Neural Network for Identification of Heat integrated Distillation Column

Abstract

The heat integrated distillation column (HIDC), one of the most advantageous energy saving technique for separation of two liquids with distinct boiling point. The dynamic and nonlinear nature of distillation column make it a challenging problem for control engineers. Artificial neural network (ANN) is used for the nonlinear identification of HIDC. The nonlinear autoregressive network with exogenous input (NARX) model is also created to map the dynamic nature. Chemical process simulator Aspen HYSYS software is used to obtain realistic model of column and also for data sets. The mole fraction of top composition (benzene) and mole fraction of bottom composition (toluene) are taken as output variables, reflux rate and tray15 temperature are the input variables. Of 2000 samples collected from HYSIS, 1500 samples used for training the network and remaining for validation of the model. This will guarantee accuracy and robustness. The performance of both (back propagation) BP-ANN and NARX-BP-ANN are compared. The regression coefficient and root mean square error (rmse) of models with 12 different algorithms and different hidden layers calculated.