Test of on-site alert-level earthquake early warning based on machine learning prediction models:A case for the Sichuan Luding M6.8 earthquake on September 5,2022
Test of on-site alert-level earthquake early warning based on machine learning prediction models:A case for the Sichuan Luding M6.8 earthquake on September 5,2022
At 12∶52 on September 5,2022,a M6.8 earthquake occurred in Luding County,Garzê Prefecture,Sichuan Province,causing serious economic losses and casualties.This paper uses the strong-motion data recorded by the station for this earthquake to off-line simulate the on-site alert-level earthquake early warning(EEW)method based on machine learning prediction models.The method first constructs magnitude estimation model and on-site peak ground velocity(PGV)prediction model based on the support vector machine(SVM),and then compares the predicted magnitude and PGV of each station with the magnitude threshold(M=5.7)and the PGV threshold(PGV=9.12cm·s-1)respectively,so as to obtain the on-site alert levels(0,1,2,3),which are used to judge whether there is potential damage near the station.The alert level 3 denotes that the predicted magnitude and PGV both exceed the thresholds,indicating that there is earthquake potential damage near the station and the event is high magnitude.The off-line simulation results show that using 3 s P-wave time window,the standard deviation of magnitude prediction error of a single station based on SVM magnitude estimation model is 0.35 and the mean absolute error is 0.27;the standard deviation of PGV prediction error based on SVM PGV prediction model is 0.34,and the mean absolute error is 0.32;the magnitude prediction error and PGV prediction error are mainly distributed within±2 times of the standard deviation.Without considering the data packaging and transmission delay,almost all triggered stations in the seismic intensity Ⅶ area issued alert level 3 at 8 s after the earthquake occurs.The on-site alert-level EEW method based on machine learning prediction models can obtain reliable alarms at the initial stage after the earthquake occurs,which provides a potential reference for the upgrading of Chinese EEW system.
关键词
现地地震预警/机器学习/震级预测/PGV预测/泸定地震
Key words
On-site earthquake early warning/Machine learning/Magnitude prediction/PGV prediction/Luding earthquake
维普
万方数据