Network-Level Travel Time Reliability Analysis via Network Partitioning with Community Detection Algorithms

Travel time reliability (TTR) has been widely used as an important evaluation index for traffic performance evaluation. On the network level, the existing analyses tend to simplify the network as a homogeneous one, i.e., ignoring the situation that the traffic states in different regions exist in difference. It may lead to errors in the evaluation of network-level TTR. To this end, the community detection algorithm is applied to help divide the network into several homogeneous clusters, for which TTR could be investigated more sophisticatedly. Concretely, the network is initially partitioned by Fast-Unfolding algorithm, and then the clustering results are enhanced by the Newman fast algorithm. Experiments based on real network traffic data from automatic number plate recognition (ANPR) demonstrate that the proposed method enables to provide of more accurate partition results of the heterogeneous network. The essential heterogeneity in network traffic states can significantly affect the evaluation of the network-level TTR.