Comparison of offline identification methods on bounded autoregressive polynomial models

In this paper we focused on methods for offline identification of bounded autoregressive polynomials models. Firstly we used classical least square (LS) method for identification. Secondly we used total least square (TLS) method and thirdly we used gradient based method Levenberg-Marquardt for identification. Bounded AR polynomial models are basically nonlinear in parameters but the models can be modified to linear dependencies on parameters if bounding function is irreversible. Levenberg-Marquardt method was applied to unmodified bounded AR polynomial models. Input/Output data was generated from the model of isothermal continuous stirred-tank reactor with and without additive noise. Finally all methods are compared on one-step and multi-step predictions.