Advancing Reservoir Characterization: A Comparative Analysis of Xgboost and Ann for Accurate Porosity Prediction

Abstract This study presents a comprehensive analysis of porosity prediction in reservoir characterization using two powerful machine learning algorithms, XGBoost and Artificial Neural Networks (ANN). Well log data, including GR log, resistivity log, density log, sonic log, neutron log, and por_insitu, were employed for training and validation, with the latter serving as core data for porosity at identical depths. The initial XGBoost prediction yielded an accuracy of 57.54%, prompting the implementation of performance metrics and hyperparameter tuning. Through these optimizations, the XGBoost model's accuracy skyrocketed to an impressive 96.75%, with refined performance metrics, including MAE of 0.766158, MSE of 1.047763, and RMSLE of 0.038324. Leveraging the well log data and core data validation, the XGBoost model demonstrated outstanding accuracy levels, holding immense promise for reservoir characterization and hydrocarbon exploration. On the other hand, ANN model underwent data standardization and architecture experimentation, ultimately finding the "tanh" model with ten hidden layers to be the best performer. This model achieved a mean absolute error (MAE) of 1.87 for the train data and 4.08 for the test data. Comparing the two models, the XGBoost model clearly outperformed the ANN model in accuracy for porosity prediction, making it the preferred choice for reservoir applications. The project's success highlights the efficacy of XGBoost in handling complex geological data and its ability to deliver accurate predictions for porosity. Additionally, the ANN model's performance demonstrates the importance of optimizing network architecture and activation functions in achieving accurate results. The findings of this study underscore the significance of machine learning in reservoir characterization, offering valuable insights for decision-making processes in the petroleum industry. The report provides valuable insights for reservoir engineers, geoscientists, and data scientists, offering a foundation for future advancements in reservoir characterization and resource exploration.