Study On Urban Traffic State Prediction Method Based On Graph Neural Network With High Frequency Sub-trajectories

With the gradual improvement of residents’ pursuit of travel comfort and flexibility,the number of urban motor vehicles has continued to grow,while making urban roads bear huge pressure of traffic congestion.Intelligent transportation system has become one of the main means to alleviate traffic congestion.The floating vehicle technology broadens the real-time traffic information collection channel of the intelligent transportation system and provides support for the realization of the traffic control and traffic flow guidance functions of the intelligent transportation system.In this paper,an association mining strategy is proposed to extract the high frequency sub-trajectoryof floating vehicles.Based on the fusion of graph neural network and spatiotemporal graph,a traffic condition prediction model is constructed for the key nodes in the high frequency sub-trajectoryof floating vehicles.In this paper,GPS data of floating cars in Beijing are used to predict traffic conditions.The content of this paper mainly includes the following four aspects:(I)The floating car data preprocessing and floating car track grid matching.In view of GPS data of floating cars in Beijing,invalid data were cleaned,effective trajectory data were extracted,vehicle type identification and coordinate conversion were carried out.On this basis,based on the coverage intensity and coverage,sample size verification and analysis of floating car data are carried out.(2)Determine the grid traffic state division method.The definition of traffic state and the current classification standard of road traffic state are analyzed and the average speed and density of floating vehicles in the grid area are selected as the classification parameters of grid traffic state.K-means algorithm and AP clustering algorithm are used to cluster the traffic state of the grid respectively.The distribution range of grid traffic state partition parameters is obtained by combining the clustering evaluation indexes and optimizing the clustering results.(3)Constructing spatio-temporal map structure of traffic grid based on association rule mining algorithm.The time sequence graph structure based on the time sequence dependence of traffic state was constructed,and the key spatial graph structure and key nodes of urban traffic grid were extracted by using Apriori and g Span association rule mining algorithms.According to the characteristics of floating vehicle trajectory data,several mining strategies for high frequency sub-trajectory of floating vehicle are proposed and an example is analyzed.(4)Establish the traffic state prediction model of urban grid based on graph neural network algorithm.A graph neural network algorithm is used to fuse the temporal and spatial maps based on the sequential graph structure and the key spatial graph structure based on the high frequency subtrajectory.The Graph SAGE algorithm framework is used to construct the urban grid traffic state prediction model,and the experimental results show that the key spatial Graph structure constructed by the combination mining strategy can significantly improve the prediction effect.In this paper,the traffic state prediction model of urban road key grid nodes based on floating vehicle high-frequency sub-trajectory is established,which provides a reference for traffic managers to reasonably reduce the number of monitoring equipment,reduce the cost of urban construction,and solve the actual traffic problems related to traffic congestion.