Robust self-tuning predictive control for industrial applications

The recently introduced self-tuning Generalized Predictive Control (GPC) algorithm based on long range prediction, has been successfully tested in a wide range of industrial control applications. The complicated nature of the GPC algorithm, however, makes it very difficult to apply to it the standard analytical robustness techniques. A novel approach, Minimax Predictive Control (MPC), is developed which is shown by simulation studies to have robustness properties superior to those of the standard GPC. The difference between MPC and GPC algorithm is that the peak of the future predicted tracking error and the incremental control spectra are penalized rather than their integral on the unit circle. Both one degree and two degree of freedom MPC algorithms are developed. The two degree of freedom design leads to mixed $H\sb2/H\sb{\infty}$ (minimax) predictive control law design. Extensions of the method to the multivariable case are also considered. The Minimax Prediction Error (MPE) identifier is introduced as the parameter estimator in the adaptive (self-tuning) version of the algorithm. The resulting self-tuning algorithm is shown to have better robustness properties than the GPC self-tuner. The constrained MPE estimation algorithm is developed to provide the long term integrity for adaptive control.U of I OnlyETDs are only available to UIUC Users without author permissio