11 January 2024 in Geophysics
We review five types of physics-informed machine-learning (PIML) algorithms for inversion and modeling of geophysical data. Such algorithms use the combination of a data-driven machine-learning (ML) method and the equations of physics to model or invert geophysical data (or both). By incorporating the constraints of physics, PIML algorithms can effectively reduce the size of the solution space for ML models, enabling them to be trained on smaller data sets. This is especially advantageous in scenarios in which data availability may be limited or expensive to obtain. In this review, we restrict the physics to be that from the governing wave equation, either as a constraint that must be satisfied or by using numerical solutions of the wave equation for modeling and inversion. This approach ensures that the resulting models adhere to physical principles while leveraging the power of ML to analyze and interpret complex geophysical data. There are several potential benefits of PIML compared to standard numerical modeling or inversion of seismic data computed by, for example, finite-difference solutions to the wave equation. Empirical tests suggest that PIML algorithms constrained by the physics of wave propagation can sometimes resist getting stuck in a local minima compared with standard full-waveform inversion (FWI).After the weights of the neural network are found by training, then the forward and inverse operations by PIML can be more than several orders of magnitude more efficient than FWI. However, the computational cost for general training can be enormous.If the ML inversion operator Hw is locally trained on a small portion of the recorded data dobs, then there is sometimes no need for millions of training examples that aim for global generalization of Hw. The benefit is that the locally trained Hw can be used to economically invert the remaining test data dtest for the true velocity m≈Hwdtest, where dtest can comprise more than 90% of the recorded data.
1 November 2017 in Geophysics
Viscoacoustic least-squares reverse time migration, also denoted as Q-LSRTM, linearly inverts for the subsurface reflectivity model from lossy data. Compared with conventional migration methods, it can compensate for the amplitude loss in the migrated images due to strong subsurface attenuation and can produce reflectors that are accurately positioned in depth. However, the adjoint propagators used for backward propagating the residual data are also attenuative. Thus, the inverted images from -LSRTM with a small number of iterations are often observed to have lower resolution when compared with the benchmark acoustic LSRTM images from acoustic data. To increase the resolution and accelerate the convergence of -LSRTM, we used viscoacoustic deblurring filters as a preconditioner for -LSRTM. These filters can be estimated by matching a simulated migration image to its reference reflectivity model. Numerical tests on synthetic and field data demonstrate that -LSRTM combined with viscoacoustic deblurring filters can produce images with higher resolution and more balanced amplitudes than images from acoustic RTM, acoustic LSRTM, and -LSRTM when there is strong attenuation in the background medium. Our preconditioning method is also shown to improve the convergence rate of -LSRTM by more than 30% in some cases and significantly compensate for the lossy artifacts in RTM images.
23 September 2021 in Journal of Geophysical Research: Solid Earth
We present a hybrid machine learning (HML) inversion method, which uses the latent space (LS) features of a convolutional autoencoder (CAE) to estimate the subsurface velocity model. The LS features are the effective low‐dimensional representation of the high‐dimensional seismic data. However, no equations exist to describe the relationship between the perturbation of an LS feature and the velocity perturbation. To address this problem, we use automatic differentiation (AD) to connect the two terms. Following this step, we use the wave‐equation inversion to invert the LS features for the subsurface velocity model. The HML misfit function measures the LS feature differences between the observed and predicted seismic data in a low‐dimensional space, which is less affected by the cycle‐skipping problem compared to the waveform mismatch in a high‐dimensional space. A low dimensional LS feature mainly contains the kinematic information of seismic data, while a large dimensional LS feature can also preserve the dynamic information of seismic data. Therefore, the HML inversion can recover the subsurface velocity model in a multiscale approach by inverting the LS features with different dimensions. Based on the different ways of utilizing AD to compute the velocity gradient, we propose full‐ and semi‐automatic approaches to solve this problem. These two approaches are mathematically equivalent; the former is easier to implement, while the latter is computationally more efficient. Numerical tests show that the HML inversion method can effectively recover both the low‐ and high‐wavenumber velocity information by inverting the LS features with different dimensions.
19 December 2019 in Geophysics
Reflection full-waveform inversion (RFWI) can recover the low-wavenumber components of the velocity model along with the reflection wavepaths. However, this requires an expensive least-squares reverse time migration (LSRTM) to construct the perturbation image that can still suffer from cycle-skipping problems. As an inexpensive alternative to LSRTM, we use migration deconvolution (MD) with RFWI. To mitigate cycle-skipping problems, we develop a multiscale reflection phase inversion (MRPI) strategy that boosts the low-frequency data and should only explain the phase information in the recorded data, not its magnitude spectrum. We also use the rolling-offset strategy that gradually extends the offset range of data with an increasing number of iterations. Numerical results indicate that the MRPI + MD method can efficiently recover the low-wavenumber components of the velocity model and is less prone to getting stuck in local minima compared to conventional RFWI.
4 June 2020 in Geophysics
We present a wave-equation inversion method that inverts skeletonized seismic data for the subsurface velocity model. The skeletonized representation of the seismic traces consists of the low-rank latent-space variables predicted by a well-trained autoencoder neural network. The input to the autoencoder consists of seismic traces, and the implicit function theorem is used to determine the Fréchet derivative, i.e., the perturbation of the skeletonized data with respect to the velocity perturbation. The gradient is computed by migrating the shifted observed traces weighted by the skeletonized data residual, and the final velocity model is the one that best predicts the observed latent-space parameters. We denote this as inversion by Newtonian machine learning (NML) because it inverts for the model parameters by combining the forward and backward modeling of Newtonian wave propagation with the dimensional reduction capability of machine learning. Empirical results suggest that inversion by NML can sometimes mitigate the cycle-skipping problem of conventional full-waveform inversion (FWI). Numerical tests with synthetic and field data demonstrate the success of NML inversion in recovering a low-wavenumber approximation to the subsurface velocity model. The advantage of this method over other skeletonized data methods is that no manual picking of important features is required because the skeletal data are automatically selected by the autoencoder. The disadvantage is that the inverted velocity model has less resolution compared with the FWI result, but it can serve as a good initial model for FWI. Our most significant contribution is that we provide a general framework for using wave-equation inversion to invert skeletal data generated by any type of neural network. In other words, we have combined the deterministic modeling of Newtonian physics and the pattern matching capabilities of machine learning to invert seismic data by NML.
2 June 2020 in Geophysics
We have recast the forward pass of a multilayered convolutional neural network (CNN) as the solution to the problem of sparse least-squares migration (LSM). The CNN filters and feature maps are shown to be analogous, but not equivalent, to the migration Green’s functions and the quasi-reflectivity distribution, respectively. This provides a physical interpretation of the filters and feature maps in deep CNN in terms of the operators for seismic imaging. Motivated by the connection between sparse LSM and CNN, we adopt the neural network version of sparse LSM. Unlike the standard LSM method that finds the optimal reflectivity image, neural network LSM (NNLSM) finds the optimal quasi-reflectivity image and the quasi-migration Green’s functions. These quasi-migration Green’s functions are also denoted as the convolutional filters in a CNN and are similar to migration Green’s functions. The advantage of NNLSM over standard LSM is that its computational cost is significantly less and it can be used for denoising coherent and incoherent noise in migration images. Its disadvantage is that the NNLSM quasi-reflectivity image is only an approximation to the actual reflectivity distribution. However, the quasi-reflectivity image can be used as an attribute image for high-resolution delineation of geologic bodies.