Estimation of Total Organic Carbon and Brittleness from Shale Rock Physics Properties Using Machine Learning Methods

Estimating the shale-specific rock properties, such as total organic content and brittleness, is important to characterize unconventional reservoir rocks, and to understand the depositional environments of shales. Since we are still far from fully understanding the physical properties of organic-rich shale rocks, it is difficult to predict total organic carbon (TOC) and mineralogical brittleness index (BI) directly from well log data. Historically, these properties have been estimated using the Schmoker density-log and the elastic brittleness methods, respectively. However, both methods are empirical and require additional local calibration due to various depositional environments of shale formations. Therefore, we wish to develop an easy and efficient method to compensate for the weakness of the traditional methods. Machine Learning (ML) can help to solve both linear and nonlinear estimation problems with large datasets. We apply supervised ML techniques to improve TOC and BI estimations from shale well logs. First, we define the influential input rock and seismic properties, which are highly correlated with the TOC and BI values. From bivariate correlation analysis, we distinguish six influential factors: porosity, density, P-wave velocity, S-wave velocity, Poisson's ratio, and Young's modulus. Second, we conduct multi-variate linear regression analysis to derive simpler and more predictive statistical models than those from traditional methods, showing higher R² values and lower RMSE values. Third, we use supervised ML algorithms with these variables, including the ensemble of bagged trees method for TOC and support vector machines (SVM) for BI, respectively. These ML methods help to find supervised nonlinear regression models with a better fit to the data than the simple linear regression models. In conclusion, our ML approach can provide a more accurate and reliable method to estimate total organic content and brittleness from the rock physics properties for improved shale characterization.