Académique Documents
Professionnel Documents
Culture Documents
Jack Bouska
BP Amoco plc
Sunbury, England
Summer 1999
Mike Cooper
Andy ODonovan
BP Amoco plc
Aberdeen, Scotland
Chip Corbett
Houston, Texas, USA
Alberto Malinverno
Michael Prange
Ridgefield,
Connecticut, USA
Sarah Ryan
Cambridge, England
21
Traditional Approach
Leading-Edge Approach
Distributed disciplines
Multidisciplinary teamwork
> Simulation approaches. In the past, single-discipline interpretation and lack of computing capability
limited the use of reservoir simulation. Now, a more sophisticated approach to simulation makes the
most of multidisciplinary teams and nearly ubiquitous computers.
22
In some cases, it is best to begin with the simplest model that fits the data and the objectives of
the project and reproduces reservoir behavior. In
all cases, the starting point should be an evaluation of what answers are required from reservoir
simulation, the accuracy needed and the level of
confidence or the acceptable range of quantitative predictions. The model complexity might be
increased as more data become available. The
reward for increasing model complexity can be
evaluated after each simulation run to decide
whether more complex simulation is justified.
Estimates of well flow rates and predictions
of reservoir performance from simulations affect
design of production facilities and should be
believed, even if they seem unlikely. For example,
a deepwater Gulf of Mexico field required expansion and de-bottlenecking of facilities soon after
initial production because the initial reservoir
model was compromised by a pessimistic view of
the interpreted reservoir continuity and flow
rates. Better predictions allow operators to size
facilities correctly the first time rather than having
to re-engineer them.
The quality of predevelopment reserve estimates, field appraisals and development strategies relates closely to reservoir architecture and
structural complexity; reserve estimates tend to
be underestimated in large, less complex fields,
whereas reserves in smaller, more complex fields
are commonly overstated. Poor reservoir models
and resultant incorrect calculations of reserves,
whether too high or too low, have negative economic consequences. In the North Sea, deficient
reservoir models have led to improper facilities
sizing and suboptimal well placement, even in
fields where simulation studies were carried out.3
Better validation of models, particularly using 3D
seismic data, might have averted over- or undersizing production facilities or drilling unnecessary
wells in some cases. In other cases, reservoir
simulation has allowed identification of the key
drivers of reservoir performance so that datagathering efforts can be targeted to reduce
uncertainty in those areas. Alternatively, facilities can be designed to be flexible within a given
reservoir uncertainty.
Oilfield Review
Situation
New discoveries
Deepwater exploration
and trajectories. The reservoir model is strengthened if a geological map of permeability values is
created by applying a porosity-to-permeability
transform to the porosity map according to permeability values interpreted from well tests, well
logs or cores. Even more rigorous results are
obtained when, in addition to inclusion of well
rates and produced or producible hydrocarbon
volumes, all available production data are input
into the model. These include pressure, gas/oil
ratios, and fluid densities, saturations, viscosities
and compressibilities.
In many instances, though, reservoir models
fail to encompass the full diversity of reservoir
data because only a few basic geological and
geophysical maps, constructed from a subset of
the data available, are used to describe variations in the data. Additional data and interpretations are needed to make reservoir models
more robust. For example, core data can serve as
calibrators for geological, petrophysical and
engineering data and interpretations, but are
often used only as guides to permeability. Core
analysis refines model values of porosity, permeability, capillary pressure and fluid saturation.
Whole cores, while not necessarily representative of the entire reservoir, offer tangible
evidence of grain size and composition, sorting,
depositional environment and postdepositional
reservoir history, such as bioturbation, cementation or diagenesis.
Desired Results
Drive mechanism
Prospect evaluation
Scenario planning
Mature fields
Answers to sudden
production problems
Implementation of secondary
recovery
Determine appropriate
recovery method
Divestment or abandonment
Determine future
production volumes
> Simulation uses. Reservoir simulation is useful during all phases of the life of a reservoir and in
both high- and low-risk projects.
Summer 1999
23
A shared earth model is a model of the geometry and properties of the reservoir constrained
by a variety of measurements. To be predictive,
the model should approximate the key features
of the actual reservoir as closely as possible
(right). In a valid reservoir model, predictions
from the model agree with the measured data. A
good fit between predictions and measurements
is not sufficient, though. Several models might
agree equally well with the data. The best model
is the one that agrees best with the data and
with prior information on the model parameters. The uncertainty of the model is defined as
the range of model parameters consistent with
the measurements.
Consider a thin bed imaged by seismic data
(below right). The uncertainty of the shared
earth model in this case is described by the
range of thickness and impedance values that
satisfy the data. This range defines a probability
density function (PDF).
Reservoir Model
Measured Data
Predicted Data
Seismic data
Well log data
Dynamic data
> Model inputs. A shared earth model begins with seismic, well log and dynamic data from the
actual reservoir. In this example, the reservoir is represented by a model (top left). Measured
seismic data (top right) are compared with data predicted by the model (bottom right),
which can be adjusted to improve the fit. The final three-dimensional shared earth model
(bottom left) incorporates all available data.
V
P
Poor fit
V
P
Uncertainty ellipse
h
Good fit
Best fit
> Quantifying uncertainty. The thin bed shown as a red layer (left) has thickness h
and acoustic impedance VP. The plot to the right displays the posterior probability
density function. Thickness and impedance values within the red uncertainty ellipse
satisfy the data, and within that ellipse are red circles denoting a good fit and the
best fit. The red circle outside the uncertainty ellipse does not satisfy the model.
24
Oilfield Review
m2
m2
m2
Uncertainty
ellipse
Best
model
m1
m1
Prior PDF
p (m)
m2
Posterior PDF
p (m d[1])
m2
m2
m1
Prior PDF
p (m d[1])
m1
Likelihood
L (m d[1])
m1
Likelihood
L (m d[2])
m1
Posterior PDF
p (m d[1],d[2])
> Reducing uncertainty. In a Bayesian approach, a prior PDF quantifies the initial
information on model parameters, expressed as vector m. The prior PDF (top left) is
refined by the inclusion of new data (top center) to create the posterior PDF (top right).
The uncertainty of the model is shown in the red uncertainty ellipse. The blue circle
represents the best model. The posterior PDF then becomes the prior PDF (bottom left)
when more data become available (bottom center). The next posterior PDF
(bottom right) has a smaller uncertainty ellipse and a slightly different optimal model.
> Validation modes. Prototype software developed by Schlumberger includes an interactive mode in which the
user assesses the effects of interpretation decisions on reservoir models. In this case, the center of the upper
panel shows predicted seismic data as dotted lines and measured data as solid lines after the upper horizon,
shown in green to the left, has been moved. The lower panel shows a better fit between the predicted and
measured data (center) and the model uncertainty in the ellipse to the right.
Summer 1999
25
Simulation model
Upscaling
Geological modeling
Well logs
Petrophysical modeling
Drilling data
Classification system
Reservoir simulations
> Shared earth model. A numerical representation of the subsurface, housed in a database shared by multidisciplinary team members, allows constant
access to and updating of the reservoir model used for simulation. As databases and software improve, the simulation model and the shared earth model,
which now must be upscaled before being used as a simulation model, will be the same.
26
Oilfield Review
Time horizon
interpretation
Data loading
Attribute extraction
Reservoir property
distribution
Depth grids
Petrophysical
interpretation
Geologic
correlation
Model construction
Weighted average
> Model construction workflow. Once data from the Stratton field were loaded, the team worked together from the outset to correlate well logs,
vertical seismic profile (VSP) data and seismic data. Interpreted seismic horizons, depth conversion results and extracted attributes were compared with
normalized well log porosities and geologic log correlations. The consistent relationship between the weighted average porosity and seismic amplitude
prompted generation of a reservoir property distribution map, a seismic-guided map of porosity distribution in this case, to complete the reservoir model.
Obtaining a good match between the production history and predictions from simulations is
inexpensive in some cases, but can become time
consuming when the model is continuously
refined and simulated. In certain situations, such
as waterfloods, tracers in the form of chlorides,
isotopes or brines are introduced into injected
water to reveal patterns in the reservoir.
Comparisons of these patterns with expected
patterns can be used to reevaluate input values,
for example, porosity, permeability and transmissibilitythe ease with which fluid flows from
one model cell to another, to improve the history
match. Whenever a new well is drilled, it offers
an opportunity to check the quality of a reservoir
simulation, principally by comparison of observed
pressure with the pressure predicted by the
model at the drilling location.
The difficulty of simulating a reservoir
underscores the need to constrain the reservoir
model with all available data. A reservoir model
constrained and validated by geological, geophysical and reservoir data before initiating
simulation extracts as much information as possible from the data and provides a better result.
Also, understanding the range and impact of
reservoir uncertainty allows a quantitative and
qualitative judgment of the accuracy or range of
model predictions.
Summer 1999
Model Validation
A data set from the Stratton field of south Texas
(USA) demonstrates the value of cross-disciplinary
interpretation and model validation in calculating
in-situ gas reserves in the Frio formation (above).8
The data include 3D seismic data, logs from nine
wells, correlations of geological markers and a
vertical seismic profile (VSP). Resistivity, neutron
porosity, bulk density, and spontaneous potential,
gamma ray or both curves were available for the
nine wells. Preliminary examination of the well
logs and VSP data guided selection of horizons in
the Frio formation for seismic horizon tracking.
The VSP provided a good tie between the well
and the seismic data along with good understanding of the phase of the seismic data.9
A thin, clean Frio sand that is easy to correlate
and ties to a mappable seismic event was
selected for both well-by-well analysis and multiwell petrophysical interpretation that ensured
consistent analysis of all the logs. The interpreters
observed that porous zones seemed to correspond
27
Well 18
Well 7
0.10
Well 13
Well 20
0.09
Well 19
0.08
Well 11
0.07
Well 12
0.06
Well 9
0.05
0.04
Well 10
30
35
40
45
50
55
60
65
Amplitude
0.17
Well 10
0.16
Well 19
Well 9
Multiwell effective porosity
0.15
Well 13
Well 12
0.14
0.13
0.12
Well 20
0.11
Well 11
0.10
Well 18
0.09
30
35
Well 7
40
45
50
55
60
Amplitude
28
65
to high seismic amplitude. To confirm this observation, crossplots of effective porosity and amplitude were prepared. The crossplots of the
well-by-well petrophysical analysis showed significant scatter, whereas the multiwell analysis
demonstrated a clear relationship between seismic amplitude and effective porosity (left).
Next, an equation that related effective
porosity to amplitude was used to generate a
map of effective porosity. The mathematical relationship between the weighted average porosity
values at the wellheads and the seismic amplitudes at those locations guided the mapping.
Combining carefully integrated core porosity, logderived porosity and seismic attributesin this
case, amplitudeproduced a single, validated
porosity map constrained by several independent
sources of porosity information (next page).
Using each type of data in isolation in the
Stratton field example obscured relationships
between data and probably would have resulted
in a set of incompatible subsurface maps that
were not physically realistic.
The difference between the single-well analytical approach and the consistent, normalized
petrophysical analysis in the Stratton field
affects the economic evaluation of the reservoir.
The single-well approach precluded integrating
the well logs with the seismic data to generate a
seismic-guided porosity map because the crossplot of effective porosity and amplitude indicated
no consistent relationship between the well logs
and seismic data. The in-situ gas volume calculated by single-well petrophysical analysis is
12% greater than that calculated from the validated, seismic-guided porosity distribution. An
overstated gas volume might lead to unnecessary
infill drilling.
In another case offshore Malaysia, 3D seismic data, well logs, wellbore image logs and
core data enabled generation of time-depth
relationships and synthetic seismograms to tie
logs to seismic data.10 The relationship between
effective porosity, seismic amplitude and acoustic impedance, expressed as a calibration function, allowed prediction of effective porosity
throughout the 3D seismic data, similar to the
previous Stratton field example. Additional
data, such as pressure measurements from
wireline tools or well tests, make the reservoir
model more robust and improve confidence in
the predictions from simulation.11
Oilfield Review
7000
6000
100
5000
90
Well 9
80
4000
3000
70
60
Well 12
50
0.040
0.060
0.080
0.100
0.120
0.140
0.160
0.180
0.200
0.220
2000
Well 11
Well 18
Well 10
40
Well 13
30
1000
Well 7
Well 20
20
10
Well 19
20
40
1000
2000
60
3000
80
4000
100
5000
120
6000
140
7000
160
8000
9000
180
200
10,000 11,000
> Seismic-guided porosity distribution. In the Stratton field of south Texas, USA, a clear relationship between effective
porosity and seismic amplitude permitted seismic-guided mapping of effective porosity. This map could not have been
created without consistent, multiwell petrophysical analysis. Yellow and orange represent areas of high seismic
amplitude; blue represents low amplitude.
Model Manipulation
Because simulation inputs are subject to revision
by the project team to improve the match between
the simulation and production history, it is important to restrict the input model as much as the
data permit and avoid unnecessary adjustments of
input values. Simulation software typically allows
interpreters to change not only the geological and
geophysical maps used to build a reservoir model,
but also variables such as pressure, temperature,
fluid composition and saturation, permeability,
transmissibility, skin, productivity index and rock
compressibility. Seasoned interpreters have different opinions about what changes to simulation
inputs are acceptable, but prudently adjusting
simulation input parameters often improves the
history match.
10. Corbett C, Solomon GJ, Sonrexa K, Ujang S and Ariffin T:
Application of Seismic-Guided Reservoir Property
Mapping to the Dulang West Field, Offshore Peninsular
Malaysia, paper SPE 30568, presented at the SPE
Annual Technical Conference and Exhibition, Dallas,
Texas, USA, October 22-25, 1995.
11. For another example of seismic-guided property mapping:
Hart BS: Predicting Reservoir Properties from 3-D
Seismic Attributes With Little Well ControlJurassic
Smackover Formation, AAPG Explorer 20, no. 4
(April 1999): 50-51.
Summer 1999
29
-800
-1200
-800
-1600
-1000
-1800
-1400
-1200
Depth, ft
415
1443
413
1443
412
1443
410
1443
408
1443
406
1443
404
1443
11,050
18,000
11,150
21,000
11,250
24,000
27,000
11,350
30,000
11,450
33,000
11,550
36,000
11,650
39,000
42,000
11,750
45,000
11,850
48,000
11,950
51,000
12,050
54,000
12,150
T67V
f/s*g/cm3
> Confirming depth conversion. Dipmeter data reduce interpretive contouring options for structure maps
if the mapper honors the data (top). Dipmeter data from the depth of interest, plotted at each well, show
reasonable conformity with structure contours in the upper right and lower sections of the map, but
refute the contouring of the upper left area in this fictitious example. Dip interpretation from an image log,
tied to an actual depth-converted seismic section, confirms dip direction and magnitude at horizons of
interest (bottom). The color variation in the seismic section represents acoustic impedance.
30
Oilfield Review
Some reservoir engineers minimize adjustments to PVT samples, which indicate reservoir
fluid composition and behavior at the pressure,
volume and temperature conditions of the reservoir. In cases of surface recombination of the
sample or sample collection long after initial production, however, the engineer might decide to
adjust PVT values. At the other extreme, production rates and volumes, and pressure data from
wells are considered inalterable by some
experts, although exceptions are made at times,
such as when production measurement equipment fails. Many experts choose to honor the
most accurate representation of production data.
Placing restrictions on the alteration of input
values makes a good history match from simulation more elusive, but many input values may be
adjusted during simulation. Transmissibility is
computed by the simulator using the input porosity and permeability. A high computed transmissibility value can be overridden if well tests,
formation tester data or seismic data provide evidence of separate sand bodies, stratigraphic
changes, faults or other types of reservoir compartmentalization. Differences in fluid chemistry
or pressure from one well to another also suggest reservoir compartmentalization. In-situ fluid
samples obtained from the OFA Optical Fluid
Analyzer component of the MDT tool are uncontaminated and can be brought to surface without
changing phase for chemical analysis.13
Production logs, well tests and pressure transient analyses indicate skin, which is a dimensionless measure of the formation damage
frequently caused by invasion of drilling fluids or
perforation residue.14 When the location, penetration and effectiveness of perforations are of
concern, production logs provide information that
may affect the model input for skin. If a field is
located in a geological trend of similar accumulations, skin values in the trend might be a useful
starting assumption if data within the field are
initially scarce.
Summer 1999
Multidisciplinary validation of reservoir models increases the value of data beyond the cost of
data-gathering activities alone. In 1998, for
example, Geco-Prakla acquired 3D multicomponent seismic data for Chevron in the Alba field in
the North Sea. The objectives of the survey were
to better image the sandstone reservoir, identify
intrareservoir shales that affect movement of
injected water and map waterflood progress.
After integration of the new shear-wave data to
improve the reservoir model, two additional
wells were drilled in the field. The first well is
producing up to 20,000 B/D [3200 m3/d]; the second well is being completed and has resulted in
the discovery of Albas highest net sand. Both
wells have confirmed some of the features
observed on the converted-wave data. Because
the first well was drilled less than a year after
seismic acquisition started, Chevron felt the new
data arrived in time to make a significant commercial impact on the fields development.17
15. For more on the shared earth model and integrated
interpretation: Beardsell M, Vernay P, Buscher H,
Denver L, Gras R and Tushingham K: Streamlining
Interpretation Workflow, Oilfield Review 10, no. 1
(Spring 1998): 22-39.
16. Major MJ: 3-D Gets Heavy (Oil) Duty Workout, AAPG
Explorer 20, no. 6 (June 1999): 26-27.
ORourke ST and Ikwumonu A: The Benefits of
Enhanced Integration Capabilities in 3-D Reservoir
Modeling and Simulation, paper SPE 36539, presented
at the SPE Annual Technical Conference and Exhibition,
Denver, Colorado, USA, October 6-9, 1996.
Sibley MJ, Bent JV and Davis DW: Reservoir Modeling
and Simulation of a Middle Eastern Carbonate
Reservoir, SPE Reservoir Engineering 12, no. 2
(May 1997): 75-81.
17. For more on the Alba field survey and shear wave
seismic data: MacLeod MK, Hanson RA, Bell CR and
McHugo S: The Alba Field Ocean Bottom Cable Seismic
Survey: Impact on Development, paper SPE 56977,
prepared for presentation at the 1999 Offshore European
Conference, Aberdeen, Scotland, September 7-9, 1999.
Caldwell J, Christie P, Engelmark F, McHugo S, zdemir H,
Kristiansen P and MacLeod M: Shear Waves Shine
Brightly, Oilfield Review 11, no. 1 (Spring 1999): 2-15.
31
32
> Forward modeling to optimize data acquisition. Predicted properties of seismic data at
time t2 (top left) are used to predict the appearance of seismic data (middle left). These
predictions are revisited after acquisition of actual seismic data at time t2 (bottom left).
Seismic properties at time t3 (top right) are predicted next from actual t2 seismic data.
By considering fluid changes in the reservoir and their effects on seismic waves, and
then modeling the seismic data that would result from surveying at time t3 (middle right),
additional seismic surveys for reservoir monitoring will be acquired at the optimal time
t3 (bottom right).
Oilfield Review
DC2
Faroe
Islands
DC1
Clair
Schiehallion
0
Foinaven
Shetland
Islands
South
1 km
1 mile
Reservoir
T35
T34
T32
T31
Study area
DC1
DC2
North
Orkney
Islands
SCOTLAND
20
40
20
60
80
40
Horizontal scale
100 km
60 mi
0
0
1 km
Vertical
1 mile scale, m
0
50
100
> Foinaven field. Located West of Shetlands (left), the Foinaven field produces from four main turbidite reservoirs. The reservoir map (top right) shows gas
caps in red and the strike of the normal faults as black lines. The platforms and well locations are shown in black. The Foinaven study area is indicated by
the blue box. The cross section (bottom right), which extends from south of platform DC1 to north of platform DC2, shows the layered reservoirs that have
been compartmentalized by normal faulting that must be drained by carefully constructed directional wells.
Summer 1999
33
Repeat
3D survey
1998
Repeat
3D survey
1998
Water
1000
2000
3000
4000
5000
6000
7000
Baseline
3D survey
1995
Baseline
3D survey
1995
Oil
Dim
Water
1000
2000
3000
4000
5000
6000
7000
> Visible changes in repeated surveys. Cross-sectional synthetic seismic displays for the baseline survey and repeated survey (left)
show the development of a gas cap. The actual seismic sections confirm the predictions from seismic modeling (right).
Small
gas caps
OWC
500
1000
1500
-3.40
-3.17
-2.94
-2.71
-2.48
-2.25
-2.02
-1.79
-1.56
-1.33
-1.10
-0.87
-0.65
-0.42
-0.19
0.04
0.27
0.50
2000 m
Free gas
Enlarged
gas caps
OWC
Free gas
N
0
500
1000
1500
2000 m
> Amplitude changes. Map views of the 1995 baseline survey and the 1998 repeat survey clearly
display the changes in seismic amplitude that result from gas cap development. The small gas
caps in the original survey and the synthetic data shown above it (top) enlarged significantly
after 10 months of production (bottom). The oil-water contact (OWC) remains consistent
between the two surveys.
34
Oilfield Review
Reservoir performance
Reservoir modeling
Interpretive data
Flow simulation
Development scenario
History
matching
Production
Reservoir
development
Production and reserves
forecasting
Uncertainty analysis
and risk management
Reservoir Characterization
Field Implementation
> Future reservoir management. Reservoir optimization is an iterative process that normally begins with reservoir characterization of a new discovery, but can be implemented at any stage in an existing field. Reservoir management will rely increasingly on
monitoring and modeling reservoir performance to optimize oil and gas production. The key additional element will be ongoing
collection of data at the reservoir scale, including seismic data and wellbore measurements, so that the development plan can
be assessed and, where necessary, modified. Monitoring the reservoir closely will overcome the current problems of history
matching using only the loose constraints of production data.
Future Possibilities
Reservoir simulation has already helped oil and
gas producers increase predicted ultimate recovery, and further improvement is likely. In addition
to ongoing software and shared earth model
enhancements, reservoir monitoring with downhole sensors, 4D seismic surveys or other methods is becoming increasingly cost-effective,
particularly when new data are acquired at optimal times (above).22
22. Watts et al: reference 20.
Summer 1999
35