Abstract: To meet the real-time demand for passenger flow control and train dispatching in urban rail transit operation organizations, an online travel destination prediction method based on passenger origin destination(OD) spatiotemporal characteristics is proposed. First, the passenger OD spatiotemporal feature matrix is defined. Moreover, the automatic fare collection (AFC) history data of individual passengers are used as training samples, and an extraction method of passenger OD spatiotemporal features is proposed using stroke density clustering. Further, based on the matching rule analysis between the card-swiping information of passengers and their OD spatiotemporal feature matrix, the corresponding real-time destination prediction method is proposed for three matching cases. The AFC data of Nanjing rail transit are used to verify the proposed method. Results show that the proposed method is effective in the prediction accuracy of rush hours and provides stable all-day prediction, which can provide a reference for subway operations.

Key words: urban rail transit, AFC data, origin destination spatiotemporal feature, stroke clustering, destination prediction, naive Bayesian , machine learning