Towards a generic control strategy for Evolutionary Algorithms: an adaptive fuzzy-learning approach

This paper presents a new method to generalize strategies in order to control parameters of Evolutionary Algorithms (EAs). A learning process establishes the relationship between optimal quality parameters and diversity, and simplifies control to just one variable, highly correlated with Exploration/Exploitation Balance, in such way that strategies can be defined in more abstract terms. The acquired knowledge is expressed in a simple fashion that helps the user to understand internal mechanics of EA. The model is built after a careful example gathering and encoded in Fuzzy Logic Controllers.