728 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO.

2, MAY 2011

Commitment of Combined Cycle Plants Using a Dual

Optimization–Dynamic Programming Approach
Juan Álvarez López, Member, IEEE, Rolando Nieva Gómez, and Isaías Guillén Moya

Abstract—Due to the existence and building of an important Constraints that are particular to thermal units include min-
number of combined cycle plants throughout electric power imum and maximum up and down time limits, start-up costs,
systems around the world, there exists the growing need to have and minimum and maximum generation limits. When CCPs
a more accurate and practical model to represent these type of
power plants when solving the unit commitment problem. A com- are incorporated to the UC problem, the following constraints
monly used optimization technique implemented to solve the unit must be considered: different configurations of the CCPs, fea-
commitment problem is dual programming. This work focuses on sible transitions between configurations, and transitions costs
solving the subproblem of scheduling a combined cycle plant using between configurations. For the regulated electricity industry,
dynamic programming under a dual optimization framework. The the objective when solving the UC problem is to minimize the
model used to represent the combined cycle plants is component
based; this new way of modeling combined cycle plants takes operating costs while satisfying the demand, whereas for the
into account such constraints as the feasible transitions between deregulated electricity industry, the goal of market participants
configurations, and the minimum and maximum time that a is to maximize their benefits rather than satisfying the demand
combined cycle plant must remain on a certain configuration. The at a minimum cost. When network constraints are incorporated
advantage of the model presented in this paper is that it accurately to the UC problem, and using an ac formulation, the UC
models all the physical constraints of combined cycle plants. Also,
this paper presents the modeling of hybrid combined cycle plants; problem is known as security constrained unit commitment
these are the ones that use an auxiliary boiler in order to increase (SCUC) [1]–[4].
the production of steam used by the steam turbine. This paper is organized as follows. Section II explains the
Index Terms—Combined cycle plant, configurations, dual pro- operation and technical constraints of CCPs. The different
gramming, dynamic programming, transitions, unit commitment. models used to represent the CCPs are discussed in Section III.
Section IV presents a brief discussion on the different existent
approaches used to solve the UC problem when incorporating
I. INTRODUCTION CCPs. The new model based on a component model using
dynamic programming for CCPs under a dual programming
scheme is detailed in Section V. Also, in this section, a com-
D UE to the recent changes in the electricity industry such as
deregulation, the opening of the electricity market to pri-
vate investors, and an increased concern for the environment, an
parison between our model and previous ones is made in order
to show the benefits of our formulation. Section VI presents
important number of combined cycle plants (CCPs) have been two numerical examples to illustrate the practicality of the new
built all over the world. This, in turn, has brought new chal- model for CCPs using an integer state index. Some concluding
lenges to the unit commitment (UC) problem. The challenges remarks are given in Section VII. An Appendix at the end of
mainly come from the need to model the CCPs in a more ac- this paper discusses how the heat energy requirement for a CCP
curate and practical way and then incorporate this model to the can be obtained.
UC problem.
The short-term UC problem is basically to determine, for II. COMBINED CYCLE PLANTS
each hour of the next day and for up to seven days, the optimum CCPs are generating units that have flexible operating con-
operating point of the available hydro and thermal units in order ditions. Other generating units with flexible operating condi-
to satisfy the forecasted level of demand with the minimum tions are: 1) fuel switching units, 2) fuel blending units, 3) con-
operating cost and, while doing so, meeting all the physical stant/variable pressure units, 4) over-fire units, and 5) dual boiler
and operational constraints of the power system. Some of the units [1], [5]. CCPs are made up of one or more combustion tur-
more common constraints incorporated to the UC problem bine (CT) generator, each one of them with its own heat recovery
are load balance, spinning reserve, scheduled reserve, offline steam generator (HRSG), and one conventional steam turbine
reserve, must-run units, and fuel consumption, among others. (ST) generator common to all the CTs as shown in Fig. 1. The
most basic CCP is formed by one CT generator, one HRSG,
Manuscript received November 11, 2009; revised April 28, 2010 and July 07, and one ST generator, and its operation is as follows: In the first
2010; accepted August 03, 2010. Date of publication September 13, 2010; date
of current version April 22, 2011. Paper no. TPWRS-00875-2009.
stage, a mixture of air and fuel is burned in the combustor of the
J. Álvarez López and I. Guillén Moya are with the Network Analysis De- CT. The released energy by the combustion is used to move the
partment, Electric Research Institute of México, Cuernavaca, Morelos, México CT, which in turn moves a generator to produce electricity. In
(e-mail: jalvarez@iie.org.mx; iguillen@iie.org.mx). the second stage, the hot gases of the CT, which otherwise would
R. Nieva Gómez is with the Division of Electric Systems, Electric Research
Institute of México, Cuernavaca, Morelos, México (e-mail: rnieva@iie.org.mx). be wasted to the atmosphere, are used by the HRSG to generate
Digital Object Identifier 10.1109/TPWRS.2010.2066584 steam and that steam is used to move the ST that in turn moves
0885-8950/$26.00 © 2010 IEEE

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 ÁLVAREZ LÓPEZ et al.: COMMITMENT OF COMBINED CYCLE PLANTS USING A DUAL OPTIMIZATION–DYNAMIC PROGRAMMING APPROACH 729

configuration is the number of periods, i.e., hours, that the CCP

must remain on that configuration. If it takes for a given CCP
three periods to complete the cold start-up sequence, then the
start-up configuration has three states. The transitions between
configurations must follow operational and time constraints;
these are imposed by the number of periods that a CCP must
remain on a given configuration and also by the rules that
dictate to which configuration it is feasible to transition to [1],
[2], [5].

III. MODELING OF COMBINED CYCLE PLANTS

In the literature, there are reported three main ways of mod-
eling CCPs. These are: 1) aggregated model, 2) mode model,
and 3) component model. A brief description of each one of
them is presented next.

A. Aggregated Model
The aggregated model ignores all the components of a CCP
and considers it as a single equivalent unit; hence, the sched-
Fig. 1. Schematics of a CCP.
uling of the CCPs is done as if they were conventional thermal
units. This is a very simplistic model since the commitment of
the equivalent unit results in an on-off state, therefore ignoring
a generator to produce electricity. In a CCP, the power output of all the different configurations and technical constraints of the
the ST generator is usually half the output of the CT generators. CCP. At the end, the configuration of the CCP in real time is left
For instance, if a CCP has a configuration with one CT generator to the plant operators. This model is currently in use by ISO NE,
producing an output of 80 MW, the ST generator would have an NYISO, MISO, and PJM [7]. The model currently used by the
output of 40 MW, making 120 MW the total output of the CCP. Electric Research Institute of México (IIE, acronym in Spanish)
This configuration of a CCP can reach an efficiency of up to is also an aggregated model. The power output of a CCP that is
60%; this is an improvement of 20%–30% with respect to con- modeled as an equivalent thermal unit is the sum of all the CT
ventional combustion turbines [2]. CCPs have high thermal effi- generators affected by a contribution factor to the ST generator.
ciency, that is, a lesser fuel consumption to generate the same en- Even though the aggregated model is a “working” representa-
ergy. Less fuel consumption implies lower operating costs and tion of a CCP, a more accurate model is needed to account for
less emission of pollutants per unit of energy generated, making the different configurations of a CCP; this is the mode model.
them not only an economical option but also an environmentally
friendly solution [6]. Even though some of the less expensive B. Mode Model
generating units are the ones that burn fossil fuel, they are also In the mode model, each configuration of a CCP is accounted
the ones that produce more pollutants. Some of these pollutants by means of an equivalent pseudo unit that has its own heat-
are: 1) sulfur dioxide , 2) nitrogen oxides , and 3) rate curve and its own minimum up and down time. Hence,
carbon dioxide . Unlike fossil fuels, the burning of nat- each pseudo unit representing a configuration of the CCP has
ural gas does not produce and produces considerably less its own quadratic cost function. The feasible transitions between
and [5]. pseudo units (configurations) is determined by a pre-established
CCPs have different configurations and each configuration state–space transition diagram. Pseudo units corresponding to
may have several states. The generation capacity of a CCP the same CCP are mutually exclusive. If one wants to model the
changes depending on which configuration the CCP is on, and minimum/maximum time that a CCP must remain on a given
the transition costs from one configuration to another are by configuration, then each configuration (pseudo unit) must have
no means negligible. CCP configurations are dependent on its own space transition diagram [2], [3], [8]. Even though this
the different combinations that can be formed with the CT model is quite an improvement on the aggregated model, a more
generators, HRSGs, and the ST generator; for instance, CT closely related model to the physical components of a CCP is
generators can be operated with or without their associated still needed; this is the component model.
HRSGs whereas the ST generator cannot be operated without
at least one HRSG available. Based on this, consider one CCP C. Component Model
formed by two CT generators, namely 1 and 2, and only one This model considers the three major components of a CCP
HRSG connected to the one ST. This CCP would have only individually, namely CT generators, HRSGs, and ST generator.
the following seven configurations: 1) CT generator 1, 2) CT Considering these three major components of the CCP has
generator 2, 3) CT generators 1 and 2, 4) CT generator 1 major advantages over the aggregated model and the mode
and HRSG-ST generator, 5) CT generator 2 and HRSG-ST model since it allows the representation by individual CT of
generator, 6) CT generators 1, 2, and HRSG-ST generator, the minimum and maximum on and off time, load ramp rates,
and 7) all components off. The number of states for a given minimum and maximum power output limits, individual heat

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 730 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

energy requirement curves, start-up costs; by ST of the min- coupling constraints can be obtained for each period of the plan-
imum and maximum power output limits; and by CCP of the ning horizon using the well-know LR technique [12]. A number
start-up sequences, configurations, states, and transition costs of Lagrange multipliers from the coupling constraints can be
between configurations and states. All the possible transitions added to the objective function of the individual CCP commit-
between configurations are represented by state–space dia- ment problem. The most representative coupling constraint in
grams. Another advantage of this model is that the availability any unit commitment problem is the total power balance con-
of each one of these components can be taken into account. straint [4].
Reference [9] gives a comprehensive comparison between the The cost of being in a feasible configuration/state in period
mode model and the component model. , used in the formulation of the DP problem, is obtained by
solving the optimization problem shown in (1) and (2):
IV. SOLUTION TECHNIQUES
There are several solution techniques to solve the UC and
SCUC problems when CCPs are included in the formulation.
Some of these techniques are mixed integer programming (1)
(MIP) [8], [9]; evolutionary algorithms like genetic algorithms (2)
(GA), evolutionary programming (EP), and particle swarm
where , , and are the coefficients of the quadratic cost
optimization (PS) [10]; and dual programming, commonly
function derived from the input-output curves of the CT gener-
known as Lagrangian relaxation (LR) when applied to the UC
problem, combined with dynamic programming. A variation of ators,1 is the unit, is the set of units that belong to config-
the latter one, called dynamic programming with Lagrangian uration , is the Lagrange multiplier that is associated with
the power balance constraint, is the power output of unit
reduction of search-range, is proposed by [11]. The LR tech-
in period , and is the contribution factor. The contribu-
nique decomposes the UC problem into a set of subproblems
in which the coupling system constraints are relaxed and in- tion factor is a dimensionless factor that when multiplied by
cluded in the objective function by means of their Lagrange the generation output of the CT generator gives as a result the
contributed generation output of the ST generator due to the ex-
multipliers. Each subproblem, one subproblem per individual
haust gases of the CT generator. The total power output of the
generating unit, minimizes the generation costs, start-up costs,
and the terms including the Lagrange multipliers subject to CCP in configuration is the sum of all the power outputs of the
minimum and maximum on and off time constraints, minimum CT generators pertaining to configuration plus the generation
of the ST generator due to the contribution of the CTs. This is
and maximum generating level constraints, and ramp rate
shown in (3), where the assumption that all the CTs are equal
constraints. These subproblems are solved individually using
DP; the commitment of each one of the generating units is done has been made, and represents the total power output of
for all the periods considered in the planning horizon [1]–[3]. the CCP while in configuration :
The techniques used to solve the SCUC problem are Benders’ (3)
decomposition and LR with DP. In [5], the SCUC problem is
divided into two problems using Benders’ decomposition. The
master problem solves the unit commitment problem while the From (3), one can see that the contribution factor is defined
slave problem minimizes the network violations. The master as
problem is solved with LR and DP using transition diagrams
(4)
for the CCPs.
As pointed out by [8], compared to the LR solution, the MIP
solution would exhibit a globally optimal solution. However, One contribution factor is obtained for each configuration of
when MIP is used to solve larger real-life systems, the compu- the CCP. Hence, the generation limits for the CCP in configu-
tation time and memory requirement are major obstacles for its ration are as shown in (5):
application. Hence, our work focuses on the solution of the sub- (5)
problem of committing the CCPs based on a component model
under a dual programming scheme.
The use of the contribution factor is an equivalent idea to
V. COMBINED CYCLE PLANTS BASED ON A COMPONENT the one presented in [9], where each CT has a MW-generated
MODEL UNDER A DUAL PROGRAMMING SCHEME steam curve which represents the amount of steam that will be
When scheduling CCPs using DP, all the configurations that produced by the HRSG when the CT generates a certain MW of
the CCP has, along with their states, minimum and maximum electricity and each ST has a consumed steam-MW curve that
number of periods for each configuration, transition costs be- represents the amount of electricity generated by the ST given
tween configurations and states, and the set of feasible transi- a certain amount of steam.
tions between configurations and states, must be defined. Also, Some CCPs may have an auxiliary boiler, as shown in Fig. 2,
for each configuration, including all its states, its own minimum in order to generate more steam to aid the HRSG to drive, or
and maximum generation levels must be defined as well. even drive independently, the ST. We name this type of CCPs
The new proposed model solves the problem of scheduling a 1For a discussion on how the input-output curves are obtained for a CCP,
CCP using forward DP where the Lagrange multipliers for the please refer to the Appendix.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 ÁLVAREZ LÓPEZ et al.: COMMITMENT OF COMBINED CYCLE PLANTS USING A DUAL OPTIMIZATION–DYNAMIC PROGRAMMING APPROACH 731

is the power output of the ST generator due to exhaust gases of

CTs (residual).
The total power output of an HCCP while in configuration
is shown in (9):

(9)

The configurations/states in this model are represented by

only one integer state index that, depending on its value, is
able to identify each and every one of the different configura-
tions/states. This is accomplished by assigning a value of 1 to
any of the configurations at the start and then, consecutively,
adding the number of periods that it takes to complete each and
every one of the configurations/states for the CCP. There are
a number of benefits in representing the configurations/states
by means of an integer state index. Unlike previous formula-
tions, [1]–[3], [9], when representing the minimum and max-
imum time for each configuration, the number of variables and
the number of states in the state transition diagram do not in-
Fig. 2. Schematics of a CCP with auxiliary boiler. crease. Hence, the state–space diagrams using our formulation
are much simpler. Another advantage is that our formulation
does not need integer variables in order to keep track of the
hybrid combined cycle plants (HCCPs). Depending on the number of hours that a CCP has been on a given configuration,
configuration and output of the exhaust gasses of the CTs, the nor does it need binary start-up/shut-down variables. These ad-
HRSG may not be able to generate enough steam to drive the vantages are exemplified next by computing the table of values
ST generator and meet its minimum generation limit. When of the integer state index.
this happens, the auxiliary boiler must be used (supplementary Consider a CCP formed by two CTs and one ST; it has three
heat, SH) in order to meet the minimum generation limit of the different stop modes and three different start-up sequences.
ST generator. The use of the auxiliary boiler comes at the cost First, start by assigning the value of 1 to the integer state index
of reducing the efficiency of the HCCP. The cost of being in a and associate it with any configuration, say the hot stop
feasible configuration/state in period for an HCCP, used in (HS). Suppose that the CCP must remain in the HS configura-
the formulation of the DP problem, is shown in (6): tion for at least a minimum number of periods, , and cannot
remain in such configuration more than a maximum number
of periods, . Normally, using previous formulations, this
would imply that the state transition diagram for this config-
(6) uration would have as many states as and would need a
separate variable to keep track of the number of periods that the
where is the component of the power output of the ST CCP has been on the HS configuration. Using the state index
generator due to the use of supplementary heat, and and formulation, the CCP is in the HS configuration for as long as
are the coefficients of the quadratic cost function derived has a value of
from the input-output curve of the ST generator when using sup-
plementary heat. A discussion on how the input-output curve is (10)
obtained for an HCCP is presented in the Appendix.
The power generation limits for the CT generators remain as Next, suppose that the next configuration to be added to the
in (2). When the exhaust gases of the CTs are not enough to sat- state index is the warm stop (WS) configuration. This configu-
isfy the minimum generation requirement of the ST generator, ration also has its own minimum/maximum time, namely
the power limits of how much power is to be generated using and . The CCP is in the WS configuration for as long as
SH are as shown in (7). When the exhaust gases of the CTs are has a value of
enough to satisfy the minimum generation requirement of the
ST generator, the power limits of how much power is to be gen- (11)
erated using SH are as shown in (8):
It can be seen then that in our formulation, it is not necessary
(7) to define a variable to keep track of the number of hours that
(8) a CCP has been on a given configuration before completing its
minimum/maximum number of hours that must remain in such
where and are the minimum and maximum gen- configuration. Neither is it necessary to define a variable to in-
eration limits of the ST generator, is the power output of dicate in which configuration the CCP is; all this information is
the ST generator due to the SH, and contained in the integer state index.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 732 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

TABLE I TABLE II
UNIT COST FUNCTIONS AND GENERATION LIMITS HCCP CONFIGURATIONS AND STATES HCCP

in $=MW h, in $/MWh, in $/h, and in MW.

The CCP only has a minimum time to remain in the cold stop
(CS) configuration. Adding the CS configuration to the integer
state variable, one has that

(12)

The three startup-sequences that the CCP has are cold

(CSUS), warm (WSUS), and hot (HSUS) start-up sequences.
Each of these sequences takes a number of periods to be com-
pleted. Equations (13)–(15) show the value of the integer state
index for the CSUS, WSUS, and HSUS, respectively:
generators and the supplementary heat remain constant over the
planing horizon regardless of the configuration of the HCCP.
(13)
Three start-up sequences are considered: cold start-up se-
quence (CSUS), warm start-up sequence (WSUS), and hot
(14) start-up sequence (HSUS); the number of hourly periods that
each one of the starp-up sequences take until their completion is
also considered. The minimum and maximum number of hourly
(15)
periods that an HCCP must remain in a given configuration is
The 1 CT and 1 ST configuration, and the 2 CT and 1 ST considered as well. In Table II, the configurations and states for
configuration have their minimum operating time: and this numerical example are given, where , ,
, respectively. Equations (16) and (17) show the value of , , , , ,
the integer state index for the 1 CT+ST and 2 CT+ST configu- , , , and .
rations, respectively: CT-only configurations are not taken into account. It is also
considered that the residual heat from the 3 CTs is not enough
to meet the minimum output power of the ST; this assumption
causes the need to use supplementary heat. Assume that there is
(16) a fixed transition cost of $73.94 every time that a CT is started.
As explained before, the integer state index , depending
on its value, represents each and every single one of the config-
urations/states of the HCCP. The value of for all the config-
(17) urations/states, after defining as the first hourly period
of the HS configuration, is shown in Tables III and IV.
In order to validate our results, the following section presents The feasible transitions between configurations and states are
two numerical examples. shown in Table V; only the commitment of the 3 CT + ST from
any start-up sequence is considered as a possible transition. It
VI. NUMERICAL EXAMPLES is important to notice that it is not necessary to explicitly con-
In order to illustrate the usefulness of our proposed model, sider in the state–space diagram configurations where only CTs
and since the focus of this paper is on the solution of the sub- are in operation; this would only make the state–space diagram
problem of committing CCPs using DP, we present the commit- more complex. Instead, this situation can be handled by setting
ment of an HCCP over a 24-h period and the commitment of a to zero the contribution factor of the CTs whenever the ST is not
CCP over a 168-h period. The model is implemented in FOR- available. Otherwise, operating a CCP without its ST, if avail-
TRAN and tested on a Pentium 4, 3.2 GHz with 1 GB of RAM. able, would be inefficient. Even though CT-only configurations
are less efficient than CT + ST configurations, the start-up time
A. Hybrid Combined Cycle Plant Over a 24-h Period from a cold start to any CT only configuration is usually shorter
Consider an HCCP (supplementary heat) that has three CT than the start-up time from a cold start to any CT + ST con-
generators and one ST generator. The planning horizon is 24 figuration. This flexible feature of the CCP could also be mod-
hourly periods. It is assumed that the contribution factor, eled, if required. It also should be noted that after completing
, and the cost functions, as shown in Table I, for the CT any start-up sequence, the CCP has all its CT units on. This is

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 ÁLVAREZ LÓPEZ et al.: COMMITMENT OF COMBINED CYCLE PLANTS USING A DUAL OPTIMIZATION–DYNAMIC PROGRAMMING APPROACH 733

TABLE III TABLE IV

INTEGER STATE INDEX  HCCP, A INTEGER STATE INDEX  HCCP, B

usually the case for representative CCPs, and this information is

obtained from the manufacturer’s data sheets. However, for spe-
cial cases, configurations where a CCP after completing any of
its start-up sequences has only some of its CTs on can be added remain constant over the planing horizon regardless of the con-
to the state–space diagram. figuration of the CCP. Like in the previous example, CT-only
The generation, after solving, per component of the HCCP configurations are not taken into account. Also, only the com-
along with the HCCP configuration are shown in Table VI. Fig. 3 mitment of the 2 CT + ST from any start-up sequence is consid-
shows the detail of the dynamic programming search path for ered as a possible transition.
the first five periods. It can be seen that the transition from the The value of for all the configurations/states, after
CSUS, when completed, to the 3 CT+ST+SH configuration is defining as the first hourly period of the HS configura-
imposed since the residual heat from the 3 CTs is not enough to tion, is shown in Table IX.
meet the minimum generation limit of the ST generator; hence, The feasible transitions between configurations and states are
it is forced to use SH to meet the minimum generation limit shown in Table X.
of the ST generator. The CPU time to solve this problem is Due to space limitations, only 24 h of the 168-h period are
0.1718750 s. reported. The generation per component of the CCP along with
the CCP configuration for hours 95 to 118 are shown in Table XI.
B. Combined Cycle Plant Over a 168-h Period The CPU time to solve this problem is 3.656250 s.
Consider a CCP that has two CT generators and one ST gen- Other tests, not reported, of our model include simultaneous
erator. The planning horizon is 168 hourly periods. Like be- commitment of nine CCP and one HCCP embedded into the
fore, three start-up sequences are considered. In Table VII the current planning tool used by the Mexican Electricity Company
configurations and states for this numerical example are given, (Comisión Federal de Electricidad) using representative data for
where , , , , , a planing horizon of up to 168 h (one week).
, and . The transition cost is variable,
depending on the time that a given CT has been off by the time VII. CONCLUSION
it is started. This paper presents a new way of modeling CCPs and HCCPs
It is assumed that the contribution factor, , and based on a component model using DP under a dual optimiza-
the cost functions, as shown in Table VIII, for the CT generators tion scheme. This new model is a practical answer to the needs

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 734 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

TABLE V TABLE VI
FEASIBLE TRANSITIONS HCCP GENERATION PER COMPONENT HCCP

in MW, in $/MWh.

of power system operators by means of providing a more ac-

curate and practical model that is able to represent to the fullest
extent possible the technical constraints of a CCP. Among these Fig. 3. Dynamic programming solution for the first five periods.
technical constraints are the different start-up sequences and the
different configurations/states that a CCP has. The minimum
and maximum number of intervals that a CCP must remain in APPENDIX
a given configuraion/state are also considered. Benefits of mod- This Appendix describes how to obtain the heat energy re-
eling the different sates/configurations by means of a single in- quirement, the input-output curve, for each one of the CT gen-
teger state index are a decrease in the number of integer vari- erators and the ST generator of a CCP and an HCCP.
ables and a simplification of the space–state diagrams. A new From actual measurements obtained from heat rate tests of
way of modeling HCCPs, the ones with a dual boiler to produce any thermal generation unit, the coefficients of the input-output
supplementary heat, is also proposed. curves can be derived. This is done after multiplying the heat

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 ÁLVAREZ LÓPEZ et al.: COMMITMENT OF COMBINED CYCLE PLANTS USING A DUAL OPTIMIZATION–DYNAMIC PROGRAMMING APPROACH 735

TABLE VII rate characteristic taken at specific output levels by the genera-
CONFIGURATIONS AND STATES CCP tion output of the unit and then fitting these values to a polyno-
mial curve, typically to a quadratic curve [4]. The authors use a
weighted least-squares method to fit the heat rate characteristic
to a quadratic convex function. The details of this technique are
out of the scope of this paper.
The following assumptions are made. It is assumed that all
the CT generators are equal and that the generation output of
the CCP and the ST generator are known. It is also assumed
that the efficiency of the ST is known. Measurements are taken
at several generating outputs of the CCP.
Before fitting the heat energy requirement data to a quadratic
curve, and after multiplying the heat rate characteristic taken at
specific output levels by the generation output of the unit, the
following computations are made.
The heat energy required by the ST generator to generate the
TABLE VIII
UNIT COST FUNCTIONS AND GENERATION LIMITS CCP power output is

(18)

where is the efficiency of the ST and is the conversion

in $=MW h, in $/MWh, in $/h, and in MW. factor from electric power to calorific power which value is 0.86
Gcal/h.
TABLE IX From Fig. 4, one can see that the total heat energy required by
INTEGER STATE INDEX  CCP the CCP to generate the power output can be expressed
by

(19)

where is the residual heat of the exhaust gasses of the CT

generator . Defining as the total residual heat
of the exhaust gasses of all the CT generators, one has that

(20)

Defining as the combined efficiency of all the heat

recovery steam generators, and with the supplementary heat
equalling zero, one has that

(21)

Since it is assumed that all the CT generators are equal, the

heat energy requirement for each one of the CT generators is

(22)

where is the cardinality of .

The input-output curves for the ST generator and each one of
the CT generators are obtained after fitting the points obtained
from (18) and (22) to a quadratic function, respectively. Once
the quadratic input-output curve has been obtained for the ST
generator, the input-output curve for the supplementary heat can
be obtained. When supplementary heat is to be used, one has that
the total heat energy to the ST generator is

(23)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 736 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 2, MAY 2011

TABLE X
FEASIBLE TRANSITIONS CCP

Fig. 4. Heat energy in a CCP.

TABLE XI
GENERATION PER COMPONENT CCP

Fig. 5. Input-output curve of an ST generator with SH.

Defining the input-output curve of the steam turbine gener-

ator as

(26)

The heat energy to the ST due to the residual heat of the CTs
is . Using the ST input-output curve, obtained after
in MW, in $/MWh.
fitting the points obtained by (18), one has that
. Using (24)–(26), one has
that
and, from (23), the heat energy required by the supplementary
heat to generate is (27)

(24) where and . The input-

output curves of the ST generator and the SH are shown in Fig. 5.
The power output of the ST generator has two components:
a component due to the residual heat of the CT generators, ,
and a component due to the supplementary heat, ; this is REFERENCES
shown in (25):
[1] A. I. Cohen and G. Ostrowski, “Scheduling units with multiple oper-
ating modes in unit commitment,” IEEE Trans. Power Syst., vol. 11,
(25) no. 1, pp. 497–503, Feb. 1996.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.
 ÁLVAREZ LÓPEZ et al.: COMMITMENT OF COMBINED CYCLE PLANTS USING A DUAL OPTIMIZATION–DYNAMIC PROGRAMMING APPROACH 737

[2] B. Lu and M. Shahidehpour, “Short-term scheduling of combined cycle Network Analysis Department. His main research interests are power system
units,” IEEE Trans. Power Syst., vol. 19, no. 3, pp. 1616–1625, Aug. planning, deregulated power systems, unit commitment for large power systems,
2004. and power plant modeling.
[3] M. R. Bjelogrlic, “Inclusion of combined cycle plants into optimal
resource scheduling,” in Proc. IEEE 2000 Power Eng. Soc. Summer
Meeting, 2000, vol. 1, pp. 189–194.
[4] B. R. Wollemberg and A. J. Wood, Power Generation Operation and Rolando Nieva Gómez received the B.Sc. degree in
Control. New York: Wiley, 1996. 1973 in mechanical-electrical engineering from the
[5] B. Lu and M. Shahidehpour, “Unit commitment with flexible gener- Instituto Tecnológico y de Estudios Superiores de
ating units,” IEEE Trans. Power Syst., vol. 20, no. 2, pp. 1022–1034, Monterrey, Monterrey, Nuevo León, México. He also
May 2005. received the Ph.D. degree in electrical engineering
[6] R. M. Rifaat, “Economic dispatch of combined cycle cogeneration from the University of Alberta, Edmonton, AB,
plants with environmental constraints,” in Proc. IEEE 1998 Int. Conf. Canada in 1979.
Energy and Management and Power Delivery, Mar. 3–5, 1998, vol. 1, Since 1979, he has been with the Electric Research
pp. 149–153. Institute of Mexico, Cuernavaca, Morelos, México,
[7] G. Anders, Commitment Techniques for Combined-Cycle Generating where he is currently the Director of the Electric Sys-
Units, Kinectrics, Inc., Rep. T053700-3103, Dec. 2005. tems Division. He has authored or coauthored more
[8] T. Li and M. Shahidehpour, “Price-based unit commitment: A case than 50 technical papers.
of Lagrangian relaxation versus mixed integer programming,” IEEE
Trans. Power Syst., vol. 20, no. 4, pp. 2015–2025, Nov. 2005.
[9] C. Liu, M. Shahidehpour, Z. Li, and M. FotuhiFiruzabad, “Compo-
nent and mode models for the short-term scheduling of combined-cycle
units,” IEEE Trans. Power Syst., vol. 24, no. 2, pp. 976–990, May 2009. Isaías Guillén Moya received the B.Sc. degree in
[10] F. Gao and G. B. Sheble, “Stochastic optimization techniques for eco- 1980 and the M.Sc. degree in 1985 in electrical engi-
nomic dispatch with combined cycle units,” in Proc. IEEE 9th Int. neering, majoring in power systems, from the Insti-
Conf. Probabilistic Methods Applied to Power Systems (CD), Royal tuto Politécnico Nacional, México City, México.
Inst. Technol., Stockholm, Sweden, Jun. 11–15, 2006, pp. 1–8. He is currently working as a researcher at the Net-
[11] R. Nieva, A. Inda, and I. Guillén, “Lagrangian reduction of search- work Analysis Department of the Electric Research
range for large-scale unit commitment,” IEEE Trans. Power Syst., vol. Institute of México, Cuernavaca, Morelos, México.
2, no. 2, pp. 465–473, May 1987. His research interests include the operation, planning
[12] D. P. Bertsekas, Nonlinear Programming, 2nd ed. Nashua, NH: and economics of electric energy systems, as well as
Athena Scientific, 1999. optimization theory and its applications.

Juan Álvarez López (M’08) received the B.Sc.

(hons.) degree in 2002 and the M.Sc. (hons.) degree
in 2004 in electrical engineering, majoring in power
systems, from the Instituto Tecnológico de Morelia,
Morelia, Michoacán, México. He also received a
Ph.D. degree in electrical and computer engineering
from the University of Waterloo, Waterloo, ON,
Canada, in 2008.
Since 2008, he has been with the Electric Research
Institute of Mexico, Cuernavaca, Morelos, México,
where he is currently working as a researcher at the

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA. Downloaded on April 17,2021 at 15:47:12 UTC from IEEE Xplore. Restrictions apply.