Prediction Method of Train Delay Using Deep Learning Technique

抄録

Commuter lines in metropolitan areas in Japan suffer frequent short train delays. Train dispatchers reschedule operations according to these delays and how they evolve. However, because of the complex way in which they evolve, it is difficult to predict delays of up to a few tens of minutes. To build an accurate prediction method, we developed prediction method using a deep learning model called Long Short Term Memory. This paper reports on the prediction performance of this method compared with the conventional method using neural networks.