Well-Posed Model Uncertainty Estimation by Design of Validation Experiments

Abstract

In deterministic model validation approaches, model errors can be attributed to both disturbances and model uncertainty, leading to an ill-posed problem formulation. The aim of this paper is to remedy the ill-posedness in model validation for robust control. A two-stage procedure is developed, where first an accurate, nonparametric, deterministic disturbance model is estimated from data, followed by the enforcement of averaging properties through an appropriate periodic experiment design. The proposed deterministic approach results in an asymptotically correctly estimated model uncertainty and is illustrated in a simulation example.