Scholarlink

<p>This paper suggests a framework combining sophisticated machine learning models with metaheuristics optimization for seismic drift ratio prediction and risk quantification in vertically irregular reinforced concrete structures. A 120-case database, with variations in stiffness, mass distributions, and geometric irregularities, was generated with nonlinear dynamic analysis in ETABS and complemented with experimental evidence. A systematic optimization with three metaheuristics, Emperor Penguin Colony (EPC), Sailfish Optimization (SFO), and Hunger Games Search (HGS), was performed with two predictive models, Convolutional Neural Networks (CNN) and Extreme Gradient Boosting (XGBoost). The optimization increased the robustness and generalization capabilities of the models, achieving gains in performance with 27% (CNN) and 29% (XGBoost) RMSE reduction, along with R² values higher than 0.95. The framework was also applied to the derivation of fragility curves with optimized outputs, which presented specific seismic vulnerability behaviors, with a particular incidence in mass-irregular structures showing moderate PGA value-dominated damage levels at relatively reduced levels. The statistical validation substantiated the importance of the improvements (<i>p</i> &lt; 0.001). The paper is impaired by a mostly adopted configuration based on RC moment-resisting frames mostly simulated for the case of Amman (Zone 2&#xa0;A) hazard context as well as a suite of ground motion that, even if code-compatible, lacks a capture of world hazard diversity. Results will consequently need recalibration with a different configuration for other construction system applications, code environment, or even seismic site effects. The paper is a demonstration of the potential framework combining machine learning with metaheuristics optimization methods with a potential for enhancing seismic performance prediction and providing a facility with a set of workable tools for a basis for seismic design for performance as well as identifying retrofit priorities with a potential for implementation in earthquake-risk territories.</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

Methodological integration of machine learning and metaheuristics for seismic risk assessment of irregular buildings

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Ahmad Habboush,
  • Bassam Elzaghmouri

摘要

This paper suggests a framework combining sophisticated machine learning models with metaheuristics optimization for seismic drift ratio prediction and risk quantification in vertically irregular reinforced concrete structures. A 120-case database, with variations in stiffness, mass distributions, and geometric irregularities, was generated with nonlinear dynamic analysis in ETABS and complemented with experimental evidence. A systematic optimization with three metaheuristics, Emperor Penguin Colony (EPC), Sailfish Optimization (SFO), and Hunger Games Search (HGS), was performed with two predictive models, Convolutional Neural Networks (CNN) and Extreme Gradient Boosting (XGBoost). The optimization increased the robustness and generalization capabilities of the models, achieving gains in performance with 27% (CNN) and 29% (XGBoost) RMSE reduction, along with R² values higher than 0.95. The framework was also applied to the derivation of fragility curves with optimized outputs, which presented specific seismic vulnerability behaviors, with a particular incidence in mass-irregular structures showing moderate PGA value-dominated damage levels at relatively reduced levels. The statistical validation substantiated the importance of the improvements (p < 0.001). The paper is impaired by a mostly adopted configuration based on RC moment-resisting frames mostly simulated for the case of Amman (Zone 2 A) hazard context as well as a suite of ground motion that, even if code-compatible, lacks a capture of world hazard diversity. Results will consequently need recalibration with a different configuration for other construction system applications, code environment, or even seismic site effects. The paper is a demonstration of the potential framework combining machine learning with metaheuristics optimization methods with a potential for enhancing seismic performance prediction and providing a facility with a set of workable tools for a basis for seismic design for performance as well as identifying retrofit priorities with a potential for implementation in earthquake-risk territories.