Identification algorithm for a class of nonlinear systems

Abstract

Nonlinear system identification based on local model networks is considered for the nonlinear process where a nonlinear element is followed by linear dynamics. The local model is chosen as a linear model, or a simple block oriented nonlinear model, whose orders are determined through a criterion function with respect to both the approximation accuracy and model simplicity. The weight of every local model varies with the operating point of the present process state, and the parameters of local models are estimated through some simple parameter estimation algorithms. The algorithm can work even under the situation where little information on nonlinearity is available, and it can be implemented easily in practical systems. Moreover, its application to the processes with saturation and backlash is investigated to show the effective of the proposed algorithm.