摘要
为提高地震预警参数的预测精度,提出一种基于卷积神经网络和门控循环单元的震中距预测模型(记为CGR),该模型输入P波初至0.5~3 s的数据,输出震中距预测值,同时提取波形和时间序列特征,实现震中距的端到端预测。利用日本KiK-net数据库的地震记录对CGR模型进行训练和测试,并与B-Δ方法进行对比。测试结果表明CGR模型在P波到时后0.5 s便拥有相比于传统方法更高的准确性,该模型避免了传统方法的人为干扰,显著改善地震预警中震中距估算的时效性和准确性。
In order to improve the prediction accuracy of earthquake early warning parameters,an epicenter distance prediction model(denoted as CGR)based on convolutional neural network and gated recurrent unit is proposed,which inputs the P-wave first arrivals of 0.5-3 seconds,outputs the predicted value of the epicenter distance,and extracts the waveform and time-series features at the same time to realize the end-to-end prediction of the epicenter distance.In this paper,the CGR model is trained and tested using seismic records from the KiK-net database in Japan,and compared with the B-Δmethod.The test results show that the CGR model has higher accuracy than the traditional method at 0.5 seconds after the arrival of the P-wave.The model avoids the human interference of the traditional method,and significantly improves the timeliness and accuracy of the epicentral distance estimation in earthquake early warning.
作者
林建平
王郭艳
王阳
王延伟
LIN Jian-ping;WANG Guo-yan;WANG Yang;WANG Yan-wei(School of Civil&Architectural Engineering,Guilin University of Technology,Guilin 541000,China)
出处
《南阳理工学院学报》
2024年第6期50-56,共7页
Journal of Nanyang Institute of Technology
基金
国家自然科学基金项目(51968016)。
关键词
深度学习
卷积神经网络
门控循环单元
震中距
deep learning
convolutional neural network
gated recurrent unit
epicenter distance
