Model validation for control and controller validation : a prediction error identification approach

Prof Michel Gevers (Universite Catholique de Louvain)

Prediction error identification delivers a nominal parametric model and a parametric uncertainty ellipsoid. Such description appears to be miles away from the classical uncertainty descriptions used in robust control analysis and design. In this talk we present recent results that connect these two sets of tools in a coherent way. Our results cover two distinct aspects.

1. We present a measure for the size of the model uncertainty set, resulting from prediction error identification or validation, that is directly connected to the size of the set of model-based controllers that stabilize all models in the model set. This allows us to establish that one identified model set is better tuned for robust control design than another, leading to control-oriented experiment design guidelines.

2. We also present necessary and sufficient conditions for a specific controller to stabilize all models - or to achieve a given level of performance for all models - in an uncertainty set defined by such ellipsoid in parameter space.

Since our results are politically correct, they rely heavily on the nu-gap metric, mu-analysis, LMI's, LFT's and other S-procedures. The presentation, however, will aim at the average man or woman in the street, if such are to be found in Cambridge.

Back to Control Seminars Page.