Том: 5 , Год издания: 2021

Место издания: Amsterdam

We present the approach to seismic modelling combining the conventional grid-based numerical simulation with DNN-based approach using the NDM-net. The first step of the proposed workflow is a generation of the training dataset. It requires forward modelling on a fine grid for a limited number of common shot gather positions (less than 10% in 2D). Next, the forward modelling for the whole acquisition on a coarse grid is done. After that, the NDM-net is trained. The last step is a numerical dispersion elimination by the trained DNN. The proof of a concept was done using a complex 2D elastic model. The presented results demonstrate the ability of NDM-net to make a high-quality seismic data prediction using the synthetics generated on a coarse grid.