Spectral matching for parameter estimation in nonlinear input-output models

Parameter estimation in nonlinear stochastic models by spectral matching is studied. Models with a measured output and both known and unknown inputs are considered. Simulated data is used to estimate the spectrum of the nonlinear model and two different cost functions are suggested for the matching of the spectrum. Theoretical results on convergence and parameter covariance are discussed. An identifiability condition for a class of nonlinear models is presented. Finally, the estimation methods are illustrated on a numerical example.