamic route replanning based on the spatiotemporal conflict map to obtain a cooperative travel scheme, wherein the replanning route is guided by the road network potential field gradient.

[0069] Specifically, after obtaining the target travel paths of all vehicles in the target road network, a spatiotemporal conflict graph is constructed. Nodes represent road network nodes, i.e., intersections or key road segments, and edges represent road connections. Each edge and node is appended with the estimated arrival and departure time intervals of vehicles calculated from the target travel path and predicted travel speed. Specifically, the time required for a vehicle to travel from the previous node to the current node is calculated using the formula road length / speed, and then the estimated arrival time of the preceding node is added to obtain the estimated arrival time of the current node. The estimated departure time is the arrival time plus the vehicle's transit time at the node, such as the average travel time. When two vehicles have overlapping spatiotemporal occupancy intervals at the same node or edge, a conflict relationship edge is established in the spatiotemporal conflict graph to identify potential conflicts.

[0070] Then, based on the spatiotemporal conflict graph, local dynamic route replanning is performed. For example, based on vehicle type, real-time speed, and potential energy value, the priority of each vehicle at the conflict node is calculated. Vehicles with higher priority maintain their original target path, while vehicles with lower priority undergo dynamic adjustment. The priority calculation expression is as follows: S = w³ × (1 - P) k )+w4×(V k / V max )+w5×(t k / T), where P k V represents the potential energy value of the node that vehicle k is expected to occupy. k Let V be the current speed of vehicle k. max For the maximum speed limit of the road, t k Let T be the remaining time for vehicle k to reach the conflict node, and let T be the length of the spatiotemporal window.

[0071] For vehicles with low priority and existing conflicts, local paths are replanned along the direction of decreasing potential energy, guided by the dynamic potential field gradient of the road network. The potential field gradient refers to the direction vector of how node potential energy changes spatially, representing the direction in which congestion or risk is most likely to spread. During replanning, vehicles select adjacent nodes along the direction of decreasing potential energy gradient to ensure local avoidance of conflicts while maintaining the original path's efficiency as much as possible. Local replanning only adjusts the paths of conflicting nodes and their N preceding and following nodes, avoiding significant deviations from the original target route and ensuring global path stability. After local path adjustment, the spatiotemporal occupancy of affected nodes and edges is recalculated, and the spatiotemporal conflict graph is updated. If conflicts still exist, the above steps are repeated until all conflicts are resolved or a minimized conflict state is achieved within the allowed spatiotemporal window. Finally, a cooperative traffic scheme is generated, ensuring vehicles proceed along the overall direction of the target traffic path while eliminating potential conflicts through local adjustments. By using vehicle-to-everything (V2X) technology, collaborative traffic plans are distributed to vehicles. Vehicles then receive driving guidance based on these plans, including lane-changing suggestions or partial route adjustments, enabling collaborative traffic flow between vehicles and ensuring smooth traffic flow across the entire road network.

[0072] By analyzing the spatiotemporal occupancy conflicts of the target path and combining local dynamic replanning with potential gradient guidance, congestion and conflicts at intersections or key road sections can be effectively avoided. In the case of multiple vehicles traveling at the same time, the safety and efficiency of coordinated passage can be improved, thereby ensuring the smoothness and safety of traffic in the entire road network.

[0073] Example 2, based on the same inventive concept as the vehicle cooperative traffic control method based on cloud collaboration and road potential field in the previous examples, such as... Figure 3 As shown, this application provides a vehicle cooperative traffic control system based on cloud collaboration and road potential field, wherein the vehicle cooperative traffic control system based on cloud collaboration and road potential field includes: The potential energy map acquisition component 11 is used to discretize the target road network into a directed graph structure. It receives real-time reported data from the cloud and uses a continuous recursive prediction mode with a preset time window in the directed graph structure. It adopts the first-order partial differential equation based on edge direction as the prediction method and obtains the dynamic potential energy map of the road network through potential energy prediction. The search module construction component 12 is used to construct a path search module based on the dynamic potential energy map of the road network. The travel path solving component 13 is used to generate candidate path maps by using breadth-first search for global path coarse screening and pruning for each vehicle based on the current position and destination of the vehicle. The depth-first search is used to evaluate the potential energy of the candidate path maps. The optimal path is converged through alternating iterations to solve the target travel path of each vehicle. The travel scheme determination component 14 is used to perform local dynamic route replanning under spatiotemporal conflict localization based on the target travel paths of each vehicle, determine the cooperative travel scheme, and use the vehicle network to provide vehicle-side targeted delivery and travel guidance.

[0074] Furthermore, the potential energy map acquisition component 11 is also used for: in the directed graph structure, nodes represent intersections or key road segments, edges represent connecting roads, and each node is identified by a state feature vector; the cloud accesses roadside radar data and data reported by connected vehicles in real time, initializes the state features of the directed graph structure, performs recursive potential energy trend prediction based on a preset time window, and obtains the predicted potential energy value of the preset time window, wherein the predicted potential energy value is identified by a confidence interval; and superimposes the predicted potential energy value onto the directed graph structure to obtain a dynamic potential energy map of the road network.

[0075] Furthermore, the potential energy graph acquisition component 11 is also used to: divide multiple short-term continuous windows according to the preset time window; within the first short-term window, with the initial state as the starting condition, perform first-order partial differential equations along each directed edge direction for the first node to calculate the potential energy propagation rate from the high-density node to the adjacent low-density node, and use it as the first window predicted potential energy, wherein the first node is any node in the directed graph structure, and the first short-term window is the first and second window segments among the multiple short-term continuous windows; with the first window predicted potential energy as the starting condition, perform multiple rounds of iterative solutions until the preset time window is covered to obtain the potential energy curve of the first node; and establish an association between the first node potential energy curve as the first predicted potential energy value and the first node in the directed graph structure.

[0076] Furthermore, the potential energy map acquisition component 11 is also used for: the state feature vector includes instantaneous features and propagation features, wherein the instantaneous features include the real-time vehicle density, average vehicle speed, road length and risk potential energy value of the node, and the propagation features include the expected arrival time of the congestion wave, the main propagation direction of the potential energy along each outgoing edge and the attenuation coefficient of the potential energy with distance.

[0077] Furthermore, the search module construction component 12 is also used to: deploy a first search layer by coarse screening and hierarchical pruning of road network accessibility under breadth-first search; deploy a second search layer by recursively refining candidate paths using depth-first search; deploy a third search layer by synergistic fusion and iteration of depth search and breadth search; cascade the first search layer, the second search layer and the third search layer, embed the road network dynamic potential energy map in the first search layer, and supervise the training of the path search module.

[0078] Furthermore, the path solving component 13 is also used to: determine a first node layer based on the road network dynamic potential energy map, taking the current position of the vehicle as the starting point and the destination as the ending point, wherein the first node layer consists of multiple reachable road network nodes extending to the neighborhood from the road network node corresponding to the starting point; perform path pruning from the starting point to each reachable road network node in the first node layer by calculating the minimum cost lower bound to obtain a first-level candidate path backbone, wherein the minimum cost lower bound is calculated using Euclidean distance and free-flow time; perform path pruning decisions from the first-level candidate path backbone to the second node layer to obtain a second-level candidate backbone node, wherein the second node layer consists of multiple reachable road network nodes extending to the neighborhood from each road network node in the first-level candidate path backbone; and perform hierarchical association on the obtained first-level candidate path backbone, second-level candidate backbone nodes, and up to N-level candidate backbone nodes through recursive iteration until the destination is reached to obtain a first candidate path graph.

[0079] Furthermore, the path solving component 13 is also used to: for the first candidate path map, perform a local evaluation based on the predicted potential value from the starting point to any road network node in the first-level candidate path backbone to obtain a first evaluation result, wherein the evaluation type is lane changing, detour, or waiting; generate a depth search instruction or a return instruction based on the first evaluation result, and construct a first evaluation map through multiple iterations until the search of the first candidate path map is completed; wherein, if it is a depth search instruction, update the road network node to the starting point, and select any road network node in the associated second-level candidate backbone nodes for evaluation; if it is a return instruction, return to the starting point, and reselect and evaluate road network nodes from the first-level candidate path backbone.

[0080] Furthermore, the path solving component 13 is also used to: write the first evaluation graph into the open set of the first candidate path graph and update it with an expanded priority coefficient; perform breadth-first search pruning on the first candidate path graph using the expanded priority coefficient to obtain a second candidate path graph; perform a depth-first search evaluation based on the predicted potential value on the second candidate path graph to obtain a second evaluation graph; write the second evaluation graph into the second candidate path graph, and obtain the target path by performing breadth-first search pruning and depth-first search evaluation under multiple rounds of search alternating iterations.

[0081] Furthermore, the traffic scheme determination component 14 is also used to: obtain the target traffic path of each vehicle, construct a spatiotemporal conflict map, wherein the spatiotemporal conflict map defines conflict by the spatiotemporal occupancy overlap of the same node and the same side; and perform local dynamic route replanning according to the spatiotemporal conflict map to obtain a cooperative traffic scheme, wherein the replanning route is performed by the direction guided by the road network potential field gradient.

[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The vehicle cooperative traffic control method and specific examples based on cloud collaboration and road potential field in the aforementioned embodiment 1 are also applicable to the vehicle cooperative traffic control system based on cloud collaboration and road potential field in this embodiment. Through the foregoing detailed description of the vehicle cooperative traffic control method based on cloud collaboration and road potential field, those skilled in the art can clearly understand the vehicle cooperative traffic control system based on cloud collaboration and road potential field in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0083] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0084] Obviously, those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A vehicle cooperative traffic control method based on cloud collaboration and road potential field, characterized in that, The method includes: The target road network is discretized into a directed graph structure. Real-time data is received from the cloud. In the directed graph structure, a staged prediction mode is used by continuous recursion within a preset time window. The prediction method is based on the first-order partial differential solution of the edge direction. The dynamic potential energy map of the road network is obtained through potential energy prediction. Based on the road network dynamic potential energy map, a path search module is constructed; Based on the vehicle's current location and destination, the path search module uses breadth-first search to perform global path screening and pruning to generate candidate path maps for each vehicle. Depth-first search is used to evaluate the potential energy of the candidate path maps. The optimal path is converged through alternating iterations to solve the target travel path for each vehicle. Based on the target travel routes of each vehicle, local dynamic route replanning is performed under the spatiotemporal conflict localization to determine a collaborative travel scheme, and vehicle-side targeted deployment and travel guidance are carried out through the vehicle network.

2. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 1, characterized in that, In the directed graph structure, nodes represent intersections or key road segments, edges represent connecting roads, and each node is identified by a state feature vector. The cloud accesses roadside radar data and data reported by connected vehicles in real time, initializes the state features of the directed graph structure, performs recursive potential energy trend prediction based on a preset time window, and obtains the predicted potential energy value of the preset time window, wherein the predicted potential energy value is marked with a confidence interval. The predicted potential energy value is superimposed on the directed graph structure to obtain the dynamic potential energy map of the road network.

3. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 2, characterized in that, Perform a recursive potential energy trend prediction based on a preset time window to obtain the predicted potential energy value for the preset time window, including: Based on the preset time window, multiple short-time continuous windows are divided; Within the first short-term window, the first node is initialized and the first partial differential is solved along each directed edge to calculate the potential energy propagation rate from the high-density node to the adjacent low-density node. This rate is used as the first window to predict the potential energy. Here, the first node is any node in the directed graph structure, and the first short-term window is the first and second window segments within multiple short-term continuous windows. Using the predicted potential energy of the first window as the starting condition, multiple rounds of iterative solution are performed until the preset time window is covered, and the potential energy curve of the first node is obtained. The potential energy curve of the first node is used as the first predicted potential energy value and associated with the first node in the directed graph structure.

4. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 3, characterized in that, The state feature vector includes instantaneous features and propagation features. The instantaneous features include the real-time vehicle density, average vehicle speed, road length, and risk potential energy value of the node. The propagation features include the expected arrival time of the congestion wave, the main propagation direction of the potential energy along each outgoing edge, and the attenuation coefficient of the potential energy with distance.

5. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 1, characterized in that, Based on the road network dynamic potential energy map, a path search module is constructed, including: The first search layer is deployed by coarse screening of road network accessibility under breadth-first search and hierarchical pruning. Recursively refine candidate paths using depth-first search, and then deploy a second search layer; A third search layer is deployed by iteratively integrating deep search and breadth search. The first search layer, the second search layer, and the third search layer are cascaded, and the dynamic potential energy map of the road network is embedded in the first search layer to supervise the training of the path search module.

6. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 5, characterized in that, A breadth-first search is used for global path coarse screening and pruning to generate a candidate path graph, including: Based on the road network dynamic potential energy map, a first node layer is determined with the vehicle's current location as the starting point and the destination as the ending point. The first node layer consists of multiple reachable road network nodes that extend to the neighborhood, centered on the road network node corresponding to the starting point. Using the reachable road network nodes from the starting point to the first node layer, path pruning is performed by calculating the minimum cost lower bound to obtain a candidate path backbone. The minimum cost lower bound is calculated using Euclidean distance and free flow time. From the first-level candidate path backbone to the second node layer, a path pruning decision is performed to obtain the second-level candidate backbone nodes. The second node layer consists of multiple reachable road network nodes that expand to the neighborhood, starting from each road network node in the first-level candidate path backbone. By iterating recursively until the destination is reached, the resulting first-level candidate path backbone, second-level candidate backbone nodes, and so on up to the Nth-level candidate backbone nodes are hierarchically associated to obtain the first candidate path graph.

7. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 6, characterized in that, Potential evaluation of the candidate path graph using depth-first search includes: For the first candidate route map, from the starting point to any road network node in the first-level candidate route backbone, a local evaluation based on the predicted potential energy value is performed to obtain the first evaluation result, wherein the evaluation type is lane change, detour or waiting; Based on the first evaluation result, a deep search instruction or a return instruction is generated, and the search of the first candidate path graph is completed through multiple rounds of iteration, thereby constructing the first evaluation graph; If it is a deep search command, the road network node is updated as the starting point, and any road network node is selected from the associated second-level candidate backbone nodes for evaluation. If it is a return command, return to the starting point and perform road network node reselection and evaluation from the first-level candidate path backbone.

8. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 7, characterized in that, Optimal path convergence is achieved through alternating iterations, including: Write the first evaluation graph into the open set of the first candidate path graph and update it with the expanded priority coefficient; Using the aforementioned expansion priority coefficient, a breadth-first search pruning is performed on the first candidate path graph to obtain a second candidate path graph; For the second candidate path map, a deep search evaluation based on the predicted potential energy value is performed to obtain the second evaluation map; The second evaluation map is written into the second candidate path map, and the target path is obtained by performing breadth-based pruning and depth-based evaluation under multiple rounds of alternating search iterations.

9. The vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in claim 1, characterized in that, Performing local dynamic route replanning under spatiotemporal conflict localization includes: Obtain the target travel path of each vehicle and construct a spatiotemporal conflict map. The spatiotemporal conflict map defines conflict based on the spatiotemporal overlap of the same node and the same side. Based on the spatiotemporal conflict diagram, local dynamic route replanning is performed to obtain a cooperative travel scheme, wherein the replanning route is guided by the road network potential field gradient.

10. A vehicle cooperative traffic control system based on cloud collaboration and road potential field, characterized in that, The steps for implementing the vehicle cooperative traffic control method based on cloud collaboration and road potential field as described in any one of claims 1 to 9 include: The potential energy map acquisition component is used to discretize the target road network into a directed graph structure, receive real-time reported data through the cloud, and use a staged prediction mode with continuous recursion of a preset time window in the directed graph structure. It adopts the first-order partial differential solution based on the edge direction as the prediction method, and obtains the dynamic potential energy map of the road network through potential energy prediction. The search module construction component is used to construct a path search module based on the road network dynamic potential energy map; The route solving component is used to solve the target travel path for each vehicle by using a breadth-first search to perform global path screening and pruning to generate a candidate path map based on the vehicle's current location and destination. The path search module then uses a depth-first search to evaluate the potential energy of the candidate path map and performs optimal path convergence through alternating iterations. The traffic scheme determination component is used to perform local dynamic route replanning under spatiotemporal conflict localization based on the target traffic paths of each vehicle, determine the cooperative traffic scheme, and conduct vehicle-side targeted deployment and traffic guidance through the Internet of Vehicles.

Images