Novel empirical correlation for estimation of the total organic carbon in devonian shale from the spectral gamma-ray and based on the artificial neural networks

The evaluation of the quality of unconventional hydrocarbon resources becomes a critical stage toward characterizing these resources, and this evaluation requires the evaluation of the total organic carbon (TOC). Generally, TOC is determined from laboratory experiments; however, it is hard to obtain a continuous profile for the TOC along the drilled formations using these experiments. Another way to evaluate the TOC is through the use of empirical correlation, and the currently available correlations lack the accuracy especially when used in formations other than the ones used to develop these correlations. This study introduces an empirical equation for the evaluation of the TOC in Devonian Duvernay shale from only gamma-ray and spectral gamma-ray logs of uranium, thorium, and potassium as well as a newly developed term that accounts for the TOC from the linear regression analysis. This new correlation was developed based on the artificial neural networks (ANNs) algorithm which was learned on 750 datasets from Well-A. The developed correlation was tested and validated on 226 and 73 datasets from Well-B and Well-C, respectively. The results of this study indicated that for the training data, the TOC was predicted by the ANN with an AAPE of only 8.5%. Using the developed equation, the TOC was predicted with an AAPE of only 11.5% for the testing data. For the validation data, the developed equation overperformed the previous models in estimating the TOC with an AAPE of only 11.9%.