Effectiveness of neighborhood crossover in multiobjective genetic algorithm

In this paper, the effectiveness of the neighborhood crossover of EMO algorithms is discussed through the numerical experiments. The neighborhood crossover chooses two parents which are close to each other in the objective space. All the individuals are sorted in order of proximity in the objective space, and then neighborhood shuffle is conducted, which randomly replaces individuals in the population at a fixed width interval of population size. This operation prevents crossing over repeatedly between the same pair of individuals.The width of neighborhood shuffle is the parameter of this operation and this parameter determines the range of the population where individuals are shuffled. Therefore, this parameter affects the quality of the solutions. We implemented the NSGA-II with the neighborhood crossover and examined the effect of the width of neighborhood shuffle to further investigate the effectiveness of neighborhood crossover. The results of the numerical experiment indicated that the effect of neighborhood crossover can be achieved by applying neighborhood crossover to the search population created through copy selection. In addition, the necessity of neighborhood shuffle and an appropriate width of neighborhood shuffle were reviewed.