于子叶^1,2,,
储日升^1,,,
盛敏汉^1,2
1. 中国科学院测量与地球物理研究所大地测量与地球动力学国家重点实验室, 武汉 430077
2. 中国科学院大学, 北京 100049

基金项目: 973计划（2013CB733203），国家自然科学基金面上项目（41474049，41661164035）联合资助


详细信息

作者简介: 于子叶, 男, 博士研究生, 主要从事地震信号检测、滑坡变形监测研究. E-mail:yuziye@whigg.ac.cn

通讯作者: 储日升, 男, 研究员, 博士生导师, 主要从事地震学和地球动力学研究. E-mail:chur@asch.whigg.ac.cn

中图分类号: P315

收稿日期:2017-11-27
修回日期:2018-10-10
上线日期:2018-12-05

Pick onset time of P and S phase by deep neural network

YU ZiYe^1,2,,
CHU RiSheng^1,,,
SHENG MinHan^1,2
1. State Key Laboratory of Geodesy and Earths Dynamics, Institute of Geodesy and Geophysics, Chinese Academy of Sciences, Wuhan 430077, China
2. University of Chinese Academy of Sciences, Beijing 100049, China


More Information

Corresponding author: CHU RiSheng,E-mail:chur@asch.whigg.ac.cn

MSC: P315

--> Received Date: 27 November 2017
Revised Date: 10 October 2018
Available Online: 05 December 2018


摘要
摘要:从地震波形数据中快速准确地提取各个震相的到时是地震学中的基础问题.本文针对上述问题提出了利用深度神经网络拾取到时的新方法，建立了用于地震到时提取的17层Inception深度网络模型，在对原始三分量数据进行高通滤波和归一化处理后输入网络直接输出到时信息.整个过程基于神经网络自适应提取波形特征，自动输出结果.通过对100组加了不同强度的噪声数据进行了可靠性检验，相比于其他方法神经网络方法对于噪声具有较高的容忍度以及稳定性，并且与地震目录数据有较高的相似性.相比于AR-AIC+STA/LTA，深度神经网络虽然运算速度稍慢，但整个过程不需设定时窗与阈值，同时具有更高的可用性，并且可以迭代升级以提高精度.此方法作为人工智能方法，为波形到时拾取提供了新思路.
关键词: 震相拾取/
深度神经网络

Abstract:One of the fundamental problems in seismology is to pick up arrival times of different phases quickly and accurately. In this study, we introduce a new method to automatically measure P-and S arrival times based on deep neural network (DNN). We build an eighteen-layer neural network with inception substructure to export arrival time. Three component seismic data, after being normalized and filtered, is fed to the DNN directly. While being trained, DNN can extract the waveform feature adaptively and export the result directly. To better understand the reliability of DNN, we undertook a test of 100 samples with noise. Compared to other method, DNN shows good stability and reliability. It has high similarity, when compared to the earthquake catalogue. In contrast to AR-AIC+STA/LTA, DNN has high computational cost. But it does not need to set threshold manually and has a better result. DNN can be updated iteratively. As an artificial intelligence method in recent years, DNN provides a new and reliable solution to pick up different waveforms.
Key words:Picking onset/
Deep neural work



PDF全文下载地址:

http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_14791