Predicting Bubble Point Pressure and Gas-Oil Ratio PVT Properties Using Bagging Ensemble Techniques

Abstract

An enhanced accurate predictive model has been developed for the estimation of reservoir oil Pressure-Volume-Temperature (PVT) properties of Bubble Point Pressure (BPP) and Gas-Oil Ratio (GOR) using Bagging Ensemble machine learning. To develop the Bagging ensemble (BE) model, three different estimators, Decision Tree Regressor, Random Forest Regressor, and Extra Trees Regressor, were used as the base estimators. An averaging method to finally predict the model performance was done using a voting regressor to fit the base estimators. Hyper-parameter tunings for optimization were determined using cross-validation grid search and the implementation of Bagging ensemble described. The ensemble methods were compared with those developed using Artificial Neural Network (ANN) and some selected empirical correlations.  Their performances were evaluated using Average Absolute Percentage Relative Error (AAPRE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and graphical cross-plot error analysis Correlation Coefficient (R2). Result showed improved performances, notably the R2 for BPP, the BE is 96.6%, ANN is 89.04%, and the best empirical model is 88.1%. For the GOR, the R2 for BE and ANN are 94.1% and 89.0%, respectively while best empirical model is 88.1%.