Scholarlink

<p>The behaviour of structural and non-structural elements of a building is an important area for earthquake engineering. Failure of these elements after the major earthquake may strongly affect the collapse of the buildings, life safety, etc. One of the most important parameters for the seismic design or evaluation of the structural and non-structural elements of a building is peak floor acceleration (<i>PFA</i>). Therefore, estimating <i>PFA</i> values of the buildings provides valuable information and rapid assessment for earthquake engineering practice. In this study, the prediction of the <i>PFA</i> of reinforced concrete (RC) frames was investigated using powerful machine learning (ML) algorithms, specifically categorical boosting (CatBoost) and eXtreme gradient boosting (XGBoost). For this aim, low- to mid-rise buildings were represented by 3-story and 8-story RC frames and for each of these RC frames, the study utilized data from nonlinear time history (NTA) analysis of 2300 ground motion records. The input parameters for ML algorithms included ten different intensity measures (IMs) and spectral acceleration (<i>S</i><sub><i>a</i></sub>) values at the fundamental periods of the structures. The particle swarm optimization (PSO) algorithm, differential evolution (DE), and genetic algorithm (GA) were used for hyperparameter optimization of the ML models. The prediction results of the PSO, DE, and GA optimized ML models were compared with those of the models using default parameters (i.e., XGBoost-RAW and CatBoost-RAW). Furthermore, the accuracy of the optimized ML models was assessed using eighteen statistical assessment criteria to study the statistical significance and accuracy of the results with more depth. The results indicated that the XGBoost-DE algorithm was yielded as the best model for prediction <i>PFA</i> of a 3-story RC building considering statistical evaluation metrics. Also, the XGBoost-GA and CatBoost-PSO models were determined to be the best models for prediction <i>PFA</i> of an 8-story RC building. The findings demonstrated the capability of proposed optimized ML models in accurately and reliably predicting <i>PFA</i> values of low-to mid-rise RC buildings, as confirmed by several performance metrics.</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 Prediction of Peak Floor Acceleration in Low- to Mid-Rise RC Buildings Using Ground Motion Intensity Measures

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Ahmet Demir,
  • Selçuk Demir,
  • Emrehan Kutlug Sahin

摘要

The behaviour of structural and non-structural elements of a building is an important area for earthquake engineering. Failure of these elements after the major earthquake may strongly affect the collapse of the buildings, life safety, etc. One of the most important parameters for the seismic design or evaluation of the structural and non-structural elements of a building is peak floor acceleration (PFA). Therefore, estimating PFA values of the buildings provides valuable information and rapid assessment for earthquake engineering practice. In this study, the prediction of the PFA of reinforced concrete (RC) frames was investigated using powerful machine learning (ML) algorithms, specifically categorical boosting (CatBoost) and eXtreme gradient boosting (XGBoost). For this aim, low- to mid-rise buildings were represented by 3-story and 8-story RC frames and for each of these RC frames, the study utilized data from nonlinear time history (NTA) analysis of 2300 ground motion records. The input parameters for ML algorithms included ten different intensity measures (IMs) and spectral acceleration (Sa) values at the fundamental periods of the structures. The particle swarm optimization (PSO) algorithm, differential evolution (DE), and genetic algorithm (GA) were used for hyperparameter optimization of the ML models. The prediction results of the PSO, DE, and GA optimized ML models were compared with those of the models using default parameters (i.e., XGBoost-RAW and CatBoost-RAW). Furthermore, the accuracy of the optimized ML models was assessed using eighteen statistical assessment criteria to study the statistical significance and accuracy of the results with more depth. The results indicated that the XGBoost-DE algorithm was yielded as the best model for prediction PFA of a 3-story RC building considering statistical evaluation metrics. Also, the XGBoost-GA and CatBoost-PSO models were determined to be the best models for prediction PFA of an 8-story RC building. The findings demonstrated the capability of proposed optimized ML models in accurately and reliably predicting PFA values of low-to mid-rise RC buildings, as confirmed by several performance metrics.