Dynamic traffic bottleneck prediction method based on causal anonymous walking

Through causal anonymous walk and CAW-Network model, the congestion level cost and transmission cost of roads are calculated, and the traffic bottlenecks in urban road networks are accurately predicted, which solves the problem of difficult to detect and predict dynamic traffic bottlenecks in the existing technology, and achieves the correct Effective relief of traffic congestion.

CN116863692BActive Publication Date: 2024-11-15CHANGHUI AUTOMOBILE STEERING SYST HUANGSHAN +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202310818719.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-07-05
Publication Date
2024-11-15
Estimated Expiration
2043-07-05

AI Technical Summary

Technical Problem

The prior art is difficult to accurately detect and predict dynamic traffic bottlenecks in urban road networks, making it difficult to effectively alleviate traffic congestion.

Method used

Using a dynamic traffic bottleneck prediction method based on causal anonymous wandering, a congestion propagation diagram is constructed by calculating the congestion level cost and congestion propagation cost of roads, and using the CAW-Network model to predict traffic bottlenecks to achieve traffic bottlenecks in different time periods in the road network. Accurate prediction.

Benefits of technology

It can accurately predict traffic bottlenecks at different time periods, alleviate traffic congestion through traffic signal control and vehicle diverting, considering the spatial and temporal characteristics of congestion propagation, and improving the overall improvement effect of traffic flow.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN116863692B_ABST
    Figure CN116863692B_ABST
Patent Text Reader

Abstract

The present invention relates to the technical field of road congestion cost calculation, in particular to a dynamic traffic bottleneck prediction method based on causal anonymous walking, comprising the following steps: S1, calculating the congestion level cost of each road in a road network; S2, calculating the congestion propagation cost of each road in the road network based on the causal anonymous walking principle; S3, adding the congestion level cost and the congestion propagation cost of the road to obtain the total congestion cost of the road; if the total congestion cost of the road is greater than a set bottleneck threshold, the road is a traffic bottleneck, otherwise it is a normal road; the present invention can accurately predict traffic bottlenecks at different times in the road network, and then timely control traffic to alleviate traffic congestion.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of road congestion cost calculation, and in particular to a dynamic traffic bottleneck prediction method based on causal anonymous walking. Background Art

[0002] With the development of urbanization, more and more vehicles are put into use in cities for cargo transportation and travel. However, the traffic capacity of urban road networks is limited, and too many vehicles are put into use, which exceeds the limit of the traffic capacity of urban road networks, resulting in increasingly serious traffic congestion problems in urban road networks.

[0003] Traffic bottlenecks are the weak link in the urban road network and the root cause of urban traffic congestion. As the main factor causing congestion, traffic bottlenecks account for 40% of traffic congestion. When congestion occurs in a bottleneck section, it is more likely to spread to other sections, causing large-scale congestion in the road network. The concept of bottleneck means that relieving congestion in traffic bottleneck sections can bring about improvements in traffic flow within the road network, not just the traffic situation at the bottleneck location. In order to improve traffic flow, patent 200810117959.8 discloses a control method for increasing the vehicle throughput at a traffic bottleneck. The method sets a buffer zone, restricts the driving rules of vehicles in the buffer zone, and controls the number of vehicles in the buffer zone; and automatically introduces a feedback mechanism in the control of the number of vehicles, so that the self-recovery ability of the system becomes stronger. Although the vehicle flow controlled by the above-mentioned prior art can be stabilized at a fixed value for a long time, the flow fluctuation caused by the entrance ramp control method is avoided. However, the existing technology fails to point out how to detect traffic bottlenecks. In the actual operation of the road network, the traffic information in the road network will change dynamically with the morning and evening peaks. Therefore, the traffic bottlenecks in the road network change at different times, and each road section may become a traffic bottleneck. Therefore, the existing technology has certain limitations in the detection of traffic bottlenecks. Summary of the invention

[0004] In order to avoid and overcome the technical problems existing in the prior art, the present invention provides a dynamic traffic bottleneck prediction method based on causal anonymous walking. The present invention can accurately predict the traffic bottlenecks at different times in the road network, and then control the traffic in time to alleviate the traffic congestion.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] The dynamic traffic bottleneck prediction method based on causal anonymous walking includes the following steps:

[0007] S1. Calculate the congestion level cost of each road in the road network;

[0008] S2, calculate the congestion propagation cost of each road in the road network based on the causal anonymous walk principle;

[0009] S3. Add the congestion level cost and congestion propagation cost of the road to obtain the total congestion cost of the road; if the total congestion cost of the road is greater than the set bottleneck threshold, the road is a traffic bottleneck, otherwise it is a normal road.

[0010] As a further solution of the present invention: the specific steps of step S1 are as follows:

[0011] S11. Obtain the average road occupancy rate of each road in the road network and form a road average occupancy rate set Y, where Y = [Y1, Y2, ..., Y N ], Y1 is the average occupancy rate of the first road in the road network, Y2 is the average occupancy rate of the second road in the road network, and Y N is the average occupancy rate of the Nth road in the road network;

[0012] S12, obtaining the average flow of each road in the road network normalized based on the maximum average flow in the road network, and forming an average flow set X, X = [X1, X2, ..., X N ], X1 is the average flow rate of the first road in the road network, X2 is the average flow rate of the second road in the road network, and X N is the average flow of the Nth road in the road network;

[0013] S13, taking the average flow of each road as a weight and assigning the average occupancy rate of the road to form a weighted congestion level of the road, and storing the weighted congestion level of each road in a weighted congestion level set M, M = [X1*Y1, X2*Y2, ..., X N *Y N ], X1*Y1 is the weighted congestion level of the first road in the road network, X2*Y2 is the weighted congestion level of the second road in the road network, X N *Y N is the weighted congestion level of the Nth road in the road network;

[0014] S14. Then, the congestion level cost calculation formula is used to calculate the congestion level cost of each road. The calculation formula is as follows:

[0015] W i =X i *Y i / max{X1*Y1,X2*Y2,...,X N *Y N}

[0016] Among them, W i Represents the congestion level cost of the i-th road in the road network.

[0017] As a further solution of the present invention: the specific steps of step S2 are as follows:

[0018] S21, judging whether each road is in a congested state according to the congestion state index;

[0019] S22, determining the congestion correlation between the road in a congested state and adjacent roads according to a congestion correlation determination criterion;

[0020] S23. Based on the graph theory principle, according to the congestion schedule of each road in the road network and the congestion correlation between roads, a congestion propagation map of the road network is constructed;

[0021] S24, inputting the congestion propagation graph into the CAW-Network model, performing causal anonymous walks, sampling a predetermined number of anonymous walk sequences, and the walk sequences constitute a congestion propagation data set; inputting the congestion propagation data set into the CAW-Network model for training, thereby obtaining an optimal CAW-Network model; and calculating the congestion propagation probability between roads at any time through the optimal CAW-Network model;

[0022] S25. Generate a maximum spanning tree based on the congestion propagation graph, and calculate the congestion propagation cost of each road according to the congestion propagation probability.

[0023] As a further solution of the present invention: the specific steps of step S21 are as follows:

[0024] S211, firstly obtain the travel speed of each road, and calculate the average travel speed of the entire road network by averaging;

[0025] S212, let n% of the average travel speed be a congestion status indicator;

[0026] S213, comparing the travel speeds of various roads at different times with the congestion status index. If the travel speed of a road is lower than the congestion status index, the road is in a congested state during the time period; otherwise, the road is not congested.

[0027] As a further solution of the present invention: the specific steps of step S22 are as follows:

[0028] S221, setting the first criterion for congestion correlation as follows: the shortest path distance K between two roads is less than the spatial threshold T s ;

[0029] S222: Set the second criterion for congestion correlation as: the shortest path distance K between the two roads and the time difference T between the two roads when congestion occurs d The ratio K / T d , is within the pre-specified congestion propagation speed interval;

[0030] S223. If two roads enter a congested state successively and satisfy the first judgment criterion and the second judgment criterion at the same time, there is congestion correlation between the two roads, and congestion is propagated from the road where congestion occurs first to the road where congestion occurs later.

[0031] As a further solution of the present invention: the specific steps of step S23 are as follows:

[0032] S231, obtaining a congestion schedule for each road in the road network, and obtaining all congestion propagation between the roads in the road network based on the congestion correlation between the roads;

[0033] S232. Construct a congestion propagation map of the road network according to the propagation direction of congestion between the roads.

[0034] As a further solution of the present invention: the specific steps of step S25 are as follows:

[0035] S251, based on the congestion propagation graph, taking the node for which the congestion propagation cost needs to be calculated as the root node, calling the maximum spanning tree algorithm to generate a maximum spanning tree, so as to obtain the congestion propagation path of the node road in the road network;

[0036] S252, using the optimal CAW-Network model to calculate the congestion propagation probability between two arbitrary roads at any time;

[0037] S253, based on the road congestion level cost, congestion propagation probability and maximum spanning tree, calculate the congestion propagation cost of all roads in the entire road network; the specific operation should be like this. In order to predict the traffic bottleneck in the road network at time t, I need to calculate the total congestion cost of all roads in the road network at this time.

[0038] The calculation process of the road congestion level cost has been described above. Next, we need to determine how to calculate the congestion propagation cost of the road. Before this step, we have obtained the congestion propagation between all roads in the road network and constructed the congestion propagation graph based on graph theory.

[0039] To calculate the congestion propagation cost of all roads, it is necessary to be detailed to a specific road. It is necessary to know the congestion propagation path of each specific road and calculate the congestion propagation cost based on the congestion propagation path.

[0040] Since the congestion propagation graph is a directed graph, if you need to obtain the congestion propagation path of road A, you only need to use the maximum spanning tree algorithm on the congestion propagation graph with road A as the root node to obtain the maximum spanning tree. The maximum spanning tree is the congestion propagation law of the root node road in the road network.

[0041] S254, delete each child node in turn in the reverse direction of the propagation direction of each propagation path in the maximum spanning tree, and the total congestion propagation cost of the parent node of the currently deleted child node is the product of the congestion level cost of the parent node plus the congestion propagation probability of the parent node and the child node and the total congestion propagation cost of the child node; if the parent node has a set number of child nodes, the total congestion propagation cost of the parent node is the product of the congestion level cost of the parent node plus the congestion propagation probability of the set number of the parent node and the child nodes and the total congestion propagation cost of the child node; the total congestion propagation cost of the child node at the end of the propagation path is the congestion level cost of the child node;

[0042] S255. Calculate the congestion propagation cost of the root node through step S254.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] 1. The present invention can accurately predict traffic bottlenecks in the road network at different time periods within a day, so that traffic police can be sent to the corresponding bottleneck sections for command at different times, and traffic congestion at the bottlenecks can be alleviated by means of traffic signal control and vehicle diversion.

[0045] 2. The present invention considers that the congestion of a road section may spread to surrounding roads and cause congestion there, and quantifies the congestion propagation effect of the road section. In this process, not only its spatial characteristics, i.e., the propagation range and propagation path, but also its temporal characteristics need to be considered. This is because the different congestion levels of road sections at different times cause changes in the probability of congestion propagation.

[0046] 3. The present invention uses the time network to abstract the congestion propagation system in the road network and construct a spatiotemporal coupled congestion propagation dynamic model. The source and target road sections within the spatial threshold are screened out, and then the timestamps of their congestion propagation are obtained according to the congestion propagation speed interval. Finally, based on the knowledge of graph theory, a congestion propagation graph containing the congestion propagation law can be constructed. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process structure of the present invention.

[0048] Figure 2 This is a technical route diagram of the present invention.

[0049] Figure 3 It is the travel speed diagram of the road in the present invention.

[0050] Figure 4 It is a schematic diagram of the road congestion correlation structure in the present invention.

[0051] Figure 5a This is the congestion schedule of road A in the present invention.

[0052] Figure 5b This is the congestion schedule of road B in the present invention.

[0053] Figure 5c Schematic diagram of the congestion propagation structure of the road section AB in the present invention.

[0054] Figure 5d This is a diagram of the road network congestion propagation network model in the present invention.

[0055] Figure 6 It is the ternary closure regular diagram in the present invention.

[0056] Figure 7a This is a schematic diagram of the causal relationship extraction structure in the present invention.

[0057] Figure 7b Schematic diagram of the set-based anonymization structure in the present invention.

[0058] Figure 8 It is a schematic diagram of the process of constructing a congestion propagation graph in the present invention.

[0059] Fig. 9 It is a structural schematic diagram of the maximum spanning tree in the present invention.

[0060] Fig.10 It is a schematic diagram of the congestion propagation cost calculation process based on the maximum spanning tree in the present invention.

[0061] Fig.11 It is a schematic diagram of the structure of the regional road network set in the present invention.

[0062] Fig.12 The total congestion cost of the regional road network in the 450th minute is set in the implementation of the present invention.

[0063] Fig.13a The bottleneck section with the highest total cost in the regional road network of 10 to 450 minutes is set in the implementation of the present invention.

[0064] Fig.13b The bottleneck section with the highest total cost in the 450 to 900 minute road network of the regional road network is set in the implementation of the present invention.

[0065] Fig.14 A heat map of bottleneck sections of a regional road network set within one day in the implementation of the present invention.

[0066] Fig.15 The travel speed improvement results of different roads in the set regional road network widening in the implementation of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] See also Figures 1-2 In an embodiment of the present invention, a dynamic traffic bottleneck prediction method based on causal anonymous walking is described in detail as follows.

[0069] 1. Dynamic traffic bottleneck

[0070] Some relevant studies only consider the congestion level of the road itself when measuring the definition of bottlenecks, without considering the congestion propagation effect that road congestion may spread to nearby roads. In addition, due to the existence of morning and evening peaks, the congestion levels of roads in the road network at different times are different, resulting in changes in the probability of congestion propagation, so the traffic bottlenecks in the road network may change in different time periods. In response to the above problems, we fully consider the spatial and temporal characteristics of congestion propagation and propose a new dynamic traffic bottleneck definition. We predict roads with a total congestion cost higher than the bottleneck threshold as traffic bottlenecks. Among them, the total congestion cost of the road is equal to the sum of the congestion level cost and the congestion propagation cost. The dynamic characteristics are reflected in the changes in the probability of congestion propagation of roads in different time periods, resulting in different congestion propagation costs. The key to this method is to quantify the road congestion level cost and the congestion propagation cost.

[0071] 2. Quantifying the cost of congestion levels

[0072] This method uses the average occupancy rate of roads to characterize the congestion level of road sections. However, due to the different locations of roads in the road network, sections with the same congestion level in the road network may have different effects on the road network. Therefore, this method uses the normalized average road flow as a weight to reflect the importance of the road.

[0073] Specifically, for a road network with N roads, let the average occupancy rates of the N roads be [Y1, Y2, ..., Y N ], the average flow of all roads normalized based on the maximum average flow is [X1, X2, ..., X N ]. Then, the weighted congestion level of all roads in the road network is: [X1*Y1, X2*Y2, ..., X N *Y N ].

[0074] Furthermore, the normalized weighted congestion level of the road is used as the congestion level cost of the road, that is:

[0075] Wi =X i *Y i / max{X1*Y1,X2*Y2,...,X N *Y N}

[0076] Among them, W i Represents the congestion level cost of the i-th road in the road network.

[0077] 3. Quantifying congestion propagation costs and total congestion costs

[0078] 3.1. Definition of congestion status

[0079] Some studies set thresholds for indicators such as travel time, travel speed, and road occupancy rate to determine whether a road is congested. These thresholds are usually fixed and specified empirically, and cannot reflect the characteristics of different roads. In order to better identify the congestion state of a road, this method uses the important indicator of the average travel speed of the road to judge whether the road is congested. Figure 3 As shown, we calculate the average travel speed based on the historical data of road travel speed, and then take n% of the average travel speed, for example 50%, as the congestion status indicator. It is considered that when the road travel speed is lower than this congestion status indicator, the road is in a congested state.

[0080] 3.2. Filtering Congestion Propagation

[0081] After clarifying the criteria for evaluating road congestion status, we can further discuss the congestion propagation effect between roads. After the current road is congested, within a certain time range, the congestion may spread to other roads within a certain spatial range. If this situation is met, the two roads are considered to have congestion correlation. The above is a description of congestion correlation, and the congestion correlation also designs corresponding judgment criteria.

[0082] Congestion correlation: If road A and road B are congested one after another and meet the judgment criteria, there is congestion correlation between road A and road B.

[0083] The first judgment criterion (spatial threshold) is: the shortest path distance K between two roads is less than the spatial threshold T s .

[0084] The second judgment criterion (congestion propagation speed interval) is: the shortest path distance K between the two roads and the time difference T when congestion occurs on the two roads respectively d The ratio K / T d , which is within the pre-specified congestion propagation speed interval.

[0085] Figure 4An example is given in the figure. Road 1 is congested at 7:00. After a period of time, four roads are congested one after another. According to the spatial threshold T s , we can filter out the shortest path distance from road 1 that is less than the spatial threshold T s Road 2, Road 3, Road 4. Among them, the shortest path distance between Road 5 and Road 1 is greater than the spatial threshold T s It can be seen that road 4 and road 1 are congested almost at the same time. According to the specified congestion speed propagation interval, it can be calculated that the congestion propagation speed between them is not within the congestion speed propagation interval, so roads 4 and 1 have no congestion correlation. In summary, when considering both the spatial threshold and the congestion speed propagation interval, roads 2 and 3 have congestion correlation with road 1. And the congestion propagation is from road 1 to roads 2 and 3.

[0086] Based on the congestion timetables, spatial thresholds, and congestion propagation speed ranges of all roads in the road network, all congestion propagation occurring in the road network within a day can be screened out.

[0087] 3.3 Congested Propagation Time Network

[0088] Temporal networks generally abstract the dynamic systems of the real world to study the laws of the evolution of the systems over time. For example, the ternary closure law in interpersonal social networks describes that if two nodes have a common adjacent node, then the two nodes will often be linked in the near future, which reflects how people establish social connections. This social network has obvious time-shift characteristics and periodicity, that is, it requires two nodes to be linked to a third node before they are linked to each other, and if there are other three nodes in the network that meet this structure, they will also undergo this evolution. Similar to the social network model, in the urban road network, after a road is congested, it may spread to the surrounding roads over time, causing congestion on the surrounding roads, which has a time-shift characteristic. And because the road network structure is fixed and the daily morning and evening peak cycles are similar, the congestion propagation in the road network also has regularity and is periodic. Therefore, we thought of using time networks to characterize the congestion propagation laws in the road network, build a dynamic model of congestion propagation, and master its laws of evolution over time, so as to predict traffic bottlenecks.

[0089] Using the road congestion schedule and the pre-specified spatial threshold and congestion propagation speed interval, we can obtain when congestion occurs between any roads in a day and how many times congestion propagation occurs. More specifically, if Figure 5a and Figure 5b is the congestion schedule of road A and road B. Assume that road segment B is within the spatial threshold of road segment A. According to Figure 5a and Figure 5bThe congestion propagation speed can be calculated by the time difference between the congestion occurrence of sections A and B and the shortest path distance of section AB. If the congestion propagation speed is within the pre-specified congestion propagation speed range, it is considered that congestion propagation occurs between sections AB. Figure 5c Congestion propagation occurred between the middle section AB at 420 and 470 minutes. Using the section as the node where the interaction occurs, the congestion propagation as the interaction, and the time of the congestion propagation as the time label, we can construct the following: Figure 5c Congestion propagation. Expand the screening scope to the entire road network, obtain all congestion propagations that occur within a day, and merge congestion propagation based on graph theory knowledge to obtain the following: Figure 5d The road network congestion propagation network model contains the congestion propagation law of the road network.

[0090] 3.4. Principle of Causal Anonymous Walk

[0091] The idea of ​​modeling temporal networks through causal anonymous walks is inspired by the research on temporal network motifs. Temporal network motifs refer to connected subgraphs with links within a finite time range, which essentially reflects the dynamic nature of the network. Figure 6 The ternary closure network is a network motif that reflects the dynamic process of changes in interpersonal relationships in society. When the inductive model captures this pattern, it should predict that nodes B and C will have a third link later, and the time t3 required for the third link to occur is greater than t1 and t2. However, this prediction should be independent of the node identity of the motif so that it can be applied to the same network structure with different nodes. At the same time, if we only focus on structural dynamics and completely ignore the node identity, the inductive model will confuse the nodes and cause prediction failures. Causal anonymous walks have two important properties that can solve the above problems.

[0092] Causal Relationship Extraction: CAW starts from the link of interest and traces back several adjacent links over time to encode the potential causal relationships of network dynamics. Each walk essentially captures a temporal network motif. Figure 7a As shown in the figure, we want to predict whether nodes U and V interact at t=10. We take nodes U and V as the starting nodes, perform M m-step walks, and obtain 2M walk samples. Because the walk is performed in time, the timestamps of the links in the walk are constantly decreasing. Among them, the default timestamp and default node are found when the historical link cannot be found.

[0093] Set-based anonymization: CAW removes node identifiers during the walk to ensure inductive learning, and encodes relative node identifiers based on the number of times a node appears at a certain position in a set of walk samples. Relative node identifiers ensure that the structure and relevance of the motif remain unchanged after removing the node identifiers. Figure 7bAs shown in Figure 2, we first perform a collection operation on the six 3-step walks sampled from nodes U and V to form a set S u and S v . Use the count information of nodes appearing in different positions to encode their relative node identifiers I CAW Finally, the node identifiers of all nodes are replaced with relative node identifiers, and the walk set is anonymized. CAW It not only encodes which walk each network motif corresponds to, but also establishes the correlations between these network motifs.

[0094] 3.5 Congestion Propagation Probability

[0095] CAW-Network: In order to predict the temporal link between two nodes of interest, some scholars have proposed a model CAW-Network (CAW-N) to sample some causal anonymous walks (CAWS) from these two nodes, and encode and aggregate these CAWs through RNNs and set-pooling respectively to predict links. CAW-N proposes causal anonymous walks based on undirected graphs, but congestion propagation in road networks is obviously directional. Therefore, this paper modifies CAW-N for causal anonymous walks on directed graphs, so that it can sample walks based on directed temporal networks and predict temporal links. Experiments have shown that modifying to causal anonymous walks on directed graphs can improve model performance. Finally, based on the congestion propagation dynamic model, a congestion propagation dataset is constructed for CAW-Network training. After the performance of the model is verified, this network can be used to predict the probability of congestion propagation between roads at any time, and to calculate the congestion propagation cost.

[0096] 3.6 Congestion Propagation Graph and Maximum Spanning Tree

[0097] Based on the definition of the speed tables and congestion correlation of all roads in the road network, we can obtain the congestion relationship of all roads. Based on the congestion relationship, we can construct the congestion propagation map of the road network. The congestion propagation map contains the range and path information of congestion propagation on the road network. A more detailed construction process is as follows Figure 8 shown.

[0098] Figure 8 The initial state is a partially constructed congestion propagation graph and a new congestion relation. All congestion relations need to be inserted into the graph to obtain a complete set of congestion propagation graphs. When one end node of the inserted congestion relation already belongs to the congestion propagation graph, and the other end node does not, it is sufficient to add a new node and introduce the corresponding edge on the basis of the existing congestion propagation graph, such as Figure 8 The first step. Of course, if the two end nodes of the inserted congestion relationship are not on the existing congestion propagation graph, a new congestion propagation graph is created, such as Figure 8Finally, if the two endpoints of the newly inserted congestion relation belong to the child nodes of different congestion propagation graphs in the congestion graph set, then a directed edge is added between the corresponding nodes to merge the two congestion propagation graphs, such as Figure 8 By following these three rules and continuously inserting congestion relations, we can obtain a complete set of congestion propagation graphs.

[0099] After obtaining the congestion propagation graph of the road network, a maximum spanning tree is constructed on it. The maximum spanning tree shows the congestion propagation path and impact range of the root node section in the road network. Based on the maximum spanning tree, we quantify the congestion propagation effect caused by a road section and calculate the corresponding congestion propagation cost. The formal definition of the maximum spanning tree is as follows:

[0100] Maximum spanning tree: A maximum spanning tree of a graph is a tree consisting of the maximum set of directed edges such that there is one and only one path from the root node of the tree to any other node of the tree.

[0101] This method proposes a maximum spanning tree algorithm based on breadth-first traversal. The core of the algorithm is to determine whether a loop path will be formed in the tree to destroy the tree structure each time a new edge is inserted into the spanning tree. The judgment method is to check whether there is a directed path from j to i in the current spanning tree. Fig. 9 In Figure 1 we show how to construct a maximum spanning tree with each node as the root in a graph consisting of nodes A, B, C, D, E, and F.

[0102] 3.7 Congestion propagation cost, total congestion cost and bottleneck prediction

[0103] Finally, based on the maximum spanning tree, we can use the congestion level cost of the road segment and the congestion propagation probability at the corresponding time to calculate the congestion propagation cost of the root node road segment. We start from the leaf node of the maximum spanning tree and recursively calculate the congestion propagation cost from the parent node of the leaf node to the leaf node until the root node, as shown in the following figure. Fig.10 shown.

[0104] Fig.10 The maximum spanning tree shown in has 6 nodes and 7 congestion propagation directed edges, and the edges contain the congestion propagation probability at the corresponding time. It reflects the congestion propagation effect of the root node A in the road network. The congestion level cost of each road section is recorded as [W A , W B , W C , W D , W E , W F ].like Fig.10As shown, we calculate from left to right. We start from leaf node F and first get the congestion propagation cost from its parent node E and parent node D to leaf node F, which are P DF W F and P EF W F Then we delete the leaf node F. At this time, nodes D and E become leaf nodes in the new graph, and their total congestion cost is W. D +P DF W F and WE+P EF W F In this way, we can calculate the congestion cost of E's parent nodes B and C, and finally get the total congestion cost of the root node A, as follows:

[0105] Total congestion cost of node B: TC b =W B +P BE (W E +P EF W F );

[0106] Total congestion cost of node C: TC c =W C +P CE (W E +P EF W F );

[0107] Total congestion cost of node A: TC a =W A +P AB [W B +P BE (W E +P EF W F )]+P AC [W C +P CE (W E +P EF W F )]+P AD [W D +P DF W F ].

[0108] Where W Ais the congestion level cost of node A itself, and the others are congestion propagation costs. By modifying the timestamp of the test set and letting CAW-N estimate the congestion propagation probability at different times, we can calculate the congestion propagation cost of node A at different times. Following this idea, we can obtain the total congestion cost of all roads in the road network at different times. By comparing the total congestion costs of different roads at the same time, we predict the roads with a total congestion cost higher than the bottleneck threshold (specified empirically) as bottlenecks in the road network during this time period. Based on the above steps, we can predict traffic bottlenecks at any time period in the road network.

[0109] 4. Embodiment

[0110] 4.1 Road network information

[0111] The traffic simulation tool SUMO is used to simulate the traffic conditions of a set area road network for one day. After using the detector to collect traffic information, a congestion propagation dataset is constructed and CAW-N is trained. The total congestion cost is calculated based on the dynamic congestion propagation probability and the bottleneck sections in the road network are predicted. Finally, we verify the effectiveness of the proposed dynamic bottleneck prediction technology by widening all roads and comparing the travel speed improvement of the road network.

[0112] like Fig.11 As shown in the figure, in the traffic simulation tool SUMO, 24 nodes are used to match the main intersections in a certain area, and 100 edges are used to match the main roads in the area. The length of each edge is approximately equal to the straight-line distance between two intersections in the real road network. In the traffic simulation tool SUMO, each edge is set to three lanes, which can better simulate the situation of the real road network. We simulate the morning and evening rush hours and use randomTrips to randomly generate traffic flow files in the road network. Finally, an E2 detector is set on each lane to collect traffic flow information within a day.

[0113] 4.2. Create a dataset and train CAW-N

[0114] Based on the output file of the E2 detector, we collect the travel speed, traffic volume and occupancy information of the road. The moment when the vehicle travel speed is lower than 50% of the average travel speed is considered as congestion, and the congestion schedule of all roads is obtained. Then, all congestion propagation moments and congestion correlations are obtained by combining the spatial threshold and the congestion propagation speed interval. The congestion propagation dataset is constructed by summarizing all the source sections, target sections and corresponding congestion propagation moments where congestion propagation occurs.

[0115] The congestion propagation dataset was used to train and test CAW-N. In the end, the model test had an accuracy rate of 87% for both the transduction task and the link prediction task. After the performance of the model was verified, we obtained all the node pairs that needed to predict the congestion propagation probability based on the maximum spanning tree. Combined with different timestamps, a test set was constructed to predict the congestion propagation probability at different times. Finally, CAW-N was used to predict the congestion propagation probability.

[0116] 4.3 Predicting traffic bottlenecks

[0117] First, the traffic volume and occupancy information in the road network is used to calculate the congestion level cost of 100 roads in the road network. Fig.12 As shown, road 33 has the highest congestion level cost.

[0118] Then, based on the maximum spanning tree, combined with the congestion propagation probability at the corresponding time and the congestion level cost itself, the total congestion cost of all roads in the road network can be calculated. Fig.12 , taking the 450th minute as an example, the roads with a total congestion cost higher than a certain threshold are regarded as bottlenecks in the road network at the 450th minute, namely sections 33 and 38.

[0119] exist Fig.13a and 13b In Figure 1 we show the bottlenecks with the highest total cost in the network from 10 to 900 minutes. Fig.14 In the figure we show a heat map of bottleneck sections in a day.

[0120] 4.4 Traffic Bottleneck Verification

[0121] In the verification phase, we widened 100 roads and changed 3 lanes to 5 lanes. Without changing the traffic file, we re-simulated and collected the travel speed information in the E2 detector output file to see the improvement in the average travel speed of the road network. Fig.15 shown.

[0122] like Fig.15 As shown, when the road section 33, which is most predicted to be a bottleneck at all times of the day, is widened, the average travel speed of the road network is improved the most, which proves the accuracy of the method in predicting traffic bottlenecks.

[0123] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A dynamic traffic bottleneck prediction method based on causal anonymous walking, characterized by: The following steps are involved: S1. Calculate the congestion level cost of each road in the road network; S2, calculate the congestion propagation cost of each road in the road network based on the causal anonymous walk principle; S3, adding the congestion level cost and congestion propagation cost of the road to obtain the total congestion cost of the road; if the total congestion cost of the road is greater than the set bottleneck threshold, the road is a traffic bottleneck, otherwise it is a normal road; The specific steps of step S1 are as follows: S11. Obtain the average road occupancy rate of each road in the road network and form a road average occupancy rate set Y, where Y = [Y1, Y2, ..., Y N ], Y1 is the average occupancy rate of the first road in the road network, Y2 is the average occupancy rate of the second road in the road network, and Y N is the average occupancy rate of the Nth road in the road network; S12, obtaining the average flow of each road in the road network normalized based on the maximum average flow in the road network, and forming an average flow set X, X = [X1, X2, ..., X N ], X1 is the average flow rate of the first road in the road network, X2 is the average flow rate of the second road in the road network, and X N is the average flow of the Nth road in the road network; S13, taking the average flow of each road as a weight and assigning the average occupancy rate of the road to form a weighted congestion level of the road, and storing the weighted congestion level of each road in a weighted congestion level set M, M = [X1*Y1, X2*Y2, ..., X N *Y N ], X1*Y1 is the weighted congestion level of the first road in the road network, X2*Y2 is the weighted congestion level of the second road in the road network, X N *Y N is the weighted congestion level of the Nth road in the road network; S14. Then, the congestion level cost calculation formula is used to calculate the congestion level cost of each road. The calculation formula is as follows: W i =X i *Y i / max{X1*Y1,X2*Y2,...,X N *Y N } Among them, W i represents the congestion level cost of the i-th road in the road network; The specific steps of step S2 are as follows: S21, judging whether each road is in a congested state according to the congestion state index; S22, determining the congestion correlation between the road in a congested state and adjacent roads according to a congestion correlation determination criterion; S23. Based on the graph theory principle, according to the congestion schedule of each road in the road network and the congestion correlation between roads, a congestion propagation map of the road network is constructed; S24, inputting the congestion propagation graph into the CAW-Network model, performing causal anonymous walks, sampling a predetermined number of anonymous walk sequences, and the walk sequences constitute a congestion propagation data set; inputting the congestion propagation data set into the CAW-Network model for training, thereby obtaining an optimal CAW-Network model; and calculating the congestion propagation probability between roads at any time through the optimal CAW-Network model; S25, generating a maximum spanning tree based on the congestion propagation graph, and calculating the congestion propagation cost of each road according to the congestion propagation probability; The specific steps of step S21 are as follows: S211, firstly obtain the travel speed of each road, and calculate the average travel speed of the entire road network by averaging; S212, let n% of the average travel speed be a congestion status indicator; S213, comparing the travel speed of each road at different times with the congestion status index, if the travel speed of the road is lower than the congestion status index, the road is in a congested state during the time period, otherwise the road is not congested; The specific steps of step S22 are as follows: S221, setting the first criterion for congestion correlation as follows: the shortest path distance K between two roads is less than the spatial threshold T s ; S222: Set the second criterion for congestion correlation as: the shortest path distance K between the two roads and the time difference T between the two roads when congestion occurs d The ratio K / T d , is within the pre-specified congestion propagation speed interval; S223: If two roads enter a congested state successively and satisfy the first judgment criterion and the second judgment criterion at the same time, then there is congestion correlation between the two roads, and congestion is propagated from the road where congestion occurs first to the road where congestion occurs later; The specific steps of step S23 are as follows: S231, obtaining a congestion schedule for each road in the road network, and obtaining all congestion propagation between the roads in the road network based on the congestion correlation between the roads; S232, constructing a congestion propagation map of the road network according to the propagation direction of congestion between various roads; The specific steps of step S25 are as follows: S251, based on the congestion propagation graph, taking the node for which the congestion propagation cost needs to be calculated as the root node, calling the maximum spanning tree algorithm to generate a maximum spanning tree, so as to obtain the congestion propagation path of the node road in the road network; S252, using the optimal CAW-Network model to calculate the congestion propagation probability between two arbitrary roads at any time; S253, calculating the congestion propagation costs of all roads in the entire road network based on the road congestion level cost, the congestion propagation probability and the maximum spanning tree; S254, delete each child node in turn in the reverse direction of the propagation direction of each propagation path in the maximum spanning tree, and the total congestion propagation cost of the parent node of the currently deleted child node is the product of the congestion level cost of the parent node plus the congestion propagation probability of the parent node and the child node and the total congestion propagation cost of the child node; if the parent node has a set number of child nodes, the total congestion propagation cost of the parent node is the product of the congestion level cost of the parent node plus the congestion propagation probability of the set number of the parent node and the child nodes and the total congestion propagation cost of the child node; the total congestion propagation cost of the child node at the end of the propagation path is the congestion level cost of the child node; S255. Calculate the congestion propagation cost of the root node through step S254.