Intelligent modeling for precise water saturation prediction in carbonate gas reservoir using Xgboost

In hydrocarbon reservoir studies, the precise determination of fluid saturations within the rock formation poses numerous challenges. Particularly in carbonate reservoirs, conventional methods for calculating water saturation based on well logs using Archie's relationship and its derived equations often exhibit inaccuracies. This research aims to present an intelligent model for more accurate determination of water saturation in a gas-bearing carbonate reservoir located in southern Iran, where the shortcomings of conventional calculation methods have been demonstrated through prior studies. To achieve this goal, well log data from three wells in this carbonate reservoir were utilized. Through the thoughtful design of a XGBoost machine learning algorithm which is tuned by grid search method; and K-fold, experimental reservoir water saturation was computed. By dividing the data from these three wells into training and testing sets, the model was constructed and its performance evaluated, revealing significantly higher accuracy in predicting reservoir water saturation compared to conventional methods such as Archie's relationship. The model exhibited training accuracy metrics with MAE=0.031, MSE=0.001, 𝑅2 =0.989, and testing accuracy metrics with MAE=0.032, MSE=0.0025, 𝑅2 =0.958, indicating the efficacy of the intelligently designed model

Intelligent modeling for precise water saturation prediction in carbonate gas reservoir using Xgboost Keywords:

Water Saturation , Gas Reservoir , Well Logging , Xgboost , Machine Learning.