Tuning and Sensitivity Study for Model Predictive Control

Abstract

The objective of this work is to evaluate the performance of model predictive controllers based on low order process models and various tuning methods. Successive quadratic programming is used to fit first order plus dead time and second order plus dead time models to a simulated fifth order plus dead time plant. The performance of four tuning methods, principle component selection, move suppression, input blocking, and an "L=1" controller is compared. Performance is similar for all four tuning methods using both the first and second order models. Controller sensitivity to model parameters is discussed for each model with the principle component selection tuning method. The first order plus dead time model is sufficient to give good control over a wide range of model parameters. We recommend using a first order plus dead time model and prefer the move suppression tuning method.