基于深度神经网络EikoNet走时计算方法及应用  

EikoNet traveltime calculation method and application based on deep neural network

作　　者：姚时 侯爵 黄跃鹏 徐涛[1,4] 白志明[1,4] 高正辉[5] Yao Shi;Hou Jue;Huang Yuepeng;Xu Tao;Bai Zhiming;Gao Zhenghui(State Key Laboratory of Lithospheric Evolution,Institute of Geology and Geophysics,Chinese Academy of Sciences,Beijing 100029,China;University of the Chinese Academy of Sciences,Beijing 100049,China;Institute of Geophysics,China Earthquake Administration,Beijing 100081,China;Innovation Academy for Earth Science,Chinese Academy of Sciences,Beijing 100029,China;College of Geo-exploration Science and Technology,Jilin University,Changchun 130026,China)

机构地区：[1]中国科学院地质与地球物理研究所岩石圈演化国家重点实验室,北京100029 [2]中国科学院大学,北京100049 [3]中国地震局地球物理研究所,北京100081 [4]中国科学院地球科学研究院,北京100029 [5]吉林大学地球探测科学与技术学院,长春130026

出　　处：《地球与行星物理论评（中英文）》2023年第1期81-90,共10页Reviews of Geophysics and Planetary Physics

基　　金：国家自然科学基金资助项目(42130807)。

摘　　要：地震波走时计算在层析成像、偏移成像和微震定位等地震学领域中都有重要作用.使用有限差分方法求解程函方程是地震波走时计算的重要方法之一.常规程函方程求解方法需要计算每一个震源激发的走时场,随着网格数量的增加会消耗大量的时间和存储空间.本文介绍了基于深度神经网络的EikoNet走时计算方法,该方法构建了一个包含速度和走时场偏差之间关系的深度神经网络,通过在三维空间中采样生成训练样本,以给定的速度模型为标签实现训练过程中对网络的优化,在计算走时过程中,能传递关于地震波场和速度结构的信息,而且高度适用于GPU,可以无网格地快速确定三维域中任意两点之间的走时,大大提高了计算效率并降低了内存消耗.EikoNet方法和常规快速推进法(FMM)在几个速度模型上的数值实验表明EikoNet方法在保持高精度的同时还具有更高的效率.Seismic wave traveltime calculation plays an important role in many areas of seismology,such as seismic tomography,migration and microseismic location.Solving the eikonal equation with the finite difference method is an essential method for calculating traveltime.The conventional method of solving the eikonal equation needs to calculate the traveltime field of each source.As the number of grids increases,it will consume a lot of time and memory.We introduce the EikoNet based on a deep neural network.Its samples are generated by sampling in the three-dimensional space,using the given velocity model as labels to optimize the network.Furthermore,it cantransmit information about seismic wavefield and velocity structure during calculation and is highly suited for GPU.The EikoNet can quickly determine the traveltime between any two points in a three-dimensional domain without meshes,significantly improving calculation efficiency and reducing memory consumption.Numerical experiments of the EikoNet and the fast marching method(FMM)on several velocity models show that the EikoNet has higher efficiency while maintaining high accuracy.

关 键 词：走时计算 程函方程 EikoNet 快速推进法 

分 类 号：P315[天文地球—地震学]