A Comparative Study on Evolutionary Algorithms for Many-Objective Optimization
A Comparative Study on Evolutionary Algorithms for Many-Objective Optimization
复制标题
DOI：
10.1007/978-3-642-37140-0_22
复制
发表时间：
2013-03
期刊：
影响因子：
--
通讯作者：
M. Li;Shengxiang Yang;Xiaohui Liu;R. Shen
中科院分区：
文献类型：
--
作者：
M. Li;Shengxiang Yang;Xiaohui Liu;R. Shen
多目标优化问题在进化多目标优化领域得到了越来越多的关注，近年来已经开发了各种方法来解决多目标问题。然而，现有的实证比较研究往往局限于少数几个测试问题的方法。本文从六个角度对八种有代表性的方法进行了系统的比较，以解决多目标问题。四组定义良好的连续和组合的测试功能，通过三个性能指标，以及在决策空间中的视觉观察比较的方法进行测试。我们的结论是，没有一种方法有明显的优势，比其他的，虽然他们中的一些是竞争力的大部分问题。此外，这些方法对不同特征问题的搜索能力不同，这表明在解决多目标问题时要谨慎选择方法。
Many-objective optimization has been gaining increasing attention in the evolutionary multiobjective optimization community, and various approaches have been developed to solve many-objective problems in recent years. However, the existing empirically comparative studies are often restricted to only a few approaches on a handful of test problems. This paper provides a systematic comparison of eight representative approaches from the six angles to solve many-objective problems. The compared approaches are tested on four groups of well-defined continuous and combinatorial test functions, by three performance metrics as well as a visual observation in the decision space. We conclude that none of the approaches has a clear advantage over the others, although some of them are competitive on most of the problems. In addition, different search abilities of these approaches on the problems with different characteristics suggest a careful choice of approaches for solving a many-objective problem in hand.