A New Decision Rule for Model Structure Identification of A Class of Nonlinear Dynamic Systems

This paper is concerned with the problem of deriving a statistical decision to rule for model structure identification of a class of nonlinear dynamic systems. The concept of a basis to describe the model structure is introduced and the model structure space corresponding algebra structure are defined. Then, based on the Kullback-Leibler mean information, a new model structure decision rule is developed by maximising the average log-likelihood function. Some analytical and simulated comparisons of this decision rule with Akaike's FPE and AIC, F-test and Bayes aposteriori decision rule are given.