An effective MPPT control based on machine learning method for proton exchange membrane fuel cell systems

Abstract

This study proposes a machine learning-based maximum power point tracking method for fuel cell systems. Initially, a Matlab model was developed to accurately represent the behaviour of a proton exchange membrane fuel cell. A dataset for training of machine learning methods was collected from fuel cells operating under different conditions. To demonstrate the effectiveness of the proposed method, comparison studies were conducted using classical methods, namely "perturb and observe" and "incremental conductance" under varying temperature and pressure conditions. Notably, under the condition of temperature and pressure variation, the output powers obtained from the system were 2005.2 W for support vector machine and 2018.5 W for linear regression at 4.5 s, while the output powers of 'incremental conductance' and 'perturb and observe' were 1989.6 W and 1986.3 W, respectively. The results demonstrate that the proposed method contributes to a more efficient operating condition and faster dynamic responses to changes.