Striking a Mean- and Parent-Centric Balance in Real-Valued Crossover Operators

This paper investigates the mean- and parent-centric balance in real-valued crossover operators, which is strongly related to the powerful and efficient optimization performance. To treat the property as a continuous value, a novel crossover operator, called asymmetrical normal distribution crossover (ANDX), has been introduced. Because the crossover operator has a tunable parameter for the mean- and parent-centric balance, an arbitrary continuous balance is achievable, whereas in previous studies, the property has been treated dualistically. Through numerically empirical analysis with ANDX, the relationship between optimization performance and balance was clearly observed by changing the balance at regular intervals. To determine a practically suitable first choice of balance, a performance comparison with various parameter settings of ANDX was conducted on large-scale objective functions. The experimental results demonstrate that we should consider accepting mean-centric crossover operators as a realistic first choice in practice.