关键词: 高维多目标优化, Pareto支配关系, 多样性

Abstract: People often need to optimize many conflicting objectives at the same time both in engineering application and real life. These optimization problems are called many-objective optimization problems (MaOPs). The evolutionary algorithms based on Pareto dominance will lose the convergence pressure when dealing with MaOPs. The decomposition based algorithm needs a set of preset reference vectors in objective space, which make the poor robustness when dealing with MaOPs with irregular PFs. In order to improve the convergence and maintain the flexible diversity strategy in dominance based algorithms, this paper proposes a two phase many-objective optimization algorithm based on Pareto dominance. The proposed algorithm centralizes computing resources to search the nadir point of optimization problems in the first phase. Then the inner and outward space can be divided according to the nadir point, and the convergence of algorithms can be improved by selecting the solutions in the inner space. In the second phase, the dynamic minimal distance method is proposed to improve the diversity of the algorithm. Compared with four state-of-art many-objective optimization algorithms, the experimental results show that the performance of the proposed algorithm is significantly better than that of compared algorithms on the MaOPs with irregular PFs. Meanwhile, the proposed algorithm also obtains competitive results on MaOPs with regular PFs. In a word, these experimental results indicate that the proposed algorithm has better robustness.

Key words: many-objective optimization, Pareto dominance relationship, diversity