Professional Documents
Culture Documents
CONTENTS
1
APPENDIX A - References
APPENDIX B - Some Overview Articles on
Reservoir Simulation
1. Reservoir Simulation: is it worth the effort?
SPE Review, London Section monthly panel
discussion November 1990.
2. The Future of Reservoir Simulation - C.
Galas, J. Canadian Petroleum Technology, 36,
January 1997.
3. What you should know about evaluating
simulation results - M. Carlson; J. Canadian
Petroleum Technology, Part I - pp. 21-25,
36, No. 5, May 1997; Part II - pp. 52-57, 36,
No. 7, August 1997.
LEARNING OBJECTIVES:
Having worked through this chapter the student should:
Be able to describe the simplifications and issues that arise in going from the
description of a real reservoir to a reservoir simulation model.
Be able to list what input data is required and where this may be found.
Know the meaning of all the highlighted terms - or terms referred to in the
Glossary - in Chapter 1 e.g. history matching, black oil model, transmissibility,
pseudo relative permeability etc.
Be able to describe and discuss the main changes in reservoir simulation over
the last 40 years from the 60's to the present - and say why these have
occurred.
Know all the types of reservoir simulation models and what type of problem
or reservoir process each is used to model.
Know or be able to work out the equations for the mass of a phase or component
in a grid block for a black oil or compositional model.
dN = .N
dt
(1)
where is a constant. We now want to solve this model by answering the question:
what is N as a function of time, t, which we denote by N(t), if we start with a bacterial
colony of size No. It is easy to show that, N(t) is given by:
N( t ) = N o .e . t
(2)
which is the well-known law of exponential growth. We can quickly check that
this analytical solution to our model (equation 1), is at least consistent by setting t
= 0 and noting that N = No, as required. Thus, equation 1 is our first example of a
simulation model which describes the process - bacterial growth in this case - and
equation 2 is its analytical solution. But looking further into this model, it seems
to predict that as t gets bigger, then the number N - the number of bacteria in the
colony - gets hugely bigger and, indeed, as t , the number N also . Is this
realistic ? Do colonies of bacteria get infinite in size ? Clearly, our model is not an
exact replica of a real bacterial colony since, as they grow in size, they start to use
up all the food and die off. This means that our model may need further terms to
describe the observed behaviour of a real bacterial colony. However, if we are just
interested in the early time growth of a small colony, our model may be adequate
for our purpose; that is, it may be fit-for-purpose. The real issue here is a balance
between the simplicity of our model and the use we want to make of it. This is an
important lesson for what is to come in this course and throughout your activities
trying to model real petroleum reservoirs.
In contrast to the above simple model for the growth of a bacterial colony, some
models are much more difficult to solve. In some cases, we may be able to write
down the equations for our model, but it may be impossible to solve these analytically
due to the complexity of the equations. Instead, it may be possible to approximate
these complicated equations by an equivalent numerical model. This model would
commonly involve carrying out a very large number of (locally quite simple) numerical
calculations. The task of carrying out large numbers of very repetitive calculations is
ideally suited to the capabilities of a digital computer which can do this very quickly.
As an example of a numerical model, we will return to the simple model for colony
growth in equation (1). Now, we have already shown that we have a perfectly simple
analytical solution for this model (equation 2). However, we are going to forget
this for a moment and try to solve equation 1 using a numerical method. To do this
we break the time, t, into discrete timesteps which we denote by t. So, if we have
the number of bacteria in the colony at t = 0, i.e. No, then we want to calculate the
number at time t later, then we use the new value and try to find the number at
time t later and so on. In order to do this systematically, we need an algorithm (a
mathematical name for a recipe) which is easy to develop once we have defined the
following notation:
Notation:
Clearly, it is the Nn+1 that we are trying to find. Going back to the main equation that
defines this model (equation 1), we approximate this as follows:
N n +1 N n
.N n
t
(3)
where we use the symbol, "", to indicate that equation 3 is really an approximation, or
that it is only exactly true as t 0. Equation 3 is now our (approximate) numerical
model which can be rearranged as follows to find Nn+1 (which is the unknown that
we are after):
N n +1 = (1 + .t ).N n
(4)
where we have gone to the exact equality symbol, =, in equation 4 since, we are
accepting the fact that the model is not exact but we are using it anyway. This
is our numerical algorithm (or recipe) that is now very amenable to solution
using a simple calculator. More formally, the algorithm for the model would be
carried out as shown in Figure 1.
Set,
t=0
Print n, t and N (N n)
Set
N n+1 = (1 + .t). N n
Set
No
Figure 1
Example of an algorithm to
solve the simple numerical
simulation model in the
text
N n = N n+1
n = n+1
t = t + t
Time to stop ?
e.g. is t > tmax or n > nmax
Yes
End
The above example, although very simple, explains quite well several aspects of
what a simulation model is. This model is simple enough to be solved analytically.
However, it can also be formulated as an approximate numerical model which is
organised into a numerical algorithm (or recipe) which can be followed repetitively.
A simple calculator is sufficient to solve this model but, in more complex systems,
a digital computer would generally be used.
Throughout this course we will use Field Units and/or SI Units, as appropriate.
Although the industry recommendation is to convert to SI Units, this makes discussion
of the field examples and cases too unnatural.
EXERCISE 1.
Return to the simple model described by equation 1. Take as input data, that we
start off with 25 bacteria in the colony. Take the value = 1.74 and take time
steps t = 0.05 in the numerical model.
(i) Using the scale on the graph below, plot the analytical solution for the
number of bacteria N(t) as a function of time between t = 0 and t = 2 (in
arbitrary time units).
(ii) Plot as points on this same plot, the numerical solution at times t = 0, 0.5, 1.0,
1.5 and 2.0. What do you notice about these ?
(iii)Using a spreadsheet, repeat the numerical calculation with a t = 0.001
and plot the same 5 points as before. What do you notice about these?
1000
N(t)
500
(i)
(ii)
Time
We note that certain quantities such as injection and production rates are subject to
physical constraints imposed on us by the reservoir itself.
(ii) Reservoir Givens: Secondly, there are the givens such as the (usually very
uncertain) geology that is down there in the reservoir. There may or may not
be an active aquifer which is contributing to the reservoir drive mechanism.
We can do things to know more about the reservoir/aquifer system by carrying
out seismic surveys, drilling appraisal wells and then running wireline logs,
gathering and performing measurements on core, performing and analysing
pressure buildup or drawdown tests, etc.
(iii) Reservoir Performance Results: Thirdly, there is the observation of the results
i.e the reservoir performance. This includes well production rates of oil, water
and gas, the field average pressure, the individual well pressures and well
productivities etc.
SSW
Fluvial
mud/sand
supply
FC
OM/CS
CRS
TC
SM
SM
TC
ay
eB
rin
a
stu
OM/CS
L
Flu
lain
dp
oo
Fl
ial/
TC
TF
TF
FTD
BM
TS
SM
TF
TC
SS
SS
rrie
Ba
SM
o
Sh
a
ref
ce
SS
TCI
ETD
Fluvial/Floodplain
Facies Asociation
FC: Fluvial channel sandstones
CRS: Crevasse channel/splay sandstones
OM/L: Overbank/lake mudstone
CS: Coal swamp/marsh mudstone and coal
Estuarine Bay-Fill
Facies Association
TC: Tidal channel sandstones
TF: Lower intertidal flat sandstones
TS: Tidal shoal sandstone
SM: Salt marsh/upper intertidal flat mudstones
BM: Brackish bay mudstones
FTD: Flood tidal delta
Tidal Inlet-Barrier Shoreline
Facies Association
TCI: Tidal inlet/ebb channel sandstones
SS: Barrier shoreline sandstone
ETD: Ebb tidal delta
.15
12
km
Figure 2
Conceptual depositional
model for the Linnhe and
Beryl formations from the
middle Jurassic period (UK
sector of the North Sea).
(G. Robertson in Cores
from the Northwest
European Hydrocarbon
Provence, edited by C D
Oakman, J H Martin and
P W M Corbett, Geological
Society, London. 1997).
Slab 1
Top
15855 ft
Slab 2
Top
15852 ft
Slab 3
Top
14591 ft
Slab 4
Top
14361 ft
Slab 5
Top
14358 ft
Medium-grained
Carbonate cemented
sandstone
( =14%, ka = 2mD)
- some thin clay and
carbonate rich lamination
1m
Figure 3
Cores from the midJurassic Beryl formation
from UK sector of the North
Sea. is porosity and ka is
the air permeability. (G.
Robertson in Cores from
the Northwest European
Hydrocarbon Provence,
edited by C D Oakman, J H
Martin and P W M Corbett,
Geological Society, London.
1997).
Medium-grained
ripple-laminated and
bioturbated carbonate
cemented sandstone
( =10%, ka = 1mD)
Medium to coarse-grained
cross-stratified
sandstone
( =21%, ka =1440mD)
- in fining-up units
15858 ft
Base
15855 ft
Base
14594 ft
Base
14364 ft
Base
14361 ft
Base
Coarse-grained
carbonaceous sandstone
( =20%, ka =2940mD)
- in cross-stratified,
fining-up units
Producer
Water Injector
Figure 4
A schematic diagram of a
waterflood simulation in a
3D layered model with an
8x8x5 grid. The information
which is input for a single
grid block is shown.
Contrast this simple model
with the detail in a
geological model (Figure 2)
and in the actual cores
themselves (Figure 3).
x
Inp
, ut:
cr
kx, ock,
n
S ky, k et t
og
w,
P i krw(z,
ros
c (S
S
s
w),
w)
kr
w(S
w ),
2.2 What Are We Trying To Do and How Complex Must Our Model Be?
Therefore, at its most complex, our task will be to incorporate all of the above features
(i) - (iii) in a complete model of the reservoir performance. But we should now
stop at this point and ask ourselves why we are doing the particular study of a given
reservoir? In other words, the level of modelling that we will carry out is directly
related to the issue or question that we are trying to address. Some engineers prefer
to put this as follows:
What is the minimum level of modelling - or which tool can I use - that
allows me to adequately make that decision?
This matter is put well by Keith Coats - one of the pioneers of numerical reservoir
simulation - who said:
The tools of reservoir simulation range from the intuition and judgement of the
engineer to complex mathematical models requiring use of digital computers. The
question is not whether to simulate but rather which tool or method to use.
(Coats, 1969).
Therefore, we may choose a very simple model of the reservoir or one that is quite
complex depending on the question we are asking or the decision which we have to
make. Without giving technical details of what we mean by simple and complex,
in this context, we illustrate the general idea in Figure 5 which shows three models
of the same reservoir. The first (Figure 5a), shows the reservoir as a tank model
where we are just concerned with the gross fluid flows into and out of the system. In
Chapter 2, we will identify models such as those in Figure 5a as essentially material
balance models and will be discussed in much more detail later. The particular
advantage of material balance models is that they are very simple. They can address
questions relating to average field pressure for given quantities of oil/water/gas
production and water influx from given initial quantities and initial pressure (within
certain assumptions). However, because the material balance model is essentially
a tank model, it cannot address questions about why the pressures in two sectors
of the reservoir are different (since a single average pressure in the system is a core
assumption). The sector model in Figure 5b is somewhat more complex in that it
recognises different regions of the reservoir. This model could address the question
of different regional pressures. However, even this model may be inadequate if the
question is quite detailed such as: in my mature field with a number of active injector/
producer wells where should I locate an infill well and should it be vertical, slanted
or horizontal ? For such complicated questions, the model in Figure 5c would be
more appropriate since it is more detailed and it contains more spatial information.
This schematic sequence of models illustrates that there is no one right model for
a reservoir. The simplicity/complexity of the model should relate to the simplicity/
complexity of the question. But there is another important factor: data. It is clear
that to build models of the types shown in Figure 5, we require increasing amounts
of data as we go from Figure 5a5b5c. It is also evident that we should think
carefully before building a very detailed model of the type shown in Figure 5c, if
we have almost no data. There are some circumstances where we might build quite
10
a complicated model with little data to test out hypotheses but we will not elaborate
on this issue at this point.
The simplicity/complexity of the model should relate to the simplicity/complexity of
the question, and be consistent with the amount of reliable data which we have.
(a) "Tank" Model of the Reservoir
Average Pressure
Average Saturations
Wells Offtake
=
=
P
So , Sw and Sg
Aquifer
Aquifer
Figure 5
Schematic illustrations of
reservoir models of
increasing complexity.
Each of these may be
suitable for certain types of
calculation (see text).
Injector
Producer
200ft
2000ft
We are now aware that various levels of reservoir model may be used and that the
reservoir engineer must choose the appropriate one for the task at hand. We will
assume at this point that building a numerical reservoir simulation model is the
correct approach for what we are trying to achieve. If this is so, we now address the
issue: What do we model in reservoir simulation and why do we model it ? There
are, as we have said, a range of questions which we might answer, only some of
which require a full numerical simulation model to be constructed. Let us now say
what a numerical reservoir simulation model is and what sorts of things it can (and
cannot) do.
Definition:
A numerical reservoir simulation model is a grid block model
of a petroleum reservoir where each of the blocks represents a local part of the
11
reservoir. Within a grid block the properties are uniform (porosity, permeability,
relative permeability etc.) although they may change with time as the reservoir
process progresses. Blocks are generally connected to neighbouring blocks are fluid
may flow in a block-to-block manner. The model incorporates data on the reservoir
fluids (PVT) and the reservoir description (porosities , permeabilities etc.) and their
distribution in space. Sub-models within the simulator represent and model the
injection/producer wells.
An example of numerical reservoir simulation gridded model is shown in Figure 6,
where some of the features in the above definition are evident. We now list what
needs to be done in principle to run the model and then the things which a simulator
calculate, if it has the correct data.
To run a reservoir simulation model, you must:
(a) Gather and input the fluid and rock (reservoir description) data as outlined above;
(b) Choose certain numerical features of the grid (number of grid blocks, time
step sizes etc);
(c) Set up the correct field well controls (injection rates, bottom hole pressure
constraints etc.); it is these which drive the model;
(d) Choose which output (from a vast range of possibilities) you would like to have
printed to file which you can then plot later or - in some cases - while the
simulation is still running.
The output can include the following (non-exhaustive) list of quantities:
Some of the above quantities are shown in simulator output in Figure 7. This field
example is for a Middle East carbonate reservoir where the structural map is shown in
Figure 7(d). Figure 7(a) shows the field and simulation results for total oil and water
cumulative production over 35 years of field life. Figure 7(b) shows the actual and
modelled average field pressure. The type of results shown in Figures 7(a) and 7(b)
are very common but the modelling of the RFT (Repeat Formation Tester) pressure
shown in Figure 7(c) is less common. The RFT tool measures the local pressure at a
given vertical depth and, in this case, it can be seen that the reservoir comprises of
three zones each of ~ 100 ft thick and each is at a different pressure. This indicates that
12
pressure barriers exist (i.e. flow is restricted between these layers). This is correctly
modelled in the simulation. This is an interesting and useful example of how reservoir
simulation is used in practice.
Figure 6
An example of a 3D
numerical reservoir
simulation model. The
distorted 3D grid covers
the crestal reservoir and a
large part of the aquifer
which is shown dipping
down towards the reader.
Oil is shown in red and
water is blue and a vertical
projection of a cross-section
at the crest of the reservoir
is shown on the x/z and
y/z planes on the sides of
the perspective box. Two
injectors can be seen in the
aquifer as well as a crestal
horizontal well. Two faults
can be seen at the front
of the reservoir before the
structure dips down into the
aquifer. The model contains
25,743 grid blocks.
Note that a vast quantity of output can be output and plotted up and the post-processing
facilities in a reservoir simulator suite of software are very important. There is no
point is doing a massively complex calculation on a large reservoir system with
millions of grid blocks if the output is so huge and complex that it overwhelms the
reservoir engineers ability to analyse and make sense of the output. In recent years,
data visualisation techniques have played on important role in analysing the results
from large reservoir simulations.
600
500
Observed Oil
400
Modelled Oil
300
200
Observed Water
Modelled Water
100
0
10
15
20
25
30
35
Year of Production
(a) Full field history match of cumulative oil and water production
3500
Institute of Petroleum Engineering, Heriot-Watt University
ure (psia)
700
3000
13
Cumulative Pro
Modelled Oil
300
200
Observed Water
Modelled Water
100
0
10
15
20
25
30
35
Year of Production
(a) Full field history match of cumulative oil and water production
3500
3000
2500
Observed Data
Modelled Data
2000
1500
10
15
20
25
30
35
Year of Production
Figure 7b
Datum
Observed
Modelled
Depth (ft.)
-100
-200
-300
1000
1500
1000
2500
3000
Figure 7c
A Lower Cretaceous
Drilled
New Location
Injector Location
1 Mile
14
Convert to Injector
Figure 7d
Figure 8
Schematic example of how
reservoir simulation might
be used in a study of four
field development options
(see text).
(a) Field A areal plan showing injector and producer well locations; lithology is
given from wells A, B, C and D
15
Sand 1
Sand 2
Sand 3
Sand 4
Figure 8 (b)
(b) Schematic vertical cross-section showing the lithology across the field through
4 wells A, B, C and D
A
A
B
Figure 8 (c)
A
A
B
NZ = 8
NZ = 8
D
D
(d) Reservoir simulation vertical cross-sectional grid showing current well locations.
16
Figure 8 (d)
The grid has 8 blocks in the z- direction representing the thickness of the
formation as shown below; NZ = 8. Note that the vertical grid is not uniform.
Periferal Injectors
A
Periferal Injectors
C
Periferal Injectors
B
B
D
D
Infill Wells
Infill Wells
Option 4
Option 3
Option 4
Cumulative
Oil Oil
Cumulative
Cumulative Oil
Figure 8 (e)
Option 3
Continue as at present (do nothing) Option 1
Option 2
Time
Continue as at present (do nothing) Option 1
Option 2
Continue as at present (do nothing) Option 1
Time
Option 2
Time
Figure 8 (f)
NPV NPV
or IRR
or IRR
NPV or IRR
(f)
3
3
1
1
2
2
4
4
Option
Option
Option
Figure 8 (g)
Institute of Petroleum Engineering, Heriot-Watt University
17
The nature of the reservoir recovery plan e.g. natural depletion, waterflooding,
gas injection etc.
The nature of the facility required to develop the field e.g. a platform, a subsea
development tied back to an existing platform or a Floating Production System
(for an offshore fileld).
The sequencing of the well drilling program and the topside facilites.
18
It is during the initial appraisal stage that many of the biggest - i.e. most expensive
- investment decisions are made e.g. the type of platform and facilities etc. Therefore,
it is the most helpful time to have accurate forward predictions of the reservoir
performance. But, it is at this time when we have the least amount of data and,
of course, very little or no field performance history (there may be some extended
production well tests). Therefore, it seem that reservoir simulation has a built-in
weakness in its usefulness; just when it can be at its most useful during appraisal is
precisely when it has the least data to work on and hence it will usually make the
poorest forward predictions. So, does reservoir simulation let us down just when we
need it most? Perhaps. However, even during appraisal, reservoir simulation can
take us forward with the best current view of the reservoir that we have at that time,
although this view may be highly uncertain. As we have already noted, if major
features of the reservoir model (e.g. the stock tank oil initially in place, STOIIP) are
uncertain, then the forward predictions may be very inaccurate. In such cases, we
may still be able to build a range of possible reservoir models, or reservoir scenarios,
that incorporate the major uncertainties in terms of reservoir size (STOIIP), main
fault blocks, strength of aquifer, reservoir connectivity, etc. By running forward
predictions on this range of cases, we can generate a spread of predicted future field
performance cases as shown schematically in Figure 9. How to estimate which of
these predictions is the most likely and what the magnitude of the true uncertainties
are is very difficult and will be discussed later in the course.
"Optimistic" Case
Figure 9
Spread of future predicted
field performances from a
range of scenarios of the
reservoir at appraisal.
2005
"Pessimistic" Case
Most Probable Case
2010
Time (Year)
2015
Different assumptions about the original oil in place (STOIIP; Stock Tank Oil
Originally In Place).
19
Mature field development: we define this stage of field development for our purposes
as when the field is in mid-life; i.e. it has been in production for some time
(2 - 20+ years) but there is still a reasonably long lifespan ahead for the field, say
3 - 10+years. At this stage, reservoir simulation is a tool for reservoir management
which allows the reservoir engineer to plan and evaluate future development options
for the reservoir. This is a process that can be done on a continually updated basis.
The main difference between this stage and appraisal is that the engineer now has
some field production history, such as pressures, cumulative oil, watercuts and GORs
(both field-wide and for individual wells), in addition to having some idea of which
wells are in communication and possibly some production logs. The initial reservoir
simulation model for the field has probably been found to be wrong, in that it fails
in some aspects of its predictions of reservoir performance e.g. it failed to predict
water breakthough in our waterflood (usually, although not always, injected water
arrives at oil producers before it is expected). By the way, if the original model
does turn out to be wrong, this does not invalidate doing reservoir simulation in the
first place. (Why do you think this is so?)
At this development stage, typical reservoir simulation activities are as follows:
Carrying out a history match of the (now available) field production history
in order to obtain a better tuned reservoir model to use for future field performance
prediction
Using the history match to re-visit the field development strategy in terms
of changing the development plan e.g. infill drilling, adding extra injection
water capability, changing to gas injection or some other IOR scheme etc.
There are many reported studies in the SPE literature where the simulation model is
re-built in early-/mid-life of the reservoir and different future development options
are assessed (e.g. see SPE10022 attached to this chapter).
Late field development: we define this stage of field development as the closing few
years of field production before abandonment. A question arises here as to whether
the field is of sufficient economic importance to merit a simulation study at this stage.
20
A company may make the call that it is simply not worth studying any further since
the payback would be too low. However there are two reasons why we may want
to launch a simulation study late in a fields lifetime. Firstly, we may think that,
although it is in far decline, we can develop a new development strategy that will
give the field a new lease of life and keep it going economically for a few more
years. For example, we may apply a novel cheap drilling technology, or a program
of successful well stimulation (to remove a production impairment such as mineral
scale) or we may wish to try an economic Improved Oil Recovery (IOR) technique.
Secondly, the cost of field abandonment may be so high - e.g. we may have to remove
an offshore structure - that almost anything we do to extend field life and avoid this
expense will be economic. This may justify a late life simulation study. However,
there are no general rules here since it depends on the local technical and economic
factors which course of action a company will follow. In some countries there may
be legislation (or regulations) that require that an oil company produces reservoir
simulation calcualtions as part of their ongoing reservoir management.
Three field cases are now presented. We reproduce the full SPE papers describing
each of these reported cases. In the text of each of these papers there are margin
numbers which refer to the Study Notes following the paper. We use these to explain
the concepts of reservoir simulation as they arise naturally in the description of a field
application. In fact, you may very well understand many of the term immediately
from the context of their description in the SPE paper.
The three field examples are as follows:
Case 1: The Role of Numerical Simulation in Reservoir Management of a West
Texas Carbonate Reservoir, SPE10022, presented at the International Petroleum
Exhibition and Technical Symposium of the SPE, Beijing, China, 18 - 26 March
1982, by K J Harpole and C L Hearn.
Case 2: Anguille Marine, a Deepsea-Fan Reservoir Offshore Gabon: From Geology
Toward History Matching Through Stochastic Modelling, SPE25006, presented at the
SPE European Petroleum Conference (Europec92), Cannes, France, 16-18 November
1992, by C.S. Giudicelli, G.J. Massonat and F.G. Alabert (Elf Aquitaine)
Case 3: The Ubit Field Rejuvenation: A Case History of Reservoir Management of
a Giant Oilfield Offshore Nigeria, SPE49165, presented at the SPE Annual Technical
Conference and Exhibition, New Orleans, LA, 27-30 September 1998, by C.A. Clayton
et al (Mobil and Department of Petroleum Resources, Nigeria)
These cases were chosen for the following main reasons:
They are all good technical studies that illustrate typical uses of reservoir
simulation as a tool in reservoir management (we have deliberately taken all
cases at the middle and the mature stages of field development since much
more data is available at that time);
They introduce virtually all of the main ideas and concepts of reservoir
simulation in the context of a worked field application. As these concepts
21
and specialised terms arise, they are explained briefly in the study notes although
more detailed discussion will appear later in the course. Compact definitions
of the various terms are given in the Glossary at the front of this module;
By choosing an example from the early 1980s, the early/mid 1990s and the late
1990s, we can illustrate some of the advances in applied reservoir simulation
that have taken place over that period (this is due to the availability of greater
computer processing power and also the adoption of new ideas in areas such
as geostatistics and reservoir description).
How you should read the next part of the module is as follows:
Read right through the SPE paper and just pay particular attention when there
is a Study Note number in the margin;
Go back through the paper but stop at each of the Study Notes and read
through the actual point being made in that note.
As noted above, all the main concepts that are introduced can also be found
in the Glossary which should be used for quick reference throughout the
course or until you are quite familiar with the various terms and concepts in
reservoir simulation.
See SPE 10022 paper in Appendix
Case 1:
The Role of Numerical Simulation in Reservoir Management of a West
Texas Carbonate Reservoir, SPE10022, presented at the International Petroleum
Exhibition and Technical Symposium of the SPE, Beijing, China, 18 - 26 March
1982. by K J Harpole and C L Hearn.
Summary: This paper presents a study from the early 1980s where a range of reappraisal strategies for a mature carbonate field are being evaluated using reservoir
simulation. For example, possible development strategies include the blowdown
of the gas cap or infill drilling. They explicitly state in their opening remarks that
their central objective is to optimise reservoir performance by choosing a future
development strategy from a range of defined options. The structure of the study is
very typical of the work flow of a field simulation study, viz Introduction; Reservoir
Description; Simulation Model; History Matching; Future Performance; Conclusions
and recommendations. Although this paper is almost 20 years old, it introduces the
reader in a very clear way to virtually all the concepts of conventional reservoir
simulation.
Location maps and general reservoir structure shown in Figures 1 and 2 of SPE
10022.
22
1. States explicitly that the objective of the study is to optimise reservoir performance
as discussed in the introductory part of this module.
2. Raises the issue of an accurate reservoir description being required and this is
emphasised throughout this paper.
3. An interesting point is raised comparing the carbonate reservoir of this study
broadly to sandstone reservoirs. It notes that the post-depositional diagenetic effects
are of major importance in the West Seminole field in that they affect the reservoir
continuity and quality i.e. the local porosity and permeability. In contrast, it is noted
that sandstone reservoir are mainly controlled by their depositional environment
and tend to show less diagenetic overprint. However, a point to note is that the
broad outline and work flow of a numerical reservoir simulation study are quite
similar for both carbonate and sandstone reservoirs.
4. Carbonate diagenetic processes include dolomitisation (dolomite = CaMg(CO3)2),
recrystallisation, cementations and leaching. This geochemical information is not
directly used in the simulation model but it is important since it leads to identification
of reservoir layer to layer flow barriers (see below).
5. Strategy: Previous gas re-injection into the cap + peripheral water injection =>
not very successful. They want to implement a 40 acre, 5-spot water flood; see
Fig. 3. A 5-spot is a particular example of a pattern flood which is appropriate
mainly for onshore reservoirs where many wells can be drilled with relatively close
spacing (see Waterflood Patterns in the Glossary).
6a. They raise the issue of vertical communication between the oil and gas zones.
This is an excellent example of an uncertain reservoir feature that can be modelling
using a range of scenarios from free flow between layers to zero interlayer flow + all
cases in between. Therefore, we can run simulations of all these cases and see which
one agrees best with the field observations (which is what they do, in fact).
6b. The vertical communication - or lack of it - will affect flow between the oil and
gas zones which may lead to loss of oil to the gas cap; see Figure 4.
7. States the structure of the simulation study work flow: Accurate reservoir description
- Develop the simulation model (perform the history match - see below - use model
for future predictions - evaluate alternative operating plans). A history match is when
we adjust the parameters in the simulation model to make the simulated production
history agree with the actual field performance (expanded on below).
8a. A lengthy geological description of the reservoir is given where the depositional
environment is described - reference is made to extensive core data (~7500 ft. of
core).
8b. The impact of the geology/diagenesis in the simulation model is discussed here.
There is evidence of field wide barriers due to cementation with anhydrite which may
reduce vertical flows. This is important since it gives a sound geological interpretation
to the existence of the vertical flow barriers. Therefore, if we need to include this to
Institute of Petroleum Engineering, Heriot-Watt University
23
match the field performance, we have some justification or explanation for it rather
than it simply being a fiddle factor in the model.
9. Figure 5 shows the 6 major reservoir layers where the interfaces between the layers
are low , low k anhydrite cement zones. Again, these may be explained from the
depositional environment and the subsequent diagenetic history of the reservoir.
10. 7500 ft. of whole core analysis for the W. Seminole field was available which
was digitised for computer analysis (not common at that time, late 1970s). This is
very valuable information and is often not available.
11. Permeability distributions in the reservoir are shown in Fig. 6 and these data
are vital for reservoir simulation. Dake (1994; p.19) comments on this type of data:
What matters in viewing core data is the all-important permeability distribution
across the producing formations; it is this, more than anything else, that dictates the
efficiency of the displacement process.
12. They note that no consistent k/ correlation is found in this system (which is
quite common in carbonates). Often some approximate k/ correlation can be found
for sandstones (e.g. see k/ Correlations in the Glossary).
13. The W. Seminole field does exhibit a distinctly layered structure and the
corresponding permeability stratification in the model is shown in Fig. 7.
14a. Pressure transient work - again gives important ancillary information on
the reservoir. The objectives of this work were to determine whether there was
(i) directional permeability effects, directional fracturing or channelling; (ii) the
degree of stratification in the reservoir; (iii) evaluation of the pay continuity
between the injectors and producers
14b. No evidence of channelling or obvious fracture flow system
14c. Distinct evidence was seen for: (a) the presence of a layered system; (b) restricted
communication between layers (P 200 - 250 psi between layers). This is vital
information since it gives an immediate clue that there is probably not completely free
flow between layers i.e. there are barriers to flow as suspected from the geology.
14d. Finally on this issue, there is pressure evidence of arithmetically averaged
permeabilities. This is again typical of layered systems.
15. Native state core tests are referred to from which they obtained steady-state
relative permeabilities. These could be very valuable results but no details given
here. NB it appears that only one native state core relative permeability was actually
measured. This is probably too little data but reflects the reality in many practical
reservoir studies that often the engineer does not have important information; however,
we just have to get on with it.
16. In this study the reservoir simulator which they used was a commercial Black Oil
Model (3D, 3 phase - oil/water/gas). Modelling carried out on the main dome portion
of the reservoir. This is done quite often in order to simplify the model and to focus
24
on the region of the field of interest (and importance in terms of oil production). A
no flow boundary is assumed in the model on the saddle with the east dome (justified
by different pressure history). Again, this is supported by field evidence but it may
also be a simplifying judgement to avoid unnecessary complication in the model.
17a. The grid structure used in the simulations is shown in Fig 8. The particular grid
that is chosen is very important in reservoir simulation. An areal grid of 288 blocks
( 16 x 18 blocks) - about 10 acre each is taken along with six layers in the vertical
direction; i.e. a total of 1728 blocks. This would be a very small model by todays
standards and could easily be run on a PC - this was not the case in late 1970s.
17b. They refer to changing the transmissibilities between grid blocks in order to
reduce flows. (See Glossary for exact definition of transmissibility.)
18. The following three concepts are closely related (see Pseudo-isation and
Upscaling in the Glossary):
18a. Grid size sensitivity: Refers to the introduction of errors due to the coarsness
of the grid known as numerical dispersion.
18b. The very important concept of pseudo--relative permeability is introduced here
(Kyte and Berry, 1975). Pseudos are introduced in order to control numerical
dispersion and account for layering. In essence, the use of pseudos can be seen as
a fix up for using a coarse grid structure.
18c. Corresponding coarse and fine grid reservoir models are shown in Fig. 9.
They note that the fine grid model uses rock relative permeabilities while the coarse
grid model uses pseudo relative permeabilities.
19. History Matching: The basic idea of history matching is that the model input
is adjusted to match the field pressures and production history. This procedure is
intended as being a way of systematically adjusting the model to agree with field
observations. Hopefully we can change the correct variables in the model to get
a match e.g. we may examine the sensitivity to changes in vertical flow barriers in
order to find which level of vertical flow agrees best with the field (indeed, this is
done in this study). See History Matching in the Glossary.
20a. Early mechanism identified as solution gas drive and assistance from expansion.
Some initial discussion of field experience and numerical simulation conclusions is
presented and developed in these points.
20b. They note some problems with data from early field life. (i) Complicated by
free gas production; (ii) channelling due to poor well completions; (ii) no accurate
records on gas production for the first 6 years.
20c. The actual field history match indicates that approx. 8 - 10 BCF of gas must have
been produced over this early period in order to match the field pressures. This is a
use of a material balance approach in order to find the actual early STOIIP (STOIIP
= Stock Tank Oil Initially In Place).
25
21a. They present a description of some adjustments to the history match - but overall
it is very good (which they attribute to extensive core data).
21b. Some highlighting of problems with earlier water injection .
21c. The actual history match of reservoir pressure and production is shown in Fig.
10. This is a good history match but think of which field observable - gas production,
water production or average field pressure - is the easiest/most difficult to match?
22. A good description of their study of the sensitivity to vertical communication
is given at this point. This is examined by adjusting the vertical transmissibilities.
They look at the following cases: (i) no barriers; (ii) moderate barrier; (iii)
strong barriers and (iv) no-flow barriers. Most of the sensitivties are for the
moderate and strong barrier cases.
23a. Results showed that => strong barrier case is best but some problem high
GOR wells are encountered randomly spaced through the field. They diagnosed
and simulated this as behind the pipe gas flow in these wells to explain the
anomalies in the field observations. This is quite a common explanation that
appears in many places.
23b. Layer differential pressures up to 200 - 250 psi can only be reproduced for
the strong barrier case. In simulation terms, this is probably the strongest evidence
that this is the best case match.
24. The strong barrier case was chosen as the base case and this was used for
the predictive runs. The base case predictions refer to the cases which essentially
continue the current operations and these are shown in Fig. 11.
25. The strategies looked at for the future sensitivities are listed as follows: (i) change
rate of water injection; (ii) management of gas cap voidage i.e. increase of gas and
blowdown at different times; (iii) infill drilling.
26a. Outlines the problems/issues for various strategies as follows: (i) shows vertical
communication is very importance - it has a major impact on predicted reservoir
performance; (ii) shows that can avoid high future P between gas cap and oil
zone by high water injection or early blowdown; (iii) shows better development
strategy is to keep low P e.g. increase gas injection or infill drill. Finally, shows
infill drilling is the most attractive option and the forward prediction for this case is
shown in Figure 12.
26b. Table showing some alternatives in text.
27a. A brief summary of the best future development option is given which is: (i)
infill drilling as the best option; (ii) water injection increased concurrently with the
drilling program to maintain voidage replacement (but prevent the over-injection of
water).
27b. For completeness, it is explained why other plans are not as attractive; i.e.
blowdown of gas cap before peak in waterflood production rate would significantly
reduce oil recovery.
26
28. A reasonably good initial forward prediction from 1978 - 1981 is shown in Figs.
13 and 14.
29. Conclusions are given which, in summary, are as follows:
1.
2.
3.
4.
5.
6.
27
second stage where the petrophysical values are assigned to each building block in the
geological model. The consequences of making various assumptions in the gridding
are evaluated for water, gas and WAG (water-alternating-gas) injection. They note
that is it very important to represent the main geological features in the gridded model.
It was also noted that, when a regular coarse grid was used, the contrast in properties
of this heterogeneous reservoir were smoothed out by the averaging process and
in most cases led to a more optimistic predicted production performance. That is,
the more stochastic models led to a reduction in predicted recovery compared with
conventional coarse gridded models.
In the proceedings of the SPE European Petroleum Conference, Cannes, France,
16-18 November 1992. A session at this conference produced the following
selection of reservoir simulation papers:
SPE25008: Reservoir Management of the Oseberg Field After Four Years,
S. Fantoft (Norsk Hydro)
Summary: The Oseberg Field (500x106 Sm3 oil; 60x106 Sm3 gas) comprises of
seven partly communicating reservoirs. Both water and gas are being injected and
modelled in this study and results indicate over 60% recovery in the main reservoir
units. The modelling results indicate that the plateau production will be extended
by the use of horizontal wells. The objective of the simulation study was exactly
this - i.e. to maximise the plateau and improve ultimate oil recovery. This is a very
competent simulation study and - although details are not given - it is stated that
the geological model is updated annually based on information from new wells.
It establishes several aspects of the reservoir mechanics and makes a number of
recommendations for operating practice in the future. In other respects, this is quite
a conventional study.
SPE25057: The Construction and Validation of a Numerical Model of a Reservoir
Consisting of Meandering Channels, W. van Vark, A.H.M. Paardekam, J.F. Brint
J.B. van Lieshout and P.M. George (Shell)
Summary: This study focuses on a reservoir which has low sandbody connectivity and
which is interpreted as a meandering channel fluvial system. Two years of depletion
data is available and one of the aims of the study was to evaluate the possibility of
performing a waterflood in this field. They identified a problem in that the sandbody
connectivity was lower than might be expected from the sedimentology alone and it
was conjectured that this might be due to minor faulting with throws of a few meters.
This study again emphasises the importance of the reservoir geology and tries to
relate the performance back to this. The geological model is also an early practical
example of using a voxel representation of the system - approx. 128,000 voxels
were used in the model. They noted that the original (sedimentological) models
gave over optimistic connectivity. An acceptable match to observed field pressures
by including some level of smaller scale faulting.
SPE25059: Development Planning in a Complex Reservoir: Magnus Field UKCS
Lower Kimmeridge Clay Formation (LKCF), A.J. Leonard, A.E. Duncan, D.A.
Johnson and R.B. Murray (BP Exploration Operating Co.)
Summary: This simulation study was carried out on the geologically complex,
low net to gross LKCF (rather than on higher net to gross Magnus sands studied
28
previously). The objective was to formulate a development plan for the LKCF which
would accelerate production from these sands. Stochastic modelling techniques
were integrated into more conventional deterministic models and various options
were screened for inherent uncertainty and risks. The study concluded that a phased
water injection scheme was the best way forward with the phasing being used to
manage and offset the considerable geological risks. Ranges of expected recovery
were generated and an incremental recovery of 60 MMstb was predicted increasing
the total reserve of the LKCF by a factor of x2.4. This study also demonstrated the
importance of inter-disciplinary team work to overcome the previously inhibiting
high risks involved.
The proceedings of Europec92 also included the following paper:
SPE25006: Anguille Marine, a Deepsea-Fan Reservoir Offshore Gabon: From
geology Toward History Matching Through Stochastic Modelling, C.S. Giudicelli,
G.J. Massonat and F.G. Alabert (Elf Aquitaine)
This paper is such a good example of contemporary studies at that time, that
this is chosen as our Case 2 example and is presented in some detail in the
next section.
Case 2:
Anguille Marine, a Deepsea-Fan Reservoir Offshore Gabon: From
geology Toward History Matching Through Stochastic Modelling, SPE25006,
presented at the SPE European Petroleum Conference (Europec92), Cannes, France,
16-18 November 1992, by C.S. Giudicelli, G.J. Massonat and F.G. Alabert (Elf
Aquitaine)
See SPE 25006 paper in Appendix
Summary: The Anguille Marine Field in Gabon has 25 years of production history.
The waterflood performance indicated severe sedimentary heterogeneity as the field
is known to have been deposited in a deep water fan sedimentary environment. This
paper is one of the first to refer to the multi-scale nature of the heterogeneity (5 scales
were studied) and to refer this back to the sequence stratigraphy of the depositional
environment. The sequence stratigraphic approach allowed the field to be divided into
the main types of turbiditic geometries (channels, lobes, slumps, laminated facies).
Fine scale models (> 2 million grid blocks) were generated using geostatistical
techniques and several issues were raised concerning both the geological model
and the upscaling process itself. This is a very good example of an early integrated
geology(sedimentology)/engineering study in reservoir simulation. The multi-scale
nature of the heterogeneity is well related back to the geology.
1. Depositional environment: the Anguille Marine field is a deep sea fan environment
(i.e a turbidite) with a low sand/shale ratio. This geological description opens the
discussion (unusual for previous simulation studies) and the geology features heavily in
the flow properties and hence in the geological and reservoir models of this field.
2. Sequence stratigraphy: A more modern feature of reservoir simulation is that the
five identified scales of heterogeneity are recognised and some attempt is made to
Institute of Petroleum Engineering, Heriot-Watt University
29
incorporate them into the 3D simulation model. These scales are also firmly linked
to the geology (sedimentology) through the principles of sequence stratigraphy.
3. Geostatistics: Reference is made to how the geological features constrain the
fine scale 3D models (of > 2 million blocks - which was large for the time) using
geostatistical techniques. By the early 90s, the use of geostatistical methods was
becoming more widespread and how it has been applied in this case is covered better
by Refs. 1 - 5 in this paper.
Location and structure maps of Anguille Marine are given in Figures 1 - 4.
4. Brief field facts: Discovery 1962; primary depletion commenced in 1966 but reservoir
pressure fell rapidly over the next 2 - 3 years and GOR increased; waterflooding from
1971 restored pressure support but channelling led to early water breakthrough; infill
drilling not very successful due to lack of current understanding of complex reservoir
geology; new approach in 1990 focused more strongly on the reservoir geology of this
heterogeneous low sand/shale ratio system recognising the characteristic geometries
of a tubditic fan - lobes channels, levees, slumps, laminated facies etc.
5. The approach: It is important in all reservoir simulation studies to have a clear
logic to how we approach the simulation of a large complex reservoir system. Here
they describe their general methodology although details are in Refs. 1 - 4 at the end
of the paper. Basically they: describe and model upper reservoir/ extend to the whole
reservoir/ try to translate the geological model to a practical simulation model. On
the latter issue they describe the use of partial models where just a smaller sector
of the reservoir is studied but lessons are taken back into the full model.
6. Reservoir description: Section 2 of the paper gives a sedimentological description
of the reservoir as a slope-apron fan of complex lithology (depositional model Figure
3) in which 14 (simplified) facies were retained; criteria of composite log recognition of
various facies shown in Figure 5. Some contradictory water breakthrough observations
were noted. Table 1 gives sedimentary body dimensions (lengths and widths) for
channels, lobes. levees/crevasse-splay, slumps, channels (Upper Anguille); Table 2
gives mean petrophysical characteristics. A very important final result for reservoir
simulation is the identification of five scales of heterogeneity - Figures 6 and 7; this
makes the geological analysis and information numerically useable.
7. Sedimentary history: In earlier reservoir simulation studies, and indeed up to the
present time, it is rare to see sedimentary history discussed in terms of a sequence
stratigraphic analysis (even mentioning the pioneering work on sea level changes of
P. R. Vail et al, Seismic stratigraphy and global changes of sea level, in Seismic
Stratigraphy, Applications to Hydrocarbon Exploration, AAPG Memoir 26, pp. 49-212,
1977). Chronostratigraphic correlations refer to the timelines of simultaneous
deposition. This analysis underpins much of the reservoir description but we will
not elaborate on it here.
8. Geostatistical modelling: Mainly discussed in Refs. 1 and 2 of this paper. Firstly,
focus on geostatistical modelling of the 3D distributions of the major flow units
(channels and lobes) and barriers (laminated facies or slumps) for the entire reservoir.
This is done as a conditional simulation where the distribution is constrained
30
(or conditioned) to the observed facies and reservoir quality observed at the
wells. Secondly, the smaller scale heterogeneities are unconditionally simulated
(synthetically) to yield average properties within the major flow units (see Ref. 2 in
paper). The geostatistical simulation method used was indicator simulation (Refs.
1 and 8) which require the average frequency and variogram information.
9. Reservoir zonation: Six unit vertical reservoir zonation shown in Figure 10.
Simulations of lateral continuity within each of the units (five - not Middle Anguille,
Figure 10) performed independently since they correspond to separate sedimentary
phases. For horizontal zonation, Figure11 shows lateral zonation on LA2
and UA2 units showing directional trends and thus variograms with spatially
variable anisotropy direction used in final model. Figures 12 - 15 show resulting
correlation structures of the various units. Ends up with >2 million grid blocks
in the full field 3D model.
10. Flow simulations: Discusses details of upscaling from fine grid stochastic model
(>2 million blocks) to coarse grid simulation model (11,000 grid blocks). 11 vertical
layers are retained to represent the reservoir layering with more blocks being used
in the best reservoir units. Upscaling of absolute permeability at some aggregation
rate (e.g. 4x4) is applied leading to areal block sizes of 200m x 200m - see Figure14.
Relative permeabilitiees were upscaled on a typical block configuration (details
in Refs. 2 and 4). Additionally: Three major zero-transmissibility faults included in
model; some WOC variation across field; depth varying bubble points assigned; 25
years of injection/production for history matching.
11. Simulation results: Initial pressure depletion results shown in Figure 16 - where
14 out of 17 wells show satisfactory pressure behaviour. Pressure behaviour and
water breakthrough are poorly predicted during injection stage - Figure 17; water
saturations around injectors shown in Figure18 - upscaling has washed out the
finer scale strong anisotropy.
12. Model changes: Table 8 lists a number of sometimes quite radical changes to
the model in order to achieve a better fit to observed field performance - Figure 15
shows differences in upscaled permeability maps. Continuing problems with injection
predictions => - is geological model correct? - what is the real effect of upscaling?
13. Partial models: Thin model - Figure 19 shows the thin partial field model to
verify reservoir geology; well AGM18 good water breakthrough match (Figure 20)
- early breakthrough for well AGM29 (Figure 21). When thin model upscaled as in
full field model (abs. k upscale + rel perm as before) - results in Figures 21 and 22
- breakthrough delayed in both wells but shape of BSW is satisfactory. Conclusions:
Thin model partly validates geological model; Some problems with upscaling not
supressing breakthrough, making reservoir too connected and eliminating strong
anisotropy. Test model - (50 x 20 x 56) model extracted from full field model. Figure
23 shows that an optimum upscaling aggregation rate (2 x 2 x 7) is found - they warn
caution on this point. We note that if very reliable and general upscaling techniques
were available, then this should be eliminated (more work has been done on this
issue since 1992 - much of it at Heriot-Watt!).
31
14. Conclusions:
Sedimentology controls heterogeneity analysis when very wide variation in
sandbody geometries is found (as in this case)
Geostatistical indicator simulation is a good tool for modelling this multi- scale
heterogeneity - trends can also be included
1. New data and techniques: The study is a very good example of the close
integration of (especially) 3D seismic data used in several ways, computer mapping
32
4. Structural reinterpretation: Figures 2a and 2b show both the original and current
interpretations of the structure. The original rubble beds are reinterpreted as being
techtonically disturbed downslope movements of the youngest sand sequences
resulting in large scale slumping and block sliding. The older interpretation saw
these facies as being essentially chaotic whereas they are now thought to be more
ordered and predictable. 3D seismic data is of central importance in the definition
of the structural geometries where the bedded and disturbed strata are shown
on a seismic section in Figure 4. Several seismic techniques were applied including
attribute analysis, rock physics and amplitude analysis, seismic facies analysis of
time slices and conventional reflector mapping. The resulting 70 internal slump
and fault blocks are shown in Figure 7.
5. Petrophysics-based facies: Seven flow controlling depositional facies were
identified as shown in Figure 8 with rock properties related to grain size (lithology,
typical log response, net/gross, k vs. , Pc and kro-krw). Depositional facies types
present are - marine turbidites and debris flow sands; lower delta plain tidal channels
and lagoonal sands; shallow marine upper shoreface and lower shoreface sands and
shelf shales (best are turbidites, shoreface and channel sands - comprise 80% pore
volume in oil column).
6. Layering and Reservoir Simulation Model: Vertical layering is shown
schematically in Figure 9 and the areal grid is given in Figure 12, with a set of rock
property maps for a single simulation layer given in Figure10. Grid is
Nx x Ny x Nz = 93 x 40 x 18 (67,000 blocks) with most oil leg cells being z = 10ft.
to resolve the gravity stable gas front. Rock property slices were loaded into the 3D
modelling software to connect up the stratigraphic layers (using new but unclear
developments by authors) as shown in Figure 11.
33
From the above field examples (Cases 1 -3), there is clearly a progression in the
engineering approach, the degree of reservoir description and the computational
capabilities as we go from reservoir simulation in the late 1970s to the present
time.
The main changes are as follows:
Computer power: There has been a vast increase in computer processing power
over this period because of :
(a) CPU: The growth of powerful CPU (central procesing units - i.e. chips) especially
as implemented in Unix machines (workstations) and RISC technology and
more recently by the development of modern PCs. The corresponding cost of
computing has fallen dramatically. A graph of processing power (Mflops/s) vs.
time and a corresponding graph of maximum practical model size vs. time is
shown in Figure 10:
1000000
Gridblocks
Mflops/s
1000
100000
100
10
1
1000
0.1
1970
10000
1975
1980
1985
1990
1995
Year
(a) State of art CPU performance
2000
100
1960
1970
1980
Year
1990
2000
(b) Parallel Processing: Part of the increase in computing power referred to above
is the growth of parallel processing in reservoir simulation. The central idea
here is to distribute the simulation calculation around a number of processors
( or nodes) which perform different parts of the computational problem
simultaneously. A bank of such processors is shown in Figure 11 (from the
34
Figure 10
(a) CPU performance
(Mflops/s) vs. time and (b)
maximum practical model
size vs. time; Mflop/s =
mega-flops per second =
million floating point
operations per second; from
J.W. Watts, Reservoir
Simulation: Past, Present
and Future, SPE Reservoir
Simulation Symposium,
Dallas, TX, 5-7 June 1997.
Figure 11
A cluster of parallel
processors; from Megacell
Reservoir Simulation A.H. Dogru - SPE
Distinguished Author
Series, SPE57907, 2000.
1.2
1
Parallel Simulators
0.8
Figure 12
The impact of parallel
reservoir simulation; from
Megacell Reservoir
Simulation - A.H. Dogru
- SPE Distinguished Author
Series, SPE57907, 2000.
0.6
0.6
Conventional Simulators
0.2
0
1988
1990
1992
1994
1996
1997
1998
35
0.6
Conventional Simulators
0.2
0
1988
1990
1992
1994
1996
1997
1998
Speedup
16
12
12
Figure 13
A cluster of parallel
processors; from Megacell
Reservoir Simulation A.H. Dogru - SPE
16
Number of Nodes
Cluster
SP2
Ideal
Reservoir
Model Size
(Millions of Gridblocks)
History
Length (Years)
CPU Hours
On 64 Nodes On 128 Nodes
Carbonate
1.2
27
1.7
Sandstone
1.3
49
4.5
2.5
Carbonate
(With gas cap)
3.9
10
2.0
Carbonate
2.5
2.5
4.0
Figure 14
The type of reservoir
simulation that becomes
more possible with parallel
processing. Comparison
with fine grid and megacell
simulation which identifies
the scale of remaining oil
in a reservoir displacement
process; from Megacell
Reservoir Simulation A.H. Dogru - SPE
Distinguished Author
Series, SPE57907, 2000.
(e) Linked to this increased power, is the ability to handle huge geocellular models
and somewhat smaller but still very large reservoir simulation models.
Geostatistics: there have been significant advances in the application of geostatistical
techniques in reservoir modelling. Such approaches were quite well known in the
mining, mineral processing and prospecting industries but only in the last 10 to 15
years have they been specifically adapted for application in petroleum reservoir
modelling. Introductory texts are now available such as:
Both pixed-based point geostatistical techniques and object based modelling have
been developed and applied in various reservoirs.
Upscaling: There have been a number of advances in approaches to upscaling (or
pseudo-isation) from fine geocellular model the reservoir simulation model. This
still an area of active research and the debate is still in progress on the question:
The basic idea of upscaling has been introduced in the SPE examples. Upscaling is
dealt with in much more detail in Chapter 7.
Organisational changes in the oil industry: A number of major organisational
changes have occurred in the oil industry since the 1970s which have affected the
practice of reservoir simulation. The main ones are as follows:
(a) Many companies have taken a more integrated geophysics/geology/engineering
view of reservoir development and many studies have made a central virtue of this
by organising reservoir studies within more multi-disciplinary asset teams (e.g.
SPE25006 clearly shows a strong integration of geology and engineering);
(b) There have been significant organisational changes in the structure of the
industry given the sucessive rounds of downsizing and outsourcing that have
occurred. For example, see the short article by Galas, The future of Reservoir
Simulation, JCPT, p.23, Vol. 36 (1), January 1997, which is reproduced in
Appendix B. This article has an interesting slant from the point of view of the
smaller consultant and it makes a number of interesting observations;
(c) There have been a number of major mergers and take-overs recently which
have formed some very large companies e.g. BP (BP - Amoco - ARCO), ExxonMobil, Total-Fina-Elf. Likewise, a number of very low cost operations have
grown up which may specialise in the successful (i.e. profitable) exploitation of
mature assets e.g. Talisman, Kerr-McGee etc. How these changes will affect
the future of reservoir simulation remains to be seen.
37
(d) Up until the late 1970s, almost every major and medium sized oil company
had a research centre where programmes of applied R&D were carried
out by oil company personnel. The view was essentially that this in-house
technology development would give the company a competitive and commercial
edge in reservoir exploration and development. Most companies have greatly
reduced the amount of in-house R that takes place and have focused much
more heavily on shorter term asset related D. Many companies do support
research in universities and other independent outside organisations - they also
ally themselves with service companies in order to have their R&D needs met
in certain areas. Again, the situation is in flux and the longer term effects of
this change is yet to be seen.
(e) More specific to reservoir simulation is the fact that, in the 1970s, most
companies would have had their own numerical reservoir simulator which was
built (programmed up) and maintained in-house. To this day, a few companies
still do. However, most oil companies use specialised software service companies
to supply their reservoir simulation (and visualisation, gridding etc.) software.
Again, the relative merits and demerits of this will emerge in the coming
years.
Detailed technical advances: In addition to the changes discussed above, many
advances have been made over the past 50 years on how we perform the simulations
i.e. on the formulation and numerical methods etc. Our practical capabilities have
also expanded greatly as discussed above. Table 1 presents a list of capabilities and
major technical advances in reservoir simulation over the last 50 years; this table
was adapted from two tables in J.W. Watts Reservoir Simulation: Past, Present and
Future, SPE38441, SPE Reservoir Simulation Symposium, Dallas, TX, 5-7 June
1997. This article is well worth reading. Most of the technical details in the advances
listed in Table 1 are beyond the scope of this course and the introductory student does
not need to have any in-depth knowledge on these.
38
Decade
Capabilities
References
1950s
Two dimensions
Two incompressible phases
Simple Geometry
1960s
Three dimensions
Three phases
Black-oil fluid model
Multiple wells
Realistic geometry
Well coning
Sheldon et al (1960)
Stone and Garder (1961)
Lantz (1971)
Stone (1968)
Compositional
Miscible
Chemical
Thermal
1970s
Table 1
Capabilities and major
technical advances in
reservoir simulation over
the last 50 years (adapted
from two tables in J.W.
Watts Reservoir
Simulation: Past, Present
and Future, SPE38441,
SPE Reservoir Simulation
Symposium, Dallas, TX, 5-7
June 1997; (all references
are given in Appendix A)
Technical Advances
Todd et al (1972)
Price and Coats (1974)
Spillette et al (1973)
Kyte and Berry (1975)
Thomas et al (1976)
Meijerink and Van der Vorst
(1977)
Vinsome (1976)
Peaceman (1978)
Yanosik and McCracken (1979)
1980s
Code vectorization
Nested factorization
Volume balance formulation
Young-Stephenson formulation
Adaptive implicit method
Constrained residuals
Local grid refinement
Cornerpoint geometry
Geostatistics
Domain decomposition
1990s
Code parallelization
Upscaling
Voronoi grid
Heinemann et al (1991)
Palagi and Aziz (1994)
Figure 15
A single forward prediction
of the oil recovery
production profile for a
given reservoir.
2000
2005
2010
Time (Year)
39
Clearly, we cannot trust this single curve since there is a considerable amount of
uncertainty associated with it for various easily appreciated reasons. The main
contributors to this uncertainty are to do with lack of knowledge about the input
data although the modelling process itself is not error free. A list of possible sources
of error is as follows:
Lack of knowledge or wide inaccuracies in the size of the reservoir; its areal
extent, thickness and net-to-gross ratios
Lack of knowledge about the reservoir architecture i.e. its geological structure
in terms of sandbodies, shales, faults, etc.
Inaccuracy in the fluid properties such as viscosity of the oil (o), formation volume
factors (Bo, Bw, Bg), phase behaviour etc., or doubts about the representativity
of these properties
Because the representational reservoir simulations model may be poor, e.g. the
numerical errors due to the coarse grid block model may significantly affect the
answer in either an optimistic or pessimistic manner.
The above list of uncertainties for a given reservoir, especially at the appraisal stage,
is really quite realistic and is by no means complete. As we have noted elsewhere,
it is at the appraisal stage when, although the future reservoir performance is at its
most uncertain, we must make the biggest decisions about the development and hence
speed most of our investment money.
At a first glance, the task of doing something useful with reservoir simulation may
seem quite hopeless in the face of such a long list of uncertainties. No matter how
bleak things look, the only two options are to give up or do something, and reservoir
engineers never give up! We must produce an answer - even if it is an educated
guess (or even just a guess) - and some estimate of the sort of error sound that we
might expect.
Before considering what we can do in practice, let us first consider what the answer
might look like for the case above in Figure 15. Figure 16 gives some idea of what
is required:
40
Figure 16
Outcome of reservoir
simulation calculations
showing a range of
recoveries for various
reservoir development
scenarios.
2000
2005
2010
Time (Year)
The results in Figure 16 can be understood qualitatively without worrying about how
we actually obtain them right now. Our single curve in Figure 15 may becomes a
most probable (or base case) future oil recovery forecast. The closer set of outer
curves is the range of future outcomes that can be expected with a 50% probability.
That is, there is a probability, p = 0.5, that the true curve lives within this envelope
of curves shown in Figure 16. Such results allow economic forecasts to be made with
the appropriate weights being given to the likelihood of that particular outcome. A
company can then estimate its risk when it is considering various field development
options.
In fact, here we will just discuss doing some simple sensitivities to various factors in
the simulation model. We can think of a given calculation as a scenario. Therefore,
we can set up various scenarios based on our beliefs about the various input values
in our model and we simply compare the recovery curves for each of the cases. For
example, suppose we have a layered reservoir as shown in Figure 17 which we think
has a field-wide high permeability streak set in background of 100 mD rock.
INJECTOR
Figure 17
This shows a layered
reservoir where we have
some uncertainties in the
various parameters such as
the permeability (khi ), the
thickness (Zhi ) and the
porosity (hi ) in the high
permeability layer.
PRODUCER
1000ft
1000ft
klow = 100mD
100ft
= 0.18
High Permeability Streak,
khigh, hi
Zhi
41
Permeability, khi
Thickness, Zhi
Porosity, hi
Low Value
Mean Value
High Value
400 mD
800 mD
1600 mD
20 ft
0.18
30 ft
0.22
40 ft
0.26
Even with just the three uncertainties in this single model, we can see that there are
3x3x3 = 27 possible scenarios or combinations of input data for which we could
run a reservoir simulation model. Alternatively, we could conclude that some input
combinations are unlikely (e.g. lower permeability with higher value of porosity) and
we could reduce the number. We could simply keep the mean value of two of the
factors while varying only the third factor, leading to 7 scenarios to simulate. Taking
this view, we can take some measure of the oil recovery e.g. cumulative oil produced
(predicted) at year 2010. The notional results could be entered in Table 2
Changed Input
Value
Oil Initially in
Cumulative
% Change in
Change in Recovery
Place (OIIP)
(res. bbl)
Recovery at
Year 2010 (stb)
Base Case
khi = 400 mD
khi = 1600 mD
Table 2
Results of sensitivity
simulations described in the
text.
Zhi = 20 ft
Zhi = 40 ft
hi = 0.18
hi = 0.25
Note: The OIIP will vary somewhat from case to case since the thickness of the high
permeability layer and its porosity both change.
In Table 2, we have noted the % change in the varied parameters relative to its base
case value. Not that different physical quantities such as k and , vary by different
percentages for realistic min./max. values. A useful way to plot the variation in recovery
is against this % change in input value since all three factors can be represented on the
same scale in a so-called spider diagram. Such a plot is shown in Figure 18.
Porosity
Permeability
X
X
Layer Thickness
% Change in Parameter
42
Figure 18
Spider diagram showing
the sensitivity of the
cumulative oil to various
uncertainties in the
reservoir model parameters
(khi; Zhi; hi) in Figure 17.
This type of spider plot is very useful since it displays the effect of the different
uncertainties on the outcome. It clearly highlights which is the most important input
quantity (of those considered) and has the most impact on the result. Thus, if we
were going to spend time and effort on reducing the uncertainty in our predictions,
then this tells us which quantity to focus on first. Indeed it ranks the effects of the
various uncertainties.
There are more sophisticated ways to deal with uncertainty in reservoir performance
but these are beyond the scope of the current course. The basic ideas presented above
give you enough to go on with in this course.
43
Figure 19
The sequence of saturation
distributions as the flood
front moves through the
reservoir. From Res_
Sim_D1.ppt Down arrow
injector, up arrow producer.
44
Different types of simulator are available to model these different types of reservoir
recovery process. Throughout the chapters of this course we will focus on the
simplest of these (which is quite complex enough!) known as the "Black Oil Model".
However, for completeness, we will also list the others and present a table comparing
experience of these various models.
The Black Oil Model: This model was used in the three SPE field case studies above
and is the most commonly used formulation of the reservoir simulation equations
which is used for single, two and three phase reservoir processes. It treats the three
phases - oil, gas and water - as if they were mass components where only the gas
is allowed to dissolve in the oil and water. This gas solubility is described in oil
and water by the gas solubility factors (or solution gas-oil ratios), Rso and Rsw,
respectively; typical field units of Rso and Rsw are SCF/STB. These quantities are
pressure dependent and this is incorporated into the black oil model.
A simple schematic of a grid block in a black oil simulator is presented in Figure
20 showing the amounts of mass of oil, water and gas present. Note that, because
the gas is present in the oil and water there are extra terms in the expression for the
mass of gas. These mathematical expressions for the mass of the various phases are
important when we come to deriving the flow equations (Chapter 5).
Reservoir processes that can be modelled using the black oil model include:
45
Vp.osc
Mass oil =
Bo
So;
Mass water =
Mass gas =
Vp.wsc
Bw
Sw
Vp.gsc
(Sg + So.Rso + Sw.Rsw)
Bg
Rock
Gas, Sg
Oil + Gas, So
Water + Gas, Sw
free gas
gas in oil
gas in water
Gas injection with oil mobilisation by first contact or developed (multi- contact)
miscibility (e.g. in CO2 flooding).
46
Figure 20
Schematic of a grid block
in a black oil simulator
showing the amounts of
mass of oil, water and gas
present. Note that, because
the gas is present in the oil
and water there are extra
terms for the mass of gas;
pore volume = Vp = block
vol. x ; osc, wsc. and gsc
are densities at standard
conditions (60F and 14.7
psi); Bo, Bw and Bg are the
formation volume factors;
Rso and Rsw are the gas
solubilities (or solution
gas/oil ratios).
Component concentrations
in each phase:
ROCK
Figure 21
The view of phases and
components taken in
compositional simulation.
Cij - is the mass
concentration of component
i in phase j (j = gas, oil or
water) - dimensions of mass/
unit volume of phase; pore
volume = Vp = block vol. x
Phase Labels:
j = 1 = Gas
j = 2 = Oil
j = 3 = Water
GAS
Sg
OIL + GAS
So
WATER + GAS
Sw
Gas:
Oil:
The Chemical Flood Model: This model has been developed primarily to model
polymer and surfactant (or combined) displacement processes. Polymer flooding
can be considered mainly as extended waterflooding with some additional effects in
the aqueous phase which must be modelling e.g. polymer component transport, the
viscosification of the aqueous phase, polymer adsorption, permeability reduction
etc. Surfactant, flooding however, involves strong phase behaviour effects where
third phases may appear which contain oil/water/surfactant emulsions. Specialised
phase packages have been developed to model such processes. For economic reasons,
activity on field polymer flooding has continued at a fairly low level world wide
and surfactant flooding has virtually ceased in recent years. However, if economic
factors were favourable (a very high oil price), then interest in these processes may
revive. Extended chemical flood models are also used to model foam flooding.
Examples of reservoir processes that can be modelled using a chemical flood model
include:
47
Foam flooding where a surfactant is added during gas injection to form a foam
which has a high effective viscosity (lower mobility) in the formation than the
gas alone which may then displace oil more efficiently.
Another near-wellbore process that can be modelled using such simulators in water
shut-off using either polymer-crosslinked gels or so-called relative permeabilty
modifiers.
Thermal Models: In all thermal models heat is added to the reservoir either by
injecting steam or by actually combusting the oil (by air injection, for example). The
purpose of this is generally to reduce the viscosity of a heavy oil which may have o
of order 100s or 1000s of cP. The heat may be supplied to the reservoir by injected
steam produced using a steam generator on the surface or downhole. Alternatively,
an actual combustion process may be initiated in the reservoir - in-situ combustion
- where part of the oil is burned to produce heat and combustion gases that help to
drive the (unburned) oil from the system.
Examples of reservoir processes that can be modelled using thermal models
include:
Steam soaks where steam in injected into the formation, the well is shut in for
a time to allow heat dissipation into the oil and then the well is back produced
to obtain the mobilised oil (because of lower viscosity). This is known as a
Huff n Puff process.
Steam drive where the steam is injected continuously into the formation
from an injector to the producer. Again, the objective is to lower oil viscosity
by the penetration of the heat front deep into the reservoir.
The above more complex reservoir simulation models are really based on the fluid flow
process. However, there are also other types of simulator that are more closely defined
by their treatment of the rock structure or the rock response. These include:
Dual-Porosity Models of Fractured Systems: These models have been designed
explicitly to simulate multiphase flow in fractured systems where the oil mainly
flows in fractures but is stored mainly in the rock matrix. Such models attempt to
model the fracture flows (and sometimes the matrix flows) and the exchange of fluids
between the fractures and the rock matrix. They have been applied to model recovery
processes in massively fractured carbonate reservoir such as those found in many
parts of the Middle East and elsewhere in the world. There is quite considerable field
experience of modelling such systems in certain companies but there are also doubts
over the validity of such models to model flow in fractured systems.
48
Coupled Hydraulic, Thermal Fracturing and Fluid Flow Models: These simulators
are still essentially at the research stage although there have been published examples
of specific field applications. The main function of these is to model the mechanical
stresses and resulting deformations and the effects of these on fluid flow. This is
beyond the scope of this course although, in the future, these will be important in
many systems.
We now consider what field experience exists in the oil industry with the various
models from the black oil model through to more complex fracture models and in situ
combustion models etc. The vast majority of simulation studies which are carried
out involve the black oil model. However, there are pockets of expertise with the
various other types of simulation model, depending on the asset base of the particular
oil company or regional expertise within regional consultancy groups. For example,
there is (or until recently, was) a concentration of expertise in both California and
parts of Canada on steam flooding since this process is applied in these regions to
recover heavy oil; in the Middle East (and within the companies that operate there)
there is great competence in the dual-porosity simulation of fractured carbonate
reservoirs.
Table 3
This is an adapted version
of a table in Chapter 11 of
Mattax and Dalton (1990).
This gives some idea of
the problems and issues
encountered in applying
advanced simulation models
relative to applying a black
oil simulator. The view
about the difficulties and
computer time consuming
these are is somewhat
subjective.
Degree of Difficulty
Relative Computing
Costs
Processes Modelled
Primary depletion
Waterflooding
Immiscible gas
injection
Imbibition
Routine
Cheap = 1
Huge
But there are still
challenges with
upscaling of large
models
>90% of cases
Compositional
Model
Gas injection
Gas recycling
CO2 injection
WAG
Difficult
Specialisd
Expensive
(x3 - x20)
Moderate
High in certain
companies
Coats, (1980a),
Acs et al (1985),
Nolen (1973),
Watts (1986),
Young and Stephenson
(1983).
Compositional
Model- Near Crit.
Difficult
Very expensive
(x5 - x30)
Low to moderate
Chemical Model
- Polymer
Polymer flooding
Near-well water
shut-off
Moderate
(x2 - x5)
Moderate to large
Bondor et al (1972),
Vela et al (1976),
Sorbie (1991)
Chemical Model
- Surfactant
Micellar flooding
Low tension polymer
flooding
Difficult
Specialisd
Expensive
(x5 - x20)
Low
Mainly research type
pilot floods
Thermal Model
- Steam
Steam soak
(Huff n Puff)
Steam flooding
Expensive
(x3 - x10)
Moderate
High in limited
geographical areas
In situ combustion
processes
Very difficult
Very specialised
Expensive
(x10 - x40)
Very low
Crookston et al (1979),
Youngren (1980),
Coats (1980b)
Amount of Industrial
Experience
Example References2
Simulator Type
as above
49
This book presents an excellent treatment of the mathematical and numerical aspects
of reservoir simulation. It discusses the discretisation of the flow equations and the
subsequent numerical methods of solution in great detail.
SPE Reprint No. 11, Numerical Simulation I (1973) and SPE Reprint No. 20,
Numerical Simulation II (19**).
These two collections present some of the classic SPE papers on reservoir simulation.
All aspects of reservoir simulation are covered including numerical methods,
solution of linear equations, the modelling of wells and field applications. Most of
this material is too advanced or detailed for a newcomer to this field but the volumes
contain excellent reference material. They are also relatively cheap!
Thomas, G W: Principles of Hydrocarbon Reservoir Simulation, IHRDC, Boston,
1982. This short volume is written - according to Thomas - from a developers
viewpoint; i.e. someone who is involved with writing and supplying the simulators
themselves. The treatment is quite mathematical with quite a lot of coverage of
numerical methods. The treatment of some areas is rather brief; for example, there
are only 7 pages on wells.
APPENDIX A:
REFERENCES
NOTE: SPEJ = Society of Petroleum Engineers Journal - there was an early version
of this and it stopped for a while. Currently, there are SPE Journals in various
subjects but reservoir simulation R&D appears in SPE (Reservoir Engineering and
Evaluation).
Acs, G., Doleschall, S. and Farkas, E., General Purpose Compositional Model,
SPEJ, pp. 543 - 553, August 1985.
Allen, M.B., Behie, G.A. and Trangenstein, J.A.: Multiphase Flow in Porous Media:
Mechanics, Mathematics and Numerics, Lecture Notes in Engineering No. 34,
Springer-Verlag, 1988.
Amyx, J W, Bass, D M and Whiting, R L: Petroleum Reservoir Engineering, McGrawHill, 1960.
Appleyard, J.R. and Cheshire, I.M.: Nested Factorization, paper SPE 12264
presented at the Seventh SPE Symposium on Reservoir Simulation, San Francisco,
CA, November 16-18, 1983.
Archer, J S and Wall, C: Petroleum Engineering: Principles and Practice, Graham
and Trotman Inc., London, 1986.
Aronofsky, J.S. and Jenkins, R.: A Simplified Analysis of Unsteady Radial Gas
Flow, Trans., AIME 201 (1954) 149-154
Aziz, K. and Settari, A.: Petroleum Reservoir Simulation, Elsevier Applied Science
Publishers, Amsterdam, 1979.
51
Bondor, P.L., Hirasaki, G.J and Tham, M.J., Mathematical Simulation of Polymer
Flooding in Complex Reservoirs, SPEJ, pp. 369-382, October 1972.
Clayton, C.A., et al, The Ubit Field Rejuvenation: A Case History of Reservoir
Management of a Giant Oilfield Offshore Nigeria, SPE49165, presented at the SPE
Annual Technical Conference and Exhibition, New Orleans, LA, 27-30 September
1998.
Coats, K.H., .......... 1969 - tools of res sim
Coats, K.H., A Highly Implicit Steamflood Model, SPEJ, pp. 369-383, October
1978.
Coats, K.H., An Equation of State Compositional Model, SPEJ, pp. 363-376,
October 1980a; Trans. AIME, 269.
Coats, K.H., In-Situ Combustion Model, SPEJ, pp. 533-554, December 1980b;
Trans. AIME 269.
Coats, K.H., Dempsey, J.R., and Henderson, J.H.: The Use of Vertical Equilibrium
in Two-Dimensional Simulation of Three-Dimensional Reservoir Performance, Soc.
Pet. Eng. J. 11 (March 1971) 63-71; Trans., AIME 251
Craft, B C, Hawkins, M F and Terry, R E: Applied Petroleum Reservoir Engineering,
Prentice Hall, NJ, 1991.
Craig, F F: The Reservoir Engineering Aspects of Waterflooding, SPE monograph,
Dallas, TX, 1979.
Crichlow, H B: Modern Reservoir Engineering: A Simulation Approach, PrenticeHall Inc., Englewood Cliffs, NJ, 1977.
Crookston, R.B., Culham, W.E. and Chen, W.H., A Numerical Simulation Model for
Thermal Recovery Processes, SPEJ, pp. 35-57, February 1979; Trans. AIME 267.
Dake, L P: The Fundamentals of Reservoir Engineering, Developments in Petroleum
Science 8, Elsevier, 1978.
Dake, L P: The Practice of Reservoir Engineering, Developments in Petroleum
Science 36, Elsevier, 1994.
Fanchi, J R: Principles of Applied Reservoir Simulation, Gulf Publishing Co.,
Houston, TX, 1997.
Fantoft, S., Reservoir Management of the Oseberg Field After Four Years, SPE25008,
proceedings of the SPE European Petroleum Conference, Cannes, France, 16-18
November 1992.
Giudicelli, C.S., Massonat, G.J. and Alabert, F.G., Anguille Marine, a DeepseFan Reservoir Offshore Gabon: From Geology Toward History Matching Through
52
53
Simulation, Soc. Pet. Eng. J. 18 (June 1978) 183-194; Trans., AIME 253.
Peaceman, D.W. and Rachford, H.H.: The Numerical Solution of Parabolic and
Elliptic Differential Equations, Soc Ind. Appl. Math. J. 3 (1955) 28-41
Peaceman, D W: Fundamentals of Numerical Reservoir Simulation, Developments
in Petroleum Science No. 6, Elsevier, 1977.
Ponting, D.K.: Corner point geometry in reservoir simulation, in The Mathematics
of Oil Recovery - Edited proceedings of an IMA/SPE Conference, Robinson College,
Cambridge, July 1989; Edited by P.R. King, Clarendon Press, Oxford, 1992.
Pope, G.A. and Nelson, R.C., A Chemical Flooding Compositional Simulator,
SPEF, pp.339-354, October 1978.
Prats, M., Thermal Recovery SPE Monograph Series No. 7, SPE Richardson, TX,
1982.
Price, H.S. and Coats, K.H.: Direct Methods in Reservoir Simulation, Soc. Pet.
Eng. J. 14 (June 1974) 295-308; Trans., AIME 257
Robertson, G., in Cores from the Northwest European Hydrocarbon Provence, C D
Oakman, J H Martin and P W M Corbett (eds.), Geological Society, London. 1997.
Schilthuis, R.J., Active Oil and Reservoir Energy, Trans. AIME, Vol. 118, p.3,
1936; (original ref. on Material Balance)
Sheldon, J.W., Harris, C.D., and Bavly, D.: A Method for Generalized Reservoir
Behavior Simulation on Digital Computers, SPE 1521-G presented at the 35th
Annual SPE Fall Meeting, Denver, Colorado, October 1960.
Sibley, M.J., Bent, J.V. and Davis, D.W., Reservoir Modelling and Simulation of
a Middle Eastern Carbonate Reservoir, SPE36540, proceedings of the SPE 71st
Annual Conference and Exhibition, Denver, CO, 6-9 October 1996.
Sorbie, K.S., Polymer Improved Oil Recovery, Blakie and SOns & CRC Press,
1991.
SPE Reprint No. 11, Numerical Simulation I (1973) and SPE Reprint No. 20,
Numerical Simulation II (19**).
Spillette, A.G., Hillestad, J.H., and Stone, H.L.: A High-Stability Sequential Solution
Approach to Reservoir Simulation, SPE 4542 presented at the 48th Annual Fall
Meeting of the Society of Petroleum Engineers of AIME, Les Vegas, Nevada,
September 30-October 3, 1973.
Stone, H.L.: Iterative Solution of Implicit Approximations of Multidimensional Partial
Differential Equations, SIAM J. Numer.Anal. 5 (September 1968) 530-558
Stone, H.L.: Probability Model for Estimating Three-Phase Relative Permeability,
J. Pet. Tech. 24 (February 1970) 214-218; Trans., AIME 249.
54
55
Wallis, J.R., Kendall, R.P., and Little, T.E.: Constrained Residual Acceleration of
Conjugate Residual Methods, SPE 13536 presented at the Eighth SPE Reservoir
Simulation Symposium, Dallas, Texas, February 10-13, 1985.
Watts, J.W.: An Iterative Matrix Solution Method Suitable for Anisotropic Problems,
Soc Pet. Eng .J. 11 (March 1971) 47-51; Trans., AIME 251.
Watts, J.W., A Compositional Formulation of the Pressure and Saturation Equations,
SPE (Reservoir Engineering), pp. 243 - 252, March 1986.
Watts, J.W., Reservoir Simulation: Past, Present and Future, SPE Reservoir
Simulation Symposium, Dallas, TX, 5-7 June 1997.
Yanosik, J.L. and McCracken, T.A.: A Nine-Point Finite Difference Reservoir
Simulator for Realistic Prediction of Unfavorable Mobility Ratio Displacements,
Soc. Pet. Eng. J. 19 (August 1979) 253-262; Trans., AIME 267.
Young, L.C. and Stephenson, R.E., A Generalised Compositional Approach for
Reservoir Simulation, SPEJ, pp. 727-742, October 1983; Trans. AIME 275.
Youngren, G.K., Development and Application of an In-Situ Combustion Reservoir
Simulator, SPEJ, pp. 39-51, February 1980; Trans. AIME 269.
This very interesting pair of articles gives a very good broad brush commentary
on a range of technical issues in reservoir simulation e.g. gridding, handling wells,
pseudo-relative permeability, error analysis and consistency checking. The views
are clearly those of someone who has been deeply involved in applied reservoir
simulation. They are well presented and quite individual although again there are
issues that would provoke disagreement. Read this and decide for yourself what you
accept and what you dont.
57