Agnostic deep neural network approach to the estimation of hydrogen production for solar-powered systems

Abstract

Countries' abilities to meet their own energy needs using renewable and clean energy regardless of external sources is an important option in terms of sustainability. Moreover, very low emission systems can be developed using a combination of hydrogen energy technologies and renewable energy systems. However, renewable energy shows certain irregularities as a fact of its very nature. However, an energy source that is not affected by irregularities is quite possible using electricity produced from renewable energy in production and storage of hydrogen. At this point, the predictability of a country's energy sources is vital when investing in stable energy systems. In this study, the hydrogen production values of a PEM electrolyser (PEM-E), as supported by GaAs Photovoltaic (GaAsPV) technology, are estimated using data from a meteorological station with high solar energy potential. The observed and predicted solar radiation values constitute the main dataset. Daily radiation estimations are achieved via an Artificial Neural Network (ANN), Adaptive Network-Based Fuzzy Logic Inference System (ANFIS), Multiple Linear Regression (MLR), Empirical Angström-Prescott (EAP), and Agnostic Deep Learning (DL) models. The system, which consists of a 10 kW (68.10 m2) GaAsPV and 4688 W PEM-E, is considered for the determination of HP values. The root mean square error (RMSE) and the coefficient of determination (R2) are used as performance criteria in comparing hydrogen production based on estimated solar radiation with observed values falling on inclined surfaces. The DL model is highly compatible with the observed data and is the most successful model of those considered with a performance value of 96.26% (R2).