A Concept for Model based Predictive Control without Explicit Process Identification

Model Based Predictive Control (MBPC) seems to be a promising strategy for control of real-life processes. Many applications have been reported. Although the concept is simple, an important obstacle in practical implementations is the acquisition of a suitable dynamic process model. Usually this process model is obtained by (a priori) off-line process identification, sometimes also by (real-time) on-line identification, e.g. adaptive strategies. Process identification in realistic circumstances is rather difficult and time-consuming. In this paper a concept is introduced which bypasses the obstacle of explicit process identification for obtaining a suitable MBPC prediction model.