← Back to Sessions

Rapid 3D Seismic Waveform Modeling using U-Shaped Neural Operators (U-NO)

Description: 

Recent development of Neural Operators such as the Fourier neural operator enables accurate approximation of the various operators used in scientific research. Within seismology, successful application of Fourier neural operators on solving the acoustic wave equation and the 2D elastic wave equation and apply them to full waveform inversion build the basis for further development of the methods. In this presentation, we extend prior efforts to use U-NO for solving the 3D elastic wave equation in seismic waveform modeling. Using the high-performance computing facilities, a general solution operator to the 3D elastic wave equation can be learned from an ensemble of numerical simulations generated by using random velocity models as well as the source locations. We will show the initial results as well as the difficulties we encountered to train the U-NO model.

Session: Opportunities and Challenges for Machine Learning Applications in Seismology

Type: Oral

Date: 4/19/2023

Presentation Time: 04:45 PM (local time)

Presenting Author: Qingkai Kong

Student Presenter: No

Invited Presentation: Yes

Authors