An adaptive neighborhood-based multi-parent crossover operator for real-coded genetic algorithms

In this paper, we propose the adaptive neighborhood-based multi-parent cross-over operator (ANMX), a novel search operator for real-coded genetic algorithms, which generates offspring by the linear non-convex combination of several relative-parents, each of which is uniformly sampled in the respective neighborhoods of randomly selected parents in the population. ANMX is a general variation operator in essence, for it not only takes on the form of a multi-parent crossover operator but also implicitly has some of the characteristics of a mutation operator. To enhance the efficiency, we introduce a mechanism for adapting the range of neighborhoods according to the evolutionary progress. Numerical experiments using a suite of test functions, which are widely studied in the field of evolutionary computation, show good search ability of the proposed operator for functions with multimodality and/or epistasis.