Development of multi variable neural hammerstein model for MTBE catalytic distillation

The major purpose of controlling MTBE reactive distillation is to maintain the MTBE purity at the desired range. This process has highly nonlinear characteristic and strong interaction among the variables, hence only an advanced nonlinear model were proposed to model this process, which can be embedded in an advanced model based control. In this work, Neural Hammerstein (N-H) model has been developed to model the multi variable tray temperatures of the MTBE reactive distillation process. The input variables chosen are reboiler duty and reflux flowrate, while output variables chosen are temperature of tray number 3 and 8. In this work, data used was generated using aspen dynamic model of MTBE reactive distillation that is connected with Simulink in Matlab environment. Using N-H model, the accuracy to predict the output 1 and 2 are 99.81% and 74.66%, respectively. In contrast, the nonlinear autoregressive exogenous model (NARX) model was only produce 99.11% and 64.28% accuracy in predicting output 1 and 2, respectively.