Multiobjective optimization based on reputation
Multiobjective optimization based on reputation
复制标题
DOI：
10.1016/j.ins.2014.07.020
复制
发表时间：
2014-12
期刊：
影响因子：
--
通讯作者：
Siwei Jiang;Jie Zhang;Y. Ong
中科院分区：
文献类型：
--
作者：
Siwei Jiang;Jie Zhang;Y. Ong
为了提高进化算法的鲁棒性和易用性，进化算子和控制参数的自适应显示出显着的优势，比固定的运营商与默认的参数设置。迄今为止，许多成功的自适应进化算法的研究工作都致力于单目标优化问题（SOP），而很少有研究已经进行了多目标优化问题（MOPs）。由于这两类问题的内在差异，在MOPs背景下直接继承SOP的适应机制面临挑战。为了填补这一空白，本文提出了一种新的基于信誉的多目标进化算法（MOEA），作为一般MOEA的统一框架。信誉的概念首次引入到测量的动态能力的进化算子和控制参数的问题和阶段的搜索MOEA。基于信誉的概念，个人的解决方案，然后选择高度信誉的进化算子和控制参数。在jMetal中对58个基准MOP进行的实验研究证实了其优于经典MOEA和其他自适应MOEA的上级性能。
To improve the robustness and ease-of-use of Evolutionary Algorithms (EAs), adaptation on evolutionary operators and control parameters shows significant advantages over fixed operators with default parameter settings. To date, many successful research efforts to adaptive EAs have been devoted to Single-objective Optimization Problems (SOPs), whereas, few studies have been conducted on Multiobjective Optimization Problems (MOPs). Directly inheriting the adaptation mechanisms of SOPs in the MOPs context faces challenges due to the intrinsic differences between these two kinds of problems. To fill in this gap, in this paper, a novel Multiobjective Evolutionary Algorithm (MOEA) based on reputation is proposed as a unified framework for general MOEAs. The reputation concept is introduced for the first time to measure the dynamic competency of evolutionary operators and control parameters across problems and stages of the search in MOEAs. Based on the notion of reputation, individual solutions then select highly reputable evolutionary operators and control parameters. Experimental studies on 58 benchmark MOPs in jMetal confirm its superior performance over the classical MOEAs and other adaptive MOEAs.