Total organic carbon (TOC) quantification using artificial neural networks: Improved prediction by leveraging XRF data

Abstract

This study develops a new artificial neural network (ANN) model for predicting the total organic carbon (TOC) of an organic-rich carbonate mudstone formation using conventional well log data and X-ray fluorescence spectroscopy (XRF) analysis. The data used in the study include conventional well logs, redox-sensitive elements from XRF, and TOC values measured in lab for a total of 150 core samples obtained from five wells. Selected well logs including gamma ray (GR), bulk density (RHOB), uranium (URAN), and XRF-derived elements, including molybdenum (Mo), copper (Cu), and nickel (Ni), were used to train and develop the ANN model to predict and generate continuous high-resolution TOC log profiles for the five wells. TOC data were classified into two groups based on geological descriptions and well locations. Statistical analyses were performed to establish the range of data used for each group and to evaluate relationships among the TOC and input parameters. The developed ANN model showed a high performance in providing a continuous profile of TOC. The difference between absolute average is less than 0.50 and the correlation coefficient (R-value) is greater than 0.70. Empirical correlations were extracted from the best performing ANN model, which will allow easy and quick estimation for TOC values. The developed correlations outperform available methods for determining TOC and reduce the estimation error by 42 %.