Fitness diversity adaptation in memetic algorithms

Abstract

This work proposes novel tailored implementations of Memetic Algorithm for some specific classes of problems and, at the same time, proposes novel general ideas in Computational Intelligence algorithmic philosophy. Much emphasis is given to adaptation and coordination of the local searchers. Several adaptive schemes have been designed; all of them resorting to a measurement of the fitness diversity as an estimation of the diversity amongst individuals of the population. According to the philosophy common to the algorithms included in this thesis, the algorithm should behave like an intelligent structure, thus able to analyze online the optimization process and then apply countermeasures necessary for continuing and efficiently finalizing the search. This thesis includes eight articles addressing applications having various natures such as biology, image processing, telecommunication, and electrical engineering. Each problem has been analyzed by considering the features of each fitness landscape being handled and each fitness function being optimized. The resulting algorithms seem to have promising performance in terms of final solution detected and convergence velocity. Extended numerical experiments have been carried out in each case in order to show the statistical significance of results.