Offline Automatic Parameter Tuning of MOEA/D Using Genetic Algorithm

The multiobjective evolutionary algorithm based on decomposition (MOEA/D) is one of the favorite algorithms in the evolutionary computation community. In this paper, a genetic algorithm is used to automatically tune the parameters for MOEA/D in an offline manner. We consider a version of MOEA/D with a normalization mechanism and two neighborhood structures (for mating and replacement). Our experimental results show that the automatically obtained implementation of MOEA/D outperforms MOEA/D with the default settings in their applications to the DTLZ and WFG test suites. The obtained implementation for each test problem also allows us to discover some potentially good parameter values that can lead to the performance improvement of MOEA/D on certain test problems.