Application of a non-linear long range predictive control to a packed distillation column

In this paper, nonlinear long range predictive control was applied to a pilot packed distillation column separating a mixture of methanol–water. The use of polynomial nonlinear autoregressive integrated moving average with external input (NARIMAX) model related with top product temperature and reflux ratio for nonlinear control was emphasised. The first part of the paper presents a rigorous dynamic model based on fundamental chemical engineering principles formulated for a packed distillation column. The actual column response to step changes in the feed mole fraction and temperature agreed well with dynamic model predictions. Then, recursive Gauss–Newton prediction error algorithm was used to determine NARIMAX model parameters. This algorithm can be used efficiently for this model and a transfer function model relating the manipulated variable (reflux ratio) to the controlled variable (top product temperature) was obtained. The role of data prefiltering prior to model parameter estimation is examined to overcome the parameter bias problem caused by disturbances was demonstrated. Next, non-linear long range predictive control algorithm was successfully applied in controlling the packed distillation column.