Scholarlink

<p>Metaheuristic algorithms are widely applied to large-scale optimization but often suffer from premature convergence and diversity loss, particularly in global optimization settings. This study introduces a modified momentum–leader–differential grey wolf optimizer (MLDGWO). Rather than introducing isolated novel operators, MLDGWO establishes a mutually compensatory framework that dynamically couples dual-mode adaptive momentum, dynamic leader adjustment, and Lévy-DE hybrid perturbations to address these challenges.On the CEC2020 benchmark, MLDGWO demonstrates superior stability, achieving Friedman ranks of 1.1 (100D) and 1.4 (50D), with best results on 26 out of 30 functions and significant accuracy improvements over MSGWO and SCSO. On the CEC2017 suite (100D), it ranks first on 22 of 28 functions, particularly excelling in hybrid and composition categories. While demonstrating highly robust generalization across the vast majority of diverse landscapes, it acknowledges inherent limitations on highly structured, flat plateaus (e.g., CEC2017 F26-F27). In terms of efficiency, MLDGWO requires only 12.00&#xa0;s at 100D, compared with 23.44&#xa0;s for MSGWO and 78.23&#xa0;s for SCSO, thereby preserving linear-time complexity without the structural bloat typical of hybrid metaheuristics. Ablation studies further validate the strict necessity of this architectural synergy. Applied to five engineering optimization problems—including pressure vessel, cantilever beam, and gear train design—MLDGWO consistently achieves the lowest objective values, with up to 8.5% improvement in the pressure vessel problem over the best competitor, while analytically proving its ability to navigate heavily penalized boundaries and satisfy all constraints. Overall, MLDGWO provides a scalable and practical framework; its dynamically coupled strategies underpin both convergence gains and robustness, offering an efficient tool for large-scale and complex optimization tasks.</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

MLDGWO: a grey wolf optimizer with momentum, leader adjustment, and differential perturbation for global optimization problems

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • Xin Su,
  • Yichen Liu

摘要

Metaheuristic algorithms are widely applied to large-scale optimization but often suffer from premature convergence and diversity loss, particularly in global optimization settings. This study introduces a modified momentum–leader–differential grey wolf optimizer (MLDGWO). Rather than introducing isolated novel operators, MLDGWO establishes a mutually compensatory framework that dynamically couples dual-mode adaptive momentum, dynamic leader adjustment, and Lévy-DE hybrid perturbations to address these challenges.On the CEC2020 benchmark, MLDGWO demonstrates superior stability, achieving Friedman ranks of 1.1 (100D) and 1.4 (50D), with best results on 26 out of 30 functions and significant accuracy improvements over MSGWO and SCSO. On the CEC2017 suite (100D), it ranks first on 22 of 28 functions, particularly excelling in hybrid and composition categories. While demonstrating highly robust generalization across the vast majority of diverse landscapes, it acknowledges inherent limitations on highly structured, flat plateaus (e.g., CEC2017 F26-F27). In terms of efficiency, MLDGWO requires only 12.00 s at 100D, compared with 23.44 s for MSGWO and 78.23 s for SCSO, thereby preserving linear-time complexity without the structural bloat typical of hybrid metaheuristics. Ablation studies further validate the strict necessity of this architectural synergy. Applied to five engineering optimization problems—including pressure vessel, cantilever beam, and gear train design—MLDGWO consistently achieves the lowest objective values, with up to 8.5% improvement in the pressure vessel problem over the best competitor, while analytically proving its ability to navigate heavily penalized boundaries and satisfy all constraints. Overall, MLDGWO provides a scalable and practical framework; its dynamically coupled strategies underpin both convergence gains and robustness, offering an efficient tool for large-scale and complex optimization tasks.