Attention-based Graph Neural Network Enabled Method to Predict Short-term Metro Passenger Flow

Effective, accurate, and reliable prediction of short-term metro passenger flow is essential to improving the operational efficiency and passenger travel experience of public transport, as well as enhancing the stakeholder emergency response capability against adverse events. Various deep learning models like the long short-term memory (LSTM) models and the graph convolutional network (GCN) have been implemented to predict short-term metro passenger flow, despite the fact that they are either computationally expensive or less accurate. To strike a balance between computational cost efficiency and accuracy concurrently, this study proposes to consider only adjacent stations and apply an attention-based graph neural network (AGNN) approach to short-term metro passenger flow prediction. The proposed method can effectively improve prediction accuracy compared to the LSTM and GCN based models with a less computational cost. Empirical studies are conducted to validate the proposed method.

I. Introduction

Metro (or subway) is an important public social infrastructure and an important means for people to travel. The study of the urban metro system's short-term passenger flow prediction problem can help optimize the metro system's real-time scheduling, alleviate urban congestion, and improve passengers' travel experience.