Rule learning in fuzzy systems using evolutionary programs
John C. Goddard, Romeo Urbieta Parrazales, Iván Barriuso López, A.D. de Luca P · 2002
The present paper considers the problem of automatically learning a set of optimised rules and membership functions from data, for the case of a rule-based fuzzy controller. The method applies evolutionary programs in a two step fashion. The first step produces the singleton conclusions for a reduced set of rules using symmetric triangular membership functions for the fuzzy variables in the premises. The second step then adjusts the triangular membership functions, whilst maintaining the fixed rules obtained in the first step. The steps are illustrated using a simulated DC motor. We present comparisons of the method for different sized rule-bases showing the average rule reduction obtained, and finally consider the problem of individual rule importance.