Real-Time Passenger Flow Anomaly Detection Considering Typical Time Series Clustered Characteristics at Metro Stations

Abstract

Real-time anomaly detection at metro stations is a very important task with considerable implications for massive passenger flow organization and train timetable rescheduling. State-of-the-art studies tend to conduct passenger flow anomaly detection; however, they fail to provide more detailed analysis of anomaly combination at metro stations. The primary motivation of this study is to develop a systematic approach to identifying the nature of passenger flow anomalies and estimating their alarm levels dynamically. Firstly, a K-means algorithm combined with hierarchical clustering is used to extract incrementally updated typical clustered features. Secondly, anomaly detection indexes that contain both mutant and migration anomalies are designed to identify the time and category of passenger flow anomalies. Then, coordinated anomaly thresholds and corresponding alarm level are listed considering active safety management and passenger travel efficiency. Finally, one of the busiest stations in the Shanghai, China, metro network is selected to demonstrate the proposed method. Application results indicate that these real-time anomalies can be detected both efficiently and accurately in changing passenger flow conditions. The insightful features extracted and fast online computation ensure that the detection results can be applied to assist real-time decision making in prewarning management and optimizing passenger flow organization strategies.