Towards the Defuzzification Procedure in an ANFIS

Abstract

The paper pays attention to particularities concerned with the defuzzification procedure in adaptive neural network based fuzzy inference systems. Specifically, the problem of constructing recursive parameter estimation algorithms is considered with regard to their convergence and stability. In the combination with various forms of the quadratic criterion, such an approach enables on to obtain strongly consistent estimation algorithms under essential generality of modeled system description; and the convergence properties are demonstrated in the comparison with conventional algorithms.