基于机器学习算法的银川台阵资料自动化震相分析和定位

近年来快速发展的机器学习算法显著提高了震相拾取的精度和效率。 采用卷积神经网络和递归神经网络的震相识别方法对银川台阵2019年6~7月的连续波形数据进行事件检测和P、 S震相拾取, 并通过快速震相关联和事件定位得到了银川地区较全的地震目录。 结果表明, 当震相数小于10时, 虽然可以检测出较多事件, 但分布呈弥散状, 与区域地震活动特征不符。 进一步对震相数≥10的事件进行了人工复核。 总体而言, 随着震相数量的增加, 事件的误检率逐步降低。 震相数16是该地区自动检测和定位结果准确性的拐点。 当震相数≥20时, 全部召回了地震目录中的13个地震事件, 二者平均定位差异4.27 km。 经过人工复核, 检测到的真实地震事件为区域内地震目录中事件数量的9倍。 本文使用的基于机器学习和快速震相关联和定位方法的流程可在确保准确率的基础上降低人工检测的难度, 提高地震检测的效率。

Automatic Seismic Phase Analysis and Earthquake Location Using Yinchuan Array Datasets based on a Machine Learning Algorithm

In recent years,the rapid development of machine learning algorithms have significantly improved the accuracy and efficiency of seismic phase picking.In this paper,a hybrid neural network algorithm,which comprised by convolutional neural networks(CNN)and recurrent neural networks(RNN)algorithms,is used to on continuous waveform data recorded by Yinchuan Array in June and July,2019 for earthquake detection and phase picking.Phases from RNN are associated and located by using the Rapid Earthquake Association and Location method(REAL),and then compared with the routine catalog.The results show that when the number of seismic phases is less than 10,although more events can be detected,the distribution is diffuse,which is inconsistent with the characteristics of regional seismicity.The events with 10 or more seismic phases are manually inspected.In general,as the number of seismic phases increases,the false detection rate is lower.16 is the critical number of seismic phases to achieve accurate phase detection and location results.When the number of seismic phases is 20 or more,13 earthquakes in the routine catalog are all recalled.The average location offset between the relocation and routine catalog result is 4.27 km.Besides,the number of manually inspected earthquakes from the automatic procedure is 9 times of that from the routine catalog.It suggests that the procedure which combined the machine learning and REAL method is capable of reducing the difficulty and improving the accuracy and efficiency of earthquake detection.