Hybrid digital twin and physics-informed surrogate optimization for dynamic power-to-methanol under renewable intermittency

Intermittent renewable power disrupts steady operation in power-to-methanol plants and depresses efficiency when controllers chase sharp ramps. A digital twin integrates a dynamic first-principles model of the methanol loop with a physics-informed surrogate in a gray-box form and a model predictive control layer tuned for constraint handling and economic objectives. Training uses multi-fidelity samples from a high-fidelity simulator plus bench microreactor anchors across 200–270 °C and 20–40 bar. Cross-validated parity metrics show R 2 above 0.99 for CO 2 conversion and above 0.98 for reactor temperature on held-out scenarios. Closed-loop studies across solar, wind, mixed, curtailment-limited, and gusty profiles show an 8 % reduction in specific energy use, a 0.07 absolute gain in CO 2 conversion, and a 3.5 % rise in CH 3 OH selectivity relative to a proportional–integral baseline. Constraint violations drop from 14 to 3 per cycle, a 79 % decrease, while actuator moves respect rate limits. Economic analysis indicates a levelized cost shift from 455 $ t −1 to 420 $ t −1 , consistent with electricity-driven savings identified in sensitivity tests. Results confirm that physics guidance in the surrogate curbs extrapolation error, accelerates optimization, and supports robust tracking under minute-scale ramps, with bench evidence aligning with digital predictions across the stated operating window.