Parameter Tuning Boosts Performance Of Variation Operators In Multiobjective Optimization

Typically, the variation operators deployed in evolutionary multiobjective optimization algorithms (EMOA) are either simulated binary crossover with polynomial mutation or differential evolution operators. This empirical study aims at the development of a sound method how to assess which of these variation operators perform best in the multiobjective context. In case of the S-metric selection EMOA our main findings are: (1) The performance of the tuned operators improved significantly compared to the default parameterizations. (2) The performance of the two tuned variation operators is very similar. (3) The optimized parameter configurations for the considered problems are very different.