Identifying Traffic Bottleneck in Urban Road Networks via Causal Inference

Abstract

Identifying traffic bottlenecks and estimating their impacts on congestion propagation is a critical component of the Intelligent Transportation System (ITS). However, it is very challenging to identify urban traffic bottlenecks since they are caused by many complicated factors that are difficult to be pre-defined in urban road networks. In the paper, the authors propose a novel method to identify urban traffic bottlenecks via causal inference. The method mainly includes two parts: (1) Model causal relationships among traffic flow data from spatially distributed sensors. We firstly extract the uptrend intervals of traffic flow sensors, then calculate the causal strengths among spatially distributed sensors using transfer entropy; (2) Construct causality graphs and perform frequent subgraphs mining to identify traffic bottlenecks. We use a real-life traffic flow dataset from 74 loop detector sensors, which is collected in the urban traffic network of Hangzhou, china over one month. Our experimental results demonstrate significant findings regarding traffic bottlenecks and congestion propagation in Hangzhou, which will be useful for governments to make policy and govern traffic congestions.