A New Comparison Function Based Direct Multisearch Method for Derivative-Free Multi-objective Optimization Problems

In this paper, for addressing Multi-Objective Derivative-Free Optimization (MODFO) problems with box constraints, a new direct multisearch algorithm named CmpFDMS is proposed. The method dose not aggregate objectives and uses the concept of Pareto dominance to update the list of non-dominated points. Same as the traditional direct search method for solving MODFO, this method consists of a search step and a poll step. To generate Pareto sets with high-performance, we introduce a new comparison function on the basis of Pareto dominance. The new comparison function can compare a pair of points against a background of non-dominated points. Computational experiments demonstrate that the proposed CmpFDMS competes effectively with the state-of-the-art direct multisearch algorithm known as DMS in the context of multi-objective black-box optimization.