Abstract
An identification algorithm that enables frequency domain weighting to reduce influences of unmodelled dynamics or noises is proposed. In this paper followings are shown using some examples: 1) It may happen that frequency bands to be empasized, in which signal-to-noise ratio is high or in which model fitting accuracy are needed, are known from some informations about plants, sensors or specifications for control. And this method is successfully used to weights such frequency bands. 2) The proposed method requires the smaller degree identification models to achieve sufficient data fittings compared to the well known least-squares method. 3) The computational power required to execute the proposed algorithm is not so big for popularly used work stations.
