Eco-driving strategy for connected automated vehicles at signalized intersections: A behavior-cloning approach in distributed model predictive control

This paper proposes an eco-driving strategy for connected automated vehicles (CAVs) at signalized intersections, aiming to reduce the overall fuel consumption of both CAVs and human-driven vehicles (HDVs) under varying CAV penetration rates and traffic volumes. The strategy integrates distributed model predictive control (DMPC) with a behavior-cloning network to control CAVs. To address this, we first designed an eco-DMPC controller using a virtual leading vehicle and dynamic desired speed. A neural network was then proposed for behavior-cloning of the Eco-DMPC controller, improving computational efficiency. Finally, simulation experiments were conducted to validate the effectiveness of the proposed strategy. The results show that as CAV penetration rate increases, fuel consumption decreases under the same traffic volume. At 100 % CAVs, fuel-savings range from 6.8 % to 9.7 % across varying traffic volumes, when compared to HDVs. Furthermore, the strategy improves intersection throughput as CAV penetration rate increases, with a 23.8–28.6 % increase in throughput under high traffic volumes when CAV penetration rate reaches 100 %. Additionally, compared to the eco-DMPC, the proposed strategy offers significant computational efficiency gains. Moreover, our strategy outperforms the green light optimal speed advisory (GLOSA) in both fuel-savings and intersection throughput, particularly in high traffic volumes across different CAV penetration rates.