A new optimal filtered version of stable generalized predictive control is presented. Using the concept of frequency weightings an additional degree of freedom for robustness tuning is introduced. The robust stability in respect to additive plant uncertainty is proven. A final control application to a cold rolling mill in an industrial noisy environment is discussed, using the Kalman design and robustness tuning. This application example exhibits enhanced tracking accuracy and improved stability margins.

A multivariable end-point state weighted generalized predictive controller for the Bridgman crystal furnace TITUS is developed. A multivariable temperature prediction is done by a reduced dynamic representation using a matrix polynomial representation of the plant. The theoretical equivalence between the proposed multivariable control and Kalman filter state prediction is proven. It is shown, that the incorporation of the stochastic model improves the control performance