Closed-loop AI forecasting for heuristic, forecast-informed dispatch in solar–green hydrogen hybrid systems (SGHHS) under seasonal variability

Solar–hydrogen hybrid systems can provide low-carbon, dispatchable power, yet many studies assume freshwater availability and rely on static dispatch strategies that are sensitive to climatic variability. This work develops a closed-loop Solar–Green Hydrogen Hybrid System (SGHHS. A five-layer supervisory architecture is implemented in which a Random Forest model predicts a physics-informed PV availability index from meteorological data and supports forecast-informed, rule-based dispatch under component and storage constraints. The system (22.75 MW PV, 2.25 MW electrolyzer, 450 kg H₂ storage, 1 MW fuel cell) is evaluated using long-horizon, hourly simulations under a consistent day/night demand profile. Results indicate an LCOE of USD 0.19–0.23 /kWh, corresponding to an 8–12% reduction relative to a static baseline, with smaller but consistent gains observed under a real-year test. The proposed controller improves dispatch tracking by approximately 22% and reduces hydrogen overproduction and curtailment by about 20%. Overall, the findings show that ML-enabled availability forecasting combined with closed-loop water–energy integration can enhance the operational robustness and sustainability of SGHHS. • Forecast-driven dispatch improves SGHHS operational stability under seasonal solar variability. • Random Forest outperforms LR/SVR/GBDT for PV-efficiency prediction in the studied dataset. • AI-informed control reduces fuel-cell cycling and hydrogen venting compared with static dispatch. • Cross-year (TMY vs 2012) validation demonstrates improved robustness of dispatch decisions. • The proposed SGHHS achieves lower LCOE through reduced curtailment and improved asset utilization.