A Fast,Noniterative Method to Generate Fuzzy Inference Rules From Observed Data

Abstract

This note describes a noniterative method to generate a fuzzy model of a nonlinear system using observed data. This method works on multiple input, single output systems. It produces a series of rules of the form “IF X₁ is A_j,1 and X₂ is A_j,2 and … and X_n is A_j,n THEN Y is S_j,” where simple membership functions fuzzify all antecedent variables and a weighted sum of singletons is the single conclusion variable. The maximum number of rules is known at the start of the procedure, but the rules are pruned to an independent robust set by a modal reduction method. This innovative procedure is demonstrated successfully with several linear and nonlinear equations.