Sie sind auf Seite 1von 6

Reservoir Performance Forecasting:

Acceleration by Parallel Planning


Nansen G. Salerl, SPE, Chevron E&P Services Co.

Summary
This paper evaluates reservoir performance Nansen G. Salerl has been a senior petroleum
forecasting . Actual field examples are dis- engineering consultant with Saudi Aramco in Dhahran .
cussed, comparing past forecasts with ob- since Oct. 1992. He previously was manager of reserVOIr
served performances. The apparently weak engineering at Chevron E&P Services Co. in Houston. His
correlation between advances in technolo- responsibilities at Chevron, which he joined in 1974,
gy and forecasting accuracy is assessed. included worldwide reservoir projects and directing the
Parallel planning is presented as an approach company's reservoir management training programs . He
Saler! holds MS and PhD degrees in chemical engineering from
that can significantly accelerate reservoir
forecasts. The recognition of inevitable fore- the U. of Virginia. Saleri was a 1991-92 Distinguished Lecturer.
casting uncertainties constitutes the
philosophical basis of parallel planning.
effect" example points to inevitable limits Discretization
Introduction of predictability. The analogies between Both geostatistical and finite-difference
weather and reservoirs have been noted 4,5 ; models discretize reservoirs. The two
To say that reservoir performance forecast- specifically, the sensitivity of reservoir per-
models, however, have dissimilar discreti-
ing is not an exact science would be an un- formance and hence forecasts to certain ge-
zation scales (geostatistical models use
derstatement. Even with all the significant ologic parameters (i.e., flow boundary
inches to several feet; finite-difference
advances occurring across a wide spectrum conditions) have been highlighted. This sen-
models use hundreds of feet) .
of related areas, questions still remain sitivity suggests that performance forecasts
Current and projected hardware and soft-
regarding the reliability of reservoir predic- will remain uncertain indefinitely.
ware limitations suggest that the discretiza-
tions. In fact, our efforts today are aimed Both internal and external reservoir fac- tion gap between finite-difference and
as much at defining the limits of uncertainty tors contribute to forecast uncertainties (Fig.
geostatistical models will not disappear for
envelopes as at producing forecasts. The dis- 1). When model forecasts diverge from ac-
giant fields. Consider the multibillion-barrel
cussion here pursues the following ques- tual performance , distinctions among
ATLlINB field in West Africa . A finite-
tions : (1) What are realistic accuracy primary causes are sometimes lost. For ex-
difference model using I_ft3 cells will re-
expectations in performance forecasts? and ample, accurate models may produce appar-
quire about 0.5 trillion cells . The cor-
(2) Are our conventional thought process- ently poor forecasts when presumed field
responding figure for the Safaniya field in
es in modeling inherently ill-structured to management strategies and facility outlays
the Middle East is about 7 trillion cells. Cells
produce rapid forecasts? are not actually implemented as a result of (1 in. 3 ) will suggest models with roughly
By their very nature, EOR processes in- external factors.
800 trillion cells for the ATLlINB field.
troduce additional levels of complexity to When model forecasts duplicate actual These numbers imply our indefinite need for
forecasting . This discussion relates mainly performance , this can also be misinterpret-
a scale-up process and hence the resulting
to conventional reservoir systems. ed as model validation. In fact , the duplica- uncertainties.
tion could simply reflect compensating
Forecasting Methods errors among the internal and external fac- Homogenization
and Uncertainty tors. The point here is that accurate fore-
An obvious outcome of the scale-up process
Methods and Limits. Current reservoir ' casts do not mean accurate models. (The is homogenization. Porosity/permeability
performance forecasting methods can be hypothetical corollary also appears notewor- transforms, often used to describe permea-
classified into two broad categories: empir- thy: poor forecasts do not necessarily equate bility fields, also contribute to homogenized
ical and mathematical. This paper focuses to poor reservoir models.) The nature of the property assignments in simulation models.
on finite-difference methods because they oil industry limits predictability of external Fig. 2 gives the porosity/permeability core
represent the predominant industry-wide ve- factors, such as exact field operating prac- data for the ATLlINB field. This field ex-
hicle in reservoir evaluations. Empirical tices. At best, multiple forecasts need to be hibits a complex lithology of predominant-
methods , such as decline curves, are use- developed for a range of external factors . ly silica sands intermixed with dolomite. The
ful, yet they have a limited application do- Of the four uncertainty causes in Fig. I, data use of a single-variable transform, represent-
main. Continuation of past production quality and mathematical solutions are be- ed by the solid line, filters the observed
practices and mechanisms is a precondition coming less pronounced, and reservoir variability in the core data. An alternative
for forecast reliability. Likewise, hybrid characterization and scale-up present the approach that would reduce the homogeni-
methods , 1,2 while suitable for a wide class primary obstacles to improving performance zation effect is the use of "cloud trans-
of problems (e.g., miscible, pattern floods) forecasts . forms " developed by Kasischke and
have not yet fully matured to offer a univer- The lack of determinism in both external Williams. 8 Cloud transforms produce
sal forecasting capability. and internal factors suggests only the obvi- property representations in models that can
Lorenz 3 recognized the stochastic nature ous: all reservoir performance forecasts car- mimic distributions observed in real data
and hence the inherent limitations of weather ry a band of uncertainty. Ballin et aL 6 (e.g., cores or logs) . Fig. 3 shows a sam-
forecasting. Lorenz's celebrated " butterfly attempted to quantify this uncertainty for a ple distribution generated by a cloud trans-
special class of problems. Haldorsen and form for the Elk Hills 26 R reservoir. The
'Now with Saudi Aramco. Damsleth 7 described a general methodolo- reservoir is an Upper Miocene submarine
Copyright 1993 Society of Petroleum Engineers gy for producing stochastic forecasts. channel deposit.

July 1993 JPT


Sources of
Uncertainty in Forecasting
'-~~~~~~-'~~~~-l~~'I~~~1~~~~------,

Reservoir Scale-Up ! TABLE 1-ATL FIELD CUMULATIVE


Characterization L -_ _ _ _--.J : PRODUCTION (JAN. 1, 1987 TO
I
I JAN. 1, 1991), MMSTB/Bscf
I

External Difference
Actual Model (%)
Factors ~-
~~

Oil 97.5 102.7 4.7


Water 8.0 6.9 14.1
Gas 73.6 85.9 16.6
Fig. i-Uncertainty sources.

All forecasting is reservoir-specific and poorly (Fig. 5). In fact, from gas produc- methods in well performance forecasting.
the historical comparisons given below are tion comparisons, 43 % of the total wells This assessment, although discomforting,
not meant to represent a statistical sampling were considered "busts." probably will remain unchanged for the
of fields. They are interesting examples be- The poor well performance forecast is foreseeable future.
cause (1) they offer two well-studied fields neither coincidental nor isolated. Its coin- Factors 1 through 3 are not strongly sen-
and (2) their accuracy aspects are fairly con- cidence with a fairly successful fieldwide sitive to ongoing advances in reservoir
sistent with industry-wide observations. In match points to a certain dichotomy between characterization. This observation also ex-
this sense, they are not atypical. field and well forecasts. In the latter, un- plains the apparent inelasticity between tech-
certainties appear to be much more ampli- nology changes and improvements in
ATL Field fied. The increased difficulty in well forecast reliability for wells. The same is
The ATL field has been on production since performance forecasting is caused by several true to a lesser degree for fieldwide fore-
Dec. 1982, with a production rate of 80,000 factors. casts where compensating errors for in-
BOPD as of Jan. 1993. It contains 33 API0 1. Larger near-wellbore potentials, which dividual wells tend to dampen field wide
crude with no associated gas cap. The field amplify the effects of slight reservoir or error amplitudes. Factors 2 and 3, for in-
has roughly 40 producers. wellbore misrepresentations. stance, can be predicted more accurately
A reservoir simulation study was under- 2. Manmade factors (poor cement jobs, fieldwide than well by well, and thus the dis-
taken during 1986-87 to predict reservoir casing leaks, stimulation effects) that can parity favors field forecast accuracies over
performance under natural depletion and seldom be described or anticipated fully. those for wells.
water injection and to assist in the planning 3. Sensitivity of well performance to sur- A distinction can also be made between
of a regional water injection project. The face conditions (e.g., backpressures), which long- and short-term forecasting. The ac-
model was history matched through March can seldom be known on a well-to-well ba- curacy of short-term predictions is more sen-
1987 and calibrated to duplicate field sis for sustained periods. sitive to near-well factors under a set of
production rates. Several prediction cases 4. Frequent absence of high-quality well external factors. Thus, one can expect, para-
also were conducted. We now can evaluate production data. Often, individual well rates doxically perhaps, better success in estimat-
the 1987 forecast accuracies by examining are allocated estimates based on metered ing ultimate recoveries than in making a
4 years (Table 1) of production information flow information on trunklines, flowlines, IO-year production forecast.
(Fig. 4). etc. The ATL field example also underscores
Fieldwide, model forecasts appear to Factor 1 is not unlike Lorenz's butterfly how external factors, often unpredictable,
match the ATL production/pressure perfor- effect. A minor perturbation in initial reser- affect forecasts. The 1987 study assumed a
mance during 1987-91 reasonably well. voir descriptions leads to a significant diver- waterflood startup of mid-1989 for the water
This is not unexpected. given the continua- gence in performance. This apparent injection evaluation. Yet the program actu-
tion of an operational mode (natural deple- forecasting "instability" is magnified by ally began in Dec. 1990. Because of the dis-
tion) on which the model was history near-wellbore potentials. In a larger sense, placement in the startup date, it is difficult
matched and calibrated. On a well-by-well the net effect of these factors is to limit se- to gauge the true success of the 1987 model
basis, individual performance forecasts fared verely the reliability of finite-difference in matching waterflood performance.

104 , - - - - - - - - - - - - - - - .

"0
E
~ 102
:0
CII
Q)
E
CD
a..

.05 .15 .25 .35


Core Porosity 0.00 0.10 0.20 0.30
Porosity

Fig. 2-ATLIINB porosity/permeability transform. Fig. 3-Cloud transform.

JPT. July 1993 653


TABLE 2-NINIAN FIELD ACTUAL
VS. MODEL DIFFERENCE

Difference *
(%)
1976 1985
Based On Model Model
-- --
Cumulative oil 7.6 16.5
production
Cumulative water 27.6 8.9
production
Cumulative water 53.0 5.5
injection
Peak water injection 32.0 14.6
'The 1976 model is based on data from Jan. 1,
1979, to June 1, 1991; the 1985 model is based
on data from Jan. 1, 1984, to June 1, 1991. The
difference is calculated as (actual
modellactual)100%.

Fig. 4-Actual vs. forecast performance (1987-90), ATL field.

Ninian Field pIing of fields or forecasts. The reported difference models still offer the most effec-
The Ninian field in the North Sea was dis- range of forecast accuracies, however; is tive tools.
covered in Sept. 1973. The reservoir is in consistent with other field performance
the Middle Jurassic Brent Group. Six ap- predictions: Casablanca (offshore Spain), Acceleration by Parallel
praisal wells followed the two discovery Anoa (Indonesia), Vuko (Angola), Mibale Planning
wells to establish the extent of the field. In (Zaire), among others. Some generaliza- This discussion addresses two central ques-
May 1990, 105 wells remained active (73 tions, albeit SUbjective, can be made. tions in finite-difference modeling: (1) Why
producers, 32 injectors). The multibillion- 1. Barring major geologic and/or reser- do modeling studies often fail to meet es-
barrel field produced 93 MSTBID in June voir data limitations, fieldwide cumulative
tablished deadlines? and (2) Are our conven-
1991, with an average water cut of 86 % . production forecast accuracies would tend
tional thought processes in modeling
The Ninian field has been the subject of to range from 10% to 40%. Availability of
inherently flawed? Two approaches-
several finite-difference modeling studies. field performance information can push
sequential planning and parallel planning-
The 1976 model was concluded 3 years be- forecasts to the lower end of the spectrum.
Accuracies below 10% are more likely to are discussed. The recognition of inevita-
fore the field came on stream in 1979. The ble forecasting uncertainties, as noted in the
reservoir and geologic information was be artifacts of compensating errors or luck
than precise engineering or geology. field examples, constitutes the philosophi-
limited to six wells. The 1985 model incor-
2. Well performance forecasts are bound cal basis for parallel planning.
porated actual production/injection history
(1979-85). to be less successful than field wide
Table 2 and Figs. 6 and 7 present graphi- predictions. Sequential Planning Approach. Historical-
cal comparisons of actual performance vs. These observations are not meant to ly, finite-difference reservoir simulation
the 1976 and 1985 model forecasts. Differ- diminish the value of finite-difference studies have been conducted with a sequen-
ences in cumulative production and injec- models. While a 10% to 40% forecast un- tial planning approach (Fig. 8). Its under-
tion varied from 5.5% to 15.5% and from certainty may appear alarming in an abso- lying philosophy is a linear logic that
7.6% to 53% for the 1976 and 1985 models, lute sense, the majority of reservoir assumes a steady progression from raw data
respectively. engineering decisions require choices based to final forecasts in five modular steps.
solely on comparative analyses (e.g., Step 1. Data analysis.
Review. As noted, the two field examples peripheral vs. pattern flood). Thus, in select- Step 2. Model construction.
are not meant to represent a statistical sam- ing optimum management strategies, finite- Step 3. History matching.

Model vs. Actual (1987 - 1990)

Water Oil Gas


Good 8% Good 31%
Bust26~
"Parallel planning is a
flnltedifference
modeling strategy that
Bust 58% \....J Bust 43%

can be considered the Fair 43%


antithesis of sequential
~ o{1! Differen!:<e
planning." Good <10
Fair 10 to 50
Bust >50
Model - Actual
% Difference = I I x100
Actual

Fig. 5-Well forecast quality based on cumulative production.

654 July 1993 JPT


500 1400
en 1985
.. " 1976 ~ 1200
~
o
400
oco ,:ir '-'.... ( ~ 1000
6300
./ \
"0
OJ
0
.... 800
tii a..
.... 600
~ 200 ~
"
o E 400 //,,/ t..
100
:::J
()
200
",,"<~ . . . . . . . ..
................ 1976
Actual
o 0
79 81 83 85 87 89 91 79 81 83 85 87 89 91
Time, Year Time, Year

Fig. 6-Actual vs. model forecasts, oil rate vs. time. Fig. 7-Actual vs. model forecasts, cumulative water produc-
tion vs. time.

Step 4. Calibration. translate directly to forecast delays. In fact, varied from a few weeks to 6 months.
Step 5. Predictions. any delay in the entire process causes a cor- Scheduling requirements necessitated the use
In sequential planning, each step can be responding delay in the final delivery. of unconventional planning methods.
started only after the preceding steps are Critical Flaws. All reservoir modeling Parallel planning emerged as an unorthodox,
completed. The process is sequentially con- studies, regardless of their complexity, are albeit experimental, solution. Advances in
ditioned. Thus, predictions will begin after likely to encounter some type of problem. hybrid simulation offer cost-effective solu-
calibration, which can begin only after all In sequential planning, success depends on tions for a special class of problems. Their
history-matching steps are finished. Also, the occurrences (or absences) and locations emergence also underscores the need for
this method presupposes that conditions and of critical flaws. An extreme case may in- more rapid turnaround times in forecasting.
information available at the inception of the volve a 4O,OOO-cell model, successfully con- In essence, parallel planning has five as-
modeling work will remain unchanged (i.e., structed and history matched, that displays sumptions.
static modeling). The noted dual attributes severe instabilities in the prediction mode. 1. Preliminary forecasts are a business
of sequential planning (sequential condition- In essence, this is a fatal flaw because necessity.
ing and static modeling) manifest themselves predictions cannot be produced owing to ex- 2. Problems will arise in all five phases
in several ways: cessive time requirements and/or computer of modeling.
1. Delays accumulate. Each "problem" costs. 3. It is not possible to predict a priori
step delays progress on all subsequent steps. New Data Incorporation. Sequential plan- where and when these problems will occur.
2. Critical flaws can be identified only at
ning offers little or no flexibility with respect
4. New information probably will become
to incorporating information that becomes
the arbitrary time that they are discovered. available while the study is in progress.
available while studies are in progress. For
3. When new data become available (e.g., 5. Identification of critical flaws is of
instance, the introduction of new reservoir
PVT or relative permeability information), paramount importance in all studies.
data during the prediction step suggests that
the entire sequential process must be repeat- prediction runs be discontinued and Steps Parallel planning has several architectur-
ed to produce forecasts. 2 through 5 be repeated. The resulting de- al and conceptual analogs. Beta testing in
These factors impose severe handicaps in lays are evident. software installations is one such conceptu-
producing forecasts, namely increased al example. Parallel processing in computer
project times and costs and delayed decisions Parallel Planning Approach. Parallel plan- hardware offers the architectural analog.
(thus missed opportunities). ning is a finite-difference modeling strate- Parallel planning uses several (as opposed
Delay Accumulation. Delays occur be- gy that can be considered the antithesis of to single) channels to advance from raw in-
cause each step depends on all preceding sequential planning. Its conceptual origins formation to a final forecast. The method
steps. Often, history matching will not be are based on numerous "rush" modeling uses multiple numerical models: the actual
started because some (but not all) of the rela- studies during the last decade. The rush model and its pseudo model analogs (Figs.
tive permeability or PVT data may be un- evaluations were conducted on a variety of 9 through 11).
available. Similarly, until the history fields to address a number of reservoir The actual model rigorously represents
matching is considered final, no predictions management decisions under strict time reservoir and geologic information and is
are undertaken. Data-acquisition delays schedules. Typically, their project deadlines used to produce the definitive forecasts. The

Planned

-
Analyze
Construct
History Match "Well performance
Actual Calibrate forecasts are bound to
Analyze be less successful than
Construct Predict
-HistOry Match
Calibrate fieldwide predictions."
-Predict
Time
Time

Fig. a-Sequential approach in modeling. Fig. 9-Parallel planning.

JPT. July 1993 655


engineering information. Their function is
Parallel Planning Approach to develop analogous test models quickly to
capture potential software, hardware, model
Actual stability, or other problems. Their numeri-
Model Final cal configuration (grid size/number, orien-
Results
tation, etc.), however, will be identical to


Starl~_...........- the actual model configuration. Pseudo-
models contain synthetic data segments that
are convenient substitutes for actual data be-
cause actual data either are unavailable or
are too time- and resource-consuming to
pseudo . Preliminary
Results construct.


The synthetic data segments Illay repre-
sent any or all of the following reservoir/sur-
face facility parameters.
I. Relative permeability.
A = Analysis MC = Model Construction 2. Permeability (vertical or horizontal).
HM = History Match C = Calibration P = Prediction 3. PVT.
4. Well and surface facility.
5. Geologic description (structure, net
Fig. i0- Parallel planning . sand, etc.).
6. Other (e.g., prediction guidelines).
In parallel planning, several pseu-
domodels may coexist for use at different
times for various reasons. Consider the
hypothetical case of a reservoir evaluation
involving a virgin field using a 30,OOO-cell
(lOOXlOOx3-1ayer) model. One can pro-
ceed with the construction and initialization
of a pseudomodel as soon as the grid dimen-
sions and orientation are determined. This
pseudomodel can have synthetic relative per-
meability and PVT data, assuming that the
actual information will be unavailable for a
while (perhaps 2 to 3 months). Pseudomodel
PVT properties can be approximated with
industry-accepted API correlations, assum-
ing that some limited information exists.
Similarly, neighboring fields on reservoirs
with analogous geologic /lithologic attributes
can be a source for relative permeability
relationships. A gross average permeabil-
ity value can be assigned with appropriate
engineering approximations. Pseudomodel 1,
thus constructed and initialized, will then be
Pseudo Model Actual Model used for test runs.
Any generic problems independent of spe-
cific reservoir parameters can thus be diag-
nosed and corrected early. The majority of
software, hardware, and model stability
problems fall into this category. Similarly,
one can formulate approximate development
scenarios (i.e., production targets, number
of wells, facility constraints) for this virgin
field that can then be used for preliminary
Synthetic Actual forecasts. The value of these early results
are three-fold: (1) to help identify problems,
(2) to assist in formulating more informed
scenarios for the actual model, and (3) to
Preliminary Results Rigorous Results provide first-order estimates of reservoir
performance.
Parallel planning also helps in mature-
field studies where history matching is a sig-
Fig. ii-Data flow in parallel planning. nificant portion (e.g., more than 75%) of
the total effort. Sequential planning would
pseudomodels act as testing tools to identi- nel progression, identification of critical advocate conducting prediction runs only af-
fy and eliminate potential problems. They flaws, and bypassing of "choke" points. Its ter the completion of history matching.
also help produce rapid, yet preliminary primary benefits are acceleration of model- History-matching nonuniqueness and the
(beta), forecasts. The information gained ing, and hence final forecasts, and rapid imprecise forecasts discussed earlier sug-
through pseudomodels is networked into the availability of preliminary forecasts. gest, however, the prudence of using par-
actual model. The main attributes of parallel Pseudomodels. Pseudomodels represent tially history-matched pseudomodels for
planning include pseudomodels, multichan- gross approximations of actual geologic and preliminary forecasts. The benefits are simi-

656 July 1993 JPT


TABLE 3-EXAMPLES OF SIMULATION STUDIES CONDUCTED WITH PARALLEL PLANNING

ANCA LlBWA-7 Elk Hills, 26R* ATUINB


---
Study type Prospect evaluation Horizontal well evaluation Mature field recovery Field development
improvement
Simulator Black oil Black oil (CHEARS) Black oil/compositional Black oil (CRS3D)
(CHEARS)
Cells
Number-Active 3,402 17,800 28,359 14,162
(27 x 42 x 3 layers) (89 x 8 x 25 layers) (50x10x74 layers) (32 x 82 x 8 layers)
History period, years 0 2 15 8
Prediction period, 10/5 5/7 15/5 5/4
years/cases
Number of wells 16 2 13 117
Total elapsed time, 2 1 10* 2
months
Total man-months 4.5 2 17 4
Number of active 3 3 4 2
participants
Objectives of study To evaluate various de- To optimize the fracture- To evaluate the effects To optimize placement
velopment scenarios space design and of heterogeneity on of proposed injection
involving gas and vertical placement of a vertical and areal wells
water injection horizontal well sweep efficiency

'Includes 2'/2 months for developing the geostatistic 1T,0del.

lar to those realized for the virgin-field 2. The apparent inelasticity between im- Parallel vs. Sequential Approach
example. provements in forecasting accuracy and Advantages
technological advances can be explained in Critical Flaws _ Identified Early
Parallel vs. Sequential Planning. Fig. 12 terms of the nature of the causal factors, Delays Do Not Accumulate
shows the principal advantages of parallel e.g., near-wellbore uncertainties and surface Elapsed TIme Less
over sequential planning. One qualification conditions. Overall Cost _ Less
needs to be made: parallel planning involves 3. Actual field examples tend to suggest (Manpower/Computer)
additional manpower at any given time be- that fieldwide cumulative production fore-
cause of the need for simultaneous proc- cast accuracies range from 10% to 40 %.
essing of the actual model and the Well performance forecasts are bound to be Fig. 12-Advantages of parallel vs. se-
pseudomodels. less successful. quential planning.
On a total-project basis. the manpower re- 4. Reservoir performance forecasting is
quirements always are less for parallel than 5. Saleri, N.G.: "Discussion of Stochastic
bound to remain probabilistic indefinitely.
Modeling," JPT (July 1990) 929.
for sequential planning. In fact, our finite-
6. Ballin, P., Journel, A., and Aziz, K.: "Pre-
difference studies suggest that the use of Acknowledgments diction of Uncertainty of Reservoir Perfonn-
parallel planning can realize resource sav- ance Forecasting," paper CIM 91-88 presented
I thank the managements of Chevron E&P
ings (labor plus computer) of 10% to 50% at the 1991 CIM/AOSTRA Conference, Banff,
Services, Cabinda Gulf, Chevron U.K.,
over sequential planning. Precise compari- April 21-24.
AGIP, Sonongal, Enterprise, Lasmo, Mur-
sons are unavailable because no modeling 7. Haldorsen, H.H. and Damsleth, E.: "Stochas-
studies were conducted with both sequen- phy/Ocean, Neste, Oryx, and Ranger for
tic Modeling," JPT (April 1990) 404.
tial and parallel planning. permission to publish this work. Sharon A.
8. Kasischke, W.F. and Williams, G.R.: "Reser-
Table 3 gives data on four studies con- Sloan's valuable help in the preparation of voir Characterization by a 'Cloud' Transfonn:
ducted with parallel planning. The Elk Hills the manuscript is acknowledged. Also, the Elk Hills 26R," report No. EPAD920050, "
26 R study involved a simulation model con- contributions of my reservoir engineer- Chevron E&P Services Co., Houston (1992).
structed with fractal geostatistics. Its objec- ing colleagues-particularly Robert M.
tive was to evaluate fine- and medium-scale Toronyi, Chul-Hee K. Chun, and Mark A. 51 Metric Conversion Factors
heterogeneity effects on recovery. Model Williams-to the intellectual foundation of
'API 141.5/(131.5+' API) = g/cm 3
complexities (28,000 cells, fine-grid size) the work are appreciated. bbl x 1.589 873 E-Ol = m)
posed a difficult flow simulation problem. ft x 3.048* E-Ol = m
Parallel planning was instrumental in this References ft3 x 2.831685 E-02 = m3
project's timely completion. in. X 2.54 E+OO = cm
1. Emanuel, A.S. et ai.: "Reservoir Perfonnance
In summary, the superiority of parallel md X 9.869 233 E-04 = I'm'
Prediction Methods Based on Fractal Geostatis- psi x 6.894757 E+OO = kPa
planning lies in its spontaneous conformance tics," SPERE (Aug. 1989)' 311; Trans.,
to the dynamic and probabilistic nature of AIME,287. Conversion factor is exact.
reservoir forecasting. 2. Hewett, T .A. and Behrens, R.A.: "Condition-
al Simulation of Reservoir Heterogeneity With Provenance
Conclusions Fractals," SPEFE (Sept. 1990) 217; Trans.,
AIME,289. Original SPE manuscript, Reservoir Per-
1. The sequential approach, the conven- 3. Lorenz, E.N.: "Nonlinear Statistical Weather formance Forecasting: Limits Field Ex-
tional finite-difference method, is inherently Predictions," paper presented at the 1980 amples and Acceleration by Parallel
ill-structured to produce timely forecasts. World Meterological Organization Symposi- Planning, received for review May 6, 1992.
Parallel planning offers an alternative that um on Probabilistic and Statistical Methods in Revised manuscript received Jan. 13, 1993.
can significantly accelerate forecasting. Weather Forecasting, Nice, Sept. 8-12. Paper (SPE 25151) accepted for publication
Recognition of inevitable forecasting uncer- 4. Saleri, N.G., Toronyi, R.M., and Snyder, May 10, 1993.
tainties constitutes the philosophical basis for D .E.: "Data and Data Hierarchy, " JPT (Dec.
parallel planning. 1992) 1286; Trans., AIME, 293. JPT

JPT July 1993 657

Das könnte Ihnen auch gefallen