Formulating the Electric Bus Fleet Scheduling Problem for Reinforcement Learning

The transition to electric public transportation introduces new challenges in scheduling and operations due to battery constraints and charging requirements. To address these challenges, we propose a simulator designed for reinforcement learning (RL) based approaches to the electric bus fleet scheduling problem. Our work focuses on defining the state space, action representation, and overall simulator functionality to enable effective training and evaluation of RL agents. To validate our solution, we conduct initial experiments using a PPO (Proximal Policy Optimization) agent with a transformer-based architecture and implement automated testing to verify the correctness of generated schedules. Our results confirm that the simulator produces feasible solutions, providing a basis for future research in applying RL to electric bus scheduling.