The Novel Fuzzy System Identification: Comparative Study and Application for Data Forecasting

Abstract

A novel fuzzy technique for identification data is proposed that combines fuzzy clustering with a type of Takagi-Sugeno fuzzy inference. The identification algorithm created in this study differs from others found in the literature, by the way the inference’s antecedents and consequents are built. Gaussian fuzzy sets with support on the cluster’s ordinal set of a determined α-level are unidimensional antecedents that have as a consequent an affine function that recovers the attributes of the collected data. Two simulations examples are performed to compare the new method: the comparison is done with well-known algorithms in terms of the normalized root mean square goodness of the fit measure, and the computational speed in the process of identification. To that end, Matlab System Identification toolbox is used in order to consolidate a unique source of comparison. The method has been tested in the prediction field, as part of a project supporting Brazilian local companies.