摘要
全波形反演利用地震记录中的振幅、走时和相位等信息,通过拟合实际地震记录和计算波场来定量提取地下介质的弹性参数,进而为勘探地震成像、速度建模以及大尺度构造演化分析等提供可靠依据.但全波形反演计算量巨大,特别是应用于三维大区块叠前数据时,生产成本仍然很高.本文介绍并比较了时间域和频率域的全波形反演方法,综合两者的优点,最终采用混合域的反演算法,并且在此基础上做了进一步的简化以提高计算效率.针对全波形反演方法应用于大规模叠前数据时易陷入局部极小值的问题,我们提出对模型数据进行分割,同时在数个小模型内进行梯度搜索,然后对比各个局域的梯度,最终找出合适的全局下降方向,以克服局部极小的隐患.该方法能够充分利用GPU的硬件特性.在GPU环境下实现本文所提出的简化混合域全波形反演算法.数值计算实例体现出新方法具有良好的计算效率、反演精度和算法可扩展性.
By fitting synthetic data to the seismic observations, full waveform inversion （FWI） uses the amplitude, traveltime, and phase information embedded in the recorded data to quantitatively extract the elastic rock properties and to provide reliable information for seismic imaging, velocity model building, and large-scale tectonic evolution interpretation, etc. However, the computational cost of FWI could be overwhelming, especially for large size 3D prestack seismic surveys. We compare FWI algorithms in both time domain and frequency domain and evaluate their pros and cons, respectively. Then we propose to implement FWI in hybrid domain and simplify the method to reduce production cost. To overcome the possible local minimum traps during the inversion, we propose to divide the model into several small pieces. The method computes the gradient of the objective function in each piece, compares all the local gradients and derives an optimal global inversion update. We implemented our simplified hybrid domain FWI method on GPUs. The numerical experiments reported in this paper demonstrate that our proposed inversion algorithm is effective, efficient, and scalable.
出处
《地球物理学报》
SCIE
EI
CAS
CSCD
北大核心
2017年第2期665-677,共13页
Chinese Journal of Geophysics
基金
国家高技术研究发展计划(2012AA061202))资助
关键词
全波形反演
混合域
三维模型
大规模
多GPU并行
Full waveform inversion
Hybrid domain
3D-model
Large-scale
Multi-GPU acceleration
