Indus
& Pre
DIALOG
‘Ten Golden Rules for
Simulation Engineers
With the proliferation of reservoir simula-
tors and extensive use ofthis powerful toch-
nology by the industry, there is increasing
danger that inexperienced users will misuse
sophisticated models available to them. To
‘minimize this danger, one should keep cet-
(ain sic rules in mind. The following rules
reflect my own experience and what Ihave
‘ined by reading papers and through dis-
cussions with colleagues in industry and
academia. I would weleome comments that
‘may result in the refinement ofthese rules,
[wrote the frst version ofthese rules in June
1989 and this version in Sept. 1989.
1, Understand Your Problem and De-
fine Your Objectives. Before you do any
simulation, understand the geological char-
acteristics of your reservoir, the fuids it
contains, and its dynamic behavior. Also
clearly site the objectives of your study on
paper before you start. Ask yourself ifthe
‘Objectives are realistic. These considerations
will help you choose the most appropriate
model for your study.
2, Keep It Simple. Start and end with the
simplest model that s consistent with the na-
tre ofthe reservoir, the objective of your
study, and the avalabiiy of data, Classic
reservoir engineering, simple analytical
‘models, or single-block simulations are often
all you need. At other times, the most
‘sophisticated model availabe to you may not
serve your needs. Understand mode! limi-
tations and capabilities.
3, Understand Interaction Between
Different Parts. Remember that a reservoir
not an isolated entity. Iemay connect with
an aquifer and, through it, even to other
reservoirs. Furhermore, reservoirs are con-
nected through wells to the surface facilities.
‘The isolation of different components ofthis
system for separate study often may lead to
inappropriate rosuls by neglecting interac-
tion between different parts ofthe systom.
However, when appropriate, don't be afraid
to break a big problem into its smaller com-
ponents, This can lead not only to substantial
‘savings but to greater understanding of the
mechanisms involved.
4. Don’t Assume Bigger Is Always Bet-
‘ter. Always question the size ofa study that
Js limited by the computer resources or the
IPT + November 1989.
budget. Simulation engineers often believe
‘that no computers big enough for whet they
want to do and tend simply to inerease the
size of their models to fit the computer.
More blocks and components do not suto-
matically translate into greater accuracy and
reliability. In fact, in some situations the
reverse is tric. Insist on seeing appropriate
justification forthe number of blocks used
ina given study.
5. Know Your Limitations and Trust
Your Judgment. Remember that simulation
isnot an exact science. All modes are based
con assumptions and provide only an approx-
imate answer to the real problem. Hence,
«8 good understanding of both the problem
and the model is essential for success. Nu-
rerical approximations may introduce
“pseudophysical” phenomena like numer-
ical dispersion. Use and trust your judg-
ment, especially if it is based on” your
‘analysis of field or laboratory observations
‘Be careful to check your input and output
Do simple material-balance calculations to
‘check simulation results. Pay particular a-
tention to such things as negative compres-
sibilities and permeabilities,
6. Be Reasonable in Your Expectations.
Don't try to get from the simulator what it
{is incapable of producing. Often the most
you can get from a study is some guidance
on the relative merits of choices available
to you. At other times you have the right
to demand a lot more, But remember that
iff you exclude a mechanism during model
development, you cannot study its effect
with that model
7. Question Data Adjustments for His-
tory Matching. Always question data
adjustments during history matching. Re-
‘member that this process does not have a
tig solution. The most reasonable solu
tion will result from paying close attention
to physical and geological reasonableness.
‘A "good! history match with inappropriate
adjustments tothe data will ead 1 poor pre-
dictions. Don't be lulled into false security
by a "good" or “close” match.
8, Don't Smooth Extremes. Pay atten-
sion to extremes in permeability (barriers
and channels). Be careful in the process of
‘averaging to avoid losing essential informa-
tion when averaging the extremes. Never
average out extremes.
9. Pay Attention to the Measurement
‘and Use Scales. Measared values atthe core
scale may not apply directly at the larger
block scale, but measurements do influence
‘values at other scales. Remember that aver-
aging may change the nature of the varia-
bles you average. For example, permeability
may be a scalar at some small scale and a
tensor ata larger scale. Even the meaning
of capillary pressure and relative permea-
Dilty can be different at diferent scales.
‘Also, the dispersive terms in our equations
fare a result of the process of averaging.
10, Don’t Skimp on Necessary Labora-
tory Work. Models do not replace god lab-
‘oratory experiments that ae designed to gin
‘an understanding ofthe nature ofthe process
being modeled or to measure essential pa-
rameters of the equations being solved by
your simulator. Plan your laboratory work
‘With the end use of tis information in mind
Learn how to seale data
1 would like to thank Aziz Odeh and
Roland Horne for their comments on these
rales.
Khalid Aziz, SPE
‘Stanford U.
Technology Summary
(From Page 1195)
Prudhoe Bay Well-Management Routine
Enhances Simulator Performance outlines
the well constrains imposed by a subrou-
tine used in a conventional reservoir sim-
lator. ‘The input data consist of limits on
injection and production facilities and of
relationships between each injector and its
iteracting producers. Three injection
faclienges, seawater, an produced
water—are considered, each with its own
{at limit The subroutine hands any com-
Bination ‘of wells on water of Water
alternating-g injection. Injection and pro-
duction are balanced within each pattern
while simitanously honoring gas pressure,
total gas, water, andl lis at the
producers. Priotization is discussed and ex-
ples are given. The subroutine has made
Tong-term predictive simulations of large,
unbounded, faiity-constrained, multiple”
pater floods practical.
tse