A robust real-coded genetic algorithm using Unimodal Normal Distribution Crossover augmented by Uniform Crossover: effects of self-adaptation of crossover probabilities
Isao Ono, Hajime Kita, Shigenobu Kobayashi · 1999
This paper presents a robust real-coded genetic algorithm using the Unimodal Normal Distribution Crossover (UNDX) enhanced by the Uniform Crossover (UX). The UNDX has an advantage, which most other crossover operators do not have, that it can efficiently optimize functions with strong epistasis among parameters. However, the UNDX has a disadvantage that there can be some areas where the UNDX cannot sufficiently generates individuals by using a given initial population. Contrary to this, the UX has an advantage that it can make individuals in areas where the UNDX cannot with the same initial population and has a disadvantage that it cannot efficiently optimize functions with epistasis among parameters. To make use of the advantages of the UNDX and the UX that are complementary to each other, we introduce a mechanism of adapting the operator probabilities according to the characteristics of a given function. Through some experiments, we show the robustness of the proposed method by demonstrating that the proposed method can solve more various functions than a GA using only the UNDX.