Machine learning-enhanced uncertainty quantification for renewable-powered hybrid green ammonia and refrigeration systems: technoeconomic and environmental effects

Abstract

Renewable-powered ammonia production is a promising route for sustainable energy and hydrogen storage but is highly sensitive to operational uncertainty from variable power supply and component performance. This study presents a novel machine learning–enhanced uncertainty quantification (ML-UQ) framework that, for the first time, integrates a high-fidelity surrogate model—an artificial neural network with autoregressive feedback—into Aspen Plus simulations of a hybrid ammonia production system coupled with a vapour absorption refrigeration unit for heat recovery. The framework captures nonlinear interactions among six critical uncertain parameters, including renewable power variability, heat exchanger effectiveness, and compressor efficiency. It reduces the computational cost by three orders of magnitude while maintaining high predictive accuracy (R2 = 0.97, MAE = 8.57, RMSE = 11.3). The ANN surrogate enables scalable uncertainty propagation via polynomial chaos expansion. Results show that, across nominal power levels of 10–20 MW, uncertainties can cause up to 18 % variation in ammonia output, 30 % in refrigeration, and 40–50 % in CO2 emissions reduction. Heat exchanger effectiveness alone accounts for nearly 50 % of total variability. Economic analysis indicates a 5 % increase in the levelized cost of ammonia and 30–40 % variation in annual refrigeration revenue. This work delivers the first computationally feasible, ML-assisted surrogate-based UQ framework for hybrid green ammonia systems. More broadly, it offers a practical and readily scalable tool for designing resilient, economically viable, and low-carbon energy and chemical manufacturing systems.

More information

Identifiers

ORCID for Nader Karimi: orcid.org/0000-0002-4559-6245

Catalogue record

Export record

Altmetrics