Improvement of Distillation Composition Control by Using Predictive Inferential Control Technique

Abstract

For on-line control of product compositions in a multicomponent distillation column, several inferential control schemes have been proposed. In this paper, inferential models, not for estimating the current product compositions but for predicting future product compositions, are developed with Partial Least Squares (PLS). Then, the predicted compositions are used as controlled variables. This control scheme is termed “predictive inferential control.” Disturbances can be compensated before they affect the product compositions, and control performance can be improved, by controlling the future product compositions instead of the current product compositions. The detailed dynamic simulation results show that the proposed predictive inferential control scheme works much better than the conventional inferential control scheme as well as the tray temperature control scheme. Furthermore, it is demonstrated that control performance as well as estimation accuracy can be improved by the use of cascade control and an iterative modeling technique.