Designing robustness into predictive control

Model-based predictive control (MBPC) contains a whole family of algorithms with the same basic idea: to perform long range prediction of future outputs based on a process model and a set of hypothesised controls; to optimise a suitable cost-function to get the `best´ control sequence; and to assert the first of the set. Previous work has shown the role of `horizons´ in MBPC to attain stability and various other performance objectives in the `good model´ case. The present author concentrates on the model structure and filter design to maintain stability when the model is mis-matched to the plant. Some basic and general properties of prediction are presented