Abstract:Full Waveform Inversion (FWI) is an inverse problem for estimating the wave velocity distribution in a given domain, based on observed data on the boundaries. The inversion is computationally demanding because we are required to solve multiple forward problems, either in time or frequency domains, to simulate data that are then iteratively fitted to the observed data. We consider FWI in the frequency domain, where the Helmholtz equation is used as a forward model, and its repeated solution is the main computational bottleneck of the inversion process. To ease this cost, we integrate a learning process of an encoder-solver preconditioner that is based on convolutional neural networks (CNNs). The encoder-solver is trained to effectively precondition the discretized Helmholtz operator given velocity medium parameters. Then, by re-training the CNN between the iterations of the optimization process, the encoder-solver is adapted to the iteratively evolving velocity medium as part of the inversion. Without retraining, the performance of the solver deteriorates as the medium changes. Using our light retraining procedures, we obtain the forward simulations effectively throughout the process. We demonstrate our approach to solving FWI problems using 2D geophysical models with high-frequency data.

What problem does this paper attempt to address?

This paper focuses on the computational efficiency issue in Full Waveform Inversion (FWI). FWI is an inverse problem used to estimate the distribution of wave velocities based on observed data at the boundaries. It usually requires solving multiple forward problems (such as the Helmholtz equation in the time or frequency domain) to simulate data and iteratively fit it to the observed data. As solving the Helmholtz equation in the frequency domain repeatedly is computationally expensive, researchers propose an encoder-solver preconditioning method based on Convolutional Neural Networks (CNNs) to accelerate this process. In the paper, the authors utilize an encoder-solver architecture consisting of two CNNs, where the encoder encodes velocity medium parameters and the solver serves as a preconditioner to efficiently solve the linear system given the encoded medium parameters. Through lightweight retraining during the optimization iterations, the solver adapts to the gradually changing medium, maintaining efficient forward simulations throughout the entire inversion process. The study points out that traditional end-to-end deep learning methods may not be suitable for different experimental settings, while physics-guided methods are more flexible and faithful to numerical simulations. The proposed method focuses on accelerating the physical simulation and reducing the computational burden of the entire inversion process, especially in high-precision applications that do not require real-time estimation, such as oil and gas exploration. The related work section mentions end-to-end neural network FWI solvers, physics-guided neural networks, and methods integrating deep learning into classical inversion processes. The paper also introduces frequency continuation and Gauss-Newton methods for solving FWI problems, as well as a multi-grid enhanced CNN preconditioner for accelerated solving. To sum up, this paper aims to address how deep learning can be used to accelerate computations in Full Waveform Inversion, particularly through a lightweight retraining strategy to improve adaptivity to continuously changing wave velocity models and reduce computational costs for solving complex inverse problems.

Physics-guided Full Waveform Inversion using Encoder-Solver Convolutional Neural Networks

Physics-guided Full Waveform Inversion using Encoder-Solver convolutional neural networks

Multiscale Data-driven Seismic Full-waveform Inversion with Field Data Study

InversionNet: A Real-Time and Accurate Full Waveform Inversion with CNNs and continuous CRFs

Physics-guided deep autoencoder to overcome the need for a starting model in full-waveform inversion

Frequency-domain Full-Waveform Inversion with an Efficient Iterative Solver Approach

Reparameterized full-waveform inversion using deep neural networks

Integrating Deep Neural Networks with Full-waveform Inversion: Reparametrization, Regularization, and Uncertainty Quantification

Unsupervised learning of full-waveform inversion: Connecting CNN and partial differential equation in a loop

Learning to Invert Pseudo-Spectral Data for Seismic Waveform Inversion

Full waveform inversion with nonlocal similarity and model-derivative domain adaptive sparsity-promoting regularization

InversionNet3D: Efficient and Scalable Learning for 3D Full Waveform Inversion

Data-driven Ringed Residual U-Net Scheme for Full Waveform Inversion

FWIGAN: Full‐waveform inversion via a physics‐informed generative adversarial network

Robust Full Waveform Inversion: A Source Wavelet Manipulation Perspective.

Accelerating Full Waveform Inversion By Transfer Learning

Robust Full Waveform Inversion with deep Hessian deblurring

Efficient Pseudo-Gauss-newton Full-Waveform Inversion in the Τ-P Domain

Joint-Domain Full Waveform Inversion and Its Applications to Ultra-Shallow Water OBN Data

An Unsupervised Deep Learning Approach for the Wave Equation Inverse Problem