Optimization algorithm of earthquake early warning based on multi-station high frequency GPS
-
摘要: 随着精密定位技术的发展，高频GPS已能够精确记录地表位移数据，研究高频GPS能为地震预警工作做出一定补充. 针对目前地震预警中单站预警误报率高的问题引入深度学习技术，利用长短期记忆网络(LSTM)联合周边区域台站对单台站进行预警以达到减少误报的目的. 首先通过对新西兰南部地区1 Hz高频GPS数据进行解算得到多个台站无震时间序列，再利用该数据训练网络得到融合区域特征的高精度模型. 该模型可以对无震时间序列进行预测并动态制定阈值区间，当实际观测值超出置信区间则判定异常. 通过与传统短时窗平均/长时窗平均算法(STA/LTA)及未融合区域特征的单站模型进行对比，结果表明：融合区域特征的单站模型可有效减少误报，在多个台站的无震长序列上较传统方法表现优异，具有一定的应用价值.Abstract: With the development of precision positioning technology, high-frequency GPS has been able to accurately record surface displacement data. Research on high frequency GPS can make a certain supplement to earthquake early warning. In view of the high false alarm rate of single station in earthquake early warning, we introduce deep learning technology and use the long short-term memory (LSTM) neural network to combine with surrounding stations to give early warning to single station. First, the seismic-free time series of multiple stations are obtained by solving the 1 Hz high-frequency GPS data in the southern region of New Zealand Then the data is used to train the network to obtain a high-precision model that integrates regional features. The model can predict the seismic-free time series and dynamically formulate a threshold interval. When the actual observation value exceeds the confidence interval, an abnormality is determined. By comparing with the traditional short-time window averaging/long-time window averaging algorithm (STA/LTA) and the single station model without regional features, the results show that the single station model fusing regional features can effectively reduce false alarms. It performs better than traditional methods on seismic-free long sequences of multiple stations and has certain application values.
-
表 1 高频GPS台站信息表
台站 经度/(°) 纬度/(°) 震中距/km MRBL 172.7752 –42.6618 24.665 WITH 173.9843 –41.5606 149.984 WEST 171.8062 –41.7447 150.339 
下载: 导出CSV
表 3 不同方法检测所用时间及误报数
方法 台站误报数 所用时间/ms MRBL WITH WEST HANM KAIK LKTA 本文模型
(引入区域特征)0 0 1 0 0 0 31.71 单台站模型
(未引入区域特征)1 2 2 2 1 1 23.59 STA/LTA 9 4 5 5 4 4 0.73
下载: 导出CSV
表 4 网络架构及参数表
方法 GPS台站 网络架构 优化算法 依赖时间/s batch size RMSE 本文模型
(引入区域特征)MRBL
WITH
WESTLSTM : 128
Dropout : 0.2
Dense : 1
Linear ActivationAdam 30 64 0.002 989
0.003 018
0.003 056单台站模型
(未引入区域特征)MRBL
WITH
WESTLSTM : 64
Dropout : 0.5
LSTM : 32
Dropout : 0.5
Dense : 1
Linear ActivationAdam 30 128 0.003 258
0.003 345
0.003 261
下载: 导出CSV
-
[1] 张红才, 金星, 李军, 等. 地震预警震级计算方法研究综述[J]. 地球物理学进展, 2012, 27(2): 464-474. DOI: 10.6038/j.issn.1004-2903.2012.02.009 [2] 单新建, 尹昊, 刘晓东, 等. 高频GNSS实时地震学与地震预警研究现状[J]. 地球物理学报, 2019, 62(8): 3043-3052. DOI: 10.6038/cjg2019M0076 [3] SATRIANO G, WU Y M, ZOLLO A, et al. Earthquake early warning: concepts, methods and physical grounds[J]. Soil dynamics and earthquake engineering, 2011, 31(2): 106-118. DOI: 10.1016/j.soildyn.2010.07.007 [4] NAKAMURA Y. On the urgent earthquake detection and alarm system (UrEDAS)[C]//The 9th World Conference On Earthquake Engineering, 1988. [5] HORIUCHI S, NEGISHI H, ABE K, et al. An automatic processing system for broadcasting earthquake alarms[J]. Bulletin of the seismological society of America, 2005, 95(2): 708-718. DOI: 10.1785/0120030133 [6] WU Y M, KANAMORI H. Experiment on an onsite early warning method for the Taiwan early warning system[J]. Bulletin of the seismological society of America, 2005, 95(1): 347-353. DOI: 10.1785/0120040097 [7] 殷海涛, 刘希强, 甘卫军. 实时高频GPS在地震学中的应用研究[J]. 地震研究, 2013, 36(3): 330-336. [8] 尹昊, 单新建, 张迎峰, 等. 高频GPS和强震仪数据在汶川地震参数快速确定中的初步应用[J]. 地球物理学报, 2018, 61(5): 1806-1816. DOI: 10.6038/cjg2018L0527 [9] TONG C, KENNETT B. Automatic seismic event recognition and later phase identification for broadband seismograms[J]. Bulletin of the seismological society of America, 1996, 86(6): 1896-1909. DOI: 10.1785/BSSA0860061896 [10] ZHANG H J, THURBER C, ROWE C A. Automatic P-wave arrival detection and picking with multiscale wavelet analysis for single-component recordings[J]. Bulletin of the seismological society of America, 2003, 93(5): 1904-1912. DOI: 10.1785/0120020241 [11] 赵岳, KTAKANO K, 赵英萍,等. 一种识别宽频带震相的人工智能网络方法[J]. 世界地震译丛, 2001(4): 55-67. [12] 于子叶, 储日升, 盛敏汉. 深度神经网络拾取地震P和S波到时[J]. 地球物理学报, 2018, 61(12): 4873-4886. DOI: 10.6038/cjg2018L0725 [13] 赵明, 陈石, DAVE YUEN. 基于深度学习卷积神经网络的地震波形自动分类与识别[J]. 地球物理学报, 2019, 62(1): 374-382. DOI: 10.6038/cjg2019M0151 [14] SAK H, SENIOR A, BEAUFAYS F. Long short-term memory recurrent neural network architectures for large scale acoustic modeling[J]. Computer science, 2014. DOI: 10.48550/arXiv.1402.1128 [15] 赵桂儒, 杨天青, 徐平, 等. 小区域GPS形变监测网GAMIT数据处理结果与IGS站选取的关系探讨[J]. 地震地磁观测与研究, 2006, 27(5): 103-106. DOI: 10.3969/j.issn.1003-3246.2006.05.019 [16] GRAVES A. Supervised sequence labelling with recurrent neural networks[J]. Studies in computational intelligence, 2012(385). DOI: 10.1007/978-3-642-24797-2 -
点击查看大图
图(13) / 表(4)
计量
- 文章访问数: 188
- HTML全文浏览量: 67
- PDF下载量: 16
- 被引次数: 0
