A traffic congestion propagation detection method based on space-time adaptive local search

Abstract

The application relates to a traffic congestion propagation detection method based on space-time adaptive local search, which comprises the following steps: calculating a traffic congestion index of each time slice based on matching data; constructing a space-time congestion subgraph at different time; calculating graph structure information and semantic information of a road section in the space-time congestion graph; obtaining a similarity feature matrix of the space-time congestion graph at different time; calculating a space-time adaptive adjacency matrix of each space-time congestion graph; constructing a topological directed graph and determining the nearest local leader to stop searching; screening multi-scale community centers based on the local leader to identify traffic congestion propagation bottlenecks; and calculating the number of traffic congestion propagation bottlenecks and its rules between different types of dates and different types of time periods. Compared with the prior art, the application has the advantages of capturing the space-time relationship driving the dynamic change of a road network at different time to accurately identify the shift rule of traffic congestion propagation bottlenecks, and adapting to data with different time resolutions to reduce errors.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of road traffic information monitoring and road network congestion analysis, and particularly relates to a traffic congestion propagation detection method based on space-time adaptive local search. BACKGROUND

[0002] Traffic congestion is an inevitable result of rapid modernization, causing a serious imbalance between traffic supply and travel demand. The cascading reaction of traffic congestion propagation can trigger secondary and tertiary congestion, because whether it is regular or irregular congestion, the congestion bottleneck often continues or shifts in time and space. If not intervened in time and effectively, it can even lead to the paralysis of the entire road network, thus bringing great challenges to urban traffic management and public travel. In view of this, it is of great practical significance to detect the traffic congestion bottleneck in the urban road network and to mine its propagation law.

[0003] Traditional traffic congestion propagation detection methods are mainly based on simulation methods, including micro and macro levels. Micro-simulation is represented by car-following model and lane-changing model, taking vehicles as the research object, and transforming the congestion propagation process into a queuing problem. Macro-simulation represents the propagation process of traffic flow by capturing the overall behavior of traffic flow, and the representative methods are cellular transmission model and susceptible-infected-removed model. However, these two types of simulation methods have the following three limitations: first, they rely on certain traffic flow theory assumptions and potential models, and lack real-time data support; second, there are too many model parameters and weak generalization ability, and a large number of model parameter modifications may cause distortion problems; third, it is difficult to extend to large-scale networks or complex scenarios, thus limiting its application range. In recent years, with the development of big data and artificial intelligence technology, data-driven methods have gradually become the mainstream technology for traffic congestion propagation analysis, mainly including dynamic Bayesian network and deep learning model. The dynamic Bayesian-based method discretizes continuous historical data into multiple traffic states, calculates the state transition probability between two adjacent road segments, and infers the traffic congestion propagation pattern, but data discretization and prior knowledge affect the effectiveness of reasoning. Deep learning models often directly regard traffic networks as graph structures, and regard road segments as nodes in the graph to construct feature matrices. The limitation of such methods is that they often rely on pre-defined road network adjacency matrices, which are difficult to accurately describe the information dynamic transmission process of congestion propagation, because the pre-defined adjacency matrix is often fixed.

[0004] In summary, although data-driven approaches overcome the limitations of simulation, they still face two key challenges: First, how to design dynamically changing adjacency matrices to capture the spatiotemporal relationships driving road network dynamics at different times, replacing traditional predefined adjacency matrices. Second, how to integrate road network congestion features to adapt to data with different time resolutions and eliminate potential errors caused by time granularity. Summary of the Invention

[0005] The purpose of this invention is to capture the spatiotemporal relationships that drive the dynamic changes of the road network at different times in order to accurately describe the dynamic information transmission process of congestion propagation, and to reduce errors by adapting to data with different time resolutions. This invention provides a traffic congestion propagation detection method based on spatiotemporal adaptive local search.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A traffic congestion propagation detection method based on spatiotemporal adaptive local search, characterized by the following steps:

[0008] S1. Obtain road network and floating car trajectory data and match them to obtain matching data;

[0009] S2. Calculate the traffic congestion index for each time slice based on the matching data;

[0010] S3. Construct a spatiotemporal congestion sub-graph for different times based on the traffic congestion index. The nodes of the spatiotemporal congestion sub-graph are congested road segments.

[0011] S4. Calculate the graph structure information and semantic information of the middle road segment of the spatiotemporal congestion graph. The graph structure information includes node curvature, node degree and spatial proximity. The set of spatiotemporal congestion subgraphs constitutes the spatiotemporal congestion graph.

[0012] S5. Based on graph structure information and semantic information, cosine similarity is used to obtain the curvature similarity feature matrix, degree similarity feature matrix, spatial distance matrix and semantic similarity matrix of the spatiotemporal congestion graph;

[0013] S6. The curvature similarity feature matrix, degree similarity feature matrix, spatial distance matrix and semantic similarity matrix are fused using the entropy weight method to obtain the spatiotemporal adaptive adjacency matrix of the spatiotemporal congestion graph.

[0014] S7. Based on the spatiotemporal adaptive adjacency matrix and combined with the topological structure of the spatiotemporal congestion graph, construct a topological directed graph;

[0015] S8. Based on the directed graph, the local breathing first search algorithm is used to identify the local leaders of the spatiotemporal congestion graph. Then, the local breadth-first search algorithm is used to find the nearest local leader of each local leader. Once the nearest local leader is determined, the search stops.

[0016] S9. Based on local leaders, screen multi-scale community centers to identify bottlenecks in the spread of traffic congestion;

[0017] S10. Merge the selected multi-scale community centers according to date type, and calculate the number and pattern of traffic congestion propagation bottlenecks between different date types.

[0018] S11. Merge the selected multi-scale community centers according to date type, and calculate the number and pattern of traffic congestion propagation bottlenecks between different time periods.

[0019] Furthermore, the traffic congestion index is:

[0020]

[0021] in, Representative road segment f i free flow velocity, Represents the timestamp t j The true speed.

[0022] Furthermore, the node curvature is the absolute value of the ratio of the tangent turning angle Δα to the arc length Δl of the road segment; the degree of a node is defined as the total number of edges directly connected to that node; and spatial proximity is Euclidean distance.

[0023] Furthermore, the traffic congestion index is:

[0024]

[0025] Among them, the spatiotemporal congestion subgraph Defined as a graph structure formed by connecting all congested instances as nodes within a given time period, where These represent the nodes, edges, edge weights, and edge weights of the graph, respectively. It is determined by the fast Fourier transform of node curvature, node degree, spatial proximity and traffic congestion index.

[0026] Furthermore, the spatiotemporal adaptive adjacency matrix is:

[0027]

[0028] in, Represents timestamp t j Spatiotemporal adaptive adjacency matrix of spatiotemporal congested subgraph, curvature similarity of spatiotemporal congested subgraph similarity matrix Spatial distance matrix Similarity to the Fourier transform of the traffic congestion index and These represent the weight values ​​corresponding to each similarity matrix obtained using the entropy weighting method.

[0029] Furthermore, based on the spatiotemporal adaptive adjacency matrix and combined with the topological structure of the spatiotemporal congestion graph, the specific steps for constructing a directed graph are as follows:

[0030] Based on the spatiotemporal adaptive adjacency matrix and combined with the topological structure of the spatiotemporal congestion graph, the nodes f are... i The sum of the weights of all directly adjacent edges is assigned to f. i And the sum of the allocated weights is denoted as node f. i The eigenvalues ​​are denoted as Form a directed graph where node f is defined if the following two conditions are met. i Point to its largest neighbor f k :

[0031] First, the node has a higher value than its neighbors, that is...

[0032] Second, each node f i They all point to their maximum neighbor, that is Where V(f) i ) represents f i The set of all its neighbors.

[0033] Furthermore, the local leader in the spatiotemporal congestion map is specifically identified as follows:

[0034] A node with in-degree but no out-degree is considered a local leader that dominates its surrounding area.

[0035] Furthermore, the most recent local leader is identified as follows:

[0036] For each local leader If satisfied but as a local leader The most recent local leader.

[0037] Furthermore, the different types of dates include weekdays, weekends, and public holidays.

[0038] Furthermore, different time periods include morning peak hours, evening peak hours, and off-peak hours.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] (1) This invention constructs a dynamic adaptive adjacency matrix based on multi-feature fusion, which can identify congestion bottlenecks from large-scale traffic congestion network datasets with different time resolutions. From the perspective of technological innovation, ablation experiments based on different feature combinations show that fusing graph structural information and semantic information is of great significance for traffic congestion analysis. Specifically, the fusion of Fourier transforms of the curvature, degree, spatial proximity, and traffic congestion index of nodes in the road network can improve the performance of multi-scale community structure identification in traffic networks.

[0041] (2) By analyzing detected multi-scale communities, this invention can effectively uncover traffic congestion propagation patterns on different types of dates (weekdays, weekends, and holidays) and different time periods (morning peak hours, evening peak hours, and off-peak hours) on the same type of date. Based on this, the periodicity of traffic congestion propagation patterns, their sensitivity to time periods, and the impact of human travel activities on congestion propagation patterns can be revealed. This can eliminate potential errors caused by time granularity, thereby providing targeted suggestions for urban traffic managers. It has strong feasibility and practicality. Attached Figure Description

[0042] Figure 1 This is a flowchart illustrating the traffic congestion propagation detection method based on spatiotemporal adaptive local search of the present invention.

[0043] Figure 2 The results of ablation experiments based on different combinations of features are shown in the figure.

[0044] Figure 3 A schematic diagram illustrating the bottlenecks and propagation patterns of traffic congestion on different types of dates (Experimental Area 1), in which... Figure 3 (a) represents the change in congestion transmission bottlenecks from weekdays to weekends. Figure 3 (b) represents the change in congestion transmission bottlenecks from weekdays to holidays. Figure 3 (c) represents the change in congestion transmission bottlenecks from holidays to weekdays.

[0045] Figure 4 The probability of congestion center shifting at different times of the week (Experimental Area 1);

[0046] Figure 5 The probability of congestion center shifting at different times of the weekend (Experimental Area 1);

[0047] Figure 6 The probability of congestion center shifting during different periods of holidays (Experimental Area 1);

[0048] Figure 7 A schematic diagram illustrating the bottlenecks and propagation patterns of traffic congestion on different types of dates (Experimental Area 2), in which... Figure 7(a) represents the change in congestion transmission bottlenecks from weekdays to weekends. Figure 7 (b) represents the change in congestion transmission bottlenecks from weekdays to holidays. Figure 7 (c) represents the change in congestion transmission bottlenecks from holidays to weekdays;

[0049] Figure 8 The probability of congestion center shifting at different times of the weekday (Experimental Area 2);

[0050] Figure 9 The probability of congestion center shifting at different times of the weekend (Experimental Area 2);

[0051] Figure 10 The probability of congestion center shifting at different times during holidays (Experimental Area 2). Detailed Implementation

[0052] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0053] This invention proposes a traffic congestion propagation detection method based on spatiotemporal adaptive local search, allowing for the detection of traffic congestion bottlenecks and congestion propagation patterns. The method involves: acquiring floating car trajectory data and matching it with the road network; extracting congested road segments based on the traffic congestion index and constructing spatiotemporal congestion subgraphs with different timestamps; designing a spatiotemporal adaptive adjacency matrix for the spatiotemporal congestion subgraph by fusing node curvature, degree, spatial proximity, and semantic information of the graph structure using the entropy weight method; performing a local search by combining the dynamically changing spatiotemporal adaptive adjacency matrix with the topological structure of the spatiotemporal congestion subgraph to detect multi-scale communities and extract traffic congestion bottlenecks; analyzing the traffic congestion propagation patterns based on the community's movement between different types of dates; and analyzing the interaction between traffic congestion and human travel based on the community's movement between different time periods on the same type of date. This invention conducts traffic congestion propagation detection from the perspective of multi-scale communities, allowing domain experts to conduct in-depth analysis of congestion propagation at multiple levels, demonstrating good practicality and feasibility. This invention provides a method for detecting traffic congestion bottlenecks and congestion propagation patterns based on spatiotemporal adaptive search, which is mainly used to solve the technical problem that current traffic congestion propagation analysis based on fixed predefined adjacency matrices cannot capture dynamically changing spatiotemporal relationships.

[0054] This invention includes the following steps:

[0055] S1. Obtain road network and floating car trajectory data and match them to obtain matching data;

[0056] S2. Calculate the traffic congestion index for each time slice based on the matching data;

[0057] S3. Construct a spatiotemporal congestion sub-graph for different times based on the traffic congestion index. The nodes of the spatiotemporal congestion sub-graph are congested road segments.

[0058] S4. Calculate the graph structure information and semantic information of the middle road segment of the spatiotemporal congestion graph. The graph structure information includes node curvature, node degree and spatial proximity. The set of spatiotemporal congestion subgraphs constitutes the spatiotemporal congestion graph.

[0059] S5. Based on graph structure information and semantic information, cosine similarity is used to obtain the curvature similarity feature matrix, degree similarity feature matrix, spatial distance matrix and semantic similarity matrix of the spatiotemporal congestion graph;

[0060] S6. The curvature similarity feature matrix, degree similarity feature matrix, spatial distance matrix and semantic similarity matrix are fused using the entropy weight method to obtain the spatiotemporal adaptive adjacency matrix of the spatiotemporal congestion graph.

[0061] S7. Construct a directed graph based on the spatiotemporal adaptive adjacency matrix and the topological structure of the spatiotemporal congestion graph;

[0062] S8. Based on the directed graph, the local breathing first search algorithm is used to identify the local leaders of the spatiotemporal congestion graph. Then, the local breadth-first search algorithm is used to find the nearest local leader of each local leader. Once the nearest local leader is determined, the search stops.

[0063] S9. Screening multi-scale community centers based on local leaders and analyzing the transfer patterns of community centers;

[0064] S10. Merge the selected multi-scale community centers according to date type, and calculate the number of community transfers and their patterns between different date types.

[0065] S11. Merge the selected multi-scale community centers according to date type, and calculate the number of community transfers and their patterns between different time periods.

[0066] The specific steps of this invention are as follows:

[0067] Step S1: Obtain road network and floating car trajectory data, and use big data analysis technology to match the timestamped floating car trajectory data with the road network;

[0068] Step S2: Select an appropriate traffic congestion rating index and calculate the traffic congestion index for each time slice based on the matching data from Step S1.

[0069] Step S3: Construct a spatiotemporal congestion sub-graph for different times based on the traffic congestion index calculated in Step S2. The nodes of the graph are congested road segments.

[0070] Step S4: Calculate the road segment structure information (curvature, degree, spatial distance) and semantic information (fast Fourier transform form of traffic congestion index) in the spatiotemporal congestion map;

[0071] Step S5: Based on the four features in Step S4, cosine similarity is used to obtain the curvature similarity feature matrix, degree similarity feature matrix, spatial distance matrix, and semantic similarity matrix of the spatiotemporal congestion map at different times;

[0072] Step S6: Use the entropy weight method to fuse the four feature similarity matrices from step S5 to obtain the spatiotemporal adaptive adjacency matrix of each spatiotemporal congestion map;

[0073] Step S7: Based on the spatiotemporal adaptive adjacency matrix fused in Step S6, and combined with the topology of the spatiotemporal congestion graph at each time step, construct a directed graph, where the direction of the edges between nodes is determined by the local dominant value of the nodes.

[0074] Step S8: Use the local breathing priority search algorithm to identify local leaders and construct a hierarchical tree based on the shortest path length between local leaders. Different levels of hierarchical trees represent communities of different scales.

[0075] Step S9: Filter the central nodes of communities at different scales, i.e., the bottlenecks that cause traffic congestion;

[0076] Step S10: Based on the detected communities for different types of dates (weekdays, weekends, and holidays), explore the propagation patterns of traffic congestion, especially its periodicity.

[0077] Step S11: Based on the detected communities in different time periods (morning peak hours, evening peak hours, and off-peak hours), explore the interaction between human travel activities and traffic congestion propagation patterns;

[0078] Step S12: Analyze and visualize the number of community transfers calculated in steps S11 and S12, and summarize the congestion propagation patterns and potential propagation modes.

[0079] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, in step S2, given a road segment f i At timestamp t j The Traffic Congestion Index (TSI) is calculated as follows:

[0080]

[0081] in, Representative road segment f i free flow velocity, Represents the timestamp t j The true speed. The range is [0,1], when When the value is not less than 0.7, it indicates that road segment f i At timestamp t j It has reached a congested state. In other words, when At that time, road section f i This is an example of congestion.

[0082] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, specifically step S4, constructing the spatiotemporal congestion subgraph, includes the following steps: timestamp t j The following is a subgraph of spatiotemporal congestion. Defined as a graph structure formed by connecting all congested instances as nodes within this time period, where These represent the nodes, edges, and edge weights of the graph, respectively. Edge weights It is determined by spatiotemporal connectivity. Specifically, spatiotemporal adjacency includes the following four aspects: node curvature, node degree, spatial proximity, and traffic congestion semantic relationship, calculated as follows:

[0083] Curvature is defined as the absolute value of the ratio of the tangent turning angle Δα to the arc length Δl of a road segment, expressed by the formula:

[0084]

[0085] The degree of a node is defined by the total number of edges directly connected to that node.

[0086] Spatial proximity is chosen here as spatial Euclidean distance, expressed by the formula:

[0087]

[0088] The semantic relationships are based on traffic congestion index measurements. To overcome the influence of different time resolutions in the data, a Fast Fourier Transform (FFT) is used to transform the traffic congestion index features in the time domain into features in the frequency domain. The formula is as follows:

[0089] F = FFT(TSI), (4)

[0090] All spatiotemporal congestion subgraphs constructed based on the above relationships at different timestamps t1, t2, t3, ..., t J The set of graphs is called a spatiotemporal congestion graph, and is represented by the mathematical notation:

[0091]

[0092] Where J represents the total number of timestamps.

[0093] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, specifically step S6, which uses the entropy weight method to fuse four feature similarity matrices to construct an adaptive adjacency matrix, involves the following steps: Based on the four spatiotemporal adjacency relationships defined in step S, cosine similarity is used to calculate the timestamp t. j Curvature similarity of spatiotemporal congestion subgraphs similarity matrix Fourier transform similarity of traffic congestion index Furthermore, the entropy weighting method and the spatial proximity matrix are fused to obtain t. j Spatial adjacency matrix of timestamps:

[0094]

[0095] in, and These represent the weight values ​​corresponding to each similarity matrix obtained using the entropy weighting method. Based on this, given a spatiotemporal congestion graph G containing a total of J timestamps, its spatiotemporal adaptive adjacency matrix can be automatically updated and calculated as follows:

[0096]

[0097] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, specifically step S8, which uses a local breathing priority search algorithm to identify local leaders, involves the following steps: In the first stage, by combining a dynamic adaptive adjacency matrix with the topology of the spatiotemporal congestion graph, nodes f... i The sum of the weights of all directly adjacent edges is assigned to f. i , recorded as The topological structure of the spatiotemporal congestion graph is an unweighted graph. The element values ​​in the spatiotemporal adaptive adjacency matrix represent the dynamic characteristic values ​​between node pairs. These values ​​are assigned to the edges between the corresponding nodes in the topological graph as the weights of the edges between the nodes. This process can construct a directed graph. If the following condition is satisfied, then node f... i Point to its largest neighbor f k First, the node has a higher value than its neighbors, that is... Second, each node f i They all point to their maximum neighbor, that is Where V(f) i ) represents f i The first phase is the set of all neighbors. The second phase identifies local leaders for each spatiotemporally congested subgraph; nodes with in-degree but no out-degree are considered local leaders that dominate their surrounding regions. These nodes are crucial for understanding the network's hierarchical structure. The third phase, for each local leader... A local breadth-first search algorithm is used to find the nearest local leader. Must meet The algorithm's advantage lies in its high computational efficiency, as it stops searching once the nearest local leader is identified, without traversing the entire network. Finally, based on the detected local leaders, multi-scale community centers are selected, and the transfer patterns of these centers are further analyzed, thus providing insights into the analysis of traffic network congestion bottlenecks and propagation patterns.

[0098] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, the specific steps of step S10: exploring the periodicity of traffic congestion propagation and its sensitivity to dates are as follows: merging the community centers detected in step S9 according to date type (weekday, weekend, holiday), and calculating the number and pattern of community transfers between different date types (from weekday to non-weekday, from weekday to weekend, from holiday to weekday);

[0099] The traffic congestion propagation detection method based on spatiotemporal adaptive local search described in this invention, the specific steps of step S11: exploring the interaction between human travel activities and traffic congestion propagation patterns are as follows: the community centers detected in step S9 are merged according to date type (weekday, weekend, holiday), and the number and pattern of community transfers between different types of time periods (morning peak hours, evening peak hours, off-peak hours) are calculated.

[0100] Compared with the prior art, the present invention has the following beneficial effects:

[0101] 1. The traffic congestion propagation detection method based on spatiotemporal adaptive local search provided by this invention can identify congestion bottlenecks from large-scale traffic congestion network datasets with different temporal resolutions by constructing a dynamic adaptive adjacency matrix based on multi-feature fusion. From the perspective of technological innovation, ablation experiments based on different feature combinations show that fusing graph structural information and semantic information is of great significance for traffic congestion analysis. Specifically, the fusion of Fourier transforms of the curvature, degree, spatial proximity, and traffic congestion index of nodes in the road network can improve the performance of multi-scale community structure identification in traffic networks. From the application perspective, experimental verification on two large-scale real traffic network datasets shows that the community detection method can be successfully applied to traffic congestion propagation analysis, including but not limited to tracking congestion bottlenecks, secondary and tertiary congestion centers in spatiotemporal congestion graphs.

[0102] 2. The traffic congestion propagation detection method based on spatiotemporal adaptive local search provided by this invention can effectively uncover traffic congestion propagation patterns on different types of dates (weekdays, weekends, and holidays) and different time periods (morning rush hour, evening rush hour, and off-peak hours) on the same type of date by analyzing detected multi-scale communities. Based on this, the periodicity of traffic congestion propagation patterns, their sensitivity to time periods, and the impact of human travel activities on congestion propagation patterns can be revealed, thus providing targeted suggestions for urban traffic managers. This method has strong feasibility and practicality.

[0103] like Figure 1 As shown, the present invention proposes a traffic congestion propagation detection method based on spatiotemporal adaptive local search, the specific process of which is as follows:

[0104] (1) Spatiotemporal congestion map construction: Match the original trajectory data with the road network to obtain the trajectory sequence, calculate the traffic congestion index, screen the road segments with a traffic congestion index (TSI) of not less than 0.7, and construct the spatiotemporal congestion sub-map.

[0105] (2) Extraction of graph structure features and semantic information: Based on the constructed spatiotemporal congestion subgraph, based on formula (2)

[0106] (3)(4) Calculate the curvature, spatial distance and Fourier transform of the TSI of the graph nodes, and calculate the degree of the nodes.

[0107] (3) Spatiotemporal adaptive adjacency matrix calculation: Based on the curvature, degree, Fourier transform features of TSI of graph nodes and spatial distance matrix, cosine similarity is used to calculate the corresponding curvature similarity matrix, degree similarity matrix and semantic similarity matrix. Based on the entropy weight method, the above four feature matrices are fused to obtain the adaptive adjacency matrix of each spatiotemporally congested subgraph.

[0108] (4) Multi-scale community detection: Based on the dynamic adaptive adjacency matrix and topology of each spatiotemporal congestion subgraph, a value is assigned to each node and a directed graph is constructed according to the relationship with neighboring nodes; based on the directed graph, a local breadth-first search algorithm is used to find the nearest local leader; based on the detected local leaders, multi-scale community centers are screened and the transfer patterns of community centers are analyzed.

[0109] (5) Traffic congestion propagation pattern analysis: The detected community centers are merged according to date type (weekday, weekend, holiday), and the number and pattern of community transfers between different date types (from weekday to non-weekday, from weekday to weekend, from holiday to weekday) are calculated. The congestion propagation pattern is visualized and analyzed using chord diagrams.

[0110] (6) Analysis of traffic congestion propagation patterns and human movement interaction: The detected community centers are merged according to date type (weekday, weekend, holiday), and the number and pattern of community transfers between different time periods (morning peak, evening peak, off-peak) are calculated. Sankey diagrams are used to visualize and analyze the interaction between traffic congestion propagation patterns and human movement.

[0111] Example 1:

[0112] Using the method of this invention, taking floating car trajectory data (with a time resolution of 1 hour) of a certain city 1 (study area 1) as an example, we mine traffic congestion bottlenecks and congestion propagation patterns. First, we obtain hourly spatiotemporal congestion sub-graphs based on the traffic congestion index. Then, we calculate similarity matrices based on different features according to spatiotemporal proximity. Next, we use the entropy weight method to weight and fuse the multi-feature similarity matrices to obtain an adaptive spatiotemporal adjacency matrix. Finally, we combine the adaptive adjacency matrix with the road network topology graph to search for multi-scale communities. Based on the detected communities, we analyze traffic congestion bottlenecks and congestion propagation patterns, as detailed below:

[0113] (1) Based on the collected floating car data trajectory, the original trajectory data is matched with the road network of a certain city, and the traffic congestion index is calculated to screen road segments with a TSI of not less than 0.7. A spatiotemporal congestion sub-map is constructed every hour, and a spatiotemporal congestion map containing 24 timestamps is constructed every day.

[0114] (2) Based on the calculation rules of the spatiotemporal adaptive adjacency matrix, update and calculate the adjacency matrix of each spatiotemporally congested subgraph. To verify the effectiveness of different feature combinations, an ablation experiment is conducted, using modularity as a metric to illustrate the effectiveness of the four feature combinations: node curvature, node degree, spatial distance, and Fourier transform of TSI. Figure 2 As shown in the figure, K, D, S, and T represent curvature similarity, degree similarity, spatial distance matrix, TSI similarity, and Fourier transform similarity of TSI, respectively.

[0115] (3) A spatiotemporal adaptive adjacency matrix combining four features is used for local search to detect multi-scale communities. The migration of communities between different types of dates (weekdays, weekends, and holidays) is visualized to further analyze the propagation patterns of traffic congestion, such as... Figure 3 As shown.

[0116] (4) Visualize the movement of community groups between different time periods (morning peak, evening peak, and off-peak) on the same type of date, and further analyze the interaction between traffic congestion and human travel, such as... Figure 4 , Figure 5 , Figure 6 As shown.

[0117] Example 2:

[0118] Using the method of this invention, taking taxi trajectory data (with a time resolution of 5 minutes) from City 2 (Study Area 2) as an example, traffic congestion bottlenecks and congestion propagation patterns are mined. First, a spatiotemporal congestion sub-graph is obtained every 5 minutes based on the traffic congestion index. Then, a similarity matrix based on different features is calculated according to spatiotemporal proximity. Next, the entropy weight method is used to weight and fuse the multi-feature similarity matrices to obtain an adaptive spatiotemporal adjacency matrix. Finally, the adaptive adjacency matrix is ​​combined with the road network topology graph to search for multi-scale communities. Based on the detected communities, traffic congestion bottlenecks and congestion propagation patterns are analyzed, as follows:

[0119] (1) Based on the collected taxi data trajectory, the original trajectory data is matched with the road network of a certain city, and the traffic congestion index is calculated to screen road segments with a TSI of not less than 0.7. A spatiotemporal congestion sub-map is constructed every 5 minutes, and a spatiotemporal congestion map containing 288 timestamps is constructed every day.

[0120] (2) Based on the calculation rules of the spatiotemporal adaptive adjacency matrix, update and calculate the adjacency matrix of each spatiotemporally congested subgraph. To verify the effectiveness of different feature combinations, an ablation experiment is conducted, using modularity as a metric to illustrate the effectiveness of the four feature combinations: node curvature, node degree, spatial distance, and Fourier transform of TSI. Figure 2 As shown in the figure, K, D, S, and T represent curvature similarity, degree similarity, spatial distance matrix, TSI similarity, and Fourier transform similarity of TSI, respectively.

[0121] (3) A spatiotemporal adaptive adjacency matrix combining four features is used for local search to detect multi-scale communities. The migration of communities between different types of dates (weekdays, weekends, and holidays) is visualized to further analyze the propagation patterns of traffic congestion, such as... Figure 7 As shown.

[0122] (4) Visualize the movement of community groups between different time periods (morning peak, evening peak, and off-peak) on the same type of date, and further analyze the interaction between traffic congestion and human travel, such as... Figure 8 , Figure 9 , Figure 10 As shown.

[0123] In summary, this application presents a traffic congestion propagation detection method based on spatiotemporal adaptive local search. From the perspective of multi-scale community detection, it involves a dynamically updated spatiotemporal adaptive adjacency matrix to identify traffic congestion bottlenecks and congestion propagation patterns. This addresses the problem in current traffic congestion propagation analysis that relies on a fixed predefined adjacency matrix and whose results are affected by the temporal resolution of the data. This invention allows domain experts to conduct in-depth analysis of congestion propagation at multiple levels, while being simple and efficient to implement, without needing to consider too many other factors.

[0124] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A traffic congestion propagation detection method based on spatio-temporal adaptive local search, characterized in that, The method comprises the following steps: S1, acquiring road network and floating car trajectory data and matching to obtain matching data; S2, calculating a traffic congestion index for each time slice based on the matching data; S3, constructing a space-time congestion subgraph at different times based on the traffic congestion index, the nodes of the space-time congestion subgraph being congestion road segments; S4, calculating graph structure information and semantic information of the middle road segments of the space-time congestion graph, the graph structure information including node curvature, node degree and spatial proximity, the set of the space-time congestion subgraphs constituting the space-time congestion graph; S5, based on the graph structure information and the semantic information, using cosine similarity to obtain a curvature similarity feature matrix, a degree similarity feature matrix, a spatial distance matrix and a semantic similarity matrix of the space-time congestion graph; S6, using an entropy weight method to fuse the curvature similarity feature matrix, the degree similarity feature matrix, the spatial distance matrix and the semantic similarity matrix to obtain a space-time adaptive adjacency matrix of the space-time congestion graph; S7, based on the space-time adaptive adjacency matrix, combining the topological structure of the space-time congestion graph, constructing a topological directed graph; S8, based on the directed graph, using a local breath-first search algorithm to identify local leaders of the space-time congestion graph, and then using a local breadth-first search algorithm to find the nearest local leader of each local leader, and stopping searching when the nearest local leader is determined; S9, based on the local leaders, screening multi-scale community centers to identify traffic congestion propagation bottlenecks; S10, merging the screened multi-scale community centers according to date types, and calculating the number and rules of traffic congestion propagation bottlenecks between different types of dates; S11, merging the screened multi-scale community centers according to date types, and calculating the number and rules of traffic congestion propagation bottlenecks between different types of time periods.

2. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, The traffic congestion index is: wherein, represents the free flow speed of the road segment f i , represents the true speed at the time stamp t j .

3. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, The node curvature is the absolute value of the ratio of the tangent angle Δα of the road segment to the arc length Δl; the degree of the node is defined as the total number of edges directly connected to the node; and the spatial proximity is the Euclidean distance.

4. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, The traffic congestion index is: wherein the spatiotemporal congestion subgraph is defined as the graph structure formed by connecting all congestion instances as nodes within the time period, wherein represent the nodes, edges, weights of edges, and weights of edges of the graph, respectively are determined by the node curvature, node degree, spatial proximity, and fast Fourier transform of the traffic congestion index.

5. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, The space-time adaptive adjacency matrix is: wherein, denotes the time stamp t j spatiotemporal adaptive adjacency matrix of the spatiotemporal congestion subgraph at time t, curvature similarity of the spatiotemporal congestion subgraph degree similarity matrix spatial distance matrix Fourier transform similarity of the traffic congestion index and respectively represent the weight values corresponding to each similarity matrix obtained by using the entropy weight method.

6. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, Based on the space-time adaptive adjacency matrix, combining the topological structure of the space-time congestion graph, the specific steps for constructing the directed graph are: Based on the spatio-temporal adaptive adjacency matrix, combined with the topological structure of the spatio-temporal congestion graph, the weight sum of all edges directly adjacent to node f i is assigned to f i , and the assigned weight sum is recorded as the characteristic value of node f i . A directed graph is formed, in which if the following two conditions are met, node f i points to its largest neighbor f k : First, the node has a higher value than its neighbors, i.e. Second, each node f i points to its maximum value neighbor, i.e. where V(f i ) represents the set of all neighbors of f i .

7. The traffic jam propagation detection method based on spatio-temporal adaptive local search according to claim 6, characterized in that, Identifying the local leaders of the space-time congestion graph specifically comprises: The node with an in-degree but no out-degree is regarded as a local leader that dominates the surrounding area.

8. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 7, characterized in that, Determining the nearest local leader specifically comprises: For each local leader If then is the most recent local leader of is the most recent local leader of 9. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, Different types of dates include weekdays, weekends and holidays.

10. The traffic congestion propagation detection method based on spatio-temporal adaptive local search according to claim 1, characterized in that, Different types of time periods include morning peak hours, evening peak hours and non-peak hours.

Images