Fast approximate identification of nonlinear systems (Q1398410)

The paper deals with the identification of continuous or discrete-time SISO nonlinear systems by using a class of block-oriented models. The composite models in the class consist of a linear model, a Hammerstein model, a Wiener model and a Wiener-Hammerstein (sandwich) model connected in parallel. It is assumed that the linear dynamics is the same in each branch of the model and static nonlinearities are polynomials. An approximate fully decentralized two-step identification routine is proposed where in the first step the best, in the least squares sense, linear approximation of the system is determined and next, in the second step, the parameter of the polynomial models in each branch of the model are computed by linear least squares for the linear dynamics obtained in the first step. Finally, the most significant part of the whole model is selected. The approach is applied to nonlinear systems with a dominant linear behaviour and small nonlinear distortions. The method is verified experimentally by a collection of examples.