《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (4): 1199-1210.DOI: 10.11772/j.issn.1001-9081.2025050534

• 先进计算 • 上一篇    

基于子区域的超多目标进化算法

郭京蕾1(), 刘诗源1, 姜守勇2   

  1. 1.华中师范大学 计算机学院,武汉 430079
    2.中南大学 自动化学院,长沙 410083
  • 收稿日期:2025-05-16 修回日期:2025-07-13 接受日期:2025-07-23 发布日期:2025-08-01 出版日期:2026-04-10
  • 通讯作者: 郭京蕾
  • 作者简介:刘诗源(2001—),男,河南周口人,硕士研究生,主要研究方向:计算智能、智能优化
    姜守勇(1988—),男,湖北襄阳人,教授,博士,主要研究方向:计算智能、机器学习、生物计算。
  • 基金资助:
    国家自然科学基金资助项目(62376288);湖南省自然科学基金资助项目(2024JJ5441)

Subregion-based many-objective evolutionary algorithm

Jinglei GUO1(), Shiyuan LIU1, Shouyong JIANG2   

  1. 1.School of Computer Science,Central China Normal University,Wuhan Hubei 430079,China
    2.School of Automation,Central South University,Changsha Hunan 410083,China
  • Received:2025-05-16 Revised:2025-07-13 Accepted:2025-07-23 Online:2025-08-01 Published:2026-04-10
  • Contact: Jinglei GUO
  • About author:LIU Shiyuan, born in 2001, M. S. candidate. His research interests include computational intelligence, intelligent optimization.
    JIANG Shouyong, born in 1988, Ph. D., professor. His research interests include computational intelligence, machine learning, biological computing.
  • Supported by:
    National Natural Science Foundation of China(62376288);Natural Science Foundation of Hunan Province(2024JJ5441)

摘要:

针对多目标进化算法(MOEA)求解超多目标优化问题(MaOP)难以平衡收敛性和多样性的困境,提出一种基于子区域的超多目标进化算法(SR-MaOEA)。首先,通过子区域划分策略构建目标空间的分布结构,采用偏移密度估计(SDE)量化子区域密度,并设计一种层级化的子区域支配关系排序策略,以增强个体选择压力。然后,提出一种收敛性与多样性自适应融合的加权选择机制,通过动态计算相邻代子区域加权和的差异评估子区域潜力值,进而优先保留高潜力子区域内的个体更新种群。在MAF基准测试集上,SR-MaOEA与多种主流超多目标进化算法的对比实验的结果表明,在大多数测试问题上,SR-MaOEA在反世代距离(IGD+)和超体积(HV)指标上均优于对比算法,验证了该算法在高维目标空间中的有效性和稳定性。

关键词: 超多目标优化问题, 进化算法, 子区域划分, 自适应加权, 密度估计

Abstract:

To address the challenge of balancing convergence and diversity in Multi- Objective Evolutionary Algorithm (MOEA) when solving Many-objective Optimization Problem (MaOP), a SubRegion-based Many-Objective Evolutionary Algorithm (SR-MaOEA) was proposed. In this algorithm, distribution structure of the objective space was constructed through a subregion partitioning strategy, the subregion density was quantified using Shifted Density Estimation (SDE), and a hierarchical subregion dominance relation ranking strategy was designed to enhance the selection pressure of individuals. Furthermore, a convergence-diversity adaptively fused weighted selection mechanism was proposed, in which the potential value of each subregion was evaluated by calculating the difference in weighted sums of adjacent generations of subregions dynamically, and then the individuals in subregions with higher potential values were retained preferentially to update the population. Experimental results on the MAF benchmark test set comparing multiple mainstream many-objective evolutionary algorithms show that on most test problems, SR-MaOEA outperforms the comparison algorithms in terms of both Inverted Generational Distance (IGD+) and HyperVolume (HV) metrics, thereby demonstrating the effectiveness and robustness of this algorithm in high-dimensional objective spaces.

Key words: Many-objective Optimization Problem (MaOP), Evolutionary Algorithm (EA), subregion partitioning, adaptive weighting, density estimation

中图分类号: