欢迎访问《地震》,

地震 ›› 2023, Vol. 43 ›› Issue (1): 137-151.doi: 10.12196/j.issn.1000-3274.2023.01.011

• • 上一篇    下一篇

白鹤滩水库库区基于深度学习的震相拾取与地震定位

章宇成, 华卫   

  1. 中国地震局地震预测重点实验室, 中国地震局地震预测研究所, 北京 100036
  • 收稿日期:2022-03-16 修回日期:2022-05-02 出版日期:2023-01-31 发布日期:2023-05-15
  • 通讯作者: 华卫, 研究员。 E-mail: huawei@ief.ac.cn
  • 作者简介:章宇成(1997-), 男, 江西抚州人, 硕士研究生, 主要从事地震学研究。
  • 基金资助:
    国家自然科学基金面上项目(41774057); 国家重点研发计划课题(2021YFC3000703)

Phase Picking and Earthquake Location based on Deep Learning in the Baihetan Reservior Area

ZHANG Yu-cheng, HUA Wei   

  1. Key Laboratory of Earthquake Prediction, Institute of Earthquake Forecasting, CEA, Beijing 100036, China
  • Received:2022-03-16 Revised:2022-05-02 Online:2023-01-31 Published:2023-05-15

摘要: 近年来深度学习技术广泛应用于震相拾取与地震定位研究, 采用深度神经网络搭建的EQTransformer模型对白鹤滩水库库区34个数字地震台站2016—2018年记录的连续数据进行P、 S波震相拾取, 并通过REAL进行震相关联和初步定位, 然后使用 VELEST和hypoDD地震定位算法优化地震位置。 研究表明, 基于深度学习的震相拾取, 与白鹤滩水库地区传统的人工处理方法相比显示出更高的效率, EQTransformer模型可拾取与人工拾取相当的P、 S波震相到时, 其时间差的均值分别为0.03 s和0.07 s, 符合正态分布。 REAL初步定位后的地震个数(13815个)接近常规目录(7862个)的2倍, 最终通过hypoDD获得了7108个高精度定位地震。 估算的震级比常规目录中的震级平均低0.27, 震级差值集中在0.7 以内, 最小完备震级由常规目录的ML1.4更改为ML0.6+0.27, 填补了部分常规目录的震级空白, 丰富了研究区域内的中小型地震。

关键词: 震相拾取, 地震定位, 深度学习, 白鹤滩水库

Abstract: In recent years, deep learning technology has become increasingly prevalent in seismic phase picking and earthquake location research. This paper employs the EQTransformer, a deep neural network, to pick P and S arrivals from continuous data recorded by 34 digital seismic stations in the Baihetan reservoir area from 2016 to 2018. Phase association and preliminary localization by REAL, followed by optimization of earthquake location using VELEST and hypoDD. The research reveals that seismic phase picking based on deep learning is significantly more efficient than traditional manual methods in the Baihetan Reservoir area. The accuracy of the EQTransformer-picked P and S first-arrivals is comparable to that of manually picked phases, with average time differences of 0.03 s and 0.07 s, conforming to a normal distribution. The number of events after preliminary location by REAL (13815) is nearly twice of the routine catalog (7862), and we ulfimately obtain high-precision locations of 7108 earthquakes by the hypoDD. The estimated magnitude is on average 0.27 lower than that in the routine catalog, and the magnitude difference is primavily within 0.7. The minimum magnitude of completeness is changed from ML1.4 in the routine catalog to ML0.6+0.27, effectively filling the magnitude gap of the routine catalog and enriching the data on moderate and small earthquakes in the Baihetan reservoir area.

Key words: Phase picking, Earthquake location, Deep learning, The Baihetan reservior

中图分类号: