Durability-enhanced decision-making with style awareness for autonomous hydrogen fuel cell vehicle based on integrated reinforcement learning approaches

Fuel cell hybrid electric vehicles (FCHEVs) are pivotal for hydrogen-powered transport, yet achieving sustainability requires co-optimizing autonomous driving and powertrain control. This paper proposes an integrated reinforcement learning framework that synergistically optimizes tactical driving behavior and powertrain energy management of FCHEVs. Building upon the upstream rainbow deep Q-network (RDQN)-based lane-changing module that enhances driving decision space and energy-saving potential, a downstream improved soft actor-critic (SAC) algorithm is developed to concurrently optimize continuous acceleration control and power distribution, incorporating durability-aware reward mechanisms. While amplifying decision interpretability, it accounts for driver preferences and ensures coordinated vehicle-environment-energy interaction. Experimental results demonstrate that this tightly coupled approach attains 98 % near-optimal energy efficiency with collaborative lateral-longitudinal decisions. Style-aware optimization yields divergent outcomes: cautious trajectories deliver comfort with reduced consumption, against aggressive ones boosting velocity at 1.3 times hydrogen cost.