Perpustakaan Digital ITB

This study focuses on predicting water saturation (Sw) value with the application of Machine Learning (ML) models and comparing it to an existing well log data. This alternative technique for predicting water saturation using machine learning models can be used in wells because it can predict Sw based on well log data and saturation equation such as Archie’s, predict Sw in a well where core analysis are not available, and also predict initial water saturation in a non-initial well. The machine learning models that were used in this study are Least Squares Support Vector Machine - Particle Swarm Optimization (LSSVM-PSO) model, Artificial Neuron Network (ANN) model, Random Forest model, AdaBoost model, and XGBoost model. The five models will predict the water saturation of a new unseen well by choosing the best parameter from the model that has been trained. This study will determine the best features for the model from the well log data. This study will also determine which model is the best in predicting water saturation in the new unseen well. The result of this study shows only the ANN model is able to predict the water saturation value of the new unseen well accurately, with the ????2, RMSE, MSE, and MAE value of 0.784, 0.152, 0.023, and 0.103 respectively. In order of the level accuracy, the best model is ANN, Random Forest, AdaBoost, XGBoost, and the last one is LSSVM-PSO.