Computer Engineering and Applications ›› 2018, Vol. 54 ›› Issue (14): 1-6.DOI: 10.3778/j.issn.1002-8331.1803-0114

Previous Articles     Next Articles

Decomposition multiobjective optimization algorithm with new neighborhood model

LI Zhixiang, LI Yun, HE Liang   

  1. National Key Laboratory of Science and Technology on Blind Signal Processing, Chengdu 610041, China
  • Online:2018-07-15 Published:2018-08-06


李智翔,李  赟,贺  亮   

  1. 盲信号处理重点实验室,成都 610041

Abstract: In multiobjective evolutionary algorithm based on decomposition, the solution used for reproduction is chosen from neighbor set defined by subproblems. However, when the test instance has complex features like multi peaks, the solutions may be far away in the decision space, which will lead to poor performance of the algorithm. A new algorithm based on new neighborhood model is proposed, which is called MOEA/D-NN. First, the neighbor model for reproduction operator is redesigned to calculate the neighborhood relation by the distance of solutions in decision space, and then realizes the reproduction calculation based on the newly defined neighborhood sets, which are saved for every subproblems and refreshed at regular intervals to ensure that the definition of the neighbor meets the needs of reproduction operator. The experimental results on a variety of test instances show that the proposed algorithm is competitive in comparison with other state-of-the-art multiobjective evolutionary algorithms.

Key words: multiobjective optimization, decomposition method, reproduction operator, neighborhood set

摘要: 在通常的基于分解的多目标进化算法中,繁殖计算时使用的解从基于子问题定义的邻居集合中选择,当目标函数存在多峰等复杂特征时,它们在决策空间的距离可能较远,这会导致算法性能变差。为了解决这一问题,提出了一种采用新邻居模型的多目标分解进化算法MOEA/D-NN。该算法重新设计了繁殖计算中使用的邻居模型,利用解在决策空间上的距离计算邻居,进而为每个子问题维护相应的邻居集合,在此基础上对邻居集合进行定时更新,实现了基于新邻居模型的繁殖计算。通过在公开测试集上的实验结果表明,提出的算法与几种经典的多目标进化算法相比,在大多数测试集上表现更优。

关键词: 多目标优化, 分解方法, 繁殖计算, 邻居集合