Machine Learning-Based Prediction of Natural Fractures and Vugs in Carbonate Reservoir

ABSTRACT: To mitigate drilling risks and optimize the economic and efficient utilization of carbonate resources, precise reservoir characterization is essential. However, predicting the distribution of natural fractures and vugs in carbonate reservoirs using seismic and well data is fraught with uncertainty. This arises from challenges such as inadequate well interpretation, poor seismic imaging, and inconsistencies between well and seismic data. This study aims to reduce these uncertainties by calibrating the interpretation of natural fractures and vugs using well log and seismic data. Additionally, a machine learning model is developed to predict the 3D spatial distribution of fractures and vugs in limestone reservoirs. The research focuses on the limestone formations of the H oilfield in the southeastern Mesopotamia Basin of Iraq. Seismic attributes are extracted from inter-well regions based on well locations and tracks from six characteristic wells, and these attributes are classified and calibrated in relation to fractures and vugs. Data balancing techniques are applied, and feature engineering is performed to optimize model performance. An ensemble learning method is then used to establish a nonlinear mapping between the predicted targets (fractures, vugs, matrix) and seismic attributes. Blind well testing is conducted to verify the reliability of the model. The study ultimately provides a 3D prediction of the spatial distribution of fractures and vugs, derived from the established relationships rather than the original properties. The model achieves an average prediction accuracy of 83%. The results show that vugs are primarily distributed in the MA layer, while fractures and vugs are mainly found in the MB layer, and fractures are predominantly located in the MC layer. When planning horizontal wells, the well trajectory should be optimized to reduce the risks of mud loss and wellbore instability, particularly in the MC layer. The model offers detailed information on the distribution of fractures and vugs at different depths and geographical locations, providing crucial guidance for the development of optimal drilling plans and offering both technical and theoretical support for reservoir development strategies.