Reservoir Properties Prediction Using Random Forest Regression

Abstract

This research is about the prediction of reservoir properties, namely Porosity, Water Saturation, and Horizontal Log Permeability using Machine Learning regression models, with the purpose to measure the accuracy of reservoir properties predictions obtained from well log data using regression model and analyze the benefits of using regression model to determine reservoir properties. The method used is by importing the well log dataset for data preprocessing (data cleaning, normalization, and outlier removal) and after that, the dataset is split into train and test dataset. Three regression models were used in this research, namely Random Forest, Extreme Gradient Boosting, and Light Gradient Boosting Machine. The regression models are trained with the train dataset, after that the trained regression models predicted reservoir properties from the test dataset. Performance of each model was tested, and after comparison of each model’s accuracy, Random Forest model shown the best result. The results showed that the model was able to find correlations between well log data and its reservoir properties, and was able to predict the reservoir properties accurately; indicated by error evaluation and plots of comparison of actual and predicted data values which are nearly aligned overall.