Author(s)

SHUHO WATANABE, YUKI MORI, YOSHINOBU TAKATORI, KAZUSHIGE YONEMOTO, NORIO TOMII

Abstract

We have developed a train traffic simulator in order to precisely simulate the delay occurrence and propagation which must be different depending on particular “situations.” By “situation” we mean the weather, a day of the week, season and so on. It is well known that occurrence and propagation of delays are different depending on these situations. Thus, it is necessary to develop a simulation algorithm which can exactly simulate the difference in occurrence of delays in order to use the simulator to evaluate effectiveness of delay reduction measures. The basic idea of our simulator is to use the historical train traffic records. By constructing regression trees from the historical train traffic records, we can know dwell times, running times and intervals of trains in each situation. By incorporating the results obtained from the regression tree to our macroscopic simulator constructed based on the longest path algorithm, we can construct a train traffic simulator which exactly simulate train traffic reflecting the “situation.” We have evaluated the simulator using the actual data and we have confirmed our approach is very promising.

Keywords

delay, simulator, regression tree, datamining