An Approach for Simultaneous Structure Determination and Parameter Estimation Using MINLP Techniques

摘要

The present paper presents an approach of simultaneously solving a structure determination and parameter estimation problem using Mixed Integer Nonlinear Programming (MINLP) techniques. It is shown that the minimization of Akaike's Information Criterion (AIC), the Bayesian Information Criterion (BIC), and the Final Prediction Error (FPE) can efficiently be modeled as MINLP problems. The problems are solved using the Extended Cutting Plane (ECP) method. The presented techniques are applied in determining the structure and the parameters of some illustrative Autoregressive Moving Average (ARMA) time series. The described techniques can also be applied on dynamical systems.