Scholarlink

<p>To mitigate the high computational cost associated with structural optimization using finite element software like SAP2000, this study introduces Jaya-Pattern Search-Cascade Forward Neural Networks (JPC), a novel four-step surrogate-assisted optimization framework. JPC synergistically integrates Cascade Forward Neural Networks (CFNN) for accurate surrogate modeling, Comprehensive Learning Jaya (CLJAYA) for efficient global exploration and initial data generation, and Pattern Search (PS) for rapid local refinement. A key feature is its dynamic surrogate updating strategy, designed to minimize the number of computationally expensive high-fidelity function evaluations (NFEs). Comparative analysis against established algorithms demonstrates JPC’s effectiveness across multiple benchmark truss problems. For the 10-bar truss, JPC achieves comparable optimal weight while significantly reducing NFEs by 61% versus adaptive elitist differential evolution (aeDE) and 37% versus SurrogateOpt. For the 25-bar truss, JPC yields a superior optimal design (483.88&#xa0;lb) with substantial NFE reductions of 35% compared to SurrogateOpt and up to 90% against established metaheuristics such as Enhanced Colliding Bodies Optimization (ECBO). Furthermore, JPC consistently achieves superior results (lower mean weights and reduced standard deviations) compared to the Hybrid Intelligent Genetic Algorithm (HIGA) across the 10-bar and 25-bar benchmarks. In the larger 160-bar truss optimization, JPC also identifies a considerably lighter design (1370.9&#xa0;kg) compared to SurrogateOpt (1536.8&#xa0;kg). The algorithm’s practical applicability and scalability are further validated on a complex steel–concrete composite I-girder bridge optimization problem within SAP2000, where JPC attains an 87% reduction in required high-fidelity analyses while satisfying all structural integrity and design code constraints. These findings underscore JPC’s potential as a robust and highly efficient framework for tackling computationally intensive structural optimization tasks.</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

Enhancing Structural Optimization: A Novel Four-Step Surrogate-Assisted Approach

  • 期刊论文

  • 发表日期：

    2025年9月20日

查看全文
  • Behrouz Ahmadi-Nedushan,
  • Reza Javanmardi

摘要

To mitigate the high computational cost associated with structural optimization using finite element software like SAP2000, this study introduces Jaya-Pattern Search-Cascade Forward Neural Networks (JPC), a novel four-step surrogate-assisted optimization framework. JPC synergistically integrates Cascade Forward Neural Networks (CFNN) for accurate surrogate modeling, Comprehensive Learning Jaya (CLJAYA) for efficient global exploration and initial data generation, and Pattern Search (PS) for rapid local refinement. A key feature is its dynamic surrogate updating strategy, designed to minimize the number of computationally expensive high-fidelity function evaluations (NFEs). Comparative analysis against established algorithms demonstrates JPC’s effectiveness across multiple benchmark truss problems. For the 10-bar truss, JPC achieves comparable optimal weight while significantly reducing NFEs by 61% versus adaptive elitist differential evolution (aeDE) and 37% versus SurrogateOpt. For the 25-bar truss, JPC yields a superior optimal design (483.88 lb) with substantial NFE reductions of 35% compared to SurrogateOpt and up to 90% against established metaheuristics such as Enhanced Colliding Bodies Optimization (ECBO). Furthermore, JPC consistently achieves superior results (lower mean weights and reduced standard deviations) compared to the Hybrid Intelligent Genetic Algorithm (HIGA) across the 10-bar and 25-bar benchmarks. In the larger 160-bar truss optimization, JPC also identifies a considerably lighter design (1370.9 kg) compared to SurrogateOpt (1536.8 kg). The algorithm’s practical applicability and scalability are further validated on a complex steel–concrete composite I-girder bridge optimization problem within SAP2000, where JPC attains an 87% reduction in required high-fidelity analyses while satisfying all structural integrity and design code constraints. These findings underscore JPC’s potential as a robust and highly efficient framework for tackling computationally intensive structural optimization tasks.