An optimal two-stage identification algorithm for Hammerstein-Wiener nonlinear systems (Q1129645)

A parametric identification problem for a composite structure of a Hammerstein-Wiener dynamical system is considered. The system consists of two nonlinear static elements separated by a discrete-time linear dynamics connected in a cascade. The nonlinear characteristics of the static elements are linear in the parameters and the linear dynamic part is asymptotically stable. Only input-output measurement data of the whole system are accessible. A two-stage least-squares identification algorithm for estimating the system parameters is proposed and investigated. In the first stage, some generalized parameter estimates are computed by using a least squares method and then, in the second stage, estimates of true system coefficients are derived by solving proper singular value decomposition problems. Efficiency of the algorithm is examined in the absence of noise and under additive white random noise in the output data.