Seismic Inversion by Hybrid Machine Learning
Сейсмическая инверсия с использованием гибридного машинного обучения
2021-09-01
SCID: 54.1/kp6wuymd
Discuss with AI
automatic differentiationconvolutional autoencodercycle-skipping problemhybrid machine learningseismic inversion
Figures from the paper
Abstract (AI)
Abstract 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.
Key Findings
1
Automatic differentiation establishes the previously unavailable relationship between latent-feature perturbations and velocity perturbations, enabling wave-equation inversion.
2
Full- and semi-automatic gradient approaches are mathematically equivalent; the semi-automatic approach is computationally more efficient, while numerical tests recover both low- and high-wavenumber velocity information.
3
Latent features of different dimensions support multiscale inversion: lower-dimensional features emphasize kinematic information, while larger-dimensional features preserve dynamic information.
4
Latent-space misfit is less affected by cycle skipping than high-dimensional waveform mismatch, improving the robustness of seismic inversion.
5
The hybrid machine-learning method estimates subsurface velocity models by inverting convolutional autoencoder latent-space features of seismic data.
Research Object
subsurface velocity model
Research Subject
multiscale recovery of low- and high-wavenumber velocity information from seismic data using latent-space inversion, with reduced cycle-skipping sensitivity
Publication Details
Publication Date
2021-09-01
Journal
Publisher
ISSN
Cited by
41
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest