Estimation of Geomechanical Parameters, In Situ Stress Measurement Techniques, and Determination of Safe Mud Weight Windows Using Machine Learning Algorithm Methods

Oil wells & boreholes logs data are interpreted/processed to identify petrophysical, mechanical, and in-situ geomechanical properties for rocks around oil wells, due to high cost and geological problems, some well logs cannot be measured. For example, sonic logs contain geophysical, and geomechanical information critical to determining the modulus of dynamic elasticity, Young's modulus, the bulk modulus, acoustic resistance/impedance, the shear modulus, and the Poisson's ratio of rocks around the well’s wall. Therefore, in this paper, two random wells were selected from one of the oil fields in southern Iran, one of which was selected as training well to determine the appropriate model and the other to predict the shear and compressional wave slowness. Data analyses were performed using a range of machine learning methods and setting hyperparameter tuning on algorithms, the best models were selected for predicting/estimating sonic logs. In this process, among the regression methods, the K-nearest neighbors algorithm (KNN), and among the combined methods, the Random Forest Regression algorithm and the Extra Tree Regression algorithm show the highest correlation coefficient. As a result, the extra tree algorithm for modeling was performed on the training sets and testing sets data of the well. Then this model was used to predict and synthesize the slowness acoustic compressional and the slowness acoustic shear of the target well. Then, by comparing the actual data of the target well, the root mean square error and the R-squared were obtained. Then, using poroelastic equations, the field stresses were determined and found that Sarvak and Ilam reservoirs are in reverse stress regime and Asmari reservoir is in normal stress regime up to Strike-slip. At the end of this article, using the rock mechanics criteria, the best optimal safe mud weight windows in the studied was presented.