Scholarlink

<p>Expensive function evaluations and the trade-off between convergence and diversity remain critical challenges in multiobjective optimization. This paper presents ST-MTO, a surrogate-assisted evolutionary framework that integrates multitasking optimization with dual-surrogate modeling. Two surrogate models targeting convergence and diversity metrics are optimized concurrently as related yet independent tasks. A source-target task transfer mechanism and adaptive knowledge transfer intensity guide the search while effectively avoiding negative transfer, and a reference vector-based pre-screening strategy preserves population diversity and enhances selection efficiency. Experimental results demonstrate that ST-MTO efficiently identifies well-distributed Pareto-optimal solutions in multi-line distance minimization and flight controller optimization, achieving a balanced trade-off among tracking accuracy, control energy, and signal smoothness. Compared with existing surrogate-assisted and multitasking evolutionary algorithms, ST-MTO demonstrates strong convergence performance, competitive diversity, and improved computational efficiency, highlighting its robustness and practical value in engineering design.</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

Surrogate-assisted multitask evolutionary optimization with adaptive knowledge transfer

  • 期刊论文

  • 发表日期：

    2026年4月11日

查看全文
  • Le Wang,
  • Xuefeng Yan

摘要

Expensive function evaluations and the trade-off between convergence and diversity remain critical challenges in multiobjective optimization. This paper presents ST-MTO, a surrogate-assisted evolutionary framework that integrates multitasking optimization with dual-surrogate modeling. Two surrogate models targeting convergence and diversity metrics are optimized concurrently as related yet independent tasks. A source-target task transfer mechanism and adaptive knowledge transfer intensity guide the search while effectively avoiding negative transfer, and a reference vector-based pre-screening strategy preserves population diversity and enhances selection efficiency. Experimental results demonstrate that ST-MTO efficiently identifies well-distributed Pareto-optimal solutions in multi-line distance minimization and flight controller optimization, achieving a balanced trade-off among tracking accuracy, control energy, and signal smoothness. Compared with existing surrogate-assisted and multitasking evolutionary algorithms, ST-MTO demonstrates strong convergence performance, competitive diversity, and improved computational efficiency, highlighting its robustness and practical value in engineering design.