Speed Planning for Integrated Eco-Driving and Bus Bunching Mitigation in Connected Electric Buses

Abstract

Eco-driving and bus bunching are two major challenges for connected and electric buses (CEBs). Eco-driving aims to minimize energy consumption, whereas bus bunching occurs when consecutive buses arrive at the same station simultaneously. The existing research rarely addresses both issues together. To address this gap, this article proposes a novel speed planning approach that addresses both problems via nonlinear model predictive control (NMPC) and imitation learning. A multiobjective NMPC model is developed that considers practical factors, such as energy consumption, time headway deviation, and traffic lights. To save computational resources, a speed planning network (SPN) based on transformer and long short-term memory (LSTM) architectures is designed to mimic the NMPC planner. Additionally, a knowledge distillation method is introduced to reduce the SPN’s memory footprint by incorporating mixed knowledge. Extensive experiments show that the NMPC model ensures nonstop passage through intersections and performs better across multiple metrics, including energy consumption and time headway deviation, than several baselines do. The SPN achieves similar performance to NMPC while significantly improving real-time efficiency, and the proposed distillation method further reduces memory usage while maintaining acceptable performance.