Scholarlink

<p>This paper proposes the Role Model Learning Algorithm (RMLA) to address complex system optimization problems. Inspired by Bandura’s theory of observational learning, RMLA introduces two search strategies: (I) direct imitation learning and (II) reflective practice learning. Meanwhile, it incorporates a dynamic motivational mechanism for balancing global and local search capabilities by adjusting their bias based on prior results. Additionally, a best-driven boundary absorption strategy is introduced to enhance solution diversity. In the experiments, RMLA’s search and convergence abilities, structural bias, as well as exploration-exploitation balance, are first analyzed. RMLA is then compared with five classical and eight recent high-performance algorithms on 144 CEC2017/2022 functions. RMLA achieves the best results in 55.56% of cases across all dimensions, particularly excelling in high-dimensional optimization with 70% of cases, and demonstrates high stability. Statistical tests further confirm RMLA’s superiority, with an average rank of 1.7 across all dimensions. In comparison with cutting-edge algorithms, RMLA achieves an average rank of 1.62, and in 100-dimensional problems, it attains an even higher rank of 1.4. After a parameter analysis, an ablation study is performed to assess the effectiveness of the dynamic motivational mechanism. Finally, RMLA is applied to four engineering design problems and one NP-hard problem. The results confirm RMLA’s effectiveness, particularly in high-dimensional optimization. The source code of RMLA is publicly available at <a href="https://github.com/hechenen/RMLA">https://github.com/hechenen/RMLA</a>.</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

Role Model Learning Algorithm: a human-inspired approach with dynamic motivational mechanism for complex system optimization problems

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Shaocong Guo,
  • Junhao Wang,
  • Zhenfeng Zhu

摘要

This paper proposes the Role Model Learning Algorithm (RMLA) to address complex system optimization problems. Inspired by Bandura’s theory of observational learning, RMLA introduces two search strategies: (I) direct imitation learning and (II) reflective practice learning. Meanwhile, it incorporates a dynamic motivational mechanism for balancing global and local search capabilities by adjusting their bias based on prior results. Additionally, a best-driven boundary absorption strategy is introduced to enhance solution diversity. In the experiments, RMLA’s search and convergence abilities, structural bias, as well as exploration-exploitation balance, are first analyzed. RMLA is then compared with five classical and eight recent high-performance algorithms on 144 CEC2017/2022 functions. RMLA achieves the best results in 55.56% of cases across all dimensions, particularly excelling in high-dimensional optimization with 70% of cases, and demonstrates high stability. Statistical tests further confirm RMLA’s superiority, with an average rank of 1.7 across all dimensions. In comparison with cutting-edge algorithms, RMLA achieves an average rank of 1.62, and in 100-dimensional problems, it attains an even higher rank of 1.4. After a parameter analysis, an ablation study is performed to assess the effectiveness of the dynamic motivational mechanism. Finally, RMLA is applied to four engineering design problems and one NP-hard problem. The results confirm RMLA’s effectiveness, particularly in high-dimensional optimization. The source code of RMLA is publicly available at https://github.com/hechenen/RMLA.