Full text loading...
-
Multichannel Wavefield Reconstruction with Slope Assisted Physics Informed Neural Networks
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 86th EAGE Annual Conference & Exhibition, Jun 2025, Volume 2025, p.1 - 5
Abstract
Summary
We introduce a novel approach to interpolate multi-component seismic measurements using Physics Informed Neural Networks. This network is trained to predict both the pressure and horizontal particle acceleration measurements whilst being guided by the local plane wave partial differential equation. The local slope parameter of the plane wave differential equation is simultaneously estimated during the interpolation process using a smaller auxiliary network. The results of the new PINN multicomponent reconstruction algorithm, named MC PINNslope are compared on a synthetic receiver gather against a previous version of PINN-based interpolation that leverages only the pressure data, showing that the addition of measured gradient information in our new implementation allowed us to interpolate sparse acquisition scenarios that even the original PINN-based interpolation could not handle. In the second example, we benchmark the MC PINNslope against a state-of-the-art slope regularized sparsity promoting inversion. The two methods are compared on field data with progressively coarser subsampling, demonstrating that MC PINNslope can outperform the conventional method at sparser recordings. Finally, we showcase the grid-based training feature of MC PINNslope (and PINN-based interpolators in general) interpolating the field seismic data at a denser trace spacing of the original fully sampled field data without the need of retraining.
© EAGE Publications BV