Multiobjective evolutionary algorithm based on mixture Gaussian models

Recombination operators used in most current multiobjective evolutionary algorithms (MOEAs) were originally designed for single objective optimization. This paper demonstrates that some widely used recombination operators may not work well for multiobjective optimization problems (MOPs), and proposes a multiobjective evolutionary algorithm based on decomposition and mixture Gaussian models (MOEA/D-MG). In the algorithm, a reproduction operator based on mixture Gaussian models is used to model the population distribution and sample new trails solutions, and a greedy replacement scheme is then applied to update the population by the new trial solutions. MOEA/D-MG is applied to a variety of test instances with complicated Pareto fronts. The extensive experimental results indicate that MOEA/D-MG is promising for dealing with these continuous MOPs.