Method and device for explaining traffic jam event of urban road network

Abstract

The invention discloses an urban road network traffic jam event interpretation method and device, and the method comprises the steps: obtaining sensor data and regional traffic data of a plurality of regional road networks, and carrying out the preprocessing of the regional traffic data, and obtaining the traffic data of a target region; generating an interpretation subgraph of a regional road network through an interpretation extractor based on the sensor data and the traffic data of the target region; based on the interpretation subgraph, generating a prediction subgraph of regional road network traffic jam through a space-time diagram neural network; and based on the interpretation sub-graph and the prediction sub-graph, generating an interpretation text of the traffic jam event in the urban road network. According to the invention, based on the interpretation sub-graph of the regional road network and the prediction sub-graph of the traffic jam of the regional road network, the interpretation text of the traffic jam event in the urban road network is generated, whole-course traceability and high-credibility decision support of a model prediction basis are realized, so that a traffic manager can understand a prediction result, and the prediction efficiency is improved. And the application value and the actual decision support capability of the model are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent transportation, and particularly relates to a method and device for explaining traffic congestion events in an urban road network. Background Art

[0002] Currently, Graph Neural Networks (GNNs) have made significant progress in multiple fields such as traffic flow prediction, intelligent traffic management, and urban planning. Among them, GNNs capture the spatio-temporal characteristics of traffic flow by modeling the spatial correlations between nodes (such as traffic sensors, intersections, etc.) in the road network and the dynamic changes of nodes. Moreover, the traffic system is essentially a complex graph structure, where each road, each sensor node, and the traffic flow relationships between them have important spatio-temporal dependencies. Based on this, GNNs are particularly suitable for traffic prediction, which can effectively improve the accuracy of traffic flow prediction, help traffic management departments monitor road conditions in real time, predict potential congestion events, and thus achieve more efficient traffic scheduling and management.

[0003] However, the interpretability of the prediction results of the above-mentioned GNNs is still significantly insufficient, making it difficult for traffic managers and non-professionals to understand how the complex neural network model obtains the prediction results, resulting in the lack of transparency of the prediction results and making it difficult to use them as a basis for decision support. Among them, the neural network models in the prior art usually focus on prediction accuracy while ignoring the interpretability of the results, which limits the practical application of the models, especially in scenarios that require human-machine collaboration. Based on this, how to effectively explain the decision-making process of the model and help users understand the reasons behind the prediction is an urgent problem to be solved. Summary of the Invention

[0004] The present invention aims to solve at least one of the technical problems in the related art to some extent.

[0005] To this end, an object of the present invention is to propose a method for explaining traffic congestion events in an urban road network. The method generates an explanation text for traffic congestion events in the urban road network based on the explanation subgraph of the regional road network and the prediction subgraph of regional road network traffic congestion, realizes the full traceability of the model prediction basis and high-trust decision support, enables traffic managers to understand the prediction results, and enhances the application value of the model and the actual decision support ability.

[0006] Another object of the present invention is to propose a device for explaining traffic congestion events in an urban road network.

[0007] To achieve the above object, an embodiment of one aspect of the present invention proposes a method for explaining traffic congestion events in an urban road network, including:

[0008] Obtain sensor data and regional traffic data of multiple regional road networks, and preprocess the regional traffic data to obtain target regional traffic data;

[0009] Based on the sensor data and the target regional traffic data, generate an explanatory subgraph of the regional road network through an interpretation extractor;

[0010] Based on the explanatory subgraph, generate a prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network;

[0011] Based on the explanatory subgraph and the prediction subgraph, generate an explanatory text for traffic congestion events in the urban road network.

[0012] The method for explaining traffic congestion events in the urban road network according to the embodiments of the present invention may further have the following additional technical features:

[0013] Further, the obtaining of sensor data and regional traffic data of multiple regional road networks, and the preprocessing of the regional traffic data to obtain target regional traffic data includes:

[0014] Construct a regional traffic information acquisition system, wherein the regional traffic information acquisition system includes loop sensors;

[0015] Based on the loop sensors in the regional traffic information acquisition system, obtain sensor data and regional traffic data of multiple regional road networks;

[0016] Preprocess the regional traffic data to obtain target regional traffic data.

[0017] Further, the generating of the explanatory subgraph of the regional road network through an interpretation extractor based on the sensor data and the target regional traffic data includes:

[0018] Based on the sensor data and the target regional traffic data, construct a directed graph;

[0019] Based on the directed graph and the first adjacency matrix in the sensor data, obtain a candidate subgraph set through a graph attention network;

[0020] Perform Monte Carlo tree search in the candidate subgraph set to generate the explanatory subgraph of the regional road network.

[0021] Further, the generating of the prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network based on the explanatory subgraph includes:

[0022] Construct a spatio-temporal graph neural network;

[0023] Determine the second adjacency matrix corresponding to the nodes in the explanatory subgraph;

[0024] Input the explanatory sub - graph and the second adjacency matrix into the spatio - temporal graph neural network to obtain the prediction sub - graph of the regional road network traffic congestion.

[0025] Further, based on the explanatory sub - graph and the prediction sub - graph, generate an explanatory text for traffic congestion events in the urban road network, including:

[0026] Determine the distance matrix corresponding to the explanatory sub - graph;

[0027] Based on the distance matrix, determine the traffic events corresponding to the explanatory sub - graph through the agglomerative hierarchical clustering algorithm;

[0028] Through the reverse coding technology of the geographic information system, obtain the road information corresponding to the sensor nodes in the explanatory sub - graph and the prediction sub - graph;

[0029] Based on the traffic events corresponding to the explanatory sub - graph, the road information corresponding to the sensor nodes, the explanatory sub - graph and the prediction sub - graph, generate an explanatory text for traffic congestion events in the urban road network through the text output template.

[0030] To achieve the above object, another embodiment of the present invention proposes an explanatory device for traffic congestion events in the urban road network, and the device includes:

[0031] A data processing module, configured to obtain sensor data and regional traffic data of multiple regional road networks, and pre - process the regional traffic data to obtain target regional traffic data;

[0032] A first generation module, configured to generate the explanatory sub - graph of the regional road network through an explanatory extractor based on the sensor data and the target regional traffic data;

[0033] A second generation module, configured to generate the prediction sub - graph of the regional road network traffic congestion through a spatio - temporal graph neural network based on the explanatory sub - graph;

[0034] A third generation module, configured to generate an explanatory text for traffic congestion events in the urban road network based on the explanatory sub - graph and the prediction sub - graph.

[0035] The explanatory method and device for traffic congestion events in the urban road network proposed by the present invention generate an explanatory text for traffic congestion events in the urban road network based on the explanatory sub - graph of the regional road network and the prediction sub - graph of the regional road network traffic congestion, realizing the full - traceability of the model prediction basis and high - credibility decision - making support, enabling traffic managers to understand the prediction results, enhancing the application value of the model and the actual decision - making support ability.

[0036] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. Description of the Drawings

[0037] The above and / or additional aspects and advantages of the present invention will become apparent and be readily understood from the following description of embodiments in conjunction with the accompanying drawings, in which:

[0038] Figure 1 is a flowchart of a method for explaining traffic congestion events in an urban road network according to an embodiment of the present invention;

[0039] Figure 2 is a simulation schematic diagram of a regional road network according to an embodiment of the present invention;

[0040] Figure 3 is a schematic diagram of an explanatory text according to an embodiment of the present invention;

[0041] Figure 4 is a schematic structural diagram of an apparatus for explaining traffic congestion events in an urban road network according to an embodiment of the present invention. Detailed Embodiments

[0042] Embodiments of the present invention will be described in detail below. Examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0043] Based on the above description, a method and apparatus for explaining traffic congestion events in an urban road network according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0044] First, a method for explaining traffic congestion events in an urban road network according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0045] Figure 1 is a flowchart of a method for explaining traffic congestion events in an urban road network according to an embodiment of the present invention.

[0046] As Figure 1 shown, the method for explaining traffic congestion events in the urban road network includes the following steps:

[0047] Step S1, obtaining sensor data and regional traffic data of multiple regional road networks, and preprocessing the regional traffic data to obtain target regional traffic data;

[0048] In one embodiment of the present invention, the method of obtaining sensor data and regional traffic data of multiple regional road networks and preprocessing the regional traffic data to obtain target regional traffic data may include the following steps:

[0049] Step S11, construct a regional traffic information collection system, where the regional traffic information collection system includes loop sensors;

[0050] Step S12, based on the loop sensors in the regional traffic information collection system, obtain sensor data and regional traffic data of multiple regional road networks;

[0051] Step S13, preprocess the regional traffic data to obtain target regional traffic data.

[0052] Among them, in one embodiment of the present invention, traffic system simulation software (such as SUMO (Simulation of Urban Mobility)) can be used to construct a regional road network including multiple (such as 40+) intersections, and real-time traffic flow can be set according to actual traffic flow characteristics, and a corresponding number (such as 327) of loop sensors can be placed, as Figure 1 shown, Figure 1 the blue highlights in are the placed loop sensors.

[0053] In addition, in one embodiment of the present invention, sensor data of multiple loop sensors can be obtained. Among them, the sensor data includes but is not limited to the ID, longitude and latitude of the sensor, the directed distance between sensors, and the first adjacency matrix. In one embodiment of the present invention, the first adjacency matrix between sensors can be obtained by a thresholded Gaussian kernel using the pairwise road network distance between sensors, as shown in the following first formula:

[0054]

[0055] where, W ij represents the edge weight between sensor v i and sensor v j , dist(v i -v j ) 2 represents the directed road network distance from sensor v i to sensor v j , σ is the standard deviation of the distance, and k is the distance threshold.

[0056] Further, in an embodiment of the present invention, regional traffic data can be periodically collected by multiple coil sensors. For example, the collection period is once every five minutes, and the collection duration is four months. And, in an embodiment of the present invention, the collection of regional traffic data may include traffic flow, lane occupancy rate, average vehicle speed, maximum vehicle speed, and average vehicle length.

[0057] And, in an embodiment of the present invention, after obtaining the regional traffic data of multiple regional road networks through the above steps, the regional traffic data can be preprocessed to obtain the target regional traffic data. Specifically, in an embodiment of the present invention, the method of preprocessing the regional traffic data to obtain the target regional traffic data may include: determining the missing rate of the regional traffic data of each sensor, discarding the regional traffic data with a missing rate exceeding the first threshold, and filling the missing values of the regional traffic data with a missing rate not exceeding the first threshold; determining whether the regional traffic data meets the actual traffic characteristics, and discarding the regional traffic data that does not meet the actual traffic characteristics; performing standardization processing on the regional traffic data through the second formula to obtain the target regional traffic data, so that its speed distribution is a smooth left-skewed normal distribution for subsequent traffic prediction.

[0058] Among them, in an embodiment of the present invention, the actual traffic characteristics may include the existence of morning and evening rush hours on weekdays, the later appearance of the morning rush hour of vehicle flow on rest days, and the larger vehicle flow at night than on weekdays.

[0059] And, in an embodiment of the present invention, the above second formula is:

[0060]

[0061] Where x is the regional traffic data, x f is a feature of x, u f is the mean of feature f, σ f is the variance, and x' f is the processed feature.

[0062] Step S2, based on the sensor data and the target regional traffic data, generate an explanatory subgraph of the regional road network through an explanation extractor;

[0063] Among them, in an embodiment of the present invention, after obtaining the sensor data and the target regional traffic data through the above steps, an explanatory subgraph of the regional road network can be generated based on the sensor data and the target regional traffic data through an explanation extractor.

[0064] Specifically, in an embodiment of the present invention, the method of generating an explanatory subgraph of the regional road network based on the sensor data and the target regional traffic data through an explanation extractor may include the following steps:

[0065] Step S21: Construct a directed graph based on sensor data and traffic data of the target area;

[0066] Step S22: Obtain a set of candidate subgraphs through a graph attention network based on the directed graph and the first adjacency matrix in the sensor data;

[0067] Step S23: Conduct Monte Carlo tree search in the set of candidate subgraphs to generate an explanatory subgraph of the regional road network.

[0068] Among them, in an embodiment of the present invention, based on the longitude and latitude information in the sensor data, combined with the topological structure characteristics presented by the road network and the actual driving directions of the roads in the traffic data of the target area, using the global road information provided by OpenStreetMap (OSM), the existing osmnx library and networkx library, a directed graph is constructed based on sumo simulation to provide a data basis for the subsequent graph attention network.

[0069] And, in an embodiment of the present invention, based on the features of each node i in the directed graph, a feature vector h of each node can be obtained i , and the feature vectors of all nodes are combined into a feature matrix H. Among them, the features of the nodes can include the longitude and latitude information of the sensors and the relevant attributes of the roads.

[0070] Furthermore, in an embodiment of the present invention, the directed graph, the first adjacency matrix, and the feature matrix can be input into the graph attention network to obtain a set of candidate subgraphs. Among them, in an embodiment of the present invention, the graph attention network includes multiple network layers, and each network layer includes attention coefficient calculation, attention coefficient normalization, and node feature update.

[0071] Specifically, in an embodiment of the present invention, the above-mentioned attention coefficient calculation may include: for each node pair (i, j), calculate the corresponding attention coefficient e ij to represent the importance of node j to node i. Among them, the third formula is:

[0072] e ij = LeakyReLU(α T [Wh i ||Wh j )

[0073] Among them, W is a learnable linear transformation matrix, α is a learnable parameter vector of the attention mechanism, || represents the concatenation operation, and LeakyReLU is an activation function.

[0074] And, in one embodiment of the present invention, normalizing the attention coefficient may include: normalizing the attention coefficient by the fourth formula to obtain a normalized attention coefficient α ij , so that the attention coefficients are comparable, where the fourth formula is:

[0075]

[0076] Among them, N i Represents the set of neighbor nodes of node i.

[0077] Further, in one embodiment of the present invention, updating the node feature may include: updating the feature h of the node i by the fifth formula according to the normalized attention coefficient: i ', where the fifth formula is:

[0078]

[0079] Among them, σ is the activation function, usually ReLU or ELU.

[0080] Furthermore, in one embodiment of the present invention, the output of each network layer in the above-mentioned multi-layer network layers is the input of the next network layer, so that the importance relationship between the nodes in the graph can be learned through multiple network layers in the graph attention network, and the nodes whose importance scores exceed the second threshold and the edges connected to them are screened out to form a candidate subgraph set. The candidate subgraph set will be used as the search space for subsequent Monte Carlo tree search, thereby narrowing the search scope.

[0081] And, in one embodiment of the present invention, the method of performing Monte Carlo Tree Search (MCTS) on the pre-screened candidate subgraph set may include: Determine all input nodes v i At time step t and event node v to be explained j The scores between the two nodes are calculated, and the input subgraph composed of the nodes with the highest scores is used as the root node of the tree. A rollback operation is performed to select a path from the root node to the leaf node, and at each branch, a node in the graph described by the tree node is removed to obtain a subtree node, and an explanation subgraph of the regional road network is generated. In one embodiment of the present invention, the leaf node complies with the sparsity threshold, that is, its graph contains a limited predefined number of nodes.

[0082] Wherein, in one embodiment of the present invention, the above-mentioned global heuristic function is:

[0083]

[0084] Where Δt is the sum of t and v jThe time distance between the last time steps; Δd is their spatial distance; Δs is the absolute difference between the velocity at time t and the average velocity in the event. i and v j in the event.

[0085] Exemplarily, in one embodiment of the present invention, it is assumed that a certain node N in a given tree i :

[0086] ① Expand a new tree node from node N according to a certain action, where the action includes removing a specific traffic node in the graph described in N i ; i ② Select a child node from the expanded nodes, where the child node that maximizes exploration and exploitation is selected;

[0087] ③ If N

[0088] is a leaf node, calculate the corresponding reward; i ④ If N

[0089] is not a leaf node, its reward is backpropagated from the leaf node until the root node, updating the information useful for exploration and exploitation in the tree search. i

[0090] Among them, in one embodiment of the present invention, the above-mentioned final result is to obtain an explanatory subgraph with the highest reward and meet the specified sparsity threshold. Based on this, the explanatory subgraph can be used for subsequent regional traffic prediction and explains the influencing factors and propagation paths of specific events in the road network.

[0091] Step S3: Based on the explanatory subgraph, generate a prediction subgraph of regional road network traffic congestion through a spatio-temporal graph neural network;

[0092] In one embodiment of the present invention, after obtaining the explanatory subgraph through the above steps, a prediction subgraph of regional road network traffic congestion can be generated based on the explanatory subgraph through a spatio-temporal graph neural network.

[0093] Specifically, in one embodiment of the present invention, the method for generating a prediction subgraph of regional road network traffic congestion based on the explanatory subgraph through a spatio-temporal graph neural network may include the following steps:

[0094] Step S31: Construct a spatio-temporal graph neural network;

[0095] Step S32: Determine the second adjacency matrix corresponding to the nodes in the explanatory subgraph;

[0096] Step S33: Input the explanatory subgraph and the second adjacency matrix into the spatio-temporal graph neural network to obtain a prediction subgraph of regional road network traffic congestion.

[0097] Among them, in one embodiment of the present invention, the above spatio-temporal graph neural network may include: an encoder, multiple spatial graph neural network (S-GNN) modules, multiple gated recurrent unit (GRU) modules, a Transformer module, and a multi-layer regression head.

[0098] In addition, in one embodiment of the present invention, the method of inputting the explanatory subgraph and the second adjacency matrix into the spatio-temporal graph neural network to obtain the predicted subgraph of regional road network traffic congestion may include the following steps:

[0099] Step S331: Extract hidden features from the input explanatory subgraph through the encoder to obtain the node feature matrix corresponding to the nodes where N is the number of sensor nodes in the explanatory subgraph, and d i input feature matrix X in is the feature dimension;

[0100] Step S332: Through the spatial graph neural network module, the node feature matrix of the explanatory subgraph and the second adjacency matrix A ∈ R N×N of the explanatory subgraph are aggregated with adjacent node information through graph convolution operations with position attention mechanisms to capture the spatial dependence relationship between roads, and the spatially enhanced feature matrix is output

[0101] Step S333: Through the gated recurrent unit module, the X processed by the spatial graph neural network module out , and the hidden state at the previous moment capture the local dependence relationship of the time series through a cyclic structure, and at the same time combine the spatial information to obtain the hidden state at the current moment for capturing the temporal relationship between consecutive time steps;

[0102] Step S334: Through the Transformer module, the hidden sequence output by the gated recurrent unit module position encoding e t directly captures the long-distance time dependence through the multi-head self-attention mechanism, enhances the model's ability to model periodic or sudden traffic patterns, and obtains the global time dependence feature

[0103] Step S335: Through the multi-layer regression head, map the spatio-temporal features in the global features output by the Transformer module to the prediction result to obtain the predicted subgraph of regional road network traffic congestion where T' is the predicted time step length.

[0104] Further, in an embodiment of the present invention, the feature matrix and the second adjacency matrix corresponding to the nodes in the explanatory subgraph can be determined through the prior art, and the feature matrix and the second adjacency matrix are input into the above spatio-temporal graph neural network to capture the spatio-temporal correlation characteristics, so as to obtain a prediction subgraph of regional road network traffic congestion for subsequent generation of explanatory statements of congestion events.

[0105] Step S4: Generate an explanatory text of a traffic congestion event in the urban road network based on the explanatory subgraph and the prediction subgraph.

[0106] Among them, in an embodiment of the present invention, after obtaining the explanatory subgraph and the prediction subgraph through the above steps, an explanatory text of a traffic congestion event in the urban road network can be generated based on the explanatory subgraph and the prediction subgraph.

[0107] Specifically, in an embodiment of the present invention, the method for generating an explanatory text of a traffic congestion event in the urban road network based on the explanatory subgraph and the prediction subgraph may include the following steps:

[0108] Step S41: Determine the distance matrix corresponding to the explanatory subgraph;

[0109] Step S42: Determine the traffic events corresponding to the explanatory subgraph through the agglomerative hierarchical clustering algorithm based on the distance matrix;

[0110] Step S43: Obtain the road information corresponding to the sensor nodes in the explanatory subgraph and the prediction subgraph through the reverse coding technology of the geographic information system;

[0111] Step S44: Generate an explanatory text of a traffic congestion event in the urban road network through a text output template based on the traffic events corresponding to the explanatory subgraph, the road information corresponding to the sensor nodes, the explanatory subgraph, and the prediction subgraph.

[0112] Among them, in an embodiment of the present invention, the distance matrix corresponding to the explanatory subgraph can be determined through the sixth formula to calculate the spatio-temporal distance and speed distance between all nodes in the prediction, where the sixth formula is:

[0113] M = W × M s +M d +M t

[0114] where M is the distance matrix, M d is the spatial distance matrix, M t is the time distance matrix, M s is the speed distance matrix; the matrices M, M d , M t , M s are all min-max standardized to compress their values into the interval [0, 1]; the spatial distance matrix M dDerived from the adjacency matrix, it describes the spatial distance between each node and is independent of the time step; the time distance matrix M t Describes the time distance between nodes at each time step; the speed distance matrix M s Describes the speed distance between nodes at each time step and is assigned a higher weight W so that the speed variable plays a dominant role in the subsequent clustering process.

[0115] Also, in an embodiment of the present invention, the agglomerative clustering algorithm can be used to cluster the distance matrix M to identify whether each node specified by the matrix M index in the explanatory subgraph belongs to an independent traffic cluster. Specifically, the number of clusters n for clustering will be tested within the range [1, 5], and finally the number of clusters n that can obtain the best score will be selected, so as to cluster the explanatory subgraph events into different types of traffic events, such as severe congestion, mild congestion, and free flow.

[0116] Furthermore, in an embodiment of the present invention, the score of the cluster can be calculated by the seventh formula to bias towards clusters with dissimilarity and low variance. The seventh formula is:

[0117]

[0118] Where CD is the dependence on inter-cluster dissimilarity; WCV is the within-cluster variance; N is the total number of clusters; n i is the number of nodes in the i-th cluster in the prediction network; is the variance of the node speeds in the i-th cluster; σ 2 is the overall speed variance of all predicted nodes; u i is the average speed value of the i-th cluster; |u i -u k | is the absolute difference in the average speeds between cluster i and cluster k.

[0119] Furthermore, in an embodiment of the present invention, the road information corresponding to the sensor node may include the road name and direction, and the specific distance of the sensor node from the starting point of the road. Among them, in an embodiment of the present invention, through the reverse coding technology of the geographic information system (GIS), the corresponding road, as well as the corresponding road name and direction, can be matched according to the longitude and latitude of the sensor node in the explanatory subgraph and the prediction subgraph, and the specific distance of the sensor node from the starting position longitude and latitude of the road can be obtained by calculating the distance between the longitude and latitude of the sensor node and the starting position longitude and latitude of the road.

[0120] Also, in an embodiment of the present invention, after obtaining the traffic events corresponding to the explanatory subgraph, the road information corresponding to the sensor nodes, the explanatory subgraph, and the prediction subgraph through the above steps, an explanatory text of the traffic congestion event in the urban road network can be generated based on the traffic events corresponding to the explanatory subgraph, the road information corresponding to the sensor nodes, the explanatory subgraph, and the prediction subgraph through a text output template. Specifically, a coherent narrative can be formed by replacing the placeholder in the text template with specific content. The explanatory text corresponding to the text template consists of multiple paragraphs. The first paragraph describes the predicted congestion event and briefly summarizes the reasons for generating the event; the subsequent paragraphs elaborate on the specific situations of each reason, that is, the reasons represented by each cluster in the re-explanatory subgraph. The paragraphs describing the reasons can be arranged in chronological order of the occurrence of the reasons to ensure the logic and clarity of the explanation. For example, a certain text template is "{section}{direction} is expected to have a {traffic event} from {start time} to {end time} at {time}, with an average vehicle speed of {speed} km / h. This is due to a series of alternating congested and unobstructed driving." At this time, the corresponding speed, location, and time can be extracted from the explanatory subgraph and the prediction subgraph respectively and replaced at the corresponding positions in the template, and then the text statement can be output in chronological order, as Figure 3 shown.

[0121] According to the method for explaining traffic congestion events in an urban road network proposed by an embodiment of the present invention, sensor data and regional traffic data of multiple regional road networks are obtained, and the regional traffic data is preprocessed to obtain target regional traffic data; based on the sensor data and the target regional traffic data, an explanatory subgraph of the regional road network is generated through an explanation extractor; based on the explanatory subgraph, a prediction subgraph of traffic congestion in the regional road network is generated through a spatio-temporal graph neural network; based on the explanatory subgraph and the prediction subgraph, an explanatory text of the traffic congestion event in the urban road network is generated. The present invention can generate an explanatory text of the traffic congestion event in the urban road network based on the explanatory subgraph of the regional road network and the prediction subgraph of traffic congestion in the regional road network, realizing the full traceability of the model prediction basis and high-trust decision support, enabling traffic managers to understand the prediction results, enhancing the application value of the model and the actual decision support ability.

[0122] Next, an apparatus for explaining traffic congestion events in an urban road network according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0123] Figure 4 It is a schematic structural diagram of an apparatus for explaining traffic congestion events in an urban road network according to an embodiment of the present invention.

[0124] As Figure 4 shown, the apparatus 10 for explaining traffic congestion events in the urban road network includes: a data processing module 401, a first generation module 402, a second generation module 403, and a third generation module 404, where

[0125] A data processing module 401, configured to obtain sensor data and regional traffic data of multiple regional road networks, and preprocess the regional traffic data to obtain target regional traffic data;

[0126] A first generation module 402, configured to generate an explanatory subgraph of the regional road network through an interpretation extractor based on the sensor data and the target regional traffic data;

[0127] A second generation module 403, configured to generate a prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network based on the explanatory subgraph;

[0128] A third generation module 404, configured to generate an explanatory text of a traffic congestion event in the urban road network based on the explanatory subgraph and the prediction subgraph.

[0129] Further, the above-mentioned data processing module 401 is specifically configured to:

[0130] Construct a regional traffic information collection system, where the regional traffic information collection system includes loop sensors;

[0131] Based on the loop sensors in the regional traffic information collection system, obtain sensor data and regional traffic data of multiple regional road networks;

[0132] Preprocess the regional traffic data to obtain target regional traffic data.

[0133] Further, the above-mentioned first generation module 402 is specifically configured to:

[0134] Construct a directed graph based on the sensor data and the target regional traffic data;

[0135] Based on the directed graph and the first adjacency matrix in the sensor data, obtain a candidate subgraph set through a graph attention network;

[0136] Perform Monte Carlo tree search in the candidate subgraph set to generate an explanatory subgraph of the regional road network.

[0137] Further, the above-mentioned second generation module 403 is specifically configured to:

[0138] Construct a spatio-temporal graph neural network;

[0139] Determine the second adjacency matrix corresponding to the nodes in the explanatory subgraph;

[0140] Input the explanatory subgraph and the second adjacency matrix into the spatio-temporal graph neural network to obtain a prediction subgraph of traffic congestion in the regional road network.

[0141] Further, the above-mentioned third generation module 404 is specifically configured to:

[0142] Determine the distance matrix corresponding to the explanatory subgraph;

[0143] Based on the distance matrix, determine the traffic events corresponding to the explanatory subgraph through the agglomerative hierarchical clustering algorithm;

[0144] Through the reverse coding technology of the geographic information system, obtain the road information corresponding to the sensor nodes in the explanatory subgraph and the prediction subgraph;

[0145] Based on the traffic events corresponding to the explanatory subgraph, the road information corresponding to the sensor nodes, the explanatory subgraph and the prediction subgraph, generate the explanatory text of the traffic congestion events in the urban road network through the text output template.

[0146] According to the explanatory device for traffic congestion events in the urban road network proposed in the embodiment of the present invention, obtain the sensor data and regional traffic data of multiple regional road networks, and preprocess the regional traffic data to obtain the target regional traffic data; based on the sensor data and the target regional traffic data, generate the explanatory subgraph of the regional road network through the explanatory extractor; based on the explanatory subgraph, generate the prediction subgraph of the traffic congestion in the regional road network through the spatio-temporal graph neural network; based on the explanatory subgraph and the prediction subgraph, generate the explanatory text of the traffic congestion events in the urban road network. The present invention can generate the explanatory text of the traffic congestion events in the urban road network based on the explanatory subgraph of the regional road network and the prediction subgraph of the traffic congestion in the regional road network, realizing the full traceability of the model prediction basis and high-trust decision support, enabling traffic managers to understand the prediction results, enhancing the application value of the model and the actual decision support ability.

[0147] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, "a plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0148] In the description of this specification, the description with reference to terms such as "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0149] Although the embodiments of the present invention have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those of ordinary skill in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for explaining traffic congestion events in an urban road network, characterized in that, The method includes: Obtaining sensor data and regional traffic data of multiple regional road networks, and preprocessing the regional traffic data to obtain target regional traffic data; Generating an explanatory subgraph of the regional road network through an explanation extractor based on the sensor data and the target regional traffic data; Generating a prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network based on the explanatory subgraph; Generating an explanatory text for traffic congestion events in the urban road network based on the explanatory subgraph and the prediction subgraph.

2. The method according to claim 1, wherein The obtaining sensor data and regional traffic data of multiple regional road networks, and preprocessing the regional traffic data to obtain target regional traffic data includes: Constructing a regional traffic information acquisition system, where the regional traffic information acquisition system includes loop sensors; Obtaining sensor data and regional traffic data of multiple regional road networks based on the loop sensors in the regional traffic information acquisition system; Preprocessing the regional traffic data to obtain target regional traffic data.

3. The method according to claim 1, characterized in that, The generating an explanatory subgraph of the regional road network through an explanation extractor based on the sensor data and the target regional traffic data includes: Constructing a directed graph based on the sensor data and the target regional traffic data; Obtaining a set of candidate subgraphs through a graph attention network based on the directed graph and the first adjacency matrix in the sensor data; Performing Monte Carlo tree search in the set of candidate subgraphs to generate an explanatory subgraph of the regional road network.

4. The method according to claim 1, characterized in that, The generating a prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network based on the explanatory subgraph includes: Constructing a spatio-temporal graph neural network; Determining a second adjacency matrix corresponding to the nodes in the explanatory subgraph; Inputting the explanatory subgraph and the second adjacency matrix into the spatio-temporal graph neural network to obtain a prediction subgraph of traffic congestion in the regional road network.

5. The method according to claim 1, characterized in that, The generating an explanatory text for traffic congestion events in the urban road network based on the explanatory subgraph and the prediction subgraph includes: Determining a distance matrix corresponding to the explanatory subgraph; Determining traffic events corresponding to the explanatory subgraph through an agglomerative hierarchical clustering algorithm based on the distance matrix; Obtaining road information corresponding to the sensor nodes in the explanatory subgraph and the prediction subgraph through the reverse coding technology of the geographic information system; Generating an explanatory text for traffic congestion events in the urban road network through a text output template based on the traffic events corresponding to the explanatory subgraph, the road information corresponding to the sensor nodes, the explanatory subgraph, and the prediction subgraph.

6. An apparatus for explaining traffic congestion events in an urban road network, characterized in that, The device includes: A data processing module for obtaining sensor data and regional traffic data of multiple regional road networks, and preprocessing the regional traffic data to obtain target regional traffic data; A first generation module for generating an explanatory subgraph of the regional road network through an explanation extractor based on the sensor data and the target regional traffic data; A second generation module for generating a prediction subgraph of traffic congestion in the regional road network through a spatio-temporal graph neural network based on the explanatory subgraph; A third generation module, configured to generate an explanation text for a traffic congestion event in an urban road network based on the explanation sub-graph and the prediction sub-graph.

7. The device according to claim 6, wherein The data processing module is specifically configured to: Construct a regional traffic information collection system, where the regional traffic information collection system includes loop sensors; Based on the loop sensors in the regional traffic information collection system, obtain sensor data and regional traffic data of multiple regional road networks; Preprocess the regional traffic data to obtain target regional traffic data.

8. The device according to claim 6, characterized in that, The first generation module is specifically configured to: Construct a directed graph based on the sensor data and the target regional traffic data; Based on the directed graph and the first adjacency matrix in the sensor data, obtain a candidate sub-graph set through a graph attention network; Perform Monte Carlo tree search in the candidate sub-graph set to generate an explanation sub-graph of the regional road network.

9. An electronic device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method according to any one of claims 1-5.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method according to any one of claims 1-5.

Images