Ind. Eng. Chem. Res.

2005, 44, 879-890 879

Optimization Model for Energy Planning with CO2 Emission

Considerations
Haslenda Hashim, Peter Douglas, Ali Elkamel,* and Eric Croiset
Department of Chemical Engineering, University of Waterloo, Waterloo, Ontario N2L 3G1, Canada

This paper considers the problem of reducing CO2 emissions from a power grid consisting of a
variety of power-generating plants: coal, natural gas, nuclear, hydroelectric, and alternative
energy. The problem is formulated as a mixed integer linear program (MILP) and implemented
in GAMS (General Algebraic Modeling System). Preprocessing and variable elimination strategies
are used to reduce the size of the model. The model is applied to an existing Ontario Power
Generation (OPG) fleet analyzed under three different operating modes: (1) economic mode, (2)
environmental mode, and (3) integrated mode. The integrated mode combines the objectives of
both the economic and environmental modes through the use of an external pollution index as
a conversion factor from pollution to cost. Two carbon dioxide mitigation options are considered
in this study: fuel balancing and fuel switching. In addition, four planning scenarios are
studied: (1) a base-load demand, (2) a 0.1% growth rate in demand, (3) a 0.5% growth rate in
demand, and (4) a 1.0% growth rate in demand. A sensitivity analysis study is carried out to
investigate the effect of parameter uncertainties such as uncertainties in natural gas price, coal
price, and retrofit costs on the optimal solution. The optimization results show that fuel balancing
can contribute to the reduction of the amount of CO2 emissions by up to 3%. Beyond 3%
reductions, more stringent measures that include fuel switching and plant retrofitting have to
be employed. The sensitivity analysis results indicate that fluctuations in gas price and retrofit
costs can lead to similar fuel-switching considerations. The optimal carbon dioxide mitigation
decisions are found, however, to be highly sensitive to coal price.

Introduction two possible options for reducing CO2 emissions by a

certain target while maintaining the electricity supplied
CO2 is the main greenhouse gas and is suspected to to the grid at nominal levels. The options being consid-
be the principal gas responsible for global warming and ered are fuel balancing and fuel switching. Fuel balanc-
climate change. Fossil-fuel power-generation plants are
ing is the optimal adjustment of the operation of
being challenged to comply with the Kyoto Protocol
generating stations to reduce CO2 emissions without
developed by the United Nations Framework Conven-
making structural changes to the fleet. In contrast, fuel
tion on Climate Change (UNFCC). For Canada, the
switching, which is switching from carbon-intensive
Kyoto Protocol prescribed a legally binding greenhouse
fuels (i.e., coal) to less carbon-intensive fuels (e.g., oil
gas emission reduction target of 6% below 1990 levels
or natural gas), involves structural changes to the fleet.
by 2008-2012.
Ontario Power Generation (OPG) produces 70% of Many authors have applied the tools of systems
Ontario’s electricity. Approximately 28.5% of OPG analysis to the economic evaluation of environmental
electricity is produced through the combustion of fossil problems.2-13 Kohn,7 for instance, proposed a simple
fuels, 27% through hydroelectricity, and 44% through linear programming model that can determine a suit-
nuclear energy, and the remaining 0.5% comes from able air-pollution-control strategy for a given situation.
renewable or other energy sources, such as wind tur- The model did not consider long-range trends or tech-
bines. In 2002, OPG had about 22211 MW total in- nological developments. Schweizer8 developed a method
service capacity, generated about 115.8 TW‚h of elec- for the optimal mix of high-and low-sulfur fuels for
tricity and emitted approximately 36.7 million tonnes power plants so that environmental criteria are met and
(Mt) of CO2, mainly from coal-fired power plants.1 plant operating schedules are preserved. Lou et al.9 used
There are several possible strategies to reduce the linear programming analysis for the optimal arrange-
amount of CO2 emitted from fossil-fuel power plants. ment of pollution-control equipment among various
Potential approaches include increasing plant efficiency, pollution sources. Jackson and Wohlers10 developed a
employing fuel balancing or fuel switching, making comprehensive model to determine the costs of control-
enhanced use of renewable energy (e.g., wind turbines, ling air pollution in a metropolitan area. Their model
solar, biomass, fuel cells), and employing CO2 capture was based on (1) identifying air pollutants and emission
and sequestration. All of these approaches are attrac- sources, (2) specifying all feasible control methods for
tive, but with the exception of fuel switching, they are each pollutant, and (3) predicting the interrelations that
unlikely to have a major effect on overall CO2 reductions can occur in the dynamic setting of the region. The
in the short to medium term. In this paper, we focus on model permits the determination of the percent use of
lowest-cost solutions. Kemner11 developed an integer
* To whom correspondence should be addressed. Tel.: (519) programming model to select a set of controls that can
888-4567 ext 7157. Fax: (519) 746-4979. E-mail: aelkamel@ meet a given emission restriction at the lowest cost. The
cape.uwaterloo.ca. model was applied to a coke plant and was solved by
10.1021/ie049766o CCC: $30.25 © 2005 American Chemical Society
Published on Web 01/12/2005
880 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

trial and error using plots of the control costs for various
desired reductions. Holnicki12 presented a similar model
for implementing a pollution-control strategy at a
regional scale. This approach is related to the optimal
allocation of funds for emission reduction in a given set
of power station installations. Elkamel et al.13 presented
comprehensive mixed integer linear programming mod-
els that can determine the optimum set of control or
retrofit options for a given pollution problem, along with
control setup times.
Of all the works on air pollution abatement, only a
few have dealt with CO2 mitigation in power-generation
expansion planning. Genchi et al.,14 for instance, de-
veloped a prototype model for designing regional energy
supply systems. Their model calculates a regional
energy demand and then recommends a most effective
Figure 1. Superstructure for an existing fleet of electricity-
combination of 11 different power supply systems to generating stations.
meet required CO2 emissions targets with minimum
cost. The new energy system to be installed included the variation in electricity demand, the electricity
cogeneration systems, a photovoltaic cell system, unused demand to be met was described by a load duration
energy in sewage and garbage incineration, and a solar- curve (LDC). The mathematical model was solved using
energy water supply. Linares et al.15 proposed a group Benders’ decomposition method (BDM). The model was
decision multiobjective programming model for electric- applied to the New England Generation Planning Task
ity planning in Spain based on goal programming (GP). Force.
The objective was to minimize the total cost of the In the present paper, a mathematical programming
electricity generation; the emissions of CO2, SO2, and model based on a superstructure of alternatives is
NOx; and the production of radioactive waste. The model presented and applied to an existing power-generation
is capable of estimating the capacity to be installed for system. Three modes of operation are considered: an
the year 2020 under four different social groups: regu- economic mode, an environmental mode, and an inte-
lators, academics, electric utilities, and environmental- grated mode. The economic mode is based solely on an
ists. The preferences of the groups were expressed as economic objective, the environmental mode uses the
weights in the model that affected the different main minimization of CO2 emissions as its objective, and the
criteria in the objective function. Mavrotas et al.16 integrated mode combines the two objectives in an
developed a mixed 0-1 multiple-objective linear pro- appropriate way as will be explained shortly.
gramming (MOLP) model and applied it to the Greek Four electricity generation scenarios are considered
electricity generation sector to identify the number and to investigate fuel-balancing options as well as struc-
output of each type of power unit needed to satisfy an tural changes for different growth alternatives: (1) a
expected electricity demand. The first objective was to base-load demand and (2) low-growth (0.1%), (3) medium-
minimize the annual electricity production cost, and the growth (0.5%), and (4) high-growth (1.0%) scenarios. A
second objective addressed the minimization of the total sensitivity analysis was also performed to evaluate the
amount of SO2 emissions. However, the model did not impact of natural gas prices, coal prices, and retrofit
consider CO2 mitigation and sequestration. Bai and costs on the optimal configuration of the OPG fleet of
Wei17 developed a linear programming model to evalu- electricity-generating stations.
ate the effectiveness of possible CO2 mitigation options
for the electricity sector in Taiwan. The strategies they
considered included fuel alternatives, reduced peak Methodological Approach
loads, energy conservation, power-generation efficiency (1) Superstructure Representation. The super-
improvements, and CO2 capture. They found that the structure representing a power-generation energy-sup-
combination of reduced peak production and increased ply system and fuel alternatives is presented here. The
power-plant efficiency with CO2 conservation was an following notation is adopted throughout the paper: C,
effective strategy to meet significant CO2 emissions NG, O, N, H, and A represent a set of coal, natural gas,
reduction targets. Climaco et al.18 developed new tech- oil, nuclear, hydroelectric, and alternative energy power
niques that incorporate multiple-objective linear pro- plants, respectively. Figure 1 presents a configuration
gramming and demand-side management (DSM). These of power stations. The mix of all electricity generated
techniques are able to determine the minimum expan- is injected into the electricity grid, and it is assumed at
sion cost by changing the levels and forms of electricity this stage that there is no CO2 capture at any existing
use by the consumers and generating alternatives from power-generating unit. Figure 2 illustrates the fuel-
the supply side. The model also considered the emissions balancing technique to decrease CO2 emissions by
caused by electricity production. Noonan et al.19 studied adjusting the operation of the fleet of existing electricity-
and developed an optimization program for planning generating stations, e.g., increasing the load on existing
investments in electricity-generating systems. The op- non-fossil-fuel power plants and decreasing the load on
timization program determined the mix of types of existing fossil-fuel power plants. In this case, it is
plants, sizes of the individual plants to be installed, and desired to determine the optimal load distribution for
allocation of installed capacity to minimize total dis- all power stations to maintain electricity to the grid
counted cost while meeting the system’s forecasted while reducing CO2 emissions to a certain target. Figure
demand for electricity. This problem is referred to as 3 illustrates another technique, the so-called fuel-
the generation planning problem (GPP). To comply with switching technique, that represents switching from
 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005 881

Figure 2. Superstructure for fuel balancing (increase output of non-fossil-fuel plants and decrease output from fossil-fuel plants).

mode includes the operating cost of electricity generated

for the fleet of generating stations and the retrofitting
cost associated with switching the fuel from coal to
natural gas for the fossil-fuel stations. Note that, in the
case of a fossil-fuel station already operating on gas (j
) 1), there is no retrofitting cost involved (i.e., Ri1 ) 0).
The objective function for this mode is given by

min Z1 ) ∑ ∑CijEij + i∈NF

i∈F j
∑ CiEi + ∑ ∑RijXij
i∈F j
(1)

where Cij is the electricity generation cost per MW‚h if

fuel j is used in fossil-fuel power plant i; Ci is the
electricity generation cost per MW‚h for non-fossil-fuel
power plant i; Ei is the electricity generated (MW‚h/
year) by non-fossil-fuel power plant i; Eij is the electricity
Figure 3. Superstructure for fuel switching (switching from more
generated (MW‚h/year) by fossil-fuel power plant i when
carbon-intensive fuels to less carbon-intensive fuels, e.g., switching
from coal to natural gas). using fuel j; Rij is the retrofitting cost to switch a coal
power station to natural gas power station, expressed
carbon-intensive fuels to less carbon-intensive fuels (e.g., in U.S. dollars per $/year; Xij is a binary variable that
switching from coal to natural gas). indicates which fuel to use or whether the plant should
(2) Model Formulation. The optimization problem be shut down at each fossil-fuel plant i; F is the set of
to be studied in this paper can be formulated as fossil-fuel power stations including coal and natural gas;
follows: For a given CO2 reduction target, what is the and NF is the set of non-fossil-fuel power stations,
best generating-plant load distribution and mix of fuels including nuclear, hydroelectric, and wind turbine
that will maintain the electricity supply to the grid? sources. Note that, in the case of non-fossil-fuel plants,
Three different points of view or modes for the optimi- no associated binary variables are defined, as the fuel
zation problem are considered: (1) economic mode, (2) type for these plants is known a priori.
environmental mode, and (3) integrated mode. The (b) Mode 2: Environmental Mode. In environmen-
corresponding optimization models are formulated be- tal mode, the objective is to minimize the CO2 emissions
low. Three types of variables are first defined: (1) while maintaining and enhancing power to the grid. The
continuous variables, Eij, representing electricity gener- optimization model can be written as
ated/load distribution from the ith fossil-fuel boilers; (2)
continuous variables, Ei, representing electricity gener- min Z2 ) ∑CO2ijEij (2)
ated/load distribution from the ith nuclear, hydroelec-
tric, or alternative energy power plant; and (3) binary where CO2ij represents the CO2 emissions for the ith
variables, Xij representing the type of fuel used in a fossil-fuel power station using fuel j and is calculated
given fossil-fuel plant using basic chemical equations that relate the genera-
tion of CO2 emissions to the quantity and quality of fuel

Xij ) {1 if fuel j is used in plant i

0 otherwise
burned

CO2ij ) 0.03667EFij (3)

We first state the objective functions for the three
modes of operation considered and discuss the model where 0.03667 is the conversion factor from coal to CO2
constraints in a separate section. and EFij is the CO2 emissions factor of the ith fossil-
(3) Objective Function. (a) Mode 1: Economic fuel station using fuel j
Mode. In this mode of CO2 mitigation, the objective is
1 %C
to satisfy a CO2 reduction target and maintain and
enhance power to the grid. The cost function in this
EFij ) (
ηij HHV ) (4)
882 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

where ηij is the efficiency of the coal-fired boiler in the the other hand, include binary decision variables and
ith fossil-fuel station operating on fuel j. %C represents are essential in the model implementation, especially
the percentage carbon content of the fuel, and HHV is in the case of plant shutdowns.
the fuel higher heating value. (e) Lower Bound on Operational Constraints.
(c) Mode 3: Integrated Mode. This mode combines The annual capacity factor for each power plant must
the previous two objective functions to give be greater than some minimum; otherwise, the plants
will be shut down
min Z3 ) ∑ ∑CijEij + i∈NF
i∈F j
∑ CiEi + ∑ ∑RijXij +
i∈F j fij g IijXij ∀i ∈ F
β∑∑(CO2ij)Eij (5)
i∈F j
fi g li ∀i ∈ NF (10)

The CO2 emissions in eq 5 are translated into U.S. where lij (li) is the minimum annual capacity factor for
dollars per $/year by introducing a CO2 emissions cost fossil-fuel (non-fossil-fuel) power station i and fij (fi) is
coefficient β. More details on this coefficient is given in the corresponding annual capacity factor. The relation-
the case study. The reason for considering this inte- ship between the annual capacity factor and electricity
grated mode is to find compromise decisions under both generation is given by
economic and environmental objectives. If only econom-
ics is considered, then emissions will be unacceptable. Eij ) fijEmax
ij ∀i ∈ F
Likewise, if only an environmental objective is consid-
ered, then plant economics will not be sustained. Ei ) fiEmax
i ∀i ∈ NF (11)
(4) Constraints. The minimization of the objective
functions represented by eqs 1, 2, and 5 is subjected to
where Emaxij (Emax
i ) is the installed capacity of the ith
the following constraints:
fossil-fuel (non-fossil-fuel) power plant.
(a) Energy Balance/Demand Satisfaction. The
(f) Emission Constraint. In this constraint, annual
total electricity generation must be equal to or greater
CO2 emissions must satisfy a CO2 reduction target
than the desired electricity demand

∑ Ei + ∑ ∑Eij g demand (6) ∑ ∑CO2ijEij e (1 - %CO2)CO2

i∈F j
(12)
i∈NF i∈F j

(b) Fuel Selection and Plant Shutdown. For each where CO2ij represents the CO2 emissions for the ith
fossil-fuel process i, either the process is operating with fossil-fuel power plant using the jth fuel per unit of
one chosen fuel, or it is shut down electricity generated (t of CO2/MW‚h). %CO2 is the
percentage reduction target. CO2 represents the current
∑j Xij e 1 ∀i ∈ F (7) CO2 emissions in millions of tonnes per year.
(g) Nonnegativity Constraints. The electricity
generated from all power plants must be greater than
(c) Capacity Constraints. zero; therefore, the amounts of electricity generated
from non-fossil-fuel and fossil-fuel power plants are
Eij e M ∑j Xij ∀i ∈ F (8a) defined as positive/nonnegative variables.

Ei g 0, Eij g 0 (13)
Ei e M ∀i ∈ NF (8b)
(h) Extra Constraints. Other constraints can be
The above constraint set places an upper bound on imposed on the model. These include (1) geographical/
energy produced from the different plants. It also management constraints, according to which certain
ensures that the energy production from fossil-fuel power plants are assigned to supply certain locations
plants (i ∈ F) is zero when no fuel is assigned to the while other plants cannot supply some specific locations,
plant and a decision to shut down the plant has been and (2) fuel or resource constraints, according to which
made. The parameter M is any large number and the fuel supply (i.e., natural gas) is limited by pipeline
represents an upper bound on energy production from capacity.
each plant i. M can be chosen to be the maximum
installed capacity of Emaxi for a non-fossil-fuel plant or Case Study for Ontario Power Generation
Emax for a fossil-fuel plant. (OPG)
ij
(d) Upper Bound on Operational Changes. The Currently, OPG operates 79 electricity-generating
electricity generated by unit i cannot exceed the current stations, 5 of which are coal-fired generating stations,
electricity generation for the unit by ri (the maximum Ci (i ) 1-5); 1 of which is a natural gas generating
increase in the base load due to operational constraints)- station, NGi (i ) 6); 3 of which are nuclear generating
Comparing constraints 8b and 9b, it is clear that both stations, Ni (i ) 7-9); 69 of which are hydroelectric
generating stations, Hi (i ) 10-78); and 1 of which is a
Eij e (1 + ri)Ecurrent
ij ∀i ∈ F (9a) small wind turbine, Ai (i ) 79). At nominal levels, OPG
generates 13765 MW of electricity that it injects into
Ei e (1 + ri)Ecurrent
i ∀i ∈ NF (9b) the grid from a mix of sources, i.e., coal, hydroelectric,
nuclear, and renewable energy. No CO2 capture process
represent an upper bound on Ei. Because constraints currently exists at any OPG power plant; about 36.7 Mt
9b are tighter, constraints 8b are redundant and do not of CO2 was emitted in 2002, mainly from fossil-fuel
have to be included in the model. Constraints 8a, on power plants. There are 27 fossil-fuel boilers at the 6
 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005 883

Figure 4. OPG’s current electricity generation by fossil fuels.

Table 1. Ontario Power-Generation Fossil-Fuel Generating Stations

installed number annual capacity operating CO2 emission
station fuel capacity (MW) of units factor cost ($/MW‚h) rate (t/MW‚h)
Nanticoke 1 (N1) coal 500 2 0.75 30 0.9300
Nanticoke2 (N2) coal 500 6 0.61 30 0.9300
Lambton1 (L1) coal 500 2 0.5 34 0.9386
Lambton2 (L2) coal 500 2 0.75 25 0.9384
Lakeview (LV) coal 142 8 0.25 35 0.9765
Lennox (LN) natural gas 535 4 0.15 50-70 0.6510
Thunder Bay (TB) coal 155 2 0.55 30 1.0230
Atitokan (A) coal 215 1 0.44 30 1.0230

fossil-fuel stations: 4 boilers at Lambton (L1-L4), 8 boilers are operating on natural gas. This figure will
boilers at Nanticoke (N1-N8), 8 boilers at Lakeview later be used in our analysis to check which optimal
(LV1-LV8), 1 boiler at Atikokan (A1), 4 boilers at modifications must be made under different modes of
Lennox (L1-L4), and 2 boilers at Thunder Bay operations.
(TB1-TB2). Currently, 4 boilers operated by Lennox are
running on natural gas. Results and Discussions
A summary of OPG’s current fossil-fuel generating
stations is contained in Table 1. The operational costs This section presents the results for the three
for nuclear, hydroelectric, and wind turbine were esti- modes of operation analyzed under four different sce-
mated to be $32, $5, and $4/MW‚h, respectively. Note narios: (1) base-load demand, (2) low-growth (0.1%
that, currently, natural gas is the most expensive fuel growth rate) demand, (3) medium-growth (0.5% growth
used by OPG.13 In this study, we assumed that all coal- rate) demand, and (4) high-growth (1.0% growth rate)
fired boilers operate at 35% efficiency and that the base- demand.
load demand is constant throughout the year at 13675 The implications relating to the amount of electricity
MW. Because the main objective of this paper is to study generated by each boiler and generating station, the
CO2 emissions reduction through fuel balancing and fuel total cost of electric power generation, and the CO2
switching, no attempt was made to study the effect of emissions are discussed. The models were implemented
improved technology. An improvement in boiler technol- in the GAMS (General Algebraic Modeling System)
ogy will, in principle, lead to an efficiency higher than optimization package and solved using the mixed inte-
our assumed efficiency of 35%. ger linear programming (MILP) solver.21 Given that the
The index i (i ) 1-79) represents all of OPG’s power generation of electricity can be increased by ri from the
plants, including those powered by fossil-fuel, nuclear, current load, the load distribution of the non-fossil-fuel
hydroelectric, and wind turbine sources. The index j (j power plants can be determined a priori. This is because
) 1 or 2) represents the fuel selection, j ) 1 (coal) or 2 they do not emit any CO2, and their operational cost is
(natural gas). The retrofitting cost was estimated to be less than that of the fossil-fuel plants. Therefore, all non-
$30 million/1000 MW with a 20-year lifetime and a 10% fossil-fuel power plants can be excluded from the
annual interest rate. To translate CO2 emissions into optimization process, and we need to solve only for the
cost, we used an external cost index of β ) $0.03 remaining variables that correspond to the electricity
US$/kg of CO2 emitted as developed in a recent United generation, Eij, from OPG’s 27 fossil-fuel units and the
Nations publication.20 The reserve margin, ri, for load binary variables that correspond to decisions on which
distribution for all of OPG’s fleet power plants was set fuel to use in each plant and whether any retrofitting
at 1% higher than the current level because of opera- action is required.
tional considerations. The lower bound was set to be Mode 1: Economic Mode. In this mode, the objec-
10% (i.e., a plant has to be operated with at least 10% tive is to minimize the total operating cost while
of its installed annual capacity factor; otherwise, it will meeting a specified CO2 reduction target. The optimiza-
be shut down). Figure 4 provides a summary of current tion results for the case of 3% CO2 reduction indicate
electricity generation from OPG’s fossil-fuel power sta- that no fuel switching is needed. In other words, this
tions by the 27 different boilers. Currently, only four objective can be obtained simply by adjusting the
884 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

Figure 5. Optimal electricity generation strategy for fossil-fuel boilers for a 6% CO2 reduction requirement.

Figure 6. Optimal electricity generation strategy by fossil-fuel boilers for a 20% CO2 reduction requirement.

operation of the current boilers, e.g., increasing the load CO2 reduction. In this case, nine coal boilers are
from existing non-fossil-fuel power plants and decreas- switching to natural gas (compared to only four for the
ing the load from existing fossil-fuel power plants (fuel case of a 6% CO2 reduction). Higher CO2 reduction
balancing). The results show that the total cost is also targets require more coal boilers to be switched to
reduced by 2.4% by reducing electricity generation from natural gas.
all four natural gas boilers by 32.1% and from the two The effect of increasing demand on the amounts of
coal-fired boilers (LV4 and LV8) by 33.4% and 59.4%, electricity generated by the different boilers was also
respectively. The electricity generation from other fossil- analyzed. Three different scenarios were studied, in-
fuel boilers and non-fossil-fuel power plants was in- volving low (0.1%), medium (0.5%), and high (1.0%)
creased by 1% higher than the nominal operational level growth rates in electricity demand. The analysis shows
to maintain the electricity to the grid. that fuel switching is required to satisfy CO2 emissions
To achieve more than 3% CO2 reduction, it was found reductions as electricity demand is increased. For a 0.1%
that fuel switching must be implemented. This involves growth rate in demand, a 3% CO2 reduction can be
fleet changes from coal to natural gas. The optimization achieved by fuel balancing by increasing electricity
results show, for instance, that to achieve 6% CO2 generation for all coal-fired boilers by 1% and decreasing
reduction (Canada’s Kyoto target at 1990) while main- electricity generation for all four natural gas boilers
taining the electricity to the grid at minimum cost, the without having to change the fleet structure. However,
electricity generation from three natural gas boilers for a 0.5% growth rate in demand and a 3% CO2
(LN1, LN2, and LN3) needs to be reduced by 32.1%; that reduction, one coal-fired boiler needs to be switched to
from one natural gas boiler (LN4) by 8.2%; and that natural gas. As the growth rate in demand increases
from two coal-fired boilers (LV1 and LV2) by 59.4% and further to 1%, three coal-fired boilers need to be
34.8%, respectively, and the electricity generation from switched to natural gas, as shown in Figure 7.
the other coal-fired boilers and non-fossil-fuel power Figures 8 and 9 illustrate the effect of growth rate in
plants needs to be increased by 1% over the nominal demand on CO2 emissions and total cost for a 3% CO2
operational level. Finally, four coal-fired boilers (L1, A1, reduction target. As the growth rate in demand in-
TB1, TB2) need to be switched to natural gas, resulting creases from 0 to 1%, the optimizer adjusts coal-fired
in a cost increase of about 1.2% (Figure 5). boilers to produce more electricity up to their maximum
It is estimated that Canada’s emissions will rise to level. CO2 emissions increase until they hit the CO2
approximately 750 Mt by 2005 from 571 Mt in 1990. emissions target (e.g., 3% CO2 reduction). Then, the
Therefore, the actual reduction target is to reduce optimizer switches coal-fired boilers to natural gas while
emissions to 179 Mt by 2008-2012, and this represents maintaining the electricity output at the current level.
a reduction of more than 20%.22 Figure 6 illustrates the This results in an instant drop in CO2 emissions because
breakdown of electricity production by boilers for a 20% natural gas emits less CO2; however, the total cost
 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005 885

Figure 7. Effect of a 1% increase in electricity demand on fleet structure for a 3% CO2 reduction target.

Figures 10 and 11 give optimal electricity generation

by the different plants to meet the base demand and
an increased demand of 1%, respectively. The figures
also illustrate the fuel-switching scenarios for the vari-
ous fossil-fuel plants.
Mode 3: Integrated Mode. The solution generated
from the different modes (economic and environmental)
produced different results. For example, minimizing the
total cost resulted in large CO2 emissions, and minimiz-
ing CO2 resulted in large costs. These are conflicting
strategies between minimizing cost and CO2 emissions,
as shown in Tables 2 and 3. This is not surprising
because there is a cost associated with CO2 emissions
reduction, and therefore, cost minimization will avoid
Figure 8. Effect of electricity demand growth rate on CO2 reducing emissions. A more realistic mode is therefore
emissions for a 3% CO2 reduction requirement. one that integrates both modes and seeks a compromise
between minimizing cost and minimizing CO2 emis-
sions. In this mode, as explained earlier, eq 5 is used
as the objective function. Figure 12 represents optimiza-
tion results for this mode of operation for a 1.0%
electricity demand growth rate and a 3% CO2 emissions
reduction. Comparing these results with the results of
Figure 7 (pure economic mode) and Figure 11 (pure
environmental mode), it is clear that a compromise is
achieved. The number of boilers switching from coal to
natural gas is substantially less than that of the
environmental mode. The same switching pattern is
observed, however, as in the economic mode, except that,
for the integrated mode, two gas boilers are required to
operate at higher loads. The optimal solutions to meet
3% and 6% reductions in CO2 emissions are reported
in Tables 2 and 3 (fourth row), respectively. The results
Figure 9. Effect of electricity demand growth rate on fleet
operating cost with and without a 3% CO2 reduction. for the base electricity demand case, the case of 1%
electricity demand increase, and other cases are re-
continues to rise because natural gas is more expensive ported in these tables.
than coal. The negative slope indicates that natural gas
boilers generate more electricity up to a maximum level Sensitivity Analysis
while coal boilers operate at their maximum level to
meet the demand. This trend is repeated to meet the In this section, the impact of the natural gas and coal
demand and CO2 emissions target. prices and retrofit costs on the configuration of power
Mode 2: Environmental Mode. In this mode, the plants for a 20% CO2 emissions reduction is discussed.
goal is to minimize CO2 emissions. The optimizer The sensitivity analysis was performed for four case
minimized the CO2 emissions by switching all 23 coal- studies: 0% (base-load demand), 0.1%, 0.5%, and 1.0%
fired boilers to natural gas. For the base-load demand, growth rates in demand. Three major parameters were
a 39.5% CO2 reduction was achieved after optimization considered as follows: (1) natural gas price (10% and
because all coal-fired boilers were switched to natural 20% higher than the base-case natural gas price), (2)
gas, resulting in an increased cost of $467.5 million/year. coal price (10% and 20% higher than the base-case coal
As a consequence, minimization of CO2 emissions is price), and (3) retrofit cost (-10%, -20%, +10%, and +
considered to be very costly because an increase in 20% changes from the base-case retrofit cost).
growth rate in demand will require the use of more Although it is possible for energy prices (natural gas
natural gas to produce the required electricity target. and coal) to decrease, it was assumed that the most
886 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

Figure 10. Optimization results for electricity generation to meet a base-load demand under the objective of minimizing CO2 emissions.

Figure 11. Optimization results for electricity generation to meet a 1% increased demand under the objective of minimizing CO2 emissions.

Table 2. Conflict Results between Cost and CO2 Table 3. Conflict Results between Cost and CO2
Emissions for a 3% CO2 Reduction Emissions for 6% CO2 Reduction
total cost CO2 emissions total cost CO2 emissions
case (106$/year) (Mt/year) case (106$/year) (Mt/year)
Nominal Mode, Reference Nominal Mode, Reference
base 2837.0 37.33 base 2837.0 37.33
Economic Mode, Economic Mode,
Minimize Total Cost Minimize Total Cost
base 2771.0 36.05 base 2804.9 35.09
0.1% 2775.2 36.19 0.1% 2818.2 35.09
0.5% 2812.0 36.17 0.5% 2852.9 35.09
1.0% 2862.4 36.21 1.0% 2909.7 34.99
Environmental Mode, Environmental Mode,
Minimize CO2 Emissions Minimize CO2 Emissions
base 3304.5 22.11 base 3304.5 22.11
0.1% 3309.8 22.18 0.1% 3309.8 22.18
0.5% 3335.9 22.49 0.5% 3335.9 22.49
1.0% 3372.1 22.89 1.0% 3372.1 22.89
Integrated Mode, Integrated Mode,
Minimize (Total Cost + CO2 Emissions) Minimize (Total Cost + CO2 Emissions)
base 3852.1 36.06 base 3866.2 35.09
0.1% 3860.4 36.17 0.1% 3871.1 35.09
0.5% 3898.5 36.22 0.5% 3905.9 35.09
1.0% 3948.4 36.19 1.0% 3951.5 34.99

reasonable scenarios were increasing costs. On the other groups, corresponding to the six fossil-fuel generating
hand, in the case of retrofitting costs, we felt that this stations. The first column presenting a pattern by black
cost was perhaps uncertain and that it was worth and white dots illustrates OPG’s current operation
considering both positive and negative deviations in the without a CO2 emissions target; this is the base case.
retrofit cost estimate. According to Canada’s Energy Other columns showing the optimal pattern for 0%,
Outlook 1996-2020, the natural gas price is expected 0.1%, 0.5%, and 1.0% growth rates in demand with 20%
to increase from $2.25/GJ in 1997 to $3.25/GJ by 2020.23 CO2 emissions reduction for base natural gas price are
Figure 13 illustrates all 27 boilers divided into six also presented. The black dots represent coal-fired
 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005 887

Figure 12. Electricity generation strategy for a high (1.0%) growth rate in demand and a 3% CO2 reduction for the integrated mode of
operation.

Figure 13. Effect of increasing natural gas prices on fuel-switching decisions for different electricity demand growth rates.

boilers, whereas the open or white dots represent and 20% from the base natural gas price for the case of
natural gas boilers. As can be seen, increasing demand 0%, 0.1%, and 0.5% growth rates in demand. However,
from the base load to 0.1% leads to switching of more the location of the optimum is changed for the case of a
coal-fired boilers to natural gas to meet CO2 emissions 1.0% growth rate in demand. For higher demand, the
reduction target. In the case of a 0.5% growth rate in optimizer favored switching the larger coal-fired boiler
demand, a large coal-fired boiler (N3) is switched to (N4) rather than several small boilers; this strategy by
natural gas to meet demand increase and CO2 emissions the optimizer results in three small boilers switching
target. It is important to note that the smaller boilers back to coal (LV1, LV2, and TB1).
(LV1-LV3) that are switched to natural gas for the The variation of coal price has a significant impact
cases of 0% and 0.1% growth rates in demand switch on the optimal configuration of power plants, as il-
back to coal. This is because the switching of N3 results lustrated in Figure 14. Increasing the price of coal by
in a large reduction in CO2 emissions and a large 10% and 20% from the base coal price results in changed
increase in cost; by switching LV1, LV2, and LV3 back patterns as compared to the base-case scenario for all
to coal, a significant reduction in cost occurs, along with of the different demand rates. A similar pattern is
a modest increase in CO2 emissions. At higher demands observed for 10% and 20% increases in coal price for
(1.0% increase growth rate in demand), 10 coal-fired all scenarios except that with a 0% growth rate in
boilers are switched to natural gas compared to eight demand.
boilers for the previous case. The pattern remains The location of the optimal configuration was less
unchanged if the natural gas price is increased by 10% sensitive to the retrofit cost in the range between -10%
888 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

Figure 14. Effect of increasing coal prices on fuel-switching decisions for different electricity demand growth rates.

Figure 15. Effect of retrofit cost on fuel-switching decisions for different electricity demand growth rates.

and +20% from the base retrofit cost, as illustrated in The results of the sensitivity analysis show that the
Figure 15. As a result, a similar pattern is produced at variation of fuel prices directly affects the total cost. The
the same growth rate in demand for different retrofit optimal configuration or fuel-switching pattern of the
costs. This agrees with the pattern produced by varying fleet also changes when the retrofit cost is varied. This
the natural gas price for all four cases. is not surprising because, if the fuel price and retrofit
 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005 889

costs are increased, this will definitely lead to an Acknowledgment

increase in total cost.
The authors thank Ontario Power Generation (OPG)
for their technical support of this opportunity to con-
Conclusions and Future Directions tribute to the academic study of CO2 capture. The
information expressed herein is that of the authors, and
This work represents the initial stage of an effort to OPG takes no position on CO2 capture and is simply
consider the large-scale optimization of CO2 emissions furthering scientific endeavors.
from the Ontario power grid. The optimization problem
was formulated as a mixed integer linear program
Nomenclature
(MILP) and was implemented in GAMS. The MILP
model was applied to OPG’s existing power plants as a A ) set of alternative energy power plants
first attempt to evaluate the best solution for OPG’s C ) set of coal energy power plants
power plants under three different operating strate- Ci ) operating cost per unit of electricity generated
gies: total cost reduction, CO2 emissions reduction, and %C ) percentage of carbon for specific type of fuel
an integrated operational mode. The optimization re- EFi ) emission factor for plant i
sults indicate that fuel balancing and fuel switching are Ei ) electricity generated by the ith type of power plant
effective ways to reduce CO2 emissions. Preliminary (MW‚h/year)
optimization results show that overall CO2 emissions Emax ) maximum amount of electricity generated by the
i
can be reduced by 3% or less by adjusting OPG’s current ith type of power plant (MW‚h/
operational level without having to change the fleet year)
structure (fuel balancing). Fuel balancing also results Enom ) nominal amount of electricity generated by the ith
in a reduction of operating costs by 2.4%. However, if type of power plant (MW‚h/year)
CO2 emissions are to be reduced further (e.g., by >6%, fi ) capacity factor of the ith type of power plant
Canada’s Kyoto target), it will be necessary to employ H ) set of hydroelectric energy power plants
structure changes such as fuel switching. Indeed, this HHVi ) high heating value of specific fuel i (MJ/t)
solution leads to increases in the total cost because it N ) set of nuclear energy power plants
involves retrofitting cost and use of natural gas, which N ) number of power plants
is currently OPG’s most expensive fuel. Minimizing the NG ) set of natural gas energy power plants
CO2 emissions implied a very large cost ($467.5 million/ O ) set of oil energy power plants
year) because the optimizer rightly decided to switch ηi ) power-plant efficiency
all coal-fired boilers to natural gas regardless of the
price of the fuel.
Literature Cited
If higher CO2 emissions reductions, greater than 40%,
are required, it will be necessary to employ CO2 capture (1) Ontario Power Generation, Ontario, Canada. Personal
and sequestration technologies. However, CO2 capture communication, 2003.
processes are energy-intensive and require large amounts (2) Guldmann, J.-M.; Shefer, D. Industrial Location and Air
of supplemental energy, not to mention the existence Quality ControlsA Planning Approach; John Wiley & Sons: New
York, 1980; pp 63-101.
of large geological reservoirs for CO2 storage. The (3) Flagan, R. G.; Seinfeld, J. H. Fundamentals of Air Pollution
supplemental energy requirements for CO2 capture Engineering; Prentice Hall: Englewood Cliffs, NJ, 1988.
processes can be supplied by the fossil-fuel plant itself, (4) Rossiter, A. P.; Kumana, J. D.; Ozima, M. K. Ranking
energy from the grid, special-purpose power plants built Options for Pollution Prevention and Pollution Control Using
to supply the CO2 capture energy, or an oversized power Graphical Methods. Adapted from “Rank Pollution Prevention and
plant built to supply both the CO2 capture energy and Control Options”. Chem. Eng. Prog. 1994, 90.
additional energy to the grid. These issues will be (5) Pirrone, N.; Batterman, S. A. Cost-Effective Strategies to
Control Radon in Residences. J. Environ. Eng. 1995, 121 (2), 120-
discussed in another paper. 131.
The sensitivity analysis shows that similar fuel- (6) Haq, I.; Kumar, S.; Chakrabarti, S. P. Cost-Benefit Analysis
switching patterns are observed when the natural gas of Control Measures in Cement Industry in India. Environ. Int.
price and retrofit cost are varied for different growth 1997, 23 (1), 33-45.
rates in demand. However, the coal price directly (7) Kohn, R. E. A Mathematical Programming Model for Air
affected the optimal configuration or fuel-switching Pollution Control. Presented at the Annual Convention of the
Control Association of Science and Mathematics Teachers, St.
pattern of the fleet. The solutions generated from Louis, MO, Nov 30, 1968.
different modes produced different results because there (8) Schweizer, P. F. Determining Optimal Fuel Mix for Envi-
are conflicts between minimizing cost and minimizing ronmental Dispatch. IEEE Trans. Autom. Control 1974, 19, 534-
CO2 emissions. It is likely that government regulations 537.
will be imposed, and therefore, economic mode (mini- (9) Lou, J. C.; Chung, K. F.; Chen, K. S. An Optimization Study
mizing cost while satisfying a CO2 reduction target) will of Total Emission Controls for Particulate Pollutants. J. Environ.
be satisfactory. However, in the absence of governmen- Manage. 1995, 43, 17-28.
(10) Jackson, W. E., Jr.; Wohlers, H. Determination of Regional
tal regulations, the use of a compromise programming
Pollution Control Costs and the Cost of Air Pollution Reduction
model will have to be considered. in the Delaware Valley. Presented at Environmental Engineering
The analysis used in this paper was static in the sense and Science, Philadelphia, PA, 1970.
that the electricity generation was held constant. How- (11) Kemner, W. F. Cost-Effectiveness Model for Pollution
ever, in real situations, there is high variability in Control at Coking Facilities; Research Report EPA-600/2-79-185;
electricity demand. The demand can be represented by U.S. Environmental Protection Agency, U.S. Government Printing
Office: Washington, DC, 1979.
load duration curves, and the model presented in this (12) Holnicki, P. Air Quality and Emulsion Abatement in
paper can be utilized for such situations as well by Regional Scale. In Computer Simulation, Air Pollution II; Bal-
invoking it for different (e.g., high, low, medium) dasano, J. M., Brebbia, C. A., Power, H., Zannetti, P., Eds.;
demands. Computational Mechanics Publications: Boston, 1994.
890 Ind. Eng. Chem. Res., Vol. 44, No. 4, 2005

(13) Elkamel, A.; Elgibaly, A.; Bouhamra, W. Optimal Air (19) Noonan, F.; Giglio, R. Planning Electric Power Genera-
Pollution Control Strategies: The Retrofitting Problem. Adv. tion: A Nonlinear Mixed Integer Model Employing Benders
Environ. Res. 1998, 2 (3), 375-389. Decomposition. Manage. Sci. 1977, 23, 946-956.
(14) Genchi, Y.; Saitoh, K.; Arashi, N.; Inaba, A. Assessment (20) ESCAP-UN. Energy Efficiency; UN Publications: New
of CO2 Emissions Reduction Potential by Using an Optimization York, 1995.
Model for Regional Energy Supply Systems. Presented at the Sixth (21) Brooke, A.; Kendrick, D.; Meeraus, A. GAMS: A User’s
International Conference on Greenhouse Gas Control Technolo- Guide; The Scientific Press: San Francisco, CA, 1998.
gies, Kyoto, Japan, Oct 1-4, 2002; Paper B5-6. (22) Analysis and Modeling Division, Natural Resources Canada.
(15) Linares, P.; Romero, C.Aggregation of preferences in an http://climatechange.nrcan.gc.ca/english/Home.asp?x)1.
environmental economics context: A goal programming approach. (23) Canada Energy Outlook. http://www2.nrcan.gc.ca/es/
Int. J. Manage. Sci. 2002, 30, 89-95. ener2000/online/html/chap3b_e.cfm.
(16) Mavrotas, G.; Diakoulaki, D.; Papayannakis, L. An Energy
Planning Approach Based on mixed 0-1 Multiple Objective Linear
Programming. Int. Trans. Oper. Res. 1999, 6, 231-244. Received for review March 24, 2004
(17) Bai, H.; Wei, J. The CO2 mitigation options for the electric Revised manuscript received October 21, 2004
sector. Energy Policy 1996, 24, 221-118. Accepted November 1, 2004
(18) Climaco, J.; Antunes, C. H.; Martins, A.; Almeida, A. A
multiple objective linear programming model for power generation
expansion planning. Int. J. Energy Res. 1995, 19, 419-432. IE049766O