Efficient pseudo-Gauss-Newton full-waveform inversion in the τ-p domain
Efficient pseudo-Gauss-Newton full-waveform inversion in the τ-p domain
复制标题
DOI：
10.1190/geo2014-0224.1
复制
发表时间：
2015-07
期刊：
影响因子：
3.3
通讯作者：
W. Pan;K. Innanen;G. Margrave;D. Cao
中科院分区：
文献类型：
--
作者：
W. Pan;K. Innanen;G. Margrave;D. Cao
摘要全波形反演（FWI）旨在通过迭代最小化合成数据与观测数据之间的差异来估计地下弹性特性。它的工业应用面临着几个明确的障碍，包括计算成本高、收敛速度慢和周期跳变现象。我们开发了一种有效的τ-p域波形反演，旨在通过相位编码技术减少FWI的计算负担。在τ-p域中使用线性相位编码构造梯度，并采用倾斜更新策略进一步减少了计算量。低比例和模糊的梯度更新可以使用精确或近似的逆Hessian版本来增强，这导致更快的收敛速度。我们开发了一种新的啁啾相位编码策略，用于对角黑森构造。使用对角相位编码近似Hessian来预处理梯度，形成我们所说的伪高斯-牛顿（PGN）步骤。为了测试…
ABSTRACTFull-waveform inversion (FWI) seeks to estimate subsurface elastic properties by iterative minimization of the difference between synthetic and observed data. Its industrial application suffers from several well-defined obstacles, including high computational cost, slow convergence rate, and the phenomenon of cycle skipping. We have developed an efficient τ-p domain waveform inversion aimed at reducing the computational burden of FWI with a phase-encoding technique. The gradient is constructed in the τ-p domain using linear phase-encoding, and a slant update strategy further reduces the computational burden. Poorly scaled and blurred gradient updates can be enhanced using exact or approximate versions of the inverse Hessian, which leads to a faster convergence rate. We developed a new chirp phase-encoding strategy for diagonal Hessian construction. Preconditioning the gradient using the diagonal phase-encoded approximate Hessian forms what we refer to as a pseudo-Gauss-Newton (PGN) step. To test t...