Real time determination of optimal switching times for a new H2 production process with CO2 capture: a supervised learning approach

The Ca-Cu looping process combines H2 production via reforming of natural gas with inherent CO2 capture, using CaO as CO2 sorbent and a Cu/CuO chemical looping for sorbent regeneration. The process comprises three dynamic stages, conducted sequentially in a packed bed reactor. This thesis addresses the determination of the optimal duration of each stage via a real-time approach. For the purpose, GPR models, a relatively recent method employed in machine learning, are used as soft-sensors.