Abstract

This paper introduces a fully automated, AIdriven framework for Reservoir Rock Typing (RRT), combining unsupervised and supervised learning methodologies tailored for subsurface characterization. The study was conducted in an deep offshore basin of deep-water turbidite depositional environment using basic and interpreted well log data from eight wells with side wall core data available for 5 wells. Two distinct pipelines were developed: ARRT (Automated Reservoir Rock Typing) for unsupervised lithofacies clustering, and SARRT (Supervised Automated Reservoir Rock Typing) for refined classification guided by core measurements and key well data. The workflow integrates advanced machine learning techniques including Self-Organizing Maps (SOM), DBSCAN, K-means, K-Nearest Neighbours (KNN), and Random Forests to map lithological variations and reservoir quality indices. Results demonstrate enhanced facies resolution, reduced manual interpretation, and high classification accuracy, supporting scalable and geologically consistent reservoir rock typing.