| 基于机器学习的地震动特征提取与模拟以2021年云南漾濞6.4级地震为例 |
| |
| 引用本文： | 靳超越,胡进军,胡磊,王中伟.基于机器学习的地震动特征提取与模拟以2021年云南漾濞6.4级地震为例[J].世界地震工程,2021,0(4):073-80. |
| |
| 作者姓名： | 靳超越 胡进军 胡磊 王中伟 |
| |
| 作者单位： | 中国地震局工程力学研究所, 中国地震局地震工程与工程振动重点实验室, 黑龙江 哈尔滨 150080 |
| |
| 摘 要： | 随着地震动数据数量的增长和质量的提高，将基于数据驱动的机器学习方法应用到地震动模拟中有重要意义。以2021年5月21日云南漾濞MS6.4地震为例，利用主成分析方法从前震及余震地震动记录中提取特征母波时程，将地震动三要素作为模拟误差约束，在求解母波的线性组合系数时使用多目标优化算法寻优，最终找到帕累托最优解作为模拟目标台站记录时的组合系数，得到模拟地震动时程。结果表明:主成分析法在对实际地震动记录进行特征提取后，得到的特征母波时程可以在一定程度上保留原始数据的主要信息；考虑幅值、频谱和持时这三要素的角度去控制模拟误差，可以使得模拟的地震动时程更加接近真实记录。提出的基于特征提取的地震动模拟方法可以为基于小震数据合成大震地震动提供参考。
|
| 关 键 词： | MS6.4地震漾濞MS6.4地震 地震动模拟 机器学习 特征提取 序列型地震 |
Machine learning-based seismic feature extraction and simulation a case study of the 2021 Yangbi M6.4 earthquake |
| |
| JIN Chaoyue,HU Jinjun,HU Lei,WANG Zhongwei.Machine learning-based seismic feature extraction and simulation a case study of the 2021 Yangbi M6.4 earthquake[J].World Information On Earthquake Engineering,2021,0(4):073-80. |
| |
| Authors: | JIN Chaoyue HU Jinjun HU Lei WANG Zhongwei |
| |
| Affiliation: | Key Laboratory of Earthquake Engineering and Engineering Vibration, Institute of Engineering Mechanics, China Earthquake Administration, Harbin 150080, China |
| |
| Abstract: | With the increase of the quantity and quality of ground motion data, it is of great significance to apply data-driven machine learning method to ground motion simulation for the development of simulation methods. Taking the Yangbi MS6.4 earthquake on May 21, 2021 as an example, the principal component analysis method is used to extract the characteristic mother wave time histories from the foreshock-aftershock records. The three elements of ground motion are taken as the simulation error constraints, and the multi-objective optimization algorithm is used to solve the linear combination coefficients of the mother waves. Finally, the pareto optimal solution is found as the combined coefficients of the simulated target stations, and the simulated ground motion time histories are obtained. From the simulation results, it can be seen that: After feature extraction of actual ground motion records by principal component analysis, the characteristic mother waves obtained can retain the main information of original data; To control the simulation errors from the perspective of amplitude, spectrum and duration can make the simulated time histories of ground motion more close to the real records; The whole process of ground motion simulation can provide reference for the synthesis of large earthquake ground motion records based on small earthquake data. |
| |
| Keywords: | MS6 4 earthquakeYangbi MS6 4 earthquake simulation ground motion machine learning feature extraction sequence-type ground motions |
|
| 点击此处可从《世界地震工程》浏览原始摘要信息 |
|
点击此处可从《世界地震工程》下载全文 |
