Beruflich Dokumente
Kultur Dokumente
Summary
In fractured reservoirs, data directly related to fractures are scarce
and 1D (e.g., core and image-log data). Other types of data are
more widespread (e.g., seismic data) but generally are related only
indirectly to fracture distribution. In such reservoirs, it is necessary
to understand and then to model the fracture network on a fieldwide scale by integrating all available data.
We propose a methodology to achieve this objective. The methodology establishes relationships between the fracturing and other
sources of data in a systematic workflow that goes from static 1D
data to a 3D dynamic model. The methodology is described and
illustrated with a case study from north Africa. In this field, fracture data from image logs and cores are related to (1) 3D seismic
attributes (e.g., amplitude, coherency), (2) fault patterns, and (3)
other types of well data (e.g., interval thickness, lithology index,
and porosity). Production data also are used to quantify the contribution of each fracture set to flow, which then can be mapped on
a reservoir basis with the more widely distributed log and seismic
data. The resultant set of maps then is entered into a dynamic
reservoir model. The methodology has been validated with a new
well, the fracture network of which was accurately predicted in the
reservoir by the model.
Introduction
Fractured reservoirs are by nature highly heterogeneous. In such
reservoirs, fracture systems control permeability and can also control porosity. Fracture modeling is therefore a key development
issue and requires an integrated approach from geology to reservoir simulation and well planning. Because fractures are below the
limit of seismic resolution, the static models of fractures are constrained mainly by well data (e.g., cores or image logs) using
conventional structural geology techniques.1 These models include
the mechanical origin (shears vs. joints), the geometry (orientation,
size, and frequency) and the typology (open vs. cemented) of the
fracture network. The fracture permeability then can be assessed
by relating the fracture aperture to the fracture excess conductivity
measured on electrical image logs,2 critically stressed fractures
within the present-day stress field,3,4 or both. It is the authors
opinion, however, that such approaches only give, at best, a relative estimate of permeability that must be calibrated against dynamic data. This requires the quantitative modeling of fracture
flow behaviors. At the drainage-radius scale of wells, discrete
fracture networks (DFNs)5,6 can be constructed and used to simulate flows and match them to well-test data.7 This allows us to
derive the fracture input parameters for reservoir simulation.8 Far
from wells, however, the lack of data makes DFN models very
uncertain. The static modeling of the spatial distribution of fractures at the field scale and the use of these models as input to
dynamic reservoir models are the purposes of this paper.
Methods have been presented to model field distributions of
fractures based solely on the fracture density measured along
wells.9 The scarcity of wells in which fracturing data are available
284
makes such a direct mapping difficult and very uncertain, however. Geometrical methods based on the fractal theory predict
subseismic fractures from seismic faults.10,11 These methods can
be hazardous when used to extrapolate over several orders of magnitude (i.e., from seismic faults down to core scale fractures) and
generally apply only to shear fractures, not to joint systems.12
Bourne et al.13 propose a geomechanical method to predict the
fracture distribution related to the elastic stress field perturbation
around faults. This technique does not allow the prediction of
fractures that do not result from the activation of seismic faults
(e.g., doming). Leroy and Sassi14 and Guiton15 suggest another
geomechanical method that relies on an idealization of the real
fractured rock by a continuum. They introduce opening and sliding
displacements to represent the reservoir-scale deformation and the
diffuse fracture patterns that accommodate it. This approach assumes a homogeneously fractured rock and therefore does not
predict fracturing or faulting localization. Heffer et al.16 propose a
geostatistical technique to interpolate strain/tensor components
supposedly related to fracturing. The estimation is conditioned to
well and structural data and is calibrated against well-test permeabilities. All these methods assume that the fracturing process is
related to a limited number of geological parameters that constrain
the mechanical behavior of fractured rocks. Although only one
parameter may be needed to characterize a fractured reservoir,1719
it is often the lack of a methodology to integrate the combined
effects of structure, thickness, and lithology that leads geologists to
focus only on the most important factor. With complex reservoirs,
however, more comprehensive descriptions are unavoidable when
producing reliable fracturing models. Ericsson et al.20 build an
empirical and deterministic approach to derive a fracture density
index that is a function of other indices related to reservoir variables like the structural curvature, the crestal distance, or the facies
type. Similarly, Agarwal et al.21 relate the fracture intensity to both
geological parameters and effective permeabilities to model field
permeability distributions. Their approach is less empirical but
remains fully deterministic (and hence inappropriate) to address
the uncertainty inherent in the spatial distribution of fractures or
permeabilities. Stochastic methods are the only way to account for
such uncertainties. One approach22,23 consists of using a multivariate, nonlinear regression function of secondary (geological)
parameters to fit well fracturing data (i.e., related to a fracture
index). The stochastic aspect addresses the uncertainty on the regression model by random sampling of the data set. Repeatedly, a
data subset is drawn and used to fit the regression function, which
can be accepted or rejected according to some correlation criteria.
If accepted, the multivariate regression function is applied to the
whole field to produce a realization of the fracture index. The
particularity of this approach is the use of a neural-network architecture to define the regression function. However, the approach
generates unconditional realizations (fracturing data are partly
honored) and does not allow the reproduction of any statistical
model as inferred from the data.
In this paper, a geostatistical approach is presented to simulate
fracture frequencies with the integration of primary fracturing data
and any variety of secondary geological, geomechanical, or seismic information reflecting the understanding of the fracturing process. In the following, the method is presented and applied to a
north Africa field. The results are validated against new drilling
data, and their use as input to a reservoir model is discussed.
August 2002 SPE Reservoir Evaluation & Engineering
Methodology
Full-field static fracture modeling requires that we first characterize the geometry of the entire fracture network: orientation, frequency, and size. Orientations are often direct measurements (e.g.,
in image log data). Gauthier et al.1 describe a method to quantify
fracture dimensions from well data using the proportion of fractures that terminate within the wellbore. We focus here on the
spatial distribution of fracture frequencies for directional sets of
fractures observed in wells. The fracture frequency (FF) is defined
as the number of fractures per unit length measured perpendicularly to the fracturing plane. It can be seen as a fracture density
geometrically corrected to be independent of well directions.1
A geostatistical approach that is based on both multivariate
statistical analysis and sequential simulation is our proposed solution to the data integration problem.
Primary and Secondary Variables. Two types of variables are
considered for each oriented fracture set.
The primary variable, the one to be modeled, is the FF. FF
data are calculated at wells in which image-log or core interpretations are available.
Secondary variables (SV) are variables derived from field
data or simulation and potentially correlated with fracturing. Secondary variables can be of a different nature: lithological (porosity
distribution), mechanical (geomechanical simulation results), or
structural (seismic attributes like coherency or curvature). They are
known over the entire field (exhaustive information).
Modeling Steps. The two-step geostatistical approach tackles the
problem of modeling the spatial distribution of a fracturing variable (the FF) and quantifying the uncertainty about it. In the first
step, discriminant analysis is used to derive a single geological
component that best capitalizes the indirect information about the
FF from all secondary variables. In the second step, the objective
is to generate FF realizations that honor well fracturing data and
reproduce different statistical models inferred from the data. These
statistical models include the global distribution of the FF (histogram); variogram models that specify the spatial correlation of the
FF and its cross-correlation with the geological component; and a
bivariate distribution model of the FF and the geological component. The bivariate distribution model provides the conditional
distribution of the FF for any geological component value. The
geological component being known everywhere over the field, this
model determines the a priori FF distribution (before simulating)
at any location according to the local geological component value.
In that sense, the geological component carries information about
spatial fracturing trends. The roles of the two steps of multivariate
statistical analysis and sequential simulation are as follows.
Multivariate Statistical Analysis. As discussed earlier, it is
several secondary (explanatory) variables, taken together, that allow us to identify spatial fracturing trends. To integrate them into
the geostatistical approach, discriminant analysis is used to determine the linear combination of SVs that best distinguishes selected
FF classes (e.g., low, medium, and high fracturing classes). By
considering classes, discriminant analysis can recognize nonlinear
relationships with the SVs. Because the SVs are not necessarily
independent or linearly correlated, improved (i.e., linear) correlation can be obtained by transforming them into distribution probabilities (cumulative distribution function values). From the discriminant analysis, only the first component (the so-called geological component) is retained, provided that it indeed shows a
monotonic relationship with the FF. At this stage, the geological
component remains the unique secondary variable that capitalizes
at best all indirect information about FF.
Sequential Simulation. To better represent the rather complex
nonlinear relationship between the FF and the geological component, sequential indicator simulation is used to simulate the FF.
The integration of the geological component (secondary information) is done by cokriging through a Bayesian formalism in which
the geological information is converted into soft (probability-like)
data.24,25 A Markov model is adopted, which leads to a collocated
August 2002 SPE Reservoir Evaluation & Engineering
i=N
S=
Vti dli
WS , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1)
i=1
Relation Between Fracturing and Production Data. The relation between fracture and production data reflects the role of fracturing in the dynamic behavior of the field. Fig. 6 shows the
distribution of the productivity index (PI) over the entire field and
for both horizontal and vertical wells. The following comments
can be made: (1) the PI distribution is log-normal, which is indicative of wells drilled in a heterogeneous reservoir; (2) although
the best two producers are horizontal wells, the performance of this
type of well is not systematically better than vertical wells; (3)
wells located in the southern part of the field (where dolomitization occurred) show better production performances than those
located in the central and northern parts; and (4) the three regions
show the same type of PI distribution. It could be argued that this
August 2002 SPE Reservoir Evaluation & Engineering
Fig. 6Relative PI distribution in the studied field for all tested wells.
secondary variables. Two extreme cases are evaluated. One involves a few (three) classes representing low-, medium-, and highFF values. The other relies on many (approximately 20) classes,
with most classes containing no more than three or four data. This
second class definition allows us to distinguish the few highestfracture-frequency data, the latter belonging to the same class
(with other data) in the three-class definition. Fig. 8 shows the
discriminant analysis correlation plots, with the first two discriminant components represented by the x and y axes. With three
classes, the geological component is dominated by the seismic
amplitude and the deep faults, with a positive correlation, and by
the gross thickness and the distance to faults, with a negative
correlation. With 20 classes, the dominating explanatory variables
become the deep faults, the S-map, and the dolomite content with
a positive correlation and the distance to faults with a negative
correlation. The scatterplots of the FF vs. the first geological component is given in Fig. 9 for the two class definitions. Though the
two clouds of points look very similar, the three-class component
better separates at least some of the very small, high-fracture frequencies (i.e., cases from the two extreme classes). The reason can
be found in the number of classes; it is simply easier to distinguish
a few classes than many of them.
Sequential Simulation. The two previous geological components (with 3 and 20 classes) can be calculated at grid-nodes where
the secondary variables are known. Each geological component is
used independently as secondary (soft) information to simulate the
northwest/southeast FF with the sequential indicator simulation
approach presented previously. For each geological component,
100 realizations are generated from which probability maps can be
derived. The probability maps in Fig. 10 show that the northwest/
southeast FF be greater than or equal to 0.5 frac/m.
These probability maps call for some comments. First of all, the
maps look very similar in terms of trends and show similar high-
Dolomite
Strain
0.75
0.5
Deep Faults
0.25
Dist2Faults
0
GrossThick
Amplitude
0.25
0.75
0.5
0.25
Dist2Faults
Dolomite
Strain
Amplitude
Deep Faults
l
0.25
0.5
0.5
0.75
0.75
GrossThick
1
1
1
0.75 0.5
0.25
0.25
0.5
0.75
3-Class Component
0.75 0.5
0.25
0.25
0.5
0.75
20-Class Component
Fig. 8Correlation plots corresponding to the first two discriminant components for different class definitions.
3-Class Component
nition that gathers all fracture frequencies greater than 0.5 frac/m
into the same class without any distinction between them. The
northern high-probability regions therefore can be seen as influenced (in extrapolation conditions) by the high-fracture-frequency
data that the 20-class discriminant analysis function tries to fit.
Because these data are located in the southern part of the field,
their influence in the north can be questioned. Otherwise, both
probability maps show uncertain regions of medium probabilities
(e.g., from 40 to 60% chance). In these regions, the available
information is not enough to predict with confidence whether the
FF is above or below 0.5 frac/m. A comparison of this method with
the method of Ouenes et al.23 can be found in Gauthier et al.29
2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0
20-Class Component
Fig. 9Scatterplots of the northwest/southeast FF vs. the geological component for different class definitions.
August 2002 SPE Reservoir Evaluation & Engineering
Fig. 10Probability maps showing that the northwest/southeast FF will be greater than or equal to 0.5 frac/m for (a) the 20-class
geological component and (b) the 3-class component.
Average fracture density within each facies type. This constraint is controlled by a set of FF simulations in each facies type.
The same FF value applies to all cells within the same facies type
in the column.
The distribution of fracture lengths is also required to characterize fully the geometry of the DFN. Without information to
constrain this parameter, as proposed by Gauthier et al.,1 a lognormal distribution is considered. Sensitivity analysis shows that,
unless unrealistic, the fracture length has a limited impact on upscaling results.
290
Equivalent Flow Parameters of the Fracture System. It is beyond the scope of this paper to discuss in detail the match of the
6-year history of the field. Instead, a procedure is presented to
derive, from simulated FF maps, equivalent dynamic properties
(e.g., kx, ky, kz) assigned to the blocks of a reservoir simulation grid.30
The simulation of flows in DFNs requires conductivities to be
assigned to fractures. Without evidence in this field of in-situ stress
controls of the fracture conductivity or of geographical aperture
variations caused by diagenesis, a constant aperture is assumed for
the entire reservoir. Consequently, lateral and vertical permeability
August 2002 SPE Reservoir Evaluation & Engineering
1 frac / 2m
Along-Hole Measured
NW/SE FF
1 frac / 5m
TVD (ft)
3,350
3,400
3,450
1 frac / 0.75 m
Along-Hole Measured
NW/SE FF
3,500
FF threshold
Dolomitic zones
Poor FMI data quality zones
4 m k
Fig. 11Comparison between interpreted and predicted northwest/southeast FFs in the dolomite facies along Well S-H2.
changes depend only on simulated facies types and FF. The constant aperture is calibrated by matching fluid-flow simulations in
DFNs to well tests or to some production data.
For computing reasons, DFNs cannot be generated in all cells
of the (full-field) 3D lithological grid to derive equivalent properties. Instead, the following procedure is proposed.
DFNs are generated in 1,000 cells drawn randomly in the 3D
lithological grid.
Equivalent dynamic properties are computed for each of
these DFNs.8
Principal component analysis is run to derive, for each dynamic property, a unique FF component that it is best correlated to it.
The relationships between FF components and dynamic
August 2002 SPE Reservoir Evaluation & Engineering
292
1.8
Predicted
Interpreted
1.4
1.0
0.6
0.2
0.0
0
2000
4000
Well Offset, m
3. Heffer, K.J. et al.: The influence of natural fractures, faults and earth
stresses on reservoir performance-geomechanical analysis by numerical modeling, North sea oil and gas reservoirs III, Norwegian Inst. of
Technology, Kluwer Academic Publishers (1994) 201.
4. Barton, C.A. et al.: Utilizing wellbore image data to determine the
complete stress tensor: application to permeability anisotropy and wellbore stability, The Log Analyst (NovemberDecember 1997) 21.
5. Swaby, P.A. and Rawnsley, K.D.: An Interactive 3D Fracture Modeling Environment, paper SPE 36004 presented at the 1996 SPE Petroleum Computer Conference, Dallas, 25 June.
6. Cacas, M.C. et al.: Nested geological modelling of naturally fractured
reservoirs, Petroleum Geoscience (2001) 7, S43.
7. Wei, L. et al.: Discriminating Fracture Patterns in Fractured Reservoirs by Pressure Transient Tests, paper SPE 49233 prepared for
presentation at the 1998 SPE Annual Technical Conference and Exhibition, New Orleans, 2730 September.
8. Bourbiaux, B. et al.: A Fast and Efficient Methodology to Convert
Fractured Reservoir Images Into a Dual-Porosity Model, paper SPE
38907 presented at the 1997 SPE Annual Technical Conference and
Exhibition, San Antonio, Texas, 58 October.
9. Guerreiro, L. et al.: Integrated Reservoir Characterization of a Fractured Carbonate Reservoir, paper SPE 58995 presented at the 2000
SPE International Petroleum Conference and Exhibition in Mexico,
Villahermosa, Mexico, 13 February.
10. Gauthier, B.D.M. and Lake, S.D.: Probabilistic modeling of faults
below the limit of seismic resolution in Pelican field, North Sea, Offshore United Kingdom, AAPG Bull. (1993) 77, No. 5, 761.
11. Cowie, P.A. et al.: Introduction to the special issue: Scaling law for
faults and fractures populationanalysis and applications, J. of Structural Geology (1996) 18, vxi.
12. Loosveld, R.J.H. and Franssen, R.C.M.W.: Extensional vs. Shear
Fractures: Implications for Reservoir Characterisation, paper SPE
25017 presented at the 1992 SPE European Petroleum Conference,
Cannes, France, 1618 November.
13. Bourne, S.J. et al.: Predictive modelling of naturally fractured reservoirs using geomechanics and flow simulation, GeoArabia (2001) 6,
No. 1, 27.
14. Leroy, Y.M. and Sassi, W.: A plasticity model for discontinua, Aspects of tectonic faulting, F. Lehner and J. Urai (eds.), Springer-Verlag
(2000).
15. Guiton, M.: Contribution of pervasive fractures to the deformation
during folding of sedimentary rocks, PhD dissertation, Ecole Polytechnique, France (2001).
16. Heffer, K.J., King, P.R., and Jones, A.D.W.: Fracture Modeling as
Part of Integrated Reservoir Characterization, paper SPE 53347 presented
at the 1999 SPE Middle East Oil Show, Bahrain, 2023 February.
17. McQuillan, H.: Small scale fracture density in Asmari formation
southwest Iran and its relation to bed thickness and structural setting,
AAPG Bull. (1973) 57, No. 12, 2367.
18. Murray, G.: Quantitative fracture studySanish Pool, McKenzie
County, North Dakota, AAPG Bull. (1968) 52, No. 1, 57.
19. Nelson, R.: Geologic Analysis of Naturally Fractured Reservoirs, Gulf
Publishing Co., Houston (1985).
20. Ericsson, J.B., McKean, H.C., and Hooper, R.J.: Facies and Curvature
Controlled 3D Fracture Models, Geological Society of London Special
Publication (1998) 147, 299.
21. Agarwal, B. and Allen, L.R.: Ekofisk Field Reservoir Characterisation: Mapping Permeability Through Facies and Fracture Intensity,
paper SPE 35527 presented at the 1996 SPE European 3-D Reservoir
Modelling Conference, Stavanger, 1617 April.
22. Zellou, A.M, Ouenes, A., and Banik, A.K: Improved Fractured Reservoir Characterization Using Neural Networks, Geomechanics, and
3-D Seismic, paper SPE 30722 presented at the 1995 SPE Annual
Technical Conference and Exhibition, Dallas, 2225 October.
23. Ouenes, A.: Practical application of fuzzy logic and neural networks
to fractured reservoir characterization, Computer and Geosciences,
Shahab Mohaghegh (ed.) (2000) 26, No. 7.
24. Gooverts P.: Geostatistics for natural resources evaluation, Oxford U.
Press (1997).
25. Zhu, H. and Journel, A.G. Formatting and Integrating Soft Data: Stochastic Imaging via the Markov-Bayes Algorithm, Geostat Troia,
Soares (ed.), Kluwer Publishing (1992) 112.
293
26. Schmaryan, L.E. and Journel, A.G.: Two Markov models and their
applications, Math Geology (1999) 31, No. 8, 965.
27. Auzias, V. et al.: Fracture orientation modelling in the vicinity of a
horizontal well, Bull., Centre Rech. Elf Explor. Prod. (1997) 21, 381.
28. Townsend, C. et al.: Small seismic-scale fault identification and mapping, Faulting, Fault sealing and Fluid flow in Hydrocarbon Reservoirs, G. Jones, Q.J. Fisher, and R.J. Knipe (eds.), Geological Society,
Special Publications (1998) 125.
29. Gauthier, B.D.M. et al.: Integrated Fractured Reservoir Characterisation: A Case Study in a North Africa Field, paper SPE 65118 presented at the 2000 European Petroleum Conference, Paris, 2425
October.
30. Daniel, J.M. et al.: The use of discrete fracture models to constrain
full field flow simulations, paper XX presented at the 2001 EAGE
Conference & Technical Exhibition, Amsterdam.
31. Cosentino, L. et al.: Integrated Study of a Fractured Middle East
Reservoir With Stratiform Super-K IntervalsPart 2: Upscaling and
Dual Media Simulation, SPEREE (February 2002) 24.
32. Van Lingen, P. et al.: Single Medium Simulation of Reservoirs with
Conductive Faults and Fractures, paper SPE 68165 presented at the
2001 SPE Middle East Oil Show, Bahrain, 1720 March.
294