Identification of benzene-toluene distillation column using neuro-fuzzy algorithm

Proper identification of nonlinear and complex distillation column is a challenging task in process industry. It is necessary for analyzing controller performance of the complex distillation column. In this work, identification of benzene-toluene distillation column is developed using neural network based on adaptive network based fuzzy inference system (ANFIS). Reflux rate and boil-up rate were used as input variables and top and bottom composition were used as output variable. Input output data used for identification are generated using HYSYS process simulation software. In this work nonlinear auto regressive with exogenous inputs (NARX) based ANFIS is proposed for getting better performance with dynamic nonlinear characteristics of the distillation column. Exhaustive search is used for choosing most influential six input arguments out of total twelve arguments. Quality of both ANFIS and NARX based ANFIS were compared with statistical criteria like root mean square error (RMSE) and correlation coefficient (R²). Result showed that NARX based ANFIS with exhaustive search performs better than ANFIS.