Scholarlink

<p>The rapid progression in engineering design has intensified the need for robust optimization tools capable of managing complex simulations effectively. In response, the adaptive regional enhanced constrained lower confidence bounding (ARECLCB) surrogate model has been introduced to address the computational challenges inherent in constrained optimization problems (COPs). This approach dynamically selects between two surrogate models: one based on decision tree ensembles and the other using neural networks, to effectively explore both local and global regions. The adaptive selection allows the chosen model to be trained either on elite samples for infilling in local regions or on all samples for broader exploration of global regions. This strategy balances local refinement with global exploration, significantly reducing evaluation time while maintaining high accuracy. Extensive testing on multiple numerical examples demonstrates that ARECLCB is highly competitive with recent algorithms in terms of efficiency and effectiveness. To further validate its capabilities, the ARECLCB model was applied to the lightweight design of two 3D irregular steel frames. The results show that the proposed approach not only reduces the average weight by 1.6% compared to other methods, but also reduces the computational time by 66.5%. This highlights the proposed approach potential in solving real-world engineering problems.</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

An adaptive surrogate-based approach for solving constrained optimization problems

  • 期刊论文

  • 发表日期：

    2025年8月28日

查看全文
  • Majid Ilchi Ghazaan,
  • Kasra Laal Dehghani

摘要

The rapid progression in engineering design has intensified the need for robust optimization tools capable of managing complex simulations effectively. In response, the adaptive regional enhanced constrained lower confidence bounding (ARECLCB) surrogate model has been introduced to address the computational challenges inherent in constrained optimization problems (COPs). This approach dynamically selects between two surrogate models: one based on decision tree ensembles and the other using neural networks, to effectively explore both local and global regions. The adaptive selection allows the chosen model to be trained either on elite samples for infilling in local regions or on all samples for broader exploration of global regions. This strategy balances local refinement with global exploration, significantly reducing evaluation time while maintaining high accuracy. Extensive testing on multiple numerical examples demonstrates that ARECLCB is highly competitive with recent algorithms in terms of efficiency and effectiveness. To further validate its capabilities, the ARECLCB model was applied to the lightweight design of two 3D irregular steel frames. The results show that the proposed approach not only reduces the average weight by 1.6% compared to other methods, but also reduces the computational time by 66.5%. This highlights the proposed approach potential in solving real-world engineering problems.