Abstract
A subspace identification method based on the orthogonal decomposition has been proposed in [3], where the stochastic realization of stationary process with exogenous inputs and its application to subspace identification have been discussed. In this paper, we consider two methods of identifying the deterministic part of the system. One is to estimate the deterministic part from the data obtained by the orthogonal projection of the output onto the space spanned by the input by the MOESP method [8], and the other one is to estimate the extended observability matrix based on the LQ decomposition of the data matrix [2]. We show by numerical examples that the identification results by the two methods are quite different for a system with sharp peaks in Bode plot.
