Simulation Throughout
the Life of a Reservoir
Gordon Adamson
Reservoir Management Ltd.
Aberdeen, Scotland
Martin Crick
Texaco Ltd.
London, England
Brian Gane
British Petroleum
Aberdeen, Scotland
Omer Gurpinar
Denver, Colorado, USA
Jim Hardiman
Henley on Thames, England
Dave Ponting
Abingdon, England
For help in preparation of this article, thanks to Bob
Archer, Chip Corbett, Ivor Ellul, Roger Goodan and Jim
Honefenger, GeoQuest, Houston, Texas, USA; Randy
Archibald, GeoQuest Reservoir Technologies, Henley on
Thames, England; Ian Beck, GeoQuest Reservoir Technologies, Abingdon, England; George Besserer,
PanCanadian Petroleum Limited, Calgary, Alberta,
Canada; Kunal Dutta-Roy, Simulation Sciences Inc.,
Brea, California, USA; and Sharon Wells, GeoQuest
Reservoir Technologies, Denver, Colorado.
ECLIPSE, FloGrid, GRID, Open-ECLIPSE, PVT and
RTView are marks of Schlumberger. NETOPT and
PIPEPHASE are marks of Simulation Sciences Inc.
1. Peaceman DW: “A Personal Retrospection of Reservoir Simulation,” Proceedings of the First and Second
International Forum on Reservoir Simulation, Alpbach,
Austria, September 12-16, 1988 and September 4-8,
1989.
2. Wycoff RD, Botset HG and Muskat M: “The Mechanics of Porous Flow Applied to Water-flooding Problems,” Transactions of the AIME 103 (1933): 219-249.
Muskat M and Wyckoff RD: “An Approximate Theory
of Water-Coning in Oil Production,” Transactions of
the AIME 114 (1935): 144-163.
3. Darcy’s law states that fluid flow velocity is proportional to pressure gradient and permeability, and
inversely proportional to viscosity.
4. Coats KH: “Use and Misuse of Reservoir Simulation
Models,” SPE Reprint Series No. 11 Numerical Simulation. Dallas, Texas, USA: Society of Petroleum Engineers (1973): 183-190.
16
Simulation is one of the most powerful tools for guiding reservoir
management decisions. From planning early production wells and
designing surface facilities to diagnosing problems with enhanced
recovery techniques, reservoir simulators allow engineers to
predict and visualize fluid flow more efficiently than ever before.
Reservoir simulators were first built as diagnostic tools for understanding reservoirs that
surprised engineers or misbehaved after
years of production. The earliest simulators
were physical models, such as sandboxes
with clear glass sides for viewing fluid flow,
and analog devices that modeled fluid flow
with electrical current flow.1 These models,
first documented in the 1930s, were constructed by researchers hoping to understand water coning and breakthrough in
homogeneous reservoirs that were undergoing waterflood.2
Some things haven’t changed since the
1930s. Today’s reservoir simulators generally
solve the same equations studied 60 years
ago—material balance and Darcy’s law.3
But other aspects of simulation have
changed dramatically. With the advent of
digital computers in the 1960s, reservoir
modeling advanced from tanks filled with
sand or electrolytes to numerical simulators.
In numerical simulators, the reservoir is represented by a series of interconnected
blocks, and the flow between blocks is
solved numerically. In the early days, computers were small and had little memory,
limiting the number of blocks that could be
used. This required simplification of the
reservoir model and allowed simulation to
proceed with a relatively small amount of
input data.
As computer power increased, engineers
created bigger, more geologically realistic
models requiring much greater data input.
This demand has been met by the creation
of increasingly complex and efficient simulation programs coupled with user-friendly
data preparation and result-analysis packages. Today, desktop computers may have
5000 times the memory and run about 200
times faster than early supercomputers.
However, the most significant gain has not
been in absolute speed, but speed at a moderate price. Computational efficiency has
reached a stage that allows powerful simulators to be run frequently.
Numerical simulation has become a reservoir management tool for all stages in the life
of the reservoir. No longer just for comparing
performance of reservoirs under different
production schemes or trouble-shooting
when recovery methods come under
scrutiny, simulations are also run when planning field development or designing measurement campaigns. In the last 10 years,
with the development of computer-aided
geological and geostatistical modeling, reservoir simulators now help to test the validity
of the reservoir models themselves. And simulation results are increasingly used to guide
decisions on investing in the construction or
overhaul of expensive surface facilities.
Motivation for Simulation
A numerical simulator containing the right
information and in the hands of a skilled
engineer can imitate the behavior of a reservoir. A simulator can predict production
under current operating conditions, or the
reaction of the reservoir to changes in conditions, such as increasing production rate;
production from more or different wells;
response to injection of water, steam, acid
Oilfield Review
Core plugs
Whole cores
Borehole geophysics
Well logs
Outcrop studies
Well testing
3D Seismic data
Large-scale structure
Geological expertise
Small-scale structure
1st generation geomodel
or foam; the effect of subsidence; and production from horizontal wells of different
lengths and orientations.
Reservoir simulation can be performed by
oil company reservoir engineers or by engineering consultant contractors. Some contractors specialize in engineering consulting,
while others offer a full range of oilfield services. In either case, the simulator is a tool
that allows the engineer to answer questions
and offer recommendations for improving
operating practice.
To make simulation worthwhile, there must
be a well-posed question of economic
importance: Where should wells be located
to maximize incremental recovery per dollar
of additional investment? How many wells
are required to produce enough gas to meet
a contractual deliverability schedule? Should
oil be recovered by natural depletion or
water injection? What is the optimum length
of a horizontal well? Is carbon dioxide [CO2]
injection feasible? Should we keep this reservoir alive? As observed by K.H. Coats while
at the University of Texas at Austin, USA,
“The complexity of the questions being
asked, and the amount and reliability of the
data available, must determine the sophistication of the system to be used.”4 In all
cases, a simulation study should result in
recommendations for intervention. This may
include a new strategy for data acquisition,
or an infill drilling plan with the number,
location and direction of wells and a completion strategy for each well.
How a Simulator Works
Calibration
Risk analysis
Surface network
input
Production
Static reservoir model
Up-gridding
Simulation model
Execution model
■ Creating models for input to reservoir simulators. The first-generation geomodel is created through the combined efforts of geologists, geophysicists, petrophysicists and
reservoir engineers. Reservoir properties are then upscaled to produce the static reservoir model. Optimizing the grid and calibrating with dynamic data yield the simulation
model. Finally, input from surface facilities analysis and risk calculations results in an
execution model that can guide reservoir management decisions.
Summer 1996
The function of reservoir simulation is to
help engineers understand the productionpressure behavior of a reservoir and consequently predict production rates as a function of time. The future production
schedule, when expressed in terms of revenues and compared with costs and investments, helps managers determine both economically recoverable reserves and the limit
of profitable production.
Once the goal of simulation is determined,
the next step is to describe the reservoir in
terms of the volume of oil or gas in place,
the amount that is recoverable and the rate
at which it will be recovered. To estimate
recoverable reserves, a model of the reservoir framework, including faults and layers
and their associated properties, must be
constructed. This so-called static model is
created through the combined efforts of
geologists, geophysicists, petrophysicists and
reservoir engineers (left ). Much of the multibillion-dollar business of oilfield services is
centered on obtaining information that
17
eventually feeds reservoir simulators, leading to better reservoir development and
management decisions.5
The simulator itself computes fluid flow
throughout the reservoir. The principles
underlying simulation are simple. First, the
fundamental fluid-flow equations are
expressed in partial differential form for
each fluid phase present. These partial differential equations are obtained from the
conventional equations describing reservoir
fluid behavior, such as the continuity equation, the equation of flow and the equation
of state. The continuity equation expresses
the conservation of mass. For most reservoirs, the equation of flow is Darcy’s law.
For high rates of flow, such as in gas reservoirs, Darcy’s law equations are modified to
include turbulence terms. The equation of
state describes the pressure-volume or pressure-density relationship of the various fluids present. For each phase, the three equations are then combined into a single partial
differential equation. Next, these partial differential equations are written in finite-difference form, in which the reservoir volume
is treated as a numbered collection of
blocks and the reservoir production period
is divided into a number of time steps.
Mathematically speaking, the problem is
discretized in both space and time.
Examples of simulators that solve this
problem under a variety of conditions are
found in the ECLIPSE family of simulators.
These simulators fall into two main categories. In the first category are three-phase
black-oil simulators, for reservoirs comprising water, gas and oil. The gas may move
into or out of solution with the oil. The second category contains compositional and
thermal simulators, for reservoirs requiring
more detailed description of fluid composition. A compositional description could
encompass the amounts and properties of
hexanes, pentanes, butanes, benzenes,
asphaltenes and other hydrocarbon components, and might be used when the fluid
composition changes during the life of the
reservoir. A thermal simulator would be
advised if changes in temperature—either
with location or with time—modified the
fluid composition of the reservoir. Such a
description could come into play in the case
of steam injection, or water injection into a
deep, hot reservoir.
18
Block-Centered Geometry
0
2000
4000
6000
8000
4000
6000
8000
5800
6200
6600
7000
7400
Corner-Point Geometry
0
2000
5800
6200
■ Block-centered
and corner-point
geometries. Blockcentered geometry
features flattopped rectangular
blocks that match
the mathematical
models behind the
simulator. Cornerpoint geometry
modifies the rectilinear grid so that
it conforms to
important reservoir
boundaries. Threedimensional grids
are constructed
from a 2D grid by
laying it on the top
surface of the
reservoir and projecting the grid
vertically or along
fault planes onto
lower layers.
6600
7000
7400
Local Grid Refinement
■ Local grid refinement (LGR). Local
grid refinement
allows engineers to
describe selected
regions of the reservoir in extra detail.
Radial refined grids
are often used
around wellbores to
examine coning or
other phenomena
resulting from rapid
variation in properties away from the
well. Refined grids
are also one way to
treat property variations near faults.
Oilfield Review
These and all other commercial reservoir
simulators envision a reservoir divided into
a number of individual blocks, called grid
blocks. Each block corresponds to a volume
in the reservoir, and must contain rock and
fluid properties representative of the reservoir at that location. The simulator models
the flow of mobile fluid through the walls of
the blocks by solving the fluid-flow equations at each block face. Parameters
required for the solution include permeability, layer thickness, porosity, fluid content,
elevation and pressure. The fluids are
assigned a viscosity, compressibility, solution gas/oil ratio and density. The rock is
assigned a value for compressibility, capillary pressure and a relative permeability
relationship.
Creating the grid and assigning properties
to each grid block are time-consuming tasks.
The framework of the reservoir, including its
structure and depth, its layer boundaries and
fault positions and throws, is obtained from
seismic and well log data. The well-bred grid
respects the framework geometry as much as
possible. Traditionally, reservoir simulation
grid blocks are rectilinear with flat, horizontal tops in an arrangement called block-centered geometry (previous page, top). This
configuration ensures that the grids remain
orthogonal and exactly match the mathematical models used in the simulators.
However, this approach does not easily
represent structural and stratigraphic complexities such as nonvertical faults, pinchouts or erosional surfaces using purely
rectangular blocks. The 1983 introduction
of corner-point geometry in the ECLIPSE
simulator overcame these problems. In a
corner-point grid, the corners need not be
orthogonal. In modeling a faulted reservoir,
for example, engineers have the flexibility to
choose between an orthogonal areal grid
with the fault positions projected onto the
grid or a flexible grid to exactly honor the
positions of important faults. Three-dimensional (3D) grids are constructed from an
areal, or 2D, grid by laying it on the top surface of the reservoir and projecting it vertically or along fault planes onto lower layers.
Engineers’ requirements for more detail in
the model, particularly to examine coning
and near-wellbore effects, has led to the
concept of local grid refinement (LGR) (previous page, bottom ). This allows parts of the
model to be represented by a large number
of small grid blocks or by implanting radial
Summer 1996
grids around wells in a larger Cartesian
grid. 6 Locally refined grids also capture
extra detail in other areas where reservoir
properties vary rapidly with distance, such
as near faults. And LGR, combined with grid
coarsening outside the region of interest,
allows engineers to retain fine-scale property variation without surpassing computer
space limitations. The interactive GRID program was designed to help construct the
complex reservoir grid efficiently (see
“Developments in Gridding,” page 21 ).
Once the grid has been constructed, the
next step is to assign rock and fluid properties from the reservoir framework model to
each grid block. Populating the grid with
properties is another time-consuming and
difficult task. Each grid block, typically a
few hundred square meters areally by tens
of meters thick, has to be assigned a single
value for each of the reservoir properties,
including fluid viscosity, relative permeability, saturation, pressure, permeability, porosity and net-to-gross ratio. 7 Log measurements made in wells yield high-density
data, typically every 6 in. [15 cm], but provide little information between wells. Data
from cores may provide high-density
“ground truth,” but these represent perhaps
one part in 5 billion of the volume of the
reservoir. Surface seismic reflections cover
the reservoir volume and more, but do not
translate directly into the desired rock and
fluid properties. How are these disparate
data sets merged?
Two processes are required: extrapolating
the well data into the interwell reservoir volume, then upscaling the fine-scale data to
the scale of a simulation grid block. Traditionally log or core data were upscaled, or
averaged, over lithological units at the wells.
Then these data were interpolated and
extrapolated through the reservoir and maps
produced for each layer—formerly a handdrafting exercise by geologists. The maps
would be passed to the reservoir engineer
who would then generate grids, run preliminary simulations on a series of grid sizes,
and attempt further upscaling based on the
reservoir flow characteristics.
In recent years, the process has been
reversed. The current trend is to use computer programs to build 3D geological models bounded by seismic data, and to populate the models using geostatistical or
deterministic methods to distribute log and
core data.8
Scaling core and log properties up to gridblock scales is still a challenging task. Some
properties, such as porosity, are considered
simple to upscale, following an arithmetic
averaging law. Others, such as permeability,
are more difficult to average. And relative
permeabilities—different permeabilities for
different fluid phases—remain the most difficult problem in upscaling. There is no universally accepted method for upscaling, and
it is an area of active research.9
After the model has been finalized, the
simulator requires boundary conditions to
establish the initial conditions for fluid
behavior at the beginning of the simulation.
Then, for a given time later, known as the
time step, the simulator calculates new pressures and saturation distributions that indicate the flow rates for each of the mobile
phases. This process is repeated for a number of time steps, and in this manner both
flow rates and pressure histories are calculated for each point—especially the points
corresponding to wells—in the system.
But even with the best possible model,
uncertainty remains. One of the biggest jobs
5. For specific examples: Bunn G, Cao Minh C, Roestenburg J and Wittman M: “Indonesia’s Jene Field: A
Reservoir Simulation Case Study,” Oilfield Review 1,
no. 2 (July 1989): 4-14.
Briggs P, Corrigan T, Fetkovich M, Gouilloud M, Lo
Tien-when, Paulsson B, Saleri N, Warrender J and
Weber K: “Trends in Reservoir Management,”Oilfield
Review 4, no. 1 (January 1992): 8-24.
Corbett P, Corvi P, Ehlig-Economides C, Guérillot D,
Haldorsen H, Heffer K, Hewitt T, King P, Le Nir I,
Lewis J, Montadert L, Pickup G, Ravenne C, Ringrose
P, Ronen S, Schultz P, Tyson S and Verly G: “Reservoir
Characterization Using Expert Knowledge, Data and
Statistics,”Oilfield Review 4, no. 1 (January 1992):
25-39.
Al-Rabah AK, Bansal PP, Breitenback EA, Hallenbeck
LD, Meehan DN, Saleri NG and Wittman M: “Exploring the Role of Reservoir Simulation,” Oilfield Review
2, no. 2 (April 1990): 18-30.
6. For more on local grid refinement: Heinemann ZE
and von Hantelmann G: “Using Local Grid Refinement in a Multiple-Application Reservoir Simulator,”
paper SPE 12255, presented at the Reservoir Simulation Symposium, San Francisco, California, USA,
November 15-18, 1983.
Forsyth PA and Sammon PH: “Local Mesh Refinement
and Modelling for Faults and Pinchouts,” paper SPE
13524, presented at the Reservoir Simulation Symposium, Dallas, Texas, USA, February 10-13, 1985.
7. Net-to-gross ratio, sometimes called just net to gross
(NTG), is the ratio of the thickness of pay to the total
thickness of the reservoir interval.
8. For examples of the technique: Schultz PS, Ronen S,
Hattori M, Mantran P and Corbett C: “Seismic-Guided
Estimation of Log Properties,” The Leading Edge 13,
no. 7 (July 1994): 770-776.
Caamano E, Corbett C, Dickerman K, Douglas D, Gir
R, Martono D, Mathieu G, Nicholson B, Novias K,
Padmono J, Schultz P, Suroso S, Thornton M and Yan
Z: “Integrated Reservoir Interpretation,” Oilfield
Review 6, no. 3 (July 1994): 50-64.
9. Thibeau S, Barker JW and Souillard P: “Dynamical
Upscaling Techniques Applied to Compositional
Flows,” paper SPE 29128, presented at the 13th SPE
Symposium on Reservoir Simulation, San Antonio,
Texas, USA, February 12-15, 1995.
19
Preproduction Planning
8674.00
■ Visualizing the reservoir model in 3D. Visualization is a reliable means of checking
reservoir models before input to a simulator. Inconsistencies in model parameters
may be flagged and corrected. After simulation, results may also be viewed, allowing
faster evaluation of comparative simulation runs and providing insight into recovery
behavior. In this example reservoir pressure is color-coded to show regions of high
and low pressure.
of a simulator is to evaluate the implications
of uncertainty in the static reservoir model.
Sometimes uncertainty or error is introduced through low data quality. Another
source of error arises because laboratory,
logging and geophysical experiments may
not directly measure the property of interest,
or at the right scale, and so some other
property is measured and transformed in
some way that adds uncertainty. There is
also uncertainty in how a property varies
between measurement points. Many reservoir descriptions rely on core sample measurements for rock and fluid property information. This information is uncertainly
extended through the reservoir volume, usually in some geostatistical or deterministic
fashion, guided by seismically derived surfaces or other geological constraints.
One way to reduce uncertainty is to spot
inconsistencies in the properties of the reservoir model before simulation. Three-dimensional visualization software, such as the
RTView application, helps engineers be
more efficient in finding inconsistencies by
allowing them to view the reservoir model in
3D. Results of simulation runs may also be
viewed, allowing faster evaluation of simulation runs and providing immediate insight
into recovery behavior and physical processes occurring in the reservoir (above ).
20
A simulation run itself can also help
reduce uncertainty. Outside the oil industry,
simulators are used to determine the reaction of a known environment to externally
applied perturbations. An example is a flight
simulator that tests varying visibility conditions. Although a reservoir environment is
largely unknown, simulators can help
improve the description. In a process known
as history matching, reservoir production is
simulated based on the existing, though
uncertain, reservoir description. That
description is adjusted iteratively until the
simulator is able to reproduce the observed
pressures and multiphase flow resulting
from applied perturbations—that is, the
known production and injection. If the production history can be matched, the engineer has greater confidence that the reservoir description will be a useful, predictive
tool. The history-matching process is timeconsuming and requires considerable skill
and insight, but is a necessary prerequisite
to the successful prediction of continued
reservoir performance.
These new techniques and programs for
loading data, computing simulations and
viewing results are allowing engineers to use
simulators to guide reservoir management
decisions throughout the life of many fields.
The following case studies highlight some of
the uses of simulators in four different stages
of reservoir maturity.
Forties e
pipelin
Forties
Everest
Lomond
Aberdeen
Erskine
elin
e
Pressure, psi
pip
6250.13
An example of early use of simulation
comes from the Texaco Erskine Project in
the North Sea Central Graben region
(below ). The Erskine field comprises four
high-pressure, high-temperature (HPHT)
condensate reservoirs, and will be the first
HPHT field in the North Sea to come on
line when production commences in 1997.
Production will be from an unmanned
platform, with a multiphase pipeline to the
Amoco Lomond Platform for separation.
Gas will be exported via the Central Area
Transmission System (CATS) pipeline, and
liquids via the Forties pipeline. Initial production with be from three wells, with three
more to be added. The production mechanism will be natural depletion, with no gas
recycling. Other operators in the region who
have similar reservoirs to develop are
watching how Texaco handles the hostile,
overpressured field.
Simulation was selected as a way to
predict production of gas for drawing up
deliverability contracts—contracts promising delivery of designated volumes of gas at
a specified time. The main challenge in simulating these reservoirs is accounting for
both the permeability reduction due to rock
compaction and the productivity loss due to
condensate banking—explained below—in
the near-wellbore region of the formation
when the reservoir pressure falls below the
dewpoint pressure.10
CA
TS
RTView 96A
N
UK
■ Texaco Erskine Project in the North Sea
Central Graben region. The high-temperature, high-pressure condensate field is
due to go on production in 1997.
Oilfield Review
Because of overpressure conditions in the
reservoir, the rock is expected to compact
with depressurization. This means the rock
is expected to decrease its porosity and
effective permeability as production progresses. To quantify these effects, laboratory
experiments were conducted on rock samples. The experiments showed that at the
assumed well abandonment pressure of
4000 psi, permeability would be reduced by
about 33% from the initial value, while
porosity would be negligibly reduced.
Modeling flow in condensate reservoirs
requires additional considerations. As pressure drops around the well, condensation,
or dropout, occurs and liquid forms. The liquid saturation increases—in what is called
condensate banking—until it is great
enough to overcome capillary trapping
forces and the liquid becomes mobile. But
until the liquid becomes mobile, the presence of immobile liquid reduces the relative
permeability to gas, resulting in a loss in
productivity. The rapid change in fluid saturation away from the well requires a fine
grid to accurately model reservoir properties. The ECLIPSE compositional simulator
modeled the regions around the wells with
a refined radial grid, and the remainder with
a Cartesian grid.
In addition, condensate yields vary
between the four different reservoirs, so
each reservoir fluid was represented by its
own equation of state. The local grid refinement and multiple equation of state capabilities were added to the ECLIPSE simulator
for this project, and now form part of the
commercial package.
The simulation was used to conduct
uncertainty analysis for risk management.
To maximize revenues, the tactic is to maximize gas rates without being penalized for
coming up short. To understand the risks
behind promising a given gas rate, it is
desirable to understand the sensitivity of the
simulation results to each important input
parameter. In this case, repeated simulations indicated that the parameters with the
Developments in Gridding
Since the first grids were built, the variety, range
Perpendicular Bisector (PEBI) Grid
and resolution of oilfield measurements have
increased, and computer power and efficiency
have grown. To take advantage of these developments, reservoir engineers require better and
more comprehensive simulation software tools.
Modern 3D seismic acquisition, processing and
interpretation techniques have resulted in more
reliable and higher-resolution definition of faults
and erosional surfaces. The engineer wants to
represent the full complexity of nonvertical faults,
curving or listric faults, and faults that intersect or
truncate against one another. Another development that requires more complex models is the
increasing use of high-angle and horizontal wells
and multilateral wells. These requirements
stretch the traditional gridding programs based on
corner-point geometry—such as the GeoQuest
GRID program—to the limit.
This has led to the development of new gridding
41
Water saturation % 100
■ A perpendicular bisector (PEBI) grid showing local
grid refinement around wells. Grid blocks may have
a variety of shapes and can fit any reservoir geometry. The smoother grid-block shape also gives a
more accurate simulation solution because there is
less chance of choosing the wrong grid orientation.
software techniques such as the FloGrid utility,
which will produce grids that conform to the reser-
voir models than exist in analytical models.
voir framework as defined by fault surfaces and
Unstructured PEBI grids are of great benefit in
lithological boundaries. Unstructured perpendicu-
these situations, allowing the radial components of
lar bisector (PEBI) and tetrahedral grid systems
flow into the wellbore to be combined with linear
are being developed and included in gridding and
or planar features such as the trajectory of a hori-
simulation programs (above right). “Blocks” in a
zontal well or a fault plane. Simulations run with
PEBI grid may have a variety of shapes, and they
PEBI grids tend to take longer than those run on
may be arranged to fit any reservoir geometry.
structured grids, but the ability to capture the
The smoother gridblock shape gives a more accu-
structural complexity of the reservoir’s flow units
rate simulation solution because there is less
outweighs the need for speed. A compromise can
chance of choosing the wrong grid orientation—
be reached by building a structured grid in the geo-
a potential problem with traditional grids. A PEBI
logically simple parts of the reservoir, and splicing
grid also allows flow in more directions from a
in an unstructured grid when geologic complexity
given grid block, important in the modeling of hor-
requires more flexibly shaped grid blocks.
izontal wells, gas injection schemes or the interaction of wells in an interference test. These grids
are also being used as a basis for a new genera-
10. Crick M: “Compositional Simulation for HPHT Gas
Condensate Reservoirs: Follow-up,” presented at the
Second ECLIPSE International Forum, Houston,
Texas, USA, April 15-19, 1996.
Hsu HH, Ponting DK and Wood L: “Field-Wide
Compositional Simulation for HPHT Gas Condensate Reservoirs Using an Adaptive Implicit Method,”
paper SPE 29948, presented at the International
Meeting on Petroleum Engineering, Beijing, China,
November 14-17, 1995.
Summer 1996
tion of upscaling techniques.
A further gridding development is the linking of
well test analysis with simulator programs to give
the engineer a greater range of numerical reser-
21
Percentage Changes in Reserves
-20
-15
-10
-5
0
5
10
15
20
Gas in place
Permeability
Pentland
continuity
Compaction
Critical
condensate
saturation
Trapped gas
saturation
Well skin
factor
Fault
transmissibility
■ Sensitivity of Erskine simulation results to input parameters. Repeated
simulations indicate parameters that have the most influence on simulation results. Quantifying the uncertainty in the most sensitive parameters
is an important step toward quantifying project risk. Additional simulations were run with the high, low and middle values of each parameter,
forming input sensitivities for the risk analysis shown below.
most influence on the results included gas
in place, permeability and compaction
(left ).
Deliverability and cumulative production
distributions were calculated from the sensitivity results using the parametric method
developed for oilfield applications by P.J.
Smith and coworkers at British Petroleum.11
A normalized average profile was combined
with these distributions in a Monte Carlo
simulator to give a probabalistic production
profile (below ).
The results of the risk analysis showed the
effects of different production scenarios on
the level of confidence in ability to deliver
various possible contracted rates of gas over
the initial plateau period. ( next page,
bottom ). The required 90% confidence
level for a three-year plateau period was
achieved by modifying the production rate
in the first year, adding a contingency well
in the third year, and commingling production in one well between the main Erskine
reservoir and the smaller but higher-permeability Kimmeridge reservoir.
As a result, Texaco has modified production plans, which now call for a lower production rate in the first year than in subse-
Initial
Deliverability Distribution
Parametric
Method
Probabilistic Production Profile
Normalized Average Profile
Sensitivities
Deliverability
Deliverability
Predicted
production
Monte Carlo
Analysis
Cumulative Production
Reserves Distribution
Parametric
Method
■Schematic of deliverability and cumulative production computed for best- and worst-case scenarios. The sensitivity profiles (left)
represent curves for best and worst cases, such as the lowest and highest permeability, lowest and highest compaction and all other
parameters mentioned above. Not all curves were plotted because of space constraints. All the sensitivities were combined through
a parametric method modified for oilfield application. (From Smith et al, reference 11.) A normalized average profile (center) was
combined with initial deliverability and reserves distributions in a Monte Carlo method to give a probabilistic—90% confidence—production profile (right). The upper curve is the deliverability and the lower curve is predicted production. The cyclic nature of the production curve reflects the alternation between summer and winter demand for gas.
22
Oilfield Review
quent years. Risk analysis suggested an
additional well in the third year, so platform
construction has allowed a slot for a contingency well. In addition, production from the
Erskine and Kimmeridge reservoirs will also
be commingled.
Bravo
Alpha
Charlie
Echo
Infill Drilling
Delta
Forties field
Claymore
■ The Forties field in
the North Sea, operated by BP with five
platforms and 103
wells.
Brae
Piper
Beatrice Britannia
Buchan
Forties
Lomond
Montrose
Aberdeen
Erskine
Fulmar
N
UK
600
Production, 103 B/D
Infill drilling is an expensive stage in the life
of a reservoir. Simulation, in conjunction
with other tools, can help guide the placement of wells and minimize their number.
British Petroleum has harnessed simulation
along with new reservoir description to optimize infill drilling in the Forties field in the
North Sea (right ).
The Forties field was discovered in 1970,
and produced its first oil in 1975 (middle ).
Current production is from five platforms,
with 78 producers and 25 peripheral injectors. Estimated recovery of the 4.2 billion
stock tank barrels (STB) of original oil in
place (OOIP) is 60%, or 90% of the movable oil.
The field is characterized by high permeability, high net-to-gross (NTG) pay thickness and a strong aquifer. A few years ago
the Forties was considered to be essentially
a homogeneous reservoir. But early water
breakthrough and water fingering indicated
a greater level of heterogeneity than
expected, and suggested the need for more
wells to be drilled to reach bypassed zones.
To understand the potential of infill drilling
in the field, a simulation study was conducted, including careful reinterpretation of
existing 3D seismic data and a new reser-
500
Current
production
400
300
200
Oil production
Water production
100
11. Smith PJ, Hendry DJ and Crowther AR: “The Quantification and Management of Uncertainty in
Reserves,” paper SPE 26056, presented at the SPE
Western Regional Meeting, Anchorage, Alaska,
USA, May 26-28, 1993.
0
1975
1980
1985
1990
Number
of wells
Commingling
2000
2004
■ Production in the Forties field since 1975.
Confidence levels, %
Yearly rate,
MMscf/D
1995
Year
Tubing
size, in.
Year
Normalized reserves
Confidence level, %
1
2
3
4
90
50
10
90/90/90
3
None
4.5
75
75
75
40
0.707
0.898
1.139
80/90/90
3
None
4.5
85
75
75
40
0.699
0.889
1.119
90/90/90
3
Erskine and
Kimmeridge in E1
4.5
85
85
75
45
0.738
0.937
1.176
80/90/90
3
Erskine and
Kimmeridge in E1
4.5
90
90
80
55
0.738
0.932
1.170
90/90/90
3
Erskine and
Pentland in E1
4.5
70
70
65
30
0.682
0.858
1.082
90/90/90
4
None
4.5
95
95
65
30
0.704
0.892
1.119
90/90/90
3
None
5.5
95
95
70
30
0.685
0.863
1.091
80/90/90
Extra well
in year 3
3
Erskine and
Kimmeridge in E1
4.5
90
90
95
85
0.789
1.000
1.264
Summer 1996
■ Results of risk
analysis ranking
some of the simulated production
scenarios. The
required 90%
confidence level
(bottom line) was
achieved by reducing the production
rate in the first year,
adding a well in
the third year and
commingling production from the
Kimmeridge and
Erskine reservoirs.
23
voir characterization to describe the heterogeneities encountered in the turbidite sandstone reservoir.
Simulation with a coarse full-field model
allowed identification of regions that might
benefit from infill wells, but the results were
not refined enough for detailed well placement. Once a region was identified as containing possible infill well locations, other
aspects were considered, such as: water cut
and production of surrounding wells; interference tests confirming continuity or lack
thereof with other layers; and reinterpretation of 3D seismic data for channel identification—prospective locations tend to be
along submarine channel margins, where
there is lower vertical permeability and so
less efficient sweep.
Having passed these tests, the area was
tapped for a new simulation study with local
grid refinement spotlighting the volume of
interest (below right ). The refined grid block
size was about 50 by 50 m [164 ft by 164 ft]
in area by 8 m [26 ft] in depth. Reservoir
properties were distributed in the LGR grid
based on a geostatistical model. Then the
flow in the LGR grid was simulated with the
ECLIPSE black-oil simulator and checked
against the production history from wells in
the grid. The property distribution was
modified and simulation rerun. This process
was repeated until a history match was
obtained, with only six iterations required.
The final simulation based on the refined
grid predicted a fluid distribution at the Forties Alpha 31 sidetrack (FA31ST) location
(above right ). The predicted fluid distribution closely resembled that encountered and
the predicted oil production matched the
current rate. However, the predicted net-togross rock volume of the upper zone was
optimistic relative to measured values.
Lessons learned from this work have been
fed back into subsequent studies with, for
example, seismic attributes helping to characterize the NTG variation in the reservoir.
Simulation played a similar role in assessing
the potential for infill drilling around the
other platforms.
Prediction
Actual
FA31ST
Shale
Water
FA31ST
Oil
■ Fluid and formation distributions predicted (left) and encountered (right) at the Forties
Alpha 31 sidetrack (FA31ST) location. The predicted distribution closely resembled the
layering encountered, and predicted oil production matched the current rate.
300-m Grid
50-m Grid
■ Steps in the simulation study of the
Forties Alpha platform area. Simulation with a coarse
full-field model
(top) identified
regions that would
benefit from infill
wells. Once a
region was identified as a possible
infill well location,
the location was
selected for a new
simulation study
with local grid
refinement (middle)
spotlighting the
volume of interest.
Reservoir properties were distributed in the LGR
grid based on a
geostatistical
model (bottom) of
the turbidite sandstones.
Geostatistical
Model
24
Oilfield Review
Weyburn Unit
Planning Enhanced Oil Recovery
In an example of simulation later in reservoir life, PanCanadian Petroleum Limited is
relying on simulation to examine the feasibility of CO2 injection in Unit 1 in the Weyburn field of Saskatchewan, Canada
(right ).12 This field was discovered in 1955
and put on waterflood in 1964. By 1994,
recovery had reached 314 million STB, or
28% of the unit’s original oil in place. Ultimate waterflood recovery is expected to be
348 million STB, or 31%, leaving a large
target for enhanced recovery methods. An
opportunity to take advantage of one
method, gravity segregation via CO2 injection, is presented by the division of the
reservoir into swept and unswept layers.
Carbon dioxide injected into the lower,
more permeable formation has the potential
to contact large amounts of unswept oil in
the tight upper formation since CO2 is 30%
less dense than the reservoir fluids at the
expected operating pressures (below right ).
Evaluating the feasibility of CO2 injection
proceeded in stages. First, using the GeoQuest fluid PVT simulation software, a ninecomponent equation of state was developed
that reproduced the behavior of the oil-CO2
system. The equation of state also had to
predict the development of dynamic miscibility in flow simulations while still representing the physical properties of the oilCO2 mixtures. The equation was validated
by comparison of simulated and laboratory
floods on cores.
Second, general performance parameters
were established for the formations to be
swept. These included CO 2 slug size, a
water-alternating-gas injection strategy, CO2
start-up pressure and post-CO2 blow-down
pressure. 13 Then various orientations of
injectors, producers and horizontal wells
were tested with the ECLIPSE compositional
R.13
R.12W2
T.7
T.6
T.5
Saskatchewan
Saskatoon
Yorkton
Swift
Current
Regina
Moose Jaw
Canada
United Sta
tes
■ Weyburn field of southeastern Saskatchewan, Canada. Discovered in 1955, the Weyburn field has produced 314 million STB, or
28% of the unit’s original oil in place.
Producer
CO2 Injection
Density Porosity
Gamma Ray
0
API
Neutron Porosity
150 45
Marly
%
-15
Unswept Zone
Vuggy
5m
12. Burkett D, Besserer G and Gurpinar O: “Design of
Weyburn CO2 Injection Project,” presented at the
Second ECLIPSE International Forum, Houston,
Texas, USA, April 15-19, 1996.
13. Blow-down pressure is the average field pressure
maintained after CO2 injection is stopped. Usually
this is lower than during CO2 injection to maximize
oil recovery due to expansion of CO2.
R.14
Swept
Zone
■ Division of the reservoir into swept and unswept layers, opening
the opportunity for gravity segregation of injected CO2. Carbon
dioxide (blue arrows) injected into the lower, more permeable formation will rise to displace the oil (green arrows) remaining in the
tight, unswept upper formation.
Summer 1996
25
■ Reservoir link with surface facility. Integrating surface network simulators with reservoir simulators will allow production managers
to optimize flow and fine-tune field planning.
Weyburn Unit
km
ax
60-acre
vertical infill
Original
80-acre infill
40-acre
vertical infill
in
km
Horizontal
sidetrack
26
■ A Weyburn
inverted nine-spot
pattern showing
vertical and
horizontal infill
well locations
and directions of
maximum and
minimum permeabilities (kmax ,
kmin ). Various
orientations of
injectors, producers and horizontal
wells were tested
with the ECLIPSE
compositional
simulator to
determine optimal
orientations and
spacings.
simulator (left ).14 Each original nine-spot
pattern was found to require two symmetrically positioned horizontal wells in the
upper zone to take advantage of the CO2
segregation process. Results of the parametric pattern studies, using a 30% pore volume CO2 slug, indicated ultimate recovery
without any new horizontal wells to be an
estimated 37% of OOIP. By adding two
horizontal wells in each injection pattern,
simulation predicted incremental recovery
of 7.2%.
On the Surface
Once hydrocarbons have made it up the
wellbore, most reservoir engineers consider
their job done. But tracking fluid movement
through a complex surface network with
chokes, valves, pumps, pipelines, separators
and compressors remains a daunting task.
Optimizing flow through the surface network allows production managers to minimize capital investment in surface facilities
and fine-tune field planning.
Reservoir simulators are not designed to
solve for fluid flow all the way through the
surface-gathering facility, but they can be
integrated with network simulators built for
this purpose. An example of such a network
simulator is the Simulation Sciences
PIPEPHASE system. The PIPEPHASE simula-
Oilfield Review
Summer 1996
Simulation Speedup with Parallel Processors
2500
2000
Run time, sec
tor, based on a pressure-balance technique
developed originally at Chevron in the
1980s, has been adapted to handle large,
field-wide, multiphase flow networks,
including wells, flowlines and associated
surface facilities. Through a joint project
between GeoQuest Reservoir Technologies
and Simulation Sciences, the PIPEPHASE
simulator and the NETOPT production optimizer are being integrated with the OpenECLIPSE system to provide a way to simulate
fluid flow seamlessly from reservoir through
surface network (previous page, top).15 Integration is achieved through an iterative algorithm that minimizes the differences
between the well flow rates calculated by
the two simulators from a given set of flowing well pressures.
The recent focus on integrated reservoir
management teams is a major step in the
direction of integrated reservoir and surface
network simulation. But the emphasis has
been on integration at the upstream end.
The next step is to focus at the production
and surface facilities end.
Traditionally, the integrated study has been
approached along two independent paths.
For a project involving pressure maintenance through water injection, for example,
the impact on the reservoir has been studied
in isolation. The reservoir simulation is carried out with a simplified well model:
hydraulic behavior of injection or production wells is approximated through flow
tables derived from single-well analysis. A
second study is typically performed by the
facilities engineering group to evaluate the
impact of the injection water requirements
on the surface facilities. The reservoir
behavior at the well is incorporated through
an injectivity index relating injection rate to
pressure drop at the formation.
A limitation of this divided approach is
that it ignores the true interaction between
the elements of the surface network, the
production and injection wells, and the
reservoir. The results of a truly integrated
study could be quite different.
The iterative approach to integrating the
PIPEPHASE and ECLIPSE systems, while rigorous, may be limited by convergence
issues in more complex applications. The
truly integrated solution, with the surface
and reservoir equations solved simultaneously, is expected to require a large effort,
since significant restructuring will be
needed in both simulators. One promising
approach is to initially develop a simple single-phase application for a gas field. The
experiences developed in this effort could
then be extended to address the larger problem of multiphase fluids.
1500
1000
500
0
1
2
4
8
16
Number of processors
■ Speeding up simulation with
parallel processors. For a typical
simulation, the 16-processor run
is more than 10 times faster than
a single-processor run.
The Next Step
The future of reservoir simulators may parallel developments in other oilfield technologies that provide a view of fluid and rock
behavior in the subsurface. For example, the
seismic industry, operating on a similar
physical scale and on equally staggering
amounts of data, has turned to massively
parallel processors (MPPs) for data processing and to high-performance graphics workstations for visualization of the results.
Simulation computer codes are being prepared for implementation on MPPs, but the
switch cannot be made quickly. A simulator
typically solves the fluid-flow equations one
grid block at a time. The solution does not
necessarily benefit by processing several
steps in parallel.
For a typical simulation, doubling the
number of processors cuts simulation time
almost in half, and increasing to 16 processors reduces the time to one-tenth (above ).
Departure from ideal speed gains—16 times
faster for 16 processors—is due to three factors. First, the parallel linear equation solution method is less efficient than the nonparallel solution. Second, it takes time to
assemble and transfer data between processes. And third, load balancing between
processors is uneven: some parts of the
reservoir are easier to solve than others, but
the simulation must wait for the slowest.
Also, the high cost of MPPs targets them for
sharing within departments or companies,
so one user is less likely to get sole access.
Early tests on parallelized versions of the
ECLIPSE simulator indicate that gains in
speed depend on the complexity of the
reservoir model. A North Sea case with two-
phase flow of oil and water in a relatively
simple reservoir with 50,000 grid blocks
exhibited a four-fold speed up using eight
processors, and even greater gains for bigger
models. But three-phase flow simulation in
a 1.2-million block model filled randomly
with geostatistically derived data with highly
variable permeability showed less dramatic
improvement.
One application of simulators that will
undoubtedly benefit from implementation
on MPPs is that of testing multiple scenarios. Simulation results are most valuable in a
comparative sense. Comparisons can be
made of the production behavior of different
reservoir models to gain understanding of
sensitivity to input parameters. Or different
production scenarios may be tested on a
single reservoir model. Running such simulations simultaneously will save time and
allow comparisons to be made efficiently.
In the family of tools designed to help oil
companies make effective use of expensive,
hard-won data, simulation plays a key role
in making sense of data acquired through
different physical experiments, at different
times, at different spatial scales. Simulation
is one of the few tools available for understanding the changes a reservoir experiences
throughout its life. Used together with other
measurements, simulation reinforces conclusions based on other methods and leads
to a higher degree of confidence in our
understanding of the reservoir.
—LS
14. Mullane TJ, Churcher PL, Tottrup P and Edmunds
AC: “Actual Versus Predicted Horizontal Well
Performance, Weyburn Unit, S.E. Saskatchewan,”
Journal of Canadian Petroleum Technology 35, no. 3
(March 1996): 24-30.
15. Dutta-Roy K: “Surface Facility Link: Production Planning with Open-ECLIPSE and PIPEPHASE,” presented at the Second ECLIPSE International Forum,
Houston, Texas, USA, April 15-19, 1996.
27