Data-driven fast prestack structurally constrained inversion

Abstract

Classic multitrace inversion, such as structurally constrained inversion (SCI), tends to arrange seismic data trace by trace to introduce constraints that are more in line with the structure or reflection features. However, this strategy of rearranging seismic traces extensively increases the number of calculations, thereby limiting the wide application of SCI in prestack inversion. We have developed two improved prestack SCI algorithms that can simultaneously invert all seismic traces with structural constraints in a short time and substantially reduce the dependence on seismic data quality. The key to the proposed techniques lies in achieving structural constraints by introducing a Hadamard product operator without rearranging the seismic traces because it avoids the generation of large-scale and memory-intensive convolution matrices and structural constraint operators. We have deduced a corresponding fast algorithm and named it fast structurally constrained inversion. In addition because the quality of seismic data plays a decisive role in inversion, we further have developed a data-driven fast structurally constrained inversion (DFSCI) algorithm wherein a local crosscorrelation coefficient reflects the reliability of the local seismic data to a certain extent. Therefore, applying DFSCI to control the contribution of seismic data at each sampling point in the inversion can reduce the dependence of seismic data quality. The noise resistance capability and the spatial continuity of the proposed methods have been verified through numerical experiments and a field data example.