Electromagnetic reservoir monitoring with machine-learning inversion and fluid flow simulators

Abstract

PreviousNext No AccessFifth International Conference on Engineering Geophysics, Al Ain, UAE, 21–24 October 2019Electromagnetic reservoir monitoring with machine-learning inversion and fluid flow simulatorsAuthors: Daniele Colombo*Weichang LiErnesto Sandoval-CurielGary W. McNeiceDaniele Colombo*Geophysics Technology, EXPEC Advanced Research Center, Saudi AramcoSearch for more papers by this author, Weichang LiAramco Research Center – Houston, Aramco Services Company, USASearch for more papers by this author, Ernesto Sandoval-CurielGeophysics Technology, EXPEC Advanced Research Center, Saudi AramcoSearch for more papers by this author, and Gary W. McNeiceGeophysics Technology, EXPEC Advanced Research Center, Saudi AramcoSearch for more papers by this authorhttps://doi.org/10.1190/iceg2019-043.1 SectionsAboutPDF/ePub ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinked InRedditEmail Abstract Reservoir monitoring would greatly benefit from the application of geophysical methods to infill the saturation predictions in the space in between the wells. Electromagnetic methods such as surface-to-borehole CSEM or cross-well EM provide large sensitivity to fluid saturation variations and, as such, are among the best candidates for implementing reservoir surveillance. Mapping EM field changes into reservoir models though inversion is an ill-posed and non-unique problem where robust and reliable results can be obtained only through strong regularization. We present a scheme where a fluid simulator is utilized to generate a large variety of saturation models corresponding to a water-alternating-gas (WAG) experiment for enhanced oil recovery (EOR). EM fields are generated assuming a cross-well acquisition geometry between the injector and the observation wells. Fields and models are used for training a deep learning neural network which is capable of generating corresponding model predictions from additional simulated EM data. Results show excellent reconstruction capability with the machine learning inversion approach that surpasses the resolution obtained by a conventional geophysical inversion method. ML inversion can become a viable tool to enable data-driven and physics-constrained reservoir monitoring. Keywords: frequency-domain, machine learning, controlled-source electromagnetic, neural network, modelingPermalink: https://doi.org/10.1190/iceg2019-043.1FiguresReferencesRelatedDetails Fifth International Conference on Engineering Geophysics, Al Ain, UAE, 21–24 October 2019ISSN (online):2159-6832Copyright: 2020 Pages: 315 publication data© 2020 Published in electronic format with permission by the Society of Exploration GeophysicistsPublisher:Society of Exploration Geophysicists HistoryPublished: 03 Apr 2020 CITATION INFORMATION Daniele Colombo*, Weichang Li, Ernesto Sandoval-Curiel, and Gary W. McNeice, (2020), "Electromagnetic reservoir monitoring with machine-learning inversion and fluid flow simulators," SEG Global Meeting Abstracts : 167-170. https://doi.org/10.1190/iceg2019-043.1 Plain-Language Summary Keywordsfrequency-domainmachine learningcontrolled-source electromagneticneural networkmodelingPDF DownloadLoading ...