Improved total organic carbon convolutional neural network model based on mineralogy and geophysical well log data

Abstract

Abstract Unconventional resources, such as shale oil and gas, are currently regarded as an essential resource in the face of depleting conventional hydrocarbon reserves. In line with this, the accurate determination of the hydrocarbon potential of a shale reservoir is critical and relies in part, on the total organic carbon (TOC) content. However, there exist challenges when determining TOC by conducting geochemical analysis of rock samples or developing a relationship between well log response and TOC values. Variations in the mineral composition and TOC values can influence the performance of well log-based models. There are few applications integrating mineral composition and well log response in an artificial intelligence (AI) model for TOC estimation. This study incorporates the mineral composition of the shale and well log data in developing a deep convolutional neural network (CNN) model for predicting TOC. The statistical results of the study have shown that the proposed mineralogy and well log-based CNN (MWL-CNN) outperformed well log-based CNN (WL-CNN). The sensitivity analysis performed indicate that the mineral constituents that significantly influenced the model outcome were feldspar and pyrite. These findings have established that mineral composition has a great effect on TOC predictions and must be incorporated in the model for a more accurate result.