Seismic waveform inversion using a neural network-based forward

Abstract

Abstract The purpose of seismic waveform inversion is to obtain a geological model that is optimally fitted to the predicted seismic record and the measured seismic record. Since the forward model is repeatedly called during the inversion process, in order to improve efficiency, an efficient forward calculation method must be employed. In this study, we take a 2D wave equation as an example and propose a deep learning method as a forward model to minimize the prediction error value of seismic records. And the velocity inversion test of the Marmousi model is carried out by conjugate gradient method. Numerical experiments show that compared with the traditional finite difference method, the method can greatly reduce the calculation amount and improve the calculation efficiency.