Scholarlink

<p>This paper proposes a dual-archive evolutionary algorithm based on multitasking optimization (DAEAMT) for multimodal multi-objective optimization problems with local Pareto optimal solution sets (MMOPLs). First, To balance the relationship between the global and local Pareto optimal solution set in MMOPLs, a convergence archive relating to the main task of DAEAMT guides the population toward global and local Pareto optimal solution set, and a diversity archive relating to the auxiliary task of DAEAMT enhances the global search capability, and a novel multitasking optimization framework is designed to achieve information transfer between different tasks. Moreover, a binary local convergence indicator is developed, which maintains population diversity by retaining the individuals with well diversity among local non-dominated individuals and strong convergence among local dominated individuals. Additionally, for the different purposes of main and auxiliary tasks in DAEAMT, the two environmental selection strategies are designed. Experimental results demonstrate that DAEAMT has significant advantages over other 9 state-of-the-art algorithms on 34 multimodal multi-objective benchmark problems and a real-world application problem.</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 dual-archive evolutionary algorithm based on multitasking for multimodal multi-objective optimization with local Pareto solutions

  • 期刊论文

  • 发表日期：

    2025年9月15日

查看全文
  • Da Feng,
  • Jianchang Liu,
  • Haibin Ouyang

摘要

This paper proposes a dual-archive evolutionary algorithm based on multitasking optimization (DAEAMT) for multimodal multi-objective optimization problems with local Pareto optimal solution sets (MMOPLs). First, To balance the relationship between the global and local Pareto optimal solution set in MMOPLs, a convergence archive relating to the main task of DAEAMT guides the population toward global and local Pareto optimal solution set, and a diversity archive relating to the auxiliary task of DAEAMT enhances the global search capability, and a novel multitasking optimization framework is designed to achieve information transfer between different tasks. Moreover, a binary local convergence indicator is developed, which maintains population diversity by retaining the individuals with well diversity among local non-dominated individuals and strong convergence among local dominated individuals. Additionally, for the different purposes of main and auxiliary tasks in DAEAMT, the two environmental selection strategies are designed. Experimental results demonstrate that DAEAMT has significant advantages over other 9 state-of-the-art algorithms on 34 multimodal multi-objective benchmark problems and a real-world application problem.