基于双分支自适应搜索的大规模多目标优化算法
CSTR:
作者:
作者单位:

空军工程大学

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)


Large-scale multi-objective optimization algorithm based on dual-branch adaptive search
Author:
Affiliation:

Fund Project:

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

  • 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:

    针对传统搜索算子在大规模多目标优化问题中采样效率低的问题, 提出一种双分支自适应搜索(dual-branch adaptive search, DAS)策略. 该策略从种群个体质量差异出发, 基于标量适应度将种群二分为精英与落后子群, 并根据子群特点设计不同的繁殖算子: 精英子群采用协方差自适应扩散策略, 通过对精英分布的截断奇异值分解提取帕累托前沿局部主成分方向, 沿各主成分施加与方差匹配的有向扰动, 将采样聚焦于前沿附近的低维子空间, 以兼顾多样性与收敛精度; 落后子群在目标空间以余弦相似度为每个个体匹配方向最近的精英个体作为引导者, 再以自适应引导迁移算子沿引导方向迁移加速收敛; 搜索方向与步长分别自适应于种群信息和进化进度. 实验结果表明, 将DAS嵌入三类不同环境选择机制的算法后, 在LSMOP测试集的大多数对比中优于原始算法, 优势随决策变量规模增大更显著; 在8个桁架优化问题上, 与先进算法对比, 所提算法具有较强的竞争力.

    Abstract:

    This study proposes a dual-branch adaptive search (DAS) strategy to address the low sampling efficiency of traditional search operators in large-scale multi-objective optimization problems. Exploiting the quality differences among individuals, DAS partitions the population into an elite subpopulation and a lagging subpopulation based on a scalar fitness, and designs a distinct reproduction operator for each according to their characteristics. For the elite subpopulation, DAS adopts a covariance-adaptive diffusion operator. The operator extracts the local principal-component directions of the Pareto front through truncated singular value decomposition of the elite distribution, and imposes variance-matched directional perturbations along each principal component, focusing the sampling on the low-dimensional subspace near the front to balance diversity and convergence accuracy. For the lagging subpopulation, DAS selects the closest-direction elite as a guide for each individual by cosine similarity in the objective space, and then migrates each individual along this guiding direction through an adaptive guided-migration operator to accelerate convergence. The search direction and the step size adapt to population information and evolutionary progress, respectively. Experimental results show that, when embedded into three algorithms with different environmental selection mechanisms, DAS enables them to outperform their original versions in most comparisons on the LSMOP suite, and the advantage becomes more pronounced as the number of decision variables increases. On 8 truss optimization problems, the proposed algorithm is highly competitive with advanced algorithms.

    参考文献
    相似文献
    引证文献
引用本文
相关视频

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:2026-07-02
  • 最后修改日期:2026-10-29
  • 录用日期:2026-11-02
  • 在线发布日期: 2026-11-17
  • 出版日期:
文章二维码