Scholarlink

<p>The most common strong motion parameters such as peak ground acceleration, velocity and displacement overlook the influence of the frequency content of ground motion and the duration of the shaking. Many researchers have shown that the energy input, also called high-order, parameters and duration of ground shaking are basic characteristics controlling the seismic response of structures. Many of them have been correlated with structural damage or important aspects of foundation and geotechnical engineering. In the past, a great number of Ground Motion Predictive Equations (GMPEs) have been proposed to predict the magnitude of those parameters, with a variety from simple to very complex functional forms capturing various seismic and site effects, such as magnitude scale, geometrical and anelastic attenuation, soil nonlinearity, types and geometry of faults, and the random effects associated to inter- and intra-event variability. Those GMPEs were mostly developed through calibration of regression coefficients to empirical data. On the other hand, Machine learning models, and more specifically the ANN-based models, are increasingly used in many fields of science such as earthquake engineering, geotechnical earthquake engineering and engineering seismology as evidenced by recently published extensive literature reviews. This work aims to propose new GMPEs for five high-order intensity parameters (I<sub>A</sub>, CAV, ASI, I<sub>H</sub>, I<sub>c</sub>), as well as, for three definitions of strong motion duration (SD<sub>5–95</sub>, SD<sub>5–75</sub>, SD<sub>20–80</sub>) for shallow earthquakes in Greece. Two classes of GMPEs are developed. The first is developed through training ANN models to strong motion data recorded in Greece. The second is based on calibrating the coefficients of a fixed functional form through the more classical forward nonlinear regression. The parallel implementation of more advanced Machine Learning Algorithms (MLA) and the more classical nonlinear regression for model calibration to empirical data and the subsequent comparison between the resulting models highlights their advantages and disadvantages in ground motion estimation.</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

Ground motion predictive equations for high order intensity and strong motion duration parameters for shallow earthquakes in Greece

  • 期刊论文

  • 发表日期：

    2025年12月19日

查看全文
  • Sotiriadis Dimitris,
  • Morfidis Konstantinos,
  • Margaris Basil

摘要

The most common strong motion parameters such as peak ground acceleration, velocity and displacement overlook the influence of the frequency content of ground motion and the duration of the shaking. Many researchers have shown that the energy input, also called high-order, parameters and duration of ground shaking are basic characteristics controlling the seismic response of structures. Many of them have been correlated with structural damage or important aspects of foundation and geotechnical engineering. In the past, a great number of Ground Motion Predictive Equations (GMPEs) have been proposed to predict the magnitude of those parameters, with a variety from simple to very complex functional forms capturing various seismic and site effects, such as magnitude scale, geometrical and anelastic attenuation, soil nonlinearity, types and geometry of faults, and the random effects associated to inter- and intra-event variability. Those GMPEs were mostly developed through calibration of regression coefficients to empirical data. On the other hand, Machine learning models, and more specifically the ANN-based models, are increasingly used in many fields of science such as earthquake engineering, geotechnical earthquake engineering and engineering seismology as evidenced by recently published extensive literature reviews. This work aims to propose new GMPEs for five high-order intensity parameters (IA, CAV, ASI, IH, Ic), as well as, for three definitions of strong motion duration (SD5–95, SD5–75, SD20–80) for shallow earthquakes in Greece. Two classes of GMPEs are developed. The first is developed through training ANN models to strong motion data recorded in Greece. The second is based on calibrating the coefficients of a fixed functional form through the more classical forward nonlinear regression. The parallel implementation of more advanced Machine Learning Algorithms (MLA) and the more classical nonlinear regression for model calibration to empirical data and the subsequent comparison between the resulting models highlights their advantages and disadvantages in ground motion estimation.