Unsupervised Machine Learning for Sweet-Spot Identification Within an Unconventional Carbonate Mudstone

Stratigraphic correlation in mudstone intervals is challenging as compared to coarser-grained sedimentary rocks because of the microscale heterogeneity and other constraints. Given critical mm- to cm-scale variability in mudstones, it is daunting to try to infer compositional variability from well logs and seismic data unless core data and laboratory analyses are available to calibrate the results. In this study, we propose a novel integrated approach combining sedimentological core description with geochemical data to establish chemofacies and chemostratigraphic zonation using a set of unsupervised statistical tools, i.e., Principal Component Analysis (PCA) and Hierarchical Clustering on Principal Components (HCPC). These techniques can be applied to elemental data acquired using x-ray fluorescence measured from core or cuttings samples or spectroscopy logs to provide robust analysis for unconventional assessment regarding sweet-spot identification, sequence stratigraphic interpretations, and drilling and completion designs. Further, the identified zones can be used to characterize/correlate zones in nearby un-cored wells, with the data generated serving as an indispensable input for establishing a well-log data zonation using unsupervised machine learning algorithms.