MPPT controller improvement for a PEM fuel cell system based on Gaussian Process Regression with a digital twin

Abstract

Hydrogen, due to its high energy density, stands out as an energy storage method for renewable energies in order to reduce the impact of the energy sector on global warming. Proton Exchange Membrane Fuel Cells (PEMFC) are responsible for converting the stored hydrogen into electrical energy and in order to obtain the highest energy conversion efficiency, the maximum power point (MPP) of the voltage-power curve of the fuel cell must be reached. Traditional Maximum Power Point Tracking (MPPT) algorithms, such as Perturb and Observe (P&O) or controllers such us Proportional Integral Derivative (PID) controller, are easy to implement, but must strike a balance between fast response and accurate control. Other more complex controllers such as Fuzzy Logic Control (FLC) or neural networks achieve better performance but at a higher computational cost. This paper presents a combination of a conventional Sliding Mode Control (SMC) and a machine learning Gaussian Process Regression (GPR) algorithm that provides a reference duty cycle reaching a point close to the MPP which is then used by the SMC to obtain the actual MPP. A Digital Twin of the PEMFC and a DC/DC converter, which allow a fast and large data-set generation, are used for the generation of the GPR algorithm. The proposed control is compared with a conventional SMC and performance improvements are observed using the Integral of the Absolute Error (IAE) metric. The results show, in a control initiation test, a 67% improvement in the IAE metric of the proposed control over the conventional SMC. In a load change test, the proposed control also outperforms the conventional SMC by 42.9%.