王腾飞,
徐文才
海洋地质国家重点实验室, 同济大学, 上海 200092
基金项目: 国家重点研发计划（2018YFC0310104；2017YFB0202903），国家自然科学基金（42074157，41630964，41674117）以及中国科学院战略性先导科技专项（A类）（XDA14010203）共同资助
详细信息
作者简介: 程玖兵, 教授, 研究方向: 地震波传播、成像和反演及其在油气勘探和岩石圈成像中的应用.E-mail: cjb1206@tongji.edu.cn
中图分类号: P631 收稿日期:2021-01-07
修回日期:2021-06-09
上线日期:2021-10-10
Second-order optimization of seismic reflection waveform inversion and its application
CHENG JiuBing,WANG TengFei,
XU WenCai
State Key Laboratory of Marine Geology, Tongji University, Shanghai 200092, China
MSC: P631
--> Received Date: 07 January 2021
Revised Date: 09 June 2021
Available Online: 10 October 2021
摘要
摘要:常规地震数据大都缺乏低频成分与长炮检距信号，经典的全波形反演方法不易获取中、深层弹性参数模型的长波长分量，波动方程反射波形反演作为替代方法近来受到极大关注.然而，现有的反射波形反演方法几乎都采用梯度类的一阶优化算法，收敛性和精度都有待提高.本文在二阶优化理论框架下，推导弹性参数背景与扰动模型的反射波敏感核、泛函梯度以及海森算子，揭示海森矩阵对泛函梯度的去模糊化作用和改善反演的工作机制.推覆体模型合成数据实验表明，相比于常用的共轭梯度法，利用近似海森矩阵的高斯-牛顿法明显提升了反射波形反演的收敛性与宽谱建模能力.在东海实例中，本文方法超越常用的反射走时层析技术，通过改善中、深层偏移速度建模，支撑逆时偏移高分辨率刻画长江坳陷内部复杂的断裂系统，改善了深部基底的成像质量.
关键词: 反射波形反演/
二阶优化/
海森矩阵/
高斯-牛顿法/
长江坳陷
Abstract:Conventional surface seismic data lack of usable low-frequency and long-offset signals, which leads to classical full-waveform inversion hardly retrieving long-wavelength contents of elastic models in the mid-to-deep parts. As an alternative approach, wave-equation reflection waveform inversion has drawn more attentions. However, almost all existing studies on reflection waveform inversion only use first-order optimization and thus have great room for improvement in convergence rate and accuracy. Here, in the framework of second-order optimization, we have derived reflection sensitivity kernels, functional gradient and Hessian operators with respect to background and perturbation components of the elastic models. We have demonstrated the debluring effect of Hessian operator for the functional gradient and revealed the working mechanism to provide optimal model updating and enhance inversion resolution. The synthetic example on an overthrust model shows that, compared with widely used conjugate-gradient method, approximate Hessian-based Gauss-Newton method significantly enhances the convergence rate and the capability of broadband model building. In the real data example of East China Sea, the proposed reflection waveform inversion outperforms widely used reflection travel-time tomography. It has improved velocity model building and thus supported reverse-time migration to delineate the fault systems and the basement in the Changjiang sag with higher resolution.
Key words:Reflection waveform inversion/
Second-order optimization/
Hessian matrix/
Gauss-Newton method/
Changjiang sag
PDF全文下载地址:
http://www.geophy.cn/data/article/export-pdf?id=a0d04287-198e-4b72-bf76-9cd556c82784
