Identification and Model Predictive Controller Design of the Tennessee Eastman Chemical Process Using ANN

Adaptation of network weights using LevenbergMarquardt (LM) training algorithm was proposed as a mechanism to improve the performance of Artificial Neural Networks (ANN) in modeling the Tennessee Eastman (TE) chemical process reactor. A Neural Network of the AutoRegressive, eXternal (NNARX) input model was developed. Four sub-models for the TE reactor were built. They are: the reactor level, the reactor pressure, the reactor cooling water temperature, and the reactor temperature. The Generalized Predictive Controller (GPC) shows a great success when handling highly nonlinear processes. We designed a GPC to reach the desired closed-loop behavior. A comparison is also provided between a feedforward network trained using Fast BackPropagation algorithm (FBP) and the Elman (ELM) network trained using LM algorithms. The effectiveness of the proposed models for both modeling and predictive control of the TE reactor has been demonstrated.