Recently, surges of passengers caused by large gatherings, temporary traffic control measures, or other abnormal events have frequently occurred in metro systems. From the standpoint of the operation managers, the available information about these outside events is incomplete or delayed. Unlike regular peaks of commuting, those unforeseen surges pose great challenges to emergency organization and safety management. This study aims to assist managers in monitoring passenger flow in an intelligent manner so as to react promptly. Compared with the high cost of deploying multisensors, the widely adopted automated fare collection (AFC) system provides an economical solution for inflow monitoring from the application point of view. In this paper, a comprehensive framework for the early warning mechanism is established, including four major phases: data acquisition, preprocessing, off-line modeling, and on-line detection. For each station, passengers’ tapping-on records are gathered in real time, to be further transformed into a dynamic time series of inflow volumes. Then, a sequence decomposition model is formulated to highlight the anomaly by removing its inherent disturbances. Furthermore, a novel hybrid anomaly detection method is developed to monitor the variation of passenger flow, in which the features of inflow patterns are fully considered. The proposed method is tested by a numerical experiment, along with a real-world case study of Guangzhou metro. The results show that, for most cases, the response time for detection is within 5 min, which makes the surge phenomenon observable at an early stage and reminds managers to make interventions appropriately.
Likelihood‐free parameter estimation for dynamic queueing networks: Case study of passenger flow in an international airport terminal