Abstract: In this paper, an improved clustering-ranking method is proposed for many-objective optimization, which aims to enhance the diversity of obtained solutions. The improvement of algorithm is consist of two parts, that is two-layer weight vector system and a new clustering operator. Compared with traditional method, the two-layer system is able to create the intermediate weight vectors in the objective space. The combination of boundary layer and inner layer is able to promote the ability of diversity in the objective space. Besides, a new clustering operator is introduced, avoiding too many solutions associated in the specific weight vector. Experimental results have shown the effectiveness of the proposed algorithm. Compared to original algorithm and two other algorithms, the proposed method shows highly competitive performance on test problems with various characteristics.

Key words: many-objective optimization, diversity, evolutionary algorithm, clustering operator

摘要： 针对高维多目标优化问题提出一种改进型的聚类排序算法，旨在提升原算法所得解的多样性。对该算法的改进，主要集中在两方面。首先，引入了一种双层权值向量系统。相对于原始权值向量方法，该方法可以建立目标空间当中的内部权值向量。内部向量与边缘权值向量的合并，可以促进整体权值向量的多样性。此外，引入一种新的聚类算子，可避免特定权值向量中附着过多的解。实验结果表明，相对比于原始的聚类排序算法和其他两种对比算法，所提出的算法在不同特性的测试问题上具有较好的性能。

关键词: 多目标优化, 多样性, 演化算法, 聚类算子