Short-Term Inbound and Outbound Passenger Flow Prediction for New Metro Stations Based on Clustering and Deep Learning

Abstract

The rapid expansion of metro networks, e.g., in many cities of China, continuously introduces the operation of new stations every year. Due to the lack of historical data and complicate variations of short-term passenger flow in the early stage of operation, it is difficult to accurately predict inbound and outbound passenger flows of new metro stations in the short term, which would be the database for train scheduling for new stations before operation, dynamic capacity optimization for new stations under operation, short-term prediction of cycle sharing demands near new stations, and so on. Traditional methods usually failed to exactly reflect the complicate rules or were unusable without the new station's historical data. In order to solve the above problems, this paper proposes a short-term inbound and outbound passenger flow prediction model for new metro stations at the early stage of operation by combining the K-means clustering algorithm, an improved spatiotemporal long short-term memory model (Sp-LSTM), and a real-time feedback error model (mean absolute error, MAE), where passenger flows' spatial-temporal characteristics and land-use relevance are considered. The application in Guangzhou Metro, China, where Line 21 is regarded as a new line, shows that the proposed K-Sp-LSTM model has the best prediction accuracy compared with traditional methods.