Scholarlink

<p>Dynamic multi-objective optimization problems (DMOPs) are common in real-world scenarios, yet efficiently solving them while preserving population diversity and achieving convergence remains a formidable challenge. Existing dynamic multi-objective evolutionary algorithms often face difficulties in adaptively responding to varying environmental change frequencies, leading to compromised trade-offs between diversity preservation and convergence efficiency during dynamic tracking. This study introduces a key-point-based hybrid prediction dynamic multi-objective evolutionary algorithm (MOEA-DAPD), aimed at ensuring rapid convergence and optimizing diversity in dynamic settings. The proposed algorithm is employed a dual key-point prediction mechanism tailored to the frequency of environmental changes: a centroid-based prediction method is utilized for infrequent changes to improve stability, while a clustering-based prediction approach is applied for frequent changes to enhance convergence. To further address population diversity and optimization, a dual archive strategy is developed. One archive retains high-quality parent solutions to facilitate local convergence, while the other preserves diverse solutions to comprehensively represent the global Pareto optimal front. Additionally, distinct mutation operators are applied separately to the main and archive populations, thereby strengthening diversity and overall performance in dynamic conditions. Experimental evaluations benchmark MOEA-DAPD against five leading dynamic multi-objective evolutionary algorithms (DMOEAs) across three standard test suites. The results highlight that MOEA-DAPD demonstrates notable competitiveness and efficiency in tackling DMOPs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A key-point-based hybrid prediction dynamic multi-objective evolutionary algorithm

  • 期刊论文

  • 发表日期：

    2025年6月20日

查看全文
  • Yanxiang Yang,
  • Yongkuan Yang

摘要

Dynamic multi-objective optimization problems (DMOPs) are common in real-world scenarios, yet efficiently solving them while preserving population diversity and achieving convergence remains a formidable challenge. Existing dynamic multi-objective evolutionary algorithms often face difficulties in adaptively responding to varying environmental change frequencies, leading to compromised trade-offs between diversity preservation and convergence efficiency during dynamic tracking. This study introduces a key-point-based hybrid prediction dynamic multi-objective evolutionary algorithm (MOEA-DAPD), aimed at ensuring rapid convergence and optimizing diversity in dynamic settings. The proposed algorithm is employed a dual key-point prediction mechanism tailored to the frequency of environmental changes: a centroid-based prediction method is utilized for infrequent changes to improve stability, while a clustering-based prediction approach is applied for frequent changes to enhance convergence. To further address population diversity and optimization, a dual archive strategy is developed. One archive retains high-quality parent solutions to facilitate local convergence, while the other preserves diverse solutions to comprehensively represent the global Pareto optimal front. Additionally, distinct mutation operators are applied separately to the main and archive populations, thereby strengthening diversity and overall performance in dynamic conditions. Experimental evaluations benchmark MOEA-DAPD against five leading dynamic multi-objective evolutionary algorithms (DMOEAs) across three standard test suites. The results highlight that MOEA-DAPD demonstrates notable competitiveness and efficiency in tackling DMOPs.