Parametric methods for nonlinear system identification

In this thesis, we have developed practical methods for the identification of linear, nonlinear and hybrid (multimode) systems which are applicable under relatively general conditions, i.e., when assumptions and conditions of the estimation technique are not violated. Since these algorithms were not designed specifically with any system(s) in mind, they should be applicable to experiments on a variety of systems in many different disciplines.Results demonstrate that the (polynomial) NARMAX (Nonlinear Autoregressive, Moving Average eXogenous) model class is useful for modeling the input-output behavior of a block-structured representation of two biological models. Extensive simulations demonstrated that our bootstrap model order selection (BMOS) and bootstrap structure detection (BSD) algorithms have a high probability of success for selecting the order and structure of NARMAX models and are robust in the presence of measurement noise. In addition, we illustrate that the NARMAX model structure is well suited for modeling dynamics of nonlinear hybrid systems and develop a modified extended least squares (MELS) algorithm to estimate coefficients of these systems. Application of this algorithm to a model of the vestibulo-ocular reflex (VOR) showed that it is a robust method for estimating the coefficients of multimode systems