The production of oil and gas from naturally fractured reservoirs requires a thorough understanding of fracture network connectivity. Predicting fracture continuity & connectivity between wells -considering the immense control of multiple parameters on natural fracture behavior at reservoir conditions-remains to be a daunting challenge. A novel approach has been developed to provide a Continuous Fracture Network (CFN) model that honors the temporal and spacial changes of drivers that influence fracture distribution. This technology has been tested and successfully implemented in different carbonate and clastic reservoirs in the region. Complex reservoirs such as Permo-Carboniferous Alkhlata formation, Cambrian Amin formation, and Cretaceous Natih /Shuaiba reservoirs in Sultanate of Oman are examples of successful fracture modelling studies. These case studies are used to demonstrate various steps of continuous fracture characterization in this paper.
Well data, geological properties and high resolution attributes derived from seismic are integrated to build a 3D fracture model. The Fracture Indicators (FI) from borehole images (BHI) interpretation are correlated and ranked with fracture drivers to generate multiple CFN models on a 3D geocellular grid. A fuzzy neural network is used to evaluate the hierarchical effect of each driver on the fractures.
The training and testing of the neural network are accomplished using existing well data. Once the neural network finds the relationship between the drivers and the fracture intensity, it can be used to predict this fracture intensity in every cell of the field since all the drivers are known and available in the entire 3D grid. The model is subsequently validated by drilling data, blind wells and dynamic data. Effective permeability distribution is computed through dynamic calibration of potential well test data with the CFN models. Both fracture and effective permeability models are used to improve history matching, production optimization as well as picking new wells.