Prediction of Formation Properties Based on Drilling Data of Wells at Utah FORGE Site Using Machine Learning

Abstract

ABSTRACT: This study shows how predictions of formation properties can be obtained from bit-rock interactions without additional operational costs using machine learning models. The models are based on drilling data sets acquired from a highly deviated well at the Utah FORGE (Frontier Observatory for Research in Geothermal Energy) site in south-central Utah, U.S.A. The drilling data is preprocessed by removing outliers that are out of the meaningful range. The data shows that pump pressure and bit RPM are highly correlated with rate of penetration (ROP). Random forest model yields the best performance for ROP prediction with top drive torque as the most important feature. Pump pressure and downhole torque are clustering in different groups for different sections of the inclined well. Successful prediction of confined compressive strength (CCS) is also achieved through random forest model. The model was trained with CCS obtained from sonic logs and calibrated by core test data from a nearby offset well. The well-trained machine learning model can be used to predict CCS of other nearby geothermal wells with drilling data only. 1. INTRODUCTION Understanding formation properties is important in drilling, completion, and production for geothermal reservoir development. Wireline logs, such as sonic logs, can be used to assess formation properties in oil/gas and geothermal reservoirs. However, logging after drilling can be expensive and is rarely conducted in geothermal wells. Alternatively, predictions of formation properties can be acquired from bit-rock interactions without additional operational costs using machine learning models (Jamshidi et al., 2013; Mahmoud et al., 2021). Drilling and logging contribute a significant portion of cost for Enhanced Geothermal System (EGS). Reliable estimation of formation properties without after-drilling wireline logging would reduce the cost EGS and enhance understanding of geothermal reservoirs. The drilling data sets investigated were acquired from the Utah FORGE (Frontier Observatory for Research in Geothermal Energy) site in south-central Utah, U.S.A. Utah FORGE is a geothermal field laboratory intended to develop and test tools and technologies required for creating, sustaining, and monitoring EGS reservoirs. Since 2017, five vertical monitoring wells, the deepest to 9500 ft, and a highly deviated injection well, 16A(78)-32, have been drilled. In this study, we focus on the drilling data from the injection well, 16A(78)-32, which is highly deviated at a tangent of 65° to the vertical.