Real-Time Implementation Of Non-Linear Model Predictive Control To A Pilot Distillation Column

Abstract

In the last decade there has been a growing interested concerning nonlinear model predictive control (NMPC). However, the number of practical implementation of modern NMPC techniques is still very small due to the difficulties that have to be overcome to develop a practical implementable NMPC controller. The most important obstacle in the implementation of NMPC comes from the real-time feasibility of the optimization problem involved in the controller, which strongly depends on the complexity of the model used for prediction. In this paper the real-time implementation of two NMPC techniques to a laboratory azeotropic distillation column is considered. The particular control hardware of the pilot distillation system leads to hybrid control architecture. In this paper a novel hybrid control approach is introduced which exploits the advantageous properties of genetic algorithm (GA) in the solution of the mixed real-binary optimization problem from the controller. The first approach is based on a detailed first principles based model. It is shown that the real-time feasibility of this technique cannot be guaranteed due to the high computational burden caused by the increased model complexity. To overcome this problem an artificial neural network (ANN) model based NMPC is implemented which leads to a fairly good control performance.