Elastic-AdjointNet: A physics-guided deep autoencoder to overcome crosstalk effects in multiparameter full-waveform inversion

Abstract

PreviousNext No AccessSecond International Meeting for Applied Geoscience & EnergyElastic-AdjointNet: A physics-guided deep autoencoder to overcome crosstalk effects in multiparameter full-waveform inversionAuthors: Arnab DharaMrinal SenArnab DharaUniversity of Texas at AustinSearch for more papers by this author and Mrinal SenUniversity of Texas at AustinSearch for more papers by this authorhttps://doi.org/10.1190/image2022-3745050.1 SectionsSupplemental MaterialAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail AbstractFull Waveform Inversion (FWI) is the most popular technique to obtain high resolution estimates of earth model parameters using all information present in seismic. Elastic FWI inverts multicomponent data for P and S-wave velocities and densities. We propose an alternative approach for FWI using a combination of machine learning and the physics of wave propagation. Unlike a conventional supervised machine learning, we do not require known answers to train our network. The multicomponent shot gathers are input to a convolutional neural network (CNNs) based auto encoder whose outputs are used as P-wave, S- wave and density models that are used to compute synthetic seismograms using the stress-velocity formulation of the elastic wave equation. The synthetic data are compared against observed input data and the misfit is estimated. The gradient of the misfit with respect to the velocity model parameters is calculated using the adjoint state method. The adjoint state gradient is then used to update the network weights using the automatic differentiation technique. Once the misfit term converges, the neural network can generate subsurface models consistent with the observed data. We observe that the neural network can capture spatial correlations at different scales and thus can introduce regularization in our inverse problem. The regularization is enough to mitigate the cross-talk problem in elastic FWI and also produce good results in areas with low illumination.Keywords: full-waveform inversion, elastic full-waveform inversion, deep learning, machine learning, seismic inversionPermalink: https://doi.org/10.1190/image2022-3745050.1FiguresReferencesRelatedDetails Second International Meeting for Applied Geoscience & EnergyISSN (print):1052-3812 ISSN (online):1949-4645Copyright: 2022 Pages: 3694 publication data© 2022 Published in electronic format with permission by the Society of Exploration Geophysicists and the American Association of Petroleum GeologistsPublisher:Society of Exploration Geophysicists HistoryPublished Online: 15 Aug 2022 CITATION INFORMATION Arnab Dhara and Mrinal Sen, (2022), "Elastic-AdjointNet: A physics-guided deep autoencoder to overcome crosstalk effects in multiparameter full-waveform inversion," SEG Technical Program Expanded Abstracts : 882-886. https://doi.org/10.1190/image2022-3745050.1 Plain-Language Summary Keywordsfull-waveform inversionelastic full-waveform inversiondeep learningmachine learningseismic inversionPDF DownloadLoading ...