Physics-informed NN-based adaptive backstepping terminal sliding mode control of buck converter for PEM electrolyzer

This paper proposes an advanced control approach to controlling a DC-DC buck converter for a proton exchange membrane (PEM) electrolyzer within the framework of a direct current (DC) microgrid. The proposed adaptive backstepping terminal sliding mode control (ABTSMC) leverages a physics-informed neural network (PINN) to accurately estimate and compensate for system uncertainty. The composite controller achieves finite-time convergence of the tracking error by combining backstepping control and terminal sliding mode control (TSMC). The proposed PINN aims to optimize the unconstrained parameters by utilizing observed training points from the solution, ensuring the network accurately interpolates a limited portion of the solution. The efficacy of the proposed hybrid control method is validated using a hardware-in-the-loop (HIL) implementation under various test settings, ensuring the preservation of the actual performance of the PEM electrolyzer during testing. The experimental verification results demonstrate that the proposed control method exhibits greater benefits, such as a faster dynamic response and greater robustness against parameter uncertainties than improved sliding mode-based controllers. In situations where operational conditions change, a rapid response is achieved within a mere $$ of settling time, exhibiting a minimal percentage overshoot of about $$ and presenting minimal fluctuations.