Identification of Ethane-Ethylene Distillation Column Using Neural Network and ANFIS

Abstract

In this work a non linear multiple input multiple output model for binary ethane-ethylene distillation column is derived. Identification is carried out on nonlinear auto regressive with exogenous inputs (NARX) structure based neural network (using both Steepest Descent algorithm and Levenberg-Marquardt algorithm) and NARX based ANFIS. Data used for identification is obtained from Daisy database. Ratio between reboiler duty and feed flow, ratio between reflux rate and feed flow, ratio between distillate and feed flow, input ethane composition and top pressure were used as input variables while top ethane composition, bottom ethylene composition and differential pressure between top and bottom were used as output variables. In this work a new method for identification of distillation column using NARX based ANFIS is proposed. Result showed neural network model and ANFIS model was able to capture nonlinear dynamic behavior of the distillation column. Results were compared with statistical criterion (Correlation Coefficient and Root Mean Square Error) for each of the neural network model and ANFIS model to understand which model performs better. Considering the results it is obvious that NARX based ANFIS model is more accurate with less number of iteration.