Many-Objective evolutionary algorithm based on bipolar preferences dominance

Abstract

Many-Objective optimization is a difficulty for classical multi-objective evolutionary algorithm and has gained great attention during the past few years. In this paper, a dominance relation named bipolar preferences dominance is proposed for addressing many-objective problem. The proposed dominance relation considers the decision maker's positive preference and negative preference simultaneously and creates a strict dominance relation among the non-dominated solutions, which has ability to reduce the proportion of non-dominated solutions in population and lead the race to the Pareto optimal area, which is close to the positive preference and far away from negative preference. To demonstrate its effectiveness, the proposed approach was integrated into NSGA-II to be a new algorithm denoted by 2p-NSGA-II and tested on a benchmark of two to fifteen-objective test problems. Good results were obtained. The proposed dominance relation was also compared to g-dominance and r-dominance which was the most recently proposed dominance relation, the results of comparative experiment showed 2p-NSGA-II was superior to g-NSGA-II and r-NSGA-II on a whole, no matter the accuracy of obtained solutions or the efficiency of algorithm.