Urban road management method and system based on AI

Through real-time holographic traffic perception data, the knowledge structure of the traffic network is generated, clusters and characteristics are determined, and traffic congestion rate prediction is carried out, which solves the problem that traditional urban road management methods are difficult to achieve real-time perception and accurate prediction, and improves traffic management efficiency and residents' quality of life.

CN118824018BActive Publication Date: 2024-11-26SICHUAN HUATI LIGHTING TECH
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202411297757.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-18
Publication Date
2024-11-26
Estimated Expiration
2044-09-18

AI Technical Summary

Technical Problem

Traditional urban road management methods are difficult to achieve comprehensive, real-time perception and accurate prediction of urban traffic conditions, resulting in frequent traffic congestion and accidents, which cannot meet the needs of complex traffic management.

Method used

By obtaining real-time holographic traffic perception data, generating urban traffic network knowledge structure, determining the traffic participant cluster and path cluster, extracting the fusion representation characteristics of connected members, and combining traffic perception characteristics to predict traffic congestion rate to achieve accurate traffic management and prediction.

Benefits of technology

It improves the efficiency and accuracy of urban traffic management, reduces traffic congestion, optimizes urban traffic flow, and improves residents' travel experience and quality of life.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118824018B_ABST
    Figure CN118824018B_ABST
Patent Text Reader

Abstract

The present application provides an AI-based urban road management method and system, which constructs the knowledge structure of the urban traffic network by acquiring holographic traffic perception data in real time, and determines the traffic participant cluster and traffic path cluster accordingly. The method further utilizes these cluster information to extract the connected member fusion representation features related to the target traffic participant and the target traffic path, and combines the traffic perception features to make accurate traffic congestion rate predictions, thus realizing the refined management and prediction of urban traffic status, which helps to improve the efficiency and accuracy of urban traffic management, reduce traffic congestion, optimize urban traffic flow, and thus improve the travel experience and quality of life of urban residents.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of computer technology, and more specifically, to an urban road management method and system based on AI. Background Art

[0002] With the acceleration of urbanization, urban traffic problems have become increasingly prominent. Problems such as traffic congestion and frequent traffic accidents have seriously restricted the development of cities and the quality of life of residents. In order to effectively respond to these challenges, traditional urban road management methods have been unable to meet the increasingly complex traffic management needs. In recent years, with the rapid development of artificial intelligence technology, its application in various fields has continued to expand, bringing new opportunities for urban traffic management.

[0003] Traditional urban traffic management methods mainly rely on manual monitoring and simple data analysis, which makes it difficult to achieve comprehensive, real-time perception and accurate prediction of urban traffic conditions. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, an embodiment of the present application provides an AI-based urban road management method, the method comprising:

[0005] Acquire real-time holographic traffic perception data, generate an urban traffic network knowledge structure based on traffic participants and traffic paths in the real-time holographic traffic perception data, and determine traffic participant clusters and traffic path clusters based on the urban traffic network knowledge structure;

[0006] Determining, based on the traffic participant cluster and the traffic path cluster, a connected member fusion representation feature corresponding to a target traffic participant member and a connected member fusion representation feature corresponding to a target traffic path member;

[0007] Traffic congestion rate prediction is performed based on the traffic perception characteristics corresponding to the target traffic participant member, the connected member fusion representation characteristics corresponding to the target traffic participant member, the traffic perception characteristics corresponding to the target traffic path member and the connected member fusion representation characteristics corresponding to the target traffic path member to obtain a predicted traffic congestion rate, which is a predicted result of the traffic congestion rate when the target traffic participant corresponding to the target traffic participant member travels to the target traffic path corresponding to the target traffic path member.

[0008] In a possible implementation of the first aspect, generating an urban traffic network knowledge structure based on traffic participants and traffic paths in the real-time holographic traffic perception data includes:

[0009] The traffic participant is used as a dynamic member identifier, and the traffic path is used as a static member identifier;

[0010] A traffic relationship interaction network is generated based on the dynamic member identifier, the static member identifier, and the traffic relationship between the traffic participant and the traffic path, and the traffic relationship interaction network is used as the urban traffic network knowledge structure.

[0011] On the other hand, an embodiment of the present application also provides an AI-based urban road management system, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0012] Based on the above aspects, the embodiment of the present application constructs the knowledge structure of the urban traffic network by acquiring holographic traffic perception data in real time, and determines the traffic participant cluster and traffic path cluster accordingly. The method further utilizes these cluster information to extract the connected member fusion representation features related to the target traffic participant and the target traffic path, and combines the traffic perception features to make accurate traffic congestion rate predictions, thus realizing the refined management and prediction of urban traffic status, which helps to improve the efficiency and accuracy of urban traffic management, reduce traffic congestion, optimize urban traffic flow, and thus improve the travel experience and quality of life of urban residents. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a schematic diagram of the execution flow of the AI-based urban road management method provided in an embodiment of the present application.

[0014] Figure 2 This is a schematic diagram of the hardware architecture of the AI-based urban road management system provided in an embodiment of the present application. DETAILED DESCRIPTION

[0015] The present application will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of an AI-based urban road management method provided by an embodiment of the present application. The AI-based urban road management method is introduced in detail below.

[0016] Step S110, acquiring real-time holographic traffic perception data, generating an urban traffic network knowledge structure based on traffic participants and traffic paths in the real-time holographic traffic perception data, and determining traffic participant clusters and traffic path clusters based on the urban traffic network knowledge structure.

[0017] Assume that the server is deployed in a traffic management center, responsible for processing and analyzing the city's traffic data.

[0018] First, the server obtains real-time holographic traffic perception data by connecting to various traffic monitoring devices (such as cameras, sensors, GPS positioning devices, etc.). These data include the location, speed, driving direction, vehicle type, location and movement direction of pedestrians, road conditions (such as whether there is construction or congestion), and the status of traffic lights.

[0019] Then, the urban traffic network knowledge structure is generated based on the traffic participants (such as vehicles, pedestrians, etc.) and traffic paths (such as roads, streets, etc.) in these real-time holographic traffic perception data.

[0020] For example, vehicles are used as dynamic member identifiers and roads are used as static member identifiers. The main traffic participant characteristics of vehicles are extracted, such as the movement trajectory characteristics of a vehicle (including the sequence of road nodes it passes through within a period of time), behavior characteristics (whether it frequently changes lanes, accelerates, decelerates, etc.) and category characteristics (private cars, buses, trucks, etc.); at the same time, the traffic path characteristics of roads are extracted, such as the geometric characteristics of a road (length, width, curve radius, etc.), attribute characteristics (one-way street, two-way street, expressway, etc.) and state characteristics (whether it is unobstructed, degree of congestion, etc.).

[0021] Based on these features, the traffic relationship features between vehicles and roads are extracted, such as the interaction frequency of vehicles on a certain road (the number of times they pass each day), the interaction duration (the total time spent on the road), and the interaction mode (peak or off-peak driving).

[0022] Next, the vehicle is used as a dynamic member identifier and the road is used as a static member identifier, and a unique identifier is assigned to each vehicle and road. Based on the above traffic relationship characteristics, the interaction relationship between vehicles and roads is defined, such as spatial proximity interaction relationship (whether the vehicle often travels on nearby roads), temporal synchronization interaction relationship (whether the vehicle always passes a certain road in a specific time period), and behavior similarity interaction relationship (whether the driving mode is similar to that of other vehicles).

[0023] Based on these dynamic member identifiers, static member identifiers, and interaction relationships, a traffic relationship interaction network with a bipartite graph structure is constructed. In this network, vehicles and roads are different types of nodes, and the interaction relationship between them is represented by edges. The weight of the edge represents the strength or frequency of the interaction relationship. For example, if a certain vehicle often travels on a certain road, the edge weight between them is higher.

[0024] This traffic relationship interaction network is optimized by removing some isolated nodes and edges, such as some remote roads with few vehicles; merging duplicate nodes and edges, such as two very similar adjacent roads with almost the same traffic conditions; and adjusting the edge weights according to the strength and importance of the interaction relationship, such as increasing the edge weights between roads with heavy traffic and vehicles during peak hours.

[0025] After constructing and optimizing the transit relationship interaction network, its network indicators are calculated, such as node degree (the number of edges connected to a node, reflecting the importance of the node in the network), clustering coefficient (a measure of the degree of interconnection between neighbors of a node), path length (the number of edges passing between two nodes), and network diameter (the maximum distance between any two nodes in the network), and these network indicators are added to the transit relationship interaction network for representation.

[0026] Based on the above urban traffic network knowledge structure, traffic participant clusters and traffic path clusters are determined. For example, the traffic relationship interaction network is clustered to obtain traffic interaction event clusters. This traffic relationship interaction network can be mapped into a traffic binary division structure diagram, and then iterative clustering is performed using a set clustering strategy.

[0027] Assume that the number of iterative clustering is set to 5, and 5 vehicles or 5 roads are randomly selected as the initial cluster centers. Based on the spatial weight matrix of each vehicle or road (reflecting its spatial importance and influence) and spatiotemporal dynamic characteristics (such as changes in traffic flow in different time periods), each vehicle or road is assigned to the cluster where the nearest cluster center is located. Then recalculate the average position or characteristics of all members in each cluster as the new cluster center. Repeat this allocation and recalculation process until the new cluster center no longer changes or the preset number of iterations is reached, and obtain the initial clustering result.

[0028] Calculate the density of each initial cluster in the initial clustering result. Suppose it is found that the density of two clusters is lower than the set density threshold. For these two candidate clusters, if the density of the adjacent clusters is greater than the set density threshold, then merge the two candidate clusters into the adjacent cluster. If the feature distance inside the candidate cluster is greater than the set distance, split it into multiple candidate sub-clusters, recalculate the center position of the optimized cluster, and generate a reference cluster sequence after iterative optimization.

[0029] The silhouette coefficient of this reference clustering sequence is calculated. If the silhouette coefficient meets the set conditions (for example, it is close to 1, indicating a good clustering effect), the final traffic interaction event clustering is output, thereby obtaining multiple traffic participant clusters and multiple traffic path clusters.

[0030] Step S120, determining the connected member fusion representation features corresponding to the target traffic participant member and the connected member fusion representation features corresponding to the target traffic path member according to the traffic participant cluster and the traffic path cluster.

[0031] In this embodiment, it is assumed that the server has determined multiple traffic participant clusters and multiple traffic path clusters.

[0032] Now we need to determine the connected member fusion representation features corresponding to the target traffic participant member (such as a specific bus), and the connected member fusion representation features corresponding to the target traffic path member (such as a specific main road).

[0033] First, obtain the behavior characteristics of the connected members corresponding to the connected members of the target traffic participant. Assume that the connected members of the target bus are other buses and related roads that frequently interact with it on its route. Obtain the behavior characteristics of these connected members, such as the departure time, driving speed, and passenger capacity of other buses, and the traffic volume of related roads in different time periods.

[0034] Based on the behavioral characteristics of these connected members, the traffic participation fusion coefficient corresponding to the target traffic participant is determined. For example, if other buses depart at a similar time and their routes overlap, the fusion coefficient is high; if the relevant roads have a large traffic volume during the time period when the target bus is traveling, the fusion coefficient is also high.

[0035] Based on the traffic participation fusion coefficient corresponding to the target traffic participant main member, the fusion representation characteristics of the connected members corresponding to the target traffic participant main member are determined.

[0036] Specifically, obtain the cluster of the first-level connected members corresponding to the target traffic participant, assuming that it is a traffic participant cluster consisting of multiple bus routes. Poll this cluster and use it as the target cluster. Fuse the connected member behavior features of the first-level connected members (other buses) associated with the target bus and in the target cluster, such as calculating the average speed and average passenger capacity of these buses, to obtain the fused behavior features corresponding to the target bus.

[0037] The traffic participation fusion coefficient corresponding to the target bus and the target cluster is determined based on the traffic perception characteristics of the target bus (such as current position, speed, passenger capacity, etc.) and fusion behavior characteristics.

[0038] The fusion behavior features and the traffic participation fusion coefficient are weighted to obtain the cluster representation features corresponding to the target cluster. After polling this cluster, continue to poll other clusters, repeat the above process, aggregate the cluster representation features corresponding to all clusters, and finally obtain the connected member fusion representation features corresponding to the target bus.

[0039] The same method is used to obtain the connected member behavior characteristics corresponding to the connected members of the target traffic path member (such as a main road), determine the traffic participation fusion coefficient based on these characteristics, and finally determine the connected member fusion representation characteristics.

[0040] Assume that the connected members of the target trunk road are other roads connected to it and vehicles that frequently travel on the road. Obtain the behavioral characteristics of these connected members, such as the traffic volume and vehicle types of other roads, and the travel patterns of frequently traveling vehicles.

[0041] Follow similar steps as above to determine the connected member fusion representation features corresponding to the target main road.

[0042] Step S130, predicting the traffic congestion rate based on the traffic perception characteristics corresponding to the target traffic participant member, the connected member fusion representation characteristics corresponding to the target traffic participant member, the traffic perception characteristics corresponding to the target traffic path member and the connected member fusion representation characteristics corresponding to the target traffic path member, to obtain a predicted traffic congestion rate, which is a predicted result of the traffic congestion rate when the target traffic participant corresponding to the target traffic participant member travels to the target traffic path corresponding to the target traffic path member.

[0043] In this embodiment, it is assumed that the traffic perception features and connected member fusion representation features corresponding to the target traffic participant member (a bus) and the traffic perception features and connected member fusion representation features corresponding to the target traffic path member (a main road) have been acquired.

[0044] First, the traffic perception features corresponding to the target bus (such as current position, speed, passenger capacity, etc.) and the fusion representation features of the connected members at each level (such as the fusion features with other buses and the fusion features with related roads) are aggregated to obtain the first aggregate representation feature.

[0045] The traffic perception features corresponding to the target main road (such as current traffic volume, road conditions, etc.) and the fusion representation features of the connected members at each level (such as the fusion features with the connecting roads and the fusion features with frequently traveling vehicles) are converged to obtain the second converged representation features.

[0046] The traffic perception features corresponding to the target bus and the fusion representation features of the connected members at each level are respectively fused with the traffic perception features corresponding to the target trunk road and the fusion representation features of the connected members at each level, and then converged to obtain a third converged representation feature.

[0047] The first converged representation feature, the second converged representation feature and the third converged representation feature are converged to obtain a fourth converged representation feature.

[0048] Finally, the fourth pooled representation feature is fully connected and mapped using the fully connected mapping unit to predict the traffic congestion rate of the bus traveling to this main road.

[0049] For example, the fully connected mapping unit performs complex calculations and analyses on the fourth convergence representation feature of the input, comprehensively considering the real-time status of the bus, the interaction with other traffic elements, and the current status of the main road, and finally outputs a value between 0 and 1, indicating the predicted traffic congestion rate. If the value is close to 0, it means that it is predicted that the bus is unlikely to encounter congestion when driving to the main road; if the value is close to 1, it means that it is very likely to encounter congestion.

[0050] In this way, the entire process from acquiring real-time traffic data to generating traffic congestion rate prediction results can be completed, providing valuable decision support for traffic management and planning.

[0051] Based on the above steps, the embodiment of the present application constructs the knowledge structure of the urban traffic network by acquiring holographic traffic perception data in real time, and determines the traffic participant cluster and traffic path cluster accordingly. The method further utilizes these cluster information to extract the connected member fusion representation features related to the target traffic participant and the target traffic path, and combines the traffic perception features to make accurate traffic congestion rate predictions, thus realizing the refined management and prediction of urban traffic status, which helps to improve the efficiency and accuracy of urban traffic management, reduce traffic congestion, optimize urban traffic flow, and thus improve the travel experience and quality of life of urban residents.

[0052] In a possible implementation, step S110 includes:

[0053] Step S111: taking the traffic participant as a dynamic member identifier and the traffic path as a static member identifier.

[0054] Step S112, generating a traffic relationship interaction network based on the dynamic member identifier, the static member identifier, and the traffic relationship between the traffic participant and the traffic path, and using the traffic relationship interaction network as the urban traffic network knowledge structure.

[0055] In a possible implementation, step S112 includes:

[0056] Step S1121, extracting traffic participant subject characteristics of the traffic participant subject and traffic path characteristics of the traffic path, the traffic participant subject characteristics include movement trajectory characteristics, behavior characteristics and category characteristics, and the traffic path characteristics include geometric characteristics, attribute characteristics and state characteristics.

[0057] Step S1122, extracting the traffic relationship characteristics between the traffic participant and the traffic path based on the traffic participant characteristics of the traffic participant and the traffic path characteristics of the traffic path, wherein the traffic relationship characteristics include interaction frequency, interaction duration and interaction mode.

[0058] Step S1123, the traffic participant is used as a dynamic member identifier, and the traffic path is used as a static member identifier. Each dynamic member and static member has a unique identifier.

[0059] Step S1124, defining the interaction relationship between the traffic participant and the traffic path based on the traffic relationship characteristics, wherein the interaction relationship includes a spatial proximity interaction relationship, a temporal synchronization interaction relationship, and a behavior similarity interaction relationship.

[0060] Step S1125, based on the dynamic member identifier, the static member identifier and the interaction relationship between the traffic participants and the traffic path, construct a traffic relationship interaction network with a bipartite graph structure. In the traffic relationship interaction network, dynamic members and static members are different types of nodes, and the interaction relationship between dynamic members and static members is represented by edges, and the weight of the edge represents the strength or frequency of the interaction relationship.

[0061] Step S1126, optimizing the transit relationship interaction network, the optimization step includes: removing isolated nodes and edges, merging duplicate nodes and edges, and adjusting edge weights according to the strength and importance of the interaction relationship.

[0062] Step S1127, after constructing and optimizing the transit relationship interaction network, calculate the network indicators of the transit relationship interaction network and add the calculated network indicators to the transit relationship interaction network for representation, the network indicators include: node degree, clustering coefficient, path length and network diameter: represents the maximum distance between any two nodes in the network.

[0063] In this embodiment, first, the server establishes a stable connection with various traffic monitoring devices (such as high-definition cameras, high-precision sensors, vehicle-mounted GPS positioning devices, etc.) widely distributed in the city to obtain a large amount of holographic traffic perception data in real time. These data cover the traffic conditions of various areas and roads in the city. For example, at a certain moment, the data received by the server includes the position (coordinates X1, Y1), speed (60 km / h), driving direction (due east), and vehicle model (small car) of vehicle A, the position (coordinates X2, Y2), speed (40 km / h), driving direction (due south), and vehicle model (medium-sized truck) of vehicle B, the position (coordinates X3, Y3) and moving direction (northwest) of pedestrian C, the condition of road 1 (unblocked), the condition of road 2 (half-closed due to construction), the status of signal light 1 (green light), etc.

[0064] Next, the server begins to generate the urban traffic network knowledge structure based on the traffic participants (such as vehicle A, vehicle B, pedestrian C, etc.) and traffic paths (such as road 1, road 2, etc.) in these real-time holographic traffic perception data.

[0065] The server first extracts the characteristics of the traffic participants and the characteristics of the traffic path. Taking vehicle A as an example, its movement trajectory characteristics may be that it has passed through road node 1, road node 2, road node 3, etc. in the past hour; its behavior characteristics are stable driving on the road, without frequent lane changes, acceleration or deceleration; and its category characteristics are small family cars. For road 1, its geometric characteristics include a length of 5 kilometers, a width of 20 meters, and a large curve radius. Its attribute characteristics are a two-way four-lane urban trunk road, and its state characteristics are that it is currently unobstructed and has moderate traffic volume.

[0066] Based on these extracted features, the server further extracts the traffic relationship features between vehicles and roads. For example, vehicle A interacts with road 1 twice a day in the past month, with an average interaction time of 15 minutes each time, and the interaction mode is mainly during the morning and evening rush hours on weekdays.

[0067] Then, the server uses vehicles (such as vehicle A, vehicle B, etc.) as dynamic member identifiers and roads (such as road 1, road 2, etc.) as static member identifiers. A unique identifier is assigned to each dynamic member and static member, for example, the identifier of vehicle A is 001, and the identifier of road 1 is 010.

[0068] Based on the previously extracted traffic relationship features, the server defines the interaction relationship between vehicles and roads. For example, since vehicle A often travels near Road 1, there is a strong spatial proximity interaction relationship between them; vehicle A always passes through Road 1 between 8 and 9 in the morning on weekdays, with an obvious time synchronization interaction relationship; vehicle A's driving speed and mode are similar to most other small cars driving on Road 1, with a behavioral similarity interaction relationship.

[0069] Based on these dynamic member identifiers, static member identifiers, and defined interaction relationships, the server constructs a bipartite graph structured traffic relationship interaction network. In this network, dynamic members such as vehicle A and vehicle B and static members such as road 1 and road 2 are different types of nodes. If the interaction between vehicle A and road 1 is frequent and close, the edge weight between them is high, assuming it is 0.8; while vehicle B has fewer interactions with road 1, the edge weight may be only 0.2.

[0070] After building the traffic relationship interaction network, the server optimizes it. The server finds that there is a node corresponding to a remote road with few vehicles, which has few edge connections and is an isolated node and edge, so it is removed. There are also two adjacent roads with almost the same traffic conditions, which are duplicate nodes and edges, so they are merged. At the same time, for roads with extremely high traffic volume during peak hours, such as Road 1, the server increases the weight of the edges between it and frequently passing vehicles.

[0071] After the optimization is completed, the server calculates the network indicators of the traffic relationship interaction network. For example, the degree of node road 1 is 10, which means that there are 10 vehicles that have a direct interaction relationship with it; the clustering coefficient is 0.7, which means that the nodes connected to road 1 are closely connected to each other; in terms of path length, the number of edges passed from vehicle A to road 1 is 2; the network diameter is 5, which means that the maximum distance between any two nodes in the network is 5 edges. The server adds these calculated network indicators to the traffic relationship interaction network for representation.

[0072] In a possible implementation, determining the traffic participant cluster and the traffic path cluster according to the urban traffic network knowledge structure includes:

[0073] Step S113: clustering the urban traffic network knowledge structure to obtain traffic interaction event clusters corresponding to the urban traffic network knowledge structure.

[0074] Step S114, decomposing the traffic participant and the traffic path in each traffic interaction event in the traffic interaction event clustering to obtain a plurality of traffic participant clusters and a plurality of traffic path clusters.

[0075] After successfully constructing and optimizing the knowledge structure of the urban transportation network - the traffic relationship interaction network, the server begins to determine the traffic participant clusters and traffic path clusters based on this traffic relationship interaction network.

[0076] First, the server clusters the urban traffic network knowledge structure to obtain traffic interaction event clusters corresponding to the structure. It is assumed that the urban traffic network knowledge structure processed by the server covers the traffic conditions in the city center and surrounding major areas.

[0077] The server maps this complex traffic relationship interaction network into a traffic binary division structure diagram. In this structure diagram, different traffic participants (such as vehicles, pedestrians) and traffic paths (such as roads, streets) are re-expressed as the relationship between nodes and edges.

[0078] The server uses the set clustering strategy to perform iterative clustering. The number of iterative clusters is preset to 8, and 8 representative traffic participants (such as 8 vehicles of different types and driving routes) or traffic paths (such as 8 roads of different levels and traffic volumes) are randomly selected as the initial cluster centers.

[0079] For each traffic participant or traffic path, the server assigns them to the sub-cluster with the nearest cluster center based on their spatial weight matrix (taking into account their importance and influence in geographic space) and spatiotemporal dynamic characteristics (such as traffic flow change patterns in different time periods). Then, the server recalculates the average position or characteristics of all members in each sub-cluster as the new cluster center.

[0080] After multiple allocation and recalculation processes, until the new cluster center no longer changes significantly or reaches a preset number of iterations (eg, 50 times), the server obtains the initial clustering result.

[0081] Next, the server calculates the density of each initial cluster in the initial clustering result. Assume that it is found that the density of 3 clusters is lower than the set density threshold.

[0082] The server further processes the three candidate clusters with lower density. If the density of the adjacent clusters is greater than the set density threshold, the server merges the three candidate clusters into the adjacent high-density cluster.

[0083] However, if the feature distance within a candidate cluster is greater than the set distance (for example, the driving routes of the members are too different or the traffic flow change patterns are completely different), the server will split the candidate cluster into multiple candidate sub-clusters and recalculate the center positions of these optimized clusters, and finally generate a reference cluster sequence after the iterative optimization is completed.

[0084] Then, the server calculates the silhouette coefficient of this reference clustering sequence. Assume that the server obtains the silhouette coefficient of each sample according to the pre-set formula and calculation method, and calculates that the average silhouette coefficient of the entire reference clustering sequence is 0.8 (close to 1, indicating a good clustering effect).

[0085] Since the calculated silhouette coefficient meets the set conditions, the server outputs these traffic interaction event clusters.

[0086] Subsequently, the server decomposes the traffic participants and traffic paths in each traffic interaction event in these traffic interaction event clusters.

[0087] For example, in a traffic interaction event cluster, the server identifies traffic participants such as vehicle types (such as buses, taxis, private cars), pedestrians' travel purposes (such as going to work, shopping, and leisure), and traffic paths such as road levels (such as main roads, secondary roads, and branch roads) and road functions (such as commercial streets and residential roads).

[0088] By carefully analyzing and classifying the traffic participants and traffic paths in these clusters, the server finally obtained multiple clear traffic participant clusters and multiple clear traffic path clusters.

[0089] For example, clusters are formed consisting of commuter buses and related roads within a specific area, or clusters are formed consisting of private cars traveling for shopping around a commercial center and surrounding streets.

[0090] Therefore, the clusters of traffic participants and traffic routes were determined based on the knowledge structure of the urban traffic network, providing valuable basic data and structural framework for subsequent traffic analysis and prediction.

[0091] In a possible implementation, step S113 includes:

[0092] Step S1131, when the urban traffic network knowledge structure is a traffic relationship interaction network, the traffic relationship interaction network is mapped into a traffic binary division structure diagram, and the traffic participants and traffic paths in the traffic binary division structure diagram are iteratively clustered using a set clustering strategy to obtain the traffic interaction event clusters.

[0093] Step S1132, when the urban traffic network knowledge structure is a traffic binary division structure diagram, iteratively clustering the traffic participants and traffic paths in the traffic binary division structure diagram according to the set clustering strategy to obtain the traffic interaction event clustering.

[0094] In a possible implementation, step S1132 includes:

[0095] Step S1132-1, pre-set the number K of iterative clustering according to the graph parameters of the traffic binary division structure graph, and randomly select K traffic participants or traffic paths as initial cluster centers.

[0096] Based on the spatial weight matrix and spatiotemporal dynamic characteristics of each traffic participant or traffic path, each traffic participant or traffic path is assigned to the sub-cluster where the nearest cluster center is located, and then the average position or characteristics of all members in each sub-cluster are recalculated as the new cluster center. Then, the step of assigning each traffic participant or traffic path to the sub-cluster where the nearest cluster center is located is returned to execute until the new cluster center no longer changes or the preset number of iterations is reached, and the initial clustering result is obtained.

[0097] Step S1132-2, calculating the density of each initial cluster in the initial clustering result, and obtaining candidate clusters or partial regions of candidate clusters whose density is lower than a set density threshold.

[0098] Step S1132-3, for the candidate cluster or a partial area of ​​a candidate cluster, if the density of an adjacent cluster is greater than a set density threshold, the candidate cluster or the partial area of ​​the candidate cluster is merged into the adjacent cluster; if the feature distance inside the candidate cluster or the partial area of ​​the candidate cluster is greater than the set distance, the candidate cluster or the partial area of ​​the candidate cluster is split into multiple candidate sub-clusters, and the center position of the optimized cluster is recalculated to generate a reference cluster sequence after the iterative optimization is completed.

[0099] Step S1132-4, calculate the silhouette coefficient of the reference clustering sequence after the iterative optimization is completed. The silhouette coefficient is an indicator to measure the clustering effect. The value range of the silhouette coefficient is between -1 and 1. The closer the silhouette coefficient is to 1, the better the clustering effect. The closer it is to -1, the worse the clustering effect. The calculation formula of the silhouette coefficient is: s(i)=(b(i)-a(i)) / max{a(i), b(i)}, where s(i) is the silhouette coefficient of the i-th sample in any reference cluster, a(i) is the average distance between the i-th sample and other samples in the reference cluster, and b(i) is the minimum value of the average distance between the i-th sample and all samples in other reference clusters.

[0100] Step S1132-5: when the calculated silhouette coefficient meets the set condition, the traffic interaction event cluster is output.

[0101] In this embodiment, the server first obtains the urban traffic network knowledge structure, which is represented by a traffic relationship interaction network. In order to effectively cluster it, the server begins to map the traffic relationship interaction network into a traffic binary division structure diagram.

[0102] In this traffic binary division structure diagram, various traffic participants (such as vehicles, pedestrians, bicycles, etc.) and traffic paths (such as highways, urban main roads, branch roads, alleys, etc.) are accurately represented as the relationship between nodes and edges. For example, vehicle nodes are connected to the road nodes they often travel on through edges, and pedestrian nodes are connected to the street nodes they often walk on.

[0103] Based on the detailed graph parameters of the traffic binary division structure graph, including the number of nodes, edge distribution density, node type distribution, etc., after comprehensive evaluation and analysis, the server pre-sets the number of iterative clusters K to 15. This number is determined based on the consideration of the complexity of the urban traffic structure and the requirement for the level of detail required for subsequent analysis.

[0104] Next, the server randomly selects 15 of the numerous traffic participants and traffic paths as the initial cluster centers. The selection of these 15 initial cluster centers is not completely random, but is based on their representativeness and diversity. For example, they may include 3 major highways, 5 urban traffic arteries, 3 common public transportation tools (such as buses, subway lines, and light rail lines), and 4 different types of travelers (such as office workers, students, shoppers, and tourists).

[0105] For each traffic participant or traffic path, the server starts clustering operations based on its spatial weight matrix and spatiotemporal dynamic characteristics.

[0106] Taking a specific private car as an example, the server first obtains its spatial weight matrix. This matrix reflects the activity range and key areas of the car in the urban space. Suppose through analysis, it is found that the car mainly moves in the southern part of the city and often travels between several specific communities and commercial areas.

[0107] At the same time, the server also considers its spatiotemporal dynamic characteristics. By analyzing its travel time, it is found that the car is active frequently between 8 and 9 in the morning and between 5 and 6 in the afternoon on weekdays, which is in line with the travel pattern of office workers. On weekends, its activity time is more scattered and mainly concentrated near shopping centers and leisure and entertainment areas.

[0108] Combining these spatial weight matrices and spatiotemporal dynamic characteristics, the server assigns the private car to the sub-cluster with the nearest cluster center. Assume that in this example, the nearest cluster center is an urban arterial road connecting the main community and commercial district in the southern part of the city.

[0109] After the server completes a round of allocation, it recalculates the average position or characteristics of all members in each cluster as the new cluster center.

[0110] For the above clusters including private cars and urban roads, the server will comprehensively consider various features such as the location information, traffic flow change patterns, and usage frequency of all members in the cluster. For example, the geometric center of all members is calculated as the average location, or the average trend of traffic flow changes is calculated as a feature representative.

[0111] The server continuously repeats the steps of assigning each traffic participant or traffic path to the sub-cluster where the nearest cluster center is located, and then recalculating the cluster center.

[0112] After many such iterations, the server continues to observe the changes in the cluster centers. Assume that in the first few iterations, the location and characteristics of the cluster centers have changed significantly, but as the number of iterations increases, the changes gradually decrease. The server obtains the initial clustering results until the new cluster centers no longer change significantly or the preset number of iterations (for example, 200) is reached.

[0113] Next, the server starts to calculate the density of each initial cluster in the initial clustering result.

[0114] The server evaluates the density of clusters in a variety of ways. First, it counts the number of members in each cluster. If a cluster contains too few traffic participants and traffic paths, it may mean that its density is low.

[0115] Secondly, analyze the density of connections between members. If the connections between members in a cluster are sparse and the interactions are not frequent, it also indicates that the density is low.

[0116] Furthermore, consider the similarity of member features. If the features of members in a cluster are very different and lack consistency, it will also affect the density assessment of the cluster.

[0117] Assume that after careful calculation and evaluation, the server finds that the density of 5 clusters is lower than the set density threshold.

[0118] The server performs further detailed processing on these five candidate clusters with lower density.

[0119] The server first checks the adjacent clusters of these candidate clusters. If the density of the adjacent clusters is significantly greater than the set density threshold and there is a strong logical association or similarity between them, the server merges the candidate clusters with lower density into the adjacent high-density clusters.

[0120] For example, there is a low-density cluster that mainly contains some new roads that have not been fully developed in the urban fringe areas, and few vehicles travel on these roads. The adjacent cluster is a high-density cluster that contains a mature road network in the main development area of ​​the city and a large number of frequently traveling vehicles. After comprehensive judgment, the server believes that these new roads may form a closer connection with the adjacent mature roads in the future as the city develops, so it merges this low-density cluster into the adjacent high-density cluster.

[0121] However, if the feature distance within a candidate cluster is greater than the set distance, it indicates that the members within the cluster are too different and lack internal consistency and logic. The server will split the candidate cluster into multiple candidate sub-clusters.

[0122] For example, a candidate cluster originally included an urban expressway and a narrow pedestrian street. Since the expressway mainly serves the rapid passage of vehicles, and the pedestrian street is mainly for pedestrians and a small number of non-motor vehicles, their functions, traffic characteristics, usage time, etc. are very different. The server determines that such a combination is not reasonable, splits it into two independent candidate sub-clusters, and recalculates the center positions of these optimized clusters.

[0123] After the above merging and splitting operations, the server generates a reference clustering sequence after iterative optimization.

[0124] Then, the server starts to calculate the silhouette coefficient of this reference clustering sequence.

[0125] For each cluster in the reference cluster sequence, the server calculates the silhouette coefficient of each sample therein in turn.

[0126] Assume that in a cluster, there is a sample that is a specific urban branch road. The server first calculates a(i), which is the average distance between this branch road and other roads in the cluster in terms of location, traffic volume, usage time, etc.

[0127] Next, calculate b(i), which is the minimum average distance between this branch and all roads in other clusters in terms of the same feature.

[0128] The silhouette coefficient s(i) of this branch is calculated using the formula s(i)=(b(i)-a(i)) / max{a(i),b(i)}.

[0129] The server performs such calculation on each sample in the reference clustering sequence to obtain the silhouette coefficient of each sample.

[0130] Then, the server combines the silhouette coefficients of all samples and calculates the average silhouette coefficient of the entire reference clustering sequence. Assume that this average silhouette coefficient is 0.85 (close to 1, indicating that the clustering effect is good).

[0131] Since the calculated silhouette coefficient meets the set conditions (for example, the average silhouette coefficient is set to be greater than 0.8), the server finally outputs these traffic interaction event clusters. These traffic interaction event clusters provide a valuable basis and basis for subsequent traffic flow prediction, road planning, traffic signal optimization and other work.

[0132] As a result, the knowledge structure of the urban transportation network is precisely divided and optimized, providing key support for improving the efficiency and quality of urban transportation management.

[0133] In a possible implementation, step S120 includes:

[0134] Step S121, obtaining the connected member behavior characteristics corresponding to the connected members of the target traffic participant main member, determining the traffic participation fusion coefficient corresponding to the target traffic participant main member based on the connected member behavior characteristics, and determining the connected member fusion representation characteristics corresponding to the target traffic participant main member based on the traffic participation fusion coefficient corresponding to the target traffic participant main member.

[0135] Step S122, obtaining the connected member behavior characteristics corresponding to the connected members of the target traffic path member, determining the traffic participation fusion coefficient corresponding to the target traffic path member based on the connected member behavior characteristics, and determining the connected member fusion representation characteristics corresponding to the target traffic path member based on the traffic participation fusion coefficient corresponding to the target traffic path member.

[0136] In this embodiment, it is assumed that the target traffic participant member is a specific bus numbered 005, and the target traffic path member is a city trunk road numbered 020.

[0137] First, the connected member behavior characteristics corresponding to the connected members of the target traffic participant (bus 005) are obtained.

[0138] The server starts searching for connected members related to bus 005. These connected members include other buses that often interact with it on its travel route, such as buses 006, 007, 008, etc., and related roads, such as roads 015, 016, 017, etc.

[0139] For bus 006, the server obtains that its departure time is 6:30 a.m. every day, the average speed is 40 km / h, and the average passenger volume is 80 during peak hours and 30 during off-peak hours. The departure time of bus 007 is 7 a.m. every day, the average speed is 35 km / h, the average passenger volume is 70 during peak hours and 25 during off-peak hours. The departure time of bus 008 is 6:45 a.m. every day, the average speed is 38 km / h, the average passenger volume is 75 during peak hours and 28 during off-peak hours.

[0140] For road 015, the server obtains its traffic flow information in different time periods. The traffic flow from 7 to 9 in the morning is 800 vehicles / hour, the traffic flow from 9 to 11 is 500 vehicles / hour, the traffic flow from 11 to 1 in the afternoon is 400 vehicles / hour, the traffic flow from 1 to 3 in the afternoon is 300 vehicles / hour, the traffic flow from 3 to 5 is 400 vehicles / hour, and the traffic flow from 5 to 7 is 700 vehicles / hour. The traffic flow of road 016 in the corresponding time periods is 700 vehicles / hour, 400 vehicles / hour, 300 vehicles / hour, 250 vehicles / hour, 350 vehicles / hour, and 600 vehicles / hour. The traffic flow of road 017 is 650 vehicles / hour, 350 vehicles / hour, 280 vehicles / hour, 220 vehicles / hour, 300 vehicles / hour, and 550 vehicles / hour.

[0141] Based on these connected member behavior characteristics, the server determines the traffic participation fusion coefficient corresponding to bus 005.

[0142] The server analysis found that the departure time of bus 006 is similar to that of bus 005, and their routes also partially overlap, so the fusion coefficient between them is relatively high, set to 0.8. The departure time of bus 007 is somewhat different from that of bus 005, but they have the same route in some sections, so the fusion coefficient is set to 0.6. The departure time and route of bus 008 are relatively consistent with bus 005, and the fusion coefficient is 0.7.

[0143] For road 015, since its traffic volume is large during the time period when bus 005 is running, and it has close interaction with bus 005, the fusion coefficient is set to 0.7. The traffic volume of road 016 is consistent with the travel time of bus 005 in certain time periods, and the fusion coefficient is 0.6. The traffic volume of road 017 is also related to the travel time of bus 005, and the fusion coefficient is 0.5.

[0144] Based on the traffic participation fusion coefficient corresponding to bus 005, the server determines the fusion representation characteristics of its connected members.

[0145] The server first obtains the cluster where the first-level connected member corresponding to bus 005 is located, assuming that it is a cluster of traffic participants consisting of multiple bus lines, and polls this cluster and uses it as the target cluster.

[0146] The behavior features of the connected members of the first level (other buses) associated with bus 005 and in the target cluster are fused. For example, the average speed of buses 006, 007, and 008 is calculated, which is (40+35+38)÷3=37.67 km / h. The average passenger capacity is (80+70+75)÷3=75 people.

[0147] Based on the traffic perception characteristics of bus 005, such as the current position at coordinates (X5, Y5), speed of 36 km / h, passenger capacity of 60 people, and the above fusion behavior characteristics, the traffic participation fusion coefficient corresponding to bus 005 and the target cluster is determined. Assume that the calculated fusion coefficient is 0.7.

[0148] The fusion behavior features (average speed 37.67 km / h, average passenger volume 75 people) and the traffic participation fusion coefficient 0.7 are weighted. Assuming the weights are 0.6 and 0.4 respectively, the weighted features are: the weighted value of the average speed is 37.67×0.6=22.6 km / h, the weighted value of the average passenger volume is 75×0.6=45 people, and the weighted value of the traffic participation fusion coefficient is 0.7×0.4=0.28. These weighted values ​​are summed up to obtain the cluster representation features corresponding to the target cluster.

[0149] After polling this cluster, the server continues to poll other clusters. Assuming there are two more clusters, they are processed according to the above method, and the cluster representation features corresponding to all clusters are aggregated to finally obtain the connected member fusion representation features corresponding to bus 005.

[0150] Next, the connected member behavior characteristics corresponding to the connected members of the target traffic path member (main road 020) are obtained.

[0151] The server searches for other roads connected to the main road 020, such as roads 021, 022, 023, etc., and vehicles that often travel on the main road, such as vehicles 010, 011, 012, etc.

[0152] For road 021, the server obtains that its traffic volume is 600 vehicles / hour from 7 to 9 in the morning, 400 vehicles / hour from 9 to 11, 300 vehicles / hour from 11 to 1 in the afternoon, 200 vehicles / hour from 1 to 3 in the afternoon, 300 vehicles / hour from 3 to 5, and 500 vehicles / hour from 5 to 7. The traffic volume of road 022 in the corresponding time periods is 550 vehicles / hour, 350 vehicles / hour, 250 vehicles / hour, 180 vehicles / hour, 280 vehicles / hour, and 450 vehicles / hour. The traffic volume of road 023 is 500 vehicles / hour, 300 vehicles / hour, 220 vehicles / hour, 150 vehicles / hour, 250 vehicles / hour, and 400 vehicles / hour.

[0153] For vehicle 010, the server obtains its travel pattern as starting from location A at 8:00 a.m. every weekday, passing through main road 020 to location B, with an average speed of 50 km / h. The travel pattern of vehicle 011 is starting from location C at 7:30 a.m. every Monday, Wednesday, and Friday, passing through main road 020 to location D, with an average speed of 45 km / h. The travel pattern of vehicle 012 is starting from location E at 5:00 p.m. every day, passing through main road 020 to location F, with an average speed of 40 km / h.

[0154] Based on these connected member behavior characteristics, the server determines the traffic participation fusion coefficient corresponding to the main road 020.

[0155] The server analysis found that Road 021 is closely connected to Main Road 020, and the traffic volume is relatively matched with Main Road 020 during peak hours, and the fusion coefficient is set to 0.7. The traffic volume of Road 022 is somewhat correlated with Main Road 020, and the fusion coefficient is 0.6. The connection and traffic volume of Road 023 are relatively weak, and the fusion coefficient is 0.5.

[0156] For vehicle 010, since it often travels during the peak hours of main road 020 and frequently interacts with main road 020, the fusion coefficient is set to 0.8. The travel pattern of vehicle 011 overlaps with the traffic flow of main road 020 to a certain extent, and the fusion coefficient is 0.7. Vehicle 012 travels during the evening peak hours of main road 020, and the fusion coefficient is 0.7.

[0157] Based on the traffic participation fusion coefficient corresponding to the main road 020, the server determines the fusion representation characteristics of its connected members.

[0158] The server obtains the cluster of the first-level connected member corresponding to the main road 020 in a similar way to determining the fusion representation features of the bus connected member. The server polls this cluster and fuses the behavior features of the connected members therein, such as calculating the average traffic flow of the road and the average travel pattern of the vehicles.

[0159] According to the traffic perception characteristics of the main road 020, such as the current traffic volume of 700 vehicles / hour and the good road condition, as well as the fusion behavior characteristics, the traffic participation fusion coefficient corresponding to the main road 020 and the target cluster is determined.

[0160] The fusion behavior characteristics and the traffic participation fusion coefficient are weighted to obtain the cluster representation characteristics corresponding to the target cluster.

[0161] After polling all relevant clusters, the cluster representation features corresponding to all clusters are aggregated, and finally the connected member fusion representation features corresponding to the main road 020 are obtained.

[0162] Through the above detailed calculations and analysis, the server successfully determined the connected member fusion representation features corresponding to the target traffic participant member (bus 005) and the target traffic path member (main road 020), providing important data support for subsequent traffic congestion rate prediction.

[0163] In a possible implementation, step S121 includes:

[0164] Step S1211, obtaining the cluster where the x-th level connected member corresponding to the target traffic participant member is located, polling the clusters, and taking any one of the clusters as the target cluster. The cluster is the traffic participant cluster or the traffic path cluster.

[0165] Step S1212, fusing the connected member behavior features of the x-th level connected members associated with the target traffic participant member and in the target cluster to obtain the fused behavior features corresponding to the target traffic participant member.

[0166] Step S1213, determining a traffic participation fusion coefficient corresponding to the target traffic participant and the target cluster according to the traffic perception characteristics corresponding to the target traffic participant and the fusion behavior characteristics corresponding to the target traffic participant.

[0167] Step S1214, weighting the fusion behavior characteristics and traffic participation fusion coefficients corresponding to the target traffic participant main members to obtain the cluster representation characteristics corresponding to the target cluster, and after the polling is completed, aggregating the cluster representation characteristics corresponding to all the clusters to obtain the connected member fusion representation characteristics corresponding to the target traffic participant main members.

[0168] Let’s continue to assume that the target traffic participant is a specific taxi, numbered 009. Set the x level to 1.

[0169] First, the cluster of the first-level connected member corresponding to taxi 009 is obtained. Through previous analysis and classification, the server determines that there are two related clusters, namely, traffic participant cluster A composed of multiple taxi routes and traffic path cluster B composed of related roads in a specific area.

[0170] Cluster A is polled and used as the target cluster. In cluster A, the first-level connected members associated with and located in taxi 009 include taxis 010, 011, and 012.

[0171] The server obtains the behavioral characteristics of the connected members of Taxi 010, such as its average daily operating time of 8 hours, and its main service areas are concentrated in the city's commercial areas; the average daily operating time of Taxi 011 is 10 hours, and its main service areas are urban residential areas; the average daily operating time of Taxi 012 is 9 hours, and its main service areas are urban tourist attractions.

[0172] By integrating the behavioral features of these connected members, we calculated that the average operating time is (8+10+9)÷3=9 hours, and the main service areas cover commercial areas, residential areas, and tourist attractions. In this way, we can obtain the integrated behavioral features corresponding to Taxi 009.

[0173] Next, based on the traffic perception characteristics corresponding to Taxi 009, such as the current location in the city center, the current speed of 30 km / h, and the fusion behavior characteristics obtained above, the traffic participation fusion coefficient corresponding to Taxi 009 and target cluster A is determined.

[0174] The server comprehensively analyzes these characteristics and finds that the operating mode and service area of ​​Taxi 009 have high similarity and correlation with other taxis in cluster A, and therefore determines the traffic participation fusion coefficient to be 0.7.

[0175] Then, the fusion behavior characteristics corresponding to Taxi 009 (average operating time of 9 hours, main service area) and the traffic participation fusion coefficient of 0.7 are weighted. Assume that the weight of the fusion behavior characteristics is 0.6 and the weight of the traffic participation fusion coefficient is 0.4.

[0176] The weighted value of the average operating time is 9×0.6=5.4 hours, the weighted value of the main service area is calculated based on the specific calculation method, and the weighted value of the traffic participation integration coefficient is 0.7×0.4=0.28.

[0177] These weighted values ​​are summed up to obtain the cluster representation features corresponding to the target cluster A.

[0178] After polling cluster A, start polling cluster B as the new target cluster.

[0179] In cluster B, the level 1 connected members associated with and located in taxi 009 are roads 020 , 021 , and 022 .

[0180] The server obtains the behavioral characteristics of the connected members of road 020, such as its traffic flow in different time periods: 500 vehicles / hour from 7 to 9 in the morning, 300 vehicles / hour from 12 to 14 in the afternoon, and 700 vehicles / hour from 18 to 20 in the evening; the traffic flow of road 021 in the corresponding time periods is 450 vehicles / hour, 280 vehicles / hour, and 650 vehicles / hour respectively; the traffic flow of road 022 is 480 vehicles / hour, 250 vehicles / hour, and 600 vehicles / hour.

[0181] The behavioral characteristics of these connected members are integrated to calculate the average traffic flow in different time periods.

[0182] Based on the traffic perception characteristics (current location, speed, etc.) corresponding to taxi 009 and the above fusion behavior characteristics, determine the traffic participation fusion coefficient corresponding to taxi 009 and target cluster B. Assume that it is determined to be 0.6 after analysis.

[0183] The fusion behavior characteristics and traffic participation fusion coefficients are weighted to obtain the cluster representation characteristics corresponding to the target cluster B.

[0184] After polling all relevant clusters, the cluster representation features corresponding to cluster A and cluster B are aggregated.

[0185] For example, in the cluster representation characteristics of cluster A, the weighted value of the average operating time is 5.4 hours, the weighted value of the main service area is a specific value X, and the weighted value of the traffic participation fusion coefficient is 0.28; in the cluster representation characteristics of cluster B, the weighted value of the average traffic flow in different time periods is a specific value Y, and the weighted value of the traffic participation fusion coefficient is 0.18.

[0186] These weighted values ​​are aggregated according to specific rules, and finally the connected member fusion representation features corresponding to taxi 009 are obtained.

[0187] Through the above detailed processing and calculation, the server successfully determined the corresponding connected member fusion representation features for Taxi 009, providing strong support for subsequent traffic analysis and prediction.

[0188] In a possible implementation, step S122 includes:

[0189] Step S1221, obtaining the cluster where the x-th level connected member corresponding to the target traffic path member is located, polling the clusters, and taking any one of the clusters as the target cluster. The cluster is the traffic participant cluster or the traffic path cluster.

[0190] Step S1222 , fusing the connected member behavior features of the x-th level connected members associated with the target traffic path member and in the target cluster to obtain a fused behavior feature corresponding to the target traffic path member.

[0191] Step S1223 , determining a traffic participation fusion coefficient corresponding to the target traffic path member and the target cluster according to the traffic perception characteristics corresponding to the target traffic path member and the fusion behavior characteristics corresponding to the target traffic path member.

[0192] Step S1224, weighting the fusion behavior characteristics and traffic participation fusion coefficients corresponding to the target traffic path members to obtain cluster representation characteristics corresponding to the target cluster. After polling is completed, the cluster representation characteristics corresponding to all the clusters are aggregated to obtain the connected member fusion representation characteristics corresponding to the target traffic path members.

[0193] In this embodiment, it is assumed that the target traffic path member is a specific urban secondary road numbered 030. The x level is set to 2.

[0194] First, the server obtains the cluster of the second-level connected members corresponding to the secondary road 030. After searching and analyzing, two related clusters are identified, namely, the traffic path cluster C composed of some similar-level roads and the traffic participant cluster D composed of specific types of vehicles.

[0195] Cluster C is polled and used as the target cluster. In cluster C, the second-level connected members associated with and located in the secondary road 030 include roads 031 , 032 , and 033 .

[0196] The server obtains the behavior characteristics of the connected members of road 031, for example, its average daily traffic volume is 800 vehicles during the morning peak and 1,000 vehicles during the evening peak; the average daily traffic volume of road 032 is 700 vehicles during the morning peak and 900 vehicles during the evening peak; the average daily traffic volume of road 033 is 750 vehicles during the morning peak and 950 vehicles during the evening peak.

[0197] By fusing the behavioral characteristics of these connected members, we calculate that the average traffic flow during the morning peak is (800+700+750)÷3=750 vehicles, and the average traffic flow during the evening peak is (1000+900+950)÷3=950 vehicles, thereby obtaining the fused behavioral characteristics corresponding to the secondary trunk road 030.

[0198] Next, based on the traffic perception characteristics corresponding to the secondary road 030, such as the current road width of 15 meters, the maximum vehicle flow of 1,200 vehicles per hour, and the fusion behavior characteristics obtained above, the traffic participation fusion coefficient corresponding to the secondary road 030 and the target cluster C is determined.

[0199] The server comprehensively analyzes these features and finds that secondary road 030 has certain similarities and correlations with other roads in cluster C in terms of traffic volume and road attributes, so the traffic participation fusion coefficient is determined to be 0.6.

[0200] Then, the fusion behavior characteristics corresponding to the secondary trunk road 030 (average traffic volume of 750 vehicles in the morning peak and 950 vehicles in the evening peak) and the traffic participation fusion coefficient of 0.6 are weighted. Assume that the weight of the fusion behavior characteristics is 0.7 and the weight of the traffic participation fusion coefficient is 0.3.

[0201] The weighted value of the average traffic flow during the morning rush hour is 750×0.7=525 vehicles, the weighted value of the average traffic flow during the evening rush hour is 950×0.7=665 vehicles, and the weighted value of the traffic participation integration coefficient is 0.6×0.3=0.18.

[0202] These weighted values ​​are summed up to obtain the cluster representation features corresponding to the target cluster C.

[0203] After polling cluster C, it starts polling cluster D as the new target cluster.

[0204] In cluster D, the second-level connected members associated with and located in the secondary road 030 are a set of vehicles whose vehicle type is small sedans.

[0205] The server obtains the behavioral characteristics of the connected members of these small cars, such as an average speed of 40 kilometers per hour and an average passenger capacity of 1.5 people.

[0206] The behavior features of these connected members are fused to obtain the fused behavior features corresponding to the secondary trunk road 030.

[0207] According to the traffic perception characteristics (road conditions, speed limit, etc.) corresponding to the secondary road 030 and the above fusion behavior characteristics, the traffic participation fusion coefficient corresponding to the secondary road 030 and the target cluster D is determined. It is assumed that it is determined to be 0.5 after analysis.

[0208] The fused behavior features and traffic participation fusion coefficients are weighted to obtain the cluster representation features corresponding to the target cluster D.

[0209] After polling all relevant clusters, the cluster representation features corresponding to cluster C and cluster D are aggregated.

[0210] For example, in the cluster representation characteristics of cluster C, the weighted value of the average vehicle flow during the morning peak is 525 vehicles, the weighted value of the average vehicle flow during the evening peak is 665 vehicles, and the weighted value of the traffic participation fusion coefficient is 0.18; in the cluster representation characteristics of cluster D, the weighted value of the average vehicle speed is a specific value X, the weighted value of the average passenger volume is a specific value Y, and the weighted value of the traffic participation fusion coefficient is 0.15.

[0211] These weighted values ​​are aggregated according to specific rules, and finally the connected member fusion representation features corresponding to the secondary trunk road 030 are obtained.

[0212] Therefore, the corresponding connected member fusion representation features were determined for the secondary trunk road 030, providing a valuable reference for subsequent traffic analysis and planning.

[0213] In a possible implementation manner, the connected member fusion representation features corresponding to the target traffic participant member and the connected member fusion representation features corresponding to the target traffic path member both include multi-level connected connected member fusion representation features.

[0214] Step S130 includes:

[0215] Step S131 , aggregating the traffic perception features corresponding to the target traffic participant subject member and the fusion representation features of the connected members of each level connected corresponding to the target traffic participant subject member to obtain a first aggregate representation feature.

[0216] Step S132 , aggregating the traffic perception features corresponding to the target traffic path member and the fusion representation features of the connected members of each level connected corresponding to the target traffic path member to obtain a second aggregate representation feature.

[0217] Step S133, the traffic perception features corresponding to the target traffic participant main member and the fusion representation features of the connected members at each level connected to the target traffic participant main member are respectively fused with the traffic perception features corresponding to the target traffic path member and the fusion representation features of the connected members at each level connected to the target traffic path member, and then converged to obtain a third converged representation feature.

[0218] Step S134: Aggregate the first converged representation feature, the second converged representation feature, and the third converged representation feature to obtain a fourth converged representation feature.

[0219] Step S135: Use a fully connected mapping unit to perform fully connected mapping on the fourth converged representation feature to obtain the predicted traffic congestion rate.

[0220] In this embodiment, it is further assumed that the target traffic participant member is a specific bus numbered 050, and the target traffic path member is a city trunk road numbered 100.

[0221] First, it is explained that the connected member fusion representation features corresponding to the target traffic participant member (bus 050) and the connected member fusion representation features corresponding to the target traffic path member (main road 100) both include multi-level connected connected member fusion representation features.

[0222] For bus 050, its first-level connected members may be other buses running on the same route, its second-level connected members may be related feeder buses, and its third-level connected members may be other bus routes that frequently intersect with this route.

[0223] For the main road 100, its first-level connected members may be other roads directly connected to it, its second-level connected members may be similar main roads in the nearby area, and its third-level connected members may be the main types of vehicles traveling through these connected roads.

[0224] Next, we will predict the traffic congestion rate.

[0225] The server aggregates the traffic perception features corresponding to bus 050, such as the current position at coordinates (X50, Y50), speed of 40 km / h, and passenger capacity of 50 people, and the fusion representation features of the connected members at each level corresponding to bus 050 to obtain the first aggregated representation feature.

[0226] Assume that the fusion representation characteristics of the first-level connected members of bus 050 are an average speed of 38 km / h and an average passenger capacity of 45 people; the fusion representation characteristics of the second-level connected members are that the departure frequency of the feeder bus is one every 20 minutes; and the fusion representation characteristics of the third-level connected members are that the traffic flow of the intersection line is 80 vehicles / hour during peak hours.

[0227] The server aggregates the fusion representation features of these connected members at different levels with the traffic perception features of bus 050. For example, the first aggregate representation feature is obtained by comprehensively calculating the average speed, considering the impact of passenger capacity at different levels, and incorporating factors such as departure frequency and traffic volume.

[0228] The server then aggregates the traffic perception features corresponding to the main road 100, such as the current traffic volume of 600 vehicles / hour and the good road conditions, and the fusion representation features of the connected members at each level corresponding to the main road 100 to obtain a second aggregate representation feature.

[0229] Assume that the fusion representation feature of the first-level connected members of main road 100 is that the traffic volume of the connected roads during the morning rush hour is 500 vehicles / hour; the fusion representation feature of the second-level connected members is that the average traffic volume of similar nearby main roads during the same period is 550 vehicles / hour; the fusion representation feature of the third-level connected members is that the main type of traveling vehicles is private cars and the average speed is 45 kilometers / hour.

[0230] The server aggregates these different-level features with the traffic perception features of the main road 100 itself, just like processing bus 050, to obtain a second aggregated representation feature.

[0231] Then, the server merges the traffic perception features corresponding to bus 050 and the fused representation features of connected members at each level with the traffic perception features corresponding to main road 100 and the fused representation features of connected members at each level to obtain a third merged representation feature.

[0232] In this fusion process, the server fully considers the relationship between bus 050 and main road 100 in terms of time, space, and traffic flow. For example, the server compares the estimated arrival time of bus 050 on main road 100 with the peak traffic time of main road 100, and considers the relationship between the route of bus 050 and the roads connected to main road 100. Through complex calculations and analysis, these interrelated features are fused and converged to obtain the third converged representation feature.

[0233] Next, the server aggregates the first aggregate representation feature, the second aggregate representation feature, and the third aggregate representation feature to obtain a fourth aggregate representation feature.

[0234] This aggregation process is a further integration of the various feature representations obtained previously. The server may use weighted average, principal component analysis and other methods to integrate these features to more comprehensively reflect the overall traffic conditions of bus 050 traveling to main road 100.

[0235] Finally, the server uses a fully connected mapping unit to perform fully connected mapping on the fourth aggregated representation feature to obtain a predicted traffic congestion rate.

[0236] After receiving the fourth converged representation feature, the fully connected mapping unit calculates it through a complex internal neural network. It comprehensively considers the real-time status of bus 050, its relationship with the connected members at each level, the current status of the main road 100, and the interaction between the two. After a series of mathematical operations and logical judgments, it finally outputs a value between 0 and 1 as the predicted traffic congestion rate.

[0237] Assume that the output value is 0.3, which means that when bus 050 travels to the main road 100, the possibility of encountering congestion is low and the traffic is relatively smooth.

[0238] For another example, in another prediction, the server performs the same process for another bus 200 and another main road 300 .

[0239] For bus 200, its traffic perception features are that its current position is at coordinates (X200, Y200), its speed is 30 km / h, and its passenger capacity is 40. The fusion representation features of its first-level connected members are the average speed of 28 km / h and the average passenger capacity of 35; the fusion representation features of the second-level connected members are that the frequency of branch bus departures is every 30 minutes; and the fusion representation features of the third-level connected members are that the traffic volume of the intersection line is 70 vehicles / hour during peak hours.

[0240] The traffic perception features of the main road 300 are that the current traffic volume is 700 vehicles / hour and the road is undergoing partial maintenance. The fusion representation features of its first-level connected members are that the traffic volume of the connected roads during the morning peak is 600 vehicles / hour; the fusion representation features of the second-level connected members are that the average traffic volume of similar main roads nearby during the same period is 650 vehicles / hour; the fusion representation features of the third-level connected members are that the main type of vehicles is trucks and the average speed is 35 kilometers / hour.

[0241] The server performs feature aggregation and fusion in sequence according to the steps described above to obtain first, second, third and fourth aggregated representation features.

[0242] The fully connected mapping unit processes the fourth pooled representation feature and finally outputs a predicted traffic congestion rate of 0.7, which indicates that bus 200 is likely to encounter congestion when it travels to the main road 300.

[0243] In this way, accurate traffic congestion rate prediction can be provided for each group of target traffic participants and target traffic path members, providing powerful decision-making support for traffic management and planning.

[0244] There are many similar examples, such as bus 080 and main road 150, bus 120 and main road 220, etc. The servers will follow the same rigorous process and complex calculations to provide accurate data prediction and analysis for the optimization and improvement of urban transportation.

[0245] Figure 2 The hardware structure of the AI-based urban road management system 100 provided in the embodiment of the present application for implementing the above-mentioned AI-based urban road management method is shown as follows: Figure 2As shown, the AI-based urban road management system 100 may include a processor 110 , a machine-readable storage medium 120 , a bus 130 , and a communication unit 140 .

[0246] In one possible design, the AI-based urban road management system 100 can be a single server or a server group. The server group can be centralized or distributed (for example, the AI-based urban road management system 100 can be a distributed system). In some embodiments, the AI-based urban road management system 100 can be local or remote. For example, the AI-based urban road management system 100 can access information and / or data stored in a machine-readable storage medium 120 via a network. For another example, the AI-based urban road management system 100 can be directly connected to the machine-readable storage medium 120 to access the stored information and / or data. In some embodiments, the AI-based urban road management system 100 can be implemented on an AI-based urban road management system. By way of example only, the AI-based urban road management system can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc., or any combination thereof.

[0247] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data acquired from an external terminal. In some embodiments, the machine-readable storage medium 120 may store data and / or instructions that the AI-based urban road management system 100 uses to execute or use to complete the exemplary methods described in this application.

[0248] During the specific implementation process, one or more processors 110 execute computer executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the AI-based urban road management method of the above method embodiment. The processor 110, the machine-readable storage medium 120 and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the sending and receiving actions of the communication unit 140.

[0249] The specific implementation process of the processor 110 can be found in the various method embodiments executed by the above-mentioned AI-based urban road management system 100. The implementation principles and technical effects are similar and will not be repeated in this embodiment.

[0250] In addition, an embodiment of the present application also provides a readable storage medium, in which computer executable instructions are preset. When the processor executes the computer executable instructions, the above-mentioned AI-based urban road management method is implemented.

[0251] It should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof. Similarly, it should be noted that in order to simplify the description disclosed in this application and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present application, multiple features are sometimes combined into one embodiment, drawings, or descriptions thereof.

Claims

1. An AI-based urban road management method, characterized in that: The method comprises: Acquire real-time holographic traffic perception data, generate an urban traffic network knowledge structure based on traffic participants and traffic paths in the real-time holographic traffic perception data, and determine traffic participant clusters and traffic path clusters based on the urban traffic network knowledge structure; Determining, based on the traffic participant cluster and the traffic path cluster, a connected member fusion representation feature corresponding to a target traffic participant member and a connected member fusion representation feature corresponding to a target traffic path member; Based on the traffic perception characteristics corresponding to the target traffic participant member, the connected member fusion representation characteristics corresponding to the target traffic participant member, the traffic perception characteristics corresponding to the target traffic path member, and the connected member fusion representation characteristics corresponding to the target traffic path member, a traffic congestion rate prediction is performed to obtain a predicted traffic congestion rate, wherein the predicted traffic congestion rate is a traffic congestion rate prediction result of the target traffic participant corresponding to the target traffic participant member traveling to the target traffic path corresponding to the target traffic path member; The connected member fusion representation features corresponding to the target traffic participant member and the connected member fusion representation features corresponding to the target traffic path member both include multi-level connected connected member fusion representation features; The method of predicting a traffic congestion rate based on the traffic perception features corresponding to the target traffic participant subject members, the connected member fusion representation features corresponding to the target traffic participant subject members, the traffic perception features corresponding to the target traffic path members, and the connected member fusion representation features corresponding to the target traffic path members to obtain a predicted traffic congestion rate includes: Aggregating the traffic perception features corresponding to the target traffic participant subject member and the fusion representation features of the connected members of each level connected corresponding to the target traffic participant subject member to obtain a first aggregate representation feature; Aggregating the traffic perception features corresponding to the target traffic path member and the fusion representation features of the connected members of each level connected corresponding to the target traffic path member to obtain a second aggregate representation feature; The traffic perception features corresponding to the target traffic participant subject members and the fusion representation features of the connected members at each level connected to the target traffic participant subject members are respectively fused with the traffic perception features corresponding to the target traffic path members and the fusion representation features of the connected members at each level connected to the target traffic path members, and then converged to obtain a third converged representation feature; Aggregating the first converged representation feature, the second converged representation feature, and the third converged representation feature to obtain a fourth converged representation feature; A fully connected mapping unit is used to perform fully connected mapping on the fourth converged representation feature to obtain the predicted traffic congestion rate.

2. The AI-based urban road management method according to claim 1 is characterized in that: The generating of the urban traffic network knowledge structure based on the traffic participants and traffic paths in the real-time holographic traffic perception data includes: The traffic participant is used as a dynamic member identifier, and the traffic path is used as a static member identifier; A traffic relationship interaction network is generated based on the dynamic member identifier, the static member identifier, and the traffic relationship between the traffic participant and the traffic path, and the traffic relationship interaction network is used as the urban traffic network knowledge structure.

3. The AI-based urban road management method according to claim 2 is characterized in that: The step of generating a transit relationship interaction network based on the dynamic member identifier, the static member identifier, and the transit relationship between the traffic participant and the traffic path, and using the transit relationship interaction network as the urban traffic network knowledge structure includes: Extracting traffic participant subject features of the traffic participant subject and traffic path features of the traffic path, wherein the traffic participant subject features include movement trajectory features, behavior features, and category features, and the traffic path features include geometric features, attribute features, and state features; Extracting the traffic relationship characteristics between the traffic participant and the traffic path based on the traffic participant characteristics of the traffic participant and the traffic path characteristics of the traffic path, wherein the traffic relationship characteristics include interaction frequency, interaction duration, and interaction mode; The traffic participant is used as a dynamic member identifier, and the traffic path is used as a static member identifier, and each dynamic member and static member has a unique identifier; Defining the interaction relationship between the traffic participant and the traffic path based on the traffic relationship characteristics, wherein the interaction relationship includes a spatial proximity interaction relationship, a temporal synchronization interaction relationship, and a behavior similarity interaction relationship; Based on the dynamic member identifier, the static member identifier, and the interactive relationship between the traffic participant and the traffic path, a traffic relationship interactive network with a bipartite graph structure is constructed, in which the dynamic members and the static members are different types of nodes, the interactive relationship between the dynamic members and the static members is represented by an edge, and the weight of the edge represents the strength or frequency of the interactive relationship; Optimizing the traffic relationship interaction network, the optimization steps comprising: removing isolated nodes and edges, merging duplicate nodes and edges, and adjusting edge weights according to the strength and importance of the interaction relationship; After constructing and optimizing the transit relationship interaction network, the network indicators of the transit relationship interaction network are calculated and added to the transit relationship interaction network for representation. The network indicators include: node degree, clustering coefficient, path length and network diameter: representing the maximum distance between any two nodes in the network.

4. The AI-based urban road management method according to claim 1 is characterized in that: The determining of the traffic participant cluster and the traffic path cluster based on the urban traffic network knowledge structure includes: Clustering the urban traffic network knowledge structure to obtain traffic interaction event clusters corresponding to the urban traffic network knowledge structure; The traffic participant and the traffic path in each traffic interaction event in the traffic interaction event clustering are decomposed to obtain a plurality of traffic participant clusters and a plurality of traffic path clusters.

5. The AI-based urban road management method according to claim 4 is characterized in that: The clustering of the urban traffic network knowledge structure to obtain traffic interaction event clusters corresponding to the urban traffic network knowledge structure includes: When the urban traffic network knowledge structure is a traffic relationship interaction network, the traffic relationship interaction network is mapped into a traffic binary division structure diagram, and the traffic participants and traffic paths in the traffic binary division structure diagram are iteratively clustered by using a set clustering strategy to obtain the traffic interaction event clusters; When the urban traffic network knowledge structure is a traffic binary division structure diagram, traffic participants and traffic paths in the traffic binary division structure diagram are iteratively clustered according to the set clustering strategy to obtain the traffic interaction event clusters.

6. The AI-based urban road management method according to claim 5 is characterized in that: The iterative clustering of the traffic participants and traffic paths in the traffic binary division structure diagram by using a set clustering strategy includes: According to the graph parameters of the traffic binary division structure graph, the number K of iterative clustering is pre-set, and K traffic participants or traffic paths are randomly selected as initial cluster centers; Based on the spatial weight matrix and spatiotemporal dynamic characteristics of each traffic participant or traffic path, each traffic participant or traffic path is assigned to the sub-cluster where the nearest cluster center is located, and then the average position or characteristics of all members in each sub-cluster are recalculated as the new cluster center. Then, the step of assigning each traffic participant or traffic path to the sub-cluster where the nearest cluster center is located is returned to execute until the new cluster center no longer changes or the preset number of iterations is reached, and the initial clustering result is obtained; Calculate the density of each initial cluster in the initial clustering result, and obtain candidate clusters or partial regions of candidate clusters whose density is lower than a set density threshold; For the candidate cluster or a partial region of a candidate cluster, if the density of an adjacent cluster is greater than a set density threshold, the candidate cluster or the partial region of the candidate cluster is merged into the adjacent cluster; if the feature distance inside the candidate cluster or the partial region of the candidate cluster is greater than a set distance, the candidate cluster or the partial region of the candidate cluster is split into multiple candidate subclusters, and the center position of the optimized cluster is recalculated to generate a reference cluster sequence after iterative optimization is completed; Calculate the silhouette coefficient of the reference clustering sequence after the iterative optimization is completed. The silhouette coefficient is an indicator for measuring the clustering effect. The value range of the silhouette coefficient is between -1 and 1. The closer the silhouette coefficient is to 1, the better the clustering effect is; the closer it is to -1, the worse the clustering effect is. The calculation formula of the silhouette coefficient is: s(i)=(b(i)-a(i)) / max{a(i), b(i)}, where s(i) is the silhouette coefficient of the i-th sample in any reference cluster, a(i) is the average distance between the i-th sample and other samples in the reference cluster to which it belongs, and b(i) is the minimum value of the average distance between the i-th sample and all samples in other reference clusters; When the calculated silhouette coefficient meets the set condition, the traffic interaction event cluster is output.

7. The AI-based urban road management method according to claim 1 is characterized in that: The method of determining the connected member fusion representation features corresponding to the target traffic participant member and the connected member fusion representation features corresponding to the target traffic path member based on the traffic participant cluster and the traffic path cluster includes: Acquire the connected member behavior characteristics corresponding to the connected members of the target traffic participant main member, determine the traffic participation fusion coefficient corresponding to the target traffic participant main member based on the connected member behavior characteristics, and determine the connected member fusion representation characteristics corresponding to the target traffic participant main member based on the traffic participation fusion coefficient corresponding to the target traffic participant main member; The connected member behavior characteristics corresponding to the connected members of the target traffic path member are obtained, the traffic participation fusion coefficient corresponding to the target traffic path member is determined based on the connected member behavior characteristics, and the connected member fusion representation characteristics corresponding to the target traffic path member are determined based on the traffic participation fusion coefficient corresponding to the target traffic path member.

8. The AI-based urban road management method according to claim 7 is characterized in that: The step of obtaining the connected member behavior characteristics corresponding to the connected members of the target traffic participant main member, determining the traffic participation fusion coefficient corresponding to the target traffic participant main member based on the connected member behavior characteristics, and determining the connected member fusion representation characteristics corresponding to the target traffic participant main member based on the traffic participation fusion coefficient corresponding to the target traffic participant main member includes: Obtain the cluster where the x-th level connected member corresponding to the target traffic participant member is located, poll the clusters, and use any one of the clusters as the target cluster; the cluster is the traffic participant cluster or the traffic path cluster; fusing the connected member behavior features of the x-th level connected members associated with the target traffic participant member and in the target cluster to obtain the fused behavior features corresponding to the target traffic participant member; Determining a traffic participation fusion coefficient corresponding to the target traffic participant and the target cluster according to the traffic perception characteristics corresponding to the target traffic participant and the fusion behavior characteristics corresponding to the target traffic participant; The fusion behavior characteristics and traffic participation fusion coefficients corresponding to the target traffic participant subject are weighted to obtain the cluster representation characteristics corresponding to the target cluster, and after the polling is completed, the cluster representation characteristics corresponding to all the clusters are aggregated to obtain the connected member fusion representation characteristics corresponding to the target traffic participant subject member; And, the acquiring of the connected member behavior characteristics corresponding to the connected members of the target traffic path member, determining the traffic participation fusion coefficient corresponding to the target traffic path member based on the connected member behavior characteristics, and determining the connected member fusion representation characteristics corresponding to the target traffic path member based on the traffic participation fusion coefficient corresponding to the target traffic path member, includes: Obtain the cluster where the x-th level connected member corresponding to the target traffic path member is located, poll the clusters, and use any one of the clusters as the target cluster; the cluster is the traffic participant cluster or the traffic path cluster; fusing the connected member behavior features of the x-th level connected members associated with the target traffic path member and in the target cluster to obtain a fused behavior feature corresponding to the target traffic path member; Determining a traffic participation fusion coefficient corresponding to the target traffic path member and the target cluster according to the traffic perception characteristics corresponding to the target traffic path member and the fusion behavior characteristics corresponding to the target traffic path member; The fusion behavior characteristics and traffic participation fusion coefficients corresponding to the target traffic path members are weighted to obtain the cluster representation characteristics corresponding to the target cluster. After the polling is completed, the cluster representation characteristics corresponding to all the clusters are aggregated to obtain the connected member fusion representation characteristics corresponding to the target traffic path members.

9. An AI-based urban road management system, characterized in that: The AI-based urban road management system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the AI-based urban road management method described in any one of claims 1 to 8 above.

Citation Information

Patent Citations

  • Traffic operation state prediction method and system for congested area

    CN112382082A

  • Traffic flow prediction method based on dynamic graph neural network

    CN114120652A

  • Traffic prediction method based on multi-scale space-time fusion graph network

    CN115240425A

  • Graph2vec and double-layer LSTM fused vehicle travel time prediction method and system

    CN115565376A

  • Smart city traffic management method and system based on multi-source data fusion

    CN117912251A