Research on a novel multi-agent deep reinforcement learning eco-driving framework

With the worsening global environmental crisis and the rapid advancement of intelligent driving technology, eco-driving has emerged as a critical research focus to balance transportation efficiency and energy consumption. Addressing the multi-objective co-optimization challenge of driving performance and energy management in fuel cell hybrid electric vehicles within a car-following scenario, this paper proposes a multi-agent deep reinforcement learning framework incorporating a model predictive control guidance mechanism. By analyzing the scale factors between multi-agent reward value functions, the optimal weight coefficients are determined, significantly enhancing training efficiency. The improved multi-agent twin delay deep deterministic policy gradient algorithmic framework decouples the optimization objectives of different domains and provides guidance, and ultimately achieves good safety, comfort, state of charge of the lithium-ion battery maintenance, and economy through comprehensive comparative analyses under multiple driving cycles.