Applying Machine Learning in Highly Laminated Formation to Differentiate Pay and Non-Pay Zones and Resolve Rt for Fit for Purpose Azimuthal Resistivity Tool Selection

Abstract In this work, firstly, we present a systematic workflow combining linear regression, polynomial regression to generate Rt profiles based on the Vshale model and CART - Classification And Regression Tree method to differentiate between pay and non-pay zones. This machine learning method is a decision tree, a classifier that does not require any knowledge or parameter setting. This approach is known as supervised learning, thus given training data; the ML algorithm induces a decision tree. From a decision tree, it can create rules about the data to predict the classification of unseen records differentiating between pay and non-pay zones from eight independent variables such as MD, TVDSS, ZONE, VSH, RT, PHIT, SW and K. Secondly, based on five planned horizontal wells, individual resistivity distribution maps generated for each well to estimate the maximum depth of detection for deep azimuthal propagation and extra-deep azimuthal propagation tools based on the highest resistivity contrast between pay and non-pay zones. Pseudo vertical offset wells created intersecting each horizontal well to extract the Rt profile for comparison among the existing offset wells for quality control purposes. Finally, the operator selected the most appropriate azimuthal resistivity tool for each well based on the pseudo resistivity map and validated post-well with the machine learning CART method. We also show the Rt comparison between the hybrid model and actual resistivity measurement from the first well drilled with extra-deep azimuthal resistivity service in the landing section.