Lithology prediction from well log data using machine learning techniques: A case study from Talcher coalfield, Eastern India

Abstract

Coal exploration in the Indian scenario is challenging due to plenty of carbon contents and dirt bands within a coal seam. Manual interpretation of geophysical logging data in such conditions is time-consuming and tedious due to the non-linear behaviors of well log signals. This paper intends to apply supervised machine learning techniques such as Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting technique (XGBoost) for meticulously interpreting such banded coal seams from geophysical logs. To investigate the efficacy of the above-mentioned five ML techniques in the Indian scenario, we have considered Gamma-ray, Density, and Resistivity logs from four boreholes drilled in Talcher coalfield, Eastern India. ML model training results indicate that all the prediction models have scored more than 88% accuracy scores in classifying carbonaceous and non-coal lithofacies. Receiver Operating Characteristics (ROC) curves of different ML models suggest that the obtained area under the curve (AUC) is positive and above the main diagonal for all lithotypes. Finalized ML models with appropriate hyper-parameters are also applied on the nearby drilled wells, and the outputs are validated against geological core data. It is found that all ML models have acquired an overall accuracy score greater than 80% in all three test wells, conveying that ML techniques are a potential solution for dealing with banded coal seam problems in the Indian scenario.