Early warning of burst passenger flow in public transportation system

Abstract

Burst passenger flow in the public transportation system is serious to public safety. Existing works mainly focused on prediction and monitoring of regular passenger flow, which are not suitable for burst passenger flow. In this article, we first formulate the problem as early warning of burst passenger flow. Next, we design a novel framework to solve this problem by our observation that a burst passenger in-flow usually comes after an abnormal passenger out-flow for a subway station, especially when there is a large-scale social crowd event. Our framework consists of two models: (1) Abnormal out-flow detection (AOFD) which detects abnormal out-flows and warns the city administration of the burst in-flow fairly ahead of time. (2) Burst in-flow peak estimation (BIFPE) which estimates burst in-flow peak time and volume. We evaluate our framework with real-world smartcard data of the largest city in China and use large-scale social crowd event data to further explain our model. The result shows that: (1) AOFD can detect abnormal out-flows that would later result in bursts in-flows with better performance and can send warning signal ahead of the time of burst passenger in-flow. (2) BIFPE can effectively estimate the peak time of burst in-flow and can reduce peak volume estimation error compared with the traditional passenger flow prediction models.