AI Optimizes Electric Bus Efficiency & Traffic Flow

Researchers at an unspecified institution present PriEco-DRL, a system using deep reinforcement learning to simultaneously improve electric bus eco-driving and adaptive traffic signals. The framework coordinates bus speed with signal timing to reduce energy use while maintaining on-time performance. A key component is a “priority-weighted max-pressure” signal controller that adjusts green light duration based on real-time traffic density.

The system was tested on a real-world bus corridor and outperformed traditional fixed-time, actuated, and rule-based control methods. Improvements resulted from fewer stops and starts and smoother acceleration/deceleration. The research demonstrates a trade-off between minimizing energy consumption and adhering to schedules, allowing transit agencies to prioritize one over the other.

The team used a centralized training, decentralized execution approach, enabling a single AI agent to learn from multiple buses and routes. Further work could explore the system’s performance in larger, more complex urban environments.