Machine-Learning-Based Deconvolution Method Provides High-Resolution Fast Inversion of Induction Log Data

Abstract

We built a deconvolution model for induction log data using machine learning (ML). Unlike iterative forward modeling inversion methods, the deconvolution model is extremely fast. Unlike linear deconvolution models in the past, ML-based deconvolution finds accurate layer resistivity and layer boundaries. For a unit induction tool 2C40, the 21-point, 10-ft window deconvolution model works satisfactorily.