A Data-Driven Predictive Optimization Framework for the Dynamic Operation of Green Ammonia Energy Systems Under Uncertainty

Renewable power fluctuations and uncertain ammonia synthesis performance complicate operational coordination in green hydrogen–ammonia systems. This study develops a closed-loop prediction–scheduling framework using dual Gaussian processes to characterize renewable power and process performance. Calibrated lower and upper bounds enter material and energy constraints, while multi-objective particle swarm optimization balances operating cost and schedule variation. A joint-risk index selects plan retention, local repair, or global re-optimization, combined with plan refresh and conditional Pareto reuse in rolling horizon scheduling. The framework is evaluated with 24 h aggregated simulations using public meteorological data and published experimental data. Paired comparisons across seven fixed scenarios show that event-triggered and periodic scheduling both completed all 20 runs per scenario while satisfying all hard constraints. Under the reported serial timing protocol, event triggering reduced total controller runtime by 15.30–34.73%. Of 234 supplementary stress-test requests, 177 completed the day, including 176 that satisfied all hard constraints and one with a terminal audit failure; 57 stopped during the search. These results show that the framework coordinates probabilistic prediction, physical constraints, and scheduling responses under the specified simulation conditions, providing a reproducible approach to operational optimization of green hydrogen–ammonia systems.