Model-free reinforcement learning-based transient power control of vehicle fuel cell systems

Transient power control of vehicle fuel cell systems is the basis for performing energy management strategies, which are critical to the control of fuel cell vehicles. Nevertheless, the nonlinearity and the time-varying property in the model of characteristics of fuel cell dynamics make transient power control challenging. This study employs a model-free reinforcement learning algorithm to control transient power in fuel cells precisely. This approach combines the feedforward mechanism of prior knowledge with the feedback mechanism of critic–actor neural networks. It validates its effectiveness through a real proton exchange membrane fuel cell system and hardware-in-the-loop (HIL) experiments. Experimental results from a real proton exchange membrane fuel cell system indicate that under rapidly changing load conditions, the proposed control strategy exhibits rapid dynamic response, precise steady-state tracking performance, and high robustness against unmodeled dynamics, effectively addressing complex dynamic response challenges.