Feasibility of Using Fourier Neural Operators for Three-Dimensional Elastic Seismic Simulations

Qingkai Kong | 23-FS-021

Project Overview

Three-dimensional (3D) seismic simulations are important in many of the seismological applications related to our lab's mission, however, high-fidelity simulations computationally costly. We investigated the capabilities of the newly developed Fourier Neural Operator (FNO) to solve the 3D elastic wave equations for seismic simulations. We generated simulation data for training the FNO model, and analyzed its performance on various test cases, such as the different performance with different number of training data, test on different resolution data, and test on more realistic structures. We found the FNO model can reproduce 3D seismic simulations to a high accuracy but with ~169 times faster in a spatial grid size of 16 x 16 x16. When applied on higher resolution data, we found fine-tuning with a small amount of data will achieve the best results. This feasibility study showed promising results of using FNO for 3D seismic simulations and laid the foundation of further research to develop this into a mature approach for different seismological applications. The potential impact of this project will provide techniques for large-scale or real-time applications of solving PDEs from national lab's programs.

Mission Impact

This research combines simulation and data science applied to capabilities supporting national security programs at Lawrence Livermore National Laboratory - areas of expertise and core relevance to the Lab. Using applications from Earth and Atmospheric Sciences, we investigated a generic approach to solve real-world PDEs using Fourier Neural Operator that has the benefits of both efficiency and accuracy. This study has a direct impact in characterizing emerging technologies and improving nuclear test detection and reservoir seismicity characterization. The overall framework will make contribution to develop science and technology tools and capabilities to meet future national and energy security challenges.

Publications, Presentations, and Patents

Q. Kong, et al, "3D Seismic Waveform Modeling using U-Shaped Fourier Neural Operator (U-NO)" (Invited Oral Presentation, Seismological Society Association Annual Meeting, San Juan, Puerto Rico, April 17-20, 2023). LLNL-PRES-847282

Q. Kong, et al, "Solving 3D Elastic Wave Equation using Neural Operators" (Oral Presentation, Working in Progress, LLNL, Livermore, CA, 2023). LLNL-PRES-849432

Q. Kong, et al, "Rapid 3D Seismic Waveform Modeling using Fourier Neural Operator" (Invited Oral Presentation, SZ4D Machine Learning & Artificial Intelligence Virtual Workshop, 2023). LLNL-PRES-852188