An approach to structure identification of fuzzy models

This paper deals with the structure identification problem for a fuzzy model, which is solved under the requirement of simplifying a fuzzy system once a satisfactory structure is available. Particularly, we propose a rule selection method to build a simplified version of the original rule base by preserving the model accuracy. The rule selection problem is formulated as a structure reduction process of the neuro-fuzzy network used to model a fuzzy system and is solved through an iterative algorithm aiming at selecting the minimal number of rules for the problem at hand. Experimental results demonstrate the algorithm´s effectiveness in identifying reduced fuzzy models with equivalent performance to the original one