Nonlinear system identification based on adaptive competitive clustering and OLS

In this paper, a new identification method for nonlinear system model from input-output data is presented. In accordance with the problem that sensitivity to initialization and noise, and some relative parameters must be determined beforehand during the fuzzy clustering process in the usual fuzzy cluster algorithm, and the existing competitive clustering algorithm have poor convergence properties, and make convergence to a local minimum more likely. A type of adaptive competitive cluster algorithm for structure identification is presented. At the same time, orthogonal least squares (OLS) method algorithm is used to remove redundant fuzzy rules and identify model parameters during the clustering process. Through simulation research, the effectiveness of the method is proved.