引用本文:田瑾然,刘建昌,张伟,田家鑫,刘圆超.基于指标和自适应分解的高维多目标进化算法[J].控制理论与应用,2026,43(5):937~950.[点击复制]
TIAN Jin-ran,LIU Jian-chang,ZHANG Wei,TIAN Jia-xin,LIU Yuan-chao.A many-objective evolutionary algorithm based on indicator and adaptive decomposition[J].Control Theory & Applications,2026,43(5):937~950.[点击复制]
基于指标和自适应分解的高维多目标进化算法
A many-objective evolutionary algorithm based on indicator and adaptive decomposition
摘要点击 579  全文点击 34  投稿时间：2024-06-02  修订日期：2025-08-02
查看全文  查看/发表评论  下载PDF阅读器   HTML
DOI编号  10.7641/CTA.2025.40307
  2026,43(5):937-950
中文关键词  高维多目标优化  进化算法  Pareto前沿  自适应分解
英文关键词  many-objective optimization  evolutionary algorithm  Pareto fronts  adaptive decomposition
基金项目  国家自然科学基金项目(62273080), 高等学校学科创新引智计划“111计划”项目(B16009)资助.
作者单位E-mail
田瑾然 东北大学信息科学与工程学院 1628954105@qq.com 
刘建昌* 东北大学信息科学与工程学院 liujianchang@ise.neu.edu.cn 
张伟 东北大学信息科学与工程学院  
田家鑫 东北大学信息科学与工程学院  
刘圆超 东北大学信息科学与工程学院  
中文摘要
      针对基于分解的高维多目标进化算法在求解不同类型Pareto前沿的问题时综合性能下降的问题, 本文提出 一种基于指标和自适应分解的高维多目标进化算法(MaOEA-IAD). 首先, 设计一个实时维护的外部档案库来保留 有期望个体, 以发现潜在的未开发但有前景的参考向量; 同时, 为了评价个体的收敛性和多样性, 提出一种小生境技 术与收敛性指标相结合的综合性能指标I(x); 此外, 为了提升算法处理不同Pareto前沿问题时的综合性能, 提出一种 参考向量自适应策略, 并与基于空间划分的精英替代策略相结合, 进一步提升算法在真实Pareto前沿上的分散能力. 实验结果表明, 所提算法能够在不同复杂类型的高维多目标优化问题上有效的平衡种群收敛性和多样性, 并具有一 定的优越性.
英文摘要
      In response to the issue of diminished overall performance of decomposition-based many-objective evolutionary algorithms (MaOEAs) when solving problems with various types of Pareto fronts (PF), a many-objective evolutionary algorithm based on indicator and adaptive decomposition is proposed. Firstly, an externally maintained archive is designed to retain promising individuals, with the aim of discovering potentially underdeveloped yet promising reference vectors. Simultaneously, to evaluate the convergence and diversity of individuals, a comprehensive performance indicator I(x) is proposed, which integrates a niche technology with a convergence metric. Furthermore, an adaptive reference vector strategy is introduced to enhance the algorithm’s comprehensive performance when dealing with different PF many-objective optimization problems (MaOPs). This strategy is combined with an elite replacement strategy based on spatial division to further improve the algorithm’s coverage on the true PF. Experimental results demonstrate that the proposed algorithm can effectively balance population convergence and diversity on various complex types of MaOPs, and has certain superiority.