Efficient Model Design for Seismic Impedance Inversion

 Collaborators: Ahmad Mustafa, Motaz Alfarraj, and Ghassan AlRegib

Goal/Motivation: Develop efficient model architectures to accurately estimate well log profiles on seismic sections while reducing dependence on labeled data

Challenges: ML models used for seismic inversion are prone to overfitting and thus producing suboptimal estimations on account of their not being suited to handle sequence data..

High Level Description of the Work: Both well and seismic data can be treated as sequences where the order of the amplitudes imposes structure on the respective modalities. We propose recurrent models and other CNN-based architectures better able to account for spatial and temporal cues in the data to perform vastly superior estimations to models lacking either or both of those. 

Related Publications:

  1. A. Mustafa, M. Alfarraj, and G. AlRegib, "Joint Learning for Spatial Context-Based Seismic Inversion of Multiple Datasets for Improved Generalizability and Robustness," in Geophysics, vol. 86, no. 4, Mar. 26 2021. [PDF][Code]

  2. A. Mustafa, M. Alfarraj, and G. AlRegib, "Spatiotemporal Modeling of Seismic Images for Acoustic Impedance Estimation," in Expanded Abstracts of the SEG Annual Meeting, Houston, TX, Oct. 11-16 2020. [PDF][Code][Video]

  3. A. Mustafa and G. AlRegib, "A Comparative Study of Transfer Learning Methodologies and Causality for Seismic Inversion With Temporal Convolutional Networks," in International Meeting for Applied Geoscience & Energy, Denver, CO, 2021.[PDF][Code]