Object-Based Global Optimization in Modeling Discrete-Fracture Network Map: A Case Study
- Nam H. Tran (The University of New South Wales) | Zhixi Chen (The University of New South Wales) | Sheik S. Rahman (The University of New South Wales)
- Document ID
- Society of Petroleum Engineers
- SPE Annual Technical Conference and Exhibition, 5-8 October, Denver, Colorado
- Publication Date
- Document Type
- Conference Paper
- 2003. Society of Petroleum Engineers
- 5.1.2 Faults and Fracture Characterisation, 7.6.6 Artificial Intelligence, 4.3.4 Scale, 5.8.6 Naturally Fractured Reservoir, 4.1.2 Separation and Treating, 5.5 Reservoir Simulation, 4.1.5 Processing Equipment, 5.1 Reservoir Characterisation, 3.2.3 Hydraulic Fracturing Design, Implementation and Optimisation, 5.1.5 Geologic Modeling, 5.7.2 Recovery Factors, 3.3.2 Borehole Imaging and Wellbore Seismic, 5.6.1 Open hole/cased hole log analysis, 5.1.7 Seismic Processing and Interpretation
- 0 in the last 30 days
- 388 since 2007
- Show more detail
- View rights & permissions
|SPE Member Price:||USD 5.00|
|SPE Non-Member Price:||USD 28.00|
Modeling of fracture network is an essential step for developing naturally fractured reservoirs study. It helps to develop a best scenario for hydraulic fracture treatment, to design an optimum production method and to evaluate reservoir potential. This paper presents an integrated methodology for modeling fracture network by utilizing object modeling and global optimization. It also describes a field study to evaluate the methodology's effectiveness. Firstly, as an object-based model, each fracture is presented and treated as a discrete object, characterized by its centre location, orientation and size. Object-based modeling allows the output of spatial distribution and details of discrete fracture network. From observed data sources such as seismics, outcrops, well logs and images, etc., we characterize fracture and field attributes, including fracture density, fracture parameters' statistics. Then, we perform a statistical analysis on fracture properties to identify density functions and distribution patterns. The essential feature of this approach is the formulation of the objective function. Various variogram measurements, modified multi histograms and other statistical properties are selected, so that the objective function is able to adequately describe the representative field data. In the next step, we use global optimization (simulated annealing algorithm) to produce result fracture network, by optimizing the objective function. This network matches the field characterized fracture attributes and characteristics. A global optimization method such as simulated annealing is able to honor more data than conventional geo-statistical simulation techniques. Case study shows that the modeling methodology mapped discrete fracture network very closely to the observed fracture distribution and properties.
Naturally fractured reservoirs have recently attracted intensive research attention, because the world market is increasingly under pressure to exploit energy from unconventional sources such as naturally fractured oil, gas and hot dry rock reservoirs.
An extensive study of a fractured reservoir should include four major steps: geological reservoir characterization, fracture network modeling, hydraulic stimulation analysis and fractured reservoir simulation.
Thus, building a fractured reservoir model is critical to optimize production and increase recovery efficiency, by:
Selecting best locations for production wells,
Studying the response of natural fractures under stimulation pressure, hence, developing best scenarios for hydraulic fracture treatment,
Designing optimum production methods and evaluating reservoirs potential.
Several techniques to simulate naturally fractured reservoirs were documented in literature, such as mathematical and geo-mechanical models1-3. These methods, however, mostly used only single source of data (seismic maps or well logs) and relied on simplistic geometrical descriptions of fracture systems (e.g. homogeneous reservoirs, parallel plate fractures). More recently, there were more computational intensive models using stochastic simulation4,5, neural network, fuzzy logics6,7 and other artificial intelligence tools8,9. These integrated methods, to some extend, succeeded in utilizing multiple field data sources and modeling fracture networks. Nevertheless, accuracy of the stochastic simulations was only reliable in near wellbore regions, and the unconditioned random filling of inter-well regions limited their uses beyond the near wellbore regions. In addition, stochastic simulations could not effectively utilize more sophisticated statistics such as variogram, correlogram, rodogram and madogram. Most of the previous works did not generate nor simulate discrete fractures. Their outputs were fracture intensity (density) maps, where details of fracture parameters (orientations, sizes and apertures) were not considered.
|File Size||1 MB||Number of Pages||12|