抄録
This paper describes a method for identification of nonlinear systems characterized by the cascade connection of a linear element and a zero-memory nonlinear element. The dynamics of the linear element is represented by the weighting coefficients derived from the impulse response and the characteristics of the zero-memory nonlinear element is approximated by the polynomial equation.
The identification method proposed in this paper is a kind of sequential approximation method and the procedure of the estimation takes the following two steps. As the first step, the approximate solutions of the parameters of linear element are obtained by the nonparametric training procedure of the threshold logic learning machines, and then the approximate solutions of the parameters of zeromemory nonlinear element are obtained by the gradient method. As the second step, the accuracy of the approximate solutions derived in the first step is improved by the modified gradient method.
In this paper, the authors explain the algorithm of this method and discuss the convergence problems of the estimated parameters and show some results of the digital simulation which support the described theory.
