A real-coded Genetic Algorithm using distance dependent alternation model for complex function optimization
Osamu Takahashi, Hajime Kita, Shigenobu Kobayashi · Tokyo Tech Research Repository (Tokyo Institute of Technology) · 2000
The multi-parental Unimodal Normal Distribution Crossover (UNDX-m) that was proposed by Ono et al. and extended by Kita et al for real-coded Genetic Algorithms (GA). shows an excellent performance in optimization problems of highly epistatic fitness functions in continuous search spaces. The UNDX-m is a crossover operator that preserves the statistics such as the mean vector and the covariance matrix of the population well. While the crossover operator preserves the statistics of the population, an alternation model used with the crossover is needed to evolve the population through the alternations of the individuals in the population to progress a search. We proposed a distance dependent alternation (DDA) model, which is based on alternations of the elite child with the nearest parent in the family, to progress a search maintaining a diversity of a population. In this paper we show a real-coded GA using the UNDX-m combined with DDA model robustly solves 30-dimensional Fletcher-Powell function which is highly multi-modal and has similarity to real-world problems, and has never been solved by every other Evolutionary Algorithms.