Uncovering Regular Travel Pattern for Metro Passenger Through Graph Neural Network

Abstract

This study tackles the increasing complexity of passenger travel patterns in urban rail transit systems by utilizing Automated Fare Collection (AFC) data and an enhanced GraphSAGE model. We construct detailed spatiotemporal travel graphs and introduce a weighted aggregation strategy to capture intricate network relationships. By clustering passengers into four distinct groups based on their temporal and spatial travel behaviors, we successfully identify regular passengers and other key travel patterns. A case study conducted at Anting Station in Shanghai demonstrates the method’s effectiveness in real-world applications. Our approach not only advances theoretical understanding of passenger behavior but also provides practical insights for metro operators, helping optimize operational efficiency and service quality in urban transit systems.