Adaptive quantification of model uncertainties by rational approximation

An adaptive rational approximation approach for quantifying the effect of uncertainty is presented. It is shown that a tight frequency upper bound on the uncertainty is obtainable by adaptive rational approximation. The idea is that the identifier consists of two parts: the plant identifier and the uncertainty identifier. The plant identifier gives a nominal model which has lower complexity than that of the true plant. Errors between the estimated nominal model and the true plant are characterized by a sequence of rational functions which converges to the accurate upper bound of the uncertainty in the frequency domain. Moreover, since approximation and identification are grouped together, the whole procedure is completely automatic. This allows robust control and adaptive control to be combined