基于动态ε支配的多目标遗传算法

Multiobjective Genetic Algorithm based on dynamic ε dominance

基于Pareto支配的MOEA存在着一些缺陷,如容易出现退化现象等。而基于ε支配的MOEA可以比较好地解决这些问题,并具有比较理想的收敛性和分布性。但是采用传统的ε-MOEA时,最大的困难就是ε的值的设定,并且传统的MOEA得出的解在边界部分个体的丢失现象也比较严重。针对这种情况提出了一种新的基于动态ε支配的多目标遗传算法(DEMOEA),它不需要手动设定ε的值,并且引入了动态网格概念来改善边界解丢失的现象。通过与其他两个经典的多目标进化算法的NSAGA-Ⅱ和SPEA-2的对比实验,表明提出的DEMOEA能在收敛性、分布性有较好的改进。

There are some limitations in MOEA based on Pareto dominance,such as it is easy to degraded and so on.Then the MOEA based on ε-dominance can solve these problems and it can make a preferable convergence and spread.But in the conventional ε-MOEA,the most difficult problem is the setting of ε and the loss of part of the extreme individuals is serious.In order to solve these problems,this paper proposes a new ε-MOEA based on dynamic ε （DEMOEA）,it doesn＇t need to set the ε by yourself and this paper imports a concept of dynamic grid to solve the loss of extreme individuals.Comparing with two other classical algorithms NSGA-Ⅱ and SPEA2 in experiment,the result shows that the algorithm suggested in the paper（DEMOEA） gets improved convergence and diversity.,