Scholarlink

<p>The performance of multi-objective differential evolution (MODE) critically depends on the appropriate setting of three key factors: the scaling factor, crossover probability, and mutation strategy. Manually tuning these factors for different problems and evolutionary stages is impractical. This paper proposes a novel superior-inferior subpopulation knowledge-driven adaptive MODE (SIAMODE) algorithm. Its core innovation lies in a differentiated learning framework centered on potential individuals. First, a dynamic population partitioning mechanism that classifies individuals as superior or inferior based on their real-time Pareto rank. Then, by identifying potential individuals, SIAMODE extracts successful evolutionary experiences (parameters and strategies) specific to each subpopulation. These experiences are stored in historical memory archives and used to adaptively guide the future evolution of superior and inferior individuals separately. This mechanism creates a targeted learning cycle that drives all individuals toward becoming potential. The algorithm is validated on 20 benchmark functions and a sodium gluconate fermentation process, demonstrating superior convergence, diversity, and robustness compared to state-of-the-art alternatives.</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

Adaptive multi-objective differential evolution algorithm driven by knowledge of superior-inferior subpopulations with application to sodium gluconate process

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • Li Tian,
  • Changyu Cai,
  • Zhichao Li,
  • Xuefeng Yan

摘要

The performance of multi-objective differential evolution (MODE) critically depends on the appropriate setting of three key factors: the scaling factor, crossover probability, and mutation strategy. Manually tuning these factors for different problems and evolutionary stages is impractical. This paper proposes a novel superior-inferior subpopulation knowledge-driven adaptive MODE (SIAMODE) algorithm. Its core innovation lies in a differentiated learning framework centered on potential individuals. First, a dynamic population partitioning mechanism that classifies individuals as superior or inferior based on their real-time Pareto rank. Then, by identifying potential individuals, SIAMODE extracts successful evolutionary experiences (parameters and strategies) specific to each subpopulation. These experiences are stored in historical memory archives and used to adaptively guide the future evolution of superior and inferior individuals separately. This mechanism creates a targeted learning cycle that drives all individuals toward becoming potential. The algorithm is validated on 20 benchmark functions and a sodium gluconate fermentation process, demonstrating superior convergence, diversity, and robustness compared to state-of-the-art alternatives.