SPE 112246
Rapid Model Updating with Right-Time Data - Ensuring Models Remain
Evergreen for Improved Reservoir Management
Stephen J. Webb, David E. Revus, Angela M. Myhre, Roxar, Nigel H. Goodwin, K. Neil B. Dunlop, John R.
Heritage, Energy Scitech Ltd.
Copyright 2008, Society of Petroleum Engineers
This paper was prepared for presentation at the 2008 SPE Intelligent Energy Conference
and Exhibition held in Amsterdam, The Netherlands, 25–27 February 2008.
This paper was selected for presentation by an SPE program committee following review
of information contained in an abstract submitted by the author(s). Contents of the paper
have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material does not necessarily reflect any position of the
Society of Petroleum Engineers, its officers, or members. Electronic reproduction,
distribution, or storage of any part of this paper without the written consent of the Society
of Petroleum Engineers is prohibited. Permission to reproduce in print is restricted to an
abstract of not more than 300 words; illustrations may not be copied. The abstract must
contain conspicuous acknowledgment of SPE copyright.
Abstract
History matching reservoir models to production data has
been a challenge for asset teams since the early days of
reservoir simulation. Keeping these models evergreen as
production data continues to arrive, knowing when a rehistory match is required and being able to re-history
match easily and efficiently is also a major challenge
which is often not addressed in a timely manner. This
need is becoming even more pressing as real-time
reservoir performance data is increasingly available.
Decisions can now be made with the support of the good
quality real-time data from the reservoir usually in the
form of pressure data from downhole gauges and rate data
from multiphase meters.
With the recent integration of existing technologies, a
rapid model updating workflow is now possible. The
history matching workflow that was once a discrete
process can now be a computer assisted continuous
process. Using statistically-based assisted historymatching technology in conjunction with real-time data
acquisition, data monitoring, and reservoir simulation
software, new production data can be quickly assimilated
into the reservoir model. Real-time data from the field
measurement devices is filtered for consumption by the
reservoir modeling software and compared with the
forecasts from the reservoir simulator to determine if rehistory matching is required. The new data can be added
to the history file and the model (revised if necessary) can
be used in operational performance optimization.
This rapid model updating workflow can be run semiautomatically on a continuous basis as new production
data is gathered, thereby keeping reservoir models
evergreen and providing the most up-to-date basis for the
making of important reservoir management decisions.
Introduction
Since the early days of reservoir simulation, history
matching has been identified1 as one of the best methods
of validating a reservoir model’s predictive capabilities.
Often long periods of time have been spent adjusting the
reservoir description so that the reservoir simulator’s
calculated results match the observed data from the
reservoir. Unfortunately, due to limitations of data
availability, computer hardware performance and
mantime, history matching has been a discrete process
which is performed only at certain stages in a reservoir’s
life cycle rather than as a continuous process that updates
the reservoir model as new data arrives. Although recent
advances in assisted matching technology2 have shown
there is potential to cut the time required to achieve
history matches, often the reservoir model is out of date
before the history matching process has been completed
because the production data used in history matching is
frozen in time when the process is started and is not
updated over the often many months that history matching
can take.
With the advent of digital oil field technology3,4,5, data
is now available from the reservoir in real-time. In
particular, downhole pressures6 and multiphase surface
flow rates7 can now be measured directly and
continuously. This real-time data is now routinely used
for reservoir monitoring, flow assurance calculations,
production optimization8,9,10 and for reservoir engineering
analysis11. However, despite many good intentions12,13,
reservoir modeling has been slow to take up the use of
real-time production data to condition its models. While
there have been some promising projects using Ensemble
Kalman Filtering to update reservoir models14,15, these
have mostly been research oriented and have not yet made
it into mainstream use.
The approach taken in this paper is different from
those reported previously in that it builds upon assisted
history matching technology that is already in the
commercial domain and is in widespread use. First, a
process framework is established in which a history match
of a reservoir model can be obtained. This process, which
2
SPE 112246
will be summarized in this paper and is described in detail
elsewhere2,16, allows the engineer to chose parameters in
the reservoir model that the assisted history matching
software can sensitize so that it can explore which
parameters need to be changed and how they need to be
changed to achieve a history match. The process also
produces an estimate of the uncertainty in the history
match. While it is obtaining history matches, the process
can be used to forecast reservoir performance and define
confidence intervals in the forecast performance. These
confidence intervals can then be used as a measure of the
forecast’s accuracy as new reservoir performance data
arrives. If the new data lies within the confidence
intervals, the model is forecasting field performance to
within the uncertainty of the modeling system. If the new
data lies outside the confidence intervals, either the field
is not being operated as forecast (and the simulator’s
operating constraints need updating) or the model itself
requires updating. The process framework provides an
environment where this can be done in an almost
automatic fashion.
This process will be illustrated by an example based
on a large mature North Sea reservoir where the rapid
model updating process is illustrated by repeatedly
assimilating new production data into the history
matching process, thereby keeping the model “evergreen”
and as up to date as possible for field operating and
development decisions.
Rapid Model Updating Process Components
The rapid model updating process is comprised of the
following components:
•
Statistically-based assisted history-matching
technology
•
Robust full field black oil and compositional
reservoir flow simulator
•
Real-time field monitoring and production
data management system
These technology components are directly interfaced with
each other to provide a comprehensive and integrated
total system which is described in the subsequent sections
of this paper.
Assisted History Matching
The availability of powerful and inexpensive computing
hardware over the last decade has led to the development
of software tools that can greatly help the history
matching process. Additionally these tools allow for the
inherent uncertainty in the reservoir description and
indeed the production data, and reflect that uncertainty in
the reservoir performance forecasts used for operational
and field development decision-making. The technologies
that have been employed are many, and include gradient
methods, simulated annealing, evolutionary algorithms
and
Bayesian
response
modeling2,16,17,18,19,20,21,22,23,24,25,26,27,28.
surface
The technology used in this work is the advanced
linear Bayesian tool, EnABLE™. It can be used with a
number of different reservoir simulators and is designed
to assist in the process of identifying multiple acceptable
history matches by using a structured workflow process
which allows many parameters to be automatically
modified simultaneously. Advanced experimental design
methods and linear Bayesian statistical routines are used
which permit the use of objective data, subjective opinion
and other indirect information in specifying a prior
distribution29. A statistical estimator based on a response
surface model (the Estimator) is created as a proxy for the
full reservoir simulation model and is updated after each
simulation run. By using the proxy to approximate the
simulator, extensive exploration of the simulated
responses across the solution space may be performed
without the computational expense of making a complete
simulator run in each evaluation. Starting with a set of
scoping runs designed to broadly explore the entire
solution space followed by additional “most informative
runs” (whereby the regions of the parametric space where
the proxy model has the greatest uncertainty are explored)
a reasonably predictive estimator is built. A number of
history match solutions are then obtained more rapidly
than would be possible by simply running a large number
of simulation cases. Fit for purpose history matching and
confidence interval estimation using the workflow
described here typically is accomplished with the use of a
number of reservoir model runs of the order of 100 to 300
runs. Methods that do not employ an estimator are
variously reported to require a thousand or more runs30.
The workflow process is shown in Figure 1:
1.
2.
3.
4.
5.
6.
7.
Identify study objectives.
Setup simulation model as usual.
Identify
parameters
(modifiers)
for
investigation in the history match (and
predictions
and
optimization31,
if
anticipated)
and
their
ranges
of
investigation.
Make modifications to the simulator input
data to sensitize these parameters.
Generate a set of ‘scoping runs’. These
runs, made using an experimental design
procedure, provide the initial basis for the
Estimator model. Typically 25 runs are used
for this step, although less may be sufficient
for small numbers of modifiers. Multiple
runs can be submitted for simultaneous
computation, if multiple processors are
available.
Import the historical data on well
performance and production.
Validate: decide by inspection whether the
model parameterization has the potential to
lead to a history match of sufficient
accuracy to meet project objectives.
SPE 112246
8.
9.
10.
11.
12.
3
Identify the observed data that is considered
to be most diagnostic of reservoir and flow
physics and pick points for history
matching.
Initialize the Estimator models using results
from the scoping runs. The tool will select
the most important modifiers for each match
point and construct an initial statistical
response surface model.
Generate additional runs to improve the
Estimator model with a set of ‘most
informative’ runs. This is a sequential
experimental design approach. These runs
explore where there is the most uncertainty
about the simulation model. Each of these
runs causes the tool to perform a Bayes
update of the Estimator model using the new
results.
Conduct a sequence of ‘best match’
simulation runs. Since each run updates the
Estimator model, a sequence of best match
runs is needed, where each run is using an
improved Estimator model. The parameter
values are chosen by the tool which
performs an optimization that is informed by
the Estimator model.
Evaluate the runs graphically using the
tool’s built-in capabilities.
Modify
weighting of match objective function to
steer match, ensuring that the phenomena
expected to have the most impact on
predictions are targeted. Repeat from 9 (or
10) until good matches are seen to be
emerging.
Forecasting Under Uncertainty
Once a set of acceptable history match runs has been
obtained, uncertainty in the production forecasts can be
explored. Simply by extending the simulated time period
to include the chosen production forecast scenario, the
range of expected production behavior can be explored.
By repeating steps in the workflow above for a time
period that includes the original history match period and
the forecast period, confidence intervals in the production
forecast can be generated. Rather than the Estimator just
providing information about the quality of the history
match, it is now providing details of the uncertainty in the
production forecasts given the known quality of the
associated history match. For simplicity, Figure 1 shows
this as “Start forecasting”. The details of the workflow
are:
13.
Generate a number of scoping runs
(typically 15) to scope the newly added
forecast period. These runs scope the new
solution space so that the new results in the
forecast period following on from the
history will be taken into account in the
Estimator before starting the refinement
runs.
14.
15.
Select forecast points.
Generate a series of most informative and
best match runs (e.g. 12 most informative
runs followed by 3 best match runs). The
most informative runs reduce the Estimator
uncertainty in the simulator response at
forecast points. The best matches ensure
that there are forecasts that are associated
with good history match runs.
The
Estimator is updated with the run modifiers
and results at the end of each simulation run.
A further step that can be added is to optimize the
operational plan of the reservoir. By adding controllable
parameters that can be subjected to optimization to the set
of modifiers that is being used (e.g. choke settings,
workover scheduling, well placement, compression
timing, etc.) recommendations for field operational and
development optimization can be derived.
Regarding confidence intervals and estimated
prediction uncertainty, the technology estimates the
confidence associated with the set of results for the match
point. This is a measure of the uncertainty in the value of
the production at a future point in time. The measure of
confidence takes the form of percentile values, and the
confidence interval (CI) is the probability that a value is
between an upper limit and a lower limit. The default
confidence intervals used in this work were 80%
intervals.
Real-time Production Monitoring
The real-time production system used in this workflow,
and shown in Figure 2, consists of a specialized field
monitoring system (at the local field level), and a
comprehensive production data management system (at
the office level). The real-time production system
acquires, collects, and stores real-time and historical
intelligent instrumentation data, and provides a common
desktop for visualization, field monitoring, analysis, and
interpretation.
It features “life of the field” storage capability at the
office level with flexible sampling rates as low as per
second with high precision. A broad range of specialized
interpretation, analysis, monitoring, and diagnostics tools
are available to streamline analyses, processes, daily
operations, production optimization, and reservoir
management.
Available configurations for the real-time production
system include direct interface with intelligent
instruments (downhole, subsea, and topside/surface),
and/or instrument interface via the operator’s automated
process control infrastructure of Distributed Control
Systems (DCS), Supervisory Control and Data
Acquisition (SCADA), Subsea Master Control Systems
(MCS), Information Management Systems (IMS),
Automated Systems, and third-party historians.
4
The real-time production system utilizes the industry
standard network communications protocol, TCP/IP to
connect the operator’s communications network, and to
connect remote locations via a telecommunications
provider, satellite services provider, over the internet, or
via a wireless network for near proximity locations.
This availability of real-time pressure, temperature
(topside, subsea or downhole) and multiphase rate data
(topside or subsea) means that high quality reservoir
performance data are now available to the reservoir
engineer. While these data have been enthusiastically
taken up and used by operations and production engineers
for field monitoring, field surveillance, flow assurance
and production optimization, the reservoir engineering
community has been somewhat slower to react. Take up
has been inhibited by the lack of tools to quickly
assimilate this real-time data into reservoir models.
One issue is the challenge of timescales. Whereas
real-time data arrives and can be stored on a second by
second basis, for this workflow, the reservoir engineer is
typically interested in daily, weekly, or monthly averaged
production data (Figure 3). So for real-time production
data to be ready for consumption by reservoir models,
some form of data aggregation is generally needed, which
may include filtering, interpolation, averaging, and other
summarization
methods,
depending
upon
the
characteristics of the data. This aggregated data may be
provided to reservoir models in the form of ASCII,
spreadsheet, or XML, using the emerging PRODML
format32. This is often referred to as “in-time” or “righttime” data rather than real-time data.
Therefore, prior to consumption by the reservoir
engineer, the real-time data used in this work were filtered
using a custom formula defined by the user, or a default
moving average smoothing algorithm supplied with the
application. These formulae enable customised noise
filtering to smooth extreme outliers without modifying the
original real-time data. The calculation is performed
automatically on the selected time span of real-time data.
Once smoothed, the resulting filtered data are averaged
and interpolated to produce data at any desired time
period, and may then be exported into an appropriate
format. Currently the export is a manual process whereby
the averaged and interpolated data are exported, with the
help of templates, in specialized ASCII format files that,
may be directly inserted as Include Files to the reservoir
flow simulator data deck, or imported as well history into
the assisted history matching application. However, in the
future it is anticipated that this data transfer will be in an
XML format (PRODML) and that the rapid model update
applications will receive the data when it becomes
available using a publish and subscribe mechanism
similar to that available with WITSML33 data.
Another issue that has inhibited the use of real-time
data in reservoir models is the lack of a robust automated
SPE 112246
process for repeatedly including new production data in
the history matching process. Because history matches
have often taken so long, engineers are reluctant to repeat
the process even though the models they are using for
decision making purposes can be out of date. The next
section addresses that challenge.
Rapid Model Updating Workflow
A process has been established whereby “right-time” data
can be extracted from a real-time monitoring application
in a suitably aggregated format and, employing the
workflow described in the sections above, used to update
reservoir flow simulation models that are interfaced to
assisted history matching technology and potentially
geological reservoir models. The key components of this
workflow are illustrated in Figure 4. It should be noted
that the rapid model update process and the individual
component applications are capable of handling any
desired time period, including intra-day production data.
This right-time data is then made available to the rapid
model updating workflow to append to the existing
history that has been used in the assisted history matching
process. The controls on the existing forecast runs are
replaced with the newly observed production data
controls and the confidence intervals around dependent
aspects of the modeled production behavior re-calculated.
This is achieved by rerunning steps 13 to 15 in the
workflow. The model set up determines whether a
production value is assigned or dependent; dependent
values are typically phase rates and pressures but may
include compositions and temperatures. New actual
recorded values are compared with model generated
values to determine where they lie with respect to the
model’s confidence intervals. If the new data lies within
the confidence intervals, then the field is operating within
the limits of the model’s ability to forecast and the
reservoir model is still capable of predicting field
performance within the uncertainty of the model.
If, however, the new data lies outside the confidence
intervals, then the reservoir model is no longer capable of
forecasting the performance of the field and it needs
updating. This can be achieved by rerunning steps 9 (or
10) to 12. If this automatic step does not lead to an
acceptable match then human intervention is required.
Action may be required of an engineer to steer the history
match differently as in step 12. Or if the mismatch is
sufficiently severe the process may need to start at steps 2
or 3 where new modifiers are introduced.
Once these steps have been performed the forecast
workflow is repeated (steps 13 to 15) so that a new
forecast and new confidence intervals are generated with
the new model. This new model can also be used for
optimization runs to re-evaluate and revise, if necessary,
the field operating and development plans.
This whole process is repeated every time new data
becomes available and is illustrated in Figure 5. In the
SPE 112246
reservoir management workflow this would normally not
be more frequent than every month (see Figure 2) but
could, if necessary be repeated on a more frequent basis if
the field performance was changing quickly.
In summary, as new history data (recorded data and
operational controls) are added they are checked to see if
they are consistent with the forecast based on the existing
history match (i.e. within the confidence intervals). If they
are, new forecasts are made based on the augmented
history. If not, a re-history match is required which
includes all the previous history and the new data. This
may be possible using the existing modifiers or new
modifiers may have to be introduced. The new forecasts
should be narrower than the previous forecasts from the
new forecast point but may be offset from previous
forecasts as operational controls have most likely been
changed. Also, the modifier ranges which generate good
history matches should be reduced. Therefore as time
progresses and more and more production data is
assimilated, both the history match and forecast are
refined. New information reduces uncertainty in the
geological and simulation models which results in the
forecast being refined (a consequence of refining the
history match). The new information may allow the
selection of a narrower range of possible geological
models.
Using this rapid model updating process to update the
history match as described above, the set of possible
reservoir models is always kept up to date ensuring that
the best possible set of models is used for operating and
field development planning decision-making.
Finally, there is no inherent requirement that this rapid
model updating procedure is performed only using the
reservoir simulator. It has been shown that during the
history matching process using assisted history matching
tools, geological models and reservoir simulation models
can be kept consistent by applying modifiers to the
geological model as well as the simulation model34. By
applying the rapid model updating workflow to the
geological model as well as the simulation model, the
geological model can be updated as new production data
is assimilated thereby keeping both the simulation and
geological models “evergreen”.
Example
An example of this workflow is based on a publicly
available version35 of a simulation model of the Gullfaks
field on the Norwegian Continental Shelf. The simulation
model is of the eastern part of the Gullfaks field (Block
30/10) and models the field from startup in 1986 until the
end of 1996. Although there are some 60 wells defined in
the model only about a quarter are active early on in the
production of the field. This workflow was executed by
using the advanced linear Bayesian tool described above,
together with the full physics reservoir flow simulator
Tempest-MORE and real-time production monitoring
software. The simulation recurrent data and historical
5
data were prepared from the actual field performance data
on a monthly basis for this example case.
The objective of the example was to illustrate how a
reservoir model that is kept current can be an excellent
reservoir surveillance tool, and thus improve the quality
of reservoir management36. Water production was the
focus of attention, because it was anticipated that this
parameter would provide the greatest disparity between
simulated and observed data. For the purposes of this
example it was assumed that today’s date (“current date”)
is the end of September 1989 and that everything prior to
that date is “past” and everything after that date is
“future”. The model was history matched up to September
30, 1989 using the workflow shown in Figure 1 and the
cumulative water production results, for the well group of
producers, are shown in Figure 6. Up until this point all
the wells in the models were controlled on oil rate and
water production was allowed to be predicted by the
models. Forecasts were then run from October 1, 1989
until January 1991 with the wells on bottom-hole pressure
control and a field oil production rate target (Figure 6).
Confidence intervals for these forecasts were generated at
approximately 3 month intervals, which was the planned
time period for reservoir surveillance.
The “current date” was then advanced to the end of
December 1989. The confidence intervals were compared
with the actual water production observed (Figure 7). It
can be seen that the actual water production rate was still
within the confidence intervals set for the forecasts.
Therefore, the models were still able to forecast
performance of the field within the confidence intervals
and no further adjustment of the models was required.
Forecasts were then made from January 1990 onwards
using the existing models.
The “current date” was then advanced to March 31,
1990 and the process repeated. Now the actual cumulative
water production was outside the confidence intervals
(Figure 8). The deviation between actual and forecast is
seen even more dramatically in the water production rate
(Figure 9). This meant that the models were no longer
able to forecast field performance and needed to be
adjusted to assimilate the new production data. The
forecast field performance controls were replaced with the
actual field performance controls and the models rematched using the existing modifiers (as described above
this may be achieved by re-running steps 9 to 12). The
results of this re-match are shown in Figure 10. A
comparison of the actual modifier values for the best
match history runs to September 30, 1989 and the run
closest to the actual history to March 31, 1990, indicated
that the Kv/Kh ratio and aquifer strength were some of the
key factors in achieving the re-history match. This match
is not perfect, however, and had a more accurate rehistory match have been required, the process could have
been re-started at step 2 and new modifiers introduced.
The re-history matching process (steps 2 through 12)
would then have resulted in a modified set of models that
6
could better have matched the production performance of
the field and represented a better set of models to make
forecasts for future performance of the field. However, for
the purposes of this example, the re-matching workflow
had been demonstrated and the process stopped here.
Conclusions
A process has been described and illustrated whereby new
production data can be quickly and continually
assimilated into a set of history matched reservoir models.
By extending the framework of an assisted history
matching process, new production data can be added to
the old production history and, if necessary, be used to
update the existing history match. Confidence intervals
from the reservoir performance forecasts are used as a
guide to whether the existing reservoir models are capable
of modeling the new production data. If they are, the
performance forecasts are updated by simply introducing
the new operating conditions. If not, the models are
adjusted, repeating the history matching process which
may, if necessary, include new history matching
parameters.
As data continues to arrive, it is monitored against the
forecast performance and the associated confidence
intervals. At discrete time intervals, new operational data
replaces the forecast operational control and the rapid
model updating process is repeated.
In this way, the set of history match models can
always be kept up to date or “evergreen” providing the
most up to date models as a basis for important reservoir
management and development decision making.
Acknowledgements
The authors would like to thank the management of Roxar
for permission to publish this work and for the use of the
EnABLE™ software, the Tempest-MORE reservoir
simulator and Tempest simulation suite, and the
Fieldwatch/Fieldmanager real-time monitoring and
production data management software. They would also
like to thank Bob Parish, Robert Frost and David Ponting
for their useful contributions to this work.
References
1. K.H. Coats, “Use and Misuse of Reservoir
Simulation Models”, SPE2367 (1969), JPT
November 1969
2. D.S. Bustamante, D.R. Keller and G.D. Monson,
“Understanding Reservoir Performance and
Uncertainty Using a Multiple History Matching
Process” SPE95401 (2005) presented at SPE
ATCE Dallas, TX, USA, October 2005.
3. T. Unneland and M. Hauser, “Real-Time Asset
Management: From Vision to Engagement – An
Operator’s Experience” SPE96390 (2005)
presented at SPE ATCE Dallas, TX, USA,
October 2005.
4. C. Reddick, “Field of the Future: Making BP’s
Vision a Reality”, SPE99777 (2006) presented at
SPE 112246
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
the Intelligent Energy Conference and
Exhibition, Amsterdam, The Netherlands, April,
2006.
H. Potters, P. Kapteijn, “Reservoir Surveillance
and Smart Fields”, IPTC11039 (2005) presented
at the International Petroleum Technology
Conference, Doha, Qatar, November, 2005.
T. Unneland and T. Haugland, “Permanent
Downhole
Gauges
Use
in
Reservoir
Management of Complex North Sea Oil Field”,
SPE26781 (1994) SPE Production and Facilities,
August, 1994.
K.H. Frantzen, M. Brandt, K, Olsvik,
“Multiphase Meters – Operational Experience in
the Asia-Pacific” SPE80502 (2003) presented at
the SPE Asia Pacific Oil and Gas Conference
and Exhibition, Jakarta, Indonesia, September,
2003
D. Reeves, R. Harvey, T. Smith, “Gas Lift
Automation: Real Time Data to Desktop for
Optimizing an Offshore GOM Platform”,
SPE84166 (2003) presented at the SPE Annual
Technical Conference and Exhibition held in
Denver, Colorado, October 2003.
W.E.J.J.Van Zandvoord, A.A. Maskery, T.
Schmid, “Application of Real Time Surveillance
and optimization Tools on a Large Asset”,
SPE100342 (2006) presented at the 2006 SPE
Asia Pacific Oil & Gas Conference and
Exhibition, Adelaide, Australia, September 2006.
L.A. Saputelli et al., “Promoting Real-Time
Optimization of Hydrocarbon Producing
Systems”, SPE83978 (2003) presented at
Offshore Europe, Aberdeen, UK, September
2003.
A.G. Mezzatesta, D.C. Shaw, S.J. Webb,
“Constant Reservoir Evaluation through RealTime Data as a Service”, SPE109856 (2007)
presented at the SPE ACTE Anaheim, CA, USA,
November 2007.
D.J. Rossi, O. Gurpinar, R. Nelson, S. Jacobsen,
“Discussion on Integrating Monitoring Data into
the Reservoir Management Process”, SPE65150
(2000) presented at the SPE European Petroleum
Conference in Paris, France, October, 2000.
O. Nygaard, C. Kramer, R. Kulkarni, J-E.
Nordtvedt, “Development of a Marginal GasCondensate Field Using a Novel Integrated
Reservoir
and
Production
Management
Approach”, SPE68734 (2001) presented at the
SPE Asia Pacific Oil and Gas Conference and
Exhibition in Jakarta, Indonesia, April 2001.
G. Naevdal, L. M. Johnsen, S. I. Aanonsen, E. H.
Vefring, “Reservoir Monitoring and Continuous
Model Updating Using Ensemble Kalman
Filtering”, SPE84372 (2003) presented at the
SPE ACTE, Denver, CO, USA, October 2003.
X.-H Wen, W.H. Chen, “Real-Time Reservoir
Model Updating Using Ensemble Kalman
Filtering”, SPE92991 (2005) presented at the
SPE 112246
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
SPE Reservoir Simulation Symposium, Houston,
TX, February 2005.
A.J. Little, H.A. Jutila, A. Fincham, Energy
Scitech Ltd. “History Matching With Production
Uncertainty Eases Transition into Prediction”
SPE100206 (2006) presented at the SPE
Europec/EAGE Annual Conference in Vienna,
Austria, 12-15 June 2006.
C.E. Romero, J.N. Carter, R.W. Zimmerman,
A.C.
Gringarten,
“Improved
Reservoir
Characterization
Through
Evolutionary
Computation”, SPE62942 (2000), presented at
the SPE ACTE Dallas, TX, USA, October 2000.
Various
Authors,
“PUNQ.
Production
Forecasting with Uncertainty Quantification”. A
research project funded in part by the European
Commission under the Non-Nuclear Energy
Programme (JOULE III), contract F3-CT950006, January 1996 – May 1999.
F. Bennett, T. Graf, “Use of Geostatistical
Modeling and Automatic History Matching to
Estimate Production Forecast Uncertainty - A
Case Study” SPE74389 (2002) presented at the
SPE International Petroleum Conference and
Exhibition, Villahermosa, Mexico, February
2002.
M. Feraille, F. Roggero, E. Manceau, L.Y. Hu, I.
Zabalza-Mezghani, L. Costa Reis, “Application
of Advanced History Matching Techniques to an
Integrated Field Case Study”, SPE84463 (2003)
presented at the SPE ACTE Denver, CO, USA,
October 2003.
G.J.J. Williams, M. Mansfield, D.G. MacDonald,
M.D Bush, “Top-Down Reservoir Modelling”,
SPE89974 (2004) presented at the SPE ACTE,
Houston, TX, USA, September 2004.
M. Litvak, M. Christie, D. Johnson, J. Colbert,
M. Sambridge, “Uncertainty Estimation in
Production
Predictions
Constrained
by
Production History and Time-Lapse Seismic in a
GOM Oil Field” SPE93146 (2005) presented at
the SPE Reservoir Simulation Symposium, The
Woodlands, TX, USA, February 2005.
Watkins A.J., Parish R.G., 1992. "A Stochastic
Role For Engineering Input to Reservoir History
Matching." SPE23738, LAPEC II; Caracas,
Venezuela, March 1992.
Watkins
A.J.,
Parish
R.G.,
1992.
"Computational Aids to Reservoir History
Matching."
SPE24435.
SPE
Petroleum
Computer Conference, Houston, Texas. July
1992.
Parish R.G., Watkins A. J., Muggeridge A.,
Calderbank V. J. 1993. “Effective History
Matching: The Application of Advanced
7
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
Software Techniques to the History Matching
Process“ SPE25250, 1993
A.J. Watkins, K.N.B. Dunlop, V.M. Alcobia,
"The Stochastic Revision of Knowledge in
Reservoir History Matching",
22nd Petr.
Itinerary Congress of OMBKE, (Hungarian
Mining & Mineral Society), Tihany Hungary,
Oct. 6-9 1993.
Parish R.G. & Little A.J.H., 1994. “A Complete
Methodology for History Matching Reservoirs”
6th ADIPEC, October 1994.
Parish R.G. & Little A.J.H., 1997. “Statistical
Tools to Improve the Process of History
Matching Reservoirs” SPE37730, MEOS,
March 1997.
I. Miller and J. Freund, “Probability and
Statistics for Engineers”, second edition,
Prentice-Hall, Inc., Englewood Cliffs, NJ (1977).
J. Lach, K. McMillen, R. Archer, J. Holland, R.
DePauw, B.E. Ludvigsen, “Integration of
Geologic and Dynamic Models for History
Matching, Medusa Field”, SPE95930 (2005)
presented at the SPE ACTE, Dallas, TX,
October, 2005.
H.A. Jutila, SPE, and N.H. Goodwin, Energy
Scitech Ltd. “Schedule Optimization to
Complement Assisted History Matching and
Prediction Under Uncertainty”, SPE100253
(2006), Presented at the SPE Europec/EAGE
Annual Conference in Vienna, Austria, 12-15
June 2006.
B. Weltevrede, A, Doniger, L. Ormerod, S,
DeVries, “Second-Stage Callenge for the
PRODML Standard: Adaptive Production
Optimization”, SPE110907 (2007) presented at
the SPE ACTE, Anaheim, CA, November, 2007.
M.A. Kirkman, M.E. Symmonds, S.W.
harbinson, J.A. Shields, M.Will, A. Doniger,
“Wellsite Information Transfer Standard Markup
Language, WITSML, an Update”, SPE84066
(2003) presented at the SPE ACTE, Denver, CO,
October, 2003.
S.J. Webb, J.S. Bayless, K.N.B. Dunlop,
“Enabling the ‘Big Loop’- Ensuring Consistency
of Geological and reservoir Simulation Models”
paper presented at the 2007 AAPG Annual
Convention and Exhibition, Long Beach,
California, April, 2007.
Gullfaks Data released by Statoil is a selection of
availiable petroleum technical data from the
Gullfaks field, finalized before 01.01.1998
C. C. Mattax, R. L. Dalton, “Reservoir
Simulation”, SPE Monograph Series Volume 13,
SPE, 1990.
8
SPE 112246
1
Agree study objectives
2,3
Geomodeling, gridding
Identify modifiers & ranges
Import model into EnABLE
Yes
4,5
Scoping runs
(Experimental design)
Revise
modifiers or
No
6
Import/Review Observed
Data
Revise
Geomodel
7
No
Model validated?
Yes
8,9
Pick Estimator (match) Points
10
Informative Runs
Review/Revise
tolerances
& match points
11
Best match runs
12
No
Good matches
emerging?
Yes
13,14,15
Start forecasting
No
Match, optimisation,
forecasts complete?
Yes
Recommend Decision
Figure 1 Assisted history matching and forecasting workflow diagram
SPE 112246
9
Topside
On-shore
offices
Subsea
Downhole
Figure 2 Offshore real-time production system
Reservoir Surveillance
Capacity Planning/
Business Design
Operational Planning
Scheduling/Real-Time Optimization
Supervisory Control
PID/Regulatory Control
s
th
on
s
ar
Ye
M
s
ay
D
rs
ou
H
es
ut
in
M
ds
n
co
Se
Data Collection Period
10
Figure 3 Application of real-time data by duration and granularity
10
SPE 112246
Real-Time
Production System
Data Historian
ASCII,
PRODML
Real-Time
Production System
Assisted History
Matching Workflow
Proxy Model for
Uncertainty
Analysis
Field Controls
Qo, Qw, Qg
Sand rate
Small Loop
Reservoir Flow
Simulation Model
Qo, Qw, Qg
Sand rate
Big Loop
P, T
Qo, Qw, Qg
Sand rate
Reservoir Geologic
Model
Figure 4 Rapid Model Updating Workflow
History match for (0, t0)
Predict for (t0, tend)
n=0
Record data from (tn, tn+1)
Production model for
(tn, tn+1) as expected?
Yes
No
Remove runs covering
original prediction
phase (t0, tend)
Redo runs
covering (t0, tend)
using new
schedule, treating
(t0, tn) as history
and (tn, tend) as
prediction
No
Change (t0, tn+1)
to history and
rematch,
including
prediction phase
(tn+1, tend)
Do (tn, tn+1) data
fit predicted CIs ?
Yes
n=n+1
Figure 5 Rapid Model Updating Flowchart
SPE 112246
11
Confidence Intervals
for forecast
P90
P50
End of history
Figure 6 History match of cumulative water production to October 1989 and forecast with confidence intervals
New history still
within Confidence
Intervals after 3
months
Figure 7 New cumulative water production history data to January 1990 showing new data still within confidence intervals
P10
12
SPE 112246
After 6 months new
history trending out
of Confidence
Intervals
Figure 8 At April 1990 water production data moving outside confidence intervals and re-history match is required
New history
Figure 9 Water production rate data shows an even more dramatic deviation from forecast
SPE 112246
13
New re-matched run
Figure 10 Effect of re-history match to April 1990 and re-forecast on water production rate match