You are on page 1of 11

Vol. 39, No. 1, JanuaryFebruary 2009, pp.

4656
issn0092-2102 eissn1526-551X 09 3901 0046
informs

doi 10.1287/inte.1080.0414
2009 INFORMS
THE FRANZ EDELMAN AWARD
Achievement in Operations Research
Optimizing the Norwegian Natural Gas
Production and Transport
Frode Rmo
Department of Operations Research and Applied Economics, SINTEF Technology and Society,
7465 Trondheim, Norway, frode.romo@sintef.no
Asgeir Tomasgard, Lars Hellemo, Marte Fodstad
Department of Industrial Economics and Technology Management (IT), Norwegian University of Science
and Technology (NTNU), 7491 Trondheim, Norway, and Department of Operations Research and Applied Economics,
SINTEF Technology and Society, 7465 Trondheim, Norway
{asgeir.tomasgard@sintef.no, lars.hellemo@sintef.no, marte.fodstad@sintef.no}
Bjrgulf Haukelidster Eidesen
StatoilHydro, 4035 Stavanger, Norway, bhauk@statoilhydro.com
Birger Pedersen
Gassco, 5501 Haugesund, Norway, bp@gassco.no
The network for transport of natural gas on the Norwegian Continental Shelf, with 7,800 km of subsea pipelines,
is the worlds largest offshore pipeline network. The gas owing through this network represents approx-
imately 15 percent of European consumption, and the system has a capacity of 120 billion standard cubic
meters (bcm) a year. In a network of interconnected pipelines, system effects are prevalent, and the network
must be analyzed as a whole to determine the optimal operation. SINTEF has developed a decision support
tool, GassOpt, which is based on a mixed-integer program, to optimize the network conguration and routing
for the main Norwegian shipper of natural gas, StatoilHydro, and the independent network operator, Gassco.
GassOpt allows users to graphically model their network and run optimizations to nd the best solutions
quickly. StatoilHydro and Gassco use it to evaluate the current network and possible network extensions. Both
companies use operations research (OR) methods in the departments that are responsible for transport planning
and security of supply. Several new OR projects have grown out from this cooperation. StatoilHydro estimates
that its accumulated savings related to the use of GassOpt were approximately US$2 billion in the period
19952008.
Key words: energy; natural gas; network transportation; mixed-integer programming; decision support system.
N
atural gas is the fastest-growing energy source
in OECD Europe (Energy Information Admin-
istration 2007). Norwegian gas covered 3.0 percent
of the worldwide production in 2007 (BP 2008), and
the Norwegian export is expected to increase by
4050 percent within the next decade. This paper
describes the development and use of a decision sup-
port tool for transport of natural gas. It gives a brief
presentation of StatoilHydro and Gassco and proceeds
by describing major decision problems, modeling
challenges, and the tools impact.
The Norwegian Natural Gas
Supply Chain
StatoilHydro was established in 2007 with the merger
of the mainly state-owned company Statoil and the
petroleum business of Norsk Hydro (the Norwegian
State owns 62.5 percent of StatoilHydro). Prior to the
46
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
Interfaces 39(1), pp. 4656, 2009 INFORMS 47
merger, both companies had been among the leading
operators on the Norwegian Continental Shelf (NCS)
from the early 1970s. StatoilHydro is the worlds third-
largest net seller of crude oil and one of the worlds
largest gas suppliers, as its website (http://www.
statoilhydro.com) describes. StatoilHydro is the main
operator and shipper on the NCS and also operates the
oil and gas resources that the Norwegian government
owns in full. It markets two-thirds of all Norwegian
gas sold in Europe, controls a portfolio of production
elds upstream in the transportation network, and
determines how much gas will be produced within its
share of each eld. In its role as shipper, StatoilHy-
dro buys transportation network capacity that allows
it to ship gas from the elds to the markets in which it
has delivery obligations or expects to nd good sales
opportunities.
In 2002, Gassco started operating as an indepen-
dent system operator for Gassled, the gas transport
network on the NCS. Figure 1 gives an overview of
the NCS pipeline network.
Gassco is entirely owned by the Norwegian state.
As an independent operator of the network, Gassco
gives access to the gas transport network on objective,
nondiscriminatory, and transparent terms. Gassco
controls the capacity available in the network and
administers to ensure efcient use of the resources
on the NCS. In addition, Gasscos role in infrastruc-
ture growth includes the evaluation and develop-
ment of alternative projects for network expansion.
As a former developer of infrastructure on the NCS,
StatoilHydro has technical resources and experience
and serves as a technical service provider to Gassco
on many NCS infrastructure projects. Gassco coor-
dinates these projects, which involve many stake-
holders, to ensure that the transport network of the
future will be integrated and well run, as its website
(http://www.gassco.no) describes.
The natural gas markets in this supply chain are
located mainly on the European continent and in
the UK, downstream of the Norwegian transport net-
work. Historically, most natural gas has been sold
on long-term contracts. These contracts typically have
a structure by which buyers and sellers agree on
yearly volumes, and the buyers make daily nomina-
tions within certain limits. Customers must pay for
the agreed volumes whether they take them or not
(take-or-pay contracts). In addition to volume obli-
gations, natural gas sales are typically restricted by
contractual limitations on pressure, energy content in
the gas, and the proportion of contaminants, such as
CO
2
and H
2
S, in the gas.
Driven mainly by the two European Union (EU) gas
directives (EU Commission 1998, 2003), the gas sales
from the NCS have moved from coordinated sales
to company-based sales. Additional effects of liberal-
ization are an increasing number of markets offering
short-term contracts and nancial derivatives, as well
as increased liquidity in existing markets. This creates
new business opportunities for both buyers and sell-
ers of natural gas (Midthun 2007).
To achieve nondiscriminatory third-party access to
the transport network, Gassco has established a trans-
port market. Shippers request booking of xed tar-
iff capacity and Gassco assigns the capacity based on
rules announced to all shippers. A secondary capac-
ity market in which shippers can trade the assigned
capacity also exists.
Background
In 1994, the Ministry of Petroleum and Energy in-
structed Statoil to substantiate the security of gas sup-
ply from the NCS. For such analyses, Statoil used a
customized system based on stochastic event simula-
tion (Rmo 1995). Because lengthy calculation times
complicated the use of this simulation system, Statoil
asked SINTEF to evaluate it.
SINTEF found the heuristic ow calculations in the
simulations to be unnecessarily time consuming and
made recommendations to speed them up. We esti-
mated the theoretical potential for reduced calcula-
tion time to a factor between 4 and 420. Initial tests
of alternative exact max-ow algorithms supported
our expectation. The evaluation also revealed some
inconsistencies in the results of the existing heuris-
tic. SINTEF built a prototype for a new network
ow-allocation implementation. Although we based
the prototype on Dinics max-ow algorithm (Dinic
1970), we made some modications to handle require-
ments related to physical and contractual limitations
(Rmo and Helgesen 1997). Our prototype reduced
calculation time by 90 percent. A graphical presen-
tation of ow allocations increased the transparency
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
48 Interfaces 39(1), pp. 4656, 2009 INFORMS
Snorre
Kvitebjrn
Huldra
Tune
Oseberg
Brage
Troll
Frigg
Fry
Heimdal
Alvheim
Grane
Sleipner
Draupner S/E
Ula
Gyda
Ekofisk
Valhall
Hod
N
O
R
P
I
P
E
S
T
A
T
P
IP
E
S
T
A
T
P
I
P
E

S
G
A
R
D
T
R
A
N
S
P
O
R
T
A
T
I
O
N
H
A
L
T
E
N
P
I
P
E
Z
E
E
P
I
P
E

ll
B
Z
E
E
P
I
P
E

ll
A
F
R
A
N
P
I
P
E
Z
E
E
P
I
P
E

l
Veslefrikk
Visund Gja
Kristin
sgard
Heidrun
Norne
Skarv
Krst
Kollsnes
Tjeldbergodden
Teesside
Dunkerque
Zeebrugge
SHETLAND
Stavanger
Bergen
Flor
Trondheim
Ormen Lange
Nyhamna
L
A
N
G
E
L
E
D
Easington
Emden
Dornum
Bacton
I
N
T
E
R
-
C
O
N
N
E
C
T
O
R
F
L
A
G
S
Draugen
Njord
Murchison
Statfjord
Gullfaks
TAMPEN LINK
V
E
S
T
E
R
L
E
D
St. Fergus
Armada
Rev
C
A
T
S
S
A
G
E
Beryl
50
52
52
54
54
56
56
58
58
60
60
62
62
64
66
12 10 8 6 4 2
0
0
2
2
4
4
6
6
8
8
10 12
GREAT
BRITAIN
FRANCE
BELGIUM
THE
NETHERLANDS
GERMANY
DENMARK
SWEDEN
NORWAY
THE ORKNEYS
FAROE
ISLANDS
Existing gas pipeline
Projected gas pipeline
Other pipelines
E
U
R
O
P
I
P
E
I
l
E
U
R
O
P
I
P
E

l
Figure 1: The transport network on the Norwegian Continental Shelf constitutes 7,800 km of pipelines
(The Norwegian Ministry of Petroleum and Energy and the Norwegian Petroleum Directorate 2008). Reprinted
with permission.
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
Interfaces 39(1), pp. 4656, 2009 INFORMS 49
Figure 2: An evaluation project, which the Ministry of Petroleum and
Energy requested in 1994, gave rise to GassOpt, an OR tool that is now
used frequently.
of the algorithm. Figure 2 gives a brief overview of
the major steps in the development of what became
the OR tool GassOpt.
Uses of GassOpt
GassOpt is based on a single-period steady-state
model; both StatoilHydro and Gassco use it for short-,
medium-, and long-term planning.
Decision variables are total ow and component
ow between nodes connected with pipelines and
inlet and outlet pressure for each pipeline. The objec-
tive function is to maximize ow to the markets
with some penalty terms for costly pressure increases,
etc. The constraints include eld capacities, market
demands, mass balance, pressure and ow constraints
in pipelines, and constraints to keep quality in all legs
downstream of a split node equal.
StatoilHydro as a commercial enterprise has other
interests and priorities than the independent system
operator Gassco. However, they both require analy-
ses of the network and its capacity and work to avoid
bottlenecks. Whereas StatoilHydro is most concerned
about the economic result from its available resources,
Gasscos main concern is to provide and maintain
sufcient network capacity at high regularity and to
assure network access on equal terms. The two com-
panies have access to different information, which
makes the scope of their analyses somewhat different
as well.
There are some common factors that are driving
forces in all types of analyses that the tool supports.
For example, changes in pressure in one part of the
network inuence the pressures and ows in other
parts of the network. In addition, the gas composi-
tion from different elds varies considerably; there-
fore, blending and processing of gas are signicant
processes.
Managing the Gas Flow
Both companies must consider common variations in
production and demand to determine the most ef-
cient use of network and production capacities. All
gas from the various elds does not conform to the
contracted quality. For example, a gas ow might be
processed or blended with other ows to adjust the
gas composition. Thus, taking a system perspective
makes it possible to exploit the total portfolio of gas
resources efciently. Gassco uses GassOpt to support
the work connected to coordinating the yearly main-
tenance plan and nominating the operational limits
for gas quality in the delivery points during the next
year. The ease of using GassOpt allows Gassco to ana-
lyze a number of scenarios to make a best possible
recommendation.
StatoilHydro manages its portfolio of production
and transport capacities to maximize its value. This
involves balancing the capacities with the contract
obligations and optimizing utilization of available
capacity. StatoilHydro uses GassOpt to evaluate both
existing contracts and new contract opportunities.
Capacity Reallocation for Production
and Transportation
Gassco uses GassOpt to assess the robustness of the
network. The network is sufciently exible to allow
multiple modes of operation; GassOpt utilizes this
exibility to compensate for production and transport
capacity reductions. For temporary capacity drops,
typical measures used are rerouting and deviations
from production plans on elds with exible produc-
tion rates, as Figure 3 illustrates.
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
50 Interfaces 39(1), pp. 4656, 2009 INFORMS
Figure 3: A ow allocation by GassOpt in three cases: The rst screen shot shows the network during normal
operations. GassOpt reallocates ow because of reduced capacity at Sleipner East (at the far left), resulting in
decreased delivery to the UK, France, and Germany in the second screen shot. Increased production at Troll and
Oseberg compensates for most of the shortage in the last screen shot.
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
Interfaces 39(1), pp. 4656, 2009 INFORMS 51
For StatoilHydro, capacity drops of longer duration
require extraordinary measures, such as buying extra
capacity downstream of the market. GassOpt can also
be used to select the appropriate measures to mini-
mize shortfall in the security of supply analyses.
Transport Booking
The shippers must book transport capacity in the net-
work from Gassco to ship the gas to the market. The
cost of transport is computed based on a xed tariff
and on the booked capacity from the point at which
the gas enters the pipeline system to the point of
exit. StatoilHydro must make booking decisions that
match the production and delivery plans to avoid
unnecessary transport costs and use GassOpt to verify
the feasibility of its plans.
Long-Term Planning and Investment
Modifying the transport infrastructure, which is
located down as far as 1,100 meters below sea level,
is demanding and costly. These investments need
solid justication. Decisions on where to build new
pipelines, compressors, or platforms, in what se-
quence, and when to phase out elds near depletion,
inuence the ability to exploit the network capaci-
ties. Because of the differing gas compositions in the
various elds, network modications have substan-
tial impact on Gasscos ability to blend gas and stay
within quality specications.
StatoilHydro uses GassOpt to evaluate operational
effects of its long-term plans and different invest-
ment proposals. Gassco performs similar analyses in
its annual work on its transport plan, the master
plan for transport prognoses for the next 10 years. The
work on this plan typically leads to proposals for new
investments. These investments are related to remov-
ing bottlenecks, increasing network transport capac-
ity, handling regularity issues, and avoiding off-spec
deliveries or underdeliveries.
Developing GassOpt
Based on the promising results from the prototype,
the users wanted more features in the model. In par-
ticular, they required the new tool to include pressure
decisions and the relationship between pressure and
ow capacity.
Gas Pressure
The capacity of a pipeline depends on its physical
properties, such as length, dimension, and friction.
Flow (Q) through a pipeline is a function of the
inlet (p
in
) and outlet (p
out
) pressure and can be
described by the nonlinear Weymouth equation
(Equation (1)). K is a constant computed from the
physical properties of the pipeline.
Q=K

p
2
in
p
2
out
(1)
The new challenge was how we could include
the nonlinear relationship without destroying the
good calculation-time results. Possible alternatives
we considered included using a more general linear
programming model with an approximation of the
nonlinearity, or a mixed-integer model with a dis-
cretization, as Beale (1983) describes. Otherwise, we
would have to use nonlinear programming methods.
We found that we could linearize the Weymouth
equation by using a Taylor series expansion around a
pair of input and output pressure values. This gives
an outer approximation of Equation (1), which turned
out to be sufcient for our tool, given that we are
maximizing ow near the capacity limit. For addi-
tional details, see Baumann et al. (2005), Dahl et al.
(2003), and Tomasgard et al. (2007).
To be prepared to include additional features that
had emerged during the prototype development, we
chose a framework with a general LP and MIP solver
(XpressMP) as the platform when we began develop-
ing the rst version of GassOpt in 1998.
Gas Quality
When the users recognized that GassOpt recommen-
dations were well founded, they began to see addi-
tional uses; therefore, they proposed extensions. The
rst version of the model included a quality de-
scription measuring the energy content of the gas
gross coloric value (GCV)using a bilinear term
(Tomasgard et al. 2007) approximated in a standard
way by reformulating to separable quadratic terms
and linearizing these (Williams 1999). However, using
this approach, it was not possible to determine other
quality properties of the gas, such as the level of
contaminants.
In a network in which gas from several elds enters
the same pipelines, there are blending possibilities.
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
52 Interfaces 39(1), pp. 4656, 2009 INFORMS
Blending sour gas from elds with low-quality gas
with sweet gas from elds with high-quality gas can
reduce the need for processing. This blending fol-
lowed by a split also introduces bilinear terms in the
model because we must ensure that the same fraction
of each gas component enters each pipeline going out
from the split point. This is a classic pooling problem
(Tawarmalani and Sahinidis 2002). Including bilinear
terms means that we must solve the problem by using
either nonlinear optimization techniques or approxi-
mations. We chose to implement an approximation in
a mixed-integer, linear programming context.
We introduced a multicommodity ow model in
which the different hydrocarbons and contaminants
in natural gas constitute the commodities. We ensured
the same gas component fraction in each direction
out from a split node by discretization of split ratios
implemented as special ordered sets. This rst formu-
lation, which can be found in Rmo et al. (2008) and
Tomasgard et al. (2007), was based on the work by
Ulstein (2000) and Ulstein et al. (2007).
The new method of modeling gas quality was not
used as frequently as we had expected, principally
because of lengthy calculation times, which were
caused by the discretization of the pooling problem.
The users found waiting minutes to hours for the
solution to be unacceptable, and they disabled the
function frequently. In other situations, the precision
of the discrete approximation was seen as unsatisfac-
tory. To respond, the team of users and researchers
joined together and found a way to modify the calcu-
lation scheme using a three-step algorithm:
(1) Solve the single-commodity problem.
(2) Solve the multicommodity problem with two to
three split fractions per split node, including the frac-
tions from the solution obtained in Step 1.
(3) If the same ow is obtained in Step 2 as in
Step 1, quit. Otherwise, solve the problem again using
more (evenly spread out) split fractions.
This approach reduced the calculation times for
multicomponent analysis signicantly for most cases.
A reformulation of the split fraction with binary vari-
ables enabled increased precision when needed. Prob-
lems are usually solved to optimality within seconds;
solutions found with a less than 0.15 percent optimal-
ity gap are usually found within 15 minutes, even for
hard test problems.
The importance of this functionality increases when
more and more of the new elds have a high con-
taminant content; thus, the quality limits have more
impact. Together with the improvements in the gas-
quality model in GassOpt, this explains the more fre-
quent use of the multicomponent functionality.
The models implementation also addresses many
other technical issues that we will not describe in
detail; for example, these include bidirectional pipe-
lines, xed pressure drops for installations that the
Weymouth formulation did not approximate well,
and compressor energy consumption. Rmo et al.
(2008) and Tomasgard et al. (2007) provide more de-
tailed descriptions.
System Effects
For single pipeline ows, the issues of pressure and
ow or gas composition are easy to address. How-
ever, when considering a larger system with a network
of interconnected pipelines, changing the pressure or
composition in one part of the network could inu-
ence capacities and ows in other parts of the network;
thus, taking a system perspective of the decision pro-
cesses is critical.
To illustrate such effects, we will consider two sim-
ple cases in a minimal network with only two pro-
duction nodes (A, B), one pool (C), four pipelines,
and two markets (D, E). One producer is in node A;
another is in node B. They sell their gas on contracts
in markets in nodes E and D, respectively. The rst
example, which is taken from Midthun et al. (2009),
is based on realistic data from the NCS (Figure 4). By
maximizing the ow through the network, we obtain
the solution that the left pane of Figure 4 shows.
This shows a ow from A to E of 36.99, and a ow
from B to D of 31.17; thus, the total ow is 68.16. Note
that there is blending of gas in node C (the pool).
Consider the case in which the producer in node A
would like to increase production to 41 to deliver
more to the market in E. Will this inuence the ow
from B to D? Imposing this as a new constraint and
maximizing the ow again gives the solution that the
right pane of Figure 4 illustrates. We can see that
although the ow from A to E increases to 41 as
intended, the ow from B to D declines to 22.68.
This occurs because of the increase in pressure in
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
Interfaces 39(1), pp. 4656, 2009 INFORMS 53
p p
p p p p
p
p
p p
f f f
f
f f
f
f
Figure 4: The pressure-ow relationships illustrate the system effects.
Increasing the ow from A to E reduces the capacity from B to C.
node C, which is necessary to deliver more from C
to E, whereas node B is unable to increase the pres-
sure further and will suffer from reduced transport to
node D (Midthun et al. 2009).
To illustrate the quality effects related to pooling,
consider the following example (Figure 5).
Again, we have two producing elds, A and B.
Field A produces sweet gas with 1 percent CO
2
,
whereas eld B is a sour eld with 10 percent CO
2
.
There is one pool, C, which is connected to both elds
by pipelines. The pool is connected to both market
nodes, D and E. Let the production capacity of both
elds be 10 and the demand in both markets be 10.
The maximum content of CO
2
entering the market
is 2.5 percent. Optimizing the ow modeling the nat-
ural gas as a homogeneous commodity gives the solu-
tion 10 in each market node. However, it does not
f = 10 f = 2
f = 6 f = 6
A
D
B
E
C
f = 10 f = 10
f = 10
f = 10
A
D
B
E
C
CO
2
= 1% CO
2
= 10% CO
2
= 1% CO
2
= 10%
CO
2
= 5.5% CO
2
= 5.5% CO
2
= 2.5% CO
2
= 2.5%
Figure 5: Quality requirements provide system effects. CO
2
limitations in
the markets reduce the production of sour gas from eld B.
comply with the gas-quality specication requiring
that the gas have a CO
2
content of below 2.5 percent.
Modeling the gas ow as a heterogeneous ow of sev-
eral gas components and constraining the content of
CO
2
that is allowed to enter the markets to 2.5 per-
cent, gives a total ow of only 12 because the sweet
eld, A, can only blend two units of production of
sour gas to the desired quality.
Impact
In this section, we point out some specic benets of
using GassOpt.
System Management and Portfolio Optimization
StatoilHydro, by optimizing the use of its production
possibilities, has been able to reduce losses during
temporary production shutdowns and improve gas-
portfolio balancing. If a shipper is unable to deliver
the contracted volume within the required quality
and pressure limits, that shipper might have to accept
a reduced payment, or even lose money on buy-
ing replacement gas. The losses in such situations
could be signicant. By using GassOpt, StatoilHydro
has been able to increase fulllments of contractual
obligations.
Most elds produce oil with associated gas, and a
companys inability to transport the gas from an oil
eld could lead to a cessation of production, resulting
in lost sales. In periods of high demand, it is impor-
tant to be able to compensate for unexpected events.
In periods of low demand and low prices, it is impor-
tant to minimize the gas production without affecting
the oil production. Analyses performed with GassOpt
have made it possible to avoid decisions that would
have reduced oil production and created new bottle-
necks and system dependencies.
By reallocating production and booked transport
capacity dynamically when it is necessary to deviate
from plans, StatoilHydro estimates that it saves tens
of millions US dollars (US$) each year. The accumu-
lated savings from 1995 to 2008 are estimated to be
US$200400 million.
Booking transport capacity in the network is expen-
sive. Reserving 50 million standard cubic feet per
day of transport capacity for one year costs approx-
imately US$1012 million. Analysis of the capacity
situation by applying GassOpt can show whether
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
54 Interfaces 39(1), pp. 4656, 2009 INFORMS
to wait before making a capacity reservation or to
reserve now. Thus, GassOpt analyses have led to more
correct booking decisions, thus avoiding booking too
much capacity at too early a stage. StatoilHydro esti-
mates that its savings have been several tens of mil-
lions US dollars since the project started.
Long-Term Planning
StatoilHydro and Gassco have access to other mod-
els that are designed to assist strategic decision
making regarding development of new elds, new
pipelines, and other infrastructure. However, these
portfolio analysis models use aggregated data and
cannot capture the same degree of details as GassOpt.
Specically, these models ignore pressure and system
effects. GassOpt is also an important tool in assessing
whether a given potential infrastructure improvement
will actually meet the required pressure, energy con-
tent, and contamination constraints.
Analyses performed using GassOpt have prevented
investment decisions that were shown to have a sig-
nicant negative effect on other promising eld devel-
opment options. Better insight into system effects
makes it possible to avoid new bottlenecks. Per-
forming detailed quality assessments enables users
to better understand consequences in advance. This
increased precision in analysis makes it possible to
address gas-quality issues and secure cost-optimal
integration of new elds into the network. By includ-
ing system effects in the analyses, synergy effects such
as the possibility of blending gas from two elds to
ensure the gas is within specication could also be
identied and taken into account.
For large-scale investments, making the wrong
decision might render a shipper unable to meet an
obligation. StatoilHydro estimates the potential oper-
ational-cost losses to be in the range of tens of mil-
lions US dollars each year until the issue is resolved.
These extra costs are mainly incurred from buying re-
placement gas when StatoilHydro is unable to satisfy
contracted volumes with equity gas. In addition, the
countermeasures could include increased investments
of several hundred million US dollars. StatoilHydro
has already avoided such bad investments; it has
conservatively estimated its resulting savings to be
US$1 billion.
GassOpt has been actively used for the develop-
ment of the dry-gas network on the NCS for the last
10 years, and will continue to be an important part of
infrastructure development in the future, particularly
for the long-term gas quality-management issues.
Improved Market Analysis
The improved understanding of the system has pro-
vided new insights about commercial decisions. Using
GassOpt, it became clear that some decisions that were
believed to make good business sense would neg-
atively impact the business. Although StatoilHydro
will not publish the details regarding these deci-
sions, it estimates the potential future revenues from
these discoveries to be up to several hundred million
US dollars.
Mutual Learning
The mutual learning that the teams of practitioners
and researchers experienced has been valuable to both
sides. One dimension is the increased insight into the
other parties expertise; the practitioners understand
the natural gas business and physics; the researchers
understand the methodology. Equally important is the
joint increased understanding of the problems and
decision processes acquired through our work with
GassOpt. In particular, one benet of this coopera-
tion has been the improved understanding of system
effects in the transport network. By gaining experience
with a tool that covers these aspects of the analysis,
both practitioners and researchers have built intuition
and increased our awareness and appreciation of for-
mal models. This learning takes place at the indi-
vidual level; however, from the start of the GassOpt
development, we have also seen this learning on the
organizational level. To quote a StatoilHydro manager
describing the importance of understanding system
effects: It is extremely valuable and maybe the most
important contribution of GassOpt.
Success Factors
In our experience, it has been invaluable to have peo-
ple within the client organizations who believe in the
project and in the value that operations research adds
to their analyses and operations. We also found it
important to create a team around the tool both in
our organization and in the user organizations. This
ensures continuity and makes the development less
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
Interfaces 39(1), pp. 4656, 2009 INFORMS 55
dependent on one single person. Through these teams,
the experience gained from this work also becomes
better dissipated into the organizations. In addition,
the possibilities of inspiring new and related work
based on what has already been achieved increase.
Involving more people also creates a better base for
generating new ideas for improvements, and their fre-
quent and informal cooperation throughout the devel-
opment process has been successful.
Spin-off Research Projects
The learning process for both researchers and practi-
tioners through our work with GassOpt has led to a
portfolio of related projects. There are several impor-
tant projects. Venoga is a tactical portfolio management
model based on stochastic programming and scenario
generation. The Research Council of Norway and
StatoilHydro funded this research that StatoilHydro is
now implementing. StatoilHydro funded SING Sup-
ply, a deterministic model for decision support for
planning its gas supply. The Research Council of
Norway, StatoilHydro, and Gassco funded Ramona, an
ongoing research project that addresses the security
of supply and value of exibility, based on stochastic
programming, scenario generation, and simulation.
Appendix. Model Formulation
This appendix provides a shortened version of the
model; we include only the parts that we addressed
in this paper.
Sets
. The set of all nodes in the network.
The set of nodes in the network with produc-
tion elds.
The set of nodes where gas ows are split into
two or more pipelines.
Market nodes, typically import terminals.
(n) The set of nodes with pipelines going into
node n.
(n) The set of nodes with pipelines going out of
node n.
The set of components dening the chemical
content of the natural gas.
The set of breakpoints used to linearize the
Weymouth equation.
. The set of fractions used to discretize possible
split fractions in split nodes.
Constants
D
m
Demand in market node m.
|
z
Percentage related to a split fraction z.

n
Flow capacity through a node.
G
g
Maximum production from eld g.
K
ij
Weymouth constant for the pipeline from i to .
P
min
ij
Minimum pressure for the pipeline from i to .
P
max
ij
Maximum pressure for the pipeline from i to .
P
m
Maximum delivery pressure to node m.
P
ij
Pressure pair inlet pressure used for Weymouth
linearization.
PO
ij
Pressure pair outlet pressure used for
Weymouth linearization.
Variables
]
c
ij
The ow of component c from node i to node .
]
ij
The total ow from node i to node .
p
in
ij
Inlet pressure on pipeline from i to .
p
out
ij
Outlet pressure on pipeline from i to .
Objective and Constraints
Objective function maximizing ow:
max

i(m)

m
]
im
. (2)
Terms representing soft constraints, such as costly
pressure increases, can be added.
Field (source) capacity:

(g)
]
g
G
g
, g . (3)
Market (sink) demand:

(m)
]
m
D
m
, m . (4)
Mass balance:

i()
]
c
i
=

n()
]
c
n
, ., c . (5)

c
]
c
ij
=]
ij
, i ., .. (6)
Linear approximation of pressure-ow relationship
(Weymouth):
]
ij
K
ij
P
ijl

P
2
ijl
PO
2
ijl
p
in
ij
K
ij
PO
ijl

P
2
ijl
PO
2
ijl
p
out
ij
i ., ., l . (7)
Rmo et al.: Optimizing the Norwegian Natural Gas Production and Transport
56 Interfaces 39(1), pp. 4656, 2009 INFORMS
Maximum delivery pressure:
p
out
im
P
m
, m , i (m). (8)
Technical pressure bounds on pipelines:
P
min
i
p
in
i
P
max
i
, ., i (). (9)
Multicomponent gas split: We assume each split
node having two downstream nodes,
1
and
2
. The
bilinear equation to be approximated is given in (10),
where c
1
is a gas component. A split fraction z in a
node is chosen with the variable 6
nz
. These variables
form one special ordered set for each node. The vari-
able c
c
nz
gives the componentwise ow through the
node restrained by the capacity of the node and the
corresponding active 6
nz
.
]
c
1
n
1
]
c
1
n
2
=
]
c
n
1
]
c
n
2
, n , c . (10)
]
c
n
1
=
Z

z=1
|
z
c
c
nz
, n . (11)
Z

z=1
6
nz
=1, z ., n . (12)

c
c
c
nz

n
6
nz
, z ., n . (13)
Z

z=1
c
c
nz
=]
c
n
1
+]
c
n
2
, n , c . (14)
Acknowledgments
We would like to thank our colleagues at the Norwegian
University of Science and Technology (NTNU) and SINTEF
for their cooperation during our work on GassOpt. In par-
ticular, we thank Thor Bjrkvoll, Bjrn Nygreen, Per Ivar
Helgesen, and Matthias Nowak for their contributions. We
also thank all the users at StatoilHydro, Gassco, and QPC
(a consultancy that uses GassOpt) for their valuable inputs.
The work on this paper was partially sponsored by the
Research Council of Norway in Project 176089 IDON.
References
Baumann, K. C., R. Albrechtsen, M. Carlsen, F. Rmo, L. B. Rvang.
2005. Transport analysis of the dry gas network system on
the Norwegian Continental Shelf using a network optimization
tool. Pipeline Simulation Interest Group PSIG 0503 116.
Beale, E. M. L. 1983. A mathematical programming model for the
long-term development of an off-shore gas eld. Discrete Appl.
Math. 5 19.
BP. 2008. Quantifying energy: BP statistical review of world energy
June 2008. Retrieved November 18, http://www.bp.com/
productlanding.do?categoryId=6929&contentId=7044622.
Dahl, H., J. F. Rmo, A. Tomasgard. 2003. An optimisation model
for rationing-efcient allocation of capacity in a natural gas
transportation network. IAEE Praha. 111.
Dinic, E. A. 1970. Algorithm for solution of a problem of maxi-
mum ow in networks with power estimation. Soviet Math.
Doklady 11 12771280.
Energy Information Administration (EIA). 2007. International
energy outlook 2007. Retrieved September 8, 2008, http://
www.eia.doe.gov/oiaf/ieo/index.html.
EU Commission. 1998. Directive 98/30/EC of the European
Parliament and of the Council of 22 June 1998 concerning com-
mon rules for the internal market in natural gas. Retrieved
September 8, 2008, http://eur-lex. europa.eu/LexUriServ/
LexUriServ.do?uri=OJ:L:1998:204:0001:0012:EN:PDF.
EU Commission. 2003. Directive 2003/55/EC of the European
Parliament and of the Council of 26 June 2003 concerning com-
mon rules for the internal market in natural gas and repealing
Directive 98/30/EC. Retrieved September 8, 2008, http://
eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2003:
176:0057:0078:EN:PDF.
Midthun, K. T. 2007. Optimization models for liberalized natu-
ral gas markets. Doctoral dissertation, Department of Indus-
trial Economics and Technology Management, Norwegian
University of Science and Technology, Trondheim, Norway.
Midthun, K. T., M. Bjrndal, A. Tomasgard. 2009. Modeling optimal
economic dispatch and system effects in natural gas networks.
Energy J. Forthcoming.
Rmo, F. 1995. Evaluation of ow algorithm-preliminary study
(In Norwegian). Technical Report STF83 F95004, SINTEF,
Trondheim, Norway.
Rmo, F., P. I. Helgesen. 1997. Analysing security of supply
A prototype of an alternative ow algorithm in MIRIAM
(In Norwegian). Technical Report STF38 A97603, SINTEF,
Trondheim, Norway.
Rmo, F., A. Tomasgard, L. Hellemo, M. Fodstad. 2008. A decision
support system for optimizing gas transport. Working paper,
Norwegian University of Science and Technology (NTNU) and
SINTEF, Trondheim, Norway.
Tawarmalani, M., N. V. Sahinidis. 2002. Convexication and Global
Optimization in Continuous and Mixed-Integer Nonlinear Program-
ming. Kluwer Academic Publishers, Norwell, MA.
The Norwegian Ministry of Petroleum and Energy and The
Norwegian Petroleum Directorate. 2008. The Norwegian
petroleum sector facts 2008 (In Norwegian). Retrieved
September 8, 2008, http://www.faktaheftet.no.
Tomasgard, A., F. Rmo, M. Fodstad, K. T. Midthun. 2007. Opti-
mization models for the natural gas value chain. G. Hasle,
K.-A. Lie, E. Quak, eds. Geometric Modelling, Numerical Simula-
tion and Optimization. Springer Verlag, New York.
Ulstein, N. L. 2000. Short term planning of gas production
(In Norwegian). Masters thesis, Department of Industrial Eco-
nomics and Technology Management, Norwegian University
of Science and Technology, Trondheim, Norway.
Ulstein, N. L., B. Nygreen, J. R. Sagli. 2007. Tactical planning of
offshore petroleum production. Eur. J. Oper. Res. 127(1) 550564.
Williams, H. P. 1999. Model Building in Mathematical Programming,
4th ed. John Wiley & Sons, Chichester, UK.

You might also like