Scholarlink

<p>This paper proposes DHCRWOA, a stage-specific extension of the Whale Optimization Algorithm (WOA) designed to improve its exploration–exploitation transition and reduce stagnation around suboptimal leaders. Five modules are introduced, each targeting a single WOA search phase: Good Nodes Set (GNS) initialization, adaptive parameter control, statistical-guided exploration, Cauchy-based local exploitation, and Rayleigh-weighted spiral updating. On the CEC2005 benchmark suite at 30 dimensions, DHCRWOA achieves the best average rank (2.13) among ten algorithms and the best or tied-best mean on 22 of 23 functions, with no baseline outperforming it on more than one function. At 50 and 100 dimensions, DHCRWOA maintains the top Friedman rank (1.69 and 1.62, respectively) at average runtimes of 0.52&#xa0;s and 1.03&#xa0;s per run. A five-variant ablation study confirms that weakening the statistical-guided exploration or Cauchy-based exploitation produces the largest rank degradation (from 2.35 to 4.09 and 4.70, respectively), while all five modules contribute to overall performance. Five constrained engineering design problems further confirm competitive results on practical optimization tasks. Theoretical analysis establishes the relationship between DHCRWOA and standard WOA, including a perturbation bound on the adaptive parameter (<InlineEquation ID="IEq1"><EquationSource Format="TEX">\(\le 0.02\)</EquationSource></InlineEquation>), boundedness of the population trajectory, monotonicity of the best-so-far process, and a global reachability theorem.</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

DHCRWOA: adaptive whale optimization algorithm with Cauchy-Rayleigh distribution for numerical and engineering design optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月26日

查看全文
  • Yanxiao Li,
  • Junhao Wei,
  • Yifu Zhao,
  • Zikun Li,
  • Zhiwen Wang,
  • Sio-Kei Im,
  • Xu Yang,
  • Yapeng Wang

摘要

This paper proposes DHCRWOA, a stage-specific extension of the Whale Optimization Algorithm (WOA) designed to improve its exploration–exploitation transition and reduce stagnation around suboptimal leaders. Five modules are introduced, each targeting a single WOA search phase: Good Nodes Set (GNS) initialization, adaptive parameter control, statistical-guided exploration, Cauchy-based local exploitation, and Rayleigh-weighted spiral updating. On the CEC2005 benchmark suite at 30 dimensions, DHCRWOA achieves the best average rank (2.13) among ten algorithms and the best or tied-best mean on 22 of 23 functions, with no baseline outperforming it on more than one function. At 50 and 100 dimensions, DHCRWOA maintains the top Friedman rank (1.69 and 1.62, respectively) at average runtimes of 0.52 s and 1.03 s per run. A five-variant ablation study confirms that weakening the statistical-guided exploration or Cauchy-based exploitation produces the largest rank degradation (from 2.35 to 4.09 and 4.70, respectively), while all five modules contribute to overall performance. Five constrained engineering design problems further confirm competitive results on practical optimization tasks. Theoretical analysis establishes the relationship between DHCRWOA and standard WOA, including a perturbation bound on the adaptive parameter (\(\le 0.02\)), boundedness of the population trajectory, monotonicity of the best-so-far process, and a global reachability theorem.