A Robust Solution to Train Shunting using Decision Trees

Abstract

This research tackles the Train Unit Shunting Problem (TUSP) in train maintenance service sites. Many researches focus on producing feasible solutions, but only a few of them concentrate on the robustness of solutions. In reality, it is preferred to generate robust plans against unpredictable disturbances. Besides, the approach is expected to replan if disturbances occur while performing the plan. We propose this Decision Tree (DT)-based sequential approach (DTS) that solves the TUSP by sequentially making a sub-decision according to the DT prediction. It generates solutions that are both feasible and robust. Furthermore, it operates fast using the pre-trained model. We conduct experiments and compare its performance with a heuristic algorithm and the Local Search algorithm (LS). The proposed approach DTS solves fewer problems than LS and the heuristic, but it outperforms others by generating more robust solutions.