IEEE Transactions on Geoscience and Remote Sensing | Informer-Based High-Order Geometrical Factor for Modeling Array Induction Logging in Layered Formations

By  李勇    2026-11-13    Visited 10 times

Abstract

This article presents an informer-based framework for efficient simulation of array induction logging (AIL) responses in layered isotropic and anisotropic formations. To overcome the limitations of computationally expensive pseudo-analytical methods and data-driven approaches with poor generalization, we propose a residual response learning scheme based on the geometrical factor (GF) theory. The method decomposes the total response into a low-order GF approximation and a high-order residual response learned through a physics-guided informer-based framework. The informer’s ProbSparse self-attention mechanism reduces time and memory complexity from O(L2) to O(LlogL), enabling efficient handling of long depth-series AIL data. Its multiscale feature fusion and generative decoder further enhance the model’s ability to capture interlayer dependencies and avoid cumulative prediction errors. In addition, a custom loss function incorporating electromagnetic (EM) continuity and geological background constraints ensures physically consistent predictions. For layered anisotropic formations, the framework employs a two-level scheme involving an inversion-based equivalent isotropic resistivity transformation followed by learning of anisotropic residual response. Numerical experiments demonstrate that the method achieves 96%–98% prediction accuracy while providing a computational speedup of nearly two to three orders of magnitude compared to conventional pseudo-analytical methods. The new method maintains robust performance across varying dipping angles and anisotropy coefficients, offering an accurate and efficient solution for real-time AIL response simulation.


Paper Information:

Ping Qiao, Lei Wang*, Shaogui Deng, Zhen Chen, Honglian Chen. Informer-Based High-Order Geometrical Factor for Modeling Array Induction Logging in Layered Formations[J]. IEEE Transactions on Geoscience and Remote Sensing, 2026, 64: 2001413. DOI: 10.1109/TGRS.2026.3674745.