Seismic Wave Propagation and Inversion with Neural Operators
Распространение и инверсия сейсмических волн с использованием нейронных операторов
2021-10-01
SCID: 54.1/hmfr6cau
Discuss with AI
full waveform inversionneural operatorsreverse-mode automatic differentiationseismic tomographyseismic wave propagation
Figures from the paper
Abstract (AI)
Abstract Seismic wave propagation forms the basis for most aspects of seismological research, yet solving the wave equation is a major computational burden that inhibits the progress of research. This is exacerbated by the fact that new simulations must be performed whenever the velocity structure or source location is perturbed. Here, we explore a prototype framework for learning general solutions using a recently developed machine learning paradigm called neural operator. A trained neural operator can compute a solution in negligible time for any velocity structure or source location. We develop a scheme to train neural operators on an ensemble of simulations performed with random velocity models and source locations. As neural operators are grid free, it is possible to evaluate solutions on higher resolution velocity models than trained on, providing additional computational efficiency. We illustrate the method with the 2D acoustic wave equation and demonstrate the method’s applicability to seismic tomography, using reverse-mode automatic differentiation to compute gradients of the wavefield with respect to the velocity structure. The developed procedure is nearly an order of magnitude faster than using conventional numerical methods for full waveform inversion.
Key Findings
1
A neural-operator framework learns general solutions to the 2D acoustic wave equation across varying velocity structures and source locations.
2
After training on simulations with random velocity models and source locations, the neural operator evaluates new wavefields in negligible time.
3
Because neural operators are grid-free, the trained model can evaluate solutions on higher-resolution velocity models than those used during training.
4
Reverse-mode automatic differentiation enables efficient computation of wavefield gradients with respect to velocity structure for seismic tomography.
5
The procedure is nearly an order of magnitude faster than conventional numerical methods for full waveform inversion.
Research Object
2D acoustic seismic wave propagation governed by the wave equation across varying velocity structures and source locations
Research Subject
Neural-operator modeling and inversion of wavefields, including gradient computation with respect to velocity structure and computational efficiency for full waveform inversion
Publication Details
Publication Date
2021-10-01
Journal
Publisher
ISSN
Cited by
144
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest