Online Learning of Neural Takagi-Sugeno Fuzzy Model

Abstract

A fuzzy artificial neural network which can embody a fuzzy Takagi-Sugeno model and curry out fuzzy inference and support structure of fuzzy rules is proposed. The algorithm of online model identification consist of new origin procedures namely input space partition with the new partition criterions of input space merging and adaptation of membership functions and regression coefficients in rules consequents were designed. The online identification is provided by the new identifying procedures and control subsystem. This subsystem decides if partitioning merging procedure must be applied ore system status becomes unchanged and parameters of rules antecedents and consequents must be adapted respectively (like resonance state of ART). The new identifying procedures and control subsystem were implemented into programme tools FUZNET. The case study presenting the prediction of artificial time series using the procedures of online learning fuzzy neural regression model (OLNFRM) is introduced.