Identification and control of distillation columns using optimized intelligent techniques

Abstract

In this thesis, identification and control of two types of distillation columns are newlinedescribed. (1) Binary distillation column (2) Heat integrated distillation column newline(HIDC). Both binary distillation column and HIDC are highly nonlinear, modelling newline newlineof such system are challenging in process industry. Intelligent techniques such as ar- newlinetificial neural network (ANN), fuzzy systems, adaptive neuro-fuzzy inference system, newline newlinesupport vector regression e.t.c, are well suited for nonlinear mapping. If the process newline newlineinput-output data are available, these intelligent techniques can be used to approxi- newlinemate output values for the given inputs. Intelligent methods can also be employed newline newlineas nonlinear controller for nonlinear systems such as binary distillation column and newlineHIDC. newlineIn a binary distillation column, identification or data driven modeling has been newlinecarried out using the nonparametric method such as artificial neural networks and newlinefuzzy systems. But identification of systems using these methods has a drawback of newlinetrapping at local minimal points. To avoid the drawback of trapping, meta heuristic newlineoptimization algorithms can be used. So, in our works, neural network and fuzzy newline newlinesystems optimized using meta heuristic algorithms are used. Since the real distil- newlinelation column is dynamic, NARX (nonlinear autoregressive with exogenous input) newline newlinestructures are used for identification. Data needed for identification are collected newlinefrom HYSYS chemical simulation software. 1000 samples of data are collected from newlineHYSYS. 750 samples of data are used for training and remaining 250 samples of newlinedata are employed for validation of proposed models. newline