A method and device for identifying critical paths of urban road congestion

Abstract

The present invention relates to a method and device for identifying critical paths for urban road congestion, belonging to the technical field of road traffic management. The method comprises: obtaining a target model and a graph neural network interpretation algorithm; the target model is a fitting model for urban road congestion constructed based on a graph neural network; obtaining an urban road congestion example graph through prediction using the target model; identifying important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges; and determining the critical congestion paths in the urban road congestion example graph based on the weights of the connecting edges in the target subgraph. The present invention can reasonably identify the critical paths that cause urban road congestion, providing a basis for preventing and alleviating urban road congestion.

Description

Technical Field

[0001] The present invention relates to the technical field of road traffic management, and in particular to a method and device for identifying a critical path of urban road congestion. Background Art

[0002] Preventing or alleviating traffic congestion is a major challenge and pain point in current urban road traffic management. Therefore, identifying the key congestion paths and implementing effective control measures in a timely manner will be of great significance in improving the travel experience of urban residents.

[0003] Because urban road traffic systems are complex, nonlinear, and fluctuating systems, traffic congestion is highly random. This makes identifying critical paths that cause traffic congestion challenging, and there is currently no recognized, reliable method for achieving this goal. Graph Neural Networks (GNNs), a machine learning method, are currently among the most accurate in the field of traffic forecasting. However, GNNs suffer from a common flaw of previous machine learning methods: their prediction process is like a black box, with no explanation for the predictions. Consequently, GNNs alone cannot help traffic managers identify critical paths that cause traffic congestion. Summary of the Invention

[0004] In view of this, it is necessary to provide a method and device for identifying critical paths of urban road congestion to solve the problem in the existing technology that traffic status prediction through graph neural networks cannot help traffic managers identify the critical paths causing traffic congestion.

[0005] In order to solve the above problems, in a first aspect, the present invention provides a method for identifying a critical path of urban road congestion, comprising:

[0006] Obtaining a target model and a graph neural network interpretation algorithm; the target model is an urban road congestion fitting model built based on the graph neural network;

[0007] Obtaining an example map of urban road congestion through prediction of the target model;

[0008] Using the graph neural network interpretation algorithm, the urban road congestion example graph is used to identify important subgraphs to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges;

[0009] Determine a critical congestion path in the urban road congestion example graph according to the weights of the connecting edges in the target subgraph.

[0010] Optionally, the graph neural network interpretation algorithm is a GNNExplainer algorithm.

[0011] Optionally, the target subgraph includes congested nodes; and determining the congested critical path in the urban road congestion example graph according to the weights of the connecting edges in the target subgraph includes:

[0012] Taking the connection edge in the target subgraph whose weight is greater than a preset weight threshold as the target connection edge;

[0013] A target critical path causing congestion of the congested node is determined according to the target connection edge and the congested node.

[0014] Optionally, determining, based on the target connection edge and the congested node, a target critical path causing congestion of the congested node includes:

[0015] Determining at least one sub-target critical path that causes congestion at the congested node according to the target connection edge and the congested node;

[0016] Determining the average weight of the connecting edges in the critical paths of each of the sub-goals;

[0017] The sub-target critical path with the largest average weight is used as the target critical path.

[0018] Optionally, obtaining an urban road congestion example map through prediction of the target model includes:

[0019] Coupling the target model with the graph neural network interpretation algorithm to obtain a coupled model;

[0020] Obtaining an example map of urban road congestion through prediction of the target model in the coupling model;

[0021] The method of identifying important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph includes:

[0022] The graph neural network interpretation algorithm in the coupling model is used to identify important subgraphs of the urban road congestion example graph to obtain a target subgraph.

[0023] Optionally, obtaining the target model includes:

[0024] Obtain node identification, node connection information, node traffic parameter information and node traffic congestion status of urban road network nodes;

[0025] A target model is constructed according to the node identification, node connection information, node traffic parameter information and node traffic congestion status.

[0026] Optionally, obtaining the node traffic congestion status includes:

[0027] Obtaining the mapping relationship between traffic parameter information and traffic congestion status;

[0028] The node traffic congestion state is determined according to the node traffic parameter information and the mapping relationship.

[0029] Optionally, the target model is trained in the following manner:

[0030] Construct the node adjacency matrix based on the node identifier and the node connection information A { w ij}, where w ij It is expressed as follows:

[0031] ;

[0032] The target model is used to make predictions based on the adjacency matrix and the node traffic parameter information; wherein the prediction process of the target model is expressed by the following formula:

[0033] ,in, Represents all nodes separately tn , t-n+1 ,…, t The characteristic vector of the node traffic parameter information detected at each moment, A represents the adjacency matrix, represents the parameters of the target model, represents the prediction process, Represents all nodes t Predicted node traffic congestion status at time +1;

[0034] According to the preset loss function of the target model, the predicted node traffic congestion status and the node traffic congestion status, the parameters of the target model are adjusted until the parameters meet the preset requirements.

[0035] Optionally, the target model is a three-layer network model, wherein the first and second layers are graph attention networks in the graph neural network, and the third layer is a fully connected linear output layer; the activation function of the graph attention network is the Mish function, and the loss function of the target model is a multi-classification cross entropy loss function.

[0036] In a second aspect, the present invention further provides a device for identifying a critical path of urban road congestion, comprising:

[0037] A model and model interpretation algorithm acquisition module is used to obtain a target model and a graph neural network interpretation algorithm; the target model is an urban road congestion fitting model built based on a graph neural network;

[0038] A congestion instance prediction module, configured to obtain a city road congestion instance map through prediction of the target model;

[0039] An important subgraph identification module is used to identify important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges;

[0040] The congestion critical path determination module is used to determine the congestion critical path in the urban road congestion instance graph according to the weights of the connecting edges in the target subgraph.

[0041] The beneficial effects of the present invention are:

[0042] A target model is obtained and an example diagram of urban road congestion is obtained through prediction using the target model; wherein the target model is an urban road congestion fitting model constructed based on a graph neural network; the present invention continues to use the graph neural network model to predict traffic congestion conditions to ensure the accuracy of the prediction.

[0043] Furthermore, a graph neural network interpretation algorithm is obtained and used to identify important subgraphs of the urban road congestion example graph, obtaining a target subgraph. The target subgraph includes nodes, edges connecting nodes, and edge weights. Based on the edge weights in the target subgraph, the critical congestion path in the urban road congestion example graph is determined. The graph neural network interpretation algorithm can compare the various subgraphs in the input urban road congestion example graph and identify the target subgraph that plays a key role in the target model prediction process. The larger the edge weight in the target subgraph, the more likely it is to be a critical congestion path. Therefore, by determining the critical congestion path based on the edge weights in the target subgraph, it is possible to reasonably identify the critical paths that cause urban road congestion, thereby providing a basis for preventing and alleviating urban road congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of an embodiment of a method for identifying critical paths of urban road congestion provided by the present invention;

[0045] Figure 2 A flowchart of another embodiment of the critical path identification method provided by the present invention;

[0046] Figure 3 A schematic diagram of a neural network model provided by the present invention;

[0047] Figure 4 A schematic diagram of a target subgraph provided by the present invention;

[0048] Figure 5This is a structural schematic diagram of an embodiment of the urban road congestion critical path identification device provided by the present invention. DETAILED DESCRIPTION

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0050] In the description of the embodiments of the present invention, unless otherwise specified, "plurality" means two or more. "And / or" describes the association relationship between associated objects, indicating that three relationships can exist. For example, "A and / or B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0051] The terms "first," "second," and so on, used in the embodiments of the present invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, technical features designated as "first" or "second" may explicitly or implicitly include at least one such feature.

[0052] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute a separate or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0053] Reference Figure 1 , which shows a flow chart of an embodiment of a method for identifying a critical path of urban road congestion provided by the present invention, the method comprising:

[0054] Step S101, obtain a target model and a graph neural network interpretation algorithm; the target model is an urban road congestion fitting model constructed based on a graph neural network.

[0055] Step S102: Obtain an example map of urban road congestion through target model prediction.

[0056] Step S103 , using a graph neural network interpretation algorithm to identify important subgraphs of the urban road congestion example graph, and obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges.

[0057] Step S104 : determining a critical congestion path in the urban road congestion instance graph according to the weights of the connecting edges in the target subgraph.

[0058] The target model is a neural network model used to predict urban road congestion. The graph neural network interpretation algorithm is used to analyze and explain the prediction results of the graph neural network.

[0059] The target model can first predict an urban road congestion instance graph, which is used to characterize the urban road congestion situation. The urban road congestion instance graph may include nodes, node identifiers, connecting edges between nodes, predicted node congestion status, etc. Nodes refer to but are not limited to any object that can be abstracted as a node, such as road endpoints, road sections, or road intersections. Node identifiers are used to identify nodes and to distinguish different nodes in the urban road network. The connecting edges between nodes represent the connection relationship between different nodes. If there is an upstream and downstream relationship between the traffic flows between nodes, there are edges between the nodes, otherwise there are no edges. The predicted node congestion status can be represented by the degree of node congestion, for example, no congestion, slight congestion, moderate congestion, etc.

[0060] A graph neural network explanation algorithm can be the GNNExplainer algorithm, which was proposed in 2019 to provide explainability of the decision-making process of the GNN model. GNNExplainer helps users understand how the model makes predictions by identifying the subgraph structures and node features that are most important to the GNN prediction results.

[0061] By using a graph neural network interpretation algorithm to identify important subgraphs in the urban road congestion example graph, we can obtain the target subgraph that plays a key role in the target model's prediction process. The target subgraph consists of nodes, the edges connecting them, and the weights of these edges. The larger the weight of an edge, the more likely it is to be a critical path causing congestion. Therefore, based on the target subgraph and the weights of the edges in the target subgraph, we can determine the critical congestion path in the urban road congestion example graph.

[0062] In the present invention, a target model is obtained and an example map of urban road congestion is obtained through prediction of the target model; wherein, the target model is an urban road congestion fitting model constructed based on a graph neural network; by using the graph neural network model to predict traffic congestion conditions, the accuracy of the prediction is guaranteed.

[0063] Furthermore, a graph neural network interpretation algorithm is obtained and used to identify important subgraphs of the urban road congestion example graph, obtaining a target subgraph. The target subgraph includes nodes, edges connecting nodes, and edge weights. Based on the edge weights in the target subgraph, the critical congestion path in the urban road congestion example graph is determined. The graph neural network interpretation algorithm can compare the various subgraphs in the input urban road congestion example graph and identify the target subgraph that plays a key role in the target model prediction process. The larger the edge weight in the target subgraph, the more likely it is to be a critical congestion path. Therefore, by determining the critical congestion path based on the edge weights in the target subgraph, it is possible to reasonably identify the critical paths that cause urban road congestion, thereby providing a basis for preventing and alleviating urban road congestion.

[0064] In one embodiment, the target subgraph includes a congested node; step S104 may specifically include: using a connection edge in the target subgraph whose weight is greater than a preset weight threshold as a target connection edge; and determining a target critical path that causes congestion of the congested node based on the target connection edge and the congested node.

[0065] In this embodiment, the connection edges in the target subgraph can be screened first to select target connection edges with weights greater than a preset weight threshold, and then the target critical path that causes congestion in the congested node can be determined based on the target connection edges and the congested node.

[0066] In one embodiment, the step of determining the target critical path that causes congestion of the congested node based on the target connection edge and the congested node may specifically include: determining at least one sub-target critical path that causes congestion of the congested node based on the target connection edge and the congested node; determining the average weight of the connection edge in each sub-target critical path; and taking the sub-target critical path with the largest average weight as the target critical path.

[0067] In this embodiment, after the target connection edges are screened out, one or more sub-target critical paths that cause congestion of the congested nodes can be determined based on the target connection edges and the congested nodes, and then the average weight of the connection edges of each sub-target critical path is calculated respectively, and the sub-target critical path with the largest average weight is used as the target critical path.

[0068] In one embodiment, step S102 includes: coupling the target model with the graph neural network interpretation algorithm to obtain a coupled model; obtaining an urban road congestion instance graph through the target model prediction in the coupled model; step S103 includes: identifying important subgraphs of the urban road congestion instance graph through the graph neural network interpretation algorithm in the coupled model to obtain a target subgraph.

[0069] Coupling the target model with the graph neural network interpretation algorithm means combining the two so that they can work together to solve a problem.

[0070] In one embodiment, step S101 includes: obtaining node identification, node connection information, node traffic parameter information and node traffic congestion status of urban road network nodes; and constructing a target model based on the node identification, node connection information, node traffic parameter information and node traffic congestion status.

[0071] In this embodiment, the node traffic parameter information may include average traffic flow speed, average traffic flow volume, average road occupancy rate, etc. The node traffic congestion status may be information indicating the node congestion situation, for example, the traffic congestion level. The node traffic congestion status may be determined based on the node traffic parameter information.

[0072] In one embodiment, the step of obtaining the node traffic congestion status may specifically include: obtaining a mapping relationship between traffic parameter information and the traffic congestion status; and determining the node traffic congestion status according to the node traffic parameter information and the mapping relationship.

[0073] In this embodiment, the node traffic congestion state corresponding to the node traffic parameter information may be determined according to the mapping relationship.

[0074] In one embodiment, the target model is trained by constructing an adjacency matrix of nodes based on the node identifiers and the node connection information. A { w ij}, where w ij It is expressed as follows:

[0075] ;

[0076] The target model is used to make predictions based on the adjacency matrix and the node traffic parameter information; wherein the prediction process of the target model is expressed by the following formula:

[0077] ,in, Represents all nodes separately tn , t-n+ 1 ,…, t The characteristic vector of the node traffic parameter information detected at each moment, A represents the adjacency matrix, represents the parameters of the target model, represents the prediction process, Represents all nodes t Predicted node traffic congestion status at time +1;

[0078] According to the preset loss function of the target model, the predicted node traffic congestion status and the node traffic congestion status, the parameters of the target model are adjusted until the parameters meet the preset requirements.

[0079] In one embodiment, the target model is a three-layer network model, wherein the first and second layers are graph attention networks in a graph neural network, and the third layer is a fully connected linear output layer; the activation function of the graph attention network is Mish(), and the loss function of the target model is multi-classification cross entropy CrossEntropyLoss().

[0080] Reference Figure 2 , shows a flow chart of another embodiment of the critical path identification method provided by the present invention.

[0081] ① Obtain urban road network node topology information.

[0082] Node topology information includes node identification and node connectivity information. Node connectivity information refers to the connection relationships between different nodes in the network, that is, whether there are edges connecting the nodes. In practice, for example, the node identification and node connectivity information of 228 nodes in a city road network can be obtained.

[0083] ② Obtain traffic information of urban road network nodes.

[0084] The node traffic information includes: node traffic parameter information and node traffic congestion status. The node traffic congestion status can be determined based on the node traffic parameter information.

[0085] Continuing with the above example, the node traffic parameter information can be the average vehicle speed at the node. The average vehicle speed detected by sensors at the 228 nodes can be obtained at fixed intervals of five minutes, with the collection period being May 1, 2012, to June 30, 2012. Based on the average vehicle speed at the node, the traffic congestion status at the node can be calculated. The calculation can be based on the national congestion quantification standard, as shown in Table 1 below:

[0086] Table 1, Congestion Quantification Standards

[0087]

[0088] ③Construct the urban road network node adjacency matrix.

[0089] Constructing the urban road network node adjacency matrix based on the urban road network node topology information A { w ij}.

[0090] Continuing the above example, we can get a 228 The adjacency matrix with 228 rows and columns is simplified as follows:

[0091] .

[0092] ④Construct an urban road congestion fitting model based on GNNs.

[0093] First, we build a three-layer initial network model, where the first and second layers are the Graph Attention Networks (GAT) in the Graph Neural Networks (GNNs), and the third layer is a fully connected linear output layer. The specific structure of the initial network model is as follows: Figure 3 shown.

[0094] Then, the activation function of the GAT layer is defined as the Mish function, and the loss function of the initial network model is defined as the multi-classification cross entropy loss function CrossEntropyLoss().

[0095] The initial network model was then fed with node traffic parameter information, node congestion status, and an adjacency matrix. The model then predicted the congestion status for the next time step using the average speed over 12 time steps. The model was then trained based on the initial network model's loss function, the predicted node congestion status, and the actual node congestion status until its parameters were optimized, resulting in the final target model. The target model achieved a fitting accuracy of 92%, effectively fitting congestion instances.

[0096] ⑤Couple the GNNExplainer algorithm with the target model.

[0097] The explainer coupling explanation model can be constructed through the Python explainer organization module. The model to be explained in the coupling model is the target model, and the explanation algorithm is the GNNExplainer algorithm.

[0098] ⑥Identify important subgraphs through coupling models.

[0099] The urban road congestion instance graph is obtained by predicting the target model in the coupling model. The important subgraphs of the urban road congestion instance graph are identified by the GNNExplainer algorithm to obtain the target subgraph.

[0100] The GNNExplainer algorithm can compare the importance of different subgraphs in the congestion instance to the model fitting and find the most important subgraph, namely the target subgraph. The importance is quantified by the mutual information MI:

[0101]

[0102] in, Y is the prediction result of the target model, that is, , Gs is the subgraph constructed by the GNNExplainer algorithm, Xs is the feature vector on the subgraph node, H(Y) is the probability that the congestion fitting model is accurate.

[0103] Reference Figure 4 , shows a target subgraph schematic diagram provided by the present invention. Taking congested node 2 as an example, the target subgraph that has the greatest impact on the congestion of node 2 identified by the GNNExplainer algorithm is as follows Figure 4 As shown in Figure 2, when the target subgraph is identified by the GNNExplainer algorithm, the weights of the connecting edges between the nodes of the target subgraph can be obtained. The more important the connecting edge is, the closer the weight is to 1, and vice versa.

[0104] ⑦ Determine the critical path where congestion occurs.

[0105] Continue with Figure 4 As shown in the example, the weight threshold is first preset to 0.6, then Figure 4 The target connection edges greater than the preset weight threshold include: the connection edge from node 15 to node 4, the connection edge from node 4 to node 2, the connection edge from node 4 to node 12, and the connection edge from node 12 to node 2.

[0106] Then, based on the target connection edge and congested node 2, the sub-target critical path that causes congestion of congested node 2 is determined. The sub-target critical path includes: node 15 to node 4 to node 2, and node 15 to node 4 to node 12 to node 2.

[0107] Finally, calculate the average weight of the edges in each sub-target critical path. The average weight of the edges connecting node 15 to node 4 to node 2 is (0.77 + 0.83) / 2 = 0.8; the average weight of the edges connecting node 15 to node 4 to node 12 to node 2 is (0.77 + 0.65 + 0.67) / 3 = 0.7. Therefore, we can determine that node 15 to node 4 to node 2 is the most important target critical path causing congestion at congested node 2. Traffic managers should prioritize this path when preventing or alleviating traffic congestion at node 2.

[0108] The present invention provides a method for identifying critical paths caused by urban road congestion. Based on a graph neural network model with high fitting accuracy, the method combines the GNNExplainer algorithm to identify the critical paths caused by congestion and provides a quantitative basis. It provides quantitative analysis and decision support for traffic managers, effectively helping to alleviate traffic congestion, improve travel efficiency, reduce resource waste and environmental pollution, and thus significantly improve the quality of life of urban residents and the sustainable development of cities.

[0109] Reference Figure 5 , shows a schematic structural diagram of an embodiment of a device for identifying a critical path of urban road congestion provided by the present invention, wherein the device 200 includes:

[0110] Model and model interpretation algorithm acquisition module 201, used to acquire a target model and a graph neural network interpretation algorithm; the target model is an urban road congestion fitting model built based on a graph neural network;

[0111] A congestion instance prediction module 202 is configured to obtain a city road congestion instance map through prediction of the target model;

[0112] The important subgraph identification module 203 is used to identify important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges;

[0113] The congestion critical path determining module 204 is configured to determine the congestion critical path in the urban road congestion example graph according to the weights of the connecting edges in the target subgraph.

[0114] It should be noted that the implementation principles or implementation processes of the above modules can refer to the above-mentioned embodiment of building facade structure extraction, and will not be described in detail here.

[0115] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0116] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for identifying critical paths of urban road congestion, characterized in that: include: Obtain the target model and graph neural network interpretation algorithm; The target model is an urban road congestion fitting model built based on graph neural network; Obtaining an example map of urban road congestion through prediction of the target model; Using the graph neural network interpretation algorithm, the urban road congestion example graph is used to identify important subgraphs to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges; Determine a critical congestion path in the urban road congestion example graph according to the weights of the connecting edges in the target subgraph.

2. The urban road congestion critical path identification method according to claim 1, characterized in that: The graph neural network interpretation algorithm is the GNNExplainer algorithm.

3. The urban road congestion critical path identification method according to claim 1, characterized in that: The target subgraph includes congested nodes; and determining the congested critical path in the urban road congestion example graph according to the weights of the connecting edges in the target subgraph includes: Taking the connection edge in the target subgraph whose weight is greater than a preset weight threshold as the target connection edge; A target critical path causing congestion of the congested node is determined according to the target connection edge and the congested node.

4. The urban road congestion critical path identification method according to claim 3 is characterized in that: The determining, based on the target connection edge and the congested node, a target critical path causing congestion of the congested node includes: Determining at least one sub-target critical path that causes congestion at the congested node according to the target connection edge and the congested node; Determining the average weight of the connecting edges in the critical paths of each of the sub-goals; The sub-target critical path with the largest average weight is used as the target critical path.

5. The urban road congestion critical path identification method according to claim 1 is characterized in that: The example map of urban road congestion obtained by predicting the target model includes: Coupling the target model with the graph neural network interpretation algorithm to obtain a coupled model; Obtaining an example map of urban road congestion through prediction of the target model in the coupling model; The method of identifying important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph includes: The graph neural network interpretation algorithm in the coupling model is used to identify important subgraphs of the urban road congestion example graph to obtain a target subgraph.

6. The urban road congestion critical path identification method according to claim 1, characterized in that: The acquiring of the target model includes: Obtain node identification, node connection information, node traffic parameter information and node traffic congestion status of urban road network nodes; A target model is constructed according to the node identification, node connection information, node traffic parameter information and node traffic congestion status.

7. The urban road congestion critical path identification method according to claim 6, characterized in that: The obtaining of the node traffic congestion status includes: Obtaining the mapping relationship between traffic parameter information and traffic congestion status; The node traffic congestion state is determined according to the node traffic parameter information and the mapping relationship.

8. The urban road congestion critical path identification method according to claim 6, characterized in that: The target model is trained as follows: Construct the node adjacency matrix based on the node identifier and the node connection information A { w ij }, where w ij It is expressed as follows: ； The target model is used to make predictions based on the adjacency matrix and the node traffic parameter information; wherein the prediction process of the target model is expressed by the following formula: ,in, Represents all nodes separately tn , t-n+1 ,…, t The characteristic vector of the node traffic parameter information detected at each moment, A represents the adjacency matrix, represents the parameters of the target model, represents the prediction process, Represents all nodes t Predicted node traffic congestion status at time +1; According to the preset loss function of the target model, the predicted node traffic congestion status and the node traffic congestion status, the parameters of the target model are adjusted until the parameters meet the preset requirements.

9. The urban road congestion critical path identification method according to claim 8, characterized in that: The target model is a three-layer network model, wherein the first and second layers are graph attention networks in the graph neural network, and the third layer is a fully connected linear output layer; the activation function of the graph attention network is the Mish function, and the loss function of the target model is a multi-classification cross entropy loss function.

10. A device for identifying critical paths of urban road congestion, characterized in that: include: Model and model interpretation algorithm acquisition module, used to obtain the target model and graph neural network interpretation algorithm; The target model is an urban road congestion fitting model built based on graph neural network; A congestion instance prediction module, configured to obtain a city road congestion instance map through prediction of the target model; An important subgraph identification module is used to identify important subgraphs of the urban road congestion example graph using the graph neural network interpretation algorithm to obtain a target subgraph; the target subgraph includes nodes, connecting edges between nodes, and weights of the connecting edges; The congestion critical path determination module is used to determine the congestion critical path in the urban road congestion instance graph according to the weights of the connecting edges in the target subgraph.

Images