Identification of state-space switched linear systems using clustering and hybrid filtering

This paper addresses the identification of switched linear MIMO state-space systems. The proposed methodology explores the use of subspace identification techniques, clustering, and data classification to obtain an estimate of the submodels. The main contribution of this work lies in the inclusion of an identification step of points generated by the same mode. For this points classification it is proposed to use a hybrid filtering technique known as interacting multiple model (IMM) algorithm. An important feature of the overall identification algorithm is that the matrices of different submodels can be directly combined because, they are obtained for the same state-space basis. The efficiency of the developed method is demonstrated via numerical examples.