Marwa Hassan
Hydrogen energy storage is increasingly recognized as a key enabler for enhancing flexibility and reliability in smart microgrids with high shares of renewable energy. However, its practical deployment remains constrained by challenges such as real-time dispatch complexity, forecasting uncertainty, and nonlinear system dynamics. This study presents a novel AI-powered decision-support framework that integrates Long Short-Term Memory (LSTM) neural networks for short-term forecasting with the Krill Herd Algorithm (KHA) for optimizing hydrogen charging and discharging schedules. To preserve computational tractability, the photovoltaic (PV) array, electrolyzer, and fuel cell are modeled using simplified constant-efficiency assumptions that capture overall system behavior without representing detailed electrochemical dynamics. A real-world case study based in Aswan, Egypt-one of the highest solar irradiance regions globally-demonstrates the effectiveness of the proposed approach. The simulated microgrid includes a 5 kW photovoltaic array and a hydrogen storage system sized for daily autonomy. Using 15-minute resolution data, the LSTM-KHA framework achieved a forecasting accuracy of 4.8% (MAPE), reduced average grid import from 1295.2 W to 833.6 W (-35.6%), and lowered average PV curtailment from 786.3 W to 618.2 W (-21.4%). It also improved energy self-sufficiency from 71.5% to 89.7% and resulted in a daily [Formula: see text] emissions reduction of approximately 7.76 kg. These results confirm the potential of combining deep learning with nature-inspired optimization to support intelligent, low-emission energy management in hydrogen-integrated microgrids.