Automatic Well Correlation by Aligning Multiple Well using Deep Neural Networks

Abstract

Abstract The interpretation of well logs is crucial for understanding subsurface geological formations and reservoir characterization. Among various roles of well logs in an exploration stage, constructing subsurface stratigraphic structure over target area is an important task to estimate oil and gas layer. Well logs correlation which connects multiple well logs with similar traits is commonly operated for stratigraphic surface picking. However, since a subsurface stratigraphic structure is very irregular and complex, correct well logs correlation is a time-consuming and high-cost task that is usually done by many geology experts. Recently, various studies suggest using supervised machine learning algorithms to automate well logs correlation. Most works firstly build formation matched pairwise dataset and train deep neural networks such as convolutional neural networks. However, such labeled data for supervision is cost expensive and most of them are not publicized. In this paper, we propose a simple and unsupervised method for automated well logs correlation with cutting-edge machine learning techniques. Unlike existing methods, we don`t utilize well logs paired data. Rather, we utilize well logs themselves, assuming nearby well logs would have similar characteristics at adjacent depth points. Most importantly, we introduce a time-series alignment concept to well logs correlation, which can deal with multiple well logs of different lengths. We show that our well logs alignment model is effective and can be applied to well logs in any oil and gas regions up to its simplicity.