MIMO Neural Network Model for Pilot Plant Distillation Column

Abstract

Abstract A distillation column is a complex multivariable system and exhibits nonlinear dynamic behavior due to the nonlinear vapor-liquid equilibrium relationships, the complexity processing configurations and high product purities. In order to gain better product quality and lower the energy consumption of the distillation column, an effective nonlinear model based control system is needed in order to allow the process to be run over a large operating range. The availability of a suitable nonlinear model is crucial important in the development of a nonlinear model based control. Neural networks framework have been used extensively in nonlinear process model development. In this paper, the multiple-input multiple-output (MIMO) neural network model to predict the top and bottom product compositions of a methanol-water pilot plant distillation column was developed. The neural network approach was applied by previous researchers in the development of nonlinear model for continuous distillation column based on simulation data (Yu, 2003; Singh et al., 2005 and Singh et al., 2007). However, only a handful of works have been carried out to validate the model with real plant data. The validation of the model with real data was important in order to ensure the ability of the model to represent the real processes being considered. Therefore, experimental works was carried out to separate the methanol water mixture in a continuous pilot plant distillation column. The error between the model and the actual pilot plant data was observed.