An improved soft actor-critic based energy management strategy of fuel cell hybrid electric vehicle

Proton Exchange Membrane Fuel Cell (PEMFC) stacks' lifetimes are increased and similar consuming fuel is achieved with the help of energy management strategies (EMSs) that incorporate Deep Reinforcement Learning (DRL). In the research, a cutting-edge Soft Actor-Critic (SAC) DRL algorithm is applied to optimize EMS. The algorithm incorporates fuel economy, fuel cell degradation factor, and lithium battery (Li-battery) degradation factor into the target design. Furthermore, an optimized Soft Actor-Criticism-Power Limit constraint (SAC-PL) algorithm is proposed by embedding instantaneous cost and power constraint into SAC algorithm. The cumulative degradation of hydrogen consumption, PEMFC stack and lithium battery (Li-battery) under four common vehicle driving cycles is evaluated using Dynamic Programming (DP) algorithm, SAC algorithm, SAC-PL algorithm and two other EMS DRL algorithms. The outcomes demonstrate that the EMS optimized based on SAC-PL algorithm has an important role in reducing fuel consumption, improving training stability, speeding up convergence, and extending the lifetime of PEMFCs.