Implementation and Properties of a Method for the Identification of Nonlinear Continuous Time Models

Abstract

The implementation of an output error method for the identification of non- linear, continuous time models using discrete time measurements is described. The model parameters are optimized by means of a quasi-Newton method combined with numerical integration of the model equations. If these equations satisfy certain conditions of piecewise continuity the performance function is proved to be continuously differentiable and the optimization iterations to converge. The parameter estimates are consistent for any additive, zero mean output noise. Numerical applications of the identification method are presented.