Traffic Congestion Inference Method, System, Device and Medium Based on Temporal Dynamic Graph

Abstract

The present invention discloses a traffic congestion inference method, system, device and medium based on a temporal dynamic graph. The method includes: using the temporal dynamic graph to characterize the characteristics of the congestion state of the traffic flow time series data calculated for each road section; calculating the set of congestion propagation road sections caused by the congestion state for each road section according to the obtained congestion state time series feature vectors; calculating the road network congestion spatio-temporal propagation graph according to the set of congestion propagation road sections; and inferring the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections according to the road network congestion spatio-temporal propagation graph. The present invention uses the temporal dynamic graph to realize the characterization of the temporal non-stationary characteristics in the congestion state, and then infers the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections based on the road network congestion spatio-temporal propagation graph, accurately inverses the diffusion and dissipation process of the real road section traffic congestion, and accurately quantifies the spatio-temporal influence range of the traffic congestion.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic management, and particularly relates to a traffic congestion inference method, system, device and medium based on a temporal dynamic graph. Background Art

[0002] Obtaining a data representation pattern (or called "state") with physical meaning from a large amount of high-dimensional traffic spatio-temporal observation data is helpful to provide accurate and interpretable decision support for road system operation and traffic management planning. Usually, it is easy to obtain the independent traffic operation states on different road sections within discrete time slices through traffic observation data. However, the traffic operation process is continuous in both time and space. At the same time, traffic flow has obvious spatio-temporal non-stationarity and spatio-temporal correlation, that is, the traffic state of a certain road section may be affected by the previous time period or surrounding road sections simultaneously, resulting in a change in the traffic state. How to model spatio-temporal non-stationarity and spatio-temporal correlation when constructing spatio-temporal correlation relationships is an important issue in current research. As a typical traffic state with strong spatial propagation and temporal non-stationarity, many studies have separately studied the spatial pattern and temporal pattern of traffic congestion states from the perspectives of road network cascading phenomena and time series related theories. However, the current research lacks further exploration of the spatial propagation link-level paths and probabilities of the road network traffic operation states within continuous time. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the defect in the prior art that the spatial propagation link sections of real road section traffic congestion and the congestion propagation probability cannot be accurately deduced, and to provide a traffic congestion inference method, system, device and medium based on a temporal dynamic graph.

[0004] The present invention solves the above technical problem through the following technical solutions:

[0005] In a first aspect of the present invention, a traffic congestion inference method based on a temporal dynamic graph is provided. The traffic congestion inference method includes:

[0006] Calculating the traffic flow time series data of at least one road section;

[0007] Using a temporal dynamic graph to characterize the characteristics of the congestion state of the traffic flow time series data of each road section to obtain a congestion state time series feature vector;

[0008] Calculating a set of congestion propagation road sections caused by the congestion state of each road section according to the congestion state time series feature vector;

[0009] Calculating a road network congestion spatio-temporal propagation graph according to the set of congestion propagation road sections;

[0010] Infer the key propagation sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation sections based on the congestion spatio-temporal propagation map of the road network.

[0011] Preferably, before the step of calculating the traffic flow time series data of at least one section, the traffic congestion inference method further includes:

[0012] Obtain the topological connectivity relationship of the road network;

[0013] The step of calculating the traffic flow time series data of at least one section includes:

[0014] Calculate the traffic flow time series data of at least one section according to the topological connectivity relationship of the road network;

[0015] The step of calculating the congestion spatio-temporal propagation map of the road network according to the set of congestion propagation sections includes:

[0016] Calculate the congestion spatio-temporal propagation map of the road network according to the topological connectivity relationship of the road network and the set of congestion propagation sections.

[0017] Preferably, the step of calculating the traffic flow time series data of at least one section according to the topological connectivity relationship of the road network includes:

[0018] Obtain the road network data of the topological connectivity relationship of the road network;

[0019] Obtain the historical trajectory data of the target vehicle;

[0020] Clean and filter the historical trajectory data to obtain the processed historical trajectory data;

[0021] Establish a spatial index for the processed historical trajectory data according to the road network data, and perform map road network matching on the indexed historical trajectory data to obtain the matched historical trajectory data;

[0022] Calculate the traffic flow time series data of at least one section according to the matched historical trajectory data.

[0023] Preferably, the step of calculating the set of congestion propagation sections caused by each section in the congestion state according to the congestion state time series feature vector includes:

[0024] Calculate the spatio-temporal causal relationship of each section in the congestion state according to the congestion state time series feature vector;

[0025] Obtain the set of congestion propagation sections caused by each section in the congestion state according to the spatio-temporal causal relationship.

[0026] Preferably, the step of calculating the road network congestion spatio-temporal propagation map according to the road network topological connectivity relationship and the congestion propagation road section set includes:

[0027] Calculating a congestion propagation probability matrix according to the road network topological connectivity relationship and the congestion propagation road section set;

[0028] Obtaining the road network congestion spatio-temporal propagation map according to the congestion propagation probability matrix.

[0029] Preferably, the step of inferring the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections according to the road network congestion spatio-temporal propagation map includes:

[0030] Calculating a congestion spatio-temporal propagation chain according to the road network congestion spatio-temporal propagation map;

[0031] Verifying the congestion spatio-temporal propagation chain according to the spatio-temporal causality to obtain a key congestion spatio-temporal propagation chain with redundant spatio-temporal causality removed;

[0032] Obtaining the key propagation road sections and the congestion propagation probability of the key propagation road sections according to the key congestion spatio-temporal propagation chain.

[0033] Preferably, after the step of inferring the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections according to the road network congestion spatio-temporal propagation map, the traffic congestion inference method further includes:

[0034] Evaluating the key propagation road sections of the congestion spatio-temporal propagation chain to obtain an evaluation result.

[0035] A second aspect of the present invention provides a traffic congestion inference system based on a time-series dynamic graph. The traffic congestion inference system includes a first calculation module, a characterization module, a second calculation module, a third calculation module, and an inference module;

[0036] The first calculation module is used to calculate traffic flow time series data of at least one road section;

[0037] The characterization module is used to characterize the characteristics of the congestion state of the traffic flow time series data of each road section by using a time-series dynamic graph to obtain a congestion state time series feature vector;

[0038] The second calculation module is used to calculate a congestion propagation road section set caused by the congestion state of each road section according to the congestion state time series feature vector;

[0039] The third calculation module is used to calculate a road network congestion spatio-temporal propagation map according to the congestion propagation road section set;

[0040] The inference module is used to infer the key propagation sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation sections based on the road network congestion spatio-temporal propagation map.

[0041] Preferably, the traffic congestion inference system further includes an acquisition module;

[0042] The acquisition module is used to acquire the topological connectivity relationship of the road network;

[0043] The first calculation module is used to calculate the traffic flow time series data of at least one section according to the topological connectivity relationship of the road network;

[0044] The third calculation module is used to calculate the road network congestion spatio-temporal propagation map according to the topological connectivity relationship of the road network and the set of congestion propagation sections.

[0045] Preferably, the first calculation module includes a first acquisition unit, a second acquisition unit, a processing unit, a matching unit, and a first calculation unit;

[0046] The first acquisition unit is used to acquire the road network data of the topological connectivity relationship of the road network;

[0047] The second acquisition unit is used to acquire the historical trajectory data of the target vehicle;

[0048] The processing unit is used to clean and filter the historical trajectory data to obtain the processed historical trajectory data;

[0049] The matching unit is used to establish a spatial index for the processed historical trajectory data according to the road network data and perform map road network matching on the indexed historical trajectory data to obtain the matched historical trajectory data;

[0050] The first calculation unit is used to calculate the traffic flow time series data of at least one section according to the matched historical trajectory data.

[0051] Preferably, the second calculation module includes a second calculation unit and a third acquisition unit;

[0052] The second calculation unit is used to calculate the spatio-temporal causal relationship of each section in the congestion state according to the congestion state time series feature vector;

[0053] The third acquisition unit is used to acquire the set of congestion propagation sections caused by each section in the congestion state according to the spatio-temporal causal relationship.

[0054] Preferably, the third calculation module includes a third calculation unit and a fourth acquisition unit;

[0055] The third calculation unit is configured to calculate a congestion propagation probability matrix according to the road network topological connectivity relationship and the set of congestion propagation sections;

[0056] The fourth acquisition unit is configured to obtain the road network congestion spatio-temporal propagation map according to the congestion propagation probability matrix.

[0057] Preferably, the inference module includes a fourth calculation unit, an inspection unit, and a fifth acquisition unit;

[0058] The fourth calculation unit is configured to calculate a congestion spatio-temporal propagation chain according to the road network congestion spatio-temporal propagation map;

[0059] The inspection unit is configured to inspect the congestion spatio-temporal propagation chain according to the spatio-temporal causality relationship to obtain a key congestion spatio-temporal propagation chain with redundant spatio-temporal causality relationships removed;

[0060] The fifth acquisition unit is configured to obtain the key propagation sections and the congestion propagation probabilities of the key propagation sections according to the key congestion spatio-temporal propagation chain.

[0061] Preferably, the traffic congestion inference system further includes an evaluation module;

[0062] The evaluation module is configured to evaluate the key propagation sections of the congestion spatio-temporal propagation chain to obtain an evaluation result.

[0063] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for traffic congestion inference based on a temporal dynamic graph as described in the first aspect is implemented.

[0064] A fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method for traffic congestion inference based on a temporal dynamic graph as described in the first aspect is implemented.

[0065] The positive and progressive effects of the present invention are as follows:

[0066] The present invention uses a temporal dynamic graph to characterize the characteristics of the congestion state of the traffic flow time series data of each section in each time segment, realizes the characterization of the time series non-stationary characteristics in the congestion state, and then combines the set of congestion propagation sections of each section calculated based on the temporal feature vector of the congestion state to obtain the road network congestion spatio-temporal propagation map. Furthermore, based on the road network congestion spatio-temporal propagation map, the key propagation sections of the congestion spatio-temporal propagation chain and the congestion propagation probabilities of the key propagation sections are inferred, which can accurately reverse the diffusion and dissipation process of traffic congestion on the real road sections and accurately quantify the spatio-temporal influence range of traffic congestion. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a flowchart of the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0068] Figure 2 It is a schematic diagram of characterizing the characteristics of the traffic congestion state by using the dynamic graph structure for the traffic flow time series data on a single road section in Embodiments 1 and 2 of the present invention.

[0069] Figure 3 It is a flowchart of step 1011 of the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0070] Figure 4 It is a flowchart of step 103 of the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0071] Figure 5 It is a schematic diagram of obtaining the road network congestion spatio-temporal propagation graph in Embodiments 1 and 2 of the present invention.

[0072] Figure 6 It is a flowchart of step 1041 of the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0073] Figure 7 It is for obtaining the road section r i in Embodiments 1 and 2 of the present invention, which is a schematic diagram of the candidate congestion spatial propagation chain subtree.

[0074] Figure 8 It is a flowchart of step 105 of the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0075] Figure 9 It is for obtaining the road section r i in Embodiments 1 and 2 of the present invention, which is a schematic diagram of the key propagation road sections of the congestion spatio-temporal propagation chain.

[0076] Figure 10 It is a schematic diagram of the structure of the traffic congestion inference system based on the time-series dynamic graph in Embodiment 2 of the present invention.

[0077] Figure 11 It is a schematic diagram of the structure of the electronic device in Embodiment 3 of the present invention. Detailed implementation manners

[0078] The present invention will be further described below by way of embodiments, but the present invention is not limited thereto within the scope of the described embodiments.

[0079] Embodiment 1

[0080] This embodiment provides a traffic congestion inference method based on a time-series dynamic graph, asFigure 1 As shown in Figure 1 , the traffic congestion inference method includes:

[0081] Step 101, calculate the traffic flow time series data of at least one road segment;

[0082] Step 102, use a time series dynamic graph to characterize the characteristics of the congestion state of the traffic flow time series data of each road segment, so as to obtain a congestion state time series feature vector;

[0083] In this embodiment, a time series dynamic graph (i.e., a dynamic graph structure) is used to model the dynamic evolution of the traffic flow state characteristics of each time series segment, retaining the time correlation with the previous time segment state and the cross-time segment state, and realizing the characterization of the time series non-stationary characteristics in the congestion state.

[0084] For the speed time series data on a single road segment obtained in step 101, in order to obtain an interpretable representative time series segment to describe the traffic operation state on the road, a dynamic graph structure is used to model the state change of adjacent time series segments, and capture the dynamic change law of the traffic time series state over time. Specifically, for the time series TS = {v t , t = 0, 1,..., m}, where v t represents the speed value at time t in TS, first different time intervals are selected to perform upsampling and downsampling of the sequence data, multiple time series data with different time resolutions on a single road segment are obtained, and further multiple time series data training samples are obtained by the sliding window method. Using a time classification algorithm based on shape similarity, the most significantly representative waveform feature subsequence is extracted from the multiple time series sample data of the single road segment, and an effective waveform feature set of the speed curve of the traffic flow time series of the single road segment is obtained. For example, the obtained time series feature segment set can be defined as Ω = {ω i , i = 0, 1,..., n}, that is, n feature classes are obtained from the time series data samples, where ω i is the vector value of the extracted feature class.

[0085] For a given time series segment TS Δt and the time series data feature set Ω, the shape distance D(TS Δt , ω i ) is used as the similarity measure P(Ω|TS Δt ) between the time series segment TS Δt and all the obtained time series feature classes. The obtained similarity measure value is used as the weight to realize the mapping from the feature vector set Ω in the feature space to the time series segment TS Δt , that is, each time series segment can be represented as the dot product of different weight vectors weight and the feature vector matrix. As Figure 2As shown, for two consecutive time series segments and a directed graph structure G(n) = {V, E, M} is introduced to model the state transition relationship between adjacent time series segments, where V represents the set of nodes, E represents the set of edges, and M represents the weight matrix. Specifically, the obtained feature vector ω i is used as a single data point in the high-dimensional feature space, and the time series feature set Ω is used as the set of graph nodes V of the dynamic graph structure. That is, each node V in the graph corresponds to a feature vector ω i , and the edge e k-1，k between nodes is used to represent the weight vector weight of the previous time series segment k-1 to the weight vector weight corresponding to the time series segment k , and its corresponding edge weight matrix is denoted as m k . Since the weight calculation of each time series segment is independent of each other, the edge weight matrix can be calculated by the product of independent event probabilities. Thus, a dynamic graph structure converted from the speed time series of a single road segment can be obtained.

[0086] Based on the dynamic graph structure of the time series evolution of the traffic state of a single road segment constructed above, a graph neural network is used to extract the time series features of the road segment in the congested state. Specifically, training sample data is constructed through the obtained trajectory data and its corresponding road segment congestion event records (for example, the congestion road segment marks observed through bayonet videos). In the graph neural network, it is defined that the obtained time series data feature set Ω is used as the initial state vector of each node The information of a single graph is encoded as U i , and a message passing mechanism is used to aggregate the local structure information of the graph. A multi-layer recurrent neural network mechanism is used to propagate the information at the graph node level and the entire graph level in time, and learn the feature vectors of the state changes from the continuous time series segments to . Finally, the feature representation (h, U) of a single road segment marked as the congested state is obtained.

[0087] Step 103: Calculate the set of congested propagation road segments caused by each road segment in the congested state according to the congested state time series feature vector;

[0088] Step 104: Calculate the road network congestion spatio-temporal propagation graph according to the set of congested propagation road segments;

[0089] Step 105: Infer the key propagation road segments of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road segments according to the road network congestion spatio-temporal propagation graph.

[0090] Based on the continuous state evolution of traffic flow time series data and the link topology connectivity of roads, this embodiment infers the key propagation sections of the spatio-temporal propagation chain of traffic congestion states and the congestion propagation probability of the key propagation sections, that is, how the congestion state of a certain section at a certain moment in the traffic flow time series data reaches a certain specific section and the duration, so as to obtain a spatio-temporally continuous link-level traffic congestion propagation process.

[0091] In an implementable solution, before step 101, the traffic congestion inference method further includes:

[0092] Step 100: Obtain the road network topology connectivity relationship;

[0093] Step 101 includes:

[0094] Step 1011: Calculate the traffic flow time series data of at least one section according to the road network topology connectivity relationship;

[0095] Step 104 includes:

[0096] Step 1041: Calculate the road network congestion spatio-temporal propagation map according to the road network topology connectivity relationship and the congestion propagation section set.

[0097] In an implementable solution, as Figure 3 shown, step 1011 includes:

[0098] Step 1011-1: Obtain the road network data of the road network topology connectivity relationship;

[0099] Step 1011-2: Obtain the historical trajectory data of the target vehicle;

[0100] Step 1011-3: Clean and filter the historical trajectory data to obtain the processed historical trajectory data;

[0101] Step 1011-4: Establish a spatial index for the processed historical trajectory data according to the road network data, and perform map road network matching on the indexed historical trajectory data to obtain the matched historical trajectory data;

[0102] In this embodiment, the hidden Markov model is used to perform map road network matching on the indexed historical trajectory data.

[0103] Step 1011-5: Calculate the traffic flow time series data of at least one section according to the matched historical trajectory data.

[0104] In the specific implementation process, based on the road network data with topological connectivity (i.e., the road network data of the road network topology connectivity relationship), the trajectory matching of the target vehicle (such as a floating car) is performed, and the traffic flow time series data of the section is calculated according to the trajectory data;

[0105] For example, taking the target vehicle as a floating car, specifically, read the historical trajectory data of the floating car GNSS (Global Navigation Satellite System), and preprocess the historical trajectory data of a single floating car (the historical trajectory data includes but is not limited to vehicle number, timestamp, longitude and latitude, driving status, and driving azimuth angle). Specifically, the preprocessing includes but is not limited to field cleaning and filtering (such as effective motion trajectory recognition) processing, etc. Combining the lane-link level road network data with topological connection relationships (i.e., road network data with topological connectivity relationships), first establish a spatial index for the historical trajectory GNSS sequence point data, then use the Hidden Markov Model to perform map road network matching on the indexed historical trajectory data, and perform Gaussian filtering on the obtained spatialized historical trajectory data for smoothing, so as to obtain the historical trajectory data of the floating car for reasonably calculating motion parameters. For the complete single historical trajectory data, segment it based on the road section, and establish a spatial connection between the segmented historical trajectory data and the road section for calculating traffic flow parameters on a single road section. For the motion trajectories with different sampling frequencies at the same moment on a single road section, calculate the trajectory speed curve according to the timestamp and trajectory points, and use the dynamic time warping algorithm to regularize the speed curves aggregated on a single road section, and finally obtain the traffic flow time series data of the speed changing with time for all road sections of the road network.

[0106] In an implementable solution, as Figure 4 shown, step 103 includes:

[0107] Step 1031: Calculate the spatio-temporal causal relationship of each road section in the congestion state according to the congestion state time series feature vector;

[0108] Step 1032: Obtain the set of congestion propagation road sections caused by each road section in the congestion state according to the spatio-temporal causal relationship.

[0109] In this embodiment, based on the congestion state time series feature vector of a single road section, calculate the joint relationship of the time series and causality of the congestion state evolution between road sections (that is, calculate the spatio-temporal causal relationship of each road section in the congestion state), and obtain the set of potential influence propagation road sections of each congested road section (that is, obtain the set of congestion propagation road sections caused by each road section in the congestion state);

[0110] Specifically, for example, for road section r i and road section r j the traffic state feature vectors obtained in the k-th time period and use transfer entropy TE to calculate road section r i and road section r jThe causal relationship between the congestion states is based on the asymmetry and no aftereffect of the transfer entropy. and The size of determines the propagation direction of the congested road section. Considering that the congestion state of the road section has a certain time delay in propagation in space, that is, the road section r i Congestion in the kth time period may cause road section r j The congestion in the k+h time period, therefore, the time lag variable h is added to the road section r i With road section r j The causal relationship of the congestion effect of the road section r is calculated by using the modified transfer entropy formula. i For road section r j The spatiotemporal causal relationship that may cause congestion during the kth to k+hth time interval is shown in formula (1):

[0111]

[0112] Among them, the spectral clustering method is used for the road section r i and road section r j The state feature vector and Perform feature classification and use kernel density estimation to classify the road section r i and road section r j The joint probability density of congestion status in h time periods from kth to k+hth Calculation is performed. At the same time, since the process of traffic flow parameters changing over time satisfies the Markov property, the forward algorithm of the Markov chain is used to calculate the conditional probability and By comparing all road segment pairs (e.g. road segment r i and road section r j ) transfer entropy, according to the congestion propagation direction of the road section, the congested road section r is obtained in space. i (Cause) The set of potential congestion propagation sections that may be triggered in the kth time period (result) Cr i ,like Figure 5 As shown. Furthermore, considering that the traffic status on the road segment will change over time, the traffic status of the road segment r can be obtained by stepping the sliding window. i The time series set of potentially congested road sections in the duration sequence TS is Cr = {Cr i , i=1,2,…,m}.

[0113] This embodiment combines the congestion state time series feature vector of traffic flow time series data and the link topological connectivity of roads, which can better detect the impact of congestion on the road network over time and capture the spatio-temporal causal relationship between different congested sections.

[0114] In an implementable solution, as Figure 6 shown, step 1041 includes:

[0115] Step 1041-1: Calculate the congestion propagation probability matrix according to the road network topological connectivity relationship and the congestion propagation section set;

[0116] Step 1041-2: Obtain the road network congestion spatio-temporal propagation map according to the congestion propagation probability matrix.

[0117] In the specific implementation process, after obtaining the time series set of potential impact propagation sections of a single section, since the propagation of congestion in space has spatial continuity, therefore, the spatial adjacency relationship between each section is established through the topological connectivity relationship of the road network, and the candidate congestion sections in the set of potential impact congestion sections are screened according to the constraints of the spatial adjacency relationship. As Figure 5 shown in the road network schematic diagram, for a single section r i , according to the set of potential impact sections (i.e., the set of congestion propagation sections caused by each section in the congestion state) C obtained in step 103 and its corresponding transfer entropy, the normalized transfer entropy is regarded as the congestion state propagation probability i of section r j in the set to any section r at the k-th time period. For sections not in its candidate section set, their congestion propagation probability is regarded as 0. Therefore, for the entire road network, a two-dimensional matrix P k can be obtained to represent the congestion propagation probability between road networks at the k-th time period. Considering the first-order connectivity of congestion transfer in space, that is, the congestion propagation chain is continuous in space. At the same time, since a time delay variable is introduced when calculating the transfer entropy, considering the spatial connectivity relationship, a multi-order adjacency relationship is introduced when constructing the section spatial adjacency matrix. For example, the spatial adjacency variable o can be taken, and the reciprocal of the Euclidean distance between sections is used as the spatial propagation probability attenuation coefficient, that is, the original obtained propagation probability is multiplied by the spatial attenuation coefficient to obtain the new section congestion propagation probability with spatial constraints For sections not in its o-order neighbor set, their propagation probability is set to 0. Further, as Figure 7 shown, the road network is abstracted into a graph structure with sections (nodes) and congestion propagation relationships (edges) between sections, where the weight matrix of the edges is composed of the new probability matrix P with spatial propagation attenuation constraints ′kObtained. Thus, by introducing the time dimension t, a set of spatio-temporal congestion probability directed graphs with potential propagation paths (i.e., the road network congestion spatio-temporal propagation graph) TG = {tg k , k = 1, 2,..., t} can be obtained.

[0118] In an implementable solution, as Figure 8 shown, step 105 includes:

[0119] Step 1051: Calculate the congestion spatio-temporal propagation chain according to the road network congestion spatio-temporal propagation graph;

[0120] Step 1052: Check the congestion spatio-temporal propagation chain according to the spatio-temporal causal relationship to obtain the critical congestion spatio-temporal propagation chain with redundant spatio-temporal causal relationships removed;

[0121] Step 1053: Obtain the critical propagation sections and the congestion propagation probabilities of the critical propagation sections according to the critical congestion spatio-temporal propagation chain.

[0122] In this embodiment, based on the statistical inference of the spatio-temporal causal relationship, a significance test is performed on the congestion propagation subtree, redundant spatio-temporal causal relationships are removed, and the critical sections and propagation probabilities of the congestion propagation chain are obtained.

[0123] In the specific implementation process, for the congestion probability directed graph tg k (i.e., the road network congestion spatio-temporal propagation graph) obtained in step 104, in order to calculate the congestion propagation chain of the congested section r i , its corresponding graph node and the strongly connected component where its graph node is located are correspondingly found in the road network congestion spatio-temporal propagation graph. First, multiple subtrees with the graph node as the root node are obtained on the strongly connected component i by breadth-first search to serve as the candidate set of spatial propagation chains of the section r k . For all nodes in the graph tg , the candidate set of spatial propagation chains of all sections can be recursively obtained. Then, the sliding window is stepped. For the candidate spatial propagation chain , the significance of the spatio-temporal causal relationship between the parent and child nodes of the candidate spatial congestion propagation chain subtree is determined by further using statistical hypothesis testing. At this time, the time component t is introduced to obtain the probability represented by the connection edge weight of the parent and child nodes, and the time series curve The original time series is shuffled and reconstructed as the null hypothesis of a hypothesis test. By calculating the sample mean and standard deviation of the spatio-temporal causal relationship strengths of multiple reconstructed sequence samples, the affected road segments without significant spatio-temporal causal relationships are deleted, thereby realizing the candidate spatial congestion propagation chain in the spatial chain reduction operation. For a road segment r in the road network i , the candidate spatial congestion key propagation chain with redundant spatio-temporal causal relationships removed within the k-th time period can be obtained Move the time window with a step size of s, and combine the time series sets of the congestion key propagation chain subtrees with the road segment r i as the root node to form a new congestion spatio-temporal propagation subgraph regarding the road segment r where the edge weights between the graph nodes are updated using the maximum probability in the time series set of the congestion key propagation chain subtree, and then for the obtained key propagation chain subgraph with spatio-temporal causal relationships i with the road segment r as the root node and the weights between the nodes being probabilities, the maximum spanning tree algorithm is used to obtain the key congestion propagation road segments with the maximum sum of propagation probabilities. As shown, by calculating each graph node in the congestion probability directed graph tg i obtained in step 104 and stepping the time window, the key propagation road segments with the maximum propagation possibility for all congested road segments in the entire road network can be obtained Figure 9 and the dynamic process of the growth and disappearance of its key propagation road segments in space over time can be observed k . In this embodiment, spatio-temporal causal relationship inference is used to identify the spatial propagation chain of the congestion state, obtaining the key propagation road segments of the interpretable congestion spatio-temporal propagation chain and the credible congestion propagation probabilities of the key propagation road segments, which can accurately reverse the diffusion and dissipation process of real road traffic congestion and accurately quantify the spatio-temporal influence range of traffic congestion

[0124] <000041 (This seems to be an incomplete tag. If it's just a typo, I've translated as best as possible. If not, please clarify.)

[0125] In an implementable solution, the traffic congestion inference method further includes:

[0126] Step 106: Evaluate the key propagation road segments of the congestion spatio-temporal propagation chain to obtain an evaluation result

[0127] In this embodiment, according to the structured queuing data records extracted from video checkpoints or the congestion duration marked manually as verification data, the effectiveness of the detected results of the congestion spatio-temporal propagation chain is evaluated

[0128] It should be noted that the verification data and the historical trajectory data are data of the same period and the same road segment

[0129] In this embodiment, a time-sequence dynamic graph is used to characterize the characteristics of the congestion state of the traffic flow time-series data of each road section in each time period, realizing the characterization of the time-series non-stationary characteristics in the congestion state. Then, combined with the congestion propagation road section set of each road section calculated based on the time-series feature vector of the congestion state, a road network congestion spatio-temporal propagation graph is obtained. Furthermore, based on the road network congestion spatio-temporal propagation graph, the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections are inferred, which can accurately reverse the diffusion and dissipation process of traffic congestion on real road sections and accurately quantify the spatio-temporal influence range of traffic congestion.

[0130] Embodiment 2

[0131] This embodiment provides a traffic congestion inference system based on a time-sequence dynamic graph, as Figure 10 shown. The traffic congestion inference system includes a first calculation module 21, a characterization module 22, a second calculation module 23, a third calculation module 24, and an inference module 25;

[0132] The first calculation module 21 is used to calculate the traffic flow time-series data of at least one road section;

[0133] The characterization module 22 is used to characterize the characteristics of the congestion state of the traffic flow time-series data of each road section by using a time-sequence dynamic graph to obtain a time-series feature vector of the congestion state;

[0134] In this embodiment, a time-sequence dynamic graph (i.e., a dynamic graph structure) is used to model the dynamic evolution of the traffic flow state characteristics in each time period, retaining the time correlation with the previous time period state and the cross-time period state, and realizing the characterization of the time-series non-stationary characteristics in the congestion state.

[0135] For the speed time-series data on the above-mentioned single road section, in order to obtain interpretable representative time-series segments to characterize the traffic operation state on the road, a dynamic graph structure is used to model the state changes of adjacent time-series segments, capturing the dynamic change law of the traffic time-series state over time. Specifically, for the time series TS = {v t , t = 0, 1,..., m}, where v tDenote the speed value at time t in the TS. First, upsample and downsample the sequence data at different time intervals to obtain multiple time-series data with different time resolutions on a single road segment. Further, obtain multiple time-series data training samples through the sliding window method. Adopt the time classification algorithm based on shape similarity to extract the most representative waveform feature subsequence from the obtained multiple time-series sample data of a single road segment, and obtain an effective waveform feature set of the speed curve of the traffic flow time series of a single road segment. For example, the obtained time-series feature segment set can be defined as Ω = {ω i , i = 0, 1,..., n}, that is, n feature classes are obtained from the time-series data samples, where ω i is the vector value of the extracted feature class.

[0136] For a given time-series segment TS Δt and the time-series data feature set Ω, use the shape distance D(TS Δt , ω i ) as the similarity measure P(Ω|TS Δt ) between the time-series segment TS Δt and all the obtained time-series feature classes. Take the obtained similarity measure value as the weight to realize the mapping from the feature vector set Ω in the feature space to the time-series segment TS Δt , that is, each time-series segment can be represented as the dot product of different weight vectors weight and the feature vector matrix. As shown in Figure 2 , for two consecutive time-series segments and , introduce a directed graph structure G(n) = {V, E, M} to model the state transition relationship between adjacent time-series segments, where V represents the node set, E represents the edge set, and M represents the weight matrix. Specifically, take the obtained feature vector ω i as a single data point in the high-dimensional feature space, and take the time-series feature set Ω as the graph node set V of the dynamic graph structure, that is, each node V in the graph corresponds to a feature vector ω i . The edge e k-1，k between nodes is used to represent the transition relationship from the weight vector weight of the previous time-series segment k-1 to the weight vector weight corresponding to the time-series segment k . Its corresponding edge weight matrix is denoted as m k . Since the weight calculation of each time-series segment is independent of each other, the edge weight matrix can be calculated by the product of independent event probabilities. Thus, the dynamic graph structure converted from the speed time series of a single road segment can be obtained.

[0137] Based on the dynamic graph structure of the time-series evolution of the traffic state of a single road segment constructed above, a graph neural network is used to extract the time-series features of the road segment in the congested state. Specifically, training sample data is constructed through the obtained trajectory data and its corresponding road segment congestion event records (for example, the congestion road segment marks that can be observed through bayonet videos). In the graph neural network, the obtained set of time-series data features Ω is defined as the initial state vector of each node The information of a single graph is encoded as U i , a message passing mechanism is used to aggregate the local structure information of the graph, and a multi-layer recurrent neural network mechanism is used to propagate the information at the graph node level and the entire graph level in time respectively, learning continuous time-series segments to the eigenvector of the state change, and finally the eigenrepresentation (h, U) of a single road segment marked as the congested state is obtained

[0138] The second calculation module 23 is used to calculate the set of congested propagation road segments caused by each road segment in the congested state according to the congested state time-series feature vector

[0139] The third calculation module 24 is used to calculate the road network congestion spatio-temporal propagation graph according to the set of congested propagation road segments

[0140] The inference module 25 is used to infer the key propagation road segments of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road segments according to the road network congestion spatio-temporal propagation graph

[0141] Based on the continuous state evolution of traffic flow time-series data and the link topological connectivity of roads in this embodiment, the key propagation road segments of the spatio-temporal propagation chain of traffic congestion states and the congestion propagation probability of the key propagation road segments are inferred, that is, how the congestion state of a certain road segment at a certain moment in the traffic flow time-series data reaches a certain specific road segment and the duration, so as to obtain a spatio-temporally continuous link-level traffic congestion propagation process

[0142] In an implementable solution, as Figure 10 shown, the traffic congestion inference system further includes an acquisition module 26

[0143] The acquisition module 26 is used to acquire the road network topological connectivity

[0144] The first calculation module 21 is used to calculate the traffic flow time-series data of at least one road segment according to the road network topological connectivity

[0145] The third calculation module 24 is used to calculate the road network congestion spatio-temporal propagation graph according to the road network topological connectivity and the set of congested propagation road segments

[0146] In an implementable solution, as Figure 10As shown in the figure, the first calculation module 21 includes a first acquisition unit 211, a second acquisition unit 212, a processing unit 213, a matching unit 214, and a first calculation unit 215;

[0147] The first acquisition unit 211 is configured to acquire road network data of the topological connectivity of the road network;

[0148] The second acquisition unit 212 is configured to acquire historical trajectory data of the target vehicle;

[0149] The processing unit 213 is configured to clean and filter the historical trajectory data to obtain processed historical trajectory data;

[0150] The matching unit 214 is configured to establish a spatial index for the processed historical trajectory data according to the road network data, and perform map road network matching on the indexed historical trajectory data to obtain matched historical trajectory data;

[0151] In this embodiment, a hidden Markov model is used to perform map road network matching on the indexed historical trajectory data.

[0152] The first calculation unit 215 is configured to calculate traffic flow time series data of at least one road section according to the matched historical trajectory data.

[0153] In the specific implementation process, trajectory matching of the target vehicle (such as a floating car) is performed based on the road network data with topological connectivity (i.e., the road network data of the topological connectivity of the road network), and the traffic flow time series data of the road section is calculated according to the trajectory data;

[0154] For example, taking the target vehicle as a floating car, specifically, read the historical GNSS trajectory data of the floating car, and preprocess the historical trajectory data of a single floating car (the historical trajectory data includes, but is not limited to, vehicle number, timestamp, longitude and latitude, driving state, and driving azimuth angle). Specifically, the preprocessing includes, but is not limited to, field cleaning and filtering (such as effective movement trajectory recognition) processing, etc. Combining the lane-link level road network data with topological connection relationships (i.e., road network data with topological connectivity relationships), first establish a spatial index for the historical trajectory GNSS sequence point data, then use the hidden Markov model to perform map road network matching on the indexed historical trajectory data, and use Gaussian filtering to smooth the obtained spatialized historical trajectory data, so as to obtain the historical trajectory data of the floating car for reasonably calculating motion parameters. For the complete single historical trajectory data, segment it based on the road section, and establish a spatial connection between the segmented historical trajectory data and the road section for calculating traffic flow parameters on a single road section. For the motion trajectories with different sampling frequencies at the same moment on a single road section, calculate the trajectory speed curve according to the timestamp and trajectory points, and use the dynamic time warping algorithm to regularize the speed curves aggregated on a single road section, and finally obtain the traffic flow time series data of the speed changing with time for all road sections of the road network.

[0155] In an implementable solution, as Figure 10 shown, the second calculation module 23 includes a second calculation unit 231 and a third acquisition unit 232;

[0156] The second calculation unit 231 is used to calculate the spatio-temporal causal relationship of each road section in the congestion state according to the congestion state time series feature vector;

[0157] The third acquisition unit 232 is used to obtain the set of congestion propagation road sections caused by each road section in the congestion state according to the spatio-temporal causal relationship.

[0158] In this embodiment, based on the congestion state time series feature vector of a single road section, calculate the joint relationship of the time sequence and causality of the congestion state evolution between road sections (that is, calculate the spatio-temporal causal relationship of each road section in the congestion state), and obtain the set of potential influence propagation road sections of each congested road section (that is, obtain the set of congestion propagation road sections caused by each road section in the congestion state);

[0159] Specifically, for example, for road section r i and road section r j the traffic state feature vectors obtained in the k-th time period and use transfer entropy TE to calculate road section r i and road section r jThe causal association relationship between congestion states. Based on the characteristics of the asymmetry and non-aftereffect of transfer entropy, by comparing and to determine the propagation direction of the congested road section. Considering that the propagation of the congestion state on the road section has a certain time delay in space, that is, the congestion of road section r i in the k-th time period may cause the congestion of road section r j in the k+h-th time period. Therefore, the time-delay variable h is introduced to calculate the causal association relationship of the congestion influence between road section r i and road section r j , that is, the modified transfer entropy formula is used to calculate the spatio-temporal causal relationship of the possible congestion influence of road section r i on road section r j in the time interval from the k-th to the k+h-th, as shown in formula (1) in Embodiment 1:

[0160] Among them, the spectral clustering method is used to classify the state feature vectors i of road section r j and road section r and . The kernel density estimation is used to calculate the joint probability density i of the congestion states of road section r j and road section r in the h time periods from the k-th to the k+h-th. At the same time, since the process of traffic flow parameters changing with time satisfies the Markov property, the forward algorithm of the Markov chain is used to calculate the conditional probabilities and . By comparing the transfer entropy magnitudes of all road section pairs (such as road section r i and road section r j ), according to the congestion propagation direction of the road section, in space, the set of potential impact congestion propagation road sections (effect) Cr i that may be caused by the congested road section r i in the k-th time period is obtained, as shown in Figure 5 . Further, considering that the traffic state on the road section changes continuously with time, through the step-by-step sliding window, the time series set of potential impact congestion road sections of road section r i in the continuous time series TS is Cr = {Cr i , i = 1, 2,..., m}.

[0161] This embodiment combines the congestion state time series feature vectors of traffic flow time series data and the link topological connectivity of the road network, which can better detect the impact of congestion on the road network changing with time and capture the spatio-temporal causal relationship between different congested road sections.

[0162] In an implementable solution, as Figure 10 shown, the third computing module 24 includes a third computing unit 241 and a fourth obtaining unit 242;

[0163] The third computing unit 241 is configured to calculate a congestion propagation probability matrix according to the road network topological connectivity relationship and the congestion propagation section set;

[0164] The fourth obtaining unit 242 is configured to obtain a road network congestion spatio-temporal propagation map according to the congestion propagation probability matrix.

[0165] In the specific implementation process, after obtaining the time sequence set of potential influence propagation sections of a single section, due to the spatial continuity of the congestion propagation in space, therefore, the spatial adjacency relationship between each section is established through the road network topological connectivity relationship, and the candidate congestion sections in the potential influence congestion section set are screened according to the constraints of the spatial adjacency relationship. As Figure 5 shown in the road network schematic diagram in i , for a single section r , according to the obtained potential influence section set (that is, the congestion propagation section set C caused by each section in the congested state) and its corresponding transfer entropy, the normalized transfer entropy i is regarded as the congestion state propagation probability j of section r in the set to any section r k at the k-th time period. For sections not in its candidate section set, their congestion propagation probabilities are regarded as 0. Therefore, for the entire road network, a two-dimensional matrix P can be obtained to represent the congestion propagation probability between road networks at the k-th time period. Considering the first-order connectivity of the congestion propagation in space, that is, the congestion propagation chain is continuous in space. At the same time, since a time delay variable is introduced when calculating the transfer entropy, considering the spatial connectivity relationship, multi-order adjacency relationships are introduced when constructing the section spatial adjacency matrix. For example, the spatial adjacency variable o can be taken, and the reciprocal of the Euclidean distance between sections is used as the spatial propagation probability attenuation coefficient, that is, the original obtained propagation probability is multiplied by the spatial attenuation coefficient to obtain a new section congestion propagation probability with spatial constraints Figure 7 . For sections not in its o-order neighbor set, their propagation probabilities are set to 0. Further, as ′k shown, the road network is abstracted into a graph structure with sections (nodes) and congestion propagation relationships (edges) between sections, where the weight matrix of the edges is obtained from the new probability matrix p k with spatial propagation attenuation constraints added. Thus, by introducing the time dimension t, a set of spatio-temporal congestion probability directed graphs with potential propagation paths (that is, the road network congestion spatio-temporal propagation map) TG = {tg , k = 1, 2,..., t} can be obtained.

[0166] In an implementable solution, as Figure 10 shown, the inference module 25 includes a fourth calculation unit 251, a verification unit 252, and a fifth acquisition unit 253;

[0167] The fourth calculation unit 251 is configured to calculate a congestion spatio-temporal propagation chain based on the road network congestion spatio-temporal propagation map;

[0168] The verification unit 252 is configured to verify the congestion spatio-temporal propagation chain according to the spatio-temporal causal relationship to obtain a key congestion spatio-temporal propagation chain with redundant spatio-temporal causal relationships removed;

[0169] The fifth acquisition unit 253 is configured to obtain a key propagation section and the congestion propagation probability of the key propagation section based on the key congestion spatio-temporal propagation chain.

[0170] In this embodiment, a significance test is performed on the congestion propagation subtree based on spatio-temporal causal relationship statistics and inference, redundant spatio-temporal causal relationships are removed, and the key sections and propagation probabilities of the congestion propagation chain are obtained.

[0171] In the specific implementation process, for the congestion probability directed graph tg k (i.e., the road network congestion spatio-temporal propagation map) obtained above, in order to calculate the congestion propagation chain of the congestion section r i , that is, to correspondingly find its corresponding graph node and the strongly connected component where its graph node is located in the road network congestion spatio-temporal propagation map. First, according to breadth-first search, multiple subtrees are obtained on the strongly connected component with the graph node as the root node to be used as the candidate space propagation chain set of the section r i . For all nodes in the graph tg k , the candidate space propagation chain sets of all sections can be recursively obtained. Then, the sliding window is stepped. For the candidate space propagation chain , by further using statistical hypothesis testing to determine the significance of the spatio-temporal causal relationship between the parent and child nodes of the candidate space congestion propagation chain subtree, the time component t is introduced at this time to obtain the probability represented by the connection edge weight of the parent and child nodes and the time series curve of the congestion propagation probability curve of the candidate section is reconstructed by the method of shuffling the original time series to be the null hypothesis of the hypothesis test. By calculating the sample mean and standard deviation of the spatio-temporal causal relationship strength of multiple reconstructed sequence samples, the affected sections without significant spatio-temporal causal relationships are deleted, so as to realize the chain reduction operation of the candidate space congestion propagation chain in space. For the section r i, the candidate spatial congestion key propagation chain with redundant spatio-temporal causal relationships removed within the k-th time period can be obtained Move the time window in steps of s, and use the obtained road segment r i as the root node, and the time series set of the congestion key propagation chain sub-tree Merge to form a new congestion spatio-temporal propagation sub-graph for road segment r i Among them, the edge weights between graph nodes are updated using the maximum probability in the time series set of the congestion key propagation chain sub-tree. Then, for the obtained key propagation chain sub-graph with spatio-temporal causal relationships using road segment r i as the root node, and the weights between nodes are probabilities. Use the maximum spanning tree algorithm to obtain the key congestion propagation road segments with the maximum sum of propagation probabilities. As Figure 9 shown, by calculating each graph node in the obtained congestion probability directed graph tg k and stepping the time window, the key propagation road segments with the maximum propagation possibility for all congested road segments in the entire road network can be obtained And the dynamic process of the growth and disappearance of its key propagation road segments in space over time can be observed.

[0172] In this embodiment, spatio-temporal causal relationship inference is used to identify the spatial propagation chain of the congestion state, obtaining the key propagation road segments of the interpretable congestion spatio-temporal propagation chain and the congestion propagation probabilities of the credible key propagation road segments, which can accurately reverse the diffusion and dissipation process of real road traffic congestion and accurately quantify the spatio-temporal influence range of traffic congestion.

[0173] In an implementable solution, as Figure 10 shown, the traffic congestion inference system further includes an evaluation module 27;

[0174] The evaluation module 27 is used to evaluate the key propagation road segments of the congestion spatio-temporal propagation chain to obtain an evaluation result.

[0175] In this embodiment, according to the structured queuing data records extracted from video checkpoints or the congestion duration manually marked as verification data, the effectiveness of the detected congestion spatio-temporal propagation chain results is evaluated.

[0176] It should be noted that the verification data and the historical trajectory data are data for the same period and the same road segment.

[0177] ​In this embodiment, a time-series dynamic graph is used to characterize the characteristics of the congestion state of the traffic flow time-series data of each road section in each time period, realizing the characterization of the time-series non-stationary characteristics in the congestion state. Then, combined with the congestion propagation road section set of each road section calculated based on the time-series feature vector of the congestion state, a road network congestion spatio-temporal propagation graph is obtained. Furthermore, based on the road network congestion spatio-temporal propagation graph, the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections are inferred, which can accurately reverse the diffusion and dissipation process of traffic congestion on real road sections and accurately quantify the spatio-temporal influence range of traffic congestion.

[0178] Embodiment 3

[0179] Figure 11 FIG. 7 is a schematic structural diagram of an electronic device provided in Embodiment 3 of the present invention. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1. Figure 11 The displayed electronic device 30 is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present invention.

[0180] As Figure 11 shown, the electronic device 30 may be presented in the form of a general-purpose computing device, for example, it may be a server device. The components of the electronic device 30 may include, but are not limited to: at least one of the above-mentioned processors 31, at least one of the above-mentioned memories 32, and a bus 33 connecting different system components (including the memory 32 and the processor 31).

[0181] The bus 33 includes a data bus, an address bus, and a control bus.

[0182] The memory 32 may include volatile memory, such as a random access memory (RAM) 321 and / or a cache memory 322, and may further include a read-only memory (ROM) 323.

[0183] The memory 32 may further include a program / utilities 325 having a set (at least one) of program modules 324. Such program modules 324 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include the implementation of a network environment.

[0184] The processor 31 executes various functional applications and data processing by running the computer program stored in the memory 32, such as the traffic congestion inference method based on the time-series dynamic graph in Embodiment 1 of the present invention.

[0185] The electronic device 30 can also communicate with one or more external devices 34 (such as a keyboard, a pointing device, etc.). Such communication can be carried out through the input / output (I / O) interface 35. Moreover, the device 30 for generating a model can also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through the network adapter 36. As Figure 11 shown, the network adapter 36 communicates with other modules of the device 30 for generating a model through the bus 33. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in combination with the device 30 for generating a model, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (redundant array of independent disks) systems, tape drives, and data backup storage systems, etc.

[0186] In addition, the electronic device can also be implemented in the form of an electronic chip, on which there are memory, processor-related electronic components, and an operating program stored on the memory and executable on the processor.

[0187] It should be noted that, although several units / modules or sub-units / modules of the electronic device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present invention, the features and functions of two or more of the above-described units / modules can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0188] Embodiment 4

[0189] This embodiment provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the traffic congestion inference method based on a temporal dynamic graph in Embodiment 1.

[0190] Among them, the more specific forms that the readable storage medium can adopt can include but are not limited to: portable disks, hard disks, random access memories, read-only memories, erasable programmable read-only memories, optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0191] In a possible implementation manner, the present invention can also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to enable the terminal device to execute the traffic congestion inference method based on a temporal dynamic graph in Embodiment 1.

[0192] Among them, the program code for implementing the present invention can be written in any combination of one or more programming languages, and the program code can be executed entirely on the user device, partially on the user device, executed as an independent software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0193] Although the specific embodiments of the present invention have been described above, those skilled in the art should understand that this is only an example, and the protection scope of the present invention is defined by the appended claims. Without departing from the principle and essence of the present invention, those skilled in the art can make various changes or modifications to these embodiments, but these changes and modifications all fall within the protection scope of the present invention.

Claims

1. A traffic congestion inference method based on a temporal dynamic graph, characterized in that The traffic congestion inference method includes: Calculating traffic flow time series data of at least one road section; Using a time-series dynamic graph to characterize the characteristics of the congestion state of the traffic flow time series data of each road section, so as to obtain a congestion state time-series feature vector; Calculating a set of congestion propagation road sections caused by the congestion state of each road section according to the congestion state time-series feature vector; Calculating a road network congestion spatio-temporal propagation graph according to the set of congestion propagation road sections; Inferring the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections according to the road network congestion spatio-temporal propagation graph; The step of calculating the road network congestion spatio-temporal propagation graph according to the set of congestion propagation road sections includes: Calculating a road network congestion spatio-temporal propagation graph according to the road network topological connectivity relationship and the set of congestion propagation road sections; The step of calculating the road network congestion spatio-temporal propagation graph according to the road network topological connectivity relationship and the set of congestion propagation road sections includes: Calculating a congestion propagation probability matrix according to the road network topological connectivity relationship and the set of congestion propagation road sections; Obtaining the road network congestion spatio-temporal propagation graph according to the congestion propagation probability matrix; The step of inferring the key propagation road sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation road sections according to the road network congestion spatio-temporal propagation graph includes: Calculating a congestion spatio-temporal propagation chain according to the road network congestion spatio-temporal propagation graph; Verifying the congestion spatio-temporal propagation chain according to the spatio-temporal causality relationship to obtain a key congestion spatio-temporal propagation chain with redundant spatio-temporal causality relationships removed; Obtaining the key propagation road sections and the congestion propagation probability of the key propagation road sections according to the key congestion spatio-temporal propagation chain.

2. The traffic congestion inference method based on the temporal dynamic graph according to claim 1, wherein Before the step of calculating the traffic flow time series data of at least one road section, the traffic congestion inference method further includes: Obtaining the road network topological connectivity relationship; The step of calculating the traffic flow time series data of at least one road section includes: Calculating the traffic flow time series data of at least one road section according to the road network topological connectivity relationship.

3. The traffic congestion inference method based on the temporal dynamic graph according to claim 2, characterized in that The step of calculating the traffic flow time series data of at least one road section according to the road network topological connectivity relationship includes: Obtaining the road network data of the road network topological connectivity relationship; Obtaining the historical trajectory data of the target vehicle; Performing cleaning and filtering processing on the historical trajectory data to obtain processed historical trajectory data; Establishing a spatial index for the processed historical trajectory data according to the road network data, and performing map road network matching on the indexed historical trajectory data to obtain matched historical trajectory data; Calculating the traffic flow time series data of at least one road section according to the matched historical trajectory data.

4. The traffic congestion inference method based on a temporal dynamic graph according to claim 1, wherein The step of calculating the set of congestion propagation road sections caused by the congestion state of each road section according to the congestion state time-series feature vector includes: Calculating the spatio-temporal causality relationship of each road section in the congestion state according to the congestion state time-series feature vector; Obtaining the set of congestion propagation road sections caused by the congestion state of each road section according to the spatio-temporal causality relationship.

5. The traffic congestion inference method based on a temporal dynamic graph according to claim 1, wherein After the step of inferring the key propagation sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation sections according to the congestion spatio-temporal propagation map of the road network, the traffic congestion inference method further includes: Evaluating the key propagation sections of the congestion spatio-temporal propagation chain to obtain an evaluation result.

6. A traffic congestion inference system based on a temporal dynamic graph, characterized in that, The traffic congestion inference system includes a first calculation module, a characterization module, a second calculation module, a third calculation module, and an inference module; The first calculation module is used to calculate the traffic flow time series data of at least one section; The characterization module is used to characterize the characteristics of the congestion state of the traffic flow time series data of each section by using a time series dynamic graph to obtain a congestion state time series feature vector; The second calculation module is used to calculate the set of congestion propagation sections caused by the congestion state of each section according to the congestion state time series feature vector; The third calculation module is used to calculate the congestion spatio-temporal propagation map of the road network according to the set of congestion propagation sections; The inference module is used to infer the key propagation sections of the congestion spatio-temporal propagation chain and the congestion propagation probability of the key propagation sections according to the congestion spatio-temporal propagation map of the road network; The third calculation module is used to calculate the congestion spatio-temporal propagation map of the road network according to the road network topological connectivity relationship and the set of congestion propagation sections; The third calculation module includes a third calculation unit and a fourth acquisition unit; The third calculation unit is used to calculate the congestion propagation probability matrix according to the road network topological connectivity relationship and the set of congestion propagation sections; The fourth acquisition unit is used to obtain the congestion spatio-temporal propagation map of the road network according to the congestion propagation probability matrix; The inference module includes a fourth calculation unit, a verification unit, and a fifth acquisition unit; The fourth calculation unit is used to calculate the congestion spatio-temporal propagation chain according to the congestion spatio-temporal propagation map of the road network; The verification unit is used to verify the congestion spatio-temporal propagation chain according to the spatio-temporal causality to obtain a congestion key spatio-temporal propagation chain with redundant spatio-temporal causality removed; The fifth acquisition unit is used to obtain the key propagation sections and the congestion propagation probability of the key propagation sections according to the congestion key spatio-temporal propagation chain.

7. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the traffic congestion inference method based on the time series dynamic graph according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the traffic congestion inference method based on the time series dynamic graph according to any one of claims 1-5.

Images