Subway Passenger Flow Forecasting With Multi-Station and External Factors

Abstract

With the rapid development of urban rail transit, more and more people choose to travel by subway. Therefore, accurate passenger flow forecasting is of great significance for passengers and municipal construction and contributes to smart city services. In this paper, we propose a multi-type attention-based network to forecast the subway passenger flow with multi-station and external factors. The proposed network has different types of attention mechanisms to adaptively extract relevant features, including multi-station, external factors, and historical data. In addition, the hierarchical attention mechanism is used to model the hierarchical relationship between subway lines and stations. In addition, the embedding method is applied to better combine the different kinds of data. The experiments on real subway passenger flow data in a city in China demonstrate that our method outperforms five baseline methods. Moreover, our method can visualize the impact of different stations and other factors on traffic, which plays an important role in passenger travel and subway dispatch.

Keywords

• Passenger flow forecasting
• attention mechanism
• recurrent neural networks