Total organic carbon (TOC) is an essential parameter used for unconventional shale resources evaluation. The current methods for TOC determination are based on conducting time-consuming laboratory experiments or by using empirical correlations; the later are developed based on assumptions that restrict their applicability. This study provides a new robust model for TOC estimation based on conventional well logs. The model was developed using an adaptive neuro-fuzzy inference system with subtractive clustering (ANFIS-SC).

Four conventional well logs of deep resistivity, sonic transit time, gamma ray, and formation bulk density collected from Barnett shale formation were used to develop the ANFIS-SC model for TOC estimation. A dataset consists of 645 records of the four well logs and TOC were used to develop the new model. The model was optimized for the different combinations of the ANFIS-SC’ design parameters and for training/testing data ratio. The optimum predictability of the ANFIS-SC TOC model was reached after 400 iterations using a cluster radius of 0.3 and a training/testing data ratio of 70/30.

The statistical analysis showed that TOC is a strong function of the bulk density, moderate function of the sonic transit time and gamma ray, and a weak function of the deep resistivity. The training and testing results proved that the developed ANFIS-SC model is able to predict the TOC based on the four mentioned well log data with high accuracy. For the training data set, the TOC was estimated with an average absolute percentage error (AAPE) of 8.62%, a coefficient of determination (R2) of 0.91, and a correlation coefficient (R) of 0.96. For the testing data set, the associated AAPE in predicting the TOC is 9.57%, while R2 and R between the actual and predicted TOC are 0.89 and 0.94, respectively.

You can access this article if you purchase or spend a download.