Deterministic versus stochastic discrete fracture network (DFN) modeling: Application in a heterogeneous naturally fractured reservoir. 9th Middle East Geosciences Conference, GEO 2010.

Fracture modeling is a multi-step process involving several disciplines within reservoir characterization and simulation. The main idea is to build on geological concepts and gathered data such as: (1) interpretation of faults and fractures from image log data and 3-D seismic; (2) use of field outcrop studies as analogs for conceptual models; and (3) use of seismic attributes as fracture drivers. The purpose of modeling fractures is to create simulation properties with the power to predict the reservoir behavior. This note applies the concept of discrete fracture network (DFN) in a unique and comprehensive study of fracture modeling in one of the naturally fractured carbonate reservoirs of the Middle East. A discrete fracture network is a group of planes representing fractures. Fractures of the same type that are generated at the same time are grouped into a fracture set. Each fracture network containing fractures has at least one fracture set but may have many. The simplest fracture sets are defined deterministically as a group of previously defined fractures, either as a result of fault plane extraction from a seismic cube, or as previously defined fractures. Fractures modeled stochastically can be described statistically either using numerical input, or properties in the 3-D grid. Properties in the 3-D grid can vary in 3-D and can easily be modeled using seismic attributes from 3-D seismic data. The scale-up fracture network converts the discrete fracture network (with its defined properties) into the properties that are essential for running a dual porosity, or dual permeability simulation. A simple simulation model is developed for three different grid types, using the software ECLIPSE 100, grid without DFN modeling, deterministic DFN modeling and stochastic DFN modeling. The results of the reservoir simulation indicate that cases with Stochastic DFN have a better result (history match) than cases with Deterministic DFN and the grid without DFN.