高维多目标进化算法中的密度评估策略研究  

Density estimation strategies in high-dimensional MOEAs

作　　者：黄林峰[1,2] 罗文坚[1,2] 王煦法[1,2] 

机构地区：[1]中国科学技术大学计算机科学与技术学院自然计算与应用实验室,安徽合肥230027 [2]安徽省计算与通讯软件重点实验室,安徽合肥230027

出　　处：《中国科学技术大学学报》2011年第4期353-361,共9页JUSTC

基　　金：国家自然科学基金委海外青年学者合作研究基金(60428202)资助

摘　　要：多目标进化算法中常引入密度评估策略来使算法获得更好的分布性和收敛性.但对于高维多目标问题,现有的密度评估策略却难于达到这一目的.为此更全面地考虑目标空间上各子目标的影响,提出了四种新的密度评估策略,并将其应用到经典多目标进化算法SPEA2中.在4～9个目标的多目标背包问题上的实验结果表明,采用新的密度评估策略的SPEA2算法能更有效地收敛到Pareto前沿.A density estimation strategy is often adopted in order to guarantee better distribution and convergence in MOEA. But the current density estimation strategies cannot achieve this goal when the number of objectives become large. Each objective was more generally considered and four novel strategies of density estimation were proposed. Then, they were applied in SPEA2, which was one of the classical MOEAs. The experimental results of the test cases of MOKP with 4 to 9 objectives show that SPEA2 with the novel strategies have better convergence to the Pareto front on all test cases.

关 键 词：多目标优化 多目标0/1背包问题 多目标进化算法 密度评估策略 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]