Urban-level path guidance method and system based on dynamic clustering in vehicle-road cloud cooperation scene

Abstract

The invention relates to a city-level path guidance method and system based on dynamic clustering in a vehicle-road cloud cooperation scene, and the method comprises the steps: constructing a vehicle-road cloud cooperation architecture, and collecting traffic data; the traffic management center updates the road weight and initializes a penalty matrix, and performs congestion detection at the same time; when a congested road section is detected, vehicles possibly affected by congestion are screened out, and a to-be-planned vehicle set is formed; performing spatial clustering on the to-be-planned vehicle set to obtain a plurality of vehicle clusters, and constructing an independent path planning sub-graph for each vehicle cluster; distributing priorities for the vehicles in each vehicle cluster; planning an alternative path for the vehicle; performing batch dynamic adjustment on the road weight and the penalty value on the path planning subgraph; the traffic management center issues the planned alternative path to the corresponding vehicle through the road side unit, and the vehicle runs according to the new running path to realize path induction; finally, reasonable distribution of traffic flow, improvement of road network traffic efficiency and balanced allocation of time-space resources are realized.

Description

Technical Field

[0001] The present invention relates to a city-level path guidance method and system based on dynamic clustering in a vehicle-road-cloud collaborative scenario, belonging to the technical field of traffic congestion control. Background Art

[0002] The accelerating pace of urbanization and the rapid growth of motor vehicle ownership have made traffic congestion a core bottleneck restricting urban transportation efficiency and sustainable development. Traditional route planning methods and navigation systems are often based on graph-based shortest path algorithms (such as Dijkstra and A*). These employ static pre-planning models, generating fixed routes before vehicle departure, with each vehicle making independent routing decisions. However, the dynamic nature of traffic flow limits the applicability of such solutions, which can easily lead to vehicle congestion in hotspots, causing traffic congestion and even regional road network disruptions.

[0003] To address these challenges, breakthroughs in next-generation information and communication technologies (including cloud computing and the Internet of Vehicles) have spawned a new traffic management paradigm: vehicle-road-cloud collaboration. This collaborative architecture, through efficient communication and collaboration between roadside units (ROUs), cloud-based traffic management centers, and vehicles, combined with sensor technology, overcomes information silos and enables real-time collection and processing of traffic data. Furthermore, cloud-based parallel computing clusters, leveraging powerful computing resources, provide real-time support for optimized traffic decision-making, driving the advancement of intelligent traffic management.

[0004] Based on this architecture, path guidance technology can be fully utilized. By acquiring real-time traffic situation data at the city road network level and combining it with cloud computing resources to plan the global optimal path for vehicles, it can simultaneously alleviate local congestion and improve road network efficiency. It is worth noting that the effectiveness of this technology in a collaborative environment is highly dependent on the scientific nature of the guidance mechanism, the collaborative efficiency of the decision-making architecture, and the accuracy of traffic data. Therefore, the design of efficient and accurate path guidance strategies is the key to achieving congestion control at the city level and has become a research hotspot in the field of intelligent transportation.

[0005] However, existing solutions still have significant flaws in collaborative scenarios. Current mainstream route guidance solutions generally use unified real-time traffic information for global route recommendations, ignoring the mutual influence of vehicle route selection behaviors. This flaw is particularly significant in complex urban traffic scenarios, which may cause vehicles to gather again after adjusting their routes, leading to secondary congestion. At the same time, existing solutions also have problems such as inappropriate intervention timing and imperfect fairness guarantee mechanisms, making it difficult to fully meet the travel needs of vehicles. In summary, the design of route guidance strategies requires breakthroughs in dimensions such as mechanism scientificity, algorithm robustness, and strategy fairness to solve the technical difficulties of imbalanced traffic flow distribution and convergence effect of route selection. Summary of the Invention

[0006] In response to the competitive occupation of urban road network resources caused by the deployment of large-scale vehicles in future transportation systems, and the homogeneity of path selection caused by the interactive influence of vehicle decision-making behaviors, the present invention provides a city-level path induction method and system based on dynamic clustering in a vehicle-road-cloud collaborative scenario; relying on the vehicle-road-cloud collaborative architecture, including the traffic management center, roadside units and vehicles in the cloud, real-time monitoring of traffic congestion status is achieved, and dynamic screening of affected vehicles is completed based on adaptive thresholds; on this basis, vehicles with spatial differences are dynamically clustered through multi-dimensional feature clustering; and then independent subgraphs are constructed, combined with dynamic adjustment of road weights and dynamic penalty mechanisms, to induce vehicles to change paths to solve urban-level regional congestion problems, and ultimately achieve reasonable traffic flow distribution, improved road network traffic efficiency and balanced allocation of spatiotemporal resources.

[0007] Explanation of terms:

[0008] 1. Traffic Management Center: The core control center in the traffic management field, usually composed of cloud servers or intelligent systems, is responsible for integrating traffic data, monitoring road network status, generating scheduling strategies, and optimizing traffic flow distribution through real-time analysis and decision-making to improve road network efficiency.

[0009] 2. Roadside Units: Intelligent terminal devices deployed on road infrastructure (such as streetlights and traffic lights). They have sensing, communication, and computing functions. They are used to collect road environment data (such as traffic volume and road conditions) and exchange information with vehicles and cloud systems to support the Internet of Vehicles and intelligent transportation applications.

[0010] 3. Greenshields Linear Traffic Model: A classic macroscopic model in traffic flow theory, it reveals a linear relationship between vehicle speed and vehicle density on a road, where speed decreases linearly with increasing density. This model quantifies the relationship between road capacity and congestion by parameterizing free-flow speed and critical density.

[0011] 4. DBSCAN algorithm: A density-based clustering analysis method that divides densely connected samples in the data space into the same cluster by defining a neighborhood radius and a minimum number of samples. It can effectively identify non-convex cluster structures and is robust to noise points. It is widely used in spatial data mining and pattern recognition.

[0012] 5. A* algorithm: A heuristic algorithm used for path search in computer science. It expands nodes based on a priority queue and ensures the shortest path or optimal solution in a directed graph by combining the actual cost g(n) from the starting point to the current node with the estimated cost h(n) from the current node to the end point. It is complete and efficient.

[0013] The technical solutions of the present invention are as follows:

[0014] A first aspect of the present invention provides a city-level path guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario, comprising:

[0015] Step 1: Build a vehicle-road-cloud collaborative architecture to collect traffic data and upload it to the traffic management center in the cloud;

[0016] Step 2: The traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection;

[0017] Step 3: When a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion and form a set of vehicles to be planned;

[0018] Step 4: Perform spatial clustering on the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster to achieve decision isolation between vehicle clusters;

[0019] Step 5: Assign priorities to vehicles within each cluster to lay the foundation for subsequent path planning;

[0020] Step 6: Process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths (new driving paths that replace the original paths) for the vehicles within the vehicle cluster according to the calculated priority order;

[0021] Step 7: Use the buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph to alleviate the phenomenon of vehicle path selection convergence;

[0022] Step 8: Repeat steps 6 and 7 until the path planning of all vehicles to be planned is completed;

[0023] Step 9: The traffic management center sends the planned alternative routes to the corresponding vehicles through the roadside units. The vehicles follow the new routes to achieve route guidance and solve the congestion problem in urban areas.

[0024] According to a preferred embodiment of the present invention, a vehicle-road-cloud collaborative architecture is constructed to collect traffic data and upload the collected traffic data to a traffic management center in the cloud; including:

[0025] The vehicle-road-cloud collaborative architecture includes roadside units, a traffic management center in the cloud, and vehicles. The urban road network is modeled as a directed graph, represented as in, represents the node set, i.e. the intersection set, and N is the total number of nodes; Represents a set of directed edges, i.e., a set of road segments, where e ij Indicates road section, n i 、n jRepresent any two nodes; a roadside unit is deployed in the middle of each road to collect the vehicle status of the road covered by the roadside unit and forward road network and decision-making information; the roadside unit collects traffic data of the covered road section, including the number of vehicles in each road section at the current moment, the real-time position coordinates of each vehicle, the starting point of each vehicle, and time sensitivity; among them, time sensitivity is used to characterize the urgency of vehicle travel tasks (such as rescue, emergency, commuting, logistics, etc.) and is an important basis for constructing the priority sorting mechanism in this invention; this indicator is set using discrete integer levels and is divided into 5 levels with a value range of {1, 2, 3, 4, 5}. The larger the value, the more urgent the task undertaken by the vehicle; and the collected data is uploaded to the traffic management center in the cloud.

[0026] According to a preferred embodiment of the present invention, the traffic management center updates the road weights and initializes the penalty matrix; including:

[0027] First, for any time t∈{1, 2, ..., T}, find any road segment e ij The vehicle density is as follows:

[0028]

[0029] Among them, k ij (t) represents the vehicle density of the road section at time t, N ij (t) represents the number of vehicles on the road section at time t; L ij Indicates the length of the road segment; is the number of lanes included in the road section;

[0030] Based on the Greenshields linear traffic model, the speed of vehicles on the road section is calculated as follows:

[0031]

[0032] Among them, u ij (t) represents the speed of the vehicle on the road section at time t, u f is the free flow velocity, and is the maximum vehicle density of the road section, that is, the density when the road is completely blocked and vehicles cannot move; Indicates the maximum number of vehicles that the road section can accommodate;

[0033] Based on the above vehicle speed, estimate the road segment e ij The travel delay (estimated travel delay) is as follows:

[0034]

[0035] The estimated travel delay τ of each road section ij (t) is used as the road weight to construct the road weight matrix Where N is the number of nodes, and the travel cost of each road section is quantitatively described. Represents a real number matrix; the road weight matrix includes elements w ij (t), element w ij (t) corresponds to road section e ij The estimated travel delay, that is, w ij (t) = τ ij (t);

[0036] Then, in order to further measure the long-term impact of the planned vehicle path on the road network, a dynamic road penalty matrix is ​​introduced Each element in the penalty matrix Defined as:

[0037]

[0038] in, Represents the set of vehicles that have completed path planning in a path planning task; is the indicator function, if the road segment e ij If it exists in the planned path of vehicle v, the value is 1, otherwise it is 0; η v Indicates the priority of vehicle v, Path v represents the path of vehicle v; the penalty value reflects the distribution of the planned vehicle's path selection. In the subsequent path planning process, it can be introduced into the path planning cost function to increase the cost of the corresponding road segment, thereby suppressing the homogeneity tendency of vehicle path decisions;

[0039] Finally, the directed graph of the urban road network is expanded into a dynamic time-varying graph, where the road weight and penalty value matrix changes over time, expressed as:

[0040]

[0041] Among them, W(t) and are the travel cost (estimated travel delay) and penalty value of the road section in time slot t, and Graph(t) represents the directed graph of time slot t;

[0042] Further preferably, the cloud-based traffic management center uses the vehicle density information of each road section to achieve real-time detection of congested road sections, including:

[0043] For any road segment e ij , when the traffic density k ij (t) reaches or exceeds the predefined congestion judgment threshold k m When , it is determined to be a congested road section; establish a binary congestion discrimination matrix M con ∈{0, 1} N×N , the elements in the congestion discrimination matrix are defined as:

[0044]

[0045] in, Indicates road segment e ij The congestion discrimination factor indicates whether the road section is in a congested state; the congestion discrimination matrix quantitatively reflects the current congestion state of each road section; in the preset detection period T detect At the beginning of each cycle, the traffic conditions of the entire road network are detected and the congestion discrimination matrix is ​​updated, thereby providing timely and accurate congestion information for subsequent route guidance and vehicle scheduling.

[0046] According to a preferred embodiment of the present invention, when a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion to form a set of vehicles to be planned; including:

[0047] After completing the congestion detection, the traffic management center will screen out vehicles that may be affected by the congestion based on their real-time location and current path. Specifically, for any vehicle Represents the set of vehicles in the road network at the current moment; if there is any road segment e in the current path ij satisfy And vehicle v and congested road section e ij The distance between them is less than the distance threshold l m When , vehicle v is added to the re-planning set In the process, a set of vehicles to be planned is formed;

[0048] In order to accurately reflect the impact of road congestion on vehicle route selection, an adaptive distance threshold based on congestion intensity is proposed; the distance threshold l m As shown in the following formula:

[0049]

[0050] in, is the initial threshold, k ij (t) represents the vehicle density of the road section at time t, is the maximum vehicle density of the road section; Formula (7) dynamically expands the detection range through the congestion intensity of the road section. When the threshold value l m will approach This allows for more flexible identification of vehicles that require rerouting during the route planning process.

[0051] Preferably, according to the present invention, spatial clustering is performed on the set of vehicles to be planned to obtain multiple vehicle clusters, and a path planning subgraph with an independent road weight matrix and penalty matrix is ​​constructed for each vehicle cluster to achieve decision isolation between vehicle clusters; including:

[0052] Traditional route guidance methods ignore the interactive impact of vehicle decision-making behaviors, which can easily lead to secondary clustering and congestion. To improve route guidance, one optimization method dynamically adjusts road weights based on planning results to influence subsequent vehicle route selection. However, due to differences in vehicle destination locations and travel times, a unified right-of-way adjustment strategy often causes decision interference between vehicles, reducing the effectiveness of route guidance.

[0053] To overcome these issues, the present invention proposes a path optimization architecture based on dynamic clustering and independent path planning subgraphs. This architecture not only reduces the convergence of vehicle path selection but also effectively eliminates decision interference between unrelated vehicles. Specifically, because the mutual influence of vehicle path decisions is positively correlated with their spatial proximity, destination similarity, and directional consistency, the present invention clusters the planned vehicles based on spatial features, including their real-time location coordinates, Euclidean distance from the destination, and driving direction angle. This process not only isolates spatial decisions but also decomposes the large-scale path planning problem into independent subproblems to support intra-cluster collaborative optimization. The clustering steps are as follows:

[0054] (1) Constructing the feature vector: For any vehicle v in the set of vehicles to be planned, the feature vector is constructed as follows:

[0055] f v =[p v ,d v ,θ v ] T ; (8)

[0056] Among them, f v represents the feature vector of vehicle v, p v =(x v ,y v ) represents the real-time position coordinates of vehicle v, x v 、y v They represent the horizontal and vertical coordinates of the vehicle in the road network coordinate system (Cartesian coordinates established with the center point of the road network as the origin, used to uniquely identify the positional relationship between each node and edge in the transportation network), d v =|p v -d v |2 represents the Euclidean distance from the vehicle to the destination, represents the destination coordinates of vehicle v, Respectively represent the horizontal and vertical coordinates of the vehicle's destination in the road network coordinate system, Indicates the direction angle of the vehicle to the destination;

[0057] (2) Normalization: To eliminate the dimensional differences between the indicators, the vehicle position coordinates p are normalized according to the following formula: v=(x v ,y v ) and the Euclidean distance d to the end point v Perform normalization:

[0058]

[0059] in, and Represents x v 、y v d v The normalized value of x min 、x max They represent the minimum and maximum values ​​of the horizontal coordinates of all vehicle positions in the set of vehicles to be planned, y min 、y max Respectively represent the minimum and maximum values ​​of the vertical coordinates of all vehicle positions in the set of vehicles to be planned, Represents the set of vehicles to be planned at the current moment;

[0060] The normalized eigenvector is recorded as:

[0061]

[0062] (3) Calculate the distance metric matrix: On this basis, in order to measure the similarity between vehicles, a comprehensive distance metric based on multi-index weighted fusion is designed. For any two vehicles v i and v j , the comprehensive distance is defined as:

[0063]

[0064] Among them, D(v i , v j ) represents vehicle v i and v j The comprehensive distance, and Represents vehicle v i and v j Normalized real-time position coordinates, is the normalized Euclidean distance of the endpoint, θ i ,θ j is the direction angle from the vehicle to the destination, w1, w2, and w3 are the weight parameters of position, distance, and direction indicators respectively;

[0065] Traverse all vehicle pairs (v i , v j ), calculate the comprehensive distance according to formula (11); finally, organize the comprehensive distances of all vehicle pairs into an M×M distance metric matrix D, is the number of vehicles to be planned; the distance metric matrix element D(i, j) corresponds to vehicle v i and v j The comprehensive distance, element D(i, i) = 0 represents the distance from each vehicle to itself;

[0066] (4) Using the DBSCAN algorithm to divide vehicles into multiple clusters: Based on the distance metric matrix D, the DBSCAN algorithm first traverses the vehicles to be planned based on the set neighborhood radius ε and the minimum sample number MinPts. If a vehicle contains no less than MinPts vehicles in the ε neighborhood, the vehicle is defined as a core point, and all density-reachable vehicles in the neighborhood are clustered into a cluster; vehicles that are not core points and are not included in any core point neighborhood are regarded as noise points, and finally the vehicles are divided into multiple clusters; the internal vehicles of each cluster have high consistency in spatial position, terminal distance and driving direction, thereby achieving collaborative optimization of path planning within a local range;

[0067] An independent path planning subgraph is constructed for each cluster. The path planning subgraph has its own road weight and penalty matrix. By combining it with the subsequent penalty value and the dynamic adjustment mechanism of road rights, it can effectively alleviate the phenomenon of vehicle path selection convergence. The path planning subgraph is represented as:

[0068]

[0069] Among them, Graph z (t) represents the path planning subgraph of vehicle cluster z at time slot t, W z (t) represents the road weight matrix of the path planning subgraph of time slot t, represents the penalty matrix of the path planning subgraph for time slot t;

[0070] Further preferably, based on the vehicle's destination distance and time sensitivity, priority is assigned to vehicles within each vehicle cluster, laying the foundation for subsequent path planning; including:

[0071] Within each vehicle cluster, prioritizing vehicles is necessary to rationally allocate planning resources and ensure that the needs of different vehicles are properly met. To this end, the present invention prioritizes vehicles based on two key principles to optimize the planning process. First, the more time-sensitive a vehicle is, the higher its priority. This is because vehicles performing emergency tasks (such as ambulances and fire trucks) require priority in planning resources. Second, the smaller the Euclidean distance from the vehicle to the destination, the higher its priority. This is because vehicles closer to the destination have fewer paths to choose from, and prioritizing planning effectively reduces the risk of overlapping paths with other vehicles.

[0072] For any vehicle v in any vehicle cluster z, the priority η v As shown in the following formula:

[0073]

[0074] Among them, ξ v ∈{1, 2, 3, 4, 5} represents the time sensitivity of vehicle v, d v represents the Euclidean distance from the vehicle to the destination, represents the set of all vehicles in vehicle cluster z, and γ is the weight parameter.

[0075] Preferably, according to the present invention, the path planning tasks of each vehicle cluster are processed in parallel, and an improved A* algorithm is used to plan alternative paths (new driving paths that replace the original paths) for the vehicles within the vehicle cluster according to the calculated priority order; the method includes:

[0076] To fully reflect the impact of planned vehicle routes on subsequent route selection, and incorporating real-time traffic information, this paper improves the traditional A* algorithm and constructs an A* path search algorithm with a penalty value. Within the A* algorithm framework, the estimated travel delay of a road section is used as the travel cost, and a road penalty value is introduced to dynamically adjust the road section cost. This effectively guides vehicles to avoid congested sections and alleviates the phenomenon of secondary vehicle gathering.

[0077] First, the cumulative cost function g(n i ) is defined as follows:

[0078]

[0079] Among them, τ jk (t) and Represented as road segment e jk The estimated travel delay and penalty value at time t, μ is the coefficient that weighs the impact of the penalty value, Path i Indicates the distance from the starting node to the current node n i All the sections passed; describes the journey from the starting point to any node n i the actual cumulative cost of

[0080] Then, the heuristic function h(n i ) is defined as:

[0081]

[0082] Among them, d Eucliean (n i , d v ) represents node n i To the vehicle destination d v The Euclidean distance, u f is the free flow speed; the heuristic function approximates the travel time from the current node to the destination as the predicted value of the future cost;

[0083] Finally, the total cost function f(n j ) is expressed as:

[0084] f(n i )=g(n i )+h(n i ); (16)

[0085] In each vehicle cluster, based on the improved A* algorithm, alternative paths from the current position to the vehicle destination are searched for the vehicle in order of priority;

[0086] Based on the definition of the above cost function, a priority queue (minimum heap) strategy is adopted to iteratively expand the total cost function f(n j ) minimum node; Specifically, each time the node with the minimum total cost is selected as the current node, all adjacent nodes of the current node are traversed, and the cumulative cost f(n i ) and the heuristic function value h(n i ), update the total cost function value f(n j ) and record the predecessor node; if the adjacent node has not been visited or a better path is found, it will be added to the priority queue; repeat the above process until the end node is visited; when the end node is visited, trace back the predecessor node sequence from the end node to generate an alternative path that avoids congested sections and comprehensively considers real-time travel costs and penalty values.

[0087] According to the preferred embodiment of the present invention, a buffer threshold trigger mechanism is used to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph to alleviate the phenomenon of vehicle path selection convergence; including:

[0088] In this paper, in order to reduce the system burden caused by frequent updates of road weights and penalty values ​​during the path search process, while taking into account the timeliness and accuracy of the overall update, a batch update mechanism based on relative thresholds is introduced; specifically, for a certain vehicle cluster z, let the current total number of vehicles be When the vehicle completes the path planning, the vehicle is temporarily stored in the buffer zone Q z Medium; passing relative threshold Control the update frequency. When the number of vehicles in the buffer zone meets When an update operation is triggered, the update operation steps are as follows:

[0089] First, for the buffer Q z For each vehicle v in the new driving path (alternative path) planned by vehicle v, the next segment is recorded as e next (v) and construct the set of road sections to be updated For each road segment e in the road segment set to be updated ij , update the road segment e using the following formula ij Number of vehicles in the corresponding routing subgraph:

[0090]

[0091] in, is the indicator function, if e ij =e next (v) then the value of the indicator function is 1, otherwise it is 0; Represents the road segment e in the path planning subgraph of vehicle cluster z ij After the vehicle count is updated, the estimated travel delay of the road section is calculated according to formula (1)-formula (3), and the road weight is updated;

[0092] Then, the planned vehicle set of each vehicle cluster is updated using the following formula:

[0093]

[0094] in, Represents the planned vehicle set; the planned vehicle set is initialized to before it starts planning alternative paths for vehicles in the cluster As the vehicle paths in each vehicle cluster are gradually planned, they continue to expand and eventually include all vehicles in the vehicle cluster. After the planned vehicle set is updated, the penalty value of each road section is adjusted according to formula (4) based on the updated set and the alternative paths of the vehicles in the set.

[0095] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a city-level path guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario.

[0096] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of a city-level path guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario.

[0097] A second aspect of the present invention provides a city-level route guidance system based on dynamic clustering in a vehicle-road-cloud collaboration scenario, comprising:

[0098] The data collection module is configured to: build a vehicle-road-cloud collaborative architecture, collect traffic data, and upload the collected traffic data to the traffic management center in the cloud;

[0099] The congestion detection module is configured as follows: the traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection;

[0100] The vehicle screening module to be planned is configured to: when a congested road section is detected, set an adaptive distance threshold to screen out vehicles that may be affected by congestion to form a set of vehicles to be planned;

[0101] The clustering module is configured to: spatially cluster the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster;

[0102] The priority allocation module is configured to: allocate priorities to vehicles within each vehicle cluster;

[0103] The alternative path planning module is configured to: process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths for the vehicles in the vehicle cluster according to the calculated priority order;

[0104] The batch dynamic adjustment module is configured to: use a buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph;

[0105] The loop execution module is configured to: loop execute the alternative path planning module and the batch dynamic adjustment module until the path planning of all the vehicles to be planned is completed;

[0106] The path distribution module is configured as follows: the traffic management center distributes the planned alternative path to the corresponding vehicle through the roadside unit, and the vehicle drives according to the new route to achieve path guidance.

[0107] The beneficial effects of the present invention are:

[0108] 1. The present invention constructs a complete vehicle-road-cloud collaborative path induction framework, covering all aspects from real-time traffic data collection, congestion detection, vehicle screening to path planning and dynamic update, realizing real-time feedback and dynamic regulation of traffic status.

[0109] 2. This invention employs a dynamic vehicle clustering strategy, combined with independent path planning subgraph construction, to decompose the global problem into multiple local subproblems, enabling parallel computation. Furthermore, through dynamic adjustment of right-of-way and penalty values, it fully considers the mutual influence of vehicles, reduces the tendency for path selection to be similar, and fully utilizes road resources.

[0110] 3. This invention introduces a road penalty matrix generated based on planned vehicle paths and integrates it into the A* algorithm, so that the cumulative cost function can simultaneously reflect the real-time travel cost and the impact of the road load of planned vehicles, effectively guiding vehicles to avoid congested sections and achieving more reasonable and efficient path planning.

[0111] 4. When selecting vehicles that need to be replanned, the present invention constructs an adaptive threshold based on the congestion level of the road section to achieve dynamic screening of affected vehicles, effectively avoiding the defects of premature intervention or delayed response that may exist in existing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] Figure 1 This is a flow chart of the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario proposed by the present invention;

[0113] Figure 2 It is a road network structure diagram used in simulation verification of the present invention;

[0114] Figure 3 1 is a schematic diagram comparing the average travel time of the method of the present invention and the benchmark method under different traffic volume scales;

[0115] Figure 4 3. It is a schematic diagram comparing the average road load balancing index of the method of the present invention and the benchmark method under different traffic volume scales. DETAILED DESCRIPTION

[0116] The present invention will be further described below with reference to embodiments and accompanying drawings, but is not limited thereto.

[0117] Example 1

[0118] A city-level path guidance method based on dynamic clustering in a vehicle-road-cloud collaborative scenario, such as Figure 1 Shown, including:

[0119] Step 1: Build a vehicle-road-cloud collaborative architecture to collect traffic data and upload it to the traffic management center in the cloud;

[0120] Step 2: The traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection;

[0121] Step 3: When a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion and form a set of vehicles to be planned;

[0122] Step 4: Perform spatial clustering on the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster to achieve decision isolation between vehicle clusters;

[0123] Step 5: Assign priorities to vehicles within each cluster to lay the foundation for subsequent path planning;

[0124] Step 6: Process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths (new driving paths that replace the original paths) for the vehicles within the vehicle cluster according to the calculated priority order;

[0125] Step 7: Use the buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph to alleviate the phenomenon of vehicle path selection convergence;

[0126] Step 8: Repeat steps 6 and 7 until the path planning of all vehicles to be planned is completed;

[0127] Step 9: The traffic management center sends the planned alternative routes to the corresponding vehicles through the roadside units. The vehicles follow the new routes to achieve route guidance and solve the congestion problem in urban areas.

[0128] Example 2

[0129] The difference between the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario described in Example 1 is that:

[0130] Build a vehicle-road-cloud collaborative architecture to collect traffic data and upload it to the cloud-based traffic management center; including:

[0131] The vehicle-road-cloud collaborative architecture includes roadside units, a traffic management center in the cloud, and vehicles. The urban road network is modeled as a directed graph, represented as in, represents the node set, i.e. the intersection set, and N is the total number of nodes; Represents a set of directed edges, i.e., a set of road segments, where e ij Indicates road section, n i 、n j Represent any two nodes; a roadside unit is deployed in the middle of each road to collect the vehicle status of the road covered by the roadside unit and forward road network and decision-making information; the roadside unit collects traffic data of the covered road section, including the number of vehicles in each road section at the current moment, the real-time position coordinates of each vehicle, the starting point of each vehicle, and time sensitivity; among them, time sensitivity is used to characterize the urgency of vehicle travel tasks (such as rescue, emergency, commuting, logistics, etc.) and is an important basis for constructing the priority sorting mechanism in this invention; this indicator is set using discrete integer levels and is divided into 5 levels with a value range of {1, 2, 3, 4, 5}. The larger the value, the more urgent the task undertaken by the vehicle; and the collected data is uploaded to the traffic management center in the cloud.

[0132] The traffic management center updates the road weights and initializes the penalty matrix; including:

[0133] First, for any time t∈{1, 2, ..., T}, find any road segment e ij The vehicle density is as follows:

[0134]

[0135] Among them, k ij (t) represents the vehicle density of the road section at time t, N ij (t) represents the number of vehicles on the road section at time t; L ij Indicates the length of the road segment; is the number of lanes included in the road section;

[0136] Based on the Greenshields linear traffic model, the speed of vehicles on the road section is calculated as follows:

[0137]

[0138] Among them, u ij (t) represents the speed of the vehicle on the road section at time t, u f is the free flow velocity, and is the maximum vehicle density of the road section, that is, the density when the road is completely blocked and vehicles cannot move; Indicates the maximum number of vehicles that the road section can accommodate;

[0139] Based on the above vehicle speed, estimate the road segment e ij The travel delay (estimated travel delay) is as follows:

[0140]

[0141] The estimated travel delay τ of each road section ij (t) is used as the road weight to construct the road weight matrix Where N is the number of nodes, and the travel cost of each road section is quantitatively described. Represents a real number matrix; the road weight matrix includes elements w ij (t), element w ij (t) corresponds to road section e ij The estimated travel delay, that is, w ij (t) = τ ij (t);

[0142] Then, in order to further measure the long-term impact of the planned vehicle path on the road network, a dynamic road penalty matrix is ​​introduced Each element in the penalty matrix Defined as:

[0143]

[0144] in, Represents the set of vehicles that have completed path planning in a path planning task, is the indicator function, if the road segment e ijIf it exists in the planned path of vehicle v, the value is 1, otherwise it is 0; η v Indicates the priority of vehicle v, Path v represents the path of vehicle v; the penalty value reflects the distribution of the planned vehicle's path selection. In the subsequent path planning process, it can be introduced into the path planning cost function to increase the cost of the corresponding road segment, thereby suppressing the homogeneity tendency of vehicle path decisions;

[0145] Finally, the directed graph of the urban road network is expanded into a dynamic time-varying graph, where the road weight and penalty value matrix changes over time, expressed as:

[0146]

[0147] Among them, W(t) and are the travel cost and penalty value of the road section in time slot t, respectively. Graph(t) represents the directed graph of time slot t.

[0148] The cloud-based traffic management center uses vehicle density information on each road section to detect congested sections in real time, including:

[0149] For any road segment e ij , when the traffic density k ij (t) reaches or exceeds the predefined congestion judgment threshold k m When , it is determined to be a congested road section; establish a binary congestion discrimination matrix M con ∈{0, 1} N×N , the elements in the congestion discrimination matrix are defined as:

[0150]

[0151] in, Indicates road segment e ij The congestion discrimination factor indicates whether the road section is in a congested state; the congestion discrimination matrix quantitatively reflects the current congestion state of each road section; in the preset detection period T detect At the beginning of each cycle, the traffic conditions of the entire road network are detected and the congestion discrimination matrix is ​​updated, thereby providing timely and accurate congestion information for subsequent route guidance and vehicle scheduling.

[0152] When a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion and form a set of vehicles to be planned; this includes:

[0153] After completing the congestion detection, the traffic management center will screen out vehicles that may be affected by the congestion based on their real-time location and current path. Specifically, for any vehicle Represents the set of vehicles in the road network at the current moment; if there is any road segment e in the current path ij satisfy And vehicle v and congested road section e ij The distance between them is less than the distance threshold l m When , vehicle v is added to the re-planning set In the process, a set of vehicles to be planned is formed;

[0154] In order to accurately reflect the impact of road congestion on vehicle route selection, an adaptive distance threshold based on congestion intensity is proposed; the distance threshold l m As shown in the following formula:

[0155]

[0156] in, is the initial threshold, k ij (t) represents the vehicle density of the road section at time t, is the maximum vehicle density of the road section; Formula (7) dynamically expands the detection range through the congestion intensity of the road section. When the threshold value l m will approach This allows for more flexible identification of vehicles that require rerouting during the route planning process.

[0157] The set of planned vehicles is spatially clustered to obtain multiple vehicle clusters, and a path planning subgraph with independent road weight matrix and penalty matrix is ​​constructed for each vehicle cluster to achieve decision isolation between vehicle clusters; including:

[0158] Traditional route guidance methods ignore the interactive impact of vehicle decision-making behaviors, which can easily lead to secondary clustering and congestion. To improve route guidance, one optimization method dynamically adjusts road weights based on planning results to influence subsequent vehicle route selection. However, due to differences in vehicle destination locations and travel times, a unified right-of-way adjustment strategy often causes decision interference between vehicles, reducing the effectiveness of route guidance.

[0159] To overcome these issues, the present invention proposes a path optimization architecture based on dynamic clustering and independent path planning subgraphs. This architecture not only reduces the convergence of vehicle path selection but also effectively eliminates decision interference between unrelated vehicles. Specifically, because the mutual influence of vehicle path decisions is positively correlated with their spatial proximity, destination similarity, and directional consistency, the present invention clusters the planned vehicles based on spatial features, including their real-time location coordinates, Euclidean distance from the destination, and driving direction angle. This process not only isolates spatial decisions but also decomposes the large-scale path planning problem into independent subproblems to support intra-cluster collaborative optimization. The clustering steps are as follows:

[0160] (1) Constructing the feature vector: For any vehicle v in the set of vehicles to be planned, the feature vector is constructed as follows:

[0161] f v =[p v ,d v ,θ v ] T ; (26)

[0162] Among them, f v represents the feature vector of vehicle v, p v =(x v ,y v ) represents the real-time position coordinates of vehicle v, x v 、y v They represent the horizontal and vertical coordinates of the vehicle in the road network coordinate system (Cartesian coordinates established with the center point of the road network as the origin, used to uniquely identify the positional relationship between each node and edge in the transportation network), d v =|p v -d v |2 represents the Euclidean distance from the vehicle to the destination, represents the destination coordinates of vehicle v, Respectively represent the horizontal and vertical coordinates of the vehicle's destination in the road network coordinate system, Indicates the direction angle of the vehicle to the destination;

[0163] (2) Normalization: To eliminate the dimensional differences between the indicators, the vehicle position coordinates p are normalized according to the following formula: v =(x v ,y v ) and the Euclidean distance d to the end point v Perform normalization:

[0164]

[0165] in, and Represents x v 、y v d v The normalized value of x min 、x max They represent the minimum and maximum values ​​of the horizontal coordinates of all vehicle positions in the set of vehicles to be planned, y min 、y max Respectively represent the minimum and maximum values ​​of the vertical coordinates of all vehicle positions in the set of vehicles to be planned, Represents the set of vehicles to be planned at the current moment;

[0166] The normalized eigenvector is recorded as:

[0167]

[0168] (3) Calculate the distance metric matrix: On this basis, in order to measure the similarity between vehicles, a comprehensive distance metric based on multi-index weighted fusion is designed. For any two vehicles v i and v j , the comprehensive distance is defined as:

[0169]

[0170] Among them, D(v i , v j ) represents vehicle v i and v j The comprehensive distance, and Represents vehicle v i and v j Normalized real-time position coordinates, is the normalized Euclidean distance of the endpoint, θ i ,θ j is the direction angle from the vehicle to the destination, w1, w2, and w3 are the weight parameters of position, distance, and direction indicators respectively;

[0171] Traverse all vehicle pairs (v i , v j ), calculate the comprehensive distance according to formula (11); finally, organize the comprehensive distances of all vehicle pairs into an M×M distance metric matrix D, is the number of vehicles to be planned; the distance metric matrix element D(i, j) corresponds to vehicle v i and v j The comprehensive distance, element D(i, i) = 0 represents the distance from each vehicle to itself;

[0172] (4) Using the DBSCAN algorithm to divide vehicles into multiple clusters: Based on the distance metric matrix D, the DBSCAN algorithm first traverses the vehicles to be planned based on the set neighborhood radius ε and the minimum sample number MinPts. If a vehicle contains no less than MinPts vehicles in the ε neighborhood, the vehicle is defined as a core point, and all density-reachable vehicles in the neighborhood are clustered into a cluster; vehicles that are not core points and are not included in any core point neighborhood are regarded as noise points, and finally the vehicles are divided into multiple clusters; the internal vehicles of each cluster have high consistency in spatial position, terminal distance and driving direction, thereby achieving collaborative optimization of path planning within a local range;

[0173] An independent path planning subgraph is constructed for each cluster. The path planning subgraph has its own road weight and penalty matrix. By combining it with the subsequent penalty value and the dynamic adjustment mechanism of road rights, it can effectively alleviate the phenomenon of vehicle path selection convergence. The path planning subgraph is represented as:

[0174]

[0175] Among them, Graph z (t) represents the path planning subgraph of vehicle cluster z at time slot t, W z (t) represents the road weight matrix of the path planning subgraph of time slot t, Represents the penalty matrix of the path planning subgraph for time slot t.

[0176] Assigning priorities to vehicles within each cluster based on their destination distance and time sensitivity, laying the foundation for subsequent path planning; including:

[0177] Within each vehicle cluster, prioritizing vehicles is necessary to rationally allocate planning resources and ensure that the needs of different vehicles are properly met. To this end, the present invention prioritizes vehicles based on two key principles to optimize the planning process. First, the more time-sensitive a vehicle is, the higher its priority. This is because vehicles performing emergency tasks (such as ambulances and fire trucks) require priority in planning resources. Second, the smaller the Euclidean distance from the vehicle to the destination, the higher its priority. This is because vehicles closer to the destination have fewer paths to choose from, and prioritizing planning effectively reduces the risk of overlapping paths with other vehicles.

[0178] For any vehicle v in any vehicle cluster z, the priority η v As shown in the following formula:

[0179]

[0180] Among them, ξ v ∈{1, 2, 3, 4, 5} represents the time sensitivity of vehicle v, d v represents the Euclidean distance from the vehicle to the destination, represents the set of all vehicles in vehicle cluster z, and γ is the weight parameter;

[0181] The path planning tasks of each vehicle cluster are processed in parallel. Within the vehicle cluster, an improved A* algorithm is used to plan alternative paths (new driving paths that replace the original paths) for the vehicles according to the calculated priority order. This includes:

[0182] To fully reflect the impact of planned vehicle routes on subsequent route selection, and incorporating real-time traffic information, this paper improves the traditional A* algorithm and constructs an A* path search algorithm with a penalty value. Within the A* algorithm framework, the estimated travel delay of a road section is used as the travel cost, and a road penalty value is introduced to dynamically adjust the road section cost. This effectively guides vehicles to avoid congested sections and alleviates the phenomenon of secondary vehicle gathering.

[0183] First, the cumulative cost function g(n i ) is defined as follows:

[0184]

[0185] Among them, τ jk (t) and Represented as road segment e jk The estimated travel delay and penalty value at time t, μ is the coefficient that weighs the impact of the penalty value, Path i Indicates the distance from the starting node to the current node n i All the sections passed; describes the journey from the starting point to any node n i the actual cumulative cost of

[0186] Then, the heuristic function h(n i ) is defined as:

[0187]

[0188] Among them, d Euclidean (n i , d v ) represents node n i To the vehicle destination d v The Euclidean distance, u f is the free flow speed; the heuristic function approximates the travel time from the current node to the destination as the predicted value of the future cost;

[0189] Finally, the total cost function f(n j ) is expressed as:

[0190] f(n i )=g(n i )+h(n i ); (34)

[0191] In each vehicle cluster, based on the improved A* algorithm, alternative paths from the current position to the vehicle destination are searched for the vehicle in order of priority;

[0192] Based on the definition of the above cost function, a priority queue (minimum heap) strategy is adopted to iteratively expand the total cost function f(nj ) minimum node; Specifically, each time the node with the minimum total cost is selected as the current node, all adjacent nodes of the current node are traversed, and the cumulative cost g(n i ) and the heuristic function value h(n i ), update the total cost function value f(n j ) and record the predecessor node; if the adjacent node has not been visited or a better path is found, it will be added to the priority queue; repeat the above process until the end node is visited; when the end node is visited, trace back the predecessor node sequence from the end node to generate an alternative path that avoids congested sections and comprehensively considers real-time travel costs and penalty values.

[0193] A buffer threshold trigger mechanism is used to dynamically adjust road weights and penalty values ​​in batches on the path planning subgraph to alleviate the phenomenon of vehicle path selection convergence; this includes:

[0194] In this paper, in order to reduce the system burden caused by frequent updates of road weights and penalty values ​​during the path search process, while taking into account the timeliness and accuracy of the overall update, a batch update mechanism based on relative thresholds is introduced; specifically, for a certain vehicle cluster z, let the current total number of vehicles be When the vehicle completes the path planning, the vehicle is temporarily stored in the buffer zone Q z Medium; passing relative threshold Control the update frequency. When the number of vehicles in the buffer zone meets When an update operation is triggered, the update operation steps are as follows:

[0195] First, for the buffer Q z For each vehicle v in the new driving path (alternative path) planned by vehicle v, the next segment is recorded as e next (v) and construct the set of road sections to be updated For each road segment e in the road segment set to be updated ij , update the road segment e using the following formula ij Number of vehicles in the corresponding routing subgraph:

[0196]

[0197] in, is the indicator function, if e ij =e next (v) then the value of the indicator function is 1, otherwise it is 0; Represents the road segment e in the path planning subgraph of vehicle cluster z ij After the vehicle count is updated, the estimated travel delay of the road section is calculated according to formula (1)-formula (3), and the road weight is updated;

[0198] Then, the planned vehicle set of each vehicle cluster is updated using the following formula:

[0199]

[0200] in, Represents the planned vehicle set; the planned vehicle set is initialized to before it starts planning alternative paths for vehicles in the cluster As the vehicle paths in each vehicle cluster are gradually planned, they continue to expand and eventually include all vehicles in the vehicle cluster. After the planned vehicle set is updated, the penalty value of each road section is adjusted according to formula (4) based on the updated set and the alternative paths of the vehicles in the set.

[0201] In order to verify the effectiveness of the method of the present invention, based on the road network data of Xuancheng City, Anhui Province and the starting and ending point information of some vehicles, three simulation scenarios with traffic volumes of 16,000, 19,000 and 22,000 were constructed, and compared with two benchmark solutions. The traffic environment simulation was completed on the SUMO1.21.0 platform, and the TraCI interface was used to achieve real-time interaction with the environment, complete the acquisition of traffic status information and the implementation of path guidance. In all scenarios, the departure time of the vehicle is randomly generated according to the Poisson distribution within 0-3600 seconds, and the road network structure is as follows: Figure 2 shown.

[0202] The comparison methods are:

[0203] 1) Static shortest path method: The Dijkstra algorithm is used to calculate the shortest path during the initial vehicle path planning, and no path adjustment is performed during subsequent driving.

[0204] 2) Shortest Time Guidance Method: Utilizes the congestion detection mechanism to trigger route guidance, and uses the time-cost A* algorithm to independently generate alternative routes. After the vehicle changes its route, the road weight is not updated after the route adjustment.

[0205] Table 1 shows the performance indicators of various methods when the traffic volume is 16,000. The time loss measures the delay caused by factors such as traffic lights or road congestion; the total number of congested sections refers to the sum of the number of congested sections sampled every 100 time steps from the 1000th time step to the 7000th time step of the simulation; the average road load balancing index is based on the formula The time average of the load balancing metric calculated every 100 time steps.

[0206] Table 1

[0207] method Driving time (s) Driving distance (m) Time loss(s) Road load balancing indicators Total number of congested road sections Static shortest path method 622.13-0％ 5113.18+0％ 247.30-0％ 5.78-0％ 299-0％ Shortest induction method 525.21-15.58％ 5134.54+0.42％ 148.93-39.78％ 2.65-54.15％ 81-72.91％ Method of the present invention 513.17-17.51％ 5131.17+0.34％ 137.14-44.55％ 2.07-64.19％ 12-95.99％

[0208] The data in the table show that the method of the present invention is significantly superior to the static shortest path method in key indicators such as average travel time, average time loss, and the total number of congested sections. This shows that the method of the present invention can effectively alleviate road network congestion and improve overall traffic efficiency. At the same time, it also shows a more obvious performance improvement compared with the shortest time induction method, further indicating that the present invention can effectively disperse vehicle path selection and reduce the probability of secondary aggregation and secondary congestion by dynamically updating road weights and introducing a penalty value mechanism.

[0209] Attachment Figure 3 The paper presents a comparison of average travel times between the proposed method and a baseline method under varying traffic volumes. As traffic increases and the road network approaches saturation, the baseline method's travel time increases significantly. However, the proposed method, leveraging dynamic guidance and local collaborative path optimization, maintains relatively low average travel times in high-traffic scenarios, demonstrating strong scalability and robustness.

[0210] Attachment Figure 4 The paper presents a comparison of the average road load balancing index of the proposed method and the baseline method under different traffic volume scales. The results show that the proposed method can maintain a lower load balancing index in all traffic volume scenarios, demonstrating its significant advantage in distributing traffic and avoiding overuse of certain key road sections.

[0211] Example 3

[0212] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario described in Example 1 or 2.

[0213] Example 4

[0214] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario described in Example 1 or 2 are implemented.

[0215] Example 5

[0216] A city-level route guidance system based on dynamic clustering in a vehicle-road-cloud collaboration scenario, including:

[0217] The data collection module is configured to: build a vehicle-road-cloud collaborative architecture, collect traffic data, and upload the collected traffic data to the traffic management center in the cloud;

[0218] The congestion detection module is configured as follows: the traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection;

[0219] The vehicle screening module to be planned is configured to: when a congested road section is detected, set an adaptive distance threshold to screen out vehicles that may be affected by congestion to form a set of vehicles to be planned;

[0220] The clustering module is configured to: spatially cluster the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster;

[0221] The priority allocation module is configured to: allocate priorities to vehicles within each vehicle cluster;

[0222] The alternative path planning module is configured to: process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths for the vehicles in the vehicle cluster according to the calculated priority order;

[0223] The batch dynamic adjustment module is configured to: use a buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph;

[0224] The loop execution module is configured to: loop execute the alternative path planning module and the batch dynamic adjustment module until the path planning of all the vehicles to be planned is completed;

[0225] The path distribution module is configured as follows: the traffic management center distributes the planned alternative path to the corresponding vehicle through the roadside unit, and the vehicle drives according to the new route to achieve path guidance.

Claims

1. A city-level path guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario, characterized by: include: Step 1: Build a vehicle-road-cloud collaborative architecture to collect traffic data and upload it to the traffic management center in the cloud; Step 2: The traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection; Step 3: When a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion and form a set of vehicles to be planned; Step 4: Perform spatial clustering on the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster; Step 5: Assign priorities to vehicles within each vehicle cluster; Step 6: Process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths for the vehicles within the cluster according to the calculated priority order; Step 7: Use the buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph; Step 8: Repeat steps 6 and 7 until the path planning of all vehicles to be planned is completed; Step 9: The traffic management center sends the planned alternative route to the corresponding vehicle through the roadside unit, and the vehicle drives according to the new route to achieve path guidance.

2. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 1 is characterized in that: Build a vehicle-road-cloud collaborative architecture to collect traffic data and upload it to the cloud-based traffic management center; including: The vehicle-road-cloud collaborative architecture includes roadside units, a traffic management center in the cloud, and vehicles. The urban road network is modeled as a directed graph, represented as in, represents the node set, i.e. the intersection set, and N is the total number of nodes; Represents a set of directed edges, i.e., a set of road segments, where e ij Indicates road section, n i 、n j Represent any two nodes; deploy roadside units in the middle of each road; the roadside units collect traffic data on the covered road sections, including the number of vehicles on each road section at the current moment, the real-time position coordinates of each vehicle, the starting point of each vehicle, and time sensitivity, and upload the collected data to the traffic management center in the cloud.

3. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 2 is characterized in that: The traffic management center updates the road weights and initializes the penalty matrix; including: First, for any time t∈{1, 2, ..., T}, find any road segment e ij The vehicle density is as follows: Among them, k ij (t) represents the vehicle density of the road section at time t, N ij (t) represents the number of vehicles on the road section at time t; L ij Indicates the length of the road segment; is the number of lanes included in the road section; Based on the Greenshields linear traffic model, the speed of vehicles on the road section is calculated as follows: Among them, u ij (t) represents the speed of the vehicle on the road section at time t, u f is the free flow velocity, and is the maximum vehicle density of the road section, that is, the density when the road is completely blocked and vehicles cannot move; Indicates the maximum number of vehicles that the road section can accommodate; Based on the above vehicle speed, estimate the road segment e ij The travel delay is as follows: The estimated travel delay η of each road section ij (t) is used as the road weight to construct the road weight matrix Where N is the number of nodes, Represents a real number matrix; the road weight matrix includes elements w ij (t), element w ij (t) corresponds to road section e ij The estimated travel delay, that is, w ij (t) = τ ij (t); Then, the dynamic road penalty matrix is ​​introduced Each element in the penalty matrix Defined as: Among them, v planned (t) represents the set of vehicles that have completed path planning in a path planning task, is the indicator function, if the road segment e ij If it exists in the planned path of vehicle v, the value is 1, otherwise it is 0; η v Indicates the priority of vehicle v, Path v represents the path of vehicle v; Finally, the directed graph of the urban road network is expanded into a dynamic time-varying graph, where the road weight and penalty value matrix changes over time, expressed as: Among them, W(t) and are the travel cost and penalty value of the road section in time slot t, respectively. Graph(t) represents the directed graph of time slot t. Further preferably, the cloud-based traffic management center uses the vehicle density information of each road section to achieve real-time detection of congested road sections, including: For any road segment e ij , when the traffic density k ij (t) reaches or exceeds the predefined congestion judgment threshold k m When , it is determined to be a congested road section; establish a binary congestion discrimination matrix M con ∈{0, 1} N×N , the elements in the congestion discrimination matrix are defined as: in, Indicates road segment e ij The congestion discrimination factor indicates whether the road section is in a congested state.

4. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 3 is characterized in that: When a congested road section is detected, an adaptive distance threshold is set to filter out vehicles that may be affected by the congestion and form a set of vehicles to be planned; this includes: For any vehicle Represents the set of vehicles in the road network at the current moment; if there is any road segment e in the current path ij satisfy And vehicle v and congested road section e ij The distance between them is less than the distance threshold l m When , vehicle v is added to the re-planning set In the process, a set of vehicles to be planned is formed; Distance threshold l m As shown in the following formula: in, is the initial threshold, k ij (t) represents the vehicle density of the road section at time t, is the maximum vehicle density of the road section.

5. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 4 is characterized in that: The set of planned vehicles is spatially clustered to obtain multiple vehicle clusters, and a path planning subgraph with independent road weight matrix and penalty matrix is ​​constructed for each vehicle cluster; including: Based on spatial features, including the vehicle's real-time location coordinates, Euclidean distance from the destination, and driving direction angle, the planned vehicles are clustered. The clustering steps are as follows: (1) Constructing the feature vector: For any vehicle v in the set of vehicles to be planned, the feature vector is constructed as follows: f v =[p v ,d v ,θ v ] T ; (8) Among them, f v represents the feature vector of vehicle v, p v =(x v ,y v ) represents the real-time position coordinates of vehicle v, x v 、y v Respectively represent the horizontal and vertical coordinates of the vehicle in the road network coordinate system, d v =|p v -d v |2 represents the Euclidean distance from the vehicle to the destination, represents the destination coordinates of vehicle v, Respectively represent the horizontal and vertical coordinates of the vehicle's destination in the road network coordinate system, Indicates the direction angle of the vehicle to the destination; (2) Normalization: The vehicle position coordinate p is normalized according to the following formula: v =(x v ,y v ) and the Euclidean distance d to the end point v Perform normalization: in, and Represents x v 、y v d v The normalized value of x min 、x max They represent the minimum and maximum values ​​of the horizontal coordinates of all vehicle positions in the set of vehicles to be planned, y min 、y max Respectively represent the minimum and maximum values ​​of the vertical coordinates of all vehicle positions in the set of vehicles to be planned, Represents the set of vehicles to be planned at the current moment; The normalized eigenvector is recorded as: (3) Calculate the distance metric matrix: For any two vehicles v i and v j , the comprehensive distance is defined as: Among them, D(v i , v j ) represents vehicle v i and v j The comprehensive distance, and Represents vehicle v i and v j Normalized real-time position coordinates, is the normalized Euclidean distance of the endpoint, θ i ,θ j is the direction angle from the vehicle to the destination, w1, w2, and w3 are the weight parameters of position, distance, and direction indicators respectively; Traverse all vehicle pairs (v i , v j ), calculate the comprehensive distance according to formula (11); finally, organize the comprehensive distances of all vehicle pairs into an M×M distance metric matrix D, is the number of vehicles to be planned; the distance metric matrix element D(i, j) corresponds to vehicle v i and v j The comprehensive distance, element D(i, j) = 0 represents the distance from each vehicle to itself; (4) Using the DBSCAN algorithm to divide vehicles into multiple clusters: Based on the distance metric matrix D, the DBSCAN algorithm first traverses the vehicles to be planned based on the set neighborhood radius ε and the minimum sample number MinPts. If a vehicle contains no less than MinPts vehicles in the ε neighborhood, the vehicle is defined as a core point, and all density-reachable vehicles in the neighborhood are clustered into a cluster; vehicles that are not core points and are not included in any core point neighborhood are considered noise points, and finally the vehicles are divided into multiple clusters; An independent path planning subgraph is constructed for each cluster. The path planning subgraph has its own road weight and penalty matrix. The path planning subgraph is represented as: Among them, Graph z (t) represents the path planning subgraph of vehicle cluster z in time slot t, W z (t) represents the road weight matrix of the path planning subgraph of time slot t, represents the penalty matrix of the path planning subgraph for time slot t; Further preferably, priority is assigned to vehicles within each vehicle cluster based on the vehicle's destination distance and time sensitivity; including: For any vehicle v in any vehicle cluster z, the priority η v As shown in the following formula: Among them, ξ v ∈{1, 2, 3, 4, 5} represents the time sensitivity of vehicle v, d v represents the Euclidean distance from the vehicle to the destination, represents the set of all vehicles in vehicle cluster z, and γ is the weight parameter.

6. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 5 is characterized in that: The path planning tasks of each vehicle cluster are processed in parallel. An improved A* algorithm is used to plan alternative paths for vehicles within the vehicle cluster according to the calculated priority order. This includes: First, the cumulative cost function g(n i ) is defined as follows: Among them, τ jk (t) and Represented as road segment e jk The estimated travel delay and penalty value at time t, μ is the coefficient that weighs the impact of the penalty value, Path i Indicates the distance from the starting node to the current node n i All the sections of the road passed; Then, the heuristic function h(n i ) is defined as: Among them, d Euclidean (n i , d v ) represents node n i To the vehicle destination d v The Euclidean distance, u f is the free flow speed; Finally, the total cost function f(n j ) is expressed as: f(n i )=g(n i )+h(n i ); (16) In each vehicle cluster, based on the improved A* algorithm, alternative paths from the current position to the vehicle destination are searched for the vehicle in order of priority; Adopting the priority queue strategy, iteratively expand the total cost function f(n j ) minimum node; each time the node with the minimum total cost is selected as the current node, all adjacent nodes of the current node are traversed, and the cumulative cost g(n i ) and the heuristic function value h(n i ), update the total cost function value f(n j ) and record the predecessor node; if the adjacent node has not been visited or a better path is found, it is added to the priority queue; the above process is repeated until the end node is visited; when the end node is visited, the predecessor node sequence is traced back from the end node to generate an alternative path.

7. The city-level route guidance method based on dynamic clustering in a vehicle-road-cloud collaboration scenario according to claim 6 is characterized in that: A buffer threshold trigger mechanism is used to dynamically adjust road weights and penalty values ​​in batches on the path planning subgraph; this includes: For a certain vehicle cluster z, let the current total number of vehicles be When the vehicle completes the path planning, the vehicle is temporarily stored in the buffer zone Q z Medium; passing relative threshold Control the update frequency. When the number of vehicles in the buffer zone meets When an update operation is triggered, the update operation steps are as follows: First, for the buffer Q z For each vehicle v in the new driving path, the next segment of the new driving path planned by vehicle v is recorded as e next (v) and construct the set of road sections to be updated For each road segment e in the road segment set to be updated ij , update the road segment e using the following formula ij Number of vehicles in the corresponding routing subgraph: in, is the indicator function, if e ij =e next (v) then the value of the indicator function is 1, otherwise it is 0; Represents the road segment e in the path planning subgraph of vehicle cluster z ij After the vehicle count is updated, the estimated travel delay of the road section is calculated according to formula (1)-formula (3), and the road weight is updated; Then, the planned vehicle set of each vehicle cluster is updated using the following formula: in, Represents the set of planned vehicles.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario described in any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the city-level path guidance method based on dynamic clustering in the vehicle-road-cloud collaboration scenario described in any one of claims 1-7.

10. A city-level route guidance system based on dynamic clustering in a vehicle-road-cloud collaborative scenario, comprising: The data collection module is configured to: build a vehicle-road-cloud collaborative architecture, collect traffic data, and upload the collected traffic data to the traffic management center in the cloud; The congestion detection module is configured as follows: the traffic management center updates the road weights and initializes the penalty matrix, while performing congestion detection; The vehicle screening module to be planned is configured to: when a congested road section is detected, set an adaptive distance threshold to screen out vehicles that may be affected by congestion to form a set of vehicles to be planned; The clustering module is configured to: spatially cluster the set of vehicles to be planned to obtain multiple vehicle clusters, and construct an independent path planning subgraph for each vehicle cluster; The priority allocation module is configured to: allocate priorities to vehicles within each vehicle cluster; The alternative path planning module is configured to: process the path planning tasks of each vehicle cluster in parallel, and plan alternative paths for the vehicles in the vehicle cluster according to the calculated priority order; The batch dynamic adjustment module is configured to: use a buffer threshold trigger mechanism to dynamically adjust the road weights and penalty values ​​in batches on the path planning subgraph; The loop execution module is configured to: loop execute the alternative path planning module and the batch dynamic adjustment module until the path planning of all the vehicles to be planned is completed; The path distribution module is configured as follows: the traffic management center distributes the planned alternative path to the corresponding vehicle through the roadside unit, and the vehicle drives according to the new route to achieve path guidance.

Images