238 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO.

1, FEBRUARY 2008

Transmission Expansion Planning

in Electricity Markets
Sebastián de la Torre, Member, IEEE, Antonio J. Conejo, Fellow, IEEE, and Javier Contreras, Senior Member, IEEE

Abstract—This paper presents a mixed-integer linear program- Size of the th block of the th generating unit in
ming (LP) formulation for the long-term transmission expansion all scenarios.
planning problem in a competitive pool-based electricity market.
To achieve optimal expansion planning while modeling market Maximum capacity of line in corridor .
functioning, we define a number of scenarios based on the fu- Weight of scenario .
ture demand in the system and we simulate the maximization
of the aggregate social welfare. Investment and operating costs, Slope of the th block of the linearization of the
transmission losses and generator offers, and demand bids are voltage angle for corridor .
considered. We propose to use a set of metrics to rate the effect
of the expansion on the generators, demands, and the system as Upper bound of the angle blocks of corridor .
a whole. The proposed model is applied to the Garver six-bus Price bid by the th block of the th demand in
system and to the IEEE 24-bus Reliability Test System. Simulation scenario .
results can be interpreted in economic terms based on the values
of the metrics obtained for different scenarios, parameters, and Price offered by the th block of the th generating
topologies. unit in scenario .
Index Terms—Electricity market, mixed-integer linear pro- Weighting factor to make investment and
gramming, social welfare, transmission expansion planning. operational costs comparable.

NOMENCLATURE B. Scenario-dependent variables

HE mathematical symbols used throughout this paper are
T classified below as follows.
A. Constants
Lossless power flow in line of corridor
in scenario .
Total power consumed by the th demand in
Susceptance of line in corridor . scenario .
Conductance of line in corridor . Power consumed by the th block of the th
Investment cost of constructing line in corridor demand in scenario .
. Total power produced by the th generating unit
Number of blocks of the piecewise linearization in scenario .
of power losses. Power produced by the th block of the th
Large enough positive constant. generating unit in scenario .
Number of blocks of the th demand in all Power injection at bus in scenario .
scenarios. Power injection in line of corridor
Number of blocks of the th generating unit in all computed at bus in scenario .
scenarios. Power losses in line of corridor in
Size of the th block of the th demand in scenario .
scenario . Voltage angle at bus in scenario .
Upper bound of the power output of the th Variable used in the linearization of the power
generating unit. losses in corridor th angle block relative
to this corridor in scenario .
Manuscript received November 29, 2005; revised December 19, 2006. The Auxiliary variables used in the linearization of the
work of A. J. Conejo and J. Contreras was supported in part by the Ministry of power losses in corridor in scenario .
Science and Technology under CICYT Project DPI2003-01362 and in part by
the Junta de Comunidades de Castilla–La Mancha under Project PBI-05-053. Nodal price at bus s in scenario .
Paper no. TPWRS-00747-2005.
S. de la Torre is with the Department of Electronic Engineering, University
of Málaga, Málaga, Spain (e-mail: storre@uma.es). C. Global variables
A. J. Conejo and J. Contreras are with the University of Castilla–La
Mancha, Ciudad Real, Spain (e-mail: Antonio.Conejo@uclm.es; Javier.Con-
treras@uclm.es). Binary variable that equals 1 if line from
Digital Object Identifier 10.1109/TPWRS.2007.913717 corridor is built and equals 0 otherwise.
0885-8950/$25.00 © 2008 IEEE
DE LA TORRE et al.: TRANSMISSION EXPANSION PLANNING IN ELECTRICITY MARKETS 239

D. Sets as: linear programming [2], [3], mixed-integer linear pro-

Set of all demands located at bus . gramming [4], [5], Benders decomposition [6], and dynamic
programming methods [7]; heuristic methods based on sigmoid
Set of all generators located at bus . functions [8], intelligent systems, such as genetic algorithms
Set of all lines connected to bus . [9], simulated annealing [10]; and others, such as game theory
Set of all scenarios of the period of study. models [11]–[13]. With the introduction of pool-based markets
Set of indices of the blocks of the th demand. and bilateral contracting, new transmission expansion models
propose social welfare maximization instead, [14], [15].
Set of indices of the demands.
Set of indices of the blocks of the th generating C. Aim and Contributions
unit. In this paper, we present a mixed-integer formulation for the
Set of indices of the generating units. transmission expansion problem in pool-based electric energy
Set of all possible transmission lines, prospective markets. To achieve that, we model the network topology, en-
and existing. ergy losses, generator offers, and demand bids, and we set an
Set of all prospective transmission lines. appropriate timeframe. Different scenarios are considered to ac-
count for different levels of the demand. They are weighted in
Set of all network buses.
the objective function to consider the number of hours in which
E. Metrics the demand has reached a certain level. Later, we calculate the
Metrics to assess the impact of new transmission generator and consumer surpluses and, consequently, the so-
lines in the network. cial surplus. We assume that the planning of the network corre-
sponds to a single transmission entity, although the functioning
I. INTRODUCTION of our pool-based market model allows for independent power
generators and demands that can be in the hands of different
A. Background companies.
In the past, both the operating and the planning aspects of The novel contributions of this paper are threefold:
the electric network were in the hands of centralized entities or- 1) modeling the behavior of the demand through demand-side
ganized in vertically-integrated structures. Nowadays, the un- bidding;
bundling of the electricity business has raised new challenges 2) a scenario-weighted formulation;
that the restructured industry must face. The restructuring of the 3) use of appropriate metrics to measure the welfare of the
electricity industry has resulted in the advent of new players, network planner, the generators, and the demands.
such as brokers, marketers, and independent power producers.
A salient characteristic of the new structure is the decentralized D. Paper Organization
decision making. One critical outcome of the large number of The remaining sections of this paper are organized as fol-
players and the increasing number of transactions is a more fre- lows: Section II presents the formulation of the transmission
quent stressing of the transmission grid due to the creation of expansion problem in competitive markets as a mixed-integer
congestion situations. problem, Section III defines the metrics used to analyze trans-
Network expansion is by its very nature a very complex multi- mission expansion solutions, Section IV presents numerical
period and multi-objective optimization problem [1]. Its non- results using both the Garver six-bus system and the IEEE
linear nature and the inherent uncertainty of future develop- 24-bus Reliability Test System, and Section V provides con-
ments constitute major complications. Its solution is therefore cluding remarks.
difficult, even in centralized environments. In the past, under the
vertically integrated structure, the construction of new transmis- II. FORMULATION
sion facilities has been associated with the addition of new gen- In this section, we formulate the transmission expansion plan-
erating resources and their integration into the existing network. ning problem. This model is an extended version of the one pre-
This was done under the strong control exerted by the regula- sented in [5], where we have expanded the formulation to ac-
tors over virtually every aspect of the regulated utility activities. count for market-driven generation offers and demand bids. The
Under the new paradigm, the economic signals that result from general formulation of the model is as follows:
the daily operations of the electric markets (prices, congestion
metrics, etc.) need to be considered together with the economics
of investment in new facilities in an environment of regulatory
and legislative uncertainty and with the operational control of
the facilities being vested in hands different than the ownership.

B. Literature Review
There are many mathematical models traditionally available
to solve the transmission expansion problem from the cost
minimization standpoint. The solution techniques proposed (1)
can be classified as mathematical optimization methods, such
240 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

subject to in (4) and (5), respectively. This multiplication of a binary vari-

able and a continuous function is the consequence of the fact
that the power through a line is zero if that line is not physi-
cally connected to the network, or equivalently, if .
(2)
The constraints in (6) enforce the line flow limits. The con-
straints in (7) and (10) define the size of the blocks of the gener-
ators (demands) per scenario. The constraints in (8) are the op-
(3) erating constraints that specify that, if a generator is dispatched,
its power output must be within a certain range represented by a
minimum output and a maximum output. The constraints in (9)
(4) and (11) define the power produced (consumed) by any gener-
ator (demand) as the summation of its corresponding production
(5) (consumption) blocks. Finally, the fact that the existing lines
have already been built is enforced in (12) and the binary in-
(6)
vestment decision variables are defined in (13).
The transmission expansion problem formulation presented
(7) in (1)–(13) can be expressed as a mixed-integer linear program-
(8) ming problem. For completeness, the mixed-integer formulation
of the model is provided in the following:
(9)

Maximize
(10)
(11)

(12)
(13)

where is the nodal price at bus in scenario calculated as

(14)
the dual variable of the power balance constraint (2). Its value
is computed once the mixed integer problem has been solved,
fixing the binary variables to their respective optimal values and subject to
solving the resulting continuous linear programming problem.
The objective function in (1) represents the scenario-
weighted social welfare, where the welfare is expressed as
the aggregate demand utility bid function minus the aggregate (15)
generator offer function, minus the investment cost in new
lines.1 To evaluate expression (1), it is necessary to add up all
the offers/bids provided by the generators and the demands and (16)
to use a set of weights corresponding to the periods in which
these bids take place. Moreover, as the proposed planning
procedure considers a single period, the weighting factor is (17)
used to make the investment and operating costs comparable.
It simulates the ratio between the annualized investment costs (18)
and the scenario-weighted operating costs. The analysis of the
results obtained by changing this parameter is explained at the
end of the case studies section.
The constraints in (2) enforce the power balance at every (19)
node. The constraints in (3) imply that the power injection at
(20)
node is the summation, over all lines connected to that node, of
a lossless line flow component and a loss component. For each
line, these two components are the product of a binary variable (21)
and a sinusoidal function of the difference of the angles pre- (22)
vailing at the sending and receiving ends of the line, as shown
1Note that the aggregate social welfare is also equal to the demands’ surplus
plus the generators’ surplus plus the merchandising surplus (total payments from (23)
the demands minus total payments to the generators) minus the investment cost
in new lines. (24)
DE LA TORRE et al.: TRANSMISSION EXPANSION PLANNING IN ELECTRICITY MARKETS 241

(25) The metric available to the generators is the change in the gen-
(26) erator surplus with respect to the investment cost in new lines.
It is defined as

(36)
(27)
(28) where is the aggregate generator surplus defined as the ag-
gregate revenue minus the total costs of the generators, and
(29) is the aggregate generator surplus if transmission expansion is
(30) not considered. This metric illustrates how the generators could
reference bus (31) benefit from the investment in new lines. The value of should
(32) be greater than the share of the cost that the generators will have
to pay for, because otherwise, the generators will most likely
(33) be against the construction of those lines; e.g., if the genera-
tors are responsible for 50% of the cost of the new lines, they
(34) would like to obtain, at least, a value of . If ,
the construction of lines will imply a loss for the generators.
recall that is the nodal price at bus in scenario calculated Also note that values of are possible in systems with a
as the dual variable of the power balance constraint (15). The high degree of congestion, because the construction of new lines
linear expression of the losses needs two additional sets of con- may connect inexpensive isolated generators with the demand,
tinuous variables: and , whose derivation is provided in which will cause prices to fall and, hence, generators profits to
[16]. Further details of this mixed-integer linear programming decrease. However, how the increase in social welfare or gen-
formulation can be found in [5], [17], and [18]. This second erators surplus or consumer surplus due to the construction of
formulation of the problem can be solved by any of the many new lines is allocated to generators or demands is outside the
commercially available programs that deal with mixed-integer scope of this paper.
linear problems; in particular, we have used the CPLEX opti- In addition, both single-generator metrics and gener-
mizer within GAMS. ating-company metrics could be calculated, taking into account
only the surpluses of the relevant units. These metrics could
be useful to compare the effect of the new lines for different
III. METRICS FOR TRANSMISSION INVESTMENT
generating companies.
In order to analyze and compare future investments in trans- Likewise, the demands can measure the increment of their
mission, we need to define a set of metrics that show the welfare surplus with respect to the investment cost in new lines. This
obtained by the different agents of the market: generators, de- metric is defined as
mands, and the transmission entity.
The metric that shows the change in the aggregate social wel- (37)
fare as a result of adding new lines with respect to the investment
cost in new lines is given by the following parameter: where is the aggregate demand surplus defined as the ag-
gregate demand utility function minus the total payment of the
(35) demands, and is the aggregate demand surplus if transmis-
sion expansion is not considered. This metric shows how the
demands could benefit from the investment in new lines. Also,
both single-consumer and consuming company metrics could be
where is the optimal aggregate social welfare, i.e., the first
calculated, taking into account only the surpluses of the relevant
term of (1), and is the aggregate social welfare if trans-
demands. These metrics could be useful to compare the effect
mission expansion is not considered. The calculation of
of the new lines for different consuming companies.
results from the solution of problem (14)–(34). is obtained
Finally, the ratio of the change in the merchandising surplus
if problem (14)–(34) is solved with the additional constraint that
with respect to the investment cost in new lines is defined as
no lines can be built, hence calculating the social welfare in the
situation without transmission expansion. From this definition,
(38)
metric should be greater than one to justify the investment in
new lines.
The previous ratio can be a useful metric for the entire system, where is the aggregate merchandising surplus, and is
but it could be objected by generators or demands that do not see the aggregate merchandising surplus if transmission expansion
a real improvement for themselves with the new transmission is not considered.
lines added. To account for that, we define two other metrics: Note that, as previously stated, , and are
one for the generators, one for the demands. Furthermore, we results obtained from solving problem (14)–(34). Also note that
define an additional metric that considers the effect of new lines , and are easily obtained solving the same
on the merchandising surplus. problem with the additional constraint that no lines can be built.
242 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

IV. CASE STUDIES

A. Introduction
The methodology proposed has been successfully applied
to two case studies. The first case study analyzes the Garver
six-bus system [2], and the second case implemented is based on
the IEEE RTS 24-bus system [19]. The market structure of the
systems considered consists of a number of generating units and
a number of loads; both generating units and loads submit of-
fers/bids to the market trying to attain their respective maximum
profits. All lines are built by a central entity, the network planner.
In the case studies presented next, a linear approximation of
the generators cost function is used; this linear approximation
consists of a set of “blocks” for each of which the marginal cost
is considered constant. Moreover, we consider that each unit
submits to the market one block of offered energy and the price
selected is the marginal cost corresponding to that block. We
also assume that the bids of the demands express their actual Fig. 1. Diagram for the Garver six-bus system used in the first case study.
utility functions. Given the fact that all internal data to the agents
are confidential, our metrics can only be computed in terms of
line built today will be operative for at least 25 years; thus, a
“declared” surpluses. However, each generator and each con-
25 year investment return period has been considered. Also, a
sumer can calculate its own metrics based on “real” surpluses. If
10% discount rate is assumed. With these two values in mind,
offers and bids do not match marginal costs and utilities, only the
the value of the capital recovery factor can be calculated as
“declared” surpluses can be used to compute the metrics used to
, where is the discount rate and the
allocate transmission expansion costs. Regarding bidding data,
number of years; this formula provides a capital recovery factor
we use reasonably sized generator units and cost data closely
value of 0.10. This value means that, for the next 25 years, the
matched with actual values observed in actual markets. Demand investment cost in new lines is yearly repaid at a rate of approx-
data are obtained from realistic cases of the day-ahead electricity imately 10% of the total initial investment. This is also known
market of mainland Spain [20] to get a reasonable number of as the annualized cost.
scenarios based on demand patterns corresponding to year 2004. In the case studies presented in this paper, the word corridor is
A different number of scenarios are considered in the case always used to describe all the lines that connect the same pair of
studies to describe the behavior of the demand. Each of the sce- nodes, whereas the word “line” is simply used to describe each
narios represents a significant number of hours during one typ- one of the lines within a corridor. Hence, subindexes “s, r, k” are
ical year of operation of the network. The different scenarios are used to refer to the particular line “k” that is placed inside the
weighted in the objective function in order to correctly consider corridor that connects nodes “s” and “r.” For example, in Fig. 2,
their relative relevance; hence, the scenarios that represent situ- the corridor between nodes 1 and 5 comprises four lines, and
ations that take place very often are given higher weights, and the corridor between nodes 1 and 2 contains just one line.
conversely, scenarios that represent market situations that take
place only seldom, are assigned lower weights. The main differ- B. First Case Study
ence among the considered scenarios is the amount of demand. The system shown in Fig. 1 contains five nodes and six lines
For each demand block, its size is different in every scenario. connecting them; moreover, a sixth node is considered, at which
The weight of each scenario has been calculated dividing the some generation is placed. This node is not initially connected
amount of hours represented by the specific scenario by the total to the other five nodes, but lines to connect it to the system could
number of hours considered. For example, if scenario number 1 be built if necessary. The constraints imposed in the formulation
represents the first three hours of every day (i.e., from midnight of the problem (14)–(34) allow for the addition of lines in new
until 3:00 A.M.) for a whole year, then the weighting factor for or already existing corridors, up to a total maximum of three
this scenario is: . lines per corridor. The market structure of the system considered
We consider a time horizon of one year, that is, a “target consists of ten generating units and five loads. Table I provides
year.” For this “target year,” we estimate the demand, the gener- line data. The first two columns provide the nodes of origin and
ation offers, and the demand bids. Therefore, our model repre- destination of the lines, the third and fourth columns show the
sents a “static transmission expansion planning” problem, since electric parameters of the lines, and the fifth column shows, in
it considers a “target year” for which the net social welfare is p.u., the capacity of the line. The cost value is shown in the sixth
maximized.2 column for all lines, where the annualized cost is 10% of that
We have made the following assumptions in order to esti- value. Finally, in the last column, the number of lines already
mate the weighting factor in (1). We have assumed that a built for every possible connection between nodes is shown.
2For a detailed explanation of these concepts, please refer to [21]. Also, [22]
Four different scenarios are considered to describe the be-
and [23] define some of the basic concepts regarding “static transmission ex- havior of the demand. The four scenarios considered can be
pansion planning.” regarded as: low demand, medium-low demand, medium-high
DE LA TORRE et al.: TRANSMISSION EXPANSION PLANNING IN ELECTRICITY MARKETS 243

TABLE I From Table III, it is clear that most of the generation is located
LINE STRUCTURE FOR THE FIRST CASE STUDY at the initially isolated node 6; hence, transmission expansion
plans will probably tend to construct lines connecting node 6 to
the rest of the system.
The solution obtained is described next: three new lines are
proposed to be built, having a total annualized cost of $9M. Out
of the three new lines built, two of them connect node 6 with
node 2 and the remaining one connects node 6 to node 4; this
solution allows the energy produced at node 6 to flow to all the
consumers in the system.
The solution obtained is analyzed below from the viewpoint
of the generators: the generators annual cost is $53.9M; on the
other hand, their total yearly revenue is $79.4M, hence obtaining
a total $25.5M annual profit. These results mean that 32% of the
generators income becomes profit.
From the consumer viewpoint, the solution obtained implies
total annual payments for energy of $88.5M, and demand utility
of $116.5M, which implies a consumer surplus of $28.0M.
TABLE II These results mean that the utility that consumers obtain is 31%
CHARACTERISTICS OF THE DIFFERENT SCENARIOS. FIRST CASE STUDY higher than their payments; i.e., they obtain a 31% profit.
Finally, from the merchandising surplus viewpoint, a total
surplus of $9.1M is obtained; as already stated, this surplus de-
rives from the fact that the market price is not uniform for the
whole system, and the demands have to pay more for energy
than the generators receive for producing it. Of course, this is a
consequence of locational marginal pricing.
TABLE III Summing up all the surpluses, and subtracting the amount
GENERATORS AND DEMANDS LOCATION. FIRST CASE STUDY
needed to built new lines, the annual social welfare results in a
total $53.6M. Moreover, if the new lines were built by the gen-
erators, the $9M invested in new lines would represent 11.3%
of the generators revenue, or 35.3% of their profit.
Table IV(a)–(c) shows relevant data that provide insights to
characterize the features of the optimal solution found. Note
that the social welfare values in Table IV(c) do not include the
annualized investment cost in new lines, since that cost cannot
be split per scenario.
From Table IV(a)–(c), it can be shown that scenario 4 (that
only represents 10% of the hours; see Table II) is very important,
because it provides high prices and high merchandising surplus
with almost the same amount of energy produced, as compared
to scenario 3. This means that in scenario 4, the system is under
congestion and this fact forces the prices up. Also note that the
losses, in percentage, are very similar for the four scenarios con-
sidered. Table IV(c) shows that the producer and merchandising
demand, and high demand, respectively. Table II presents the surpluses increase as the demand increases, due to higher av-
weights and relative demands for the four scenarios considered. erage nodal prices well above the generators offer prices, and
Scenario 1 has a weight of 0.412; hence, it represents 41.2% of higher levels of congestion, respectively. On the contrary, the
the hours of a year; its demand coefficient is 0.47, which means consumer surplus first increases, since the average nodal prices
that demand for that scenario is 47% of the preselected refer- are well below the demands bid prices, but decreases later, since
ence demand level. Table III provides the location of generators higher levels of congestion imply that prices are much closer to
and demands in the network along with other relevant informa- the demands bid prices.
tion. For the generators, the maximum power production and Table V shows how several generation and transmission
the offer price is shown. For the demands, the total amount of limits are reached for each of the scenarios with the proposed
energy bid in the market is shown. solution (recall from Table II that scenario 4 is the scenario
This total amount of energy demanded is divided in five with the highest demand).
blocks of equal size with the five prices also shown in the table. As previously stated, the system is highly congested in sce-
Note that, using this information about the bids made by the nario 4, for which four lines and five generators are at maximum
demands, their utility functions could be constructed. capacity.
244 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

TABLE IV
(a) SOLUTION FOR THE FIRST CASE STUDY. PRODUCTION, CONSUMPTION AND
LOSSES. (b) SOLUTION FOR THE FIRST CASE STUDY. PRICES. (c) SOLUTION FOR
THE FIRST CASE STUDY. SURPLUSES AND SOCIAL WELFARE

(a)

(b)

Fig. 2. Second case study: diagram for the 24-bus system with the addition of
the eight new lines proposed.
(c)
TABLE V
LIMITS REACHED FOR THE SOLUTION OBTAINED disregarded in this case study. The case study presents the fol-
lowing characteristics: the system considered contains 24 nodes
and 34 lines connecting them; in this case, for the sake of sim-
plicity, new lines can only be constructed in parallel to existing
lines, up to a total number of four, i.e., only new lines for old
corridors are allowed. The market structure of the system con-
sidered consists of 11 generating units and 17 loads.
Table VII provides the location of generators and demands
TABLE VI
METRICS FOR THE FIRST CASE STUDY in the network along with other relevant information. For the
generators, the maximum power production and the energy of-
fers are shown. Note that in this case study, four blocks are used
to represent the offers of the generators; for simplicity, each of
these four blocks represents 25% of the capacity of each gen-
erator. For each demand, the total amount of the energy bid in
The metrics presented in Section IV have been calculated for the market is provided; this total amount of energy demanded is
this case study. Table VI provides the results. divided into three blocks of equal size with the three prices also
From Table VI, note that for each dollar invested in new lines, provided in the table.
a total $2.84 is obtained as social welfare for the participants From Table VII, it is clear that both generation and consump-
in the market. It can be concluded that this investment will be tion are evenly distributed throughout the network.
very profitable for the system. Particularly, the increase in social One hundred different scenarios are considered in this case
welfare is divided as follows: $0.51 goes to the generators, $1.91 study to describe the behavior of the demand. For the sake of
goes to the consumers, and $0.42 goes to the network operator. simplicity, all the scenarios have the same weight: 1%. The ac-
A total computing time of 1.46 s is needed to solve the above tual procedure used to obtain the different scenarios used for this
problem under a Linux-based server with four Xeon processors case study is described next: firstly, based on real data, we ob-
clocking at 1.60 GHz and 2 GB of RAM. The software used is tain a demand forecast for each and every hour of the target year;
CPLEX under GAMS [24]. then, we sort these demand values in decreasing order; next, we
divide the previously formed decreasing curve into 100 different
C. Second Case Study portions, one for each scenario; finally, we calculate for each of
The second case study is based on a system similar to the one these portions the average amount of energy demanded; demand
described in [19] and with the topology shown in Fig. 2. Note coefficients can now be calculated for each scenario dividing its
that the original parallel lines considered in [19] are initially average demand by the reference level of demand. In this way,
DE LA TORRE et al.: TRANSMISSION EXPANSION PLANNING IN ELECTRICITY MARKETS 245

TABLE VII TABLE VIII

GENERATORS AND DEMANDS LOCATION. SECOND CASE STUDY (a) SOLUTION FOR THE SECOND CASE STUDY. PRODUCTION, CONSUMPTION,
AND LOSSES. (b) SOLUTION FOR THE SECOND CASE STUDY. PRICES. (c)
SOLUTION FOR THE SECOND CASE STUDY. SURPLUSES AND SOCIAL WELFARE

(a)

we obtain 100 different scenarios, all of which are based on real

data and that represent in detail the demand in the system. In the
case study presented here, the maximum demand coefficient is
2.20. The energy demanded at each node for each scenario is (b)
calculated multiplying the MW demand provided in Table VII
by the coefficient associated to the scenario.
The cost of construction of new lines is as follows: we assume
for simplicity that, on average, lines connecting two nodes in the
lower half of the system have an annualized cost of $2M per line,
lines connecting two nodes in the upper part of the system have
an annualized cost of $4M per line, and lines connecting the
upper and the lower parts, i.e., new lines between 3–24, 9–11,
9–12, 10–11, and 10–12, are not allowed due to their high costs.
Note that the annualized cost values are equivalent to 10% of the
total construction costs.
The solution obtained for the case study presented is de-
scribed next. Eight new lines are proposed to be built: three
lines connect nodes 1 and 5; two lines are built between nodes 2
and 4; two lines between nodes 20 and 23; and finally one line is
built between nodes 2 and 6, thus increasing the transportation (c)
capacity from the large generators at nodes 1, 2, and 23. The
total annualized cost is $20M.
Summing up the surpluses obtained by the generators, the all the scenarios considered. Table VIII(c) shows that the pro-
consumers and the network planner and subtracting the amount ducer and merchandising surpluses increase as the demand in-
needed to built new lines, the annual social welfare results in a creases, due to higher average nodal prices and higher levels of
total $200.51M. congestion, respectively. Note that for very high demand, mer-
Table VIII(a) and (b) shows the optimal solution found for chandising surplus is slightly reduced, due to very high satu-
selected scenarios. ration with nodal prices reaching their maximum values (see
From Table VIII(a)–(c), it can be observed that the last sce- Table VIII(b) allowing cheaper generators to enter the auction.
narios are important, because of their high prices and high mer- However, the consumer surplus first increases, but decreases
chandising surplus. This means that system congestions make later with some slight oscillation, due to the combined effects of
prices increase. Also note that the losses are very similar in congestion and demand bidding (see the two different demand
246 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

TABLE IX demand is more expensive; hence, it is difficult to obtain the net

OVERALL METRICS FOR THE SECOND CASE STUDY. social welfare that pays for ambitious network expansion plans.
GENERATORS AND DEMANDS EVENLY DISTRIBUTED
For this modified case study, the total computing time is
14.6 min.

E. Analysis of the Effect of the Parameter

In order to establish the importance of the weighting param-
TABLE X eter from (1), some variations on the second case study (see
METRICS FOR THE SECOND CASE STUDY.
GENERATORS AND DEMANDS NOT EVENLY DISTRIBUTED Section V-C) are performed. These tests prove that the election of
this weighting parameter is important, as small variations have
a significant impact on the number of lines to be constructed.
Initially, the value of the weighting factor is set to 1000/8760.4 For
a first test, the value of is increased to 2000/8760. Under these
conditions, the simulation results change dramatically, and no
new lines are proposed. In other words, for , no
bidding prices in Table VII). Recall that the social welfare values
social welfare increase can be attained by constructing new lines.
in Table VIII(c) do not include the annualized investment cost
On the other hand, a second test is run reducing the value of to
in new lines, since that cost cannot be split per scenario. 500/8760; in this test, the number of new lines proposed increases
Table IX provides the metrics defined in Section III for this from 8 to 13, but the costs are reduced from $20M to $17M.3 In
case study. this case, the lines to be built are the same eight lines of the base
For this case study, for each dollar invested in new lines, a case plus another five lines: one more line connecting nodes 2 and
total $1.49 is recovered as net social welfare; particularly, gen- 4, two more lines connecting nodes 2 and 6, one line between
erators increase their profits $0.73 for each dollar invested in nodes 17 and 22, and one line between nodes 18 and 21.
new lines; similarly, consumers benefit by earning $0.29 more From an economic point of view, the increment of the pa-
and the system operator by earning $0.47 more. rameter can be seen as an increase in the cost associated to the
For this second case study, a total computing time of 44 min construction of new lines, and conversely, a decrease in its value
is needed to solve the problem. can be regarded as a decrease in line construction costs. Also,
this analysis allows for the heuristic determination of a priority
D. Non-Homogeneous Generation and Demand Distribution list for line construction, i.e., if is initially set to a high value
A new case study is constructed based on the previous one by and then it is smoothly decreased, prospective lines are selected,
changing the location of some generators and demands. For this one at a time, in an orderly manner, according to their economic
system, the upper portion can be considered a generation area importance. In other words, there must be a value of for which
(area above nodes 11-12-24 in Fig. 2, all included) and the rest only one line is proposed to be built, this line can be regarded
can be considered mainly as a consumption area. as the most profitable line, and hence, the first line that should
The changes made to obtain the new system are described be funded.
next. Demand at D2 is increased by 300 MW and at D6 by
400 MW; demand is decreased at D15 by 300 MW and at D17 V. CONCLUSION
by 400 MW. Generation at G1 is decreased by 200 MW, at G2 by We have presented a mixed-integer formulation of the trans-
200 MW, and at G3 by 200 MW. Finally, generation at G8 is in- mission expansion planning problem in pool-based electricity
creased by 100 MW, at G9 by 300 MW, and at G10 by 200 MW. markets. We have set the problem considering investment costs
The solution obtained for this new case study is described and weighting the aggregate social welfare according to the dif-
next: only two new lines are proposed to be built, having a total ferent levels of demand scenarios that can occur. Different eco-
annualized cost of $8M. These new lines connect nodes 12 and nomic indicators are used to appraise the quality of the transmis-
23 and nodes 20 and 23, respectively. Note that these lines are sion expansion plans proposed by the algorithm. Results from
built to help the transfer of energy generated at nodes 18, 21, 22, different case studies show that changes in the topology, sce-
and 23 to the consumption nodes in the lower part of the system. narios, and the weighting factor relating investment and oper-
Table X shows the values for the metrics defined in Section III. ating costs have appropriate economic interpretations for the
In this case, note that, for each dollar invested in new lines, generators, demands, and the network planner.
$1.19 are obtained as net social welfare. This implies that in
the modified case study, new lines produce less benefit. Consid- APPENDIX
ering that the natural tendency for a system like this should be
to invest in new lines to strengthen the interconnection between In this Appendix we present a multi-year formulation of the
its two areas, the solution obtained is counter-intuitive. That is, problem presented in the paper. This multi-year model has not
although the modified system is in a greater need of intercon- been used in the simulations and is presented only to illustrate
necting lines to serve the demands, the solution of the problem to the readers about its complexity and difficulties.
shows that not many lines are constructed. The explanation is 3Note that in this test, lines are considered to be 50% cheaper to build because
that the modified system has a generation and demand pattern parameter  has been divided by 2.
very unevenly distributed; therefore, connecting generation and 4Note that 8760 is the number of hours in a year and 1000 is a scaling factor.
DE LA TORRE et al.: TRANSMISSION EXPANSION PLANNING IN ELECTRICITY MARKETS 247

In the multi-year model, we have considered that the sce- (51)

narios are now defined per year, and consequently, the variables (52)
related to scenarios are now defined per scenario per year. Note
that we have dropped the weighting factor and we have in-
cluded a new constraint (52) to account for the fact that one line where
can only be built once in the entire multi-year period of study.
The general formulation of the multi-year model is as follows: current year of the period of study;
base year;
discount rate;
Maximize
binary variable: if line from corridor
is built in year of the study period;
if not;
set of all years of the period of study.

REFERENCES
[1] J. Rosellón, “Different approaches towards electricity transmission ex-
pansion,” Rev. Network Econ., vol. 2, no. 3, pp. 238–269, Sep. 2003.
[2] L. L. Garver, “Transmission network estimation using linear pro-
gramming,” IEEE Trans. Power App. Syst, vol. PAS-89, no. 7, pp.
1688–1697, Sep./Oct. 1970.
[3] R. Villasana, L. L. Garver, and S. J. Salon, “Transmission network plan-
(39) ning using linear programming,” IEEE Trans. Power App. Syst., vol.
PAS-104, no. 1, pp. 349–356, Feb. 1985.
[4] R. Romero and A. Monticelli, “A hierarchical decomposition approach
subject to for transmission network expansion planning,” IEEE Trans. Power
Syst., vol. 9, no. 1, pp. 373–380, Feb. 1994.
[5] N. Alguacil, A. L. Motto, and A. J. Conejo, “Transmission expansion
planning: A mixed-integer LP approach,” IEEE Trans. Power Syst., vol.
18, no. 3, pp. 1070–1076, Aug. 2003.
(40) [6] S. Binato, M. V. F. Pereira, and S. Granville, “A new Benders decom-
position approach to solve power transmission network design prob-
lems,” IEEE Trans. Power Syst., vol. 16, no. 2, pp. 235–240, May 2001.
[7] Y. P. Dusonchet and A. H. El-Abiad, “Transmission planning using
discrete dynamic optimization,” IEEE Trans. Power App. Syst., vol.
PAS-92, no. 2, pp. 1358–1371, Apr. 1973.
[8] E. J. de Oliveira, I. C. da Silva, J. L. R. Pereira, and S. Carneiro, Jr.,
“Transmission system expansion planning using a sigmoid function to
handle integer investment variables,” IEEE Trans. Power Syst., vol. 20,
(41) no. 3, pp. 1616–1621, Aug. 2005.
[9] R. A. Gallego, A. Monticelli, and R. Romero, “Transmission system
expansion planning by extended genetic algorithm,” Proc. Inst. Elect.
Eng., Gen., Transm., Distrib., vol. 145, no. 2, pp. 329–335, May 1998.
(42) [10] R. Romero, R. A. Gallego, and A. Monticelli, “Transmission system
expansion planning by simulated annealing,” IEEE Trans. Power Syst.,
vol. 11, no. 1, pp. 364–369, Feb. 1996.
[11] J. Contreras and F. F. Wu, “Coalition formation in transmission expan-
(43) sion planning,” IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1144–1152,
Aug. 1999.
[12] J. Contreras and F. F. Wu, “A kernel-oriented coalition formation algo-
rithm for transmission expansion planning,” IEEE Trans. Power Syst.,
(44) vol. 15, no. 4, pp. 919–925, Nov. 2000.
[13] J. M. Zolezzi and H. Rudnick, “Transmission cost allocation by coop-
erative games and coalition formation,” IEEE Trans. Power Syst., vol.
17, no. 4, pp. 1008–1015, Nov. 2002.
(45) [14] G. B. Shrestha and P. A. J. Fonseka, “Congestion-driven transmission
expansion in competitive power markets,” IEEE Trans. Power Syst.,
vol. 19, no. 3, pp. 1658–1665, Aug. 2004.
[15] J. Contreras, V. Bósquez, and G. Gross, “A framework for the analysis
(46) of transmission planning and investment,” in Proc. 15th Power Systems
Computation Conf., Liége, Belgium, Aug. 22-26, 2005.
[16] E. Castillo, A. J. Conejo, P. Pedregal, R. García, and N. Alguacil,
Building and Solving Mathematical Programming Models in Engi-
neering and Science. New York: Wiley, 2001.
(47) [17] S. de la Torre, A. J. Conejo, and J. Contreras, “Simulating oligopolistic
pool-based electricity markets: A multiperiod approach,” IEEE Trans.
Power Syst., vol. 18, no. 4, pp. 1547–1555, Nov. 2003.
(48) [18] A. L. Motto, F. D. Galiana, A. J. Conejo, and J. M. Arroyo, “Network-
constrained multi-period auction for a pool-based electricity market,”
IEEE Trans. Power Syst., vol. 17, no. 3, pp. 646–653, Aug. 2002.
[19] Reliability Test System Task Force, “The IEEE Reliability Test
System—1996,” IEEE Trans. Power Syst., vol. 14, no. 3, pp.
(49) 1010–1020, Aug. 1999.
[20] OMEL, Operadora del Mercado Ibérico de Energía—Polo Español,
(50) S.A. [Online]. Available: http://www.omel.es. available online
248 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 1, FEBRUARY 2008

[21] G. C. Oliveira, A. P. C. Costa, and S. Binato, “Large scale transmission Antonio J. Conejo (F’04) received the M.S. de-
network planning using optimization and heuristic techniques,” IEEE gree from Massachusetts Institute of Technology,
Trans. Power Syst., vol. 10, no. 4, pp. 1828–1834, Nov. 1995. Cambridge, in 1987 and the Ph.D. degree from the
[22] G. Latorre, R. D. Cruz, J. M. Areiza, and A. Villegas, “Classification Royal Institute of Technology, Stockholm, Sweden,
of publications and models on transmission expansion planning,” IEEE in 1990.
Trans. Power Syst., vol. 18, no. 2, pp. 938–946, May 2003. He is currently a Professor of electrical engi-
[23] I. J. Ramírez-Rosado and T. Gönen, “Pseudodynamic planning for ex- neering at the Universidad de Castilla–La Mancha,
pansion of power distribution systems,” IEEE Trans. Power Syst., vol. Ciudad Real, Spain. His research interests include
6, no. 1, pp. 245–254, Feb. 1991. control, operations, planning and economics of
[24] A. Brooke, D. Kendrick, A. Meeraus, and R. Raman, GAMS/CPLEX electric energy systems, as well as statistics and
9.0. User Notes. Washington, DC, 2003, GAMS Development Corp. optimization theory and its applications.

Sebastián de la Torre (M’04) was born in Málaga,

Spain. He received the Ingeniero Industrial degree Javier Contreras (SM’05) received the B.S. degree
from the Universidad de Málaga, Málaga, Spain, in in electrical engineering from the University of
1999 and the Ph.D. degree in power systems oper- Zaragoza, Zaragoza, Spain, in 1989, the M.Sc.
ations planning at the Universidad de Castilla–La degree from the University of Southern California,
Mancha, Ciudad Real, Spain, in 2003. LosAngeles, in 1992, and the Ph.D. degree from the
His research interests include operations and eco- University of California, Berkeley, in 1997.
nomics of electrical energy systems, restructuring of His research interests include power systems
electric systems, and development of new electricity planning, operations and economics, and electricity
markets. He is currently an Assistant Professor in the markets. He is currently an Associate Professor at
Department of Electronic Engineering at the Univer- the University of Castilla–La Mancha, Ciudad Real,
sidad de Málaga. Spain.