Detecting Traffic Anomalies During Extreme Events via a Temporal Self-Expressive Model

Motivated by rapid urbanization and increasing natural hazards, this study aims to develop a data-driven method for detecting urban traffic anomalies during extreme events. Past experiences have shown that abnormal traffic patterns caused by extreme events can disrupt traffic in a large portion of the road network. Timely and reliable traffic monitoring for detection of such anomalies is crucial for congestion mitigation and successful emergency operation plans. An effective traffic monitoring system should detect disruptions at both network and local levels. However, the existing methods are not capable of addressing this need. This study proposes a temporal self-expressive network monitoring method to achieve this purpose. The proposed method first utilizes a temporal self-expressive model to uncover the dynamic interdependencies between local zones of the traffic network. Next, a statistical monitoring method detects network-wide anomalies based on regular traffic interdependencies. Finally, the method identifies the zones most affected by the anomalous event. We applied the proposed method to the road network of Manhattan in New York City to evaluate its performance during Hurricane Sandy. The outcomes confirmed that the temporal self-expressive model, augmented with statistical monitoring tools, could accurately detect anomalous traffic at both network and zone levels.