李志强^1,,
田有^1,2,3,,,
赵鹏飞^1,2,
刘财^1,2,
李洪丽^1,2
1. 吉林大学地球探测科学与技术学院, 长春 130026
2. 吉林大学地球信息探测仪器教育部重点实验室, 长春 130026
3. 长白山火山综合地球物理教育部野外科学观测研究站, 长春 130026

基金项目: 国家自然科学基金(41874049)、国家重点研发计划"深地资源勘查开采"项目(2017YFC0601301)与中央高校基本科研业务费项目联合资助


详细信息

作者简介: 李志强, 男, 1996年生, 硕士研究生, 从事接收函数研究工作.E-mail: zqli18@mails.jlu.edu.cn

通讯作者: 田有, 男, 1979年生, 教授, 从事地球内部结构成像研究工作.E-mail: tianyou@jlu.edu.cn

中图分类号: P315

收稿日期:2020-10-09
修回日期:2020-12-08
上线日期:2021-05-10

Receiver functions auto-picking method on the basis of deep learning

LI ZhiQiang^1,,
TIAN You^1,2,3,,,
ZHAO PengFei^1,2,
LIU Cai^1,2,
LI HongLi^1,2
1. College of Geo-exploration Science and Technology, Jilin University, Changchun 130026, China
2. Key Laboratory of Geophysical Exploration Equipment, Ministry of Education, Jilin University, Changchun 130026, China
3. Changbai Volcano Geophysical Observatory, Ministry of Education, Changchun 130026, China


More Information

Corresponding author: TIAN You,E-mail:tianyou@jlu.edu.cn

MSC: P315

--> Received Date: 09 October 2020
Revised Date: 08 December 2020
Available Online: 10 May 2021


摘要
摘要:接收函数作为地震学研究中的重要方法之一，在间断面成像、S波速度结构反演方面应用广泛.然而，接收函数方法需要耗费大量的人工成本挑选可用的数据，这不利于我们快速准确地获得地下结构，因此发展快速准确的数据自动处理方法具有十分重要的意义.本文针对这一问题，提出利用深度学习方法自动挑选接收函数，并使用中国地震局的牡丹江地震台(MDJ)和北京地震台(BJT)于2000年至2019年记录的波形数据提取的接收函数进行试验.结果表明，应用深度学习方法挑选接收函数是可行的，使用自动挑选的接收函数对台站下方地壳结构进行估计，结果与使用人工挑选的接收函数估计的结果具有较高的一致性.通过使用不同样本比例的训练集和测试集进行分析，本文提出的方法具有对训练集的数据量要求较低、利用率高、适合多台数据联合训练等特点.该方法在建立台网接收函数自动挑选模型字典、区域接收函数自动挑选模型等方面具有巨大应用潜力.
关键词: 深度学习/
接收函数/
H-k叠加/
各向异性

Abstract:As a commonly used seismic tool, receiver functions analysis plays significant roles in detecting discontinuous interface of the earth and S wave velocity inversion. However, picking receiver functions needs lots of manpower, which is a barrier for us to obtain the underground structure fast and precisely. In this condition, a fast and efficient method is urgently needed. In this work, we establish a deep learning network to auto-pick receiver functions. Receiver functions from 2000 to 2019 calculated from MDJ and BJT stations, belonging to China Earthquake Administration, are used to test our method, and the results show that the deep learning method is effective in receiver functions auto-picking. Those auto-picked data are used to estimate the crustal structure beneath the two stations. It shows a high degree of consistency compared with manual picked data. Several groups' experiments are carried out to analyze the influence of testing data size, helping us make a conclusion that our method has following advantages as less dependence for size of training set, fully mining useful seismic data and suitable for joint analysis of multi-stations. Once accuracy receiver functions auto-picking models of the seismic stations are built, it will have an enormous potential for auto-picking receiver functions in the future.
Key words:Deep learning/
Receiver functions/
H-k stacking/
Anisotropy



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=b53b975f-f9d9-4303-a22e-8375fab5b3ce