Dynamic Estimator of Artificial Neural Networks in High-Pressure Copolymerization application

Abstract

The main purpose of this research is to construct an Artificial Neural Network (ANN) for the prediction of an important quality variable, Melting Index, in a high-pressure EVA copolymerization reactor, where free radical copolymerization reaction takes place. The ANN is used to explore the feasibility of inferential feedback control and to enhance the grade transition. The modeling building for ANN makes uses of the mathematical models from Gan (2003) and Chen (2004). The model is simulated and excited by PRBS inputs to generate the required dynamic data for modeling. The ANN prediction of MI is corrected every four hours by a measurement data. The corrected ANN prediction is used as feedback to maintain the MI at a specified value. Robustness of the control system to modeling error is also demonstrated. The resulting control of MI is then applied to the case where grade transition is demanded.