Petrophysical methods are widely used to study physical and chemical properties of rocks and fluids. Due to the complexity of the lithological makeup of different rock types, quantitative analysis of mineral composition is a challenging task. In this study, we describe the use of a hybrid machine learning (ML) model to estimate mineral compositions by combining a convolutional neural network (CNN) and XGBoost algorithm. The selected inputs are preprocessed into a square matrix of format and passed to the convolutional layers of the feature learning section, and the XGBoost is utilized to solve the regression problem. The conventional and geochemical logs from the Horn River Basin, in western Canada, are selected for the model training and validation. Monte Carlo dropout (MCD) is added to model training to decrease the sensitivity and increase repeatability of model prediction. The comparison of metrics and correlation coefficients shows that the hybrid ML is slightly better than U-net CNN and XGBoost individually, which demonstrates the reliability and effectiveness of the proposed algorithm. The presented method could be applied to other exploration settings such as geothermal, unconformity uranium or sediment-hosted mineral deposits.

This content is only available via PDF.
You can access this article if you purchase or spend a download.