Bayesian Network Inference on Departure Time Choice Behavior

摘要

Departure time choice behavior plays an important role in travel decision for metro passengers during morning peak hours. Different from statistical models, this paper proposed Bayesian networks (BNs) to model the departure time choices of metro passengers. The structure of BNs is learned through K2 algorithm and its parameters are estimated by maximum likelihood estimation (MLE) method using the stated preference (SP) survey data. Main results are obtained as follows: (1) passengers can accept departure earlier than usual in the range of 0-20 min; (2) passengers will prefer to choose departure earlier if they enjoy a 20% or more discount on metro fare; and (3) passengers are willing to departure at usual time with slight crowding while they prefer to departure earlier under serious crowding. These findings contribute to making strategies for passenger flow control and safety operation for metro stations.