Self-adaptative multi-kernel algorithm for switched linear systems identification

This paper deals with the problem of switched linear system identification. This is one of the most difficult problems since it involves both the estimation of the linear sub-models and the switching instants. In fact, we propose an identification approach based on self-adaptation multi-kernel clustering algorithm to estimate simultaneously the linear sub-models and the switching signal. The estimation of the sub-models consists of decomposing the regression vector into several blocks and assigning a kernel function to each block. However, the estimation of the switching signal is provided by an unsupervised classification algorithm with self-adaptive capacities. Simulation results are presented to illustrate the effectiveness of the proposed approach.