A semiempirical identification method by using a multiestimation technique via reduced-order nominal models

A multiestimation scheme is presented to identify a partially unknown plant. Several reduced-order linear nominal models of the plant are considered to compose the multiestimation scheme. Each reduced-order nominal model is built as a parallel connection of first-order filters and contains some, but not all, natural modes, which are supposedly known, of the true plant to be identified. The assumption that the elementary filters are of first-order is identical to assume that all poles are real and distinct, which is feasible in many practical situations. The assumption that the modes are known may work in an acceptable way when nominal values and a small range of uncertainty are known. A supervisor with a switching law selects the most appropriate estimation model of the plant at certain time instants according to an index related with the identification error of each estimator. In this way, a system identification scheme that incorporates model order reduction issues can be designed.