求解DRS多目标优化问题的新颖ε-支配进化算法  

Novelε-dominance multi-objective evolutionary algorithms for solving DRS multi-objective optimization problems

作　　者：李敏强[1] 刘鎏[1] 林丹[2] 

机构地区：[1]天津大学管理学院,天津300072 [2]天津大学理学院,天津300072

出　　处：《系统工程学报》2008年第1期101-105,共5页Journal of Systems Engineering

基　　金：国家自然科学基金(70571057；70171002);新世纪优秀人才支持计划资助项目(NCET-05-0253)

摘　　要：首先分析DRS(dominance resistant solutions)多目标优化问题的特点,证明基于Pareto-支配关系的多目标优化问题算法求解该类问题很难收敛.然后,提出一种新的基于ε-支配关系的进化算法—ε—支配进化算法(EDMOEA),给出该算法框架和详细流程.最后,将ε-支配进化算法和NSGA-Ⅱ算法应用于求解一组典型的DRS多目标优化问题和常用的多目标优化测试问题,基于算法的收敛性和Pareto最优解集分布性进行评价和比较分析,表明ε-支配进化算法的有效性.The DRS（ dominance resistant solutions）existing in the multi-objective optimization problems （MOP） is characterized, and it is proved that the evolutionary algorithms fail to find the true Pareto fronts for the DRS MOP. A new evolutionary algorithm based on ε-dominance relationship, called the ε-dominance MOEA （EDMOEA） , is proposed to solve the DRS MOP, and algorithms procedure is designed. Then it is proved that the EDMOEA is able to fix the DRS issue of the DRS MOP. Finally, a set of DRS MOP and other MOP commonly used in literatures are adopted to test and compare the EDMOEA and the NSGA-Ⅱ, and the experiment results illustrate that the EDMOEA is superior to the NSGA-Ⅱ on the MOP of both the DRS and others.

关 键 词：进化算法 多目标优化问题 ε-支配 PARETO支配 

分 类 号：O22[理学—运筹学与控制论]