An intelligent full-knowledge transferable collaborative eco-driving framework based on improved soft actor-critic algorithm

Eco-driving is a promising technology for fuel cell vehicles (FCVs) that simultaneously achieves safe driving and energy saving in the urban transport sector, particularly through the application of cutting-edge deep reinforcement learning (DRL). However, developing specific DRL-based eco-driving strategies for different FCVs is a laborious task, since repetitive training is required when encountering various FCV types. To tackle this challenge, this paper proposes an intelligent transferable collaborative eco-driving framework across FCV types. Firstly, the eco-driving problem in the vehicle-following scenario is formulated by collaboratively integrating adaptive cruise control (ACC) with energy management strategy (EMS), and then an improved soft actor-critic (ISAC) algorithm is designed to solve this problem. After that, a source eco-driving strategy based on I-SAC is pretrained for a light fuel cell hybrid electric vehicle (FCHEV). Finally, all learned knowledge in the source strategy is fully transferred and reused for a heavy-duty fuel cell hybrid electric bus (FCHEB) to get the target eco-driving strategy. Experimental simulations show that the proposed framework can expedite the development of the ecodriving strategy for FCHEB by 94.83% while reducing hydrogen consumption by 10.05%.