Estimation distribution algorithms with differential mutation for multi-objective optimization problems

Estimation distribution algorithms (EDAs) have been widely used in single objective optimization problems. In this paper EDAs are combined with differential mutation (DM) to find Pareto optimal front for multi-objective optimization problems (MOPs). First, a modified extreme elitism selection method is used to choose some promising solutions as the parent solution. This selection represents some leading best solutions in the evolution to make EDAs form a primary evolutionary direction. Also, DM leads to a diversified population, which helps the algorithm to avoid premature convergence. A set of benchmark MOPs are used to test the proposed EDA/DM algorithm. The experimental results demonstrate that EDA/DM can achieve better performance on some MOPs than by several state-of-the-art multi-objective evolutionary algorithms.