Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Reservoir Modeling For Horizontal-Well Exploitation of A Giant Heavy-Oil Field

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Reservoir Modeling for Horizontal-Well

Exploitation of a Giant Heavy-Oil Field


T.H. Tankersley, SPE, and M.W. Waite, SPE, Petrolera Ameriven

Summary
The Hamaca field, located in Venezuelas Orinoco heavy-oil belt,
is a giant extra heavy-oil accumulation operated by Ameriven, an
operating agent company for PDVSA, ConocoPhillips, and ChevronTexaco. Over the 35-year life of the field, more than 1,000
horizontal laterals are planned to deliver 190,000 BOPD to a
heavy-oil upgrader facility. Reservoir models are built to support
a broad continuum of activities to meet this objective. This paper
will review the Hamaca reservoir-modeling process, the challenge
of integrating many sources of geologic and geophysical constraints (including horizontal-well information), the focus on continuous model improvement, and issues unique to Hamaca rock
and fluid properties. We will show that efficiently evolving a very
large geocellular model in an active project like Hamaca can be
accomplished through the use of object-oriented process automation. In addition, the paper will illustrate that careful consideration
should be paid to issues related to horizontal-well sampling bias
and positional uncertainty before constraining a geocellular model
with horizontal-well data. The paper will discuss the multiple sophisticated modeling techniques that were used to address the objectives of the Hamaca modeling program.
Background
The Hamaca field is located in Venezuelas Orinoco heavy-oil
belt, which is reported to contain more than 1.2 trillion barrels of
heavy and extra heavy oil in a huge stratigraphic trap on the
southern flank of the Oriente basin (Fig. 1). The Hamaca concession area, which covers 160,000 acres, contains 8 to 10API gravity oil trapped in shallow fluvial-deltaic reservoirs of the Oficina
formation (Miocene age). Sandstone reservoirs of the Oficina formation at Hamaca were generally deposited in a bed-loaddominated, fluvial-deltaic environment. Reservoir properties are
excellent, with porosity values of up to 36% and permeability
values of up to 30 darcies. Hamaca crude is considered foamy
and is generally saturated with gas at reservoir conditions.1
Over the 35-year life of the field, more than 1,000 horizontal
laterals are planned to deliver 190,000 BOPD to a heavy-oil upgrader facility, which is currently under construction.2 To date,
more than 110 horizontal wells have been drilled to produce from
the Hamaca reservoirs. Oil is being produced under cold production methods (no added heat) using progressing cavity pumps to
bring oil to the surface. Cold production is possible because of the
extended length of the horizontal wells (5,000 ft), excellent reservoir properties, and the foamy-oil nature of Hamaca crude.2 The
heavy oil will be mixed with diluent just downstream of the wellheads to facilitate transport to the upgrader facility. The Hamaca
crude will be converted to a sweeter crude product of approximately 26API at the upgrader.
The combined use of both well and seismic data is critically
important for characterizing the stratigraphic complexity of the
Hamaca fluvial-deltaic systems. To assist in targeting sweet spots
for horizontal-well placement, a 250-km2 3D seismic survey was
acquired along with the drilling of 91 stratigraphic information wells
with an average separation distance of approximately 1.5 km.

Copyright 2003 Society of Petroleum Engineers


This paper (SPE 87308) was revised for publication from paper SPE 78957, first presented
at the 2002 SPE International Thermal Operations and Heavy Oil Symposium and International Horizontal Well Technology Conference, Calgary, 47 November. Original manuscript received for review 18 November 2002. Revised manuscript received 29 August
2003. Paper peer approved 8 September 2003.

Modeling Approach
Hamaca reservoir-modeling activity has been motivated by strategic and tactical business drivers that include reservoir sweet-spot
identification for development planning, reserves studies for a contractual acreage-relinquishment decision, and horizontal wellbore
design and steering. The continuous flow of new information from
stratigraphic and horizontal-well drilling programs and the analysis of original and reprocessed vintages of 3D seismic data have
required flexibility in evolving geologic and engineering concepts.
One Size Fits All. An ambitious undertaking began in 2000 to
build a single evergreen reservoir model that satisfied each of
the Hamaca business drivers. The design goals in building such a
model were to maintain a high level of consistency across all the
operational activities in a resource- and time-efficient manner.
Creating a one-size-fits-all model required that the model cover a
large enough area (3733 km) to satisfy strategic activities such
as flow simulation for reserves assessments and sweet-spot identification, and that it be sufficiently detailed to plan and steer the
horizontal wellbores.
In selecting the appropriate model-cell dimensions, the tradeoff
between resolution and computational efficiency was considered.
Some of the factors that weighed on the selection of cell dimensions were:
Sufficient sampling of the expected vertical and horizontal
heterogeneity as defined by variography.
Accuracy in the simulation of pressure drawdown and associated production of gas in the vicinity of wells.
The ability to update the geologic model with new information with a 24-hour turnaround time.
Flow-simulation and history-matching turnaround time.
Hardware and software memory limitations.
After extensive testing, the optimal cell dimensions of 100
100 m in the horizontal and .6 m in the vertical were chosen,
resulting in approximately 38 million cells distributed among 10
major stratigraphic units. For computational flexibility, major
stratigraphic units were built as separate 3D grids (SGrids) and
then merged as needed for flow simulation and well-planning and
-drilling activities. Efficiently evolving such a large model as new
information became available posed unique challenges that required a creative approach to the model-building process.
Process Automation. The continuous inflow of new information
from the stratigraphic-well drilling program, horizontal-well drilling program, and seismic interpretation required that Ameriven
focus effort on building a reservoir-modeling process that is efficient, consistent, and repeatable. These goals were achieved by the
use of automation.
Model-building automation was accomplished by the use of
process scripting within an object-oriented modeling system.
Scripts are files containing a sequence of commands that modify
object properties and control object-to-object interaction. Nested
layers of scripts were written to manage the entire model-building
process including data loading, well-log processing, framework
construction, region definition, reservoir parameter population,
generation of numerous quality-control products, and export of the
model to the simulator (Fig. 2).
With such a large model, many of the model-building functions
require hours to complete; the use of scripting allows those processes to continue running overnight without the need for intervention. For example, when information from a recently drilled

December 2003 SPE Reservoir Evaluation & Engineering

421

www.petroman.ir

Fig. 1Location of Hamaca field in the Orinoco heavy-oil belt of


Venezuela.

well indicates the need for revision of a structural surface, the


surface can be modified and the entire model updated for use the
next day.
Property Modeling
One of the greatest challenges in computer-based model building
is to parameterize and then incorporate the available geologic interpretative data, concepts, and insights. Often, the most difficult
challenge is to capture this conceptual information in a form that
is compatible with the rigid constructs of computer model building. For the Hamaca modeling effort, regional net sand maps generated from the interpretation of seismic and well data were available. Each map embodies the most current Ameriven concepts
regarding source direction, sand fairways, and predominant depositional facies types in the model area. Hence, the model was
constructed in a way that would incorporate these maps as key
drivers in the horizontal and vertical distribution of the reservoir.
Reservoir Distribution. A technique called sequential Gaussian
simulation with block kriging (SGSBK) was chosen as the ideal
method for integrating the available net sand information because
of its ability to incorporate a 2D interval-type map as a soft constraint on the 3D interpolation of well-log point data.3 SGSBK is
an SGS technique that treats a 2D map as a soft estimate of the
column averages in the 3D model to be populated. A major
strength of the technique is that it explicitly addresses the differences in variance caused by differences in volume support between
the 2D average constraint map and the well-log-scale point data.
With SGSBK, the resulting model contains the vertical variability

Fig. 3An image showing a flattened SGrid (bottom) painted


with VSH and the corresponding net sand map used to constrain the population of VSH (top). Note the correspondence
between channel fairways in the net sand map (warmer colors)
and the sandier areas of the model.

Fig. 2A diagram illustrating the use of scripting for efficient


and repeatable model updating. With automation, the 38million-cell model can be updated within the design goal of less
than 24 hours.

driven by the well-log data and the vertical variogram and


the horizontal variability driven by the 2D average map and horizontal variogram.
To use the net sand maps to drive model reservoir distribution,
the net sand maps must first be transformed into average volumeof-shale (VSH) maps. This is accomplished by computing net-togross maps from the net sand and interval isopach maps. The
net-to-gross maps are then scaled to pseudoaverage shale-volume
maps using a linear transform developed from well-log analysis.
SGSBK is used to populate the model with shale volume using the
pseudoaverage VSH maps as a soft constraint. Fig. 3 illustrates the
high degree of correlation between the constraining net sand map
and the resulting model VSH distribution.
Effective porosity was modeled as a function of rock type,
well-log data, and model VSH. First, three petrophysical facies are
created by applying cutoffs to the model VSH property. The cutoffs are selected to differentiate the predominately channel, splay,
and nonreservoir depositional rock types found in Hamaca. Each
of these facies regions is then populated with porosity independently using SGS with a locally varying mean (SGSLVM). In this
technique, a porosity-trend model is created by linearly scaling
model VSH to porosity using a transform generated from well-log
analysis. The residuals (differences) between the model porosity
trend and the well-log porosities are distributed using SGS and the
residual porosity variogram. The distributed residuals are added to
the trend to produce a porosity field that is softly constrained
by VSH.
The advantages of running separate region-based passes of
SGSLVM are:
The distinctly different probability density functions (PDFs)
and bivariate relationships in each region are honored
more closely.
The normal-score forward and reverse transforms that are
performed with SGS are more robust because we avoid transforming the skewed, bimodal property distribution associated with a
single region.
Each facies region can be assigned a model of spatial continuity (variogram model) that is appropriate for its rock type.
The Hamaca reservoirs are very permeable owing to their wellsorted, loosely consolidated nature. Measurements from continuous cores have shown that air permeabilities range up to 30 darcies. These elevated permeabilities play a critical role in allowing
high rates of cold production from the reservoirs. Analysis of core
measurements indicates that permeability can be modeled with
satisfactory accuracy as a function of grain size (VSH) and porosity. Fig. 4 shows the porosity/permeability crossplot color coded
with well-log-matched VSH. Permeability was populated into the

422

December 2003 SPE Reservoir Evaluation & Engineering

www.petroman.ir

Fig. 5(a) Illustration showing an example oil/water contact


surface and the oil region and water region; (b) structural
cross section from the model showing Sw property and oil/water
contact.
Fig. 4Porosity-to-permeability relationship based on cores.
Separate transforms were used for VSH <= 0.20 and for VSH >=
0.30, with interpolation for values between 0.20 and 0.30.

model as a function of model porosity and VSH using the relationship shown in Fig. 4.
Fluid Distribution. A thorough understanding of water distribution in a heavy-oil project like Hamaca is imperative to avoid
earlier-than-expected water encroachment and the operational
costs associated with remediation (water shutoff). Because the
specific gravities of Hamaca oil and water are very similar (.965
vs. 1.01 g/cm3), the small difference in buoyancy results in poncheros, or perched pockets of water above the oil/water contacts.

Oil/water contact depths are highly variable because of the small


difference between the net oil/water buoyancy pressure and the
capillary displacement pressure across semipermeable and permeable faults.4 If not avoided in drilling, water will be produced
preferentially because of the large difference in viscosity between
the heavy oil and water. Therefore, accurate modeling of water
distribution is critical to the operational success of the project.
Water is distributed in the Hamaca model as a function of
well-log data, model facies, model VSH, and structural position
relative to the interpreted oil/water contacts. Ten oil/water contact
maps, which represent the base of the oil/water transition zone in
each stratigraphic interval, are used to create regions above and
below the contacts (Fig. 5). For the water region, or the area
below the oil/water contact, water saturation (Sw) is set to 1. For
the oil region, Sw is populated from well-log data using
SGSLVM. As with porosity, Sw is populated into each petrophysical-facies region separately, using VSH as a soft constraint to
better control the distribution.
By populating oil and water regions with Sw separately, low Sw
values from wells in the oil region are not interpolated into cells
below the oil/water contact and, conversely, high Sw values from wet
wells are not carried updip into cells above the oil/water contact.
Heat Distribution. In many reservoir-modeling projects, temperature variation for a constant depth is small and can be neglected in flow simulation without significantly impacting performance. In Hamaca, however, large variations in vertical temperature gradient do exist. Gradients calculated from formation
temperature logs measured in stratigraphic wells (at static conditions) range from .12 to .18 F/ft (Fig. 6a). At a constant reservoir
depth of 2,500 ft below surface, this variation in gradient corresponds to an approximately 15F change in temperature. Lab measurements of Hamaca oil samples indicate that a 10F increase in
reservoir temperature can result in a 50% reduction in viscosity,
underscoring the impact of temperature on viscosity and, ultimately, recovery. Therefore, it is important to model these observed temperature variations accurately.
The Hamaca model was populated with temperature by first
mapping the spatial variation in gradient measured from temperature logs (Fig. 6b). Each cell column of the model was then assigned a temperature gradient from the map. The depth of overburden was calculated for each model cell using remote-sensing
image data to account for topography above datum (mean sea
level, or MSL). The temperature of each model cell was estimated
from the associated depth of overburden and gradient properties.
The resulting 3D model of temperature (Fig. 6c) is input to
the fluid-flow simulation to more accurately model variations in
oil viscosity.

Fig. 6(a) Crossplot of measured temperatures vs. depth and


line fits used to calculate gradients; (b) map of temperature
gradients; (c) temperature volume.

Multiple Scales of Heterogeneity. Multiple scales of heterogeneity generally exist to a degree in most depositional settings. In
Hamacas fluvial-deltaic depositional environment, dual scales of
variability are clearly evident in variograms of reservoir properties.
Long correlation ranges in these variograms are related to large-

December 2003 SPE Reservoir Evaluation & Engineering

423

www.petroman.ir

Fig. 7Illustration showing dual scales of continuity evident in


Hamaca variograms.

scale variations in the form of channel complexes that are developed in some areas of the field but are absent in others. Within
these overall channel complexes, individual channels anastomose
and downcut into previously existing channels, with the resulting
stratigraphy showing discontinuous channel features (bars) with
shorter scale variability (Fig. 7). Short correlation ranges in
Hamaca variograms are related to these individual channel features. The different scales of variability were modeled using nested
(linearly combined) variograms. The relating of structures observed in experimental variograms with scale-dependent geologic
processes resulted in a more geologic interpretation and modeling
of reservoir spatial continuity.
Variable Azimuth. A unique aspect of the Hamaca modeling
approach is the extensive use of the variable-azimuth (VA) technique for populating model properties.5 In most modeling projects,
reservoir properties are usually distributed into a geocellular
framework using a variogram model describing a single direction
of maximum anisotropic correlation. For the Hamaca model, the
azimuth of greatest continuity was allowed to vary geographically
so that the trend of the reservoir properties would be oriented
similarly to the overall channel-complex trends that have been
interpreted. The result is greater model pore-space connectivity
and a more realistic representation of the geology (Fig. 8).
Incorporating Horizontal-Well Data
Three-dimensional geocellular models are routinely used in many
development projects to justify and plan horizontal wells. How-

Fig. 9VSH histograms from one stratigraphic interval including (lower left) and excluding (upper left) horizontal-well
samples. Inclusion of the biased and overrepresented horizontal-well samples skews the VSH property distribution toward
cleaner rock. On the right is a net sand map showing the location of wells used in the histograms.

Fig. 8Net sand map (left) used as a constraint for VSH population and resulting VSH property for a single layer of the model
using the variable-azimuth technique (middle) and using a constant variogram azimuth (right). Note the higher degree of sand
continuity and better consistency with the net sand map that
results from using the variable-azimuth technique.

ever, it is far less common to use the results of horizontal-well


drilling to update a geocellular model. The Hamaca project is
unique in this respectas the project progresses, a very large
number of horizontal wells will be available to inform the evolving
geocellular model. These horizontal-well data provide important
information about short-scale lateral heterogeneity that is not obtainable from other sources of information. However, careful consideration should be paid to several important issues that are
unique to the incorporation of horizontal-well data.
Sample Debiasing of Horizontal-Well Data. One of the design
goals of stochastic simulation techniques such as SGS is to generate property realizations that are consistent with an a priori
global probability density function (PDF). It is common practice to
assume that the hard data PDF from well logs is representative
of the global model PDF (stationarity assumption). Often, this
assumption is not satisfied to a sufficient degree because of biases
in well sampling. For example, preferential drilling of reservoir
sweet spots can skew the estimated global PDF, resulting in property simulations with overestimated reservoir volumes and continuity. The geostatistical modeler must decide whether the stationarity assumption is satisfied to a sufficient degree and, if it is not,
take measures to calculate a debiased estimate of the global property PDF for stochastic simulation.
Before using horizontal-well data as input to stochastic simulation, the effects of horizontal sampling on the global property
PDF should be considered. A horizontal well provides a much
denser sampling of rock heterogeneity than a vertical well because
of horizontal stratification (i.e., layered rock strata are more continuous laterally than they are vertically). Because the intention of
drilling a horizontal well is to drill as much high-quality reservoir
rock as possible, the dense sampling from horizontal-well logging
is generally biased toward better reservoir rock properties. When
these samples are combined with samples from vertical wells to
estimate the global PDF, the biased and overrepresented horizontal-well data will optimistically skew the property PDF. In the case
of Hamaca modeling, the inclusion of horizontal-well data in the
estimate of the PDF for VSH results in 10% overestimation of
model reservoir volume, even though the horizontal-well drilling
to date has been limited to only a small fraction of the model area
(Fig. 9).
Several approaches were tested to address the horizontal sampling bias. They all have in common the goal of reducing the
influence of the overabundant and biased horizontal-well-log
samples in the calculation of the global property PDF. These included assigning declustering weights based on the closeness of
samples,6 resampling the horizontal-log data to a coarser sampling
interval based on variography, and ignoring the horizontal-log data
altogether (for purposes of global PDF estimation only). Each of
these approaches was found to address the problem effectively. For
purposes of implementation convenience related to the modeling
software being used, we chose to resample the horizontal-well-log
data. This reduced sample rate was selected to sample lateral rock

424

December 2003 SPE Reservoir Evaluation & Engineering

www.petroman.ir

heterogeneity (as measured by the horizontal variogram range) at


roughly the same rate as a vertical well samples vertical rock
heterogeneity (as measured by the average vertical variogram).
In our case, this required a 200 to 1 reduction in horizontal
sample rate.
Variogram Analysis Using Horizontal-Well Data. Careful consideration is also required to use horizontal-well data in variogram
analysis. As with the calculation of the global property PDF,
sample bias can skew the results of variography. If the goal of
horizontal wells is to drill continuous reservoir rock, then the
horizontal variogram range will be overestimated, and the resulting
geocellular model could be overly continuous.
Conversely, the horizontal variogram range can be underestimated because of wellbore and model positional uncertainty. This
occurs because horizontal variogram analysis is generally performed in a stratigraphic space defined by the model framework.
Relative structural misalignment between the wellbore and the
model framework will result in the horizontal variography confusing vertical heterogeneity for horizontal heterogeneity (Fig.
10). Because it is difficult to tie model structural surfaces to horizontal-well-log data, a large amount of relative positional uncertainty can exist. Because of this uncertainty, variogram analysis for
the Hamaca model was restricted to using only the build sections
of the horizontal wells, which are located with a greater degree of
certainty. A large number of build sections (51) were available for
variogam analysis and were key to defining short-scale reservoir
variability and recognizing multiple scales of heterogeneity (as
discussed earlier in this paper).
Conclusions
1. Efficiently evolving a very large geocellular model in an active
project like Hamaca can be accomplished through the use of
object-oriented process automation.
2. Heat is an important element in understanding flow behavior in
the Hamaca field because of its control of oil viscosity. Therefore, accurately modeling temperature variations in 3D geocellular space is critical.

3. The relating of structures observed in experimental variograms


with scale-dependent geologic processes resulted in a more geologic interpretation and modeling of reservoir spatial continuity.
4. Careful consideration should be paid to issues related to horizontal-well sampling bias and positional uncertainty before including horizontal-well data in a geocellular model.
5. Multiple sophisticated modeling techniques were used to address the objectives of the Hamaca modeling program.
Acknowledgments
The authors would like to thank PDVSA, ConocoPhillips, ChevronTexaco, and Petrolera Ameriven for permission to publish the
material in this paper.
References
1. Maini, B.B.: Foamy-Oil Flow, JPT (October 2001) 54.
2. Gipson, L.J., Owen, R., and Robertson, C.R.: Hamaca Heavy Oil
ProjectLessons Learned and an Evolving Development Strategy,
paper SPE 78990 presented at the 2002 SPE International Thermal
Operations and Heavy Oil Symposium and International Horizontal
Well Technology Conference, Calgary, 47 November.
3. Behrens, R.A. et al.: Incorporating Seismic Attribute Maps in 3D
Reservoir Models, paper SPE 36499 presented at the 1996 SPE Annual Technical Conference and Exhibition, Denver, 69 October.
4. Watts, N.L.: Theoretical Aspects of cap-rock and fault seals for singleand two-phase hydrocarbon columns, Marine and Petroleum Geology
(November 1987) 4, 274.
5. Xu, W.: Conditional Curvilinear Stochastic Simulation Using PixelBased Algorithms, Mathematical Geology (1996) 28, No. 7, 937.
6. Deutsch, C.V.: DECLUS: A Fortran 77 program for determining optimum spatial declustering weights, Computers & Geosciences (1989)
15, No. 3, 325.

SI Metric Conversion Factors


acre 4.046 873
E + 03
bbl 1.589 873
E 01
F (F32)/1.8
ft 3.048 *
E 01
E + 01
in.3 1.638 706
mile 1.609 344*
E + 00
sq mile 2.589 988
E + 00

m2
m3
o
C
m
cm3
km
km2

*Conversion factor is exact.

Fig. 10(a) Schematic cross section with the horizontal wellbore positioned correctly relative to model framework, resulting
in correct variogram analysis; (b) schematic cross section with
the horizontal wellbore not positioned correctly relative to
model framework, resulting in an incorrect model of variability.

Terrell H. Tankersley is a senior development geologist specializing in reservoir modeling for the Hamaca Project in Puerto La
Cruz, Venezuela. e-mail: tankersley@chevrontexaco.com.
Since 1981, he has held various positions with ChevronTexaco,
focused primarily on reservoir modeling and development geology. He has worked in numerous basins in Indonesia, Venezuela, and the Gulf of Mexico. Tankersley holds a BS degree
in geology from the U. of Georgia. Michael W. Waite currently
manages reservoir-modeling and flow-simulation activities for
the Hamaca Project in Venezuela. e-mail: waitemw@
chevrontexaco.com. Since 1982, he has held positions with
ChevronTexaco involving seismic acquisition and processing,
development and application of subsurface imaging techniques, prospect evaluation, reservoir characterization using
3D/4D seismic data, and geostatistical model building. He
holds a BS degree in physics from the U. of New Orleans. Waite
served as a 199798 SPE Distinguished Lecturer on the topic of
time-lapse seismic monitoring.

December 2003 SPE Reservoir Evaluation & Engineering

425

www.petroman.ir

You might also like