Genetic Algorithm crossover strategy for enhanced solution space exploration
Kurt S. Anderson, YuHung Hsu · 7th AIAA/USAF/NASA/ISSMO Symposium on Multidisciplinary Analysis and Optimization · 1998
Without prior knowledge of the design or performance hyperplane, an optimizer with robust capability in both exploration and exploitation is a critical requirement for effective optimization. GAs are considered a flexible and powerful candidate in this field and have been broadly adapted as function optimizers. The key aspects of GAs as function optimizers are their domain independent technique which do not restrict to particular problem formats and their ability to generate and test various parameter combination efficiently. However, GAs using single point crossover strategy to perform parameter value recombination exhibit several flaws which greatly restrain the GAs exploration capability. This paper proposes a family of crossover strategies which emphasize parameter recombination so to enhance solution space exploration, yet are not so disruption as to defeat the local exploitation aspects of the algorithm. The empirical results show that GAs with the proposed crossover strategies can identify optimal or near optimal regions of the design hyperplane with fewer generations than traditional single-point crossover.