Abnormal Situation Simulation and Dynamic Causality Discovery in Urban Traffic Networks (Short Paper)

Yadi Wang, Yicheng Pan, Meng Ma, Ping Wang

Abstract: Various participants in urban traffic systems intertwine a highly complicated coupling network. An interpretable analysis of underlying correlations is one of the keys to understanding traffic anomalies. Unfortunately, abnormal situation analysis in real scenarios faces severe limitations in negative sample deficiency, data integrity, and verifiability. In view of this, we developed a simulation tool - the Traffic Anomaly Situation Simulator (TASS). Through configurable scripts, TASS simulates real traffic networks by road editing, data collection, and fault injection. Given the generated cases, we designed a dynamic causal discovery algorithm, Dycause-Traffic, to demonstrate the features of causality in traffic anomalies.