Closed-loop subspace identification based on KPLS

Closed-loop subspace identification methods have enjoyed tremendous development in last decade. This paper presents a novel method combined with KPLS, aiming at the situation without persistence of excitation. In this method, KPLS is utilized to obtain Markov parameters, then, state sequence is estimated by SVD decomposition. Based on the estimated state sequence, the model parameters are estimated by linear regression. 30 Monte Carlo simulation examples are presented in the end of the paper, the results are shown to be competitive and robust in the situation without persistence of excitation in MIMO system, and the new method is more applicable for process industry.