A Study on Large Population MOEA Using Adaptive ε-Box Dominance and Neighborhood Recombination for Many-Objective Optimization

Multi-objective evolutionary algorithms are increasingly being investigated to solve many-objective optimization problems. However, most algorithms recently proposed for many-objective optimization cannot find Pareto optimal solutions with good properties on convergence, spread, and distribution. Often, the algorithms favor one property at the expense of the other. In addition, in some applications it takes a very long time to evaluate solutions, which prohibits running the algorithm for a large number of generations. In this work to obtain good representations of the Pareto optimal set we investigate a large population MOEA, which employs adaptive

ε

-box dominance for selection and neighborhood recombination for variation, using a very short number of generations to evolve the population. We study the performance of the algorithm on some functions of the DTLZ family, showing the importance of using larger populations to search on many-objective problems and the effectiveness of employing adaptive

ε

-box dominance with neighborhood recombination that take into account the characteristics of many-objective landscapes.