An Urban Topological Map Generation Method for Traffic Flow Prediction Based on Road Segment Clustering with Floating Vehicle Trajectory Dataset

Geographic division directly affects regional traffic flow prediction result. Classical methods like latitude-longitude meshing and POIs (point of interest) clustering can’t effectively describe traffic transfer between regions. In this study, a topology map is generated by road segment clustering. Several road segments in a network are the basic clustering units. The adjacency, connectivity, and congestion between road segments and the congestion degree of road segments are made as attributes of the cluster similarity measurement. Road segments within urban areas are clustered by spectral clustering algorithm. Topological map composes converged clustering centers and edges made of connection ports among the roads of each node. The experiment data consists of 120,000 road segments and vehicles’ trajectory dataset in Beijing. Topological maps are generated at different granularities. The results show that the topological map has a better visualization and the amount of traffic flow information is superior to POIs clustering and meshing.