基于频域衰减的时域全波形反演

时域全波形反演由于采用了全频段信息,因此在迭代过程中不同波长的信息不能由低到高的逐步重建,极易陷入局部极小值.本文通过分频段的方式,对地震数据做正反傅里叶变换,利用频域指数衰减的方法逐级分离出地震数据中的高频成分,在时域上实现由低频向高频的波形反演,从而降低了反演的非线性,使不同波长的信息得到稳步恢复.同时,在高频成分衰减的过程中,后至波的能量也被削弱,由此也降低了深层反射在初始反演过程中的干扰.整个反演仅增加对数据做正反傅里叶变换过程,相较于混合域反演,无需提取全部波场的相应频率成分.在计算效率方面,利用GPU进行加速,并采用CUDA自带函数库中cufft来提高计算效率.通过对Marmousi模型测试,验证了所述方法的有效性.

Time domain full waveform inversion based on frequency attenuation

Because the full frequency range information is used in time domain full waveform inversion, different wavelengths can not be reconstructed gradually. So it is easy to fall into local minimum value. In this paper, we extract different frequency bands from seismic data by Fourier transform. Different level of high frequency components can be separated by attenuating corresponding frequency ranges, and time domain full waveform inversion is implemented from low to high frequency. At the same time, latter arrivals are attenuated and this can reduce the interference in the process of shallow part inversion. Fourier transform is only required for the shot gathers, compared with hybrid domain inversion, and don't need to extract the corresponding frequency components of all wavefields. In the aspect of computing efficiency, the algorithm is based on GPU and uses cufft to further improve the efficiency of calculation. Through the Marmouse model test, the validity of this method is verified.