A hierarchical fuzzy-clustering approach to fuzzy modeling (Q1770727)

The study is concerned with the comprehensive development of a Takagi-Sugeno rule-based system through the use of hierarchical fuzzy clustering. A generic topology of the model based upon a collection of ``local'' linear subsystems assumes the form ``if \(x\) is \(\Omega_i\) then \(y= L(x, a_i)\), \(i= 1,2,\dots,c\) where \(\Omega_i\) denotes a certain region of the multivariable input space and \(L\) stands for a linear function (conclusion) of the rule. The crux of the design concerns a construction of an optimal partition of the multivariable input space into clusters. This process consists of two phases: first, the numeric data are grouped into clusters by applying the nearest neighbor rule and afterwards the resulting centers are combined (clustered) by means of the weighted fuzzy \(C\)-means. At this stage, a blueprint of the model has been formed. In the sequel, the parameters of the model are tuned through a series of gradient-based adjustments. The study offers a number of experimental studies and demonstrates the performance of the resulting models vis-à-vis some others existing in the literature.