Nonminimal Model based Output Predictors

In this paper, a new output predictor formulation is proposed for processes represented by ARMAX models. The proposed predictor uses an FIR filter structure and hence it is inherently stable and possesses dead-beat prediction properties. Furthermore, unlike some other FIR filter predictors, this predictor does not produce truncation errors. Good noise rejection properties are achieved by designing the predictor based on a nonminimal process model. By selecting J sufficiently large, (where J-1 is the nonminimal degree of the predictor model), the predictor will generate predicted output values arbitrarily close to those of the ideal, future, noise-free process output. The structural flexibility of the proposed predictor allows J to be time varying to accommodate unknown initial conditions and/or changing process operation conditions.