Uncertainty Quantification on the Inversion of Geosteering Measurements using Deep Learning

Abstract

We propose the use of a Deep Learning (DL) algorithm for the real-time inversion of electromagnetic measurements acquired during geosteering operations. Moreover, we show that when the DL algorithm is equipped with a properly designed two-step loss function without regularization, it is possible to recover an uncertainty quantification map by analyzing certain cross-plots. We illustrate these ideas with a synthetic example based on piecewise 1D earth models. The resulting uncertainty quantification map could be used to design better measurement acquisition systems for geosteering operations.