Adaptive cruise control and energy management for fuel cell electric bus based on multi-agent reinforcement learning

The coordinated optimization of adaptive cruise control (ACC) and energy management strategy (EMS) is promising for improving the performance of fuel cell electric vehicle. This paper proposes an ensemble learning-based multi-agent proximal policy optimization (EL-MAPPO) strategy to address the complex multi-objective optimization problem in car-following scenarios. In specific, multi-agent deep reinforcement learning is constructed for ACC and EMS respectively to optimize the vehicle speed and power distribution, while ensemble learning is devised to coordinate the interactions among multiple agents. Moreover, compared with existing methods optimizing the car-following safety, comfort and efficiency, the proposed method innovatively takes into account the lifespan of onboard power sources for optimization to reduce the overall driving cost in car-following scenarios. The comprehensive indices of following distance and fuel consumption indicate the superiority of the proposed strategy to MAPPO and PPO. The proposed EL-MAPPO demonstrates preferred performance over the single-objective EL-MAPPO, reducing the fuel cell degradation and overall driving cost by 38 % and 5.7 %.