Structures Inversion and Optimization in Cased-Wells Based on Deep Learning

Abstract

Acoustic logging is a vital branch of geophysical logging and is a geophysical logging method for downhole measurement of rock acoustic properties of formation profiles and evaluation of wellbore formation properties. In this paper, we propose a novel approach based on machine learning to tackle the mapping challenge from time-series data to spatial images in the field of geophysical logging, that is, using a fully connected neural network (FCNN) to reconstruct the slowness model from wellbore data. Specifically, forward modeling is to study borehole acoustic signals using finite-difference time-domain method, and generate training and test data sets. The relevant research results indicate that the inversion in borehole imaging based on FCNN approach has a good effect in terms of structure detection and interlayer information presentation, and can also recover detailed slowness information between different layers of the wellbore. And the inversion results are more consistent with the target in terms of slowness values, downhole structures, as well as geological interfaces. Besides, we also optimize the image quality by using bilateral filtering method.