Closed-loop identification using canonical correlation analysis

Abstract

We consider the identification of linear state space innovations model from closed-loop data. We suggest to use the subspace closed-loop identification algorithm of [3] to obtain an initial estimate of the deterministic part of the system and then plug this estimate into the second stage of the 2CCA algorithm of Peternell et. al. [9]. The main result of this paper is to show that given closed-loop data and consistent estimates of a number of Markov parameters of the deterministic part of the system, the second stage of the 2CCA algorithm delivers consistent estimates of the system matrices of the innovations model.