摘要
在多目标进化算法(multiobjective evolutionary algorithms,MOEAs)的文献中,对算法的选择策略进行系统研究的还很少,而MOEAs的选择策略不仅引导算法的搜索过程、决定搜索的方向而且对算法的收敛性有重要的影响,它是算法能否成功求解多目标优化问题的关键因素之一。在统一的框架下,首先讨论了多目标优化问题中适应度函数的构造问题,然后根据MOEAs的选择机制和原理将它们的选择策略重新分成了6种类型。一般文献中很少对多目标进化算法的操作算子采用符号化描述,这样不利于对算子的深层次理解,符号化描述了各类选择策略的操作机制和原理,并分析了各类策略的优劣性。最后,从理论上证明了具备一定特征的多目标进化算法的收敛性,证明的过程表明了将算法运行终止时得到的Pknown作为多目标优化问题的Pareto最优解集或近似最优解集的合理性。
It is scarce for literatures devoted to the multiobjective evolutionary algorithms （MOEAs） to systematically research on selection strategies, however, these strategies are crucial to MOEAs for solving some multiobjective optimization problems successfully, as they not only guide the process of search and determine the search directions, but also exert great effect on the convergence of MOEAs. With the unified framework, the paper first discussed how to construct an appropriate fitness function in multiobjective optimization problem, then, selection strategies were classified as six categories based on MOEA＇s selection mechanism and principle through systematically analyzing various MOEAs. As it is rare for expressing the operators of the MOEAs symbolized in most literatures, which is not conducive to comprehend them deeply. This paper described the principle and mechanism of each selection strategy symbolized and analyzed its advantages and weaknesses respectively. At last, the paper proved the convergence of MOEAs with certain features, and the process of proof has shown that it is reasonable to regard Pknown achieved from the final results of MOEAs as Ptrue or the approximated Pareto optimal set.
出处
《计算机科学》
CSCD
北大核心
2009年第9期167-172,共6页
Computer Science
基金
教育部博士点基金项目(编号:20070486081)资助
关键词
多目标进化算法
适应度函数
选择策略
收敛性
Multiobjective evolutionary algorithms, Fitness function,Selection strategy,Convergence
