Abstract

Two categories of natural fractures are generally recognized: regional orthogonal fractures and structure-related or tectonic fractures. Over the past eight years or so, we have developed a methodology using neural networks and fuzzy logic to assist with the characterization of naturally fractured reservoir rocks. Conventional methods use only one or two geological parameters to characterize a naturally fractured reservoir. However, an integrated approach that utilizes all the information available (including lithology, thickness, state of stress and fault patterns) is required. Our approach makes use of fuzzy logic to quantify and rank the importance of each geological parameter on fracturing; a neural network is used to find the complex, non-linear relationship between these geological parameters and the fracture index.

Three case studies are described in this paper to illustrate the use of this methodology. They report respectively on a faulted limestone oil reservoir in North Africa which is deformed by fault-related fractures; a carbonate oil reservoir in New Mexico which is, characterized by fold-related fractures; and a sandstone gas reservoir in NW New Mexico which has regional orthogonal fractures.

These three case studies illustrate fracture identification and prediction using static information (i.e. image logs) and/or dynamic information (i.e. well performance) as a fracture index. The results of these studies are described as they relate to infill drilling, understanding the fractured reservoir and improved reservoir simulation.