Abstract
In this paper, we propose a new set-membership(SM) identification method using periodgrams obtained from experimental data. One of the difficulties in the existing SM identification methods is that identified model sets are conservative, since they adopt the unknown-but-bounded noise sets which ignore empirical noise properties.In the proposed method, we consider the unknown-but-bounded noise set of cross-periodgrams of input and noise. This noise set is prescribed such that the upper bound of cross-periodgrams gets smaller by increasing the data number for the low correlation properties of noise. As a result, the noise effect in model set identification is decreased by increasing the data number. Also, since this noise set is convex, the identification problem is reduced to a convex optimization. Numerical examples show the effectiveness of the proposed method.
