Data Driven Modelling to Predict Poisson's Ratio and Maximum Horizontal Stress

Abstract

During the design phase of oil and gas well drilling plans, predicting geomechanical parameters is an indispensable job. Accurate estimation of the Poisson's ratio and the maximum horizontal stress is essential where inaccurate estimation may result in wellbore instability and casing collapse increasing the drilling cost. Obtaining mechanical rock properties using mechanical tests on cores is expensive and time-consuming. Machine learning algorithms may be utilized to get a reliable estimate for Poisson's ratio and the maximum horizontal stress. This research aims to estimate the static Poisson's ratio and the maximum horizontal stress based on influencing factors from well-log input data through an Extreme gradient boosting algorithm (XGBoost). In addition, the XGBoost model was also compared with Random Forest. A real data set comprised of 22,325 data points was collected from the literature representing influencing variables which are compressional wave velocity, share wave velocity, bulk density, and pore pressure. The data set was split into 70% for training, and 30% for testing the model. XGBoost and random forest were used for training and testing the model. Mean absolute percentage error (MAPE), root mean squared error (RMSE), and coefficient of determination (R2) were assessed in the error metrics to obtain the optimum model. XGBoost and random forest were implemented using the k-fold cross-validation method integrated with grid search. The proposed XGBoost model shows an effective correlation between the geomechanical parameters (static Poisson's ratio and the maximum horizontal stress) with the input variables. The performance of the XGBoost model was found better than that of the random forest. The evaluation estimates more than 90% of R2 and approximately 4% of MAPE for the training and testing data. The key contribution of this work is the proposal of an intelligent model that estimates the geomechanical parameters without the need for destructive mechanical core testing. A reliable XGBoost model to predict the static Poisson's ratio and the maximum horizontal stress will allow improved wellbore stability analysis which significantly introduces efficiency gains.