摘要
地震参数是防震减灾的重要参数,随着地震台网的进一步扩建和地震行业技术的不断提升,地震参数自动快速计算与信息发布水平有了很大的进步,但参数的准确性有待进一步提高.为了改善决定地震破坏力强弱的地震震级关键参数的准确性,本文尝试使用了多种方法来预测地震自动速报震级与人工正式速报震级的误差值.样本数据选取2013—2019年国家地震局自动速报AU系统与人工正式速报所发布的地震参数,震级下限为MS 3.0级,研究区域为国界线以内,将AU速报的震级M结果和人工正式速报结果进行对比分析,利用不同方法总结了两种结果的震级差异分布,首先利用直线拟合以及曲线拟合的方式来研究二者的关系,再利用误差逆向传播神经网络以及基于遗传优化的误差逆向传播神经网络对其进行参数误差学习预测.本文结果显示自动速报震级与人工正式结果的差异性跟震中经纬度有关,具有区域性的特点.神经网络的方法可以改善自动速报与人工正式速报的震级差异,为自动速报的参数准确性提出了新的参考思路.
Seismic parameters are important parameters for earthquake prevention and disaster reduction. With the further expansion of seismic networks and the continuous improvement of seismic industry technology, the level of automatic and rapid calculation and information release of seismic parameters has made great progress, but the accuracy of parameters needs to be further improved. In order to improve the accuracy of earthquake magnitude, which is the key parameter to determine the destructive power of an earthquake, this paper attempts to use a variety of methods to predict the error value between the automatic quick report magnitude and the artificial official quick report magnitude. The lower limit of magnitude is MS 3.0, and the study area is within the national boundary. The magnitude M results of Au quick report and the results of the artificial official quick report are compared and analyzed, and the magnitude difference distribution of the two results is summarized by using different methods, Firstly, the relationship between them is studied by the way of line fitting and curve fitting, and then the error backpropagation neural network and the error back propagation neural network based on genetic optimization is used for parameter error learning prediction. According to the parameter analysis of the two earthquake catalogs, the difference between the automatic quick report magnitude and the artificial official result is related to the longitude and latitude of the epicenter, which has regional characteristics. The spatial distribution is generally small in the East and large in the west, and the difference is small in the middle region. The magnitude difference is generally between 0.2 and 0.4 across the country, Due to the sparsity of regional network stations in some areas, the residual error of magnitude parameters between automatic quick report and official quick report is about 1. It is found that the neural network method based on genetic optimization can improve the magnitude difference between automatic quick reports and manual quick reports, and reduce the error value of the automatic quick reports. The results of this paper provide a new reference for improving the accuracy of the automatic quick reports.
作者
王维欢
尹欣欣
蔡润
张远富
陈继锋
WANG WeiHuan;YIN XinXin;CAI Run;ZHANG YuanFu;CHEN JiFeng(Earthquake Administration of Gansu Province,Lanzhou 730000,China;Chengdu Surveying Geotechnical Research Institute Company Limited of MCC,Chengdu 610063,China)
出处
《地球物理学进展》
CSCD
北大核心
2021年第3期961-967,共7页
Progress in Geophysics
基金
中国地震局地震预测研究所基本科研业务费专项(2019IESLZ07)
中国地震局地震科技星火计划项目(XH19043)联合资助
甘肃省自然科学基金(18JR3RA414)。
关键词
地震自动速报
地震预警
地震震级
神经网络
Earthquake automatic quick report
Earthquake early warning
Earthquake magnitude
Neural network
