Abstract
Total organic carbon (TOC) is the amount of carbon present in the formation, which is an important criterion for assessing unconventional shale resources. Current TOC determination methods rely on time-consuming laboratory tests or empirical correlations created on the basis of assumptions that limit their application. Using artificial neural networks (ANN), a new robust model for TOC estimate based on traditional well logs was constructed in this study.
In this study, 891 TOC data points at different depths and their corresponding well logs of deep resistivity, gamma ray, sonic transit time, and bulk density and spectral logs collected to train the model, and then tested on 291 different data points. The ANN model was optimized for the different design parameters using inserted for loops in Matlab. The optimized model was then validated in another unseen 82 data points.
The average absolute percentage error (AAPE) and correlation coefficient (R) between the measured and the ANN-based TOC were used to evaluate the models. With R values higher than 0.93 and AAPE values less than 14%, the new model produced an outstanding agreement with the real TOC values. The model surpassed the available empirical correlations in the validation dataset, resulting in lower than 10% AAPE, compared to more than 20% AAPE in other models. As a result of these findings, all of the correlation's parameters are reported, allowing it to be used to a variety of datasets. The novelty of this new research is the simplicity and high accuracy of the developed model on estimating the TOC based on available well log data.
