基于新的网格存优策略的多目标归档算法

EMO archive algorithm using new grid-based elitist-reserving strategy

网格方法被多个进化算法用来保持解集的分布性。基于ε支配概念的ε-MOEA本质上也是基于网格策略的。虽然ε-MOEA通常情况下都能在算法性能的各方面之间取得较为合理的折衷,但是由于其存在固有缺陷,很多时候表现出不容忽视的问题——当PFtrue对某一维的变化率在该维不同区域的差异较大时,解集中边界个体或代表性个体丢失——严重影响解集的分布性。针对这一问题,定义了一种新的δ支配概念和虚拟"最优点"的概念,提出了一种新的网格存优策略,并将之应用于更新进化多目标归档算法的归档集。实验结果显示,基于新的存优策略的进化多目标归档算法(δ-MOEA)具有良好的性能,尤其在分布性方面比NSGA2和ε-MOEA好得多。

Grid-based measure is commonly used to maintain diversity in many MOEAs.The ε-MOEA,which is based on the ε- dominance concept,is essentially based on grid-strategy.Though Often gaining an appropriate tradeoff between the aspects of the performance,the ε-MOEA has its inherent vice and behaves unacceptably sometimes.That is,when the slope to one dimension of the PFtrue changes a lot along it,the algorithm loses many extreme or representative individuals,which has a severely influence on the diversity of the solution set.In order to solve this problem,a new δ-dominance concept and suppositional optimum point concept is defined,then a new grid-based elitist-reserving strategy is proposed,finally it is applied in an EMO archive algorithm（δ-MOEA）.The experimental results illustrate δ-MOEA＇s good performance,which is much better especially at the diversity than NSGA2 and ε-MOEA.