Implementation of Machine Learning Approach in Pore Size Evaluation Workflows for Carbonate Reservoirs

Abstract

Abstract The petrophysical evaluation of carbonate reservoirs presents a significant challenge due to complex diagenetic processes that alter reservoir properties. Advanced log measurements, such as NMR T2 relaxation time and borehole images, are often necessary to evaluate the carbonate pore systems (micro, meso, and macro pores), and permeability and infer the impact of diagenesis on reservoir properties. The absence of these measurements may lead to inappropriate estimation of reserves and inadequate development strategies. To address this challenge, we propose a novel interpretation method that utilizes machine learning algorithms to perform carbonate porosity partitioning (micro, meso, and macro) using standard log measurements (e.g., resistivity, density, neutron, sonic, and gamma-ray). We trained the machine learning model in key wells with NMR and/or borehole images to establish relationships between standard logs and carbonate pore partitions. Subsequently, we applied the model to other wells to predict the target pore volumes. Standard ML model workflow (data cleaning, feature engineering, train/split, hyperparameter tuning) was followed. We present a comprehensive workflow for accurate evaluation and characterization of carbonate reservoirs where NMR measurements are not available. We successfully applied this workflow to datasets from heterogeneous carbonate reservoirs in South Iraq to obtain reliable carbonate pore partitions. These pore volumes were then used to estimate permeability and variable cementation exponent to improve water saturation and perform carbonate rock typing. Our proposed methodology provides a reliable and efficient approach (if applied with domain knowledge and reservoir considerations) for the petrophysical evaluation of inter- and intragranular, vuggy carbonate reservoirs, particularly in cases where advanced log measurements are not available, overcoming challenges with the traditional interpretation methods in such complex reservoirs.