Structure identification of fuzzy model (Q1107481)

A main idea of a class of models studied in this paper concerns a representation of a model of multi-input single-output nonlinear systems by means of a family of local fuzzy models. They make use of fuzzy implication statements; for the i-th model we get: if \(X_ 1\) is \(A^ i_ 1\) and \(X_ 2\) is \(A^ i_ 2\) and... and \(X_ p\) is \(A^ i_ p\) then \(y^ i=c_ 0+\sum^{p}_{j=1}c^ i_ j,x_ j\), \(i=1,2,...,N\) where \(A^ i_ 1,A^ i_ 2,...,A^ i_ p\) are fuzzy sets describing linguistic labels for successive input variables refering to the i-th model while \(y^ i\) results from a linear dependence between input and output variables. This type of dependence holds, however, only for a certain range of the values of the input variable. Then for any value of input variables \(x^ 0_ j,j=1,2,...,p\), a truth value of the condition part is derived as a product of all components, namely, \(w^ i=\prod^{p}_{j=1}A^ i_ j(x^ 0_ j)\) and the output y is calculated as \[ y=\sum^{N}_{i=1}w^ iy^ i/\sum^{N}_{i=1}w^ i. \] Structure identification for this class of models is introduced and relevant verification criteria are studied. Moreover, assuming a piecewise linear character of membership functions of fuzzy sets standing in the rules \((A^ j_ i)\), an algorithm of the adjustment of their parameters is provided.