Identification of Non-linear Chemical Systems with Neural Networks

This study proposes the use of neural networks, specifically NARX networks, in the modeling of non-linear chemical systems with the use of the control field systems identification methodology. The chemical reactor of the Tennessee Eastman, responsible for the greater non-linearities of the plant, is studied. First, a simple decentralized control scheme is proposed for the stabilization of the plant, an identification experiment is designed, and two sub-models are trained for the level and pressure of the reactor, obtaining satisfactory results.