Partitioning urban road network based on travel speed correlation

Abstract

• Massive taxi GPS data is used in capturing city-wide travel speed correlation. • Spatial analysis on travel speed correlation berween road section is conducted. • Road network partition method based on traffic speed correlation is proposed. • Methodology can yield evenly distributed and spatially compact road partitions. • Methodology can handle citywide road network with a short computing time. Urban traffic management is increasingly critical in the future to ensure the livability, efficiency, and sustainability of the city. Urban road network partition is a fundamental step in traffic management and it directly affects the effectiveness of the overall traffic management system. In the existing road network partition algorithm, the spatial relationship of road sections are introduced for generating spatially compact partitions. However, this way of consideration fails to capture the travel speed correlation between road sections with far distance. To solve this problem, this paper proposed a road network partition method base on community detection. Travel speed correlation between road sections is captured by self-expressiveness. On the graph constructed by travel speed correlation, fast unfolding method is used to divide urban road network into sub-partitions of densely correlated road sections. A case study is conducted by using taxi GPS dataset in Shanghai. The case study examines the travel speed correlation for specific road section, which shows that the travel speed will generate high correlation even if the road sections are not spatially connected or close. The fast unfolding algorithm divides the road network in Shanghai into 77 sub-partitions with strong intro-correlation of travel speed pattern. Comparing the result with Ncut algorithm with different spatial constraints, the method proposed can consider travel speed correlation between every two road sections and generate evenly distributed and spatially compact sub-partitions.