Prediction of Water Saturation from Well Log Data by Machine Learning Algorithms: Boosting and Super Learner

Intelligent predictive methods have the power to reliably estimate water saturation (Sw) compared conventional experimental commonly performed by petrphysicists. However, due nonlinearity and uncertainty in data set, prediction might not be accurate. There exist new machine learning (ML) algorithms such as gradient boosting techniques that shown significant success other disciplines yet been examined for Sw or reservoir rock properties petroleum industry. To bridge literature gap, this study, first time, a total of five ML code programs belong family Super Learner along with algorithms: XGBoost, LightGBM, CatBoost, AdaBoost, are developed predict without relying on resistivity log data. This is important since rely can become problematic particular formations shale tight carbonates. Thus, do so, two datasets were constructed collecting several types well logs (Gamma, density, neutron, sonic, PEF, PEF) evaluate robustness accuracy models comparing results laboratory-measured It was found XGBoost produced highest accurate output (R2: 0.999 0.993, respectively), considerable distance, Catboost LightGBM ranked third fourth, respectively. Ultimately, both negligible errors but latest considered best amongst all.