Ali Gohari Nezhad - M.Sc holder of Petroleum Production Engineering, Faculty of Chemical Engineering;Engineering Faculty, University of Tehran

Precise estimation of effective hydrocarbon porosity is crucial in the field of hydrocarbon reservoir characterisation. This research explores the careful comparison of four well-known machine learning algorithms: Support Vector Machine (SVM), Random Forest, K-Nearest Neighbors (KNN), and Decision Tree. The objective of these studies is to model and predict effective hydrocarbon porosity in the intricate geological formation off Upper Dalan. To guarantee peak performance, each algorithm was carefully designed and then had its hyperparameters fine-tuned using a grid search method. The dataset employed in this research originates from conventional well logging reports extracted from five wells strategically drilled within the Upper Dalan Formation. To facilitate robust model evaluation, the dataset was intelligently partitioned, allocating ۷۰% for training and ۳۰% for testing. The selected input features, derived from conventional well logs, underwent rigorous pre-processing to enhance their predictive capabilities. The overarching goal was to predict effective hydrocarbon porosity based on outputs from the Nuclear Magnetic Resonance (NMR) log, with an emphasis on minimizing costs associated with the prediction process. The study offers an exploration of the strengths and limitations of each algorithm, providing insights into their efficacy for hydrocarbon reservoir characterization in the Upper Dalan Formation. Results from the comparative analysis indicate that the Random Forest algorithm emerged as the top performer, boasting an impressive R-squared value of ۰.۹۸۲ and a minimal Mean Absolute Error of ۰.۳۸۶ for the test dataset. These results highlight Random Forest's promise as a reliable and affordable method for precisely estimating effective hydrocarbon porosity in the complex geological formations of the Upper Dalan Formation. The present study provides significant contributions to the reservoir characterisation domain by facilitating the identification of ideal machine learning techniques for analogous hydrocarbon-containing formations.

Effective Porosity, Hydrocarbon Reservoir, Well Logging, Random Forest, Machine Learning.