A modification of Sugeno-Yasukawa Modeler to improve Structure Identification Phase

Abstract Structure identification is one of the most significant steps in Fuzzy modeling of a complex system. Efficient structure identification requires good approximation of the effective input data. Misclassification of effective input data can highly degrade the efficiency of the inference of the fuzzy model. In this paper we present a modification to Sugeno-Yasukawa modeler to improve structure identification by increasing the accuracy of effective input data detection. There exist some middle models in the Sugeno-Yasukawa modeling process which a combination of them will result in the final fuzzy model of the system. In the original modeling process parameter identification is only done for the final fuzzy model. By doing the parameter identification for the middle fuzzy models, we have highly improved the accuracy of theses middle models. The RC (Regularly Criterion) error has been reduced 53% for middle fuzzy models and 67% in the final model for the sample function in formula (3). This accuracy increase, result in a better detection of effective parameters among input data records of a system. We have also used our new modeling method for a sample application and by modeling the system we have reduced input data needed for reasoning from 17 to 6. This caused a 60 % boost in the reasoning process of input data.