Estimation Of Congestion Duration On Urban Rail Transit

Management of passenger flow congestion is an important part of the urban rail transit (URT) operation and management. In order to solve the problem of inadequate research on accuracy, effectiveness, and real-time of decision support data for current passenger flow congestion management, this paper focus on the research of the regularity and estimation model of URT passenger flow congestion duration, based on the URT historical data and passenger flow feature data collected.For the URT passenger flow congestion duration estimation, the method combining decision tree and survival analysis is chosen. The chi-square automatic interaction detection (CHAID) algorithm is adopted to establish a classification tree which sorts the passenger flow congestion events into three categories of weekday morning rush period, weekday non-morning rush period, and non-weekday. Then according to the Akaike information criteria (AIC), the distributions of Logistic, Weibull and Weibull are applied to build three hazard-based congestion duration estimation models for these categories. In addition, referencing the idea of the minimizing risk decision rules in Bayesian classifier, a correction method is proposed to adjust the prediction results of duration estimation model based on minimizing risk of lossAccording to the evaluation results of the duration estimation model based on "decision tree+ hazard" established by this paper, the results of weekday morning rush and weekday non-morning rush are of high reliability, but the result of non-weekday is not satisfactory. As for the whole model, the estimation performance is superior according to the mean absolute percentage error (MAPE) and the estimation effect achieves a significantly improvement compared to the duration estimation model based on single survival analysis theory. And the results of test for correction method show that this method can reduce the loss caused by the forecast error to some extent.