LETTER Reproduction Strategy Based on Self-Organizing Map for Real-coded Genetic Algorithms

A Real-coded genetic algorithm (RcGA) [1][2] is a modified genetic algorithm (GA) [3] employing realvalued vectors for representation of the chromosomes, and is widely-applied to many optimization problems [4]. The RcGA generally employs a reproduction strategy based on roulette wheel selection (RWS). However, this strategy may lose genetic diversity of population in an early stage [5], because it can not generate new chromosomes which are different from present chromosomes . In this paper, we propose a novel reproduction strategy employing an idea of Self-Organizing Map (SOM) [6][7] to cope with this problem. In the proposed reproduction, a set of new chromosomes of next generation is generated by learning of the SOM. The updating equation used in learning of the SOM is modified by adding new coefficients with respect to fitness values of chromosomes. The continuous update facilitates the preservation of genetic diversity and the effective search. The effectiveness of the proposed reproduction is verified by applying it to benchmark optimization problems of DeJong.