Young’s Modulus Estimation Using Machine Learning Methods and Daily Drilling Reports

To avoid drilling damages, it is very important to determine the field stress. Prediction of elastic parameters such as Poisson's ratio and Young's modulus is of great importance in determining in-situ stress and completing geomechanical modeling. These parameters are calculated statically through laboratory tests on drilling cores or dynamically through log data. However, such data may not be available in the oil field data-bank. Therefore, Daily Drilling Reports (DDR) can be introduced as a suitable alternative for predicting rock’s elastic modulus. In this study, for the first time, an attempt has been made to estimate the Dynamic Young’s modulus using DDR data with the application of a variety of conventional machine learning methods. In this regard, linear, support vector machine (SVM), artificial neural network (ANN), Random Forest (RF) LSBoost, and Baysian have been used. Input data to these algorithms also include depth, string rotary speed (RPM), rate of penetration (ROP), weight on bit (WOB), density (RHOB), porosity (Φ), pump pressure (PP), and tangential velocity (TV). Each of these algorithms was then compared in terms of accuracy using correlation coefficient (R2), mean squared error (MSE), and root mean square error (RMSE) criteria. Finally, using conventional experimental correlations and using core data, the resulting values were converted to static values. The results show that using daily drilling reports, based on the above criteria, a good estimate of the elastic parameters can be achieved. Also, among the methods used, Baysian and LSBoost methods have slightly higher and better accuracy than other methods.