Machine Learning-Based Rock Facies Prediction Using Geothermal Data: A Comparative Analysis of Algorithms

Abstract

An approach to predicting rock facies has been generated. Machine learning techniques have generated a novel approach to predicting rock facies. The objective was to develop a reliable facies predictor capable of categorizing rocks into 5 distinct facies: Sand, sandy shale, shady sand, volcanic sand, and shale. Electric logs, including gamma-ray, resistivity, and density, were utilized to generate a synthetic facies log that could accurately represent the true geological facies. The study evaluated the performance of four machine learning algorithms: k-nearest Neighbors (KNNs), Random Forest (RF), Decision Tree, and Stochastic gradient descent (SGD). These algorithms were employed to build predictive models using the Utah FORGE geothermal project data as input. The goal was to determine which algorithm performed better in accurately predicting rock facies based on the electric logs. The study results indicate that the developed facies prediction model successfully generated a synthetic facies log that closely matched human-defined facies logs. Moreover, the comparative analysis revealed insights into the strengths and weaknesses of the various machine learning algorithms in the context of rock facies prediction. This research has significant implications for geology and geothermal energy exploration, as it offers a data-driven approach to enhance our understanding of subsurface rock formations. By leveraging machine learning, we can improve the accuracy and efficiency of facies prediction, ultimately aiding geothermal resource assessment and development.