Early detection of risky spatio-temporal congestion in urban traffic

Abstract Risky congestion in city is challenging for traffic reliability. Previous research has mainly focused on link-level congestion forecasting. However, large-scale congestion in city exhibits an unique spatio-temporal growth pattern. In this article, we develop a detection method for risky congestion based on its spatio-temporal evolution feature, which can detect risky spatio-temporal congestion clusters (SCCs) when they are small. The evolution of SCCs is modelled as Markov stochastic process. A risk signal of SCC is developed based on the deviation from normal patterns. Early diagnosis of risky congestion is achieved by risk signal distribution estimation and Bayesian inference. The proposed method is verified using real traffic data of Beijing. The proposed method can support risky congestion control in traffic reliability management.