A noise resilient multi-step ahead deep learning forecasting technique for solar energy centered generation of green hydrogen

Abstract

Generation of green hydrogen based on solar energy is significantly dependent upon photovoltaic (PV) panel energy, which is reliant on the stochastic and intermittent global horizontal irradiance (GHI). Further, if inconsistent power output from PV panels due to influence of noise interferes with continuous supply of energy needed for electrolysis process, then reliable generation of green hydrogen may be impacted. Therefore, in this study, preliminary fast fourier transform (FFT) technique for noise removal is utilized and thereafter, singular spectrum analysis (SSA) algorithm is integrated with Gated Recurrent Unit (GRU) in order to enhance the accuracy during forecasting of green hydrogen reliant on GHI at multiple time steps. The suggested methodology is assessed using GHI dataset for Jaipur site (Rajasthan, India) obtained from National Institute of Wind Energy (NIWE) portal. The monthly average prediction for solar energy centered green hydrogen generation by employing the proposed methodology at Jaipur location corresponds between 0.010 kg/m2 (10 × 103 kg/km2) to 0.020 kg/m2 (20 × 103 kg/km2) in diverse circumstances. The suggested forecasting algorithm provides a reliable and precise methodology that will significantly supports low-carbon economy goals.