On the unification of predictive control algorithms

The design of the unified predictive controller (UPC) is presented. It is shown that the UPC control law is equivalent to a control law which generates the controller output by means of a linear combination of past process outputs, controller outputs and the reference trajectory. After defining a unified process model which can be used for a number of well-known process models, the authors derive its minimum-variance (MV) i-step-ahead predictor. It is shown that this predictor can be rewritten in a form consisting of two parts of which one does not depend on the noise model and the other does. This form of the unified MV i-step-ahead predictor shows that deviations between process and model, caused by model mismatch or disturbances, are filtered before they are added to the predictions.<>