Identification of non-linear dynamic systems with decomposed fuzzy models

This paper presents an approach which is useful for the identification of discrete non-linear dynamic systems based on fuzzy relational models. Fuzzy systems are characterized by a rule-base specification. If the complexity of a rule-base increases, knowledge acquisition may become tedious because the number of rules increases with an increasing number of fuzzy variables. Decomposed fuzzy models are proposed and applied to dynamic systems modeling. The evolution of the identification algorithms for the decomposed fuzzy model is suggested. A comparative study of the dynamic system identification with the conventional relational model and the decomposed relational model is presented for a well-known identification problem, namely the Box-Jenkins gas furnace data.