Data-driven multivariable controller design using Ellipsoidal Unfalsified Control

Unfalsified Control is a data-driven, plant-model-free control design approach. It does not require a plant model, only measured input/output data is used. It employs the approach of controller unfalsification, where it is recursively checked which controllers would have met the predefined performance requirement without the need to actually implement them.In Ellipsoidal Unfalsified Control, the region of unfalsified controllers is updated each sample, and the continuous region is described by an ellipsoid. In this work, the framework of Ellipsoidal Unfalsified Control is extended to cover general multivariable controllers. A new controller structure and a sequential update procedure are proposed, such that the same arithmetics as in the single-input single-output case can be employed. The proof of stability of the proposed method is deduced from the stability theory of single-input single-output Ellipsoidal Unfalsified Control.