| 本文已被：浏览 3157次 下载 5579次 |
码上扫一扫！ |
|
|
| 一种基于分解和协同的高维多目标进化算法 |
|
谢承旺1, 余伟伟2, 闭应洲1, 汪慎文3, 胡玉荣4
|
|
1.南宁师范大学 计算机与信息工程学院, 广西 南宁 530299;2.北京工业大学 软件学院, 北京 100124;3.河北地质大学 信息工程学院, 河北 石家庄 050031;4.荆楚理工学院 科技处, 湖北 荆门 448000
|
|
| 摘要: |
| 现实中大量存在的高维多目标优化问题对以往高效的多目标进化算法提出了严峻的挑战.通过将分解策略和协同策略相结合提出一种高维多目标进化算法MaOEA/DCE.该算法利用混合水平正交实验方法在聚合系数空间产生一组均匀分布的权重向量以改善初始种群的分布性；其次，算法将差分进化算子和自适应SBX算子进行协同进化，以产生高质量的子代个体，并改善算法的收敛性.该算法与另外5种高性能的多目标进化算法在基准测试函数集DTLZ{1，2，4，5}上进行对比实验，利用改进的反转世代距离指标IGD+评估各算法的性能.实验结果表明，MaOEA/DCE算法与其他对比算法相比，在总体上具有较为显著的收敛性和分布性优势. |
| 关键词: 高维多目标优化 分解策略 混合水平正交实验设计 高维多目标进化算法 |
| DOI：10.13328/j.cnki.jos.005617 |
| 分类号:TP181 |
| 基金项目:国家自然科学基金（61763010，61402481，61165004）；广西八桂学者项目；河北青年拔尖人才支持计划（冀字[2013]17）；河北省自然科学基金（F2015403046）；河北省教育厅科技重点项目（ZD2018083）；湖北省教育厅科研项目（B2015240）；荆楚理工学院科学研究重点基金（ZR201402）；荆楚理工学院科学研究引进人才科研启动金（QDB201605） |
|
| Many-objective Evolutionary Algorithm Based on Decomposition and Coevolution |
|
XIE Cheng-Wang1, YU Wei-Wei2, BI Ying-Zhou1, WANG Shen-Wen3, HU Yu-Rong4
|
|
1.School of Computer and Information Engineering, Nanning Normal University, Nanning 530299, China;2.School of Software Engineering, Beijing University of Technology, Beijing 100124, China;3.School of Information Engineering, Hebei Geo University, Shijiazhuang 050031, China;4.Department of Science and Technology, Jingchu University of Technology, Jingmen 448000, China
|
| Abstract: |
| In real-world, there exist lots of many-objective optimization problems (MaOPs), which severely challenge well-known multi-objective evolutioanry algorithms (MOEAs). A many-obective evolutioanry algorithm combining decomposition and coevolution (MaOEA/DCE) is presented in this paper. MaOEA/DCE adopts mix-level orthogonal experimental design to produce a set of weight vectors evenly distributed in weight coefficient space, so as to improve the diversity of initial population. In addition, the MaOEA/DCE integrates differential evolution (DE) with the adaptive SBX operator to generate high-quality offspring for enhancing the convergence of evolutionary population. Some comparative experiments are conducted among MaOEA/DCE and other five representative MOEAs to examine their IGD+ performance on four MaOPs of DTLZ{1,2,4,5}. The experimental results show that the proposed MaOEA/DCE has overall performance advantage over the other peering MOEAs in terms of convergence, diversity, and robustness. |
| Key words: many-objective optimization decomposition strategy mix-level orthogonal experimental design many-objective evolutionary algorithm |
