Near-term train delay prediction in the Dutch railways network
Zhongcan Li, Chao Wen, Rui Hu, Chuanlin Xu, Ping Huang, Jiang Xi
Abstract
Due to the unsuitable train delay prediction methods currently used in the Netherlands, a more accurate delay prediction method is needed. In this work, based on the data provided by the 2018 RAS Problem Solving Competition: Train Delay Forecasting, a data-driven model is established to predict the delay 20 min later. By combining the current delay with the operating conditions, the influencing factors that may influence delay propagation are extracted after analysing the delay propagation mechanisms and train movement data structure. These factors are considered as model input features for random forest regression, via which a prediction model is established. It is found that the random forest model exhibits high prediction accuracy and fast callback in terms of the training model, and ANN, XGBOOST, GBDT, and statistical algorithms are applied as benchmark algorithms. Finally, to complete the study, the importances of different delay influencers are investigated, calculated, and discussed.