| Unsupervised Machine Learning Supported by Principal Component Analysis and Seismic Attributes for Heterogeneous Reservoirs Classification in Scarab Field, Offshore Nile Delta, Egypt |
| Paper ID : 1094-ISCH |
| Authors |
|
Amir Ismail *1, Tarek Khalifa2, Amin Khalil3 1Department of Geology, Faculty of Science, Helwan University, Helwan, Cairo, Egypt 2Department of Geology & Geophysics, Texas A&M University, Texas, 77843, United States of America 3Geology Department, Faculty of Science, Helwan University, Cairo, P.O. 11795, Egypt |
| Abstract |
| Identifying reservoirs associated with incised channel systems in the Nile Delta Deep Marine (NDDM) region is challenging due to complex stratigraphic features and highly heterogeneous reservoirs. Subsurface characterizations of lithofacies in incised channel systems of the NDDM region include varied lithologic and fluid contents, vertical stacking, and depositional patterns. To address these challenges, we propose a developed ML flowchart of unsupervised machine learning techniques, specifically Kohonen's Self-Organizing Maps (SOMs) and Principal Component Analysis (PCA), to enhance reservoir characterization from seismic amplitude data. The proposed workflow involves data conditioning, dimensionality reduction using PCA to extract the most informative features, and subsequent clustering with SOMs to identify and differentiate distinct lithofacies. The integration of both SOMs and well-log analysis positively impacts the precision and resolution of prediction and classification results, leading to more detailed recognition and highly accurate clustering of the 2D seismic dataset in the Scarab field into well-defined lithofacies. Furthermore, the method successfully delineates the gas-water contact within the reservoir interval. From these findings, we show that the selected attributes from PCA used as inputs into the SOM enhance the interpretation of the features of interest. The key achievement is to offer a precise flowchart showing visual evidence to improve highly heterogeneous reservoir characterization and enhance seismic data interpretation in the NDDM, ultimately reducing exploration risks. This approach offers a valuable tool for geoscientists working on complex stratigraphic features, including varied lithologic and fluid contents, using conventional seismic analysis or even a single ML algorithm on a limited seismic dataset. |
| Keywords |
| Reservoir characterization, self-organizing maps, machine learning, seismic attributes, principal component analysis, Scarab field, Nile Delta. |
| Status: Abstract Accepted (Oral Presentation) |