Adaptive neuro-fuzzy inference system (ANFIS) – based model predictive control (MPC) for carbon dioxide reforming of methane (CDRM) in a plug flow tubular reactor for hydrogen production

Abstract

The current sources of our energy supply are plagued with many problems and the impact on the climate is of grave concern. Hydrogen (H2), when extracted from one of its many sources with carbon dioxide (CO2) capture, is considered a non-polluting, efficient and environmentally sustainable energy source. In this study, the control of a pilot-scale reformer for the production of H2 was studied. H2 was produced through the CO2 reforming of methane (CH4). An ANFIS approach was used to model the process. Type-2 fuzzy sets and systems was implemented as the fuzzy inference engine, which renders a piece-wise linearization of the nonlinear model. The generalized bell-shape membership functions were used and optimized based on empirical training data. This model was further used within an MPC framework to devise optimal control strategies. An unconstrained optimization method on the piece-wise quasi-linear model was carried out by adopting appropriate weights on the control inputs in the objective function. These weights were tuned to avoid actuator saturation via simulation. Experimental results showed that the ANFIS model was able to accurately replicate the response of the process to changes in temperature. Based on the ANFIS model, an MPC strategy was formulated for the process.