Dynamic estimation of Origin-Destination passenger flow for large-scale urban rail transit network based on improved particle filter algorithm

Accurate and dynamic estimation of Origin-Destination (OD) passenger flow is essential for rail transit agencies to efficiently monitor network status and promptly adjust transportation organization strategies effectively. However, in large-scale urban rail transit networks, the OD matrix typically exhibits both high dimensionality and sparsity, which renders accurate short-term estimation of the global OD matrix via deep learning–based methods extremely challenging. This paper establishes a matrix-form state space model for OD estimation by considering the dynamic conservation relationship of passenger flow. To avoid the high dimensionality and sparsity of data, an improved particle filter (IPF) algorithm, incorporating historical data augmentation and importance pre-sampling, is proposed to construct a dynamic short-term OD estimation model that accounts for both real-time and historical trends. The model is implemented with the automated fare collection (AFC) data from Nanjing rail transit, China, encompassing 159 stations. Results indicate that the IPF-based dynamic OD passenger flow estimation model outperforms benchmark models, including the historical value estimation method (HIS) and the Kalman filter (KF), as evaluated by three error indicators: root means square error (RMSE), weighted absolute percentage error (WAPE), quantile absolute error (QAE). The model demonstrates excellent prediction accuracy and operational performance, making it a practical and effective model for estimating dynamic OD passenger flow in large-scale urban rail transit systems.