The Application of Discrete Fracture Network Models to Fractured Reservoir Engineering: Analytical Approach, Data Sets and Early Results in Yates Field, West Texas
|Paul La Pointe
Golder Associates Inc.
Marathon Oil Co.
Project Web Page: http://www.golder.com/niper/niprhome.htm
Overview of Discrete Fracture Network (DFN) Approaches
The movement of hydrocarbons and other fluids in fractured reservoirs or in conventional reservoirs with significant fracture permeability often is not as expected or predicted. This behavior is seen in early water breakthroughs; reduced tertiary recovery efficiency due to channeling of injected gas or fluids; dynamic calculations of recoverable hydrocarbons that are much less than static mass balance ones due to reservoir compartmentalization; and dramatic production changes due to changes in reservoir pressure as fractures close down as conduits. These problems often lead to reduced ultimate recoveries or higher production costs.
Production experience with fractured reservoirs has repeatedly shown how understanding and exploiting the fracture connectivity at the reservoir scale is an important factor for optimizing reservoir performance. Discrete fracture network (DFN) models portray fractures and their fracture connectivity very differently from other methods. For example, the conventional method to simulate fracture-dominated reservoirs is to represent the rock as a dual-porosity, dual-permeability continuum. The matrix is represented as blocks or slabs. The fractures are mathematically represented as another continuum spatially coincident with the block faces. The properties of each matrix block are represented as a symmetrical tensor, and the properties are continuous throughout the entire block. Smaller fractures within each matrix block are coupled to the rock by means of factors that are related to the fracture geometry, such as the Sigma factor. Values for the fractures are generally specified as transmissibility multipliers.
This type of numerical model cannot reproduce many commonly observed types of fractured reservoir behavior because it does not accurately reflect the geometry of fluid flow pathways. Conventional dual-porosity models do not explicitly model the geometry of discrete fractures, solution features, and bedding that control flow pathway geometry. This inaccurate model of discrete feature connectivity results in inaccurate flow predictions in areas of the reservoir where there is not good well control.
DFN models afford an alternative approach. These models more realistically model the connectivity of the faults and joints that give rise to reservoir-scale and well-scale non-continuum flow behavior. In the DFN model, each conductive fracture is modeled explicitly as one or more 1D, 2D, or 3D element. Physical properties such as transmissivity or storage, and geometrical properties such as size, elongation and orientation are assigned to each polygon based upon measured data or geologically conditioned statistical distributions derived from measured values.
A DFN model typically combines deterministic and stochastic discrete fractures. The deterministic fractures are those directly imaged through seismic or intersected in wells. Other, usually smaller-scale fractures may not have been detected through seismic, yet may be very important for reservoir performance. These fractures are generated stochastically. The geometrical and physical properties for these stochastic fractures are assigned through Monte Carlo sampling of relevant distributions, which may also be conditioned to both structural geology and depositional framework. Examples of fracture models conditioned to fault systems, folding, doming and three-dimensional stratigraphic reconstructions are shown.
One of the many advantages of the DFN approach is that it makes consistent use of a wide variety of disparate geological, geophysical and production data, which conventional dual-porosity models cannot incorporate to the same extent. Data which can be used for constructing DFN models can be derived from lineament maps, outcrops, 2D and 3D seismic, well logs of various types, core, single well and multi-well production tests, flow logs, injectivity profiles, as well as structural or depositional conceptual models. Specialized tools have been developed to derive the necessary input data for DFN models from these sources, as illustrated in the presentation.
DFN models have been used for a wide variety of exploration and production purposes over the last decade. The current study represents an extension of DFN technology to improve process modeling in tertiary recovery efforts at Yates Field, West Texas.
Review of Yates Field, West Texas
An on-going cooperative project between Golder Associates Inc., Marathon Oil Co. and the Massachusetts Institute of Technology, sponsored in part by the U. S. Dept. of Energy, has focused on advancing existing DFN approaches and testing them in Marathons Yates Field in West Texas. The Yates Field was discovered in 1926 and has produced over a billion barrels of oil. The San Andres is the thickest and most prolific formation within the Yates Field. The reservoir properties of the San Andres are complicated by extensive natural fracturing, karsting, dolomitization, and precipitation of secondary calcite cement.
Before 1976, the Yates Field was operated under depletion. Beginning in 1976, gas was injected to retard water invasion. Waterflooding began on the western flanks of the field in 1979, and a polymer flood was expanded into additional portions of the west-side oil column from 1983 to 1986. During late 1985, carbon dioxide injection commenced in the northern, eastern and crestal areas of the field, and was abandoned in 1991. A field-wide co-production project was initiated in late 1992 to de-water reservoir areas containing oil bypassed by water encroachment. Water invasion occurred in the field's fracture network, bypassing the oil located in the highly oil saturated matrix. from the fracture network. Co-production successfully reversed aquifer encroachment by removing water from the fracture network and allowing the oil to flow from the matrix into the fracture system
Applications of DFN Models to the Yates Field
DFN models have been used to support Yates field development since1992. The wells in the field produce matrix fluids by means of the extensive natural fracture network. By focusing on the areas where the fracture network is best developed, it is possible to take advantage of the natural drainage system by maximizing withdrawals from high rate, high efficiency wells in these areas. For the existing wells it was possible to identify those located in highly fractured areas of the reservoir based on their production history. For new wells, DFN models were used to study the spatial distribution of fractures in the field and to optimize the location and orientation of the new horizontal wells. In 1993 and 1994 more than 30 new short-radius horizontal wells were drilled and almost 400 wells where shut in while keeping a stable total daily oil production rate.
The current DFN modeling work is focusing on optimizing steam injection for tertiary recovery. The reservoir has a large gas cap that slowly drains oil from the matrix situated above a thin fractured oil column. The fractured nature of the reservoir makes displacement processes ineffective, which limits the effectiveness of conventional recovery processes. Gravity drainage controls oil flow from the matrix into the fracture system. Heat flow through the fracture system (and thus heat transfer from the fractures into the matrix system) will assist oil drainage by reducing oil viscosity and improving the oil relative permeability. Because of the fractured nature of the reservoir, this process can not be designed as a classic thermal flood. Instead Marathon developed and patented a process called Thermally Assisted Gravity Segregation (TAGS). It is in this area that DFN technology is being expanded to provide a modeling tool for simulating TAGS steam injection into the Yates Field, specifically in the following two areas of performance enhancement:
In addition, several other new DFN tools have been developed to study reservoir compartmentalization and tributary drainage volume for fractured reservoirs produced through pressure depletion. These new approaches are also being developed and tested to the extent possible at Yates.
The application of DFN modeling to the Yates Field TAGS project consists of four stages: development of the DFN model for the two pilot study tracts; development of specialized tools for tertiary recovery process modeling; verification of tools and DFN model; and application to TAGS.
The construction of DFN models for the two test sites in Yates, and the development of all of the new DFN tools for tertiary recovery process modeling have been completed. These tools include methods for analyzing reservoir compartmentalization and tributary drainage volumes. In addition, a particle-tracking code to track "heat" particles through the fractures and matrix has been developed to model the movement of the condensation front of steam as it moves out from injection wells into the reservoir. Heat particles are injected into the fracture system at each well. Each particle carries a quantum of heat. The temperature is a function of the number of heat particles per volume. Heat particles may move through the fractures by convective movement within the fractures, and by conductive transport between the fractures and the rock matrix using standard diffusion equations, where the fractures are treated as radiative heat boundaries for the matrix blocks. This code has been verified for a DFN model consisting of a single rectangular fracture.
Verification of the DFN model for the Tract 17 area is currently underway. Verification consists of modeling a multiwell tracer experiment. Successful matching of the field results verifies that the DFN fracture model is modeling the larger-scale fracture network geometry and connectivity.
The tracer experiment was carried out around a deep water injection well located in the northern portion of the field. Both bromide and thiocyanate tracers were used. Major tracer breakthroughs were observed at four wells lying to the southeast of the injection well. Minor breakthroughs were observed in two wells lying to the east. These tests confirmed a dominant northwest-southeast directionality, and a much slower dispersion perpendicular to this direction. Tracer test simulations using the DFN model for Tract 17 are currently underway. Upon successful matching of the tracer experiments, the DFN model together with the heat particle tracking code will be used to model steam injection experiments in 1998 in order to optimize TAGS implementation.