| Abstract [eng] |
This study presents a reinforcement learning approach for reducing energy consumption in freight train operations while still meeting scheduled arrival times. A simplified physical simulation, combined with a learned power control policy, allows the system to select engine power levels based on the remaining travel time, speed limits, train mass, and the upcoming terrain along the route. The model is trained through repeated interaction with the simulated environment, where it learns to balance punctuality and energy consumption based on the reward function designed for this task. The results show that the trained model can learn energy-efficient driving behaviour and outperform basic control strategies in terms of energy consumption. Despite relying on simplified models and route-specific training, the work demonstrates that reinforcement learning is a practical and effective tool for supporting train operators in executing energy-efficient train power control. |
