Accurate identification of the train delays causes and estimation of the corresponding delays are essential to improve the train real-time dispatching under abnormal events. In this paper, we propose an ensemble prediction model for delay analysis under abnormal events. First, according to disruption and timetable characteristics, a FCM clustering algorithm was used to classify delayed trains into three scenarios. Then, an ensemble prediction model which includes Extreme Learning Machines (ELM), Random Forest (RF) and support vector machine (SVR) is developed that can capture the relation between train delays and various characteristics of a railway system in difference scenarios. Further, delayed train number, station code, speed limit, scheduled time of arrival at a station, time travelled, distance travelled, percent of journey completed distance-wise are selected as the explanatory variables, and the delay time is the target variable in the prediction model. The model is applied on a set of historical traffic realization data from the part of a busy line in China to forecast delays. The results demonstrate that the ensemble prediction model has a higher prediction precision and outperforms the support vector machine (SVR) model and the random forest (RF) model.

CICTP