Scholarlink

<p>While traditional seismic performance assessment is rigorous, it relies on dynamic analyses that are too complicated for large-scale use. This study presents a reliable, interpretable machine learning framework for quickly predicting the seismic demands on reinforced concrete frames. The methodology integrates a comprehensive database of 500 unique analysis cases, tailored to the seismic hazard conditions of Tehran, Iran, with a physics-informed feature engineering strategy. A systematic comparison of the leading regression algorithms identified the optimal models, which were evaluated using a comprehensive suite of statistical metrics, such as the Coefficient of Determination (R²) and the Root Mean Squared Error (RMSE). The top-performing models demonstrated high predictive accuracy. Crucially, an in-depth sensitivity analysis using the SHAP framework moved beyond a “black box” paradigm, confirming that the models learned physically meaningful relationships. The analysis identified spectral acceleration (Sa(T1)) and structural mass as dominant predictors, consistent with fundamental principles of structural dynamics. The findings establish this integrated approach as a viable pathway toward developing efficient and trustworthy tools for rapid urban seismic vulnerability assessment and performance-based 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

Efficient and interpretable seismic demands prediction for RC frames: a physics-informed, region-specific approach

  • 期刊论文

  • 发表日期：

    2025年10月9日

查看全文
  • Ali Kaveh,
  • Alireza Amiri

摘要

While traditional seismic performance assessment is rigorous, it relies on dynamic analyses that are too complicated for large-scale use. This study presents a reliable, interpretable machine learning framework for quickly predicting the seismic demands on reinforced concrete frames. The methodology integrates a comprehensive database of 500 unique analysis cases, tailored to the seismic hazard conditions of Tehran, Iran, with a physics-informed feature engineering strategy. A systematic comparison of the leading regression algorithms identified the optimal models, which were evaluated using a comprehensive suite of statistical metrics, such as the Coefficient of Determination (R²) and the Root Mean Squared Error (RMSE). The top-performing models demonstrated high predictive accuracy. Crucially, an in-depth sensitivity analysis using the SHAP framework moved beyond a “black box” paradigm, confirming that the models learned physically meaningful relationships. The analysis identified spectral acceleration (Sa(T1)) and structural mass as dominant predictors, consistent with fundamental principles of structural dynamics. The findings establish this integrated approach as a viable pathway toward developing efficient and trustworthy tools for rapid urban seismic vulnerability assessment and performance-based design.