Electric Bus Fleet Dispatch with Wireless Charging
Transformer-enhanced reinforcement learning for joint vehicle routing and charging-schedule optimization.
Tongji University
Role: Team Leader (three members) · Advisor: Dr. Jia Xu
Period: August 2024 – November 2024
I led the development of a Transformer-enhanced deep reinforcement learning framework for large-scale electric bus fleet dispatch. The system jointly optimizes multi-vehicle routing and charging schedules under dynamic electricity prices and time-varying passenger demand.
The Transformer serves as the policy-network backbone, using self-attention to encode high-dimensional sequential states, including changing demand and multi-vehicle status. This design captures long-range temporal dependencies that conventional multilayer perceptron policies struggle to represent.
The project won second prize among approximately 100 teams in the National Campus Artificial Intelligence Algorithm Competition. It was evaluated on energy consumption and total time cost and outperformed heuristic dispatch baselines in simulated real-world scenarios.
Technologies: Python, PyTorch, Transformers, deep reinforcement learning, combinatorial optimization