Vehicle system path adjustment method and device based on graph neural network

Abstract

The invention relates to the technical field of automatic driving, and discloses a vehicle system path adjustment method and device based on a graph neural network, and the method comprises the steps: dividing a target traffic network based on geographic position data and traffic flow characteristic data, and obtaining a plurality of independent regions; constructing a region graph neural network model, training the region graph neural network model to obtain road feature embedded representation, determining node feature information and edge feature information of boundaries of adjacent independent regions, and sharing the node feature information and the edge feature information to the region graph neural network model of the adjacent independent regions. Shared information is obtained; and updating the area graph neural network model by using real-time traffic dynamic data, determining an optimal path in the independent area, and determining a global optimal vehicle system path based on the shared information and the optimal path in the independent area. According to the invention, real-time path adjustment of the vehicle system in a complex traffic environment is realized.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to a vehicle system path adjustment method and device based on a graph neural network. Background Art

[0002] Vehicle route planning refers to the rational arrangement of vehicle routes to achieve a certain optimal goal under given customer demand points, vehicle starting positions and related constraints; reasonable vehicle route planning can effectively reduce transportation costs and improve user experience.

[0003] When faced with multi-objective optimization needs, relevant vehicle path planning requires manual setting of target priorities, or only single-objective optimization, which makes it difficult to take into account multi-dimensional needs, resulting in the vehicle system being unable to make real-time path adjustments in complex traffic environments. Summary of the invention

[0004] In view of this, the present invention provides a vehicle system path adjustment method and device based on graph neural network to solve the problem that the vehicle system cannot perform real-time path adjustment in a complex traffic environment.

[0005] In a first aspect, the present invention provides a vehicle system path adjustment method based on a graph neural network, the method comprising:

[0006] Acquire geographic location data and traffic flow characteristic data of a target traffic network, and divide the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas;

[0007] Constructing a regional graph neural network model in each of the independent regions, and training the regional graph neural network model to obtain a road feature embedding representation;

[0008] Determine the node feature information and edge feature information of the boundary of adjacent independent areas based on the road feature embedding representation, share the node feature information and edge feature information of the boundary of adjacent independent areas with the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between the adjacent independent areas;

[0009] Acquire real-time traffic dynamic data in the independent area, update the regional graph neural network model using the real-time traffic dynamic data, and determine the optimal path in the independent area using the updated regional graph neural network model;

[0010] A global optimal vehicle system path is determined based on the boundary node sharing information between the adjacent independent areas and the optimal path within the independent area.

[0011] The vehicle system path adjustment method based on graph neural network provided in this embodiment divides the target traffic network by geographic location data and traffic flow characteristic data to obtain multiple independent areas; constructs a regional graph neural network model in each independent area, trains the regional graph neural network model to obtain a road feature embedding representation; determines the node feature information and edge feature information of the boundaries of adjacent independent areas based on the road feature embedding representation, shares the node feature information and edge feature information of the boundaries of adjacent independent areas into the regional graph neural network model of the adjacent independent area, and obtains the boundary node sharing information between adjacent independent areas; updates the regional graph neural network model by using real-time traffic dynamic data, and determines the optimal path within the independent area by using the updated regional graph neural network model; determines the global optimal vehicle system path based on the boundary node sharing information between adjacent independent areas and the optimal path within the independent area; through the rapid update of the regional graph neural network model and the cross-region boundary node sharing information, the vehicle system can perform real-time path adjustment in a complex traffic environment, realizes the planning of the global optimal vehicle system path, and improves the dynamic adaptability of the vehicle system path planning.

[0012] In an optional implementation manner, the target traffic network is divided based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas, including:

[0013] Acquire geographic location data and traffic flow characteristic data of the target traffic network, and determine the road type, traffic node density and traffic flow distribution of each road based on the geographic location data and the traffic flow characteristic data;

[0014] The target traffic network is clustered and divided based on the road types of adjacent roads, the traffic node density and the traffic flow distribution to obtain the multiple independent areas.

[0015] The vehicle system path adjustment method based on graph neural network provided in this embodiment divides the vehicle system into independent areas, and the vehicle system can reasonably plan the path according to different road types, different traffic node densities and different traffic flow distributions, laying the foundation for determining the global optimal vehicle system path.

[0016] In an optional implementation, constructing a regional graph neural network model in each of the independent regions, training the regional graph neural network model, and obtaining a road feature embedding representation includes:

[0017] Constructing a regional road network graph model and a regional graph neural network model in each of the independent regions;

[0018] Inputting the node feature vector and the edge feature vector of the road network graph model in the region into the regional graph neural network model, and calculating the node embedding feature and the edge embedding feature by using a message passing mechanism;

[0019] Acquire historical traffic data in independent areas, use the node embedding features and the edge embedding features as initial features of the regional graph neural network model, use the historical traffic data in the independent areas to perform offline training on the regional graph neural network model, and obtain the road feature embedding representation of each independent area.

[0020] The vehicle system path adjustment method based on graph neural network provided in this embodiment effectively describes the static and dynamic characteristics of each intersection and road segment in the independent area by inputting the node feature vector and edge feature vector of the road network graph model in the area into the regional graph neural network model, and calculating the node embedding features and edge embedding features by using the message passing mechanism; obtaining the historical traffic data in the independent area, taking the node embedding features and edge embedding features as the initial features of the regional graph neural network model, and using the historical traffic data in the independent area to perform offline training on the regional graph neural network model, obtaining the road feature embedding representation of each independent area, and laying the foundation for the calculation of the path feature value.

[0021] In an optional implementation, the updating of the regional graph neural network model using the real-time traffic dynamics data, and determining the optimal path in the independent area using the updated regional graph neural network model, includes:

[0022] Mapping the real-time traffic dynamic data into the regional graph neural network model, and updating the node feature vector and the edge feature vector in the regional graph neural network model;

[0023] Perform forward propagation on the updated regional graph neural network model to obtain a set of feasible paths;

[0024] Calculate the feature value of each feasible path in the feasible path set based on the node embedding features and edge embedding features corresponding to the updated regional graph neural network model;

[0025] Acquire vehicle task requirements and user preferences, and determine path selection weights based on the vehicle task requirements and the user preferences;

[0026] A path comprehensive score is calculated based on the characteristic value of each feasible path and the path selection weight, and the feasible path corresponding to the minimum path comprehensive score is selected as the optimal path in the independent area.

[0027] The vehicle system path adjustment method based on graph neural network provided in this embodiment can accurately capture emergencies in the road network and their impact on traffic flow by dynamically updating the node embedding features and edge embedding features corresponding to the regional graph neural network model; by dynamically setting the path selection weight in combination with user preferences or task requirements, multi-objective optimization of path planning is achieved, and in practical applications, the best balance between safety, economy and efficiency can be found, thereby improving the comprehensive performance of the vehicle system; the comprehensive score of the path is calculated based on the characteristic value and path selection weight of each feasible path, and the comprehensive score is calculated by normalizing the characteristic value and path selection weight of each feasible path, a flexible path selection mechanism is proposed, and the feasible path corresponding to the minimum path comprehensive score is selected as the optimal path in the independent area, laying the foundation for determining the global optimal vehicle system path.

[0028] In an optional implementation, the node feature information and edge feature information of the boundary of adjacent independent regions are determined based on the road feature embedding representation, and the node feature information and edge feature information of the boundary of adjacent independent regions are shared with the regional graph neural network model of the adjacent independent regions to obtain the boundary node sharing information between the adjacent independent regions, including:

[0029] Obtain the region boundary node set between adjacent independent regions and the edge set connecting the region boundary nodes;

[0030] Based on the road feature embedding representation, assigning a shared node feature vector to each node in the region boundary node set, and assigning a shared edge feature vector to each edge in the edge set;

[0031] The region boundary node set, the edge set, the shared node feature vector and the shared edge feature vector are shared as the boundary node sharing information between the adjacent independent regions into the region graph neural network model of the adjacent independent regions, and the node embedding feature and the edge embedding feature are updated;

[0032] Dynamically calculate cross-region path weights based on updated node embedding features and edge embedding features;

[0033] The updated node embedding features, the edge embedding features, and the cross-region path weights are used as boundary node sharing information between the adjacent independent regions.

[0034] The vehicle system path adjustment method based on graph neural network provided in this embodiment constructs a regional boundary node sharing mechanism by sharing the regional boundary node set, edge set, shared node feature vector and shared edge feature vector as boundary node sharing information between adjacent independent regions into the regional graph neural network model of the adjacent independent regions, and realizes collaborative optimization within and across regions. The vehicle can adjust the path in real time through the rapid update of the regional graph neural network model and the sharing of cross-regional boundary nodes, which significantly improves the dynamic adaptability of path planning and lays the foundation for determining the global optimal vehicle system path.

[0035] In an optional implementation, the determining the global optimal vehicle system path based on the boundary node sharing information between the adjacent independent areas and the optimal path within the independent area includes:

[0036] Calculate a cross-region path feature value based on the updated node embedding feature and the edge embedding feature;

[0037] A global optimization is performed based on the cross-region path characteristic value, the cross-region path weight and the optimal path in the independent area to obtain the global optimal vehicle system path.

[0038] The vehicle system path adjustment method based on graph neural network provided in this embodiment calculates the cross-regional path feature value through the updated node embedding feature and edge embedding feature, performs global optimization based on the cross-regional path feature value, the cross-regional path weight and the optimal path in the independent area, integrates the characteristics of the optimal path in the independent area and the cross-regional path, and adjusts the cross-regional path based on the cross-regional path weight to obtain the global optimal vehicle system path, so that the global optimal vehicle system path has optimality and coherence.

[0039] In a second aspect, the present invention provides a vehicle system path adjustment device based on a graph neural network, the device comprising:

[0040] A partitioning module is used to obtain geographic location data and traffic flow characteristic data of a target traffic network, and to partition the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain multiple independent areas;

[0041] A training module is used to construct a regional graph neural network model in each independent area, train the regional graph neural network model, and obtain a road feature embedding representation;

[0042] A module is obtained, which is used to determine the node feature information and edge feature information of the boundary of the adjacent independent areas based on the road feature embedding representation, share the node feature information and edge feature information of the boundary of the adjacent independent areas into the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between the adjacent independent areas;

[0043] An updating module, used to obtain real-time traffic dynamic data in the independent area, update the regional graph neural network model using the real-time traffic dynamic data, and determine the optimal path in the independent area using the updated regional graph neural network model;

[0044] A determination module is used to determine a global optimal vehicle system path based on boundary node sharing information between adjacent independent areas and an optimal path within the independent area.

[0045] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the vehicle system path adjustment method based on graph neural network of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the vehicle system path adjustment method based on a graph neural network according to the first aspect or any corresponding embodiment thereof.

[0047] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions, which are used to enable a computer to execute the vehicle system path adjustment method based on a graph neural network according to the above-mentioned first aspect or any corresponding embodiment. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 is a flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention;

[0050] Figure 2 is a flow chart of another vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention;

[0051] Figure 3 It is a schematic diagram of the process of constructing a regional road network graph model and inputting features of a regional graph neural network according to an embodiment of the present invention;

[0052] Figure 4is a flow chart of another vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention;

[0053] Figure 5 is a flow chart of another vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention;

[0054] Figure 6 is a schematic flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention;

[0055] Figure 7 is a structural block diagram of a vehicle system path adjustment device based on a graph neural network according to an embodiment of the present invention;

[0056] Figure 8 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0058] In the relevant vehicle system path planning methods, the algorithms used include: Dijkstra algorithm (an algorithm for solving the shortest path problem in a road graph), A* algorithm (a heuristic search algorithm for graph traversal and path search) or their improved variants. The above algorithms determine the shortest path or the optimal path by statically calculating the weights of nodes and edges in the road network. Although the above methods can quickly obtain results in a static environment, in actual applications, traffic conditions are often changing rapidly. Vehicles will face sudden road congestion, traffic accidents, weather changes and other dynamic interference factors during driving, making it difficult to respond to complex traffic environments in real time. In addition, when responding to multi-objective optimization needs such as comprehensive consideration of driving time, fuel efficiency and safety, it is often necessary to manually set target priorities or perform single-objective optimization, which makes it difficult to take into account multi-dimensional needs.

[0059] In order to cope with the dynamic traffic environment, the function of real-time traffic information update is gradually introduced, such as collecting dynamic data on traffic flow and road conditions through traffic monitoring systems and vehicle-mounted terminal equipment, and then integrating them into the path planning system. However, the above method is usually implemented by regularly refreshing the path planning results. It is impossible to dynamically adjust the path during vehicle driving, and the data fusion capability is insufficient in complex scenarios, resulting in lag and inefficiency in path planning. In addition, the vehicle system's modeling of road networks is usually based on regular weight distribution and topological structure, lacking a deep understanding of the complex connections between roads and dynamic learning capabilities.

[0060] At the same time, machine learning algorithms are used to predict traffic patterns, such as predicting road congestion or vehicle travel time based on neural networks. However, such methods are usually limited to the prediction of a single indicator, and it is difficult to combine the topological characteristics of the road network and dynamic traffic data for global optimization. In addition, the above methods rely on the quality and coverage of historical data, and the accuracy of predictions is often difficult to guarantee when facing emergencies or abnormal scenarios, further limiting the application effect of the above path planning methods in real-time path adjustment.

[0061] To solve the above technical problems, an embodiment of the present invention provides a vehicle system path adjustment method based on a graph neural network. By dividing the overall road network into multiple independent areas, a regional graph neural network model is constructed in the independent area, and a regional boundary node sharing mechanism is designed to achieve collaborative optimization within and across regions. The rapid update of the regional graph neural network model and the information sharing of cross-regional boundary nodes enable the vehicle to adjust the path in real time, which significantly improves the dynamic adaptability of path planning. By combining multi-dimensional path features, a flexible path selection mechanism is proposed using normalized eigenvalues ​​and weight calculation comprehensive scores, which achieves multi-objective optimization of path planning. In practical applications, the best balance can be found between safety, economy and efficiency, thereby improving the comprehensive performance of the vehicle system. Based on the adaptive learning mechanism of the regional graph neural network model, by dynamically updating the feature embedding of nodes and edges, it is possible to accurately capture emergencies in the road network and their impact on traffic flow. The real-time traffic dynamic data is quickly modeled and learned through the regional graph neural network model, which effectively improves the vehicle system path adjustment method's ability to handle abnormal scenarios, and significantly improves the system's robustness and practical applicability.

[0062] According to an embodiment of the present invention, an embodiment of a vehicle system path adjustment method based on a graph neural network is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0063] In this embodiment, a vehicle system path adjustment method based on a graph neural network is provided, which can be used in the above-mentioned electronic device. Figure 1 is a flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0064] Step S101, obtaining geographic location data and traffic flow characteristic data of a target traffic network, and dividing the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas.

[0065] Specifically, the geographic location data L includes the latitude and longitude coordinate information of the road; the traffic flow characteristic data T includes the traffic volume, road capacity and traffic congestion degree information; the road type D of each road in the overall road network is calculated based on the geographic location data L and the traffic flow characteristic data T. i , traffic node density N i and traffic flow distribution V i , where i represents the road number.

[0066] Step S102: construct a regional graph neural network model in each independent area, train the regional graph neural network model, and obtain a road feature embedding representation.

[0067] Step S103, determining the node feature information and edge feature information of the boundaries of adjacent independent areas based on the road feature embedding representation, sharing the node feature information and edge feature information of the boundaries of adjacent independent areas into the regional graph neural network model of the adjacent independent areas, and obtaining the boundary node sharing information between the adjacent independent areas.

[0068] Specifically, a regional boundary node sharing mechanism is established between every two adjacent independent regions, and the node and edge feature information of the boundaries of adjacent independent regions are shared to the intra-regional graph neural networks of both parties through the inter-regional communication module. Then, according to the shared information of the regional boundary nodes, the path priority and weight distribution between different regions are dynamically adjusted.

[0069] Step S104, obtaining real-time traffic dynamic data in the independent area, using the real-time traffic dynamic data to update the regional graph neural network model, and using the updated regional graph neural network model to determine the optimal path in the independent area.

[0070] Specifically, real-time traffic dynamics data such as traffic volume, road occupancy rate, emergencies and traffic signal status in the area where the vehicle starting point is located are collected, and the real-time traffic dynamics data are mapped to the regional graph neural network model of the corresponding geographical area. The node features and edge features in the regional graph neural network model are dynamically updated, and then the updated regional graph neural network model is used to determine the optimal path in the independent area.

[0071] Step S105 , determining a global optimal vehicle system path based on the boundary node sharing information between adjacent independent areas and the optimal path within the independent area.

[0072] Specifically, when the vehicle driving path involves multiple independent areas, the characteristic value of the cross-region path is calculated based on the shared information of boundary nodes between adjacent areas, and the optimal choice of the path in each independent area is combined with the characteristic value of the cross-region path to generate the global optimal vehicle system path.

[0073] The vehicle system path adjustment method based on graph neural network provided in this embodiment divides the target traffic network by geographic location data and traffic flow characteristic data to obtain multiple independent areas; constructs a regional graph neural network model in each independent area, trains the regional graph neural network model to obtain a road feature embedding representation; determines the node feature information and edge feature information of the boundaries of adjacent independent areas based on the road feature embedding representation, shares the node feature information and edge feature information of the boundaries of adjacent independent areas into the regional graph neural network model of the adjacent independent area, and obtains the boundary node sharing information between adjacent independent areas; updates the regional graph neural network model by using real-time traffic dynamic data, and determines the optimal path within the independent area by using the updated regional graph neural network model; determines the global optimal vehicle system path based on the boundary node sharing information between adjacent independent areas and the optimal path within the independent area; through the rapid update of the regional graph neural network model and the cross-region boundary node sharing information, the vehicle system can perform real-time path adjustment in a complex traffic environment, realizes the planning of the global optimal vehicle system path, and improves the dynamic adaptability of the vehicle system path planning.

[0074] In this embodiment, a vehicle system path adjustment method based on a graph neural network is provided, which can be used in the above-mentioned electronic device. Figure 2 is a flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention. Figure 2 As shown, the process includes the following steps:

[0075] Step S201, obtaining geographic location data and traffic flow characteristic data of a target traffic network, and dividing the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas.

[0076] Specifically, the above step S201 includes:

[0077] Step S2011, obtaining the geographic location data and traffic flow characteristic data of the target traffic network, and determining the road type, traffic node density and traffic flow distribution of each road based on the geographic location data and traffic flow characteristic data.

[0078] Specifically, the road type D is determined according to the design attributes of the road. a , road type D b Including expressways, main roads, secondary roads and branch roads; calculate the number of traffic nodes per unit area in the independent area where road a is located, and get the traffic node density N a ; Calculate the number of vehicles passing through road a per unit time and obtain the traffic flow distribution V a .

[0079] Step S2012, clustering the target traffic network based on the road types, traffic node density and traffic flow distribution of adjacent roads to obtain multiple independent areas.

[0080] Specifically, the partition threshold parameters are set to measure the similarity of road types, traffic node density and traffic flow distribution.

[0081] Furthermore, based on the road type D a , traffic node density N a and traffic flow distribution V a The overall road network in the target traffic network is divided into regions, that is, the roads that meet the following conditions are divided into the same independent region using a clustering algorithm:

[0082] |D a -D b |≤ε D (1)

[0083] |N a -N b |≤ε N (2)

[0084] |V a -V b |≤ε V (3)

[0085] Among them, D b 、N b and V b are the road type, traffic node density and traffic flow distribution of road b adjacent to road a, D a 、N a and V aare the road type, traffic node density and traffic flow distribution of road i, ε D , ε N and ε V They are the division thresholds for road type, traffic node density and traffic flow distribution respectively.

[0086] Furthermore, all roads in the target traffic network are traversed, and the clustering algorithm is repeatedly used to divide all roads that meet the conditions, until the target traffic network is divided into multiple independent areas that meet the conditions, thereby completing the regional division of the target traffic network.

[0087] Step S202: construct a regional graph neural network model in each independent area, train the regional graph neural network model, and obtain a road feature embedding representation.

[0088] Specifically, Figure 3 As shown, the above step S202 includes:

[0089] Step S2021, constructing a regional road network graph model and a regional graph neural network model in each independent area.

[0090] Specifically, the road intersections or key locations in each independent area are taken as nodes, which contain information such as geographic coordinates, number of connected roads, and traffic signal status. The road segments representing the connecting nodes in each independent area and the characteristics describing the roads are taken as edges, including length, width, maximum speed limit, real-time traffic flow, etc. Then, based on the nodes and edges, a road network graph model G is constructed in each independent area. k , k represents the independent area number, and the road network graph model G in the area k The expression is as follows:

[0091] G k =(N k ,E k ) (4)

[0092] Where k represents the independent area number, N k is the set of nodes in the independent region k, node N k,i Indicates a road intersection or key location point, i is the node number, E k is the set of edges in the independent region k, edge E k,i Represents the connection node N k,i and node N k,j The road segment of j is the adjacent node number.

[0093] Furthermore, for each node N in the road network graph model in the region k,i Assign node feature vector The expression of the node feature vector is as follows:

[0094]

[0095] Among them, x k,i and k,i Respectively represent the geographic horizontal coordinate and the geographic vertical coordinate of the node, d k,i Indicates the number of connecting roads, s k,i The traffic signal status includes the cycle and phase information of the traffic light.

[0096] Furthermore, for each edge E in the edge set k,i Assign edge eigenvectors The expression of the edge eigenvector is as follows:

[0097]

[0098] Among them, l k,ij represents the road length, w k,ij Indicates the width of the road. Indicates the maximum speed limit. represents the current traffic flow speed, d k,ij Indicates the road type code.

[0099] Furthermore, a regional graph neural network model GNN is constructed for each independent region k .

[0100] Step S2022, input the node feature vector and edge feature vector of the road network graph model in the region into the regional graph neural network model, and use the message passing mechanism to calculate the node embedding features and edge embedding features.

[0101] Specifically, a regional road network graph model is constructed in each independent region, and a regional graph neural network model is trained based on the regional road network graph model. By inputting node features and edge features, the road topology structure and dynamic traffic characteristics in the region are learned, and the embedded features of each road and intersection, namely, node embedding features and edge embedding features, are output.

[0102] Furthermore, based on the road network graph model G in the region k , using the traffic data in the area to the regional graph neural network model GNN k For training, the node feature vector and edge eigenvector As input, the message passing mechanism is used to calculate the embedding features of nodes and edges. The updated node embedding features of the tth layer for:

[0103]

[0104] in, For node N k,i The set of neighbor nodes, W (t) is the learnable weight matrix of the tth layer in the regional graph neural network model, β 1 is the weight coefficient of the node feature, σ is the activation function, Embedding features for nodes in layer t, is the attention coefficient, and its calculation formula is as follows:

[0105]

[0106] Among them, a (t) is the attention weight vector of the tth layer, ∥ is the connection operation of the vector, and LeakyReLU is the activation function.

[0107] Furthermore, assuming that the current edge e ij Connect node v i and v j , the calculation formula of its edge embedding feature can be expressed as:

[0108]

[0109] in, For edge e ij The embedded features at the tth layer, and For node v i and v j The embedded features at the t-1th layer, For edge e ij The initial characteristics, W e is the learnable weight matrix, b e is the bias term and φ is the activation function.

[0110] Furthermore, by training the regional graph neural network model, the embedded features of each node are obtained. and the embedding features of each edge

[0111] Step S2023, obtain historical traffic data in the independent area, use node embedding features and edge embedding features as initial features of the regional graph neural network model, use the historical traffic data in the independent area to perform offline training on the regional graph neural network model, and obtain the road feature embedding representation of each independent area.

[0112] Specifically, a regional graph neural network model is constructed for each independent region, and the node features, edge features and road topology structure within the independent region are input. The regional graph neural network model is trained offline through historical traffic data to learn the topological characteristics and dynamic feature change laws of the traffic network in the region, and generate an embedded representation of the road features of each independent region.

[0113] Furthermore, the constructed regional road network graph model G k As a regional graph neural network model GNN k Input, embedding features of each node and the embedding features of each edge As the initial features of nodes and edges respectively; collect historical traffic data in independent area k, which includes traffic flow data in the past time period Road occupancy rate Traffic signal status and incident records information.

[0114] Furthermore, using historical traffic data H k Regional Graph Neural Network Model GNN k Perform offline training, and minimize the loss function L during the training process k The topological characteristics and dynamic characteristics of the traffic network in the learning area are changed. The loss function L k Defined as:

[0115]

[0116] Among them, y k,i For node N k,i The actual traffic status label, is the traffic status predicted by the model, |N k | is the total number of nodes in independent region k.

[0117] Among them, the data sources of the actual traffic status labels are: historical traffic data, including traffic volume, road occupancy, traffic signal status and average vehicle speed in the past time period; traffic management system data is real-time or historical records from the urban traffic management system, including traffic flow statistics, traffic congestion index and emergencies; external data sources include real-time data from satellite positioning equipment, intelligent traffic equipment and vehicle-mounted GPS, which records data such as the speed and location of the vehicle at a specific time.

[0118] Furthermore, during the training process, the back propagation algorithm is used to update the regional graph neural network model GNN k The parameter θ k, by adjusting the weights of the regional graph neural network model, the relevant content of the road traffic status in the prediction area can be realized. The expression of the back propagation algorithm is as follows:

[0119]

[0120] Among them, η 1 is the learning rate, is the loss function for the parameter θ k gradient.

[0121] Furthermore, after training is completed, an embedded representation of road features is generated for each independent region, including embedded features of nodes and embedded features of edges. The embedded features capture the topological structure and dynamic traffic characteristics of the road network in the region.

[0122] Furthermore, in order to avoid overfitting and limit the training time, during the training process of the regional graph neural network model, after 100 or 200 iterations, the training of the regional graph neural network model will be stopped regardless of whether the loss value or verification performance is further improved.

[0123] Step S203, based on the road feature embedding representation, determine the node feature information and edge feature information of the borders of the adjacent independent regions, share the node feature information and edge feature information of the borders of the adjacent independent regions into the regional graph neural network model of the adjacent independent regions, and obtain the shared information of the border nodes between the adjacent independent regions. For details, please refer to Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0124] Step S204, obtain real-time traffic dynamic data in the independent area, use the real-time traffic dynamic data to update the regional graph neural network model, and use the updated regional graph neural network model to determine the optimal path in the independent area. For details, please refer to Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0125] Step S205: Determine the global optimal vehicle system path based on the shared information of the boundary nodes between the adjacent independent areas and the optimal path within the independent area. Figure 1 Step S105 of the illustrated embodiment will not be described in detail here.

[0126] The vehicle system path adjustment method based on graph neural network provided in this embodiment determines the road type, traffic node density and traffic flow distribution of each road through geographic location data and traffic flow characteristic data to divide independent areas. The vehicle system can reasonably plan the path according to different road types, different traffic node densities and different traffic flow distributions, laying a foundation for determining the global optimal vehicle system path; by inputting the node feature vectors and edge feature vectors of the road network graph model in the region into the regional graph neural network model, the node embedding features and edge embedding features are calculated using the message passing mechanism, which effectively describes the static and dynamic characteristics of each intersection and road segment in the independent area; the historical traffic data in the independent area is obtained, the node embedding features and edge embedding features are used as the initial features of the regional graph neural network model, and the regional graph neural network model is trained offline using the historical traffic data in the independent area to obtain the road feature embedding representation of each independent area, laying a foundation for the calculation of the path feature value.

[0127] In this embodiment, a vehicle system path adjustment method based on a graph neural network is provided, which can be used in the above-mentioned electronic device. Figure 4 is a flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention. Figure 4 As shown, the process includes the following steps:

[0128] Step S401, obtain the geographic location data and traffic flow characteristic data of the target traffic network, and divide the target traffic network based on the geographic location data and traffic flow characteristic data to obtain multiple independent areas. Figure 2 Step S201 of the illustrated embodiment will not be described in detail here.

[0129] Step S402: construct a regional graph neural network model in each independent region, train the regional graph neural network model, and obtain a road feature embedding representation. Figure 2 Step S202 of the illustrated embodiment will not be described in detail here.

[0130] Step S403, based on the road feature embedding representation, determine the node feature information and edge feature information of the boundaries of adjacent independent areas, share the node feature information and edge feature information of the boundaries of adjacent independent areas into the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between the adjacent independent areas.

[0131] Specifically, the above step S403 includes:

[0132] Step S4031, obtaining a set of region boundary nodes between adjacent independent regions and a set of edges connecting the region boundary nodes.

[0133] Specifically, between every two adjacent independent regions k and l, a region boundary node set B is determined: kl and the edge set E of the region boundary nodes kl .

[0134] Step S4032: Based on the road feature embedding representation, a shared node feature vector is assigned to each node in the region boundary node set, and a shared edge feature vector is assigned to each edge in the edge set.

[0135] Specifically, it is the regional boundary node set B kl Each node N in kl,i Assign shared nodes feature vectors The shared node feature vector includes the geographic coordinates (x kl,i ,y kl,i ), the number of connecting roads d kl,i and traffic signal status s kl,i Information; is the edge set E connecting the boundary nodes of the region kl Each edge E in kl,ij Assign shared edge eigenvectors The shared edge feature vector includes the road length l kl,ij 、Road width w kl,ij , Maximum speed limit Current traffic speed and road type code d kl,ij information.

[0136] Step S4033, share the region boundary node set, edge set, shared node feature vector and shared edge feature vector as boundary node sharing information between adjacent independent regions into the regional graph neural network model of adjacent independent regions, and update the node embedding features and edge embedding features.

[0137] Specifically, through the inter-regional communication module, the regional boundary node set B kl and its shared node feature vector and the edge set E kl and its shared edge eigenvector Regional graph neural network model GNN shared to adjacent independent regions k and GNN l In this way, the regional graph neural network models of both independent regions contain information about shared boundary nodes and edges.

[0138] Furthermore, in the adjacent independent regions, the regional graph neural network model GNN k and GNN l In the above method, the node embedding features and edge embedding features are updated based on the shared node feature vector and the shared edge feature vector.

[0139] Step S4035, dynamically calculating the cross-region path weight based on the updated node embedding features and edge embedding features.

[0140] Specifically, according to the updated node embedding features Edge Embedding Features Dynamically adjust the path priority and weight distribution between different independent areas, and define the weight function w of the cross-area path kl,ij The calculation formula is as follows:

[0141]

[0142] Among them, ψ represents the weight calculation function, which is used to comprehensively consider the embedding features of boundary nodes and edges, and w kl,ij Represents slave node N kl,i To Node N kl,j The cross-region path weight.

[0143] Step S4036, using the updated node embedding features and edge embedding features, as well as the cross-region path weights as boundary node sharing information between adjacent independent regions.

[0144] Specifically, in the path planning process, using w kl,ij The cross-regional path is calculated so that the vehicle can dynamically adjust the path selection based on the shared information of the boundary nodes between adjacent independent areas when traveling across regions, ensuring the optimality and consistency of the global path planning.

[0145] Step S404, obtain real-time traffic dynamic data in the independent area, use the real-time traffic dynamic data to update the regional graph neural network model, and use the updated regional graph neural network model to determine the optimal path in the independent area. For details, please refer to Figure 2 Step S204 of the illustrated embodiment will not be described in detail here.

[0146] Step S405: Determine the global optimal vehicle system path based on the shared information of the boundary nodes between the adjacent independent areas and the optimal path within the independent area. Figure 2 Step S205 of the illustrated embodiment will not be described in detail here.

[0147] The vehicle system path adjustment method based on graph neural network provided in this embodiment constructs a regional boundary node sharing mechanism by sharing the regional boundary node set, edge set, shared node feature vector and shared edge feature vector as boundary node sharing information between adjacent independent regions into the regional graph neural network model of the adjacent independent regions, and realizes collaborative optimization within and across regions. The vehicle can adjust the path in real time through the rapid update of the regional graph neural network model and the sharing of cross-regional boundary nodes, which significantly improves the dynamic adaptability of path planning and lays the foundation for determining the global optimal vehicle system path.

[0148] In this embodiment, a vehicle system path adjustment method based on a graph neural network is provided, which can be used in the above-mentioned electronic device. Figure 5 is a flow chart of a vehicle system path adjustment method based on a graph neural network according to an embodiment of the present invention. Figure 5 As shown, the process includes the following steps:

[0149] Step S501, obtain the geographic location data and traffic flow characteristic data of the target traffic network, and divide the target traffic network based on the geographic location data and traffic flow characteristic data to obtain multiple independent areas. Figure 4 Step S401 of the illustrated embodiment will not be described in detail here.

[0150] Step S502: construct a regional graph neural network model in each independent region, train the regional graph neural network model, and obtain a road feature embedding representation. Figure 4 Step S402 of the illustrated embodiment will not be described in detail here.

[0151] Step S503, based on the road feature embedding representation, determine the node feature information and edge feature information of the borders of the adjacent independent regions, share the node feature information and edge feature information of the borders of the adjacent independent regions into the regional graph neural network model of the adjacent independent regions, and obtain the shared information of the border nodes between the adjacent independent regions. For details, please refer to Figure 4 Step S403 of the illustrated embodiment will not be described in detail here.

[0152] Step S504, obtain real-time traffic dynamic data in the independent area, use the real-time traffic dynamic data to update the regional graph neural network model, and use the updated regional graph neural network model to determine the optimal path in the independent area. For details, please refer to Figure 4 Step S404 of the illustrated embodiment will not be described in detail here.

[0153] Specifically, the above step S504 includes:

[0154] Step S5041, mapping the real-time traffic dynamic data to the regional graph neural network model, and updating the node feature vectors and edge feature vectors in the regional graph neural network model.

[0155] Specifically, in the independent area k where the vehicle starts, obtain real-time traffic dynamic data Including current traffic flow, road occupancy rate and traffic signal status, real-time traffic dynamic data Mapping to the Regional Graph Neural Network Model GNN k In , update the node feature vector and edge feature vector, and the calculation formula is as follows:

[0156]

[0157] in, is the updated node feature vector, is the updated edge feature vector, ⊕ is the feature connection operation; Including node N k,i Real-time traffic signal status Including edge E k,ij Current traffic volume and road occupancy rate

[0158] Step S5042, forward propagation is performed on the updated regional graph neural network model to obtain a set of feasible paths.

[0159] Specifically, using the updated regional graph neural network model GNN k , the forward propagation of the regional graph neural network model is used to calculate the vehicle from the starting node N start To destination node N end The set P of all feasible paths k , every feasible path p∈P k Can be represented as a node sequence

[0160] Step S5043, calculate the feature value of each feasible path in the feasible path set based on the node embedding features and edge embedding features corresponding to the updated regional graph neural network model.

[0161] Specifically, for each feasible path, its characteristic values ​​are calculated, including the travel time T p , driving distance L p and fuel consumption F p , driving time T p The calculation formula is as follows:

[0162]

[0163] Among them, lk,ij For edge E k,ij The road length, δ k,ij Due to emergencies The additional delay caused by For node N k,i The average waiting time at is the effective driving speed, and its calculation formula is as follows:

[0164]

[0165] in, is the maximum speed limit of the road, λ is the blocking coefficient, which reflects the impact of road congestion on vehicle speed. is the maximum road occupancy rate;

[0166] Node N k,i Average waiting time at The calculation formula is as follows:

[0167]

[0168] Among them, C k,i is the traffic signal cycle, g k,i is the green light time ratio, ρ k,i is the node saturation, and its calculation formula is as follows:

[0169]

[0170] in, For node N k,i Current traffic volume, is the traffic capacity of the node.

[0171] Furthermore, the travel distance L p The calculation formula is as follows:

[0172]

[0173] Furthermore, the fuel consumption F p The calculation formula is as follows:

[0174]

[0175] Among them, α 1 is the basic fuel consumption coefficient per unit distance, β 2 is the slope influence coefficient, a k,ij is the road slope, γ 2 is the waiting time fuel consumption coefficient, The additional fuel consumption caused by the emergency is calculated as follows:

[0176]

[0177] Among them, β is the impact coefficient of emergency fuel consumption.

[0178] Step S5044, obtaining vehicle task requirements and user preferences, and determining path selection weights based on the vehicle task requirements and user preferences.

[0179] Specifically, the path selection weight is set according to the vehicle mission requirements and user preferences. The calculation formula of the path selection weight is as follows:

[0180] w T +w L +w F =1 (22)

[0181] w T ,w L ,w F ≥0 (23)

[0182] Among them, w T is the travel time weight, w L is the driving distance weight, w F is the fuel consumption weight; the path selection weight is adjusted according to user preferences to achieve multi-dimensional path optimization.

[0183] Step S5045, calculating the path comprehensive score based on the characteristic value and path selection weight of each feasible path, and selecting the feasible path corresponding to the minimum path comprehensive score as the optimal path in the independent area.

[0184] Specifically, each path p corresponds to a comprehensive score S p The calculation formula is as follows:

[0185]

[0186] Among them, T min 、T ,ax is the minimum and maximum travel time among all feasible paths, L min , L max is the minimum and maximum value of the driving distance, F min 、F max is the minimum and maximum value of fuel consumption, θ 1 is the risk preference weight, R p is the path risk assessment value; path risk assessment value R p The calculation formula is as follows:

[0187]

[0188] Among them, κ is the risk weight coefficient, is the risk assessment value of the edge, calculated based on historical accident data and road conditions; according to the comprehensive score S p , select the feasible path corresponding to the minimum path comprehensive score as the optimal path in the current independent area. The expression of the optimal path is as follows:

[0189]

[0190] Among them, p * It is the feasible path corresponding to the minimum path comprehensive score.

[0191] Step S505 , determining a global optimal vehicle system path based on the boundary node sharing information between adjacent independent areas and the optimal path within the independent area.

[0192] Specifically, the above step S505 includes:

[0193] Step S5051, calculating the cross-region path feature value based on the updated node embedding features and edge embedding features.

[0194] Specifically, the cross-regional path is decomposed into multiple independent intra-regional paths and cross-regional boundary paths, and the characteristic values ​​such as time, distance, and fuel consumption are calculated for each cross-regional path segment based on the updated node embedding features and edge embedding features.

[0195] Step S5052, performing global optimization based on the cross-region path characteristic value, the cross-region path weight and the optimal path in the independent area to obtain the global optimal vehicle system path.

[0196] Specifically, the weight of the cross-regional path and the characteristic value of the optimal path within the independent area are included in the comprehensive scoring formula. The cross-regional path weight directly affects the final score of the path, and paths with high cross-regional path weights and better cross-regional path characteristic values ​​are given priority. Based on the comprehensive scoring results, the global optimal path of the cross-regional and independent area paths is selected.

[0197] The vehicle system path adjustment method based on graph neural network provided in this embodiment can accurately capture emergencies in the road network and their impact on traffic flow by dynamically updating the node embedding features and edge embedding features corresponding to the regional graph neural network model; by dynamically setting the path selection weight in combination with user preferences or task requirements, multi-objective optimization of path planning is achieved, and in practical applications, the best balance point can be found between safety, economy and efficiency, thereby improving the comprehensive performance of the vehicle system; the path comprehensive score is calculated based on the characteristic value and path selection weight of each feasible path, and the comprehensive score is calculated by normalizing the characteristic value and path selection weight of each feasible path, a flexible path selection mechanism is proposed, and the feasible path corresponding to the minimum path comprehensive score is selected as the optimal path in the independent area, laying the foundation for determining the global optimal vehicle system path; secondly, global optimization is performed based on the cross-regional path characteristic value, the cross-regional path weight and the optimal path in the independent area, the characteristics of the optimal path in the independent area and the cross-regional path are integrated, and the cross-regional path is adjusted based on the cross-regional path weight, so as to obtain the global optimal vehicle system path, so that the global optimal vehicle system path has optimality and coherence.

[0198] The following is a specific example to illustrate the specific steps of a vehicle system path adjustment method based on graph neural network.

[0199] Embodiment 1:

[0200] like Figure 6 As shown, the specific steps of the vehicle system path adjustment method based on graph neural network include:

[0201] 1) According to the geographical location and traffic flow characteristics of the target traffic network, the overall road network is divided into multiple independent areas. The independent areas are determined according to the road type, traffic node density and traffic flow distribution.

[0202] 2) Construct a regional road network graph model in each independent region, and train the regional graph neural network model based on the regional road network graph model. By inputting node features and edge features, the road topology structure and dynamic traffic characteristics in the region are learned, and the embedded feature representation of each road and intersection is output.

[0203] 3) Build a regional graph neural network model for each independent region, input node features, edge features and road topology structure in the region, train the regional graph neural network model offline through historical traffic data, learn the topological characteristics and dynamic feature change rules of the traffic network in the independent region, and generate the road feature embedding representation of each independent region.

[0204] 4) A regional boundary node sharing mechanism is established between every two adjacent independent regions, and the node feature information and edge feature information of the boundaries of adjacent independent regions are shared to the intra-regional graph neural network of adjacent independent regions through the inter-region communication module. According to the shared information of the independent region boundary nodes, the path priority and weight distribution between different independent regions are dynamically adjusted.

[0205] 5) Collect traffic dynamic data including vehicle flow, road occupancy, emergencies, and traffic signal status in real time, map the collected real-time traffic dynamic data to the regional graph neural network model of the corresponding independent area, and dynamically update the node features and edge features in the regional graph neural network model.

[0206] 6) Based on the regional graph neural network model in the area where the vehicle's starting point is located, combined with real-time traffic dynamics data, the characteristic value of each feasible path in the set of feasible paths in the independent area is calculated. The characteristic values ​​include driving time, driving distance and fuel consumption. The path selection weight is set according to the vehicle's mission requirements and user preferences. The comprehensive score is calculated using the characteristic value and path selection weight of each feasible path to select the optimal path in the current independent area.

[0207] 7) When the vehicle driving path involves multiple independent areas, the cross-region path characteristic value is calculated based on the shared information of boundary nodes between adjacent independent areas, and the optimal path in each independent area is combined with the cross-region path characteristic value to generate the global optimal vehicle system path.

[0208] The beneficial effects of the above embodiment 1 include:

[0209] 1) By dividing the overall road network into multiple independent areas to construct a regional graph neural network model, and designing a regional boundary node sharing mechanism, collaborative optimization within and across regions is achieved. Related methods rely on a single global optimization path planning, which is difficult to respond quickly to dynamically changing traffic environments. This embodiment enables vehicles to adjust their paths in real time through rapid updates of models within the region and information sharing of cross-regional boundary nodes, significantly improving the dynamic adaptability of path planning. In actual complex traffic scenarios, this embodiment reduces the response time of path adjustment by about 30% compared with related methods, greatly improving the efficiency of the system.

[0210] 2) By combining multi-dimensional path features, a flexible path selection mechanism is proposed by using normalized path feature values ​​and path selection weights to calculate comprehensive scores. Related path planning technologies are usually limited to the optimization of a single objective, such as the shortest time or the shortest distance. This embodiment achieves multi-objective optimization of path planning by dynamically setting path selection weights in combination with user preferences or task requirements. In practical applications, it can find the best balance between safety, economy and efficiency, thereby improving the overall performance of the vehicle system. Tests show that its path comprehensive score is improved by about 15% compared with related methods.

[0211] 3) The adaptive learning mechanism based on graph neural network can accurately capture emergencies in the road network and their impact on traffic flow by dynamically updating the feature embedding of nodes and edges. The relevant path planning methods based on rules or historical data often show lag and low accuracy when dealing with emergencies. The present embodiment can quickly model and learn real-time traffic dynamic data through graph neural network, effectively improving the ability to handle abnormal scenarios. In the simulation test, the path planning success rate of this embodiment in dealing with traffic emergencies was improved by 25%, which significantly improved the robustness and practical applicability of the system.

[0212] In this embodiment, a vehicle system path adjustment device based on a graph neural network is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" may be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0213] This embodiment provides a vehicle system path adjustment device based on a graph neural network, such as Figure 7 As shown, including:

[0214] The division module 701 is used to obtain the geographic location data and traffic flow characteristic data of the target traffic network, and divide the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain multiple independent areas.

[0215] The training module 702 is used to construct a regional graph neural network model in each independent area, train the regional graph neural network model, and obtain a road feature embedding representation.

[0216] Obtain module 703, which is used to determine the node feature information and edge feature information of the boundaries of adjacent independent areas based on the road feature embedding representation, share the node feature information and edge feature information of the boundaries of adjacent independent areas into the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between the adjacent independent areas.

[0217] The updating module 704 is used to obtain real-time traffic dynamic data in the independent area, update the regional graph neural network model using the real-time traffic dynamic data, and determine the optimal path in the independent area using the updated regional graph neural network model.

[0218] The determination module 705 is used to determine the global optimal vehicle system path based on the boundary node sharing information between adjacent independent areas and the optimal path within the independent area.

[0219] The further functional description of each of the above modules is the same as that of the above corresponding embodiments and will not be repeated here.

[0220] The vehicle system path adjustment device based on graph neural network in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0221] The embodiment of the present invention also provides a computer device having the above Figure 7 The vehicle system path adjustment device based on graph neural network is shown.

[0222] See also Figure 8 , Figure 8 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 8 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.

[0223] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0224] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0225] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0226] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0227] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0228] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0229] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0230] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A vehicle system path adjustment method based on graph neural network, characterized in that: The method comprises: Acquire geographic location data and traffic flow characteristic data of a target traffic network, and divide the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas; Constructing a regional graph neural network model in each of the independent regions, and training the regional graph neural network model to obtain a road feature embedding representation; Determine the node feature information and edge feature information of the boundary of adjacent independent areas based on the road feature embedding representation, share the node feature information and edge feature information of the boundary of adjacent independent areas with the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between the adjacent independent areas; Acquire real-time traffic dynamic data in the independent area, update the regional graph neural network model using the real-time traffic dynamic data, and determine the optimal path in the independent area using the updated regional graph neural network model; A global optimal vehicle system path is determined based on the boundary node sharing information between the adjacent independent areas and the optimal path within the independent area.

2. The method according to claim 1, characterized in that The target traffic network is divided based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas, including: Acquire geographic location data and traffic flow characteristic data of the target traffic network, and determine the road type, traffic node density and traffic flow distribution of each road based on the geographic location data and the traffic flow characteristic data; The target traffic network is clustered and divided based on the road types of adjacent roads, the traffic node density and the traffic flow distribution to obtain the multiple independent areas.

3. The method according to claim 1, characterized in that The constructing of a regional graph neural network model in each of the independent regions and training the regional graph neural network model to obtain a road feature embedding representation includes: Constructing a regional road network graph model and a regional graph neural network model in each of the independent regions; Inputting the node feature vector and the edge feature vector of the road network graph model in the region into the regional graph neural network model, and calculating the node embedding feature and the edge embedding feature by using a message passing mechanism; The historical traffic data in the independent area is obtained, and the node embedding features and the edge embedding features are used as the initial features of the regional graph neural network model. The regional graph neural network model is trained offline using the historical traffic data in the independent area to obtain the road feature embedding representation of each independent area.

4. The method according to claim 3, characterized in that The method of updating the regional graph neural network model by using the real-time traffic dynamic data and determining the optimal path in the independent area by using the updated regional graph neural network model includes: Mapping the real-time traffic dynamic data into the regional graph neural network model, and updating the node feature vector and the edge feature vector in the regional graph neural network model; Perform forward propagation on the updated regional graph neural network model to obtain a set of feasible paths; Calculate the feature value of each feasible path in the feasible path set based on the node embedding features and edge embedding features corresponding to the updated regional graph neural network model; Acquire vehicle task requirements and user preferences, and determine path selection weights based on the vehicle task requirements and the user preferences; A path comprehensive score is calculated based on the characteristic value of each feasible path and the path selection weight, and the feasible path corresponding to the minimum path comprehensive score is selected as the optimal path in the independent area.

5. The method according to claim 1, characterized in that The step of determining node feature information and edge feature information of the boundary of adjacent independent regions based on the road feature embedding representation, sharing the node feature information and edge feature information of the boundary of adjacent independent regions to the regional graph neural network model of the adjacent independent regions, and obtaining boundary node sharing information between adjacent independent regions includes: Obtain the region boundary node set between adjacent independent regions and the edge set connecting the region boundary nodes; Based on the road feature embedding representation, assigning a shared node feature vector to each node in the region boundary node set, and assigning a shared edge feature vector to each edge in the edge set; Sharing the region boundary node set, the edge set, the shared node feature vector and the shared edge feature vector into the regional graph neural network model of the adjacent independent region, and updating the node embedding feature and the edge embedding feature; Dynamically calculate cross-region path weights based on updated node embedding features and edge embedding features; The updated node embedding features, the edge embedding features, and the cross-region path weights are used as boundary node sharing information between the adjacent independent regions.

6. The method according to claim 5, characterized in that The determining of the global optimal vehicle system path based on the boundary node sharing information between the adjacent independent areas and the optimal path within the independent area includes: Calculate a cross-region path feature value based on the updated node embedding feature and the edge embedding feature; A global optimization is performed based on the cross-region path characteristic value, the cross-region path weight and the optimal path in the independent area to obtain the global optimal vehicle system path.

7. A vehicle system path adjustment device based on graph neural network, characterized in that: The device comprises: A partitioning module, used to obtain geographic location data and traffic flow characteristic data of a target traffic network, and to partition the target traffic network based on the geographic location data and the traffic flow characteristic data to obtain a plurality of independent areas; A training module, used for constructing a regional graph neural network model in each of the independent regions, training the regional graph neural network model, and obtaining a road feature embedding representation; An obtaining module is used to determine the node feature information and edge feature information of the boundary of adjacent independent areas based on the road feature embedding representation, share the node feature information and edge feature information of the boundary of adjacent independent areas to the regional graph neural network model of the adjacent independent areas, and obtain the boundary node sharing information between adjacent independent areas; An updating module, used to obtain real-time traffic dynamic data in an independent area, update the regional graph neural network model using the real-time traffic dynamic data, and determine the optimal path in the independent area using the updated regional graph neural network model; A determination module is used to determine a global optimal vehicle system path based on the boundary node sharing information between the adjacent independent areas and the optimal path within the independent area.

8. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the vehicle system path adjustment method based on graph neural network according to any one of claims 1 to 6 by executing the computer instructions.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the vehicle system path adjustment method based on graph neural network according to any one of claims 1 to 6.

10. A computer program product, characterized in that It includes computer instructions, which are used to enable a computer to execute the vehicle system path adjustment method based on graph neural network as described in any one of claims 1 to 6.

Images