Subspace Identification of Observation Outliers with color noise Based On Leave-k diagnostics

Abstract

In this paper, we develop a subspace system identification method for linear stochastic systems subject to observation noise with outliers, where the observation noise contains large values with a low probability. By using the leave-k-out diagnostics method, we detect the outliers and substitute the median of the output data for them. Then we apply the orthogonal decomposition (ORT) based method [3] to get state space models. A numerical example demonstrates the effectiveness of the proposed method.