基于双重贡献分配的多目标混合算子进化算法
作者:
作者单位:

南京信息工程大学

作者简介:

通讯作者:

中图分类号:

TP273

基金项目:

国家自然科学基金项目(面上项目,重点项目,重大项目),国家重点研发计划


Multi-objective optimization algorithm based on double credit assignment
Author:
Affiliation:

Nanjing University of Information Science & Technology

Fund Project:

The National Natural Science Foundation of China (General Program, Key Program, Major Research Plan), National Key R&D Program of China

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 资源附件
  • |
  • 文章评论
    摘要:

    针对多目标混合算子进化算法中各算子有效选择的自适应问题,提出了一种基于双重贡献分配的多目标混合算子进化算法(DCA-MOEA/D).首先采用两种现有的进化算子与两种基于方向引导的差分进化组成算子池,每代个体以轮盘赌的方式从中选择一种进化算子产生子代;然后根据子代的表现,结合两种方法来为各算子分配贡献值,从而确定算子的选择概率;接着引入外部归档集,根据支配关系与拥挤度策略来对其进行维护;最后将整个进化过程划分为5个阶段,以达到算子选择中"探索"与"探究"之间的平衡.通过与其他4种多目标进化算法在23个测试函数上的对比,以IGD与HV为性能评价指标,表明所提出算法在收敛性与分布性上具有显著优势.

    Abstract:

    In order to make the operator selection more efficient in multi-objective evolutionary algorithms (MOEAs) with multiple operators, this paper proposes a MOEA based on decomposition with double credit assignment (DCA-MOEA/D). First, the operator pool in proposed algorithm consists of two existing operators and two variants of differential evolution (DE) based on direction-guided search strategy. Individuals use a roulette wheel-like process to pick up an operator to generate offspring at each generation. Subsequently, the credit value of each operator is determined by combining two credit assignment methods according to the performance of offspring, and the selection probability of each operator is updated by the credits. Meanwhile, an extra archive is defined and uses non-dominated sorting and crowded distance strategies to maintain it. And finally, the whole evolutionary process is divided into several steps to achieve the balance between exploration and exploitation in operator selection. Empirical study validates the effectiveness of our proposal through the contrast experiment with four MOEAs in terms of IGD and HV value on 23 benchmark problems.

    参考文献
    相似文献
    引证文献
引用本文
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:2020-09-14
  • 最后修改日期:2021-03-11
  • 录用日期:2021-03-16
  • 在线发布日期: 2021-04-01
  • 出版日期: