This paper presents two
complementary methodologies for operation support and
improvement of the production conditions. The first one is
based on data reconciliation between process measurements
and flow modelling. It brings an additional level of
information to the problem of continuous metering of deepwater
subsea wells. As periodic well testing is required to
achieve this predictive metering, the second methodology
provides the optimal test sequences of well permutations. It
involves flow process simulation and algorithmical sorting,
according to production constraints and operating strategies.
Scanning the Internet for External Cloud Exposures via SSL Certs
OTC 14009 Deep Offshore Well Metering and Permutation Testing
1. OTC 14009
Deep Offshore Well Metering and Permutation Testing
Erich Zakarian, RSI; Arnaud Constant, TotalFinaElf Exploration & Production Angola; Lionel Thomas, TotalFinaElf; Martin
Gainville, Institut Français du Pétrole; Pierre Duchet-Suchaux, TotalFinaElf; and Philippe Grenier, RSI
Copyright 2002, Offshore Technology Conference
This paper was prepared for presentation at the 2002 Offshore Technology Conference held in Whereas conventional solutions, such as hardware
Houston, Texas U.S.A., 6–9 May 2002.
multiphase metering, supply a limited information, our
This paper was selected for presentation by the OTC Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
methodology works as an overall field supervisor for:
presented, have not been reviewed by the Offshore Technology Conference and are subject to • estimating individual well production with respect to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Offshore Technology Conference or its officers. Electronic reproduction, appropriate pressure and temperature measurements;
distribution, or storage of any part of this paper for commercial purposes without the written
consent of the Offshore Technology Conference is prohibited. Permission to reproduce in print
• detecting abnormal behavior (sensor drift for
is restricted to an abstract of not more than 300 words; illustrations may not be copied. The instance);
abstract must contain conspicuous acknowledgment of where and by whom the paper was
presented. • validating hardware measurements, and replacing
them in case of failure. Typically, in deep offshore
Abstract production, hardware sensors are not replaced in case
As production in very deep waters becomes a crucial of failure for financial and feasibility reasons.
challenge for many oil companies, a better management of the This methodology is based on data reconciliation. It
production is constantly required. This paper presents two assumes that measurements are not necessarily correct and can
complementary methodologies for operation support and be corrected within a confidence interval. Meanwhile,
improvement of the production conditions. The first one is unmeasured variables derive from redundancy between flow
based on data reconciliation between process measurements modelling and field data.
and flow modelling. It brings an additional level of Data reconciliation has been already successfully applied
information to the problem of continuous metering of deep- to a small production network, see Ref. 1. Our paper intends to
water subsea wells. As periodic well testing is required to go further in the study of this innovative technology and
achieve this predictive metering, the second methodology presents its application to the Girassol field.
provides the optimal test sequences of well permutations. It
involves flow process simulation and algorithmical sorting, Well monitoring
according to production constraints and operating strategies. Given a set of temperature and pressure measurements, our
Finally, comparisons between numerical simulations and plant methodology aims to provide an estimate of the phase flow
data demonstrate the ability of these two methodologies to rates produced by each individual well of an oil field.
provide strong and reliable information for deep offshore
producers. Problem modelling. Real-time plant data are completed with
a global steady state simulation of the production network
Introduction involving:
The knowledge of the phase flow rates coming from each • mass, force, and heat balance equations;
individual well of an oil field is mandatory for a better • thermodynamic calculations;
production and reservoir management. Generally, this • hydrodynamic modelling.
information comes from a series of direct well testing, where a For instance, assuming process data at both ends of a
single well flows directly to a test separator. choke and an estimate of the fluid composition, one can derive
In deep offshore, this procedure turns out to be a local estimate of the liquid and gas mass flow rates from
inappropriate: production developments are based on hydrodynamic and thermodynamic calculations.
gathering network, where manifolds merge the production Combination between physical modelling and plant data is
from several wells into a single flowline. This is the case on applied to the whole production network, leading to multiple
the Girassol oil field in Angola, see Fig. 1. Moreover, direct estimates of the same information. This multiplicity derives
well testing implies deferred production, valve reliability and from our uncertain interpretation of the reality: this statement
flow assurance issues: hydrate formation in dead branches, is precisely the main strength of the data reconciliation
slugging at low flow rates, etc. technology.
2. 2 E. ZAKARIAN, A. CONSTANT, L. THOMAS, M. GAINVILLE, P. DUCHET-SUCHAUX, AND P. GRENIER OTC 14009
Whatever the complexity of a model, one should be aware (calibrated) might be trusted with a relative confidence as the
that it always remains an approximation. However, the latter fluid composition may change between two consecutive
can be restricted to a small number of modelling parameters measurements performed on a test separator. Therefore,
embedded in residual equations. Again, any hardware sensor equations (1) to (4) are replaced by residual equations and
contribution is a residual equation weighted by vendor weighted by the standard deviation (uncertainty):
uncertainty.
ePi = (Pi - Pmi)/σPi, .......................................................... (9)
Both hardware measurements and modelling equations are
involved on the same level of analysis through a global data
eTi = (Ti - Tmi)/σTi,......................................................... (10)
reconciliation and parameter estimate, leading to an
optimization problem. Conversely, experience feedback is eBSW = (BSW - BSWm)/σBSW, ......................................... (11)
expected to get optimal values of model uncertainties.
A major innovative aspect of this work is the systematic eGOR = (GOR - GORm)/σGOR,........................................ (12)
computation of the accuracy of any estimated variable. We
notice that the accuracy of a measured variable can be slightly where i = 1, 2, and 3.
increased since data reconciliation works as a global As reminded before, an additional parameter calibrates
computation where any piece of information is likely to be each modelling equation with respect to a reference value
improved by the other ones. (tuned from test measurements) and an uncertainty. Four
We also emphasize that information on the solution residuals complete the system:
uncertainty is as important as the solution itself: whenever eWP = (ΓWP - ΓWP,ref)/σWP, ............................................. (13)
redundancy remains, a physically incorrect solution can satisfy
the problem. Therefore, a supervisor checks the consistency of eWT = (ΓWT - ΓWT,ref)/σWT, .............................................. (14)
the solution against intuitive expectations.
eCP = (ΓCP - ΓCP,ref)/σCP, ............................................... (15)
Example. Let us consider the following scenario. Given one
well tubing followed by a choke, six sensors are installed eCT = (ΓCT - ΓCT,ref)/σCT,................................................ (16)
upstream and downstream these equipment to measure the
We finally get 16 equations, namely (5) to (16), and 13
pressure and the temperature of the fluid, see Fig. 2. We derive
variables: BSW, GOR, Flow, ΓWP, ΓWT, ΓCP, ΓCT, Pi, and Ti,
six equations:
where i = 1, 2, and 3. The system seems redundant. However,
Pi - Pmi = 0,.....................................................................(1) redundancy can only be ensured from a detailed analysis: to
avoid any singularity, the rank of the jacobian must be equal to
Ti - Tmi = 0, .....................................................................(2) the number of the variables at any operating point. Note: the
where i = 1, 2, and 3 refer respectively to the following jacobian of the system is the matrix given by the partial
positions: upstream the well tubing, downstream the well derivatives of the equations (modelling equations and
tubing, and downstream the choke. residuals) with respect to the variables.
Calibrated values of the water-liquid ratio and gas-oil ratio This condition is necessary to find a solution, which is a
give two additional equations: minimization of an objective function given as a sum of the
squares of the residuals. Meanwhile, the modelling equations
BSW - BSWm = 0,............................................................(3) are exactly satisfied (optimization constraints).
GOR - GORm = 0............................................................(4)
Three phase flow modelling. Several models are involved in
A well tubing model provides a pressure drop relation and the simulation of a Girassol production loop, see Table 1. As a
a heat balance between positions 1 and 2: mixture of oil, gas and water is expected during the field life,
fWP (P1, T1, P2, T2, Flow, BSW, GOR, ΓWP) = 0,..............(5) intensive efforts are required to get an acceptable physical
representation, due to the complexity of three-phase flows.
fWT (P1, T1, P2, T2, Flow, BSW, GOR, ΓWT) = 0. ..............(6) A gridded modelling is used for long tubings (well, sea-
line, riser) since local effects such as slope changes may
A choke model gives the same kind of relations between strongly impact pressure and thermal profiles. A complex
positions 2 and 3: three-phase hydraulic module ensures a correct representation
fCP (P2, T2, P3, T3, Flow, BSW, GOR, ΓCP) = 0,...............(7) of the different flow regimes. Local flash and thermal
calculations improve the physical modelling as well.
fCT (P2, T2, P3, T3, Flow, BSW, GOR, ΓCT) = 0................(8) A rather sophisticated modelling is used for the chokes
since flow criticity and gas expansion may seriously affect
Every measurements involved in this system are not pressure and temperature variations. The choke discharge
necessarily correct since hardware sensors are subject to errors coefficient must be initially calibrated and periodically
and the flow is not perfectly stable in the entire production validated against field data. Three-phase flow meters could be
system. As far as BSW and GOR, their a priori values used. Meanwhile, test separator instruments provide sufficient
3. OTC 14009 DEEP OFFSHORE WELL METERING AND PERMUTATION TESTING 3
and accurate information for calibration. executive framework. The latter runs the whole application
This tuning procedure is described in a further section. In and its components, like the configuration of a production
terms of computation, tuning is a particular use case of this network or the use cases of a well monitoring application.
well metering methodology.
Sensor failure detection. The data interface component
Thermodynamic issues. Physical phenomena such as makes the connection between the simulator executive
vaporization in wells or expansion in chokes require accurate framework and external components that provide hardware
thermodynamic calculations. Therefore, the mixture measurements: Distributed Control System, database.
composition is tracked along the flow line, and the whole unit To avoid any undesirable effect on data reconciliation, the
operations perform local vapor/liquid/liquid equilibrium data interface performs a preliminary analysis to detect any
calculations to get an estimate of the phase properties. possible failure (unlikely value, excessive variation) or
However, it is an illusion to believe that an accurate estimate unsteady behavior when the average of the measurements
of the reservoir fluid molar composition can be found: neither depends significantly on time.
the system, nor the available sensors are able to catch the A sensor is declared invalid in case of failure and its
effect of a composition change (at constant phase properties) contribution is removed from the system. It does not
between C9 and C10 cuts, for instance. Conversely, gas participate to the data reconciliation, leaving the determination
coning or water breakthrough does affect sensor data. of the measurement to the optimizer. This preliminary
Therefore, the fluid composition is corrected with BSW and detection is as important as the data reconciliation itself. Let
GOR variations by addition or removal of water and/or gas us present an example to confirm this statement.
from the first stage separator. We consider a network with a well producing in a single
flowline through a choke and a manifold, see Fig. 4. Pressure
Software issues. The basic domains involved in an industrial and temperature sensors are located upstream and downstream
well monitoring tool are: these equipment. Assuming an initial calibration of the system,
• compositional thermodynamic calculation; we define a particular scenario with hardware sensor failures,
• hydrodynamic modelling in pipes and wells; see Fig. 5: at a fixed period of time (five minutes), the data
• thermal modelling in pipes and wells; interface sends sensor measurements to the simulator
• valve modelling; executive framework and a new solution is computed. Forty-
• reservoir PI modelling; five minutes after starting up, the pressure sensor at the bottom
• data reconciliation; hole returns zero, which is of course an unacceptable value.
• real-time process data recording and analysis. This failure lasts half an hour. Two hours after starting up,
both pressure sensors at the bottom hole and the manifold
For each domain, several approaches have to be tested and
return zero again.
selected. For example, a particular thermodynamic server
might be suitable for certain operating conditions but First, let us consider the case where the data interface does
not perform a detection of sensor failure. At the beginning of
unacceptable to other cases. Note: a server is a software
the run, the simulator computes reconciliated data, see Fig. 6.
component that provides services for other software
components (client components) through defined interfaces. The latter are very close to the real measurements reproduced
on the figure as dotted lines.
Therefore, it should be easy to combine different modules
At 45 minutes where the first pressure sensor collapses, the
with the minimum effort. In addition, many pieces of software
from different sources might fulfill our requirements. For simulator manages to rebuild a measurement at the bottom
hole. The rebuilt value is physically acceptable but likely far
these reasons, we decided to build our well monitoring
from reality. At the same time, the production is overestimated
simulator as an open software, using the CAPE-OPEN
(the productivity index remains the same in the whole run) and
(Computer-Aided Process Engineering) standard, see Ref. 2.
the gas-oil ratio decreases, see Fig. 7. The wellhead pressure
The compliance to this standard is another major and
is also strongly affected because of its direct dependence on
innovative aspect of this work. It provides high model
the bottom hole pressure through the well tubing model. The
flexibility for the end user, it makes implementation of new
pressure drop increases by 106 Pa.
features much easier and faster: plug and play integration of
any CAPE-OPEN compliant component is carried out with the At 75 minutes where the sensor starts running again, the
initial solution is recovered but at 120 minutes, both pressure
minimum effort.
sensors at the tubing ends stop to run. The simulator fails to
Built on a component-based architecture, see Fig. 3, our
well monitoring simulator includes: find a solution.
Let us run the same simulation but with detection of sensor
• CAPE-OPEN compliant components: unit operations,
failure. We notice that the data reconciliation works perfectly
thermodynamic servers;
in this case, showing its ability to rebuild unmeasured
• external components: solver, man machine interface,
variables, see Fig. 8, 9. The sensor failure does affect the
data interface, supervisor.
redundancy of the problem, which is lower than before, but it
CAPE-OPEN standard interfaces ensure the
does not affect the results.
communication between components through a simulator
4. 4 E. ZAKARIAN, A. CONSTANT, L. THOMAS, M. GAINVILLE, P. DUCHET-SUCHAUX, AND P. GRENIER OTC 14009
At 120 minutes, both sensor failures are detected at the We calibrate the simulator with data recorded on
well tubing ends. The solution remains acceptable. However, December 7th, between 02:00 and 08:00 am. Then, we run a
the manifold pressure drops significantly and its a posteriori metering, up to December 12th.
confidence interval as well. In other words, there is not enough With a sampling DCS period of five minutes, production is
redundancy to trust the reconciliated value of the manifold estimated every thirty minutes, using data filtered on the past
pressure. hour. If we assign the same level of confidence to the models
This second computation confirms the ability of our (choke, inflow, tubings), the system overestimates the
methodology to replace hardware sensors in a production production, but the predicted trend is consistent with reality,
system. It also shows its weakness whenever the number of see Fig. 12.
valid sensors is not sufficient to allow redundancy. A detailed analysis shows that the choke model is
responsible for this deviation: tuning with subsequent tests
Real-scale validation shows that the discharge coefficient drops from 0.92 to 0.6 on
Production at the Girassol field started in late 2001. We December 8th, see Fig. 13. Meanwhile, productivity index and
propose to use a first series of measurements to validate our friction factors remain roughly constant. This observation
methodology and confirm its ability to provide reliable reveals some inconsistency between plant data and the choke
information. These measurements were recorded from modelling. Further analysis will be required to get a better
December 4th to 24th, 2001. understanding of the real choke behavior.
If we set a lower relative confidence on the choke model
Calibration. We focus our presentation on a single well (by increasing the discharge coefficient uncertainty), we verify
flowing into the right branch of the P10 loop. We consider the that the initial tuning is sufficient to get, five days latter, a
network from the well tubing to the test separator on FPSO good estimate of the expected oil and gas production, see Fig.
(Floating Production Storage and Offloading), see Fig. 10. 14. This implies that the initial tuning of the GOR was good
Note that riser tubing and riser choke are not simulated in this enough for the five following days of metering. We effectively
presentation. notice that the GOR derived from tuning remains constant, see
Modelling parameters derive from a single simulation with Fig. 15.
the following configuration:
• At the test separator: set the uncertainties of the phase Permutation testing Assistance tool
flow-rate sensors to vendor accuracy (we assume Periodic calibration is required to keep the modelling close to
accurate measurements) the real process. Three-phase flow meters could be used but
• At the upstream equipment and inflow model: remove test separator measurements through well testing can also
the residuals of the modelling parameters (GOR, BSW, provide sufficient and accurate information.
choke discharge coefficient, productivity index and Production at the Girassol field is based on a loop
friction factors). configuration where sea-lines are connected to each other
The vendor accuracy of the flow-rate sensors is small. through subsea manifolds, see Fig. 1. Each well of a loop is
Therefore, the phase flow-rates computed by the optimizer are routed to a production line, either left or right. There is no
necessarily close to their measured values. Meanwhile, the specific line for well testing.
phase flow-rates derived from the modelling have to match This network architecture and flow assurance issues
these measured values. Assuming a small uncertainty on the strongly impact the well testing strategy:
pressure/temperature measurements, the optimizer is forced to • direct testing leads to increase deferred production;
change the values of the modelling parameters (a particular • direct testing at low flow-rates may lead to instability
situation occurs here since the system has no redundancy, and in the flowlines (slugging);
solution is independent on the sensor accuracy). • direct testing at flow rates below 10 000 bbl/d may
Twice a day in December 2001, measurements on a test lead to a fluid temperature lower than the paraffin
separator were carried out at the Girassol field to estimate the formation temperature (about 40°C);
production of the well. We propose to calibrate our system on • direct testing for the nearest wells leaves the upstream
one of these tests. Then, after a certain period of time, we will flowline full of dead fluid during the test, and hydrate
compare the oil, water, and gas flow-rates predicted by the inhibition with methanol is required.
simulation and those derived from real testing. Therefore, permutation testing (in addition to direct
testing) has been included in the Girassol well monitoring
Well metering. A demonstrative test can be found from strategy:
December 7th to 12th. During this period of time, the • a direct testing connects a single well to a single
production of a well was progressively increased with a choke production line; estimating the phase flow rates of the
opening from 36 % to 49 %, see Fig. 11. Meanwhile, the well is straightforward;
pressure drop in the well tubing remained approximately • a permutation testing connects several wells to both
constant and the one through the choke dropped from 5 106 Pa production lines (but a well is necessarily allocated to
to 2.5 106 Pa. a single line). A series of different permutations leads
5. OTC 14009 DEEP OFFSHORE WELL METERING AND PERMUTATION TESTING 5
to different measurements of the phase flow rates of the maximal accuracy. For instance, if we consider a
each production line. The phase flow rates of each production loop with four wells, sixteen different sequences of
individual well derives from solving a set of linear four direct tests will estimate the production, without any loss
equations. of accuracy. But, only few of them may satisfy operation
constraints.
Testing strategy. Since the wells of a production loop may According to our assistance tool, only one sequence does
produce into either left or right production line, a permutation not require hydrate inhibition with methanol, see Fig. 17: no
can be considered as a set of two well arrangements, left and dead branch is created if we consider the first arrangement as
right. The number of possible arrangements is necessarily the initial loop configuration. Conversely, if we accept a
much greater than the number of wells. For example, let us relative loss of accuracy, permutation testing will keep the
consider four producers: WA, WB, WC, WD. Any sequence production at its optimal level, see Fig. 18.
involving these four wells can be acceptable. One of them is Since both maximal accuracy and minimal production loss
shown on Table 2; notation: WA + WB + WD means a test with strategies may be required, a global weighted criterion is
WA, WB, and WD. actually used to bring all the strategies together and perform
The number of possible test sequences increases drastically the sequence sorting.
with the number of wells, see Table 3. This observation and Theoretically, there is no limitation on the number of wells
the complexity of involved phenomena prevent us from to consider. However, let us remind that the number of
deriving a simple synthetic rule that could be used in operation possible test sequences increases drastically with the number
to select the best test sequences. The latter have to be of wells, see Table 3. In the case of six wells or more, the
compliant with the whole production constraints. computation time can be prohibitive unless one or several
A permutation testing assistance tool has been specifically wells are exclusively allocated to a production line.
designed to achieve this work. Basically, a steady state process
simulator is used for network and gas–lift computation, Conclusion
providing well production in any arrangement, see Table 4 for This paper demonstrates the ability of a well monitoring
few examples. Then, sequence sorting is carried out versus software to provide reliable information for producers:
user strategy, see Fig. 16. production estimate of each individual well, abnormal
behavior detection, validation of hardware measurements and
Flow modelling. Flow rates and production losses are replacement in case of failure.
estimated from a simplified flow modelling: well performance Although the described methodology can be applied to any
curves and pressure loss tables derive from experiments or type of onshore/offshore development scheme, this work is
simulations performed on predictive multiphase software. mainly intended to deep offshore developments, such as the
Specific unit operations implement these tables in the process Girassol field in Angola.
simulator. Thus, any code can be used to generate the flow Based on data reconciliation between field data and flow
modelling. modelling, our well monitoring tool requires a periodic
Contrary to the well monitoring tool, this permutation calibration to keep its modelling close to the real process. This
testing assistant is an off-line software. However, a periodic tuning derives from test separator measurements.
tuning against real process data is recommended in order to Since a combination of direct and permutation well testing
adjust BSW or GOR of each individual well. is presently involved at the Girassol field, we also designed a
second tool to compute the optimal test sequences versus usual
Sequence sorting. After network configuration and production and operating constraints.
calculations, a sorting service provides an ordered succession Intensive use and positive feedback will confirm the
of permutations. usefulness and reliability of this work. This will be the main
The initial number of possible sequences is Ckj where k is topic of a second paper.
the length of the sequence and j is the total number of possible
arrangements, see Table 3. Some of them do not comply with Nomenclature
thermal constraints and are removed. The same applies for BSW = Basic Sediment and Water (water volume
singular sequences. flow/liquid volume flow), m3/m3
For one sequence of k arrangements, there are k! orders. BSWm = Measured value of BSW, m3/m3
Applying an order-dependent criterion, the best order is Flow = Total mass flow rate, kg.s-1
selected. The simulator computes the expected test accuracy, GOR = Gas-Oil Ratio (gas volume flow/oil volume
namely the accuracy of each individual well production, GOR, flow), Sm3/m3
and BSW. Finally, the remaining sequences are sorted with GORm = Measured value of GOR, Sm3/m3
respect to user strategy, see Fig. 16. Typical simulation results P1 = Well tubing upstream pressure, Pa
are shown on Table 5. P2 = Well tubing downstream pressure, Pa
P3 = Choke downstream pressure, Pa
Features. Direct testing is intuitively the best solution to reach Pmi = Measured value of Pi, (i = 1, 2, 3), Pa
6. 6 E. ZAKARIAN, A. CONSTANT, L. THOMAS, M. GAINVILLE, P. DUCHET-SUCHAUX, AND P. GRENIER OTC 14009
T1 = Well tubing upstream temperature, K
T2 = Well tubing downstream temperature, K Choke Pm3, Tm3
T3 = Choke downstream temperature, K
Tmi = Measured value of Ti, (i = 1, 2, 3), K Pm2, Tm2
ΓWP = Tuning parameter for well tubing pressure drop
ΓWT = Tuning parameter for well tubing heat balance
ΓCP = Tuning parameter for choke pressure drop
ΓCT = Tuning parameter for choke heat balance Well tubing
ΓWP,ref = Calibrated value of ΓWP
ΓWT,ref = Calibrated value of ΓWT
ΓCP,ref = Calibrated value of ΓCP
ΓCT,ref = Calibrated value of ΓCT
σPi = Uncertainty of Pmi (i = 1, 2, 3) Pm1, Tm1
σTi = Uncertainty of Tmi (i = 1, 2, 3)
Fig. 2: Example of a production network
σBSW = Uncertainty of BSWm
σGOR = Uncertainty of GORm
EXTERNAL
σWP = Uncertainty of ΓWP COMPONENTS
σWT = Uncertainty of ΓWT Man Machine
Interface
Supervisor
σCP = Uncertainty of ΓCP
σCT = Uncertainty of ΓCT Solver DCS
References
1. Van der Geest, R., “Reliability Through Data Reconciliation”,
OTC 13000 presented at the 2001 Offshore Technology CAPE-OPEN
Conference held in Houston, Texas (2001). Simulator Executive SIMULATION
2. Braunschweig, B., Paen, D., Roux, P., and Vacher, P., “The Use of Framework ENVIRONMENT
CAPE-OPEN Interfaces for Interoperability of Unit Operations and
Thermodynamic Packages in Process Modelling”, The European
Refining Technology Conference, Paris, France (2001). See also
http://www.colan.org.
Unit Operation Thermo Server
Figures
CAPE-OPEN
COMPONENTS
Wellhead Manifold Fig. 3: Well monitoring software overview
Fig. 1: Girassol subsea loop
Fig. 4: Sensor failure simulation
7. OTC 14009 DEEP OFFSHORE WELL METERING AND PERMUTATION TESTING 7
3.00E+07 3.00E+07
2.50E+07
2.50E+07
2.00E+07
Pressure (Pa)
2.00E+07
1.50E+07 Reconciliated bottom hole pressure
Reconciliated wellhead pressure
Pressure (Pa)
Reconciliated manifold pressure
1.50E+07
1.00E+07
Bottom hole pressure
Wellhead pressure
1.00E+07 M anifold pressure 5.00E+06
0.00E+00
5.00E+06 0 20 40 60 80 100 120 140 160
Time (min)
Fig. 8: Reconciliated pressure sensor measurements (activated
0.00E+00 failure detection)
0 20 40 60 80 100 120 140 160
Time (min) 1.60E+02
Fig. 5: Pressure sensor measurements
1.40E+02
3.00E+07
1.20E+02
2.50E+07
1.00E+02
2.00E+07 Gas-Oil Ratio (Sm3/m3)
8.00E+01
Pressure (Pa)
Total mass flow rate (kg/s)
1.50E+07 6.00E+01
Reconciliated bottom hole pressure
Reconciliated wellhead pressure
Reconciliated manifold pressure
1.00E+07 4.00E+01
2.00E+01
5.00E+06
0.00E+00
0.00E+00 0 20 40 60 80 100 120 140 160
0 20 40 60 80 100 120 140 160 Time (min)
Time (min) Fig. 9: Production estimate (activated failure detection)
Fig. 6: Reconciliated pressure sensor measurements (no
preliminary failure detection)
1.60E+02
1.40E+02
1.20E+02
1.00E+02
8.00E+01 Gas-Oil Ratio (Sm3/m3)
Total mass flow rate (kg/s)
6.00E+01
4.00E+01
2.00E+01
0.00E+00
0 20 40 60 80 100 120 140 160
Time (min)
Fig. 7: Production estimate (no preliminary failure detection)
Fig. 10: Typical Girassol production line
8. 8 E. ZAKARIAN, A. CONSTANT, L. THOMAS, M. GAINVILLE, P. DUCHET-SUCHAUX, AND P. GRENIER OTC 14009
Bottom hole pressure
Wellhead pressure
Choke downstream pressure
Choke opening
Time
Fig. 11: Pressure measurements and choke opening (%) at a Fig. 14: Girassol well: production simulation (second case)
Girassol well
Fig. 15: Girassol well: GOR calibration
Fig. 12: Girassol well: production simulation (first case)
Production
loss Oil production
uncertainty
Subsea valve Water production
operation uncertainty
Dead branch
Gas production creation
uncertainty
Methanol
consumption
Fig. 16: Various strategies for well permutation sequence sorting
Fig. 13: Girassol well: calibration of the modelling parameters
9. OTC 14009 DEEP OFFSHORE WELL METERING AND PERMUTATION TESTING 9
P1021 P1011 Fluid source Gas-lift model
P1022
Right Manifold Connection between wells and production loop
Three-phase flow model. Gridded model.
M102 M101 Pipeline
TEST
Used for well tubing, sea-line, and riser
Left Piping Simulation of small scale piping networks
P1012 Sensor Hardware sensor model
Definition of fluid composition and Productivity
Inflow
P1021 Index relation
P1022 P1012
Table 1: Well monitoring unit operations
Right
M102 M101 Test number Well arrangement
Left
TEST 1 WC
2 WA + W B + WD
P1011
3 WB + WC
4 WB + WD
P1011
P1021 P1012 Table 2: Example of a well test sequence
Right
Number of Number of well Number of well testing
M102 M101 wells arrangements sequences
Left
TEST 3 12 220
4 28 20475
P1022
5 60 2.12E+06
P1011 6 124 1.52E+09
P1012
P1022 Table 3: maximal number of well testing sequences
Right
M102 M101
TEST
Left
P1021
Fig. 17: direct testing sequence
P1021
P1022
Right
M102 M101
TEST
Left
P1012
P1011
P1021 P1012
Right
M102 M101
TEST
Left
P1022 P1012 Table 4: Well permutation tests
P1021 P1012
Right
M102 M101
TEST
Left
P1022 P1011
P1021 P1012
TEST
Right
M102 M101
Left
P1022 P1011
Fig. 18: Permutation testing sequence
Tables
Unit operation Description
Block valve Routing valve (open/closed)
Choke Three-phase model
Table 5: Test sequences (production loss minimization)