Artificial intelligence-based modeling and multi-objective optimization of industrial hydrogen production with minimization of CO and CO2 generation

This study presents an integrated approach for modeling and multi-objective optimization of the reaction section of an industrial steam methane reforming hydrogen production plant. The process was simulated in Aspen Plus using a detailed kinetic model and validated against industrial data. Response surface methodology and an artificial neural network (ANN) were developed as surrogate models and coupled with the NSGA-II algorithm to maximize hydrogen production while minimizing CO and CO 2 emissions. The ANN achieved predictive accuracy (R 2 > 0.999). Sobol global sensitivity analysis identified reformer inlet temperature and steam-to-methane ratio as the most influential variables, while less significant parameters were fixed to reduce optimization dimensionality. Both optimization approaches converged to a similar operating region, yielding hydrogen production rates of 4300–4600 mol h −1 and CO + CO 2 production rates of 900–980 mol h −1 . Validation using Aspen Plus confirmed the reliability of the proposed method, with the ANN-based approach providing the highest prediction accuracy.