摘要

Time-domain full waveform inversion algorithms store the source wavefield to calculate the cross-correlation between the source and receiver wavefields to obtain the gradient direction. Saving the full source wavefield imposes an enormous burden on computer memory resources. We apply the excitation amplitude method to a full waveform inversion to reduce the memory overload. This method removes the time dimension of the source wavefield by only exploiting the maximum amplitude signals. By adopting the excitation amplitude method to store the source wavefield and to calculate the cross-correlation, we can reduce the memory requirement for a full waveform inversion by three orders of magnitude. Since the excitation amplitude method cannot handle multipathing within the source wavefield, a gradient direction obtained using this method is an approximation to the original gradient. Nevertheless, synthetic inversion examples using the Marmousi and overthrust models demonstrate the efficiency and accuracy of the proposed scheme.

  • 出版日期2017-10