2006 年 19 巻 4 号 p. 157-165
In this paper, we consider a subspace system identification problem for linear stochastic systems subjected to observation outliers. First, the Least-Trimmed-Squares (LTS) technique due to Bai [1] is extended to Multiple-Input Multiple-Output (MIMO) regression model. Then, we identify the outliers in the output process of the MIMO state space model by using the LTS technique, and replace them by the median to obtain a preprocessed input-output data without outliers. We apply the Orthogonal (ORT) decomposition method to the preprocessed input-output data to get state space models. Numerical examples demonstrate the effectiveness of the proposed method.