Bipolar preferences dominance based evolutionary algorithm for many-objective optimization

Many-objective optimization is a difficulty in the present evolutionary multi-objective optimization community. Integrating decision makers' preferences into multi-objective evolutionary algorithm is considered to be an effective approach. This paper presents a new scheme named bipolar preferences dominance for many-objective optimization problems. In the proposed scheme, the solutions are first sorted by the g-dominance to enhance the efficiency of Pareto sorting, and the non-dominated ones are sorted again based on their similarities to increase the proportion of solutions' comparability in high-dimension space. With bipolar preferences dominance, the race is led to the Pareto optimal area which is close to the positive preference and far away from the negative preference. After combining the proposed scheme with NSGA-II methodology, the effectiveness of 2p-NSGA-II was validated on two to fifteen-objective test problems. Moreover, 2p-NSGA-II provides better result when compared with g-dominance based algorithm g-NSGA-II.