[3K3-GS-10] AI application: trafic

[3K3-GS-10-05] Acquisition of energy-saving driving strategies while ensuring on-time performance

〇Ryota Toriumi¹, Sachiyo Arai¹ (1. Chiba University)

Keywords:Deep reinforcement learning, Hierarchical reinforcement learning, On-time performance, Energy-saving operation

In recent years, energy conservation in railroad systems has become one of the most critical issues, and there are several previous studies on reducing the energy required for train operation. An approach using deep reinforcement learning (DRL) has been proposed because it can find the optimal operation sequence from arbitrary operation conditions. Previous studies applying DRL cannot guarantee on-time performance because the system learns by scalarizing the rewards related to on-time performance and energy-saving. This study proposes a hierarchical reinforcement learning method to acquire energy-saving driving strategies while ensuring on-time performance. Computer experiments verify the performance of the proposed method, and the proposed method generally achieves on-time performance and saves 27.4% of energy compared to the case where the train runs in the shortest time.