Intercity train dwell time estimation by using robust regression method: A study case on Surabaya-Yogyakarta line

Abstract

Dwell time is important an issue to optimise not only for train operation efficiency but also punctuality. This paper proposes a model based on a field survey of intercity trains along the Surabaya-Yogyakarta line by using a robust regression analysis of 138 observations from 13 trainsets operating on the line. Assuming no differences in door or platform width, a model has been formulated to include the number of people who are alighting and boarding, the type of platform and the number of train doors on one side. There were three robust estimators applied to the model. With a 2.037% mean absolute percentage error, the robust regression using the M estimator formed the dwell time model for this study. The results show that, between the dummy variable for platform type using a binary number (0 or 1), dummy variables for high-level platforms and mid-level platforms have opposing tendencies. High-level platform use reduces dwell time by 2.014 sec based on continuous variables; in contrast, dwell time increases by 9.353 sec based on continuous variables by using mid-level platforms. Furthermore, people who are alighting need less time than those boarding. Dwell time will increase by 1.222 sec per alighting passenger and 1.314 per boarding passenger, while it will decrease by 2.727 sec per door on a one-sided trainset. Moreover, dwell time estimates with the model have reduced trainset travel time.