Short-term passenger flow prediction for urban rail systems: A deep learning approach utilizing multi-source big data

Abstract Predicting short-term passenger flow in urban rail transit is crucial for intelligent and real-time management of urban rail systems. This study utilizes deep learning techniques and multi-source big data to develop an enhanced spatial-temporal long short-term memory (ST-LSTM) model for forecasting subway passenger flow. The model includes three key components: (1) a temporal correlation learning module that captures travel patterns across stations, aiding in the selection of effective training data; (2) a spatial correlation learning module that extracts spatial correlations between stations using geographic information and passenger flow variations, providing an interpretable method for quantifying these correlations; and (3) a fusion module that integrates historical spatial-temporal features with real-time data to accurately predict passenger flow. Additionally, we discuss the model's interpretability. The ST-LSTM model is evaluated with two large-scale real-world subway datasets from Nanjing and Beijing. Experimental results show that the ST-LSTM model effectively captures spatial-temporal correlations and significantly outperforms other benchmark methods.