A dynamic traffic flow direction walking method for traffic key road identification

Abstract

This invention relates to a dynamic traffic flow direction walking method for identifying critical roads, comprising the following steps: selecting a traffic database X and preprocessing X using the networkx and pandas libraries to obtain a traffic network line graph G; constructing a critical road identification model M, and then using G as input to M to obtain two embedded feature vectors Path. v and Path v The data is concatenated to obtain the final embedded vector. A clustering algorithm is then used to cluster these vectors, resulting in λ classes of traffic segments. For each traffic segment class, the average traffic demand index of its constituent nodes is calculated and compared. The class with the highest average index is selected as the key class. The traffic demand indices of all nodes in the key class are then sorted in descending order, and the top W values ​​are selected as the final list of key roads, which is then output. This invention provides higher accuracy in identifying key roads in unknown areas, offering better reference for traffic resource planning and utilization.

Description

Technical Field

[0001] This invention relates to the field of critical traffic road identification, and in particular to a dynamic traffic flow walking method for identifying critical traffic roads. Background Technology

[0002] In the field of critical road segment identification, researchers are committed to improving the accuracy and practicality of identifying key nodes in road networks by integrating traffic features with advanced algorithm models, providing important technical support for urban traffic management. With the development of complex network theory and spatiotemporal data mining techniques, critical road segment identification frameworks based on graph neural networks have been widely applied. The introduction of a dual network structure of road entity topology and traffic flow topology has broken through the limitations of traditional single road network representation, providing underlying data structure support for identification methods based on graph representation learning. This allows for the extraction of low-dimensional embedding features of road segment nodes using graph neural networks, effectively capturing the spatial correlation and traffic propagation characteristics of the road network. The evaluation system for critical road segments is also gradually shifting from single-feature index evaluation to multi-dimensional feature index evaluation. Analysis of recent research results shows that the selection and evaluation methods for critical road segments in road networks are shifting from traditional index-based evaluation to key road segment selection and evaluation methods using dynamic feature learning.

[0003] Although these methods have achieved some success in the field of critical road segment identification, some problems still exist:

[0004] (1) Traditional critical road segment research mainly employs two methods: one is the index evaluation method, including single-index evaluation and multi-index evaluation, which identifies key nodes in the network by selecting one or more measurement indicators to establish an evaluation system; the other is to identify key nodes in the network by using node deletion or node shrinkage methods, which identifies key nodes by comparing the changes in network performance before and after removing a certain node. Traditional critical road segment evaluation methods rely on artificially designed indicators, which may lead to biased results. The iterative node deletion method is not suitable for large-scale traffic networks, cannot fully obtain the spatial correlation between nodes from a graph perspective, and cannot capture complex network characteristics, such as dynamic changes and high-order relationships between nodes.

[0005] (2) With the development of graph neural networks, key node selection methods based on graph neural networks have been gradually applied to traffic networks. Graph neural networks automatically learn the features of nodes and edges through node aggregation mechanisms, and can capture features that are difficult to obtain by traditional methods, which has advantages in handling complex large-scale graph classification tasks. In most existing works based on graph neural networks, key road segment acquisition methods only obtain node embeddings from the perspective of random walks, fail to accurately identify the dynamic transitivity features of traffic flow, ignore the fusion of dynamic and static features in multi-dimensional regions, and the graph random sampling mechanism lacks specificity for traffic networks. Summary of the Invention

[0006] To address the aforementioned problems in existing technologies, the technical problem this invention aims to solve is: how to more accurately identify critical roads. To solve this technical problem, this invention adopts the following technical solution:

[0007] A dynamic traffic flow direction walking method for identifying traffic-critical roads includes the following steps:

[0008] Select the traffic database X for a certain region:

[0009] S100: The networkx and pandas libraries in Python are used to process X to obtain the traffic data information table corresponding to X. The traffic data information table is converted into a matrix to obtain the traffic network line diagram G. The traffic data information is based on the records of all license plates passing through the detectors within each hour. The hourly traffic flow between the detectors belonging to each group of related nodes is calculated, and the connection relationship of the detector road network is used.

[0010] The expression for the traffic network line diagram G is as follows:

[0011]

[0012] Where V=( ) is a set of nodes. Let G be the nth node. This represents the total number of nodes in G. This represents a node in G, and also represents a segment of a road in a certain region; Let G represent the set of edges. In G and One edge between them, ; () represents the set of average velocities of each node in G within a specific time T. Let G represent the average speed of a vehicle passing through the nth node; () represents the set of traffic flow at each node in G within a specific time T. Let G represent the traffic flow passing through the nth node; ) represents the set of actual road segment lengths corresponding to nodes. This represents the length of the actual road segment corresponding to the nth node in G; =( ) represents the set of node traffic densities. Let G represent the traffic density passing through the nth node; =( ) represents the set of node degrees. This represents the degree of the nth node in G. =( ) represents the set of node betweennesses. Let G be the betweenness number of the nth node in G; ) represents the set of node proximity scores. T represents the proximity of the nth node in G; the value of T is generally a constant.

[0013] Compared to the original method's traffic network approach, this approach emphasizes the crucial role of links within the traffic system. It enhances the clarity of the topological relationships between different links. The traffic network model captures both the static and dynamic characteristics of road links. The structure of the traffic network reflects the connectivity between different links, depicting the static characteristics of each road link. Simultaneously, traffic flow, average speed, length, and other metrics for each road segment reflect the dynamic demand characteristics of the traffic network.

[0014] S200: Construct a critical road identification model M, which includes a Path2vec layer model and a GraphSAGE layer model. The Path2vec layer model improves DeepWalk by utilizing traffic flow, transforming DeepWalk's completely random walk function into a walk function that conforms to traffic flow. The DeepWalk and GraphSAGE layer models are existing technologies.

[0015] S300: Input G into M:

[0016] G outputs the path embedding feature vector through the Path2vec layer model. G obtains the node embedding feature vectors through the output of the GraphSAGE layer model. , Represents a node The neighborhood;

[0017] S400: Obtained from splicing S300 and Then, through a linear transformation, we obtain... In the The final embedded feature vector of the layer The specific formula is as follows:

[0018]

[0019] in, Indicates a connection operation. Let u represent any node. Indicates v in the first place The embedded feature vector of the layer, Indicates u in the first place The embedded feature vector of the layer, This represents the embedding vector obtained by passing v through the Path2vec layer model. It is a non-linear activation function. It is the first The learnable weight matrix of the layer;

[0020] S500: Set the clustering algorithm parameter λ, and use the clustering algorithm to... Clustering is performed to obtain λ-class traffic segments;

[0021] S600: Calculate the average traffic demand index of all nodes in each type of traffic segment obtained from S500. The calculation formula is as follows:

[0022]

[0023]

[0024] in, h is The total number of nodes in a traffic segment This indicates the length of the actual road segment corresponding to the h-th node. This represents the traffic flow passing through the h-th node. This represents the average speed of the vehicle as it passes the h-th node.

[0025] S700: Select the category with the highest average traffic demand index as the key category, sort the traffic demand index of all nodes in the key category in descending order, select the actual road segments corresponding to the traffic demand index of the top W nodes as key road segments, output the W key road segments, and obtain the final list of key roads.

[0026] S800: Select the traffic database of the region Y to be tested, and repeat S100-S700 to obtain the list of key roads in Y.

[0027] Preferably, in S300, G is obtained by outputting the path embedding feature vector through the Path2vec layer model. The specific steps are as follows:

[0028] S310: Calculate the adjacency relationship between i and j. The specific expression is as follows:

[0029]

[0030] The non-zero elements in the adjacency matrix A represent the connections between road segments. Only adjacent road segments can have a transition probability; otherwise, the probability is 0.

[0031] S320: Obtain the adjacency relationship between i and G. Traffic flow of J ;

[0032] S330: Calculate the transition weights from i to j. Then, normalization is performed, and the specific expression is as follows:

[0033]

[0034]

[0035] in, In graph G The distance between them;

[0036] S340: Calculate the transition matrix from i to j in the next step. The calculation formula is as follows:

[0037]

[0038] in, It is the set of adjacent nodes of i. Let k represent the k-th adjacent node of i. This represents the traffic flow passing through k;

[0039] S350: Calculate the transition probability of i transitioning to j. The calculation formula is as follows:

[0040]

[0041] S360: Obtain the walk sequence of node i. The specific expression is as follows:

[0042]

[0043] in, Indicated by The walk sequence starting at node t, where m represents the length of the walk sequence, t=1,2, m

[0044] S370: Calculation The specific expression is as follows:

[0045]

[0046] in, The embedding vector of node v. This represents an existing word vector training model imported from a Python library.

[0047] Preferably, in S300, G is obtained by outputting the node embedding feature vector through the GraphSAGE layer model. The calculation formula is as follows:

[0048]

[0049]

[0050] in, This represents the minimum distance between node v and node u. Indicates the number of sampling layers. Indicates aggregation operation, This indicates that node u is at the 1st rank. -1 layer embedded feature vectors.

[0051] Preferably, in S500, the clustering algorithm is the kmene algorithm, with λ=3. The importance of road connections is generally divided into three categories: critical, normal, and unimportant. The clustering algorithm is an existing model, and the number of categories required by the output model can be set according to the parameters.

[0052] Compared with the prior art, the present invention has at least the following advantages:

[0053] 1. A line graph mapping method based on traffic networks is used, where road connectors are represented as nodes and the connections between them are represented as edges. Dynamic attributes such as traffic flow and speed, as well as static attributes such as length and degree of traffic segments, are integrated into the node attributes of the traffic network, which expands the data collection scope and improves the accuracy of the final identification.

[0054] 2. To address the problem that graph embedding algorithms based on graph random walks like Node2vec cannot effectively encode the dynamic transitivity sequence features of traffic flow in traffic network node representation due to the physical conflict between the hyperparameter randomness and the spatiotemporal heterogeneity of traffic flow, this invention innovatively proposes a dynamic traffic flow direction walking model, Path2Vec. This model constructs a flow transfer matrix and obtains walking path sequence features that match historical traffic flow transfer probabilities through walking path sampling.

[0055] 3. To address the problem that traditional GraphSAGE in traffic network representation suffers from single feature extraction and failure to effectively integrate dynamic and static attributes of road network nodes, this invention innovatively combines Path2vec with GraphSAGE. Through a graph sampling aggregation model, it constructs multi-dimensional regional dynamic and static features of nodes to obtain key road segments.

[0056] 4. From the perspective of dynamic and static characteristics of the traffic network, the accuracy of selected road segments is evaluated based on node network efficiency, network resilience, and dynamic contagion indicators. Experimental data from a regional network and PEMS08 show that the method proposed in this invention has higher accuracy compared with traditional indicator evaluation methods. Data analysis shows that compared with traditional indicator selection methods based on flow, degree, betweenness, and proximity, the model proposed in this invention significantly outperforms traditional methods in obtaining key road segments in network efficiency indicators. It also performs better in terms of subgraph quantity and node quantity ratio indicators, and has a significant advantage over other methods in network efficiency indicators. This indicates that the model fully captures the static regional characteristics of the road network and the dynamic transitivity sequence characteristics of traffic flow, and has a greater impact on the efficiency of information exchange in the network and the integrity of the road network, thus enhancing the influence on the dynamic and static indicators of the network. Attached Figure Description

[0057] Figure 1 This is a diagram of the hybrid model architecture of the present invention;

[0058] Figure 2 This is a diagram of the Path2vec model architecture.

[0059] Figure 3 This is a diagram of the GraphSAGE model architecture.

[0060] Figure 4 The process of constructing a transportation network map;

[0061] Figure 5 This is a map showing the distribution of original detection equipment in the surrounding area of ​​a certain region.

[0062] Figure 6 A map showing the transportation network around a certain area;

[0063] Figure 7 This describes the network efficiency performance in a certain region during the morning rush hour from 8:00 to 9:00.

[0064] Figure 8 This study simulates the number of infected nodes in a network of an infectious disease model during the morning rush hour (8:00-9:00) in a certain region.

[0065] Figure 9 This section compares the performance of sub-map quantity indicators during the morning rush hour (8:00-9:00) in a certain region.

[0066] Figure 10 This section compares the performance of the ratio of the number of nodes in the largest connected component of the network to the number of nodes in the original graph after deleting nodes during the morning rush hour (8:00-9:00) in a certain region.

[0067] Figure 11This is a comparison chart showing the locations of the top 15 key road segments in a certain region during the morning rush hour from 8:00 to 9:00, obtained by the model of this invention and by traffic flow model alone. Detailed Implementation

[0068] The present invention will now be described in further detail.

[0069] Identifying key road segments in a road network and using them as the core for regional traffic control and optimization plays a crucial role in traffic management. This invention proposes an improved version of the traditional biased random walk model Path2vec, which obtains path feature embeddings that conform to traffic flow direction. It then aggregates graph embedding vectors using a graph sampling and GraphSAGE key road segment hybrid model. Finally, it clusters these embedding vectors to identify key road segments in the network.

[0070] See Figures 1-11 A dynamic traffic flow direction walking method for identifying traffic-critical roads includes the following steps:

[0071] Select the traffic database X for a certain region:

[0072] S100: The networkx and pandas libraries in Python are used to process X to obtain the traffic data information table corresponding to X. After the traffic data information table is converted into a matrix, the traffic network line diagram G is obtained.

[0073] The expression for the traffic network line diagram G is as follows:

[0074]

[0075] Where V=( ) is a set of nodes. Let G be the nth node. This represents the total number of nodes in G. Let G represent the set of edges. In G and One edge between them, ; () represents the set of average velocities of each node in G within a specific time T. Let G represent the average speed of a vehicle passing through the nth node; () represents the set of traffic flow at each node in G within a specific time T. Let G represent the traffic flow passing through the nth node; ) represents the set of actual road segment lengths corresponding to nodes. This represents the length of the actual road segment corresponding to the nth node in G; =( ) represents the set of node traffic densities. Let G represent the traffic density passing through the nth node; =( ) represents the set of node degrees. This represents the degree of the nth node in G. =( ) represents the set of node betweennesses. Let G be the betweenness number of the nth node in G; ) represents the set of node proximity scores. This represents the proximity of the nth node in G;

[0076] S200: Construct a critical road identification model M, which includes a Path2vec layer model and a GraphSAGE layer model; the Path2vec layer model improves DeepWalk by utilizing traffic flow, transforming DeepWalk's completely random walk function into a walk function that conforms to traffic flow.

[0077] S300: Input G into M:

[0078] G outputs the path embedding feature vector through the Path2vec layer model. G obtains the node embedding feature vectors through the output of the GraphSAGE layer model. , Represents a node The neighborhood;

[0079] In S300, G obtains the path embedding feature vector through the Path2vec layer model output. The specific steps are as follows:

[0080] S310: Calculate the adjacency relationship between i and j. The specific expression is as follows:

[0081]

[0082] S320: Obtain the adjacency relationship between i and G. Traffic flow of J ;

[0083] S330: Calculate the transition weights from i to j. Then, normalization is performed, and the specific expression is as follows:

[0084]

[0085]

[0086] in, In graph G The distance between them;

[0087] S340: Calculate the transition matrix from i to j in the next step. The calculation formula is as follows:

[0088]

[0089] in, It is the set of adjacent nodes of i. Let k represent the k-th adjacent node of i. This represents the traffic flow passing through k;

[0090] S350: Calculate the transition probability of i transitioning to j. The calculation formula is as follows:

[0091]

[0092] S360: Obtain the walk sequence of node i. The specific expression is as follows:

[0093]

[0094] in, Indicated by The walk sequence starting at node t, where m represents the length of the walk sequence, t=1,2, m

[0095] S370: Calculation The specific expression is as follows:

[0096]

[0097] in, The embedding vector of node v. This represents an existing word vector training model imported from a Python library.

[0098] In S300, G obtains node embedding feature vectors through the output of the GraphSAGE layer model. The calculation formula is as follows:

[0099]

[0100]

[0101] in, This represents the minimum distance between node v and node u. Indicates the number of sampling layers. Indicates aggregation operation, This indicates that node u is at the 1st rank. -1 layer embedded feature vectors.

[0102] S400: Obtained from splicing S300 and Then, through a linear transformation, we obtain... In the The final embedded feature vector of the layer The specific formula is as follows:

[0103]

[0104] in, Indicates a connection operation. Let u represent any node. Indicates v in the first place The embedded feature vector of the layer, Indicates u in the first place The embedded feature vector of the layer, This represents the embedding vector obtained by passing v through the Path2vec layer model. It is a non-linear activation function. It is the first The learnable weight matrix of the layer;

[0105] S500: Set the clustering algorithm parameter λ, and use the clustering algorithm to... Clustering is performed to obtain λ-class traffic segments;

[0106] In the S500, the clustering algorithm is the kmeans algorithm, and λ=3 is set.

[0107] S600: Calculate the average traffic demand index of all nodes in each type of traffic segment obtained from S500. The calculation formula is as follows:

[0108]

[0109]

[0110] in, h is The total number of nodes in a traffic segment This indicates the length of the actual road segment corresponding to the h-th node. This represents the traffic flow passing through the h-th node. This represents the average speed of the vehicle as it passes the h-th node.

[0111] S700: Select the category with the highest average traffic demand index as the key category, sort the traffic demand index of all nodes in the key category in descending order, select the actual road segments corresponding to the traffic demand index of the top W nodes as key road segments, output the W key road segments, and obtain the final list of key roads.

[0112] S800: Select the traffic database of the region Y to be tested, and repeat S100-S700 to obtain the list of key roads in Y.

[0113] Experiments and Results

[0114] Data Description

[0115] This study used a private dataset from a specific region: this dataset contains raw traffic data acquired by external road detection devices in the surrounding road network from July 1st to July 31st, 2024. After data cleaning and processing, traffic flow data, average road speed data, and road length data for each external road detection device during the morning peak (8:00-9:00), afternoon peak (12:00-13:00), and evening peak (17:00-18:00) were obtained. Figure 5 and Figure 6 This document displays the distribution of original detection devices in the surrounding area of ​​a certain region, and establishes a traffic network map based on the connectivity of these devices. PEMS08: This is the California Highway Network PeMS traffic flow dataset. PEMS08 consists of data collected from 170 detectors every 5 minutes for a total of 62 days, identical to a private dataset for a certain region. Traffic data during peak hours is extracted from this dataset.

[0116] Baseline Description

[0117] This invention compares the key road segments obtained by the proposed key road segment fusion model based on pathtoVec and graph sampling aggregation Graphsage with traditional key road segment selection methods based on traffic flow, centrality indicators, and graph neural networks. The accuracy of the selected key road segments is evaluated from the perspectives of node network efficiency, network resilience, and dynamic contagion indicators.

[0118] Key segment selection methods based on centrality indicators include node degree selection, node betweenness centrality selection, and proximity centrality selection. The degree of a node describes the number of edges connected to it, and the node degree formula is shown below:

[0119]

[0120] in, Indicates node degree. This represents the number of edges connected to node i.

[0121] Node betweenness centrality refers to the number of shortest paths passing through nodes in a network. The formula for node betweenness centrality is as follows:

[0122]

[0123] in, This represents the number of paths that pass through node i and are the shortest paths. This represents the number of shortest paths between s and t;

[0124] Node proximity reflects the average shortest path length from a specific node to any other point in the network. The formula for node proximity centrality is as follows:

[0125]

[0126] in, This represents the average distance from node i to other points. This indicates the proximity centrality of nodes;

[0127] The following baseline methods were used to compare the data:

[0128] Onlyflow ranking: Traffic flow is an important data point in traffic flow. Traffic segments with high traffic flow often have a greater impact on the traffic network. This method selects the top K traffic segments in the network based on their traffic flow.

[0129] OnlyDegree: Node degree is one of the indicators of network node centrality. Degree reflects the number of connections with other nodes in the network. The higher the degree of a node, the more edges are connected to that node. When a node fails, the number of nodes that may be affected will be greater. This method selects the top K traffic segments in the network by degree.

[0130] Only-Betweenness Ranking: The betweenness of a node is one of the indicators of node centrality in a network. Nodes with high betweenness are often bridges in the network, connecting different communities or groups, such as information dissemination intermediaries in social networks. This method selects the top K traffic segments in terms of the number of betweennesses in the network.

[0131] OnlyCloseness Ranking: Node proximity is one of the indicators of network node centrality, representing the reciprocal of the average shortest distance from a node to all other nodes. It reflects the efficiency of information propagation among nodes; nodes with high proximity can quickly reach other nodes. This method selects the top K traffic segments in the network based on their proximity.

[0132] Path2vec: In order to reduce the uncertainty caused by the fixed parameters p and q in the traditional node2vec model and to make the random walk process more consistent with the path tendency in the road traffic network, the Path2Vec Embedding model builds a transition matrix that reflects the traffic directionality in the network based on the node2vec model. It represents the path of traffic vehicles as an embedding vector, obtains important road segments through clustering, and selects the top K traffic segments.

[0133] GraphSAGE is an improvement on graph convolutional networks. It integrates the feature attributes of nodes into a feature matrix, samples a certain number of neighboring nodes, obtains the node embedding vector through an aggregation function, and obtains important road segments through clustering. From these, it selects the top K traffic segments.

[0134] Experimental results

[0135] Onlyflow, OnlyDegree, OnlyBetweenness, OnlyCloseness, Path2vec, and GraphSAGE were trained on a private dataset and the PEMS08 dataset in a certain region and compared with the model proposed in this invention. Experiments simulated key road segments obtained from the seven models more than 100 times, calculating the overall network efficiency index, network resilience index, number of subgraphs remaining in the network after node deletion, ratio of the number of nodes in the largest connected component of the network after node deletion to the number of nodes in the original graph, and the number of infected nodes in the network in a simulated infectious disease model. The overall network efficiency index is selected as the ratio of the network efficiency of the largest connected subgraph to the network efficiency of the original network after node deletion. The lower the value of this index, the greater the impact on the overall network communication efficiency when the node list in the network becomes invalid. The number of subgraphs remaining in the network after node deletion measures that when nodes are removed from the node list, the network may be divided into multiple subgraphs, leading to a significant decrease in communication efficiency. The more subgraphs remaining in the network after node deletion, the stronger the node's ability to partition the network, and the more important the node. The ratio of the number of nodes in the largest connected component of the network after node deletion to the number of nodes in the original graph is used. A low ratio indicates that the propagation of traffic flow will be significantly suppressed after the key node is deleted. Furthermore, a smaller ratio indicates a greater impact on the connectivity of the remaining network. In the simulated infectious disease model, the number of infected nodes in the network reflects the impact of congestion at a key node on the spread of congestion to surrounding roads from a dynamic propagation perspective. The selected key node is simulated as a congested node spreading congestion to its surroundings for a certain time step. A higher number of congested nodes indicates a stronger initial node's ability to spread congestion and a greater impact on the network. In this invention, the infection rate is set to 0.8, the recovery rate to 0.2, and the time step to 15, simulating a propagation scenario with a high infection rate and a low recovery rate. Figures 7 to 11This paper demonstrates how, in a dataset covering morning, noon, and evening peak hours, key road segments were identified using onlyflow, OnlyDegree, OnlyBetweenness, OnlyCloseness, Path2vec, GraphSAGE, and the proposed aggregation model. The network efficiency after key road segment failure was calculated, along with the number of subgraphs remaining in the network after node deletion, the ratio of the number of nodes in the largest connected component to the original number of nodes, and the number of infected nodes in the network in a simulated infectious disease model. The average values ​​for these road segments in each result and the total values ​​after 100 experiments were obtained. The traffic network map based on the local area contains 134 road segment nodes, while the traffic network map based on the PEMS08 dataset contains 277 road segment nodes. In actual traffic control and planning, due to cost considerations, too many nodes are not selected as key nodes. Therefore, K is set to 15 in the local dataset model and 30 in the PEMS08 dataset model as appropriate numbers of key road segments.

[0136] Table 1 shows the performance metrics of each model on the PEMS08 dataset. The Path2vec and GraphSAGE fusion model proposed in this invention has the highest number of infected nodes in the network and the number of subgraphs in the network after node deletion in the simulated infectious disease model. This is followed by the key road segments obtained by the Path2vec model. Compared with key road segments selected based on flow, degree, and betweenness indices, it shows a significant lead. This indicates that on the PEMS08 dataset, the method proposed in this invention is superior to traditional methods in terms of dynamic propagation and network segmentation. The Path2vec and GraphSAGE fusion model proposed in this invention has the lowest network efficiency and the ratio of the number of nodes in the largest connected component of the network after node deletion to the number of nodes in the original graph. This indicates that the nodes found by the method proposed in this invention simultaneously carry the functions of structural hubs and dynamic transmission cores, and have a greater impact on network communication and connectivity.

[0137] Table 1. Model metrics for the PEMS08 dataset

[0138]

[0139] Figure 7The data shows the network efficiency performance in a certain region during the morning rush hour from 8:00 to 9:00. The proposed Path2vec and GraphSAGE fusion model achieved a total network efficiency score of 58.9 in 100 evaluations. This is significantly higher than OnlySAGE's 71.6, Path2vec's 72.9, and even surpasses the average network efficiency scores of flow-based and centrality-based selection methods (both exceeding 80). The proposed method also identifies critical road segments that act as bottlenecks in cross-regional traffic transfer, demonstrating greater global influence.

[0140] Figure 8 This data shows the number of infected nodes in a simulated infectious disease model during the morning rush hour (8:00-9:00) in a certain region. The proposed Path2vec and GraphSAGE fusion model, after 100 simulations, yielded a total infected node count of 917. This is compared to OnlySAGE's 840, Path2vec's 820, OnlyCloseness's 829, and other selection methods with total values ​​less than 800. This demonstrates that the proposed method more stably predicts key nodes with higher infectivity across multiple simulations, leading to wider spread and more sustained effectiveness in predicting transmission dynamics.

[0141] Figure 9 The results show the performance of subgraph quantity index in a certain region during the morning rush hour from 8:00 to 9:00. The proposed Path2vec and GraphSAGE fusion model achieved a total subgraph quantity index of 359.4 over 100 iterations, with OnlySAGE achieving 355.7, close to the proposed method. Other methods did not exceed 300, indicating that OnlySAGE's ability to aggregate neighbor information, capture local and global features of road segments, and obtain the topological attributes of road segments in the graph already possesses good performance and has significant advantages over traditional methods. While the proposed Path2vec fusion model and the OnlySAGE model share some similarities in mechanism, leading to similar results, there is still a certain improvement.

[0142] Figure 10 This data shows the performance of the ratio of the number of nodes in the largest connected component of the network to the number of nodes in the original graph after deleting nodes during the morning rush hour (8:00-9:00) in a certain region. The node count ratio of the Path2vec and GraphSAGE fusion model proposed in this invention is superior to the other six methods, indicating that the proposed method simultaneously considers local connectivity patterns and global path dependencies, and more comprehensively evaluates the function of nodes in complex traffic networks. The network cannot maintain high integrity after deleting key road segments obtained by this invention, and the nodes obtained by the fusion model cause greater disruption to network connectivity.

[0143] Figure 11 The graph shows the locations of the top 15 critical road segments obtained by the Path2vec and GraphSAGE fusion model proposed in this invention during the morning rush hour (8:00-9:00) in a certain region, as well as the locations of the top 15 critical road segments obtained solely through traffic flow analysis. These segments are depicted in red on the graph. In the actual road network, critical road segments may be distributed along main roads connecting different areas, bridges, tunnels, or connecting roads around transportation hubs. The results show that traffic flow-based critical road segments only include high-traffic main roads. However, the critical road segments obtained by the method proposed in this invention not only include main roads but also important weaving areas, which better matches frequently congested traffic segments and aligns more closely with empirically observed traffic segment locations.

[0144] Table 2. Model Indicators for a Certain Region's Dataset

[0145]

[0146] Table 2 shows the average total of the four indicators mentioned above during the morning, noon, and evening peak hours in a certain region after 100 simulation experiments. Data analysis shows that, compared with traditional methods based on flow rate, degree, betweenness, and proximity, the graph neural network-based methods Path2vec, GraphSAGE, and the aggregation method proposed in this invention significantly outperform traditional methods in obtaining key road segments for all four indicators. This indicates that graph neural network-based methods can fully acquire the connectivity and topological information between road segments in traffic networks, possessing stronger structure recognition capabilities and greater ability to mitigate traffic congestion and damage network structures. The key road segments obtained after incorporating the GraphSAGE module show a significant advantage over traditional indicator-based methods on the dataset of this region, having a substantial impact on the efficiency of traffic flow propagation. The key road segment acquisition method of the fusion model proposed in this invention is superior to the single GraphSAGE and Path2vec methods in terms of network efficiency, number of SIR models, number of subgraphs, and ratio of nodes mentioned above. This indicates that the fusion model of GraphSAGE and Path2vec has a greater impact on the efficiency of information exchange in the network, a stronger ability to disrupt the network, a stronger ability to spread congestion, and obtains more critical nodes. It can also more fully integrate road segment attributes with the traffic flow direction in the actual traffic network.

[0147] From the perspective of dynamic and static characteristics of traffic networks, this invention evaluates the accuracy of road segment selection based on node network efficiency, network resilience, and dynamic contagion indicators. Experimental data from a regional network and PEMS08 show that the proposed method has higher accuracy, better acquisition of traffic flow characteristic information, stronger structure recognition capabilities, and a greater impact on the overall network communication efficiency when the node list in the network fails. The node's ability to segment the network is also stronger than traditional methods, resulting in a greater impact on network connectivity and a stronger ability to spread congestion.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A dynamic traffic flow direction walking method for identifying critical traffic roads, characterized in that: Includes the following steps: Select the traffic database X for a certain region; S100: The networkx and pandas libraries in Python are used to process X to obtain the traffic data information table corresponding to X. After the traffic data information table is converted into a matrix, the traffic network line graph G is obtained. The expression for the traffic network line diagram G is as follows: Where V=( ) is a set of nodes. Let G be the nth node. This represents the total number of nodes in G. Let G represent the set of edges. In G and One edge between them, ; () represents the set of average velocities of each node in G within time T. Let G represent the average speed of a vehicle passing through the nth node; () represents the set of traffic flows at each node in G within time T. Let G represent the traffic flow passing through the nth node; ) represents the set of actual road segment lengths corresponding to nodes. This represents the length of the actual road segment corresponding to the nth node in G; =( ) represents the set of node traffic densities. Let G represent the traffic density passing through the nth node; =( ) represents the set of node degrees. This represents the degree of the nth node in G. =( ) represents the set of node betweennesses. Let G be the betweenness number of the nth node in G; ) represents the set of node proximity. This represents the proximity of the nth node in G; S200: Construct a critical road identification model M, which includes a Path2vec layer model and a GraphSAGE layer model; the Path2vec layer model improves DeepWalk by utilizing traffic flow, transforming DeepWalk's completely random walk function into a walk function that conforms to traffic flow. S300: Input G into M: G outputs the path embedding feature vector through the Path2vec layer model. G obtains the node embedding feature vectors through the output of the GraphSAGE layer model. , Represents a node The neighborhood; S400: Obtained from splicing S300 and Then, through a linear transformation, we obtain... In the The final embedded feature vector of the layer The specific formula is as follows: in, Indicates a connection operation. Let u represent any node. Indicates v in the first place The embedded feature vector of the layer, Indicates u in the first place The embedded feature vector of the layer, This represents the embedded feature vector obtained by passing v through the Path2vec layer model. It is a non-linear activation function. It is the first The learnable weight matrix of the layer; S500: Set the clustering algorithm parameter λ, and use the clustering algorithm to... Clustering is performed to obtain λ-class traffic segments; S600: Calculate the average traffic demand index of all nodes in each type of traffic segment obtained from S500. The calculation formula is as follows: in, h is The total number of nodes in a traffic segment This indicates the length of the actual road segment corresponding to the h-th node. This represents the traffic flow passing through the h-th node. This represents the average speed of the vehicle as it passes the h-th node. S700: Select the category with the highest average traffic demand index as the key category, sort the traffic demand index of all nodes in the key category in descending order, select the actual road segments corresponding to the traffic demand index of the top W nodes as key road segments, output the W key road segments, and obtain the final list of key roads. S800: Select the traffic database of the region Y to be tested, and repeat S100-S700 to obtain the list of key roads in Y.

2. The dynamic traffic flow direction walking method for identifying key traffic roads as described in claim 1, characterized in that: In S300, G obtains the path embedding feature vector through the Path2vec layer model output. The specific steps are as follows: S310: Calculate the adjacency relationship between i and j. The specific expression is as follows: S320: Obtain the adjacency relationship between i and G. Traffic flow of J ; S330: Calculate the transition weights from i to j. Then, normalization is performed, and the specific expression is as follows: in, In graph G The distance between them; S340: Calculate the transition matrix from i to j in the next step. The calculation formula is as follows: in, It is the set of adjacent nodes of i. Let k represent the k-th adjacent node of i. This represents the traffic flow passing through k; S350: Calculate the transition probability of i transitioning to j. The calculation formula is as follows: S360: Obtain the walk sequence of node i. The specific expression is as follows: in, Indicated by The walk sequence starting at node t, where m represents the length of the walk sequence, t=1,2, m S370: Calculation The specific expression is as follows: in, The embedding vector of node v. This represents an existing word vector training model imported from a Python library.

3. The dynamic traffic flow direction walking method for identifying key traffic roads as described in claim 2, characterized in that: In S300, G obtains node embedding feature vectors through the output of the GraphSAGE layer model. The calculation formula is as follows: in, This represents the minimum distance between node v and node u. Indicates the number of sampling layers. Indicates aggregation operation, Indicates that node u is at the th -1 layer embedded feature vectors.

4. The dynamic traffic flow direction walking method for identifying key traffic roads as described in claim 3, characterized in that: In the S500, the clustering algorithm is the kmeans algorithm, and λ=3 is set.

Images