Abstract
In this paper, we consider a subspace identification method for linear stochastic systems subject to observation outliers, where the observation noise contains large values with a low probability. We derive a subspace identification method by combining the orthogonal decomposition-based subspace identification method (ORT-method) and a weighted LQ decomposition. We apply the ORT-method to the input-output data, coupled with the standard LQ decomposition to obtain residuals of the output sequence. By using the median of residuals, outliers are detected by a simple scheme in robust statistics. Based on detected outliers, a weighting matrix is generated automatically, and is incorporated in the weighted LQ decomposition to get an improved estimate of the system matrices. A numerical example is included to show the effectiveness of the proposed method.
