A Stable Recurrent Equilibrium Network-Based Controller for Proton Exchange Membrane Fuel Cell in DC Microgrids

This paper proposes a model-insensitive artificial intelligence control method for proton exchange membrane fuel cells (PEMFCs) in DC microgrids. The proposed method employs the recently developed recurrent equilibrium network (REN) to achieve model-independent PEMFC regulation. While REN has built-in stability, how to ensure the stability of the closed-loop control system formed by the REN and the target plant remains a challenge, which hinders the practical application of REN. To address this key obstacle, this study proposes a novel control strategy that provides a systematic closed-loop stability proof by incorporating the REN into PEMFC control systems through Youla parameterization. The guaranteed control stability makes the proposed method superior over most PEMFC control approaches that rely on offline trained neural network. Furthermore, compared to few online learning-based PEMFC control approaches with stability proof, the proposed method offers much lower computational burden. The superiority of the proposed method is validated through hardware-in-the-loop (HIL) experiments and digital signal processor (DSP).