Computer-aided many-objective optimization framework via deep learning surrogate models: Promoting carbon reduction in refining processes from a life cycle perspective

This study proposed a hybrid model employing molecular-level process mechanism and data-driven surrogate optimization. We applied the hybrid model to realize the many-objective optimization for industrial chemical engineering process. The hybrid model is applied as an optimization paradigm in a novel processing route that maximizes chemicals from offshore crude oil. Molecular-level modeling procedures driven by mechanism models were employed to boost the deep learning database. Our methodology can be extended to general chemical engineering processes. A comparative analysis and evaluation by the hybrid model that is compared with the conventional direct solving model are implemented under varying operating modes. The proposed framework highlights significant arithmetic power advantages. Ulteriorly, two scenarios are assessed via employing the life cycle outlook. Results indicate that the orientation towards producing high-octane gasoline exhibits significantly 2.91% less non-renewable energy and 4.85% less CO₂ emissions compared with the orientation towards maximizing chemicals.