Dynamic traffic bottleneck identification method based on space-time diagram attention network

Abstract

The invention discloses a dynamic traffic bottleneck identification method based on a space-time diagram attention network, and belongs to the field of intelligent traffic, and the method comprises the following steps: obtaining road network data and track data, and carrying out the preprocessing of the track data based on the road network data, and obtaining the preprocessed track data; constructing a space-time diagram attention network, and inputting the road network data and the preprocessed trajectory data into a space-time diagram attention network model to obtain node representation with time sequence dynamics; and a dynamic traffic bottleneck identification algorithm is adopted to process the node representation with the time sequence dynamics to identify the dynamic traffic bottleneck. According to the method, the road network node space proximity can be effectively extracted, the traffic flow propagation space-time dynamics can be accurately captured, the operation speed of the recognition algorithm is high, and the real-time requirement is met.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent transportation technology, and in particular relates to a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network. Background Art

[0002] Existing traffic bottleneck identification methods primarily explore the relationship between traffic demand and road network service levels. Among these, methods based on percolation theory have been widely studied. This method evaluates the reliability of traffic networks based on percolation theory and identifies bottlenecks by analyzing the overall impact of each road segment on the flow of travel demand within the network. This type of method is suitable for identifying fixed traffic bottlenecks caused by network planning and relatively stable, excessive traffic demand, and offers stability and predictability. However, for dynamic traffic bottlenecks caused by random changes in traffic flow and sudden increases in traffic demand, existing research primarily relies on probabilistic models and statistical methods, which cannot guarantee accurate bottleneck identification. This is because dynamic traffic bottlenecks are concurrent, transient, and propagating, requiring the study of the interrelationships between network nodes under real-time congestion conditions. Existing metrics struggle to quantify their dynamic spatiotemporal characteristics. Therefore, to capture rich network features, graph representation learning maps network data (i.e., graph data) into low-dimensional feature vectors. This method preserves most network information, such as the graph's topology and the relationships between nodes. It is widely used in downstream tasks such as node classification and traffic flow prediction. To quantify the traffic correlation between road segments, Road2Vec and Seg2Vec use vehicle trajectories to learn feature representations of road segments, demonstrating their outstanding performance in traffic flow prediction and congestion mitigation, respectively. However, these models are trained based on specific time periods, which limits their ability to capture the dynamic spatiotemporal characteristics of traffic flows.

[0003] Dynamic traffic bottlenecks are concurrent, instantaneous, and propagating, necessitating the study of the interrelationships between road network nodes under real-time congestion conditions. Existing metrics struggle to quantify their dynamic spatiotemporal characteristics, severely impacting the accuracy of bottleneck identification and the development of effective congestion control strategies. Therefore, this paper proposes a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a dynamic traffic bottleneck identification method based on spatiotemporal graph attention network to solve the problems existing in the above-mentioned prior art.

[0005] To achieve the above objectives, the present invention provides a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network, comprising:

[0006] Acquiring road network data and trajectory data, and preprocessing the trajectory data based on the road network data to obtain preprocessed trajectory data;

[0007] Constructing a spatiotemporal graph attention network, inputting the road network data and the preprocessed trajectory data into the spatiotemporal graph attention network model to obtain node representations with temporal dynamics;

[0008] A dynamic traffic bottleneck identification algorithm is used to process the node representation with temporal dynamics to identify dynamic traffic bottlenecks.

[0009] Optionally, the process of preprocessing the trajectory data based on the road network data to obtain preprocessed trajectory data includes:

[0010] Performing geospatial coordinate conversion on the trajectory data to obtain coordinate-converted trajectory data; wherein the coordinate-converted trajectory data and the road network data are data in the same coordinate system;

[0011] performing data cleaning on the trajectory data after the coordinate system conversion to obtain cleaned trajectory data;

[0012] The ST-Matching algorithm is used to map the cleaned trajectory data onto the road network to generate preprocessed trajectory data.

[0013] Optionally, the process of inputting the road network data and the preprocessed trajectory data into the spatiotemporal graph attention network to obtain a node representation with temporal dynamics includes:

[0014] The graph attention enhancement module of the spatiotemporal graph attention network extracts spatial correlation and traffic flow propagation relationships based on the road network data and the preprocessed trajectory data;

[0015] Calculating an attention coefficient based on a multi-head graph attention mechanism, and modeling the spatial correlation relationship and traffic flow propagation relationship based on the attention coefficient to obtain a single time slice node representation;

[0016] Based on the long short-term memory model, several single time slice node representations are processed to obtain node representations with temporal dynamics.

[0017] Optionally, the graph attention enhancement module includes: space enhancement GAT and traffic flow enhancement GAT;

[0018] The process of extracting spatial correlation and traffic flow propagation relationships based on the graph attention enhancement module includes:

[0019] The spatial enhancement GAT processes the road network data based on the node spatial coordinates and the road section length to obtain a spatial correlation relationship;

[0020] The traffic flow enhancement GAT processes the pre-processed trajectory data to obtain a traffic flow propagation relationship.

[0021] Optionally, the calculation expression of the attention coefficient is:

[0022]

[0023] Among them, a ij represents the importance of node j to node i, A is the attention network, and is the feature vector of node i and node j, For edge e ij The eigenvectors of W and W e are the weight matrices of nodes and edges respectively.

[0024] Optionally, the process of processing the plurality of single time slice node representations based on the long short-term memory model to obtain node representations with temporal dynamics includes:

[0025] Combining the plurality of single time slice node representations into a time series according to time granularity;

[0026] Input the time series into the long short-term memory model, and output the time series enhanced node representation through the input gate, the forget gate and the output gate;

[0027] A node representation with temporal dynamics is obtained based on the temporal enhanced node representation.

[0028] Optionally, the process of using a dynamic traffic bottleneck identification algorithm to process the node representation with temporal dynamics to identify the dynamic traffic bottleneck includes:

[0029] Calculate congested road sections based on roads, road speeds, and free-flow speeds, and construct a congested road section set based on a number of congested road sections;

[0030] Calculating the congestion propagation impact of each road section based on the set of congested road sections;

[0031] Arrange the congestion propagation impact of each road section in descending order to identify dynamic traffic bottlenecks.

[0032] Optionally, the expression for calculating the congestion propagation impact of each road segment is:

[0033]

[0034] Where, e ij For road section, E c is a local congestion cluster, N n () represents the n-order neighbor set of a node, Sim() is the traffic correlation measurement formula, h u 、hi 、h j are the feature vectors of nodes u, i, and j respectively, and E t is the set of congested road sections, PI(·) is the congestion propagation impact of the road section, e ui is the road section between nodes u and i, e ju is the road section between nodes j and u.

[0035] Compared with the prior art, the present invention has the following advantages and technical effects:

[0036] This paper proposes a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network. This method obtains road network data and trajectory data, preprocesses the trajectory data to match it with the road network data, and provides an accurate data foundation for subsequent processing. A spatiotemporal graph attention network model is then constructed, and the road network data and preprocessed trajectory data are input into the model to obtain node representations with temporal dynamics. This model integrates the graph attention mechanism with the long-short-term memory network architecture, effectively characterizing the coupling relationship between the spatial topology of the road network and the spatiotemporal characteristics of traffic flow, and accurately measuring the traffic correlation between road network nodes. A dynamic traffic bottleneck identification algorithm is then used to process the node representations with temporal dynamics, efficiently and accurately identifying dynamic traffic bottlenecks. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0038] Figure 1 This is an example of a road network directed graph according to an embodiment of the present invention;

[0039] Figure 2 A spatiotemporal graph attention network according to an embodiment of the present invention;

[0040] Figure 3 is the cosine similarity of the 3n-order neighbors in the embodiment of the present invention, where (a) is the morning peak and (b) is the evening peak;

[0041] Figure 4 is the dynamic cosine similarity of an embodiment of the present invention;

[0042] Figure 5 The operating efficiency of the embodiment of the present invention is shown in Figure 1, where Figure 1 (a) shows the online processing efficiency and Figure 1 (b) shows the model training time.

[0043] Figure 6 This is a visualization of the congestion propagation influence of a local congestion cluster near the west side of Tianfu Square from 8:55 to 9:00 on October 5, 2016, according to an embodiment of the present invention;

[0044] Figure 7 The traffic control effect of the embodiment of the present invention is shown in Figure 1, where Figure 1 (a) shows the improvement in traffic speed and Figure 1 (b) shows the duration of congestion.

[0045] Figure 8 This is the SUMO traffic control interface of an embodiment of the present invention, where Figure (a) shows the traffic congestion status before control, and Figure (b) shows the traffic status after control of road section 1;

[0046] Figure 9 This is a flow chart of a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0047] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0048] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0049] The present invention aims to construct a graph representation learning model that integrates the spatial topological structure of the road network and the dynamic spatiotemporal characteristics of traffic flow, and to quantitatively analyze the spatiotemporal propagation characteristics of traffic bottlenecks. To achieve this technical goal, the present invention proposes a road network representation model based on the spatiotemporal graph attention network ST-GAT (Spatio-Temporal Graph Attention Networks), which captures the topological structure associations between road nodes through spatially enhanced GAT, and models the dynamic impact of traffic flow propagation through traffic flow enhanced GAT, and adopts a multi-head attention mechanism to adaptively learn different spatial association patterns to highlight the relationship between key nodes. This model innovatively combines the graph attention mechanism with the long short-term memory model LSTM (Long Short-Term Memory), and breaks through the limitations of traditional methods for complex road network representation through dual modeling of spatial topological constraints and dynamic propagation of traffic flow. In addition, local congestion clusters are defined for frequent traffic congestion conditions, a congestion propagation impact quantification model based on traffic correlation between nodes is constructed, and a dynamic traffic bottleneck identification algorithm is proposed.

[0050] As a specific implementation of this embodiment, the following definitions are made:

[0051] (1) Road network: A road network is abbreviated as road network and is defined as a directed graph G = (V, E). V is a set of nodes, representing intersections in the road network, and E is a set of directed edges. Each edge eij ∈E consists of a source node v i and a target node v j Sure.

[0052] (2) n-hop neighborhood: Given a directed (or undirected) graph G = (V, E), the n-hop neighbors of a node u∈V are represented by N n (u), is defined as the node in graph G whose shortest distance to node u is no greater than n. Figure 1 As shown, Figure 1 The first-order neighbors of the middle node v1 are {v2,v3,v5,v7}, and the second-order neighbors are {v2,v3,v4,v5,v6,v7,v8 } .

[0053] (3) Node representation: Given a road network graph G = (V, E), node representation aims to learn the feature representation f of the node: Map each node to a d-dimensional feature vector.

[0054] (4) Road speed: Given a road e, the average speed S of all vehicles passing through the road e within a specified time interval t e ={s1,...,s k} (k is the number of vehicles), and the confidence threshold min_sup, the road speed is defined as the average of the average speeds of all vehicles, see formula (1).

[0055]

[0056] (5) Congested road: given a road e and its road speed v e,t , free flow speed v f When v e,t Less than or equal to v f When the traffic volume is C times of that of the road e, the road e is defined as a congested road section. C is the default parameter, 0≤C<1.

[0057] (6) Local congestion cluster: The set of congested road sections E within a certain time interval t is known. t , the local congestion cluster c is defined as E t The graph structure is composed of a subset of , that is, c=(V c ,E c ), so that E c Each segment e in ij At least one of the following conditions must be met:

[0058] 1)eij It is a congested road section: all n-order neighbors of nodes i and j belong to V c ;

[0059] 2)e ij Not a congested road segment: At least one of the n-th order neighbors of node i or j belongs to E c .

[0060] (7) Congestion propagation influence: The set of congested road sections E in a certain time interval t is known. t , and local congestion cluster c=(V c ,E c ), the node representation h of the next period t+1 , road section e ij In E c The congestion propagation effect in is calculated by formula (2):

[0061]

[0062] Among them, Sim() is a traffic correlation measurement formula, such as cosine similarity. ij For road section, E c is a local congestion cluster, N n () represents the n-order neighbor set of a node, h u 、h i 、h j are the feature vectors of nodes u, i, and j respectively, and E t is the set of congested road sections, PI(·) is the congestion propagation impact of the road section, e ui is the road section between nodes u and i, e ju is the road section between nodes j and u. The above definition takes into account the directionality of the impact of traffic congestion propagation: when e ij When it is a congested section, the section will only affect the traffic conditions of its upstream section; ij When it is a non-congested road section, the traffic flow on this road section only affects its downstream road sections.

[0063] (8) Dynamic traffic bottleneck identification: Given a local congestion cluster c = (V c ,E c ), given a positive integer K(K≤|E c |), the dynamic traffic bottleneck is defined as E c A subset of E B , so that the sum of the congestion propagation effects of this subset is maximized, and |E B |=K.

[0064] This embodiment provides a dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network, including the following steps:

[0065] Problem Description: Traffic congestion typically causes vehicles on a particular road to slow down for short but recurring periods. This method defines congested sections by comparing road speeds to free-flow speeds (FFS). FFS refers to the average speed drivers would travel in non-congested or otherwise unfavorable conditions. This method uses the F percentile of all valid speeds on a road section as its FFS, with a default value of 85. By monitoring vehicle speeds, traffic congestion can be detected quickly and accurately.

[0066] Spatiotemporal Graph Attention Network (ST-GAT): The spatiotemporal graph attention network proposed in this paper is as follows Figure 2 As shown. The model first extracts the spatial correlation relationship and traffic flow propagation relationship of each node with its neighboring nodes based on the graph attention enhancement module, namely the spatial enhancement GAT and the traffic flow enhancement GAT. The attention coefficient is calculated based on the multi-head graph attention mechanism. Each group of attention mechanisms can learn different node associations and feature weights, thereby highlighting the complex spatial structure of the road network, and injects the graph structure information into the model through masked attention, that is, the calculation of the attention coefficient only considers the n-order neighbors of the node, such as the 2nd-order neighbors. Then, the output node representation vector is composed of a time series according to the specified time granularity (such as 5 minutes). The feature vector of each time interval expresses the road network structure characteristics and traffic flow propagation relationship within the time period. Since the traffic flow changes dynamically over time, the present invention extracts the spatiotemporal variation characteristics of the traffic flow based on the LSTM neural network, thereby measuring the dynamic traffic correlation between nodes. Finally, the prediction result is output through the fully connected layer (Fully Connected Layer), that is, the traffic flow at the next moment.

[0067] Furthermore, spatial feature enhancement GAT is used to extract the spatial correlation between nodes in the road network based on the spatial coordinates of the nodes and the length of the road section. ij Indicates the importance of node j to node i. The attention coefficient is calculated by formula (3). The features of node i and node j, as well as the edge features, are linearly transformed, then concatenated and mapped to a real number, namely the attention coefficient.

[0068]

[0069] Where a ij represents the importance of node j to node i, A is the attention network, and is the feature vector of node i and node j, For edge e ij The eigenvectors of W and W e are the weight matrices of nodes and edges respectively.

[0070] Finally, Softmax is used for normalization to obtain the standardized attention coefficient, as shown in formula (4):

[0071]

[0072] Furthermore, traffic flow features enhance GAT. The input node feature is the traffic flow passing through the node within a certain time interval. ij The characteristics of include the number of vehicles from node i to node j and the number of vehicles from node j to node i in a certain time interval t, that is, The attention coefficient between a central node and its neighboring nodes is affected by the traffic flow between the nodes. Because traffic flow is directional, the importance of a neighboring node to the central node depends on the amount of traffic flowing from the central node to the neighboring nodes. This means that when anomalies such as traffic congestion occur at neighboring nodes, the amount of communication flowing out of the central node will be affected.

[0073] Furthermore, for the spatiotemporal feature extraction of traffic flow, the present invention adopts LSTM network as the core architecture of time series modeling. The LSTM module is composed of several interrelated memory units, each of which contains three gate structures: input gate, forget gate, and output gate. By parameterizing the information screening and state update process, it effectively solves the gradient vanishing problem in traditional recurrent neural networks. In the specific implementation, the node representation vector generated by the graph attention enhancement module is constructed into an ordered data sequence in the time dimension according to the specified time granularity (such as 5 minutes) as the time series input of the LSTM network.

[0074] This architectural design enables the model to effectively capture the nonlinear variations in traffic flow within consecutive time slots, particularly during peak hours. Through a multi-level feature fusion mechanism, the resulting node representation vector not only incorporates the spatial correlation characteristics of the road network topology but also dynamically models the propagation effects of traffic bottlenecks through the temporal memory function of gated recurrent units.

[0075] Dynamic traffic bottleneck identification algorithm: Based on the vehicle GPS trajectory data continuously collected by the real-time traffic monitoring system, this paper proposes a dynamic traffic bottleneck identification method (as shown in Algorithm 1). This method uses a three-stage processing flow to screen key nodes with cascading effects from the detected congested sections:

[0076] (1) Local congestion cluster generation: Based on the road network graph structure, a spatially correlated congestion cluster within the n-th order neighborhood is constructed for each congested road segment (line 1). This operation is implemented using a breadth-first search algorithm to ensure that the road segments within each cluster have direct or indirect connectivity.

[0077] (2) Dynamic propagation impact calculation: Calculate the congestion propagation impact of each road segment in the local congestion cluster based on Equation (2) in Definition 7 (line 6).

[0078] (3) Generation of key congested road sections: Arrange the road sections in the local congestion cluster in descending order according to the congestion propagation impact value (line 9), and select the top K road sections as key congested road sections, i.e., traffic bottlenecks (line 10).

[0079]

[0080]

[0081] Furthermore, as another specific implementation of this embodiment, Figure 9 As shown in Figure 2, an open-source real-world dataset is used for evaluation experiments.

[0082] Obtain road network data and trajectory data. The object dataset is an open-source real dataset, including OSM road network data and the taxi trajectory dataset provided by DiDi. See Table 1 for details.

[0083] Table 1

[0084]

[0085] This embodiment evaluates the effectiveness and operating efficiency of the proposed method through comparative experiments and case analysis. The effectiveness of the proposed ST-GAT model is evaluated by learning the representation vectors of road network nodes, calculating the cosine similarity between neighboring nodes, and conducting comparative experiments with Road2Vec and Seg2Vec. The efficiency of the algorithm is evaluated by the execution time of offline mining and online processing processes. In addition, the effectiveness of key congested road identification is evaluated through case analysis based on SUMO (Simulation of Urban MObility) traffic simulation software. The operating environment of the experiment is a 64-bit server with Ubuntu 20.04.4 (operating system), which has an Intel Xeon Gold 6226R CPU @ 2.90GHz × 32 and 256GB RAM, and a GPU of GeForce RTX3090 (24G).

[0086] The trajectory data is preprocessed based on the road network data to obtain preprocessed trajectory data. Since the original trajectory data is disordered and full of noise, this embodiment constructs a process-based preprocessing operation to improve data quality. The preprocessing process includes the following three key steps:

[0087] (1) Geospatial coordinate system conversion: Since the geospatial coordinate system of the collected taxi trajectory data is the Martian coordinate system, which is inconsistent with the WGS84 coordinate system of the OSM road network data, the trajectory data is converted to the location points of the WGS84 coordinate system to support subsequent map matching and other operations, and the trajectory data after coordinate system conversion is obtained.

[0088] (2) Data cleaning: After the coordinate system is converted, the trajectory is cleaned by performing data cleaning operations such as deduplication, deletion of small trajectories and outliers to obtain the cleaned trajectory data. Among them, outliers include trajectory points with speeds far exceeding the legal speed limit and trajectory points with too long time intervals. The method for handling outliers is to disconnect the trajectory from the outlier point and divide it into two trajectories.

[0089] (3) Map Matching: Due to errors in the data collection process, there is a certain error distance between the spatial position of the trajectory points and the road network. Therefore, the present invention uses an efficient map matching algorithm, namely the ST-Matching algorithm, to map the trajectory data onto the road network to obtain pre-processed trajectory data, and then generate trajectory data composed of road network nodes.

[0090] After the data preprocessing stage, the raw trajectory data were converted into 3.68 million time-ordered trajectory records, including approximately 657 million location points.

[0091] A spatiotemporal graph attention network is constructed, and the road network data and the preprocessed trajectory data are input into the spatiotemporal graph attention network model to obtain a node representation with temporal dynamics; the process includes: the graph attention enhancement module of the spatiotemporal graph attention network extracts spatial correlation and traffic flow propagation relationships based on the road network data and the preprocessed trajectory data; the attention coefficient is calculated based on the multi-head graph attention mechanism, and the spatial correlation and traffic flow propagation relationships are modeled based on the attention coefficient to obtain a single time slice node representation; and a number of the single time slice node representations are processed based on the long short-term memory model to obtain a node representation with temporal dynamics.

[0092] Furthermore, the graph attention enhancement module includes: spatial enhancement GAT and traffic flow enhancement GAT; the process of extracting spatial correlation relationships and traffic flow propagation relationships based on the graph attention enhancement module includes: the spatial enhancement GAT processes the road network data based on the node spatial coordinates and road section length to obtain spatial correlation relationships; the traffic flow enhancement GAT processes the preprocessed trajectory data to obtain traffic flow propagation relationships.

[0093] Furthermore, the process of processing several of the single time slice node representations based on the long short-term memory model to obtain node representations with temporal dynamics includes: organizing several of the single time slice node representations into a time series according to time granularity; inputting the time series into the long short-term memory model, and outputting a temporal enhanced node representation through an input gate, a forgetting gate, and an output gate; and obtaining a node representation with temporal dynamics based on the temporal enhanced node representation.

[0094] A dynamic traffic bottleneck identification algorithm is used to process the temporally dynamic node representations to identify dynamic traffic bottlenecks. This process includes: calculating congested road sections based on road conditions, road speeds, and free-flow speeds; constructing a congested road section set based on a number of congested road sections; calculating the congestion propagation impact of each road section based on the congested road section set; and sorting the congestion propagation impacts of each road section in descending order to identify dynamic traffic bottlenecks.

[0095] A comparative experiment was conducted based on the above technical solution, and the results of the comparative experiment are as follows.

[0096] (1) Spatial adjacency between road network nodes: Based on the representation vector of the road network node, the cosine similarity between n-order neighbor nodes is calculated to measure the spatial adjacency between nodes. The morning and evening peak time intervals are selected to train Road2Vec, Seg2Vec and ST-GAT respectively to obtain the representation vector of the road network node, and the average cosine similarity of the n-order neighbors is calculated. Figure 3 As can be seen, the cosine similarity of all three models shows a downward trend with increasing neighbor order, indicating that the greater the spatial topological distance between nodes, the lower the cosine similarity. Among them, the downward trend is more pronounced for ST-GAT within third-order neighbors, indicating that this model better reflects changes in spatial adjacency. Furthermore, the cosine similarity between nodes changes more significantly during the evening rush hour than during the morning rush hour. This is because the morning rush hour is dominated by commuting, with relatively fixed routes (e.g., residential area to workplace), and traffic flow exhibits a strong regularity.

[0097] However, during the evening peak, travel destinations are more dispersed (e.g., workplace → residential area, commercial area, school, etc.), and vehicles frequently switch routes, resulting in more unstable neighbor relationships among nodes in the road network.

[0098] (2) Dynamic propagation characteristics of traffic flow: The ST-GAT model proposed in this paper is based on the graph attention enhancement module, which extracts the spatial correlation between road network nodes and the traffic flow propagation relationship. The node feature vector output for each time interval expresses the road network structure characteristics and traffic flow propagation relationship within the time period. The local nodes and their first-order and second-order neighbors are selected to compare the cosine similarity of Road2Vec, Seg2Vec, and ST-GAT in the time period of 8:00-9:00. Figure 4As shown in the figure, the cosine similarity of Road2Vec and Seg2Vec does not change over time because the node feature vectors obtained by their training are static and unchanged. The cosine similarity between nodes obtained by the ST-GAT model proposed in the present invention fluctuates over time, and its fluctuation pattern has a certain correlation with the change in traffic flow. The green lines in the figure are the traffic flow values ​​in different time periods. In the first half of the time interval, the traffic flow is in the growth period, and the correlation between road network nodes is not stable. In the later time interval, the traffic flow reaches a certain magnitude, and the correlation between nodes also shows a similar rise and fall trend with the rise and fall of traffic flow. Therefore, the proposed ST-GAT model can reflect the dynamic correlation of traffic flow between road network nodes. In addition, through model parameter adjustment, the model training effect of the following parameters is the best: the learning rate is 0.001, the number of heads is 8, and the number of hidden layers is half of the output number, i.e. 64.

[0099] (3) Operational efficiency: In terms of operational efficiency, Figure 5 (a) shows the online processing runtime of the traffic bottleneck identification method proposed in this invention, including data cleaning, map matching, congestion detection and bottleneck identification. Figure 5 As shown in (a), when the number of trajectory points increases from 10,000 to 300,000, the running time of the online processing process is basically within 1 minute. The results show that the traffic bottleneck identification method proposed in this paper is highly efficient and suitable for real-time traffic monitoring scenarios.

[0100] in addition, Figure 5 (b) shows the comparative experimental results of model training time. Seg2Vec-r-30 and Seg2Vec-r-50 respectively represent the training time when the random walk trajectory length r is 30 and 50. When r increases, the training time will increase accordingly. Figure 5 As shown in (b), the training time of the proposed spatiotemporal graph attention model consists of three stages: S-GAT, T-GAT, and LSTM. S-GAT, a spatial feature enhancement GAT, has the shortest runtime; T-GAT, a traffic flow feature enhancement GAT, has a similar training time to Seg2Vec-r-50; the LSTM module training time is shorter than Seg2Vec-r-50; and the efficiency of all three stages of the model is superior to Road2Vec. Therefore, the proposed model has certain scalability in terms of training efficiency.

[0101] Furthermore, as another specific implementation method of this embodiment, in order to further verify the effectiveness of the proposed model, SUMO traffic simulation software is used to simulate the traffic status after traffic control based on key congested sections, thereby verifying the effect of traffic bottleneck identification on traffic control.

[0102] The local congested road sections in Chengdu during the morning rush hour (8:55-9:00) on October 5, 2016 were selected. The node representation vectors of the next time interval were predicted based on the proposed ST-GAT model, and the congestion propagation influence (PI value) was calculated based on formula (2). The visualization results of the PI value ranking (as shown in Table 2) are as follows: Figure 6 As shown, the road sections are represented by nodes, and the road sections with larger PI values ​​have larger nodes.

[0103] Based on the SUMO traffic simulator, each road section was selected for traffic control. The traffic speed improvement after 20 minutes of control is as follows: Figure 7 As shown in (a), the top 5 controlled sections are Section-1, Section-4, Section-2, Section-5, and Section-13, which are basically consistent with the PI value ranking. In addition, the congestion duration results of the top 5 controlled sections are as follows: Figure 7 As shown in (b), the control effect on Section 1 was the best, shortening the congestion duration by over 50%, followed by Sections 4 and 2, which is consistent with the PI ranking. Notably, the congestion duration did not decrease when both Sections 1 and 4 were controlled. This is because controlling more sections may lead to more vehicles taking detours, increasing travel time, and thus being less effective than a solution with fewer sections under control. Figure 8 This is the SUMO visualization interface, showing the traffic conditions before and after the controlled road section 1. The numbers in the figure represent the road section numbers, corresponding to Figure 6 The traffic conditions on the previously congested sections 9 and 10 have improved significantly.

[0104] Table 2

[0105]

[0106] Real-time governance of urban traffic congestion is a key challenge to improving urban operation efficiency, and its core lies in accurately identifying the root cause of congestion - dynamic traffic bottlenecks. In response to this urgent need, the present invention deeply explores the spatiotemporal propagation characteristics of traffic congestion and proposes an innovative dynamic traffic bottleneck identification method. The core contribution of the present invention lies in the construction of a spatiotemporal graph attention network (ST-GAT) model, which creatively integrates the graph attention mechanism with the long short-term memory (LSTM) network architecture, effectively depicts the complex spatial topological structure of the road network and its coupling relationship with the dynamic spatiotemporal characteristics of traffic flow, thereby accurately measuring the traffic correlation between road network nodes. Based on this, the present invention further defines the concept of local congestion clusters, and constructs a model to quantify the impact of congestion propagation, ultimately achieving efficient and accurate identification of dynamic traffic bottlenecks.

[0107] Experimental verification fully demonstrates the superiority of the proposed method: the ST-GAT model can not only effectively extract the spatial proximity of road network nodes, but also accurately capture the complex spatiotemporal dynamics implied by traffic flow propagation; the developed dynamic bottleneck identification algorithm has an efficient running speed and meets real-time requirements. More importantly, through a case analysis of typical traffic scenarios in Chengdu on the SUMO simulation platform, the real-time identification of dynamic traffic bottlenecks was successfully achieved. The targeted traffic control strategy formulated based on the identification results significantly shortened the duration of congestion (by more than 50%), which effectively verified the effectiveness and great potential of this method in actual traffic control applications.

[0108] In summary, the dynamic traffic bottleneck identification method based on a spatiotemporal graph attention network proposed in this paper provides a powerful theoretical model and practical tool for deepening understanding of the propagation mechanism of traffic congestion and accurately locating the root causes of congestion. Its efficient identification capabilities and significant application results provide important scientific basis and technical support for urban traffic managers to make real-time and accurate traffic control decisions, and are of great significance in promoting the development of intelligent and refined urban traffic congestion management.

[0109] To address the need for real-time management of urban road traffic congestion, a quantitative analysis model for the propagation characteristics of traffic congestion is developed to identify the root causes of traffic congestion, namely, traffic bottlenecks. This paper proposes a spatiotemporal graph attention network (ST-GAT). This model innovatively combines a graph attention mechanism with a long short-term memory (LSTM) architecture to construct a graph representation learning model that integrates the spatial topology of road networks with the spatiotemporal characteristics of traffic flows. This model measures the traffic correlations between nodes in the road network. It also defines local congestion clusters for frequent traffic congestion and constructs a quantitative model for the propagation of congestion based on the traffic correlations between nodes, thereby identifying dynamic traffic bottlenecks. Experimental results demonstrate that the proposed spatiotemporal graph attention network can both extract the spatial proximity of road network nodes and represent the spatiotemporal dynamics of traffic flow propagation. Furthermore, the dynamic traffic bottleneck identification algorithm operates efficiently. Furthermore, a case study of traffic conditions in Chengdu, using the SUMO traffic simulation software, demonstrates the real-time identification of dynamic traffic bottlenecks. Traffic control strategies based on the identified bottlenecks can shorten the duration of congestion by over 50%, demonstrating the effectiveness of this method in formulating traffic control measures and providing decision support for real-time management of traffic congestion.

[0110] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A dynamic traffic bottleneck identification method based on spatiotemporal graph attention network, characterized by: The following steps are involved: Acquiring road network data and trajectory data, and preprocessing the trajectory data based on the road network data to obtain preprocessed trajectory data; Constructing a spatiotemporal graph attention network, inputting the road network data and the preprocessed trajectory data into the spatiotemporal graph attention network model to obtain node representations with temporal dynamics; A dynamic traffic bottleneck identification algorithm is used to process the node representation with temporal dynamics to identify dynamic traffic bottlenecks.

2. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 1 is characterized in that: The process of preprocessing the trajectory data based on the road network data to obtain preprocessed trajectory data includes: Performing geospatial coordinate conversion on the trajectory data to obtain coordinate-converted trajectory data; wherein the coordinate-converted trajectory data and the road network data are data in the same coordinate system; performing data cleaning on the trajectory data after the coordinate system conversion to obtain cleaned trajectory data; The ST-Matching algorithm is used to map the cleaned trajectory data onto the road network to generate preprocessed trajectory data.

3. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 1 is characterized in that: The process of inputting the road network data and the pre-processed trajectory data into the spatiotemporal graph attention network to obtain a node representation with temporal dynamics includes: The graph attention enhancement module of the spatiotemporal graph attention network extracts spatial correlation and traffic flow propagation relationships based on the road network data and the preprocessed trajectory data; Calculating an attention coefficient based on a multi-head graph attention mechanism, and modeling the spatial correlation relationship and traffic flow propagation relationship based on the attention coefficient to obtain a single time slice node representation; Based on the long short-term memory model, several single time slice node representations are processed to obtain node representations with temporal dynamics.

4. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 3 is characterized in that: The graph attention enhancement module includes: space enhancement GAT and traffic flow enhancement GAT; The process of extracting spatial correlation and traffic flow propagation relationships based on the graph attention enhancement module includes: The spatial enhancement GAT processes the road network data based on the node spatial coordinates and the road section length to obtain a spatial correlation relationship; The traffic flow enhancement GAT processes the pre-processed trajectory data to obtain a traffic flow propagation relationship.

5. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 3 is characterized in that: The calculation expression of the attention coefficient is: Among them, a ij represents the importance of node j to node i, A is the attention network, and is the feature vector of node i and node j, For edge e ij The eigenvectors of W and W e are the weight matrices of nodes and edges respectively.

6. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 3 is characterized in that: The process of processing the single time slice node representations based on the long short-term memory model to obtain node representations with temporal dynamics includes: Combining the plurality of single time slice node representations into a time series according to time granularity; Input the time series into the long short-term memory model, and output the time series enhanced node representation through the input gate, the forget gate and the output gate; A node representation with temporal dynamics is obtained based on the temporal enhanced node representation.

7. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 1 is characterized in that: The process of using a dynamic traffic bottleneck identification algorithm to process the node representation with temporal dynamics to identify the dynamic traffic bottleneck includes: Calculate congested road sections based on roads, road speeds, and free-flow speeds, and construct a congested road section set based on a number of congested road sections; Calculating the congestion propagation impact of each road section based on the set of congested road sections; Arrange the congestion propagation impact of each road section in descending order to identify dynamic traffic bottlenecks.

8. The dynamic traffic bottleneck identification method based on spatiotemporal graph attention network according to claim 7 is characterized in that: The expression for calculating the congestion propagation impact of each road segment is: Where, e ij For road section, E c is a local congestion cluster, N n () represents the n-order neighbor set of a node, Sim() is the traffic correlation measurement formula, h u 、h i 、h j are the feature vectors of nodes u, i, and j respectively, and E t is the set of congested road sections, PI(·) is the congestion propagation impact of the road section, e ui is the road section between nodes u and i, e ju is the road section between nodes j and u.

Images