2022年1月8日青海门源6.9级地震机器学习地震预警震级估计与现地阈值报警的回溯验证  

Backtracking verification of machine learning earthquake early warning magnitude estimation and on-site threshold alarm for Menyuan M6.9 earthquake in Qinghai on January 8,2022

作　　者：宋晋东[1,2] 朱景宝 韦永祥 刘艳琼[4] 何斌 李继龙 李山有[1,2] SONG JinDong;ZHU JingBao;WEI YongXiang;LIU YanQiong;HE Bin;LI JiLong;LI ShanYou(Key Laboratory of Earthquake Engineering and Engineering Vibration,Institute of Engineering Mechanics,China Earthquake Administration,Harbin 150080,China;KeyLaboratory of Earthquake Disaster Mitigation,Ministry of Emergency Management,Harbin 150080,China;Fujian Earthquake Agency,Fuzhou 350003,China;China Earthquake Networks Center,Beijing 100045,China)

机构地区：[1]中国地震局工程力学研究所地震工程与工程振动重点实验室,哈尔滨150080 [2]地震灾害防治应急管理部重点实验室,哈尔滨150080 [3]福建省地震局,福州350003 [4]中国地震台网中心,北京100045

出　　处：《地球物理学报》2023年第7期2903-2919,共17页Chinese Journal of Geophysics

基　　金：中国地震局工程力学研究所基本科研业务费专项(2016A03);国家自然科学基金项目(U2039209和51408564);黑龙江省自然科学基金项目(LH2021E119);地震科技星火计划(XH22008B);福建省地震局科技基金专项(SF202103);国家重点研发计划项目(2018YFC1504003)资助。

摘　　要：2022年1月8日1时45分青海省海北州门源县发生6.9级地震,周边地区普遍有感,并导致多条高铁线路临时停运.本文利用这次地震获取的大量烈度计加速度记录,基于正在进行系统研发的机器学习地震预警方法模块,对地震预警震级估计与现地阈值报警进行了回溯验证.结果表明:在地震发生后3.1 s,震级估计为6.5级,且震级估计误差不受信噪比和震中距变化的影响,随着首台触发后时间的增加,震级估计逐渐接近实际震级.对于现地地震动速度峰值PGV(Peak Ground Velocity)预测,各个台站在P波到达后3 s时,预测PGV与观测PGV呈现1:1线性关系,随着P波到达后时间窗的增加,预测PGV逐步接近观测PGV,且PGV预测误差不受信噪比和震中距变化的影响.现地台站仪器烈度阈值设置为Ⅵ度时,报警成功、误报、漏报的百分比分别为99.53%、0%、0.47%,平均预警时间为19.62 s,且地震烈度Ⅵ度区内没有发生误报和漏报;现地台站仪器烈度阈值设置为Ⅶ度时,报警成功、误报、漏报的百分比分别为99.77%、0%、0.23%,平均预警时间为9.69 s,且地震烈度Ⅶ度区内没有发生误报和漏报.此次回溯验证结果表明:机器学习方法在这次地震中可以得到鲁棒的震级估计和现地阈值报警结果,并为该方法的在线测试以及中国地震预警系统升级提供可行性依据;其次,在这次地震事件中,烈度计可为预警提供额外的作用,这也为烈度计在未来地震预警的研究和应用中提供了更多的可能性.At 1:45 on January 8,2022,an earthquake event with M6.9 occurred in Menyuan County,Haibei Prefecture,Qinghai Province,which was felt in the surrounding areas,resulting in the temporary shutdown of several high-speed rail lines.Based on the machine learning earthquake early warning(EEW)method module of the systematic research and development,this paper makes a backtracking verification of EEW magnitude estimation and on-site threshold alarm by using a large number of acceleration records of low-cost micro-electro-mechanical system-based(MEMS-based)stations obtained from this event.The results show that at 3.1 s after the earthquake occurs,the magnitude estimation is 6.5,and the magnitude estimation error is not affected by the change of signal-to-noise ratio(SNR)and epicentral distance.With the increase of the time after the first triggered station,the estimated magnitude based on multi-station gradually approaches the real magnitude.For on-site peak ground velocity(PGV)prediction,at 3 s after the arrival of P wave for each station,there is a 1:1 linear relationship between the predicted PGV and the observed PGV.With the increase of time window after the arrival of P wave,the predicted PGV gradually approaches the observed PGV,and the PGV prediction error is not affected by the change of SNR and epicentral distance.When the instrument intensity threshold of the on-site station is set toⅥ,the percentages of successful alarm,false alarm and missed alarm are 99.53%,0%and 0.47%respectively,the average lead time is 19.62 s,and there is no false alarm and missed alarm in the area where the seismic intensity isⅥ;When the instrument intensity threshold of the on-site station is set toⅦ,the percentages of successful alarm,false alarm and missed alarm are 99.77%,0%and 0.23%respectively,the average lead time is 9.69 s,and there is no false alarm and missed alarm in the area where the seismic intensity isⅦ.The retrospective verification results show that the machine learning method can obtain robust magnitude estimati

关 键 词：地震预警 机器学习 震级估计 现地地震动速度峰值预测 门源地震 

分 类 号：P315[天文地球—地震学]