Upgoing and Downgoing Wave Separation in Vertical Seismic Profiling Data Based on Deep Learning

Abstract

Summary Vertical seismic profile (VSP) data contains upgoing and downgoing waves, make the VSP wavefield chaotic and complex, which is not conducive to subsequent imaging, interpretation, and inversion. It is difficult to obtain accurate results through the conventional separation methods of upgoing and downgoing waves. Aiming at the high-precision intelligent separation of VSP data, we propose an upgoing and downgoing wavefield separation method for VSP data based on deep learning. The procedure of our method includes two steps. The first step is we simulate the training sets of unseparated data and separated data. The second step is that the designed multi-task learning network to train the training sets and obtain the relationship model of unseparated data and separated data. The application of forward modeling data and the real data by training model indicates that our method is efficacious for separating upgoing and downgoing wave for VSP data. And the results of our method are more accurate than the traditional methods such as f–k filtering.