A deep learning formulation of viscoelastic VTI full waveform inversion

Tianze Zhang, Jian Sun, Kristopher A. Innanen, Daniel O. Trad

In this study, we use the recurrent neural network (RNN) to achieve viscoelastic VTI full waveform inversion. Eight parameters are simultaneously inverted, which are elastic parameters C11, C13, C33, C44 and their corresponding attenuation parameters Qc11, Qc13, Qc33, Qc44. The recurrent neural network is built according to the stress velocity VTI vis- coelastic wave equation. We also study the acquisition influence on the inversion results. Numerical inversion results show that the combination of cross well data and the surface data can help to better recover the elastic parameters compared with only surface acquisi- tion in which the receivers and shots are all on the surface of the model. To mitigate the cross-talk between the parameters, we also use the high order total variation (TV) to miti- gate the cross-talk. The simple structure model and complex part of the overthrust model proves the validation of this method.