Fuzzy model identification using a hybrid mGA scheme with application to chaotic system modeling

In constructing a successful fuzzy model for a complex chaotic system, identification of its constituent parameters is an important yet dificult problem, which is traditionally tackled by a time-consuming trial-and-error process. In this chapter, we develop an automatic fuzzy-rule-based learning method for approximating the concerned system from a set of input-output data. The approach consists of two stages: (1) Using the hybrid messy genetic algorithm (mGA) together with a new coding technique, both structure and parameters of the zero-order Takagi-Sugeno fuzzy model are coarsely optimized. The mGA is well suited to this task because of its flexible representability of fuzzy inference systems: (2) The identified fuzzy inference system is then fine-tuned by the gradient descent method. In order to demonstrate the usefulness of the proposed scheme, we finally apply the method to approximating the chaotic Mackey-Glass equation.