Scholarlink

<p>This paper addresses the significant challenge of excessive computation time in optimizing nonlinear inelastic steel frames, primarily caused by time-consuming constraint evaluations. To overcome this issue, we develop an efficient optimization framework utilizing a novel self-adaptive pbest differential evolution (AEpDE) algorithm in conjunction with machine learning (ML) techniques. AEpDE leverages the pbest approach, with a newly formulated method to automatically determine the number of best individuals used per generation, reflecting the diversity of the current population. To minimize the need for structural analyses, ML algorithms are strategically integrated with AEpDE, ensuring the algorithm’s robustness despite potential surrogate model errors. The framework is tested on several practical nonlinear inelastic steel frames, including two-story space, 3 × 10, and 5 × 14 steel frames. Numerical results reveal that AEpDE produces superior optimal structural designs compared to existing methods such as effective pbest differential evolution (EpDE), Jaya, and particle swarm optimization (PSO). Furthermore, the LightGBM-AEpDE combination significantly reduces computational costs by approximately 70% to 90%, enhancing efficiency over AEpDE, EpDE, Jaya, and PSO.</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

Machine learning-based method using adaptive differential evolution for optimizing nonlinear steel frames

  • 期刊论文

  • 发表日期：

    2025年10月24日

查看全文
  • Viet-Hung Truong,
  • Thanh-Thuy T. Nguyen,
  • Huu-Tai Thai,
  • Sawekchai Tangaramvong

摘要

This paper addresses the significant challenge of excessive computation time in optimizing nonlinear inelastic steel frames, primarily caused by time-consuming constraint evaluations. To overcome this issue, we develop an efficient optimization framework utilizing a novel self-adaptive pbest differential evolution (AEpDE) algorithm in conjunction with machine learning (ML) techniques. AEpDE leverages the pbest approach, with a newly formulated method to automatically determine the number of best individuals used per generation, reflecting the diversity of the current population. To minimize the need for structural analyses, ML algorithms are strategically integrated with AEpDE, ensuring the algorithm’s robustness despite potential surrogate model errors. The framework is tested on several practical nonlinear inelastic steel frames, including two-story space, 3 × 10, and 5 × 14 steel frames. Numerical results reveal that AEpDE produces superior optimal structural designs compared to existing methods such as effective pbest differential evolution (EpDE), Jaya, and particle swarm optimization (PSO). Furthermore, the LightGBM-AEpDE combination significantly reduces computational costs by approximately 70% to 90%, enhancing efficiency over AEpDE, EpDE, Jaya, and PSO.