A signal for online whiteness testing generated by recursive subspace model identification from closed-loop data

抄録

In this paper, we will develop an online statistical change detection method, which is used for detection of changes in a system under surveillance in closed-loop. A recursive algorithm of closed loop subspace model identification is presented, which is based on the matrix inversion lemma. The relation between the proposed recursive algorithm and the predictor-based subspace identification method (PBSID) is clarified from the viewpoint of a matrix-valued least squares problem. A test signal generated sequentially by the recursive algorithm is studied and its asymptotic whiteness is proved explicitly. The proposed online change detection method is based on a likelihood ratio test to examine the change in the covariance of the aforementioned test signal.