Scholarlink

<p>Three different ground motion prediction models (GMPMs) have been developed in this paper using machine learning (ML) methods to estimate the horizontal peak ground acceleration (<i>HPGA</i>) for Iran. Two of these models are based on artificial neural networks (ANNs) of the multilayer perceptron (MLP) type, while the third employs the support vector regression (SVR). Each model utilizes moment magnitude (<i>M</i><sub><i>w</i></sub>), fault type, epicentral distance, and soil type as features (predictors) to produce a numerical prediction for <i>HPGA</i>. The models have been trained, validated, and tested using a strong-motion dataset comprising 2472 corrected horizontal accelerograms from 1100 earthquakes recorded at 815 stations across Iran from 1974 to 2022. Given the significant imbalance in the number and magnitude of recorded accelerations for Iran, an algorithm called the Repeating function has been devised to mitigate this problem within the training dataset. Besides, we designed an innovative training loop that automatically trains a model multiple times until specified criteria for the model are confirmed. Notably, three developed ML models (DMLMs) accurately predict <i>HPGA</i>, even in cases where <i>V</i><sub><i>S30</i></sub> is not defined. Although we have trained the three DMLMs to predict <i>HPGA</i> as the maximum value of the two horizontal components of the accelerogram (<i>HPGA</i><sub>max</sub>), they demonstrate a strong generalization in predicting the arithmetic and geometric means of the two mentioned components (<i>HPGA</i><sub>am</sub> and <i>HPGA</i><sub>gm</sub>). To evaluate the performance of the models, sensitivity and residual analyses, fitting curves, root-mean-square error (<i>RMSE</i>), and Pearson correlation coefficient (<i>PCC</i>) have been conducted.</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

Development of three machine learning models for predicting the horizontal peak ground acceleration for Iran

  • 期刊论文

  • 发表日期：

    2025年11月29日

查看全文
  • Mohammad-Bagher Bahraini,
  • Noorbakhsh Mirzaei,
  • Morteza Eskandari‐Ghadi,
  • Hamidreza Javan‐emrooz

摘要

Three different ground motion prediction models (GMPMs) have been developed in this paper using machine learning (ML) methods to estimate the horizontal peak ground acceleration (HPGA) for Iran. Two of these models are based on artificial neural networks (ANNs) of the multilayer perceptron (MLP) type, while the third employs the support vector regression (SVR). Each model utilizes moment magnitude (Mw), fault type, epicentral distance, and soil type as features (predictors) to produce a numerical prediction for HPGA. The models have been trained, validated, and tested using a strong-motion dataset comprising 2472 corrected horizontal accelerograms from 1100 earthquakes recorded at 815 stations across Iran from 1974 to 2022. Given the significant imbalance in the number and magnitude of recorded accelerations for Iran, an algorithm called the Repeating function has been devised to mitigate this problem within the training dataset. Besides, we designed an innovative training loop that automatically trains a model multiple times until specified criteria for the model are confirmed. Notably, three developed ML models (DMLMs) accurately predict HPGA, even in cases where VS30 is not defined. Although we have trained the three DMLMs to predict HPGA as the maximum value of the two horizontal components of the accelerogram (HPGAmax), they demonstrate a strong generalization in predicting the arithmetic and geometric means of the two mentioned components (HPGAam and HPGAgm). To evaluate the performance of the models, sensitivity and residual analyses, fitting curves, root-mean-square error (RMSE), and Pearson correlation coefficient (PCC) have been conducted.