Error control and loss functions for the deep learning inversion of borehole resistivity measurements

Abstract

Deep learning (DL) is a numerical method that approximates functions. Recently, its use has become attractive for the simulation and inversion of multiple problems in computational mechanics, including the inversion of borehole logging measurements for oil and gas applications. In this context, DL methods exhibit two key attractive features: (a) once trained, they enable to solve an inverse problem in a fraction of a second, which is convenient for borehole geosteering operations as well as in other real-time inversion applications. (b) DL methods exhibit a superior capability for approximating highly complex functions across different areas of knowledge. Nevertheless, as it occurs with most numerical methods, DL also relies on expert design decisions that are problem specific to achieve reliable and robust results. Herein, we investigate two key aspects of deep neural networks (DNNs) when applied to the inversion of borehole resistivity measurements: error control and adequate selection of the loss function. As we illustrate via theoretical considerations and extensive numerical experiments, these interrelated aspects are critical to recover accurate inversion results. K E Y W O R D S deep learning, deep neural networks, error estimation, geophysical applications, real-time inversion

Performance of Refined Isogeometric Analysis in Solving Quadratic Eigenvalue Problems

“…We investigate the electromagnetic wave propagation eigenproblem. In particular, we consider a three-layer heterogeneous Earth model as a common case study in geosteering applications (see, e.g., [7,68]). This 2D Earth model is essential in studying the full 3D wave propagation eigenproblem in transversely isotropic media [69].…”

A Finite Element based Deep Learning solver for parametric PDEs

“…These traditional methods evaluate the inverse solution pointwise (i.e., for a given set of measurements), but they rarely provide a global representation of the inverse operator. To overcome this problem and approximate the full inverse function, it is possible to use Deep Learning (DL) methods (see, e.g., [9,10,11,12,13,14]), which allow to approximate complex mappings via a composition of linear and non-linear functions.…”

Modeling extra-deep electromagnetic logs using a deep neural network

“…Such processes require the solution of possibly thousands of forward problems under strict time constraints (Pardo et al, 2015;Shahriari et al, 2018). However, deterministic inversion suffers from nonuniqueness of the solutions (Shahriari et al, 2020c).…”