Prediction of Total Organic Carbon in Organic-Rich Shale Rocks Using Thermal Neutron Parameters

Abstract

Total organic carbon (TOC) content is one of the crucial parameters that determine the value of the source rock. The TOC content gives important indications about the source rocks and hydrocarbon volume. Various techniques have been utilized for TOC quantification, either by geochemical analysis of source rocks in laboratories or using well logs to develop mathematical correlations and advanced machine learning models. Laboratory methods require intense sampling intervals to have an accurate understanding of the reservoir, and depending on the thickness of the interested formation, it can be time-consuming and costly. Empirical correlations based on well logs (e.g., density, sonic, gamma ray, and resistivity) showed fast predictions and very reasonable accuracies. However, other important parameters such as thermal neutron logs have not been studied yet as a potential input for providing reliable TOC predictions. Also, different studies estimate the TOC based on the well-logging data for various formations; however, limited studies were reported to predict the TOC for the Horn River Formation. Therefore, the objective of this study is to estimate the TOC variations based on the thermal neutron logs using one of the largest source rocks in Canada: The Horn River Formation. More than 150 data sets were collected and used in this work. The parameters of the artificial neural network (ANN) model were fine-tuned in order to improve the model’s prediction performance. Furthermore, an empirical correlation was developed utilizing the optimized ANN model to allow fast and direct application for the developed model. The developed correlation can predict the TOC with an average absolute error of 0.52 wt %. The proposed TOC model was able to outperform the previous models, and the coefficient of determination was increased from 0.28 to 0.73. Overall, the proposed TOC model can provide high accuracy for TOC ranges from 0.3 to 6.44 wt %. The developed model can provide a real-time quantification for the organic matter maturity, helping to allocate the zones of mature organic matter within the drilled formations.

Characterization and Feature Ranking of Well Log Variables Using Data-Driven Algorithms for Total Organic Carbon Estimation of Organic-Rich Shales

“…Nevertheless, to render this result quantitatively significant, it is imperative to have accurate data regarding the rock matrix and porosity. To overcome such limitations, a plethora of studies in recent years have progressively employed artificial intelligence techniques and machine-learning (ML) approaches to establish important relations between wireline log data and TOC for continuous TOC prediction with high accuracy. − Tan et al applied a support vector machine (SVM) for regression, using various kernel functions, to predict TOC values within gas-bearing shale from the Huangping syncline, China. This approach has encountered obstacles in enhancing the accuracy of the TOC content estimation.…”

“…Recently, a deep feedforward neural network has been applied to analyze seismic and well-log data to describe the spatial and vertical variations in the TOC content within Sembar shales . Hassan et al employed an artificial neural network to predict the TOC content using thermal neutron logs and delineating zones containing mature organic matter in the Horn River Formation. The potential for exploring innovative data-driven algorithms in TOC prediction, shale resource play evaluation, and feature characterization of influential variables, like bulk density, sonic slowness, gamma-ray, and others, for resource estimation is promising, because it has yet to be explored thus far.…”