Home

Quick Links

Search

 
Deep Convolutional Neural Networks for 1D Inversion of Electromagnetic DataNormal access

Authors: V. Puzyrev, S. Meka and A. Swidinsky
Event name: 81st EAGE Conference and Exhibition 2019
Session: Electromagnetic Methods II
Publication date: 03 June 2019
DOI: 10.3997/2214-4609.201901485
Organisations: EAGE
Language: English
Info: Extended abstract, PDF ( 1.08Mb )
Price: € 20

Summary:
We investigate the potential of deep learning based approaches to 1D EM inversion. Our approach to inversion based on deep neural networks is entirely data-driven, does not employ traditional gradient-based techniques and provides a guess to the model instantaneously in a single step. The results of this study show that a deep neural network can accurately reconstruct resistivity distribution in the subsurface from data measured at the seafloor. Shallow resistive and conductive structures do not significantly impede the detection of deeper targets. Once trained, the network can predict resistivity models from new data in a few milliseconds. For multidimensional problems, data-driven inversion based on deep learning is expected to deliver sufficiently accurate results orders of magnitude faster than conventional inversion methods.


Back to the article list