Online train delay recognition and running time prediction

Data mining and analysis of standard track occupation data is used to compute accurate actual train delays at stations. Based on historical train describer records the delays can be automatically classified into initial and consecutive ones through comparison of actual with scheduled blocking times. The distributions of running times and dwell times of each line and direction can be estimated conditional on the amount of original delays, route conflicts and factors such as type of rolling stock, peak hours or even weather conditions. The running times of actual trains until the next downstream stations can be predicted at a high precision by means of a new online model whose parameters have been calibrated and tested in the Dutch railway corridor Rotterdam - The Hague.