IPTC 17009

Architecture and Implementation of an Optimization Decision Support

System
Gbel, Derek (Sarawak Shell Bhd); Briers, Jan (Shell Global Solutions (Malaysia) Sdn. Bhd.); Chin, Yee Men
(Shell Global Solutions (Malaysia) Sdn. Bhd.)
Copyright 2013, International Petroleum Technology Conference
This paper was prepared for presentation at the International Petroleum Technology Conference held in Beijing, China, 2628 March 2013.
This paper was selected for presentation by an IPTC Programme Committee following review of information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the International Petroleum Technology Conference and are subject to correction by the author(s). The material, as presented, does not necessarily
reflect any position of the International Petroleum Technology Conference, its officers, or members. Papers presented at IPTC are subject to publication review by Sponsor Society
Committees of IPTC. Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes without the written consent of the International Petroleum Technology
Conference 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 where and by whom the paper was presented. Write Librarian, IPTC, P.O. Box 833836, Richardson, TX 75083-3836, U.S.A., fax +1-972-952-9435

0. Abstract
This paper describes the implementation of an optimization advisory system by Sarawak Shell Bhd. on the
integrated gas production system in Sarawak, to help monitor and optimize in real-time the more than 100 wells
on 30+ facilities of the system, governed by different Production Sharing Contracts. We highlight how multiple
disciplines are part of the solution, how the solution has been embedded to provide real-time advisory to planners
and operators, and the challenges faced.
The technology described in this paper provides real-time advisory on flow rates for individual wells and platforms
to network coordination operators to achieve a gas supply that meets overall gas demand for the onshore LNG
plants, maximizes revenue from condensates production and CO2 blending, and transparently shows the
business value impact of giving preference to short- or long-term optimization objectives. It takes into account
existing commercial and contractual constraints, as well as financial and economical elements. The optimization
approach is data-driven and features multiple, mutually dependent objectives and constraints.
The data-driven approach has proven significantly faster and more practicable than prior physical model-based
approaches. Field trials have proven a condensate production increase of up to 7,000bbl/d while at the same time
improving Ultimate Recovery of the sweet and sour gas blend of the production system by up to 2 months per
year of applied optimizer advise. At the same time, gas nominations are met as they were, and commercial and
quality constraints are adhered to. The Return On Investment of this solution has been less than a month.
This is one of the first successful attempts to implement truly-real-time optimization in a production environment of
this size and complexity, including a complicated set of commercial and contractual constraints. The approach
opens the door to application to a wide variety of real, complex, mixed-integer non-linear optimization problems,
and their organizational application, such as emissions management and decision support systems for integrated
IOR/EOR systems.

IPTC 17009

1. Background and Introduction

1.1 Background
Traditionally important aspects of security of gas supply relate to reservoir management, development planning,
and sustaining plant availability and integrity. With increased computing capabilities, increased attention is also
given to non-CAPEX-intensive ways to provide additional, smaller scale improvements to the production and
performance of an asset [11].
Besides having applied gas lift optimization in the Sabah oil assets, Sarawak Shell has tried a number of ways to
achieve automated, continuous optimization of its condensates and CO2 production in Sarawak through modelbased optimization. An implementation [9] of a Mixed-Integer Non-Linear Programming [7, 10] model has
rendered valuable results for the Medium Term and Long Term planning horizons. However the complexity of the
network and the resulting amount of computing power required, has stood in the way of applying it to optimization
in the Short Term, day-to-day operational domain.
A number of approaches have been tried: rigorous Process Engineering simulation and optimization tools linked
to the data historian, and two commercial software packages for optimization. These approaches did not produce
the desired results in a way that was stable and fast enough and sustainable by local staff. The approaches are
typically based on physical models which require a great deal of accuracy (often relying on details that are not
readily available), and secondly require long calculation times around 3 hours and more per optimization run.
1.2 Introduction
This paper describes a successful approach that features easier to apply and sustain data-driven models in a
software package that is a standard within Shell, and with which hundreds of users are familiar [4, 12]. The
objective has been to maximize condensates production in the gas network, as well as to produce more CO2 out
of the reservoirs, closer to the technically allowable limit, by providing to operators an optimization solution that is
fast enough to be applied in the real-time domain, and simple enough to be used and maintained locally as part of
the routine work processes.
In the cause of the project to achieve the above, we effectively realized three benefits that will be described in
detail in this paper:
1. Increased, detailed transparency of production network status in all nodes, and an ability to accurately
predict gas compositions entering the LNG trains 8 to 16 hours in advance, making the network more
manageable;
2. Maximized production of condensates and contaminated gas (and therewith maximized revenue and
long-term value), and providing real-time estimates of the short-term and long-term value impact of
decisions to operators and asset leaders, enabling more transparent and value-based decision making on
a day-to-day basis;
3. Solved technical and non-technical problems and identified improvements based on learnings that were
purely the result of the implementation process itself.
The resulting product can be summarized as an MS Excel interface that links to an optimization engine in the
data center, that is primarily used by production planners and network operators. The application is mainly in the
monthly to daily operational domain.
In this paper, first the asset context, its complexity and constraints are described, followed by a methods section
where details are given on: the optimization approach and objective, the architecture of the prediction and
optimization solution, and considerations on implementation, embedding, change management, and sustaining. In
the following results section, data and examples are given on the three benefits listed above, followed by a
discussion section where more background is given, as well as how the achieved solution could be applied in
different contexts.

IPTC 17009

G16

DP
KT
G33
N11
N13S

SC

TD

N14
F21
PA/PB

N13

3rd Party
3rd Party

C21

H17

CB
C22

TER
3rd Party

RX

RY

RZ
3rd Party

LNG Plant M

LNG Plant Q

LNG Plant P

Onshore
Customers
= Different Production Sharing Contracts

Figure 1 Simplified overview of the integrated production system. Names and lines are fictitious but do reflect the
true complexity of the system.
1.3 Asset Context
The Sarawak integrated gas asset is located in the South China Sea off the coast of the state of Sarawak in EastMalaysia, and comprises of: 3 onshore LNG plants, a few onshore customers and an LNG cargo export facility,
50+ offshore hubs and jackets, 100+ wells, numerous connecting pipelines, managed under 8 Production Sharing
Contracts, and where a number of different operators are active - see Figure 1, above.
The asset produces offshore natural gas of variable composition that is blended in the source system pipelines
and liquefied in the onshore LNG plants. Associated condensates are also collected and processed in the LNG
complex. Constraints in the system are physical, planning, and contractual / commercial in nature:
Physical and Infrastructural Constraints
Pipeline diameter, maximum flow and pressure
Pipelines and risers routing
Well and reservoir production envelopes (maximum drawdown, sand production)
1)
Facility capacity constraints (water handling, compressor capacity)
LNG train handling capacity:
2)
Gas volumes (that vary depending on contaminant contents)
CO2, H2S, C5+ handling
Intake pressure
1)
Plant trips
Liquid handling capacity: condensate volumes and tank levels
1)
The system does not explicitly take into consideration equipment status or equipment constraints, but the operator can manually adjust the
overall, current platform capacity based on the latest operational information.
2)

The system does not carry LNG train capacity constraints, but applies the gas demand as a constraint.

IPTC 17009

Planning Constraints
LNG demand and gas production nominations
Shutdowns
Well tests and well interventions
Well and reservoir management planning constraints
Maintenance and pigging activities
Commercial Constraints
Maximum allowable CO2 percentage per LNG train as stipulated in the Gas Sales Agreements
Gas quality / Gross Heating Value (GHV) of the blended gas mix
Ultimate Recovery: long term it is important to ensure that sufficient sweet gas is available to blend down gas
that contains a too high percentage of contaminants;
Production deficits at LNG plants can be compensated by spilling gas between risers, backfilling with supply
sourced from other PSCs. Aim is to keep this at a minimum and return the gas in kind when supply/demand
allows, instead of reconciling the balance financially.
Priority of Supply: stipulates which PSC and which field is allowed to produce first, second, etc. to meet an
overall demand objective.
Supply Ratios: in a situation where multiple operators produce into the same PSC, their production is
constrained according to a static, contractually agreed supply ratio: e.g. operator 1 can produce 2.5 times as
much as operator 2 can produce 1.3 times as much as operator 3. The ratio is based on the operator capacity
and overall average PSC demand level. If one or more operators are in deficit, the remaining operators can fill
the gap, again according to their respective ratios.
Economic, commercial, and financial stipulations in the PSC contracts. See next section.
All constraints mentioned above have been implemented into the optimizer to ensure that the optimized advice is
within the constraints that exist in the system and therewith applicable.
2. Methods
2.1 Optimization Approaches
What is optimization? An optimum is not the single point it implies to be, but very much depends on ones point
of view. The traditional Smart Fields objective is to Maximize business value by maximizing production and
ultimate recovery, and minimizing costs and HSE exposure. But typically short-term maximum value (e.g. more
production), is at odds with long-term maximum value (e.g. more ultimate recovery), and more ultimate recovery
can be achieved, but at an investment price (e.g. higher costs). Where the maximum should be, is ultimately an
Asset Leaders decision. And the key to being able to make such decision, is to have transparency on the value
and impact of the various options.
Business Value also has different meanings to different people. Three value definitions have been considered for
our optimization: financial, economical, and operational, as follows:
Financial Value impractical and too complex
This constitutes the dollars that hit the bottom line at the end of the day: total profit after royalty, tax and interest.
The dollars are generated from gas and condensate sales revenues at the going rates, minus taxes and the costs
of investments and operations.
The calculations would have to include some or all of: operator shares, contractual pricing scheme, spot prices,
the revenue or penalties associated with gas borrowing, cost recovery ullage or ceiling, PSC investment levels
and capital cost recovery status, liquids allocation scheme, currency agreements, royalty, tax and profit gas
percentages. It should in theory also include some accounting practices such as: assets and liabilities, cost and
revenue categorization, exchange rates, asset depreciation, cost of capital, CAPEX and OPEX levels, and tax
situation.
Further drawbacks of this approach are: 1. the complexity to go from a daily production number to a bottom-line
dollar this way, is enormous and would make optimization practically impossible, and 2. it does not take into
account the (long-term) time-value of gas and condensates production and reserves.

IPTC 17009

Economical Value too theoretical

Perhaps a more correct, but also more theoretical approach would be to optimize on asset economical value [8].
This would take into account a slightly simplified version of the short-term financial approach, and augment it by
placing the financial returns and expected returns in the context of the long term, and comparing the time value of
money in real terms by applying a Nett Present Value of expected cash flows.
This would shift the focus away from tomorrows bottom line profits, and bring into the picture the value of
expected Ultimate Recovery, and for instance the value of being able to produce the contaminated gas reservoirs
for longer, as a result of good short-term sweet gas production decisions.
The drawback here is that this approach also makes the model complex and too theoretical. Furthermore, when it
comes to comparing the value of reserves versus a daily production decision of 100 MMscf/d more or less from
one platform, the production variation becomes almost invisible in the bigger picture.
Operational pragmatic and fits in existing work processes
A more practical approach is to focus on daily production targets, actual production, and condensate/gas
deferments this is not as rigorous as the Financial and Economical approaches, however, the data is
measurable and exists in real-time without the need to invoke and maintain complicated calculations and
assumptions, and output can be generated in terms that operational staff can work with.
2.2 Applied Optimization Approach and Objective
In terms of the optimization objective of the optimizer in the solution, we have chosen to keep the optimization in
the operational domain, but enhance it with the ability to apply relative factors to inputs, constraints, and target
values.
This means for instance that the optimization function is not on MMscf/d production levels, but:
(Gas Production [MMscf/d] x Gas Quality [MMBtu/MMscf] x Gas Price [US$/MMBtu]) +
(Condensate Production [bbl/d] x Condensate Price [US$/bbl])
This resembles revenue more closely, and is more correct from an optimization perspective. Gas quality can be
set on a well-by-well or per-platform basis, and gas price per contract or well or platform. We do not include the
exact pricing and quality values, as this information is normally privileged, but it is possible to set relative factors
to gas streams will skew the optimizer to prefer producing more valuable gas while meeting overall gas
production targets as long as blending constraints are adhered to.
At the same time, and in the same way, it is possible to make certain, often conflicting, objectives relatively more
important than others see Figure 2. E.g.: it may be more important to maximize daily condensate revenue, than
it is to produce as much contaminated gas as allowable within contractual limits, or the other way around. The
optimizer provides an easy way of including such priorities into the solution, and therewith provides a fast and
transparent way of comparing the options to support decision making.

IPTC 17009

Stream

Figure 2 The configuration interface where relative importance can be given to objectives,
objective min/max constraints
can be set, what-ifif scenarios tested, and currentcurrent and optimum operating point are given. Numbers dont reflect
actual data. The given example preferences the production of high-CO2 gas over condensates. If one were to
prefer condensates, the factors could be reversed, and the optimizer would render corresponding results.
Our typical objective function is: To
o maximize condensate revenue while meeting gas demand at maximum
maxim
achievable gas revenue (value) without gas borrowing, whilst ensuring that CO2 content is always at its
i allowable
maximum percentage, and reservoir management priorities are taken into consideration
consideration. This satisfies most
short-term and long-term objectives.
The most important constraint is the gas demand at the three LNG plants. The existing overcapacity in the gas
system provides the room for optimizing how we meet that demand.

IPTC 17009

2.3 Architecture of the Solution

The optimizer is a powerful addition to the suite of asset management applications and solutions, but for the
overall solution to work, quite a number of contributing parts need to be in place and in working order, as can be
seen in Figure 3. This has an impact on the sustainability and maintenance requirements, as well as on user
training. See section 2.5 below.
The important pieces are: the real-time data flow from offshore instrumentation to onshore data historian, the
added virtual metering tags for where physical instruments are not available but you do need data, the
optimization models and constraints, the end-user interface, and the organizational change required to make the
usage of the optimizer a part of the day-to-day workflows of planners and operators.

OFFSHORE DOMAIN

DATA CENTER

OPERATIONS CONTROL CENTER

CONSTRAINTS

OPTIMIZER

physical
planning
commercial

optimization model
2,500+ field
instrumentation

application- and
excel user interfaces

3,000+ tags
RESULTS
data-driven MV-PV rels

100+ wells
50+ jackets
flow-, and pipelines

virtual metering (PU RTM)

real-time data

10+ distributed control systems

planners
network operations
offshore operations
offshore data historians

DECISIONS

onshore data historian

flow settings for wells and

platforms

Copyright of Sarawak Shell Bhd.

offshore to onshore telecoms links

RESTRICTED

Figure 3 The high-level architecture of the solution

2.4 Inside the Optimizer

The Sarawak RTO system has been configured using the FieldWare Production Universe (PU) software. PU is
used extensively throughout the Shell Group. Originally developed to monitor and optimize the production of
individual wells and platforms [1], more recent PU implementations have focused on asset-wide optimization [6].
PU provides of a number of modules, two of which are:
1. PU Real-Time Monitoring which estimates dry gas flow, as well as associated condensate, water and CO2
flows for each well in the system, continuously in real-time. This module also provides the virtual metering
where physical meters are not available (e.g. wells that dont have dedicated flow meters, see figure).
2. PU Real-Time Optimization which combines well rate estimates and data-driven Manipulated Variable
(MV) Process Variable (PV) relations with electronic field data to continuously maximize an objective
function whilst simultaneously honoring constraints in real-time.

IPTC 17009

The requirement to apply optimization in the real-time environment translates into a number of key requirements
for the solution. For one, the system needs to be fast: capable of delivering a new optimal production set point
for the entire field in less than 5 minutes. Also, the system needs to be flexible enough to cope with sudden and
unplanned events, such as a platform-, or LNG train trip. And the system needs to be user friendly, with an
intuitive Graphical User Interface (GUI) and an easy means to change constraints and objectives.
Using the PU application as the vehicle to implement the asset-wide RTO system has advantages. All real-time
monitoring and reporting capabilities of PU are already available at the Asset-Wide Monitoring level in PU-RTM.
With the RTO system running online using real-time data available from the RTM system and the data historian,
the users can access a real-time overview of how the entire production network is performing, what the real-time
well flow rates are, how the pressure profiles look, compositions of the various streams (CO2, H2S), how the
current demand is matched by the current production, etc.
The Exception Based Surveillance (EBS) [13] capabilities of PU can email diagnostic messages to staff in case of
faulty transmitters, or if certain constraints are violated. And automatically emailed daily reports and historical
trends make it easy to analyze productivity and performance against plan on a daily basis.
At the core of the Sarawak RTO system there are three main components:
1. The topology of the production system;
2. The models for the wells, platforms, and pipelines;
3. The optimization engine with objective function.

RX

CM

RXtoRY
CX

RYfromRX

E43
E45

RY

JT

RYtoRZ

RZfromRY

RZ

LT

Figure 4 Overall topology of the Sarawak network

Topology
The entire Sarawak production system is represented in a topology, see figures 4 and 5. Some wells, all
platforms, interconnecting pipelines, riser platforms, down to the trunk lines that feed the LNG plants are included.
The topology is constructed such that it gives an accurate view of the production system. However, a number of
simplifications have been implemented.
For the majority of the wells there are no individual topological elements. Only if wells in a cluster or field have
production of a unique composition, are they represented individually. For the majority of the fields the

IPTC 17009

composition of the natural gas (CGR, CO2, H2S, BTU) has been assumed identical for all wells based on what we
have seen in well tests or PVT data.. These wells are lumped together
together in the topology into a platform object.

Figure 5 Station topologies of the Sarawak Gas System. Wells from these fields are grouped in
into a cluster.

Also the pipelines are represented in a simplified manner. Not every bend, restriction, change in elevation,
diameter, or the hold-up
up of liquids and slugs is modeled. The pipelines are simply represented by a model that
relates pipeline flows to pressure drops
rops across the pipe. The models are constructed using linear models with gas
composition and the associated stream feature of PU. The pressure-to-flow
pressure flow data is derived from the process
historian,, or from a pipeline simulator.
For the optimization problem
m the pipelines are important, as they allow the user to set constraints on e.g.
maximum CO2 level (process constraint) or maximum flowrate (equipment constraint). Also, the pipeline
topological element allows the user to monitor in real-time
real
the flow and composition at the different nodes in the
system. One aspect of the Sarawak topology that increases the complexity (and non
non-linearity)
linearity) of the problem, is
the cross-flows
flows between the risers. Depending on the gas demand, there can be cross-flow
cross
from Riser X (RX) to
RY, and from RY to RZ. The RX, RY
Y and RZ riser platforms behave like multiplexer nodes where the flow can go
down two paths.
Models
The PU RTO software was originally developed to optimize a cluster of wells for lift gas allocation,
allocation typically on a
platform or production station [4,12] level.
level The optimization models for these wells are data
data-driven. They are
constructed using historic well tests, well production variations (for instance by adjusting amount of lift gas
injected in that well), or multi-rate
rate well tests to relate well productivity to changes in a real
real-time well head
parameter and thus modeling the performance of each well. The RTO modeling engine uses the individual data
driven well models to compute the objective
objecti function in real-time. Overall optimization is achieved by adding the
individual contributions of each well given constraints such as total lift gas available,, and taking into account back
pressures. The data-driven
driven approach speeds up the optimization computation,
omputation, and makes it accurate enough and
robust.
The models are constructed in such a way that the relationship between the MV (Manipulated Variable) and the
PV (Process Variable) is clear, and that the MV can easily be implemented by an operator in the
t field e.g. set
point for a given well gas flow rate. The well and platform models contain the relationship between the measured
wet gas flow (the MV of the system), and the total dry gas, CO2 and condensate flows. Since the relationship
between wet- or dry gas and CO2 output is assumed linear, it can be translated straightforwardly into the model as
a constraint to be tested against. See Figure 6.

10

IPTC 17009

Figure 6 A linear RTO model. The green diamond shows the current working point

As mentioned, for fields where all well flows have the same composition, the wells are clustered into a single field
object to simplify the model. The MV here is the total field wet gas flow (WGF). When the optimizer computes an
optimal total WGF for a field, it is left to the Offshore Installation Manager how to distribute this demand across
the available wells.
The last simplification is the assumption that there are no back-pressure effects between individual wells or
between platforms. This is a reasonable assumption given that we use flow as the MV in the system, and not
pressure. PU does support back-pressure effects in its modeling engine, so this is something that could be added
in the future if this would lead to a required improvement to the accuracy of the model in case this is found to be
necessary.
The simplicity of the optimization models provides a beneficial robustness to the system: e.g. if the reservoir
pressure declines, or there is scale deposition in the tubing, or the WGR increases, the CGR/CO2/H2S/BTU
compositions stay valid. If required it is easy to change the corresponding ratios in the model.

Engine
The optimization engine is required to generate convex, stable and repeatable solutions in a short time, that are
global, not local, optima. Some optimization approaches pose the risk of generating instable solutions: an advised
flow rate distribution across the network where a large percentage of wells and platforms is required to change
flow rate to achieve a marginal (and theoretical) increase in condensate output. Our engine is required to provide
set-points that are close to the current state of the network. With speed in mind, ideally this optimum solution is
derived via linear programming. This is not feasible because of the following.
The standard oil field optimization problem is non-linear to a degree, and the non-linearity increases in
complexity when there is interaction between wells, e.g. through backpressure effects in the production headers.
Having assumed no backpressure interaction and linear flow response, still non-linearity is introduced into this
optimization by the links between the riser platforms and the constraints on the composition of the gas through the
trunk lines. This forces the application of more sophisticated techniques to handle this non-linear optimization
problem which are not described in this paper.

IPTC 17009

11

2.5
5 Implementation and Sustainability
The implementation follows a phased approach, where the initial and most important step is to ensure the
availability of a real-time data infrastructure.
ructure. Then virtual metering models are created, and models that define
well- and process response given the variation of important manipulated variables. Assuming the real
real-time data
infrastructure is there and does not require a lot of fixing, the bulk of the work is in really understanding the
composition and dynamics of the production network. Dozens of people from many different disciplines are
involved in the
he management of this network, making data gathering and understanding a challenging effort. Yo
You
will find that each discipline creates their own practical view on reality.
For instance, out of the 35,000+ real--time data tags that are available, process engineers
ngineers or planners can look at
different tags to do their simulations
lations or make decisions,
decisions than operators do, and their interpretations may vary.
Especially in areas where multiple ways of measuring are available, each discipline tends to use the way that
suits their requirements best. For the reconciliation of condensates, this leads to interesting
intere
insights. Subsurface
engineers go by the PVT samples to base their long
long-term
term condensate production forecasts on. Hydrocarbon
Allocations use the contractual methods to allocate condensate production to fields (not wells) on a monthly basis
using gas composition as a basis for comparison. And prior to this optimization implementation, the Operations
Control Center mainly looked at onshore condensate meters to make sure that the tanks didnt overflow.
overflow
So multiple disciplines in the same network can give
give different perspectives on condensate origin and amounts.
amount
Information and measurements to use as a basis for optimizing condensates production may vary as a result.
Before the model will produce simulations and predictions that match reality, the understanding
understa
of the production
system must be very correct, clear and detailed
detailed, see Figure 7. One of the positive side effects of doing an
implementation like this, is that it brings all information and assumptions together and helps create a more shared
understanding of the production system.

Figure 7 A graph showing both a measured and an estimated CO2 curve. It is important to realize that the blue
curve is a single measurement from a fiscal meter, and the red line is a model estimate based on hundreds of
data tags and models, compounded. It is the red curve that the optimizer simulates to predict what-if scenarios
and select the optimum one, so it is imporant that these curves are well calibrated with what happens in the field.

12

IPTC 17009

Spending time with operations during implementation and testing has also shown that, despite the vast amount of
data and information that is available, a lot of operator experience and assumptions is not captured in any
systems or documentation. This has led to many parameter adjustments in the topologies and models after first
trial of the theoretical models. Especially when engineering documentation is used to define constraints and
physical properties of elements of the system, operators can have different working assumptions: We know the
pipeline flow constraint is 750 MMscf/d (and we also know it can easily do 800), but we tend to use 650 for the
moment because there has been some liquid build-up in the line since the last shutdown, and wed like to slowly
produce that liquid out of the line first.
Having put in place the architecture of section 2.3, and the optimizer of section 2.4, and having calibrated the
tags, measurements, assumptions, constraints, models, etc. leaves us with a large collection of things that need
documenting and maintaining. In terms of sustainability, two elements we find are important to put into place:
1. A central team, like the Smart Fields team or a combination of Control & Automation / Maintenance and
IT teams, needs to own the end-to-end solution as well as the models etc. It helps if there are dedicated
resources to whom users can direct or escalate problems with data tags, models, user interface, the
working of the system etc. for those resources to resolve.
2. A collection of inventories - also owned by this team, that are documented, signed-off, and stored at a
managed location to be kept up-to-date and maintained, if need be with a maintenance plan just like for
pieces of equipment. An inventory is a list of related items that are an integral element of the optimization
architecture. Examples:
Applications and owners
Critical tags
Calculated tags
Tags hierarchy
Models (RTM, RTO)
Users and passwords
Super and mega-users
Competence and training records
Instruments and meters, and to which historian
tags they are linked

PCD IT Hardware
OD IT Hardware
Service Level Agreements in-, and external
Contracts
IT system Interfaces
Business processes
Roles and Responsibilities
Telecoms and telemetry hardware, contracts
Etc.

Finally, change management is an important consideration in this implementation. The effort to adjust operator
workflows and train new users should not be underestimated. It requires weeks of working closely together with
both day-, and night-shift of the operations center to really understand how they make decisions, what
assumptions they carry, and what concerns or problems they have using the solution and interface. Secondly,
support is required from all levels in the organization. If operators follow an advice from the system, they do so to
help the operations manager and asset leader maximize their production. If something should trip in the field as a
result of them following an advice, they should not be held responsible. This is probably one of the biggest
hurdles to operator acceptance of an optimization support system.
3. Results
A single optimization run at the moment takes around 10 minutes to execute with the engine running on a normal
laptop, which is just about acceptable for operators and quite an improvement to alternative optimization
approaches most of which include physical models, where a single run on a fast computer can take 3 hours or
more. A scenario check / what-if analysis without optimization takes a few seconds. We are developing some
changes to the optimization engine such that an optimization will also take around 1 minute.
As stated in the introduction, there are three areas where benefits have been demonstrated: transparency,
additional production, implementation learnings.

IPTC 17009

13

3.1 Transparency of production network status

Given the amount of measured tags, virtual tags, and model output we now have at our disposal, the optimizer
user interface allows the user to double-click through the entire production network and show graphs of gas-,
condensate-, CO2-, H2S-, and water flows, pressures, temperatures, gas compositions, modeled vs. measured at
each node of the network. This helps with monitoring, understanding, and managing the network, platforms, and
wells.
An item that was picked up with particular interest by operations staff, is the ability to reliably predict the gas
composition entering the slug catchers of LNG M, Q, and P some 8 to 16 hours in advance of the gas actually
arriving at the slug catchers. The predictions are based on the compounded well-, platform- and pipeline models
that monitor and calculate gas production as soon as the gas leaves the well. See Figure 8.
In the past, when there were upsets or changes in the network, network operators would issue a phone call and
fax message to LNG operations counterparts every two hours to communicate an expected change in CO2
content in the gas arriving at the LNG plants. If the change constitutes a spike CO2 percentage above the allowed
technical CO2 limit, LNG operators need to take measures to prevent the LNG plant from tripping. This
communication was re-active and discontinuous.

Figure 8 A chart that is given in the operator user interface clearly shows the difference between what is
currently being / has been measured at the entry point of the onshore slug catcher 1 of LNG M, overlaid with what
the models predicted 8 to 16 hours before, when the molecules left the wellbores. Based on the latter, the model
gives a reliable prediction of the gas composition that is about to hit the slug catchers as per graph information
of strong interest to LNG operators.
Now the continuous CO2 prediction charts can be shared with LNG operations in real time, and network operators
still have time to adjust blending in the offshore network to mitigate for off-spec CO2 content reaching the onshore
slug catchers. This also allows for reducing the margin of error that operators tend to keep, and therewith for
increasing the amount of CO2 that can be produced, and therewith increasing asset business value.
3.2 Maximized production of condensate and contaminated gas
Figure 7 already shows the effect of increased CO2 output as a result of a change that was made in the
production network based on optimizer advice. The field trials have demonstrated that the optimizer enables an
increase in CO2 production of up to 30% while staying within technical limits of the receiving LNG plants.
Translated, this roughly means that for each year that such increase would be achieved consistently, up to 3
months of ultimate recovery would be gained for the production network assuming sweet gas reserves declining
at the current rates and no new wells are added to the system.

14

IPTC 17009

Figure 9 The ultimate impact of the optimization: at the time of the dashed, brown line, the optimizer advice was
implemented in the field: the production of three platforms was adjusted. In blue on the chart below can be seen
that the overall total gas production remains the same, whereas the associated condensate production increases.
The second optimization objective is clearly demonstrated in Figure 9: condensate output maximization. The chart
(derived from the optimizer application user interface) shows what happens when the advise that is given to
operators in the excel user interface of Figure 10, is implemented in the field, and illustrates therewith the principle
of the optimization. Flow targets of wells and platforms are adjusted in such a way that overall gas production
remains the same as per demand requirements, but associated condensate (and CO2-) production maximizes
within the existing physical, commercial, and planning constraints.
The amount of condensate that can be gained is obviously dependent on operational circumstances. E.g. if a high
condensate producer is shut down, the gains and/or room for optimization will be more limited. Implementation of
optimizer advise across the network has so far proven that condensate gains as high as 7,000bbl/d can be
achieved in comparison to operating in modes that would have been otherwise selected based on traditional
methods without the availability of an optimizer to support decisions.
As shown in sections 2.3, 2.4, and 2.5, an implementation requires quite a bit of effort. But all in all the effort and
cost are far less than the new wells that would have to be drilled and hooked up to achieve similar production
results.

IPTC 17009

15

Gas Operation and Coordination Centre

Read Values

LNG
Plant M
Plant Q
Plant P
LNG
Plant M
Plant Q
Plant P
Header

Current
3.80
4.10
4.57

Current
682
535
172

Min

N11
N11toRZ
RZfromN11
F21 PPA
F21 PPB
G16 PPA
G33 PPA
C21 DRA
C21 HUB
C21 Export
N13
TL1_TL2
TL3_TL4
TL5_TL6
Slug Catcher
SC1
SC2
SC3
Supply Ratio

Write Values
CO2, mol %
What-if Optimized
4.91
6.10
4.61
6.50
5.64
6.20

Current
9.00
1.00
14.00

Gas Supply, MMscf/d

What-if Optimized Demand
629
690
690
528
529
530
167
170
170

Write 3rd Party

H2S, ppmv
What-if Optimized
31.00
86.00
28.00
139.00
42.00
5.00

Gas Imbalance, MMscf/d

Current
What-if Optimized Demand
-30.0
-31
-3
-18.3
26
-8
7.0
6
9

Gas, MMscf/d
Max
Current
What-if Optimized
690
680.0
178.0
690.0
664
387.0
70.0
311.0
664
387.0
220.0
311.0
420
87.0
48.0
230.0
720
622.0
665.0
665.0
480
283.0
298.0
258.0
450
340.0
156.0
183.0
210
195.0
31.0
178.0
280
93.0
134.0
231.0
540
293.0
118.0
363.0
480
39.0
477.0
347.0
930
590.0
121.0
927.0
960
394.0
28.0
326.0
900
864.0
311.0
446.0
CO2, mol %
Delay (h)
6.5
4.80
4.91
6.1
7.5
6.5
4.10
4.61
6.5
12.0
6.5
4.57
5.64
6.2
12.0

Remark

Gas, MMscf/d
Third Party Supply
Other 1
82
Other 2
46
Other 3
138
Other 4
102
Other 5
25
Other 6
164
Other 7
69

Meet gas demand

Current

Gas, MMscf/d
What-If Optimized Demand
910
1179.0
578
763.3

User Entry

Auto Generated

Platform
F21C
F21W
F21S
T1DR
G23W
G23E
F18
G23E
G23W
G33
D1
G16
N11
K1
T18
N13
T19
N13S
H8
C21_W102
C21_W103
C21_W106
C21_W107
C21_W108
C22
RXtoRY
RYfromRX
RYtoRZ
RZfromRY
N11toRZ
RZfromN11
Other 1
Other 2
Other 3
Other 4
Other 5
Other 6
Other 7

Improve condensate

Units

Improve CO2

Other 1
Other 2

Condensate, bbl/d
Current
What-if Optimized
9122
8092
12987
11960
10990
13672
8237
9532
9721

Optimize

What-If

Min
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
90
0
0
0
0
0
0
0
0
0
0
0
0
0

Gas, MMscf/d
Max
Current
What-if Optimized
564
498.0
199
560.0
213
114.0
86
200.0
58
52.0
16
52.0
232
232.0
13
232.0
121
82.0
30
82.0
77
65.0
75
72.0
11
6.0
1
10.0
304
300.0
119
300.0
81
80.0
30
80.0
567
560.0
278
560.0
18
18.0
3
18.0
41
41.0
19
41.0
690
680.0
178
690.0
316
316.0
199
310.0
37
37.0
20
30.0
770
730.0
120
750.0
349
342.0
328
310.0
6
2.0
4
6.0
760
750.0
385
760.0
146
146.0
57
146.0
52
50.0
6
50.0
614
601.0
481
614.0
21
9.0
10
15.0
59
24.0
44
40.0
375
267.0
224
200.0
119
99.0
99
10.0
119
99.0
2
10.0
518
428.0
109
242.0
518
428.0
38
242.0
664
387.0
70
311.0
664
387.0
220
311.0
330
151.0
42
151.0
59
53.0
0
53.0
789
725.0
424
725.0
364
286.0
239
286.0
23
15.0
13
15.0
556
514.0
460
514.0
392
360.0
32
360.0

Figure 10 Screenshot of the operator user interface. Orange fields are for user entry, green fields are measured
or calculated. This interface allows operators to manually enter / change current system constraints and recalculate optimized production setpoints in the far right column. Encircled is how the optimized settings change
condensate, gas, and CO2 production. This interface is for manual use, there is also a planning interface that
allows for generation of optimized 30d-plans taking into account well tests and shutdowns. The interface of Figure
2 allows for the changing of optimization preferences to shift between short-term and long-term optimization etc.
Numbers, their order of magnitude, and names dont reflect actual data.
3.3 Improvements based on implementation learnings
The process of implementing all the elements of the optimization solution is in its own right a good exercise to go
through, for it can reveal a number of items that are worth correcting. Examples in random order:

Faulty or non-calibrated instruments. Sometimes when we struggle to get a model to match reality, it is
not a model problem, but a reality problem.
Different CGR or WGR assumptions can be held between different teams / disciplines / departments
A great deal of knowledge and experience is in peoples heads, but undocumented elsewhere.
Various people in the process keep a small margin of error in their targets or assumptions to for instance
ensure they dont overshoot CO2 limits.
Small errors in daily internal reports presented correct offtake volumes with slightly mismatched produced
volumes
Small errors were discovered in some of the manumatic tools that were created and used by operators to
calculate / spot-check CO2 production, and keeping a bit too much margin.
Some people turned out to monitor a different tag or meter to the one that would have been a better
alternative for their use.
Limited understanding of operational constraints and not up-to-date documentation on manipulated
variables such as well potentials and 3rd-party production lead to miss optimization opportunities.
All individuals in the production system management teams and disciplines tend to know a part of the
production system very well, but not many understand the full system end to end. Working with everyone
has helped to create a more shared and larger understanding of the bigger picture.

16

IPTC 17009

It is hard to put a monetary value on the above, but the small corrections that were made as a result can be
regarded as useful and valuable, and help improve overall quality of reservoir-, well-, and production system
models which in turn helps improve recovery.
4. Discussion
Overall, the value of this implementation has far outweighed the cost. Payback time including staff costs for the
full duration of the project has been less than a month and the investment has been recuperated more than
tenfold by now. This investment provides a more cost effective alternative to drilling new wells and/or constructing
new platforms with a comparable net result.
However it must be said that this approach does not compare entirely to the value of a new well. The approach
works as long as there is enough room between demand and capacity to optimize. Trips and shutdowns take
away from this room. Also, you are not unlocking new reserves by this approach, rather accelerating recovery of
condensates. (This is different to the impact of the CO2 optimization where indeed ultimate recovery of the
network is impacted due to maintaining the possibility for blending sweet- with sour gas for longer into the future
without the need to add new or additional sweet gas reserves.)
Furthermore the approach is not necessarily a way of working that is firmly embedded in the field managementand operating processes of todays oil and gas operating units. For instance for offshore equipment we tend to
have maintenance plans and procedures. For monitoring and optimization models this is less common and
typically they are owned and maintained by mega-users in the organization as and when such interested and
qualified individuals are available.
Put differently, an operating unit will have a mature organizational framework for traditional disciplines including
provisions for development and progression of staff, but for integrated, digital system optimization, such
framework is still in development. Implementations like this demonstrate the value that can be achieved, and as
such help the organization to firmly embed such a framework. Without this organizational framework, it is difficult
to add this approach as a sustainable and traditional element of oil and gas management, and an optimization
solution such as this one could remain an odd, out-lying, and non-standard item.
In the long term, however, integrated system and optimization solutions will move from interesting ideas to
basic necessities. This shift happened already more than a decade ago in downstream environments where
asset-wide optimization today is commonplace and a prerequisite to being able to profitably manage the very
tight-margin assets. As reservoirs become deeper, more complex, more fragmented, and recovery requires more
money and closer monitoring and management for lower margin returns, we find that field management decisions
inherently have more variables and trade-offs than used to be the case. On top of that, field equipment that is
fabricated and deployed today, produces a multitude of the amount of data that what was considered normal a
few years ago, and more advanced systems are required to keep making sense of all this information.
In light of this, solutions that support decisions by making their impact transparent in real time will become
commonplace and organizations will adapt to support them.
Looking forward, the next step from this optimization capability, could be to equip wells with remotely operated
chokes. This would create the possibility for closed-loop control of the entire production system: software could
translate the optimum system production levels into well choke movements automatically. This would allow
continuous, life cycle optimization rendering incremental gains as depicted by the blue area of Figure 11 below.
At the moment, a similar optimization approach is applied in Shell operating units across the globe in the following
areas:
Gas Lift Optimization: better use of available lift gas, reduced compression requirements and compressor
maintenance [1];
Optimizing the use of energy and fuel gas: applying RTO to beam pumps, ESPs and gas lift results in this;
Use of chemicals: PU continuously calculates required ppm of chemicals injected based on well flow rate
estimates and advises operator accordingly;
Reducing FTEs: applying of Real-Time Optimization has lead to the reduction of a team involved in
optimization and allocation [3];

IPTC 17009

17

Figure 11 Difference between continuous and manual optimization [1]

[

Reducing logistics exposure:: after introduction of PU and well intervention by exception, a Shell operator
successfully eliminated one boat, saving some US$1mln per annum;
Emissions Reduction [2]:
]: PU estimates provide a real
real-time
time handle on flaring and fuel gas consumption levels.
If correlated to plant output, the cost of emissions can be weighed a
against
gainst the output of the plant and
optimized.
Future plans include true Asset-Wide
Wide Optimization, where the optimizer will be expanded to also include
managing compressor suction levels, energy consumption, and other OPEX, while optimizing product quality and
throughput. A combination of the above applications is required to successfully and optimally manage an EOR
asset [8].
5. Acknowledgements
The authors wish to
o thank Sarawak Shell Bhd. for providing a conducive environment for this opportunity to be
realized and shared, and to Shell Global Solutions (Malaysia) Sdn. Bhd. for providing the talent and knowledge
required to get this solution to where it is today.
6. Nomenclature
CAPEX
Capital Expenditure
CGR
Condensate-to-Gas
Gas Ratio
CO2
Carbon Dioxide
DOF
Digital Oil Field
EOR
Enhanced Oil Recovery
ESP
Electric Submersible Pump
FTE
Full-Time Equivalent
GHV
Gross Heating Value
GUI
Graphical User Interface
H2S
Hydrogen Sulphide
LNG
Liquefied Natural Gas
MMscf/d
Million Standard Cubic Feet per Day
MS
Microsoft
MV
Manipulated Variable
OPEX
Operating Expenditure
PI
OsiSoft Plant Information the real-time data historian
ppm
Parts per Million

18

IPTC 17009

PSC
PSO
PU
PU (RTM)
PU (RTO)
PV
RX, RY, RZ
ST, MT, LT
THP
TL
WGF
WGR

Production Sharing Contract

Production System Optimisation
FieldWare Production Universe
Production Universe (Real-Time Monitoring module)
Production Universe (Real-Time Optimization module)
Process Variable
Riser X, Riser Y, Riser Z
Short-Term, Medium-Term, Long-Term
Tubing Head Pressure
Trunk Line
Wet-Gas Flow
Water-to-Gas Ratio

7. References
1. Cramer, R.; Goh, K.C.; Dolan, M.; Moncur, C. (2009) Data driven surveillance and optimization, paper SPE
122554, Digital Energy, Houston, 2009
2.

Cramer, R.; Goh, K.C.; Iyer, M.; Wali, N.; Spence, B.; Kessler, G.; Kanten, R. (2011) Continuous GHG
Emission, Surveillance and Control, paper OTC 21166, OTC, Houston, May 2011

3.

Cramer, R.; Hofsteenge, H.; Moroney, T.; Gbel, D.H.; Akpoghiran, Murthy (2011) Remote Operations a
remote possibility, or the way we do things round here, paper SPE 145293, ATCE, Denver, Oct. 2011

4.

Gerrard, C.; Taylor, I.C.; Goh, K.C.; De Boer, F. (2007) Implementing Real-time Production Optimisation in
Shell Exploration & Production in Europe - Changing the way we work and run our business, paper SPE
108515, Offshore Europe, Aberdeen, Scotland, U.K, 4-7 September 2007

5.

Goh, K.C.; Moncur, C.E.; Van Overschee, P.; Briers, J. (2007) "Production Surveillance and Optimization
with Data Driven Models, paper IPTC 11647, International Petroleum Technology Conference Dubai,
U.A.E., December 2007

6.

Ibrahim, K.; Narayasamy, D.; Jaberi, M.Y.; Briers, J.; Goh, K.C.; De Boer, F. (2010) Asset-wide reconciled
production monitoring - a key enabler to successful real-time field management, paper SPE 128654,
Intelligent Energy Conference and Exhibition, Utrecht, The Netherlands, 2325 March 2010

7.

Rao, S. (2009) Engineering Optimization - Theory and Practice, Chapter 10.3.3 Pg. 599 - Gomory's Method
th
for Mixed-Integer Programming Problems, 4 Edition, 2009, John Wiley & Sons

8.

Regtien, J.M.M. (2010) Extending The Smart Fields Concept To Enhanced Oil Recovery, paper SPE
136034, SPE Russion Oil & Gas Technical Conference and Exhibition, Moscow, Russia, 26-28 October 2010

9.

Selot, A.; Loi, K.K.; Robinson, M.; Mason, T.L.; Barton, P.I. (2007) A Model for Short-Term Supply Chain
Management of a LNG Production System, paper IPTC 11795, IPTC Conference Dubai, U.A.E., 4-6
December 2007

10. Stadtler, H. ; Kilger, C. (2008) Supply Chain Management and Advanced Planning - Concepts, Models,
Software, and Case Studies, Chapter 29.2 Pg. 517 - Pure Integer and Mixed Integer Programming, 2008,
Springer-Verlag
11. Van den Berg, F.; Goh, K.C.; Van Donkelaar, E.; Parchewsky, R. (2010) Remote Monitoring and Production
Optimization in Shell, paper SPE 136384, SPE Russion Oil & Gas Technical Conference and Exhibition,
Moscow, Russia, 26-28 October 2010
12. Wee, T.; Turco, A.; McIIroy, R.; De Boer, F.; Cramer, R.; Dolan, M. (2008) Brunei Shell Petroleum
Champion Field Gas-lift Optimization Project FieldWare Production Universe Implementation in a Brown
Field, paper SPE 120461, SPE Gulf Coast Section Digital Energy Conference and Exhibition, Houston,
Texas, 2021 May 2008
13. Yero, J.; Moroney, T. (2010) Exception Based Surveillance, paper SPE 127860, Intelligent Energy
Conference and Exhibition, Utrecht, The Netherlands, 2325 March 2010