Electric Power Systems Research 109 (2014) 141149

Contents lists available at ScienceDirect

Electric Power Systems Research

journal homepage: www.elsevier.com/locate/epsr

Optimal bidding strategy in transmission-constrained electricity

markets
Evaggelos G. Kardakos, Christos K. Simoglou, Anastasios G. Bakirtzis
Department of Electrical Engineering, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece

a r t i c l e i n f o a b s t r a c t

Article history: This paper addresses the problem of developing an optimal bidding strategy for a strategic producer
Received 29 April 2013 in a transmission-constrained day-ahead electricity market. The optimal bidding strategy is formulated
Received in revised form as a bi-level optimization problem, where the rst level represents the producer prot maximization
23 September 2013
and the second level represents the ISO market clearing. The transmission network is incorporated into
Accepted 23 December 2013
the ISO problem under two different approaches based on the Nodal and PTDF formulation, respec-
Available online 22 January 2014
tively. The bi-level problem is converted to a mathematical program with equilibrium constraints
(MPEC) which, in turn, is transformed into a mixed-integer linear programming (MILP) model using
Keywords:
Electricity market
the KarushKuhnTucker (KKT) optimality conditions and the strong duality theory. Test results on the
Mathematical program with equilibrium IEEE 24-bus and 118-bus systems show that the PTDF formulation of the transmission constraints is
constraints (MPEC) computationally far more efcient than the Nodal formulation.
Mixed-integer linear program (MILP) 2013 Elsevier B.V. All rights reserved.
Optimal bidding strategies
Transmission network

1. Introduction In recent years many optimization models have been presented

in the literature for the solution of the producer optimal bidding
The day-ahead electricity market is the most common form of strategy problem. These models can be categorized according to the
centralized daily transactions in competitive electricity markets. In extent that the strategic producer takes into account the behavior
such markets, the electricity producers face every day the funda- of his rivals. In this sense, there are two main modeling direc-
mental problem of how to bid into the market so as to maximize tions, namely (a) the single rm optimization models, which ignore
their expected prots. the strategic interaction of the market players and can be further
For the construction of his optimal bidding strategy, the pro- divided into price-taker [19] and price-maker [1022] producer
ducer must formulate and solve an appropriate optimization model models and (b) the market equilibrium models, which take into
taking into account various techno-economic data and operating account the strategic interaction of all players, i.e. their reaction to
constraints of his generating units, such as start-up and shut-down the decisions of their rivals [2330].
costs, ramp-rate limits, minimum-up/down time constraints, start- Although various models including detailed description of the
up and shut-down procedures, etc. Additionally, the producer must operating constraints of the generating units have been proposed
take into account the behavior of his competitors, as well as specic in the literature, only a few have considered the effect of the trans-
characteristics and enacted rules of the electricity market. In this mission network on the producer bidding strategy and on market
context, the transmission network plays an important role in an clearing. A single-rm optimization model, formulated as an MPEC,
electricity market, since it can have a notable impact on the daily is presented in [23], where a linearized DC optimal power ow
market operation. Although the transmission network is necessary (OPF) model is considered in the lower level and a penalty inte-
for the proper formulation of the producer optimal bidding strategy rior point algorithm is used to compute a local optimal solution of
problem, many challenges arise owing to its modeling complexity. the MPEC. In this model, a single-period modeling is adopted, thus
ignoring all inter-temporal and commitment constraints related to
the producer units realistic operation. In addition, the use of the
penalty interior point algorithm cannot guarantee that the global
optimal solution of the MPEC has been found. The authors of [14]
Corresponding author at: Power Systems Laboratory, Division of Electrical propose a binary expansion (BE) solution approach to the MPEC
Energy, Department of Electrical Engineering, Aristotle University of Thessaloniki, problem in order to transform it into an MILP model. They describe
AUTh Campus, 54124 Thessaloniki, Greece. Tel.: +30 2310 996383; fax: +30 2310
996302.
how the transmission network constraints can be incorporated into
E-mail address: [email protected] (A.G. Bakirtzis). the problem, without, however, providing further analysis or test

0378-7796/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.epsr.2013.12.014
142 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149

there is no guarantee that the algorithm converges to the global

Nomenclature optimum. A network-constrained market-clearing algorithm is
considered in [17], where the authors proposed an MPEC model
(L) index (set) of branches that is subsequently reduced to an MILP model using the duality
d(D) index (set) of loads theory and the KarushKuhnTucker (KKT) rst-order optimality
f(Fi ) index (set) of steps of the offer curve of unit i conditions. The authors incorporated a DC power ow modeling
i(I) index (set) of generating units I = IS IR for the transmission network representation under the Nodal
Is set of generating units of the strategic producer formulation. In addition, although a 24-h scheduling horizon
IR set of generating units of the rivals (non-strategic is considered, only simple unit inter-temporal constraints (i.e.
producers) ramp-rate limits) are modeled. In [18] the producer prot in a
n(N) index (set) of buses network-constrained electricity market is optimized based on the
ni /nd bus at which generator i/load d is connected transmission-constrained residual demand derivative (TCRDD),
t(T) index (set) of hours of the planning horizon thus avoiding the representation of the full network model. A lim-
n set of buses connected to bus n via branches itation of the proposed model is that all the units of the strategic
n set of units located on bus n producer must be located on the same bus. In addition, a local
n set of loads located on bus n search algorithm is used, which cannot guarantee convergence
r reference (slack) bus index r N to the global optimum. Finally, single-period modeling is used,
cit (qit ) total hourly production cost of unit i in hour t at level ignoring the inter-temporal and unit commitment constraints.
qit , in $/h To sum up, the models of [1416,18,23] are formulated for
Anr power transfer distribution factor (PTDF) of branch single-period time intervals, ignoring the inter-temporal and unit
 for power transfer from node n to node r commitment constraints for the strategic producer units modeling.
B susceptance of branch  connecting node n and node A 24-h scheduling horizon with only simple inter-temporal con-
m B = Bnm straints (i.e. unit ramp-rate limits) considered in the formulation of
Ldt load demand of load d in hour t, in MW the producers prot maximization problem (rst-level problem) is
Flmax capacity of branch  (connecting node n and node adopted in [17].
m), in MW (Flmax = Flnmmax )
In this paper a bi-level model is proposed for the development
Qimax maximum power output of unit i, in MW of the optimal bidding strategy of a strategic producer, who owns a
SDCi shut-down cost of unit i, in $ number of generating units located on different nodes throughout
SUCi start-up cost of unit i from standby until load with an electricity network and is able to manipulate the market clearing
synchronization, in $ prices to his own benet through his offering strategy. The bi-level
pift offer price of step f of the ith units offer curve in model consists of a rst level, which represents the producer prot
hour t, in $/MWh maximization problem and a second level, which represents the
Qift offer quantity of step f of the ith units offer curve in ISO market clearing problem, as in [1417].
hour t, in MW In the rst-level a detailed unit-commitment model is used
qift portion of step f of the ith units offer curve accepted for the strategic producers units, while in the second-level it is
by the ISO in hour t in the day-ahead energy auction, assumed that the market clearing is performed by the ISO on an
in MW hour-by-hour basis ignoring the associated unit commitment con-
uit binary variable which is equal to 1 if unit i is com- straints, so that the second-level problem is convex and can thus
mitted during hour t be replaced by its rst-order KKT optimality conditions, yielding a
yit binary variable which is equal to 1 if unit i is started- mathematical program with equilibrium constraints (MPEC). Sub-
up during hour t sequently, the MPEC is transformed into an MILP model using the
zit binary variable which is equal to 1 if unit i is shut- methodology proposed in [17]. The transmission network model-
down during hour t ing is incorporated into the ISO market clearing algorithm under
nt voltage angle of bus n in hour t, in rad two different approaches, the Nodal and the PTDF (preferred) for-
nt locational marginal price of node n in hour t, in mulation.
$/MWh The main contributions of this paper are threefold:
i set dening the feasible operating region of unit i

results. An MPEC model that uses the same BE approach for the dis- An MPEC problem is formulated and solved for the determina-
cretization of the offer quantities and prices taking also into account tion of the optimal offering strategy of a strategic producer over
the effect of the transmission network on the producer strategic a 24-h scheduling horizon. A detailed unit commitment modeling
behavior is analyzed in [15]. In both works [14,15], the proposed in the strategic producer prot maximization (rst-level) prob-
models were also formulated for single-period time intervals only. lem is used, in contrast to [17] that takes into account only the
Moreover, the proposed binary expansion methodology failed to capacity and the ramping limits of the generating units. In addi-
converge when tested to large test systems due to computational tion, the transmission network is effectively incorporated into the
issues. ISO (second-level) problem. The proposed MPEC model is effec-
The transmission network is also taken into account in [16], tively transformed into an MILP model, which is solved using a
where the proposed model is based on the bi-level programming commercial MILP solver.
and a primal-dual interior point method (PDIPM) is developed for A PTDF modeling of the transmission network constraints is pro-
the solution of the ISO market clearing problem, which included posed and compared to the respective Nodal formulation used
a DC power ow. In addition, sensitivity functions are developed in [17]. Although PTDF modeling has been extensively used for
from linear programming for the modeling of the producers the formulation and solution of the standard unit commitment
payoffs with respect to bidding strategies to solve the bi-level problem solved by the ISO in pool-type day-ahead electricity mar-
problem. In this model, a single-period modeling is used, ignoring kets in order to clear the market [31], it is the rst time that
the inter-temporal and unit commitment constraints. In addition, this transmission modeling is used for the formulation of the
 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149 143

producer optimal bidding strategy problem as an MPEC problem (a) Nodal formulation
in transmission-constrained electricity markets.  
A theoretical analysis along with extensive simulations in two qift Ldt
different real test systems show that the PTDF formulation is far i n , f F i dn
more efcient than the Nodal formulation, especially when the 
size of the system and the number of the congested branches = Bnm (nt mt ) : nt n N, t T (6)
increase. In addition, the ability of the producer to take advantage mn
of the strategic position of his generating units in the network to
his own benet is properly demonstrated. max
Flnm max
Bnm (nt mt ) Flnm n N, m n , t T (7)

The paper is organized as follows. Section 3 describes the formu-  nt  n N, t T (8)

lation of the strategic producer prot maximization problem as a
bi-level problem and its transformation into an MILP model. Section rt = 0 t T (9)
4 provides numerical results from the application of the proposed
model to the IEEE 24-bus and 118-bus systems. Finally, conclusions (b) PTDF formulation
are drawn in Section 4. Appendix provides further details on the  
qift Ldt = 0 : rt t T (10)
conversion of the MPEC into a MILP model.
iI, f F i dD

2. Mathematical formulation
  
Flmax Anr qift Ldt
The optimal bidding strategy of a strategic producer (henceforth,
Producer) is mathematically formulated as a bi-level optimiza- nN in , f F i dn

tion problem, where the rst level represents the producer prot
Flmax : min , max  L, t T (11)
maximization problem and the second level represents the ISO t t
market clearing problem. The transmission network is included in
the ISO clearing problem under two different approaches based on 
Nodal and PTDF formulation, respectively. nt = rt Anr (max
t
min
t
) n N, t T (12)
L

2.1. The producer prot maximization bi-level problem

The dual variables of constraints (5), (6), (10) and (11) are
dened following a colon. The dual variables of constraints (7)(9)
  
Max ni t qift cit qift (1)
are not dened since the Nodal formulation has been presented in
[17].
iI S , tT f F i f F i
The decision variable of the proposed bi-level optimization
model is the quantityprice pair (Qift , pift ) for each generating unit of
Subject to the following two groups of constraints:
the Producer, which is computed so as to maximize the Producers
prot. The competitors energy offers (Qift , pift ) and the system load
i) Producer units constraints are considered known, based on deterministic estimates. Since the
 rivals multi-step offers and the system load are deterministically
Qift Qimax i I S , t T (2) known, a one-step offer curve is sufcient for the denition of
f F i the strategic behavior of the Producer (Fi = {1} , i IS ). The sin-
gle quantityprice pair (Qi1t , pi1t ), computed by the solution of the
optimization problem, denes the Producer one-step offer curve.
xit = [qit , uit , yit , zit , ...] i i I S , t T (3) The rst level optimization problem is dened by (1)(3). The
objective function (1) represents the Producer prot, during the
ii) Transmission-constrained ISO market clearing problem scheduling period, which consists of the revenue from the day-
ahead energy market minus the Producer units total production
 cost. The rst term of the difference in (1) is determined through
Min pift qift (4) the solution of the second level problem. The total production cost,
iI, f F i , tT
cit (qit ), includes (a) the hourly operating cost dened by the units
no-load cost, NLCi and the units step-wise marginal cost func-
Subject to: tion, (b) the units start-up cost, SUCi , and (c) the units shut-down
cost, SDCi . Constraints (2) restrict the offer quantity (one-step offer
min
0 qift Qift : ift max
, ift i I, f F i , t T (5) quantity, Qi1t ) of each Producer unit to its respective capacity. Con-
straints (3) represent the operating constraints of the Producer
The system constraints, comprising power balance and trans- units, i.e. start-up and shut-down procedures, minimum-up/down
mission constraints, can be formulated following two different time constraints, min/max power output restrictions, ramp-rate
approaches: the Nodal formulation [17] and the PTDF formulation1 limits, fuel limitations, etc., described in detail in [5].
proposed in our model. Both approaches have been developed in The second level problem regarding the ISO market clearing is
this paper, in order to compare their performance. determined by the objective function (4), constraints (5) and the
group of constraints (6)(9) or (10)(12) depending on whether the
Nodal or the PTDF modeling of the transmission network is used.
1
PTDF stands for power transfer distribution factor. In fact, PTDF is the fraction
Eq. (4) denotes the minimization of the total as-bid production
of the amount of a power transfer from one node to another that ows over a given cost. Eq. (5) reects the energy offer quantity constraints of the
transmission line [32]. generating units of the Producer and his rivals.
144 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149

The Nodal formulation of the system constraints comprises: power ow limits (7) and voltage phase angle limits (8), and 1
equality constraint for the denition of the voltage phase angle of
(a) The power balance constraint at every node (6), which enforces the reference bus (9). The number of the unknown variables in this
that the algebraic summation of the power injections minus formulation is I F + N, where I F is the number of the unknown
the load withdrawals on every node is equal to the algebraic variables qift denoting the producer optimal offer quantities and
summation of the power ows on the branches connected to N is the number of the unknown variables nt related to the volt-
the respective node; the corresponding Lagrange multipliers age phase angles during hour t. On the other hand, for every hour
nt denote the wholesale electricity price (marginal electricity t the PTDF formulation requires only 1 system-wide power bal-
cost) at node n (during hour t). ance equality constraint (10) and L two-sided branch power ow
(b) The capacity limits of each branch (7), which ensure that the inequality constraints (11). Constraint (12) is used ex-post for the
power ow of each branch does not exceed its respective capac- calculation of the resulting LMPs in each node of the system, once
ity limits. the optimization problem has been solved. The number of unknown
(c) The upper and lower bound for the voltage phase angle of each variables in this formulation is restricted to I F denoting the pro-
bus (8). ducer optimal offer quantities qift .
(d) The requirement that the voltage phase angle of the system Finally, the numerous binary variables needed in the Nodal
reference (slack) node be equal to zero (9). formulation for the linearization of the non-linear complementar-
ity constraints of the two-sided inequalities on the voltage phase
The PTDF formulation comprises: angles (8) are avoided in the PTDF formulation, making the latter
far more efcient.
(a) The system-wide power balance constraint (10), which in this
case is enforced system-wide (one constraint for the entire sys-
2.2. MILP formulation of the producer prot maximization
tem) and not separately for each node. This equation enforces
bi-level problem
that the algebraic summation of the power injections minus
the load withdrawals of all nodes must be equal to zero. The
A common method to address a bi-level optimization problem is
dual variable associated with the power balance equation, rt ,
to transform it into a single-level problem. A well-known strategy
is the system marginal price. It represents the increase in the
that is followed in optimization theory is to replace the lower-level
hourly system operating cost ($/h) for the supply of an addi-
problem (in this case the ISO market clearing problem) with its
tional MW of load demand. Since the increase in load demand
KKT rst-order optimality conditions. This is allowed mathemat-
will be supplied by the marginal generator, located at the sys-
ically, since the lower-level problem is linear and, hence, convex.
tem reference bus, the system marginal price is equal to the
The KKT rst-order optimality conditions of the ISO market clear-
reference bus marginal price rt of the Nodal formulation.
ing problem are presented in detail in Appendix, only for the PTDF
(b) The capacity limits of each branch (11), which ensure that the
formulation, since a detailed representation of the resulting MPEC
power ow of each branch does not exceed its respective capac-
model for the Nodal formulation is presented in [17].
ity limit. The dual variables associated with the upper and lower
After replacing the lower-level problem by its KKT rst-order
transmission line ow limits max t
, min
t
represent the trans-
optimality conditions, the MPEC problem is formulated, which con-
mission capacity scarcity premium (in either direction of ow).
sists of:
(c) Constraints (12), which dene the locational marginal price
(LMP) of bus n in hour t, in terms of the second level problem
dual variables rt , max
t
, min
t
[33]. (i) The Producer units constraints (1)(3), which constitute the
upper-level problem.
In the Nodal formulation (6)(9) for every time interval t T the (ii) The KKT optimality conditions (13)(18) of the lower-level
nodal voltage phase angles nt are explicitly modeled and an active problem, which are analyzed in detail in Appendix.
power balance equation is written for every node of the system in (iii) Eq. (12) which determines LMP of bus n in hour t, which is com-
(6); the corresponding Lagrange multipliers nt denote the whole- puted by post-processing of the second-level problem solution.
sale electricity price (marginal electricity cost) at node n (during
hour t). Since the nodal voltage phase angles are explicitly modeled,
The resulting MPEC is a non-linear and non-convex optimization
the branch power ows are expressed in terms of the phase angles
problem that is difcult to solve. The nonlinearities of the MPEC
in the two-sided branch ow constraints (7).
are restricted to the existence of the product of variables ni t qift
In the PTDF formulation (10)-(11) for every time interval t T
in the objective function (1) and the complementarity conditions
the nodal voltage phase angles are eliminated from (6) and (7)
(15)(18).
resulting in a single, system-wide power balance Equation (10),
In [17] the authors proposed an efcient method to nd a linear
where Lagrange multiplier rt is the system marginal price; the
expression for the term ni t qift , based on the KKT conditions and
branch power ows are now expressed in terms of the nodal
the strong duality theory. Based on this method, a linear expression
injections (nodal supply minus demand), multiplied by the cor-
for the term ni t qift is derived, as shown in Appendix, for the PTDF
responding PTDFs in the transmission constraints (11). The two
formulation only, since a similar analysis for the Nodal formulation
formulations are equivalent. Since in the PTDF formulation only a
has already been presented in [17].
single system-wide electricity price, rt , is computed from the sin-
The non-linearities arising due to the complementarity condi-
gle power balance constraint (10), the LMPs in the PTDF formulation
tions are linearized based on the linear expressions proposed in
are computed by post-processing of the second level problem solu-
[34]. A detailed description of the resulting linearizations is also
tion, using (12) which expresses the LMP of node n (during hour t) in
presented in Appendix.
terms of the reference price rt , the PTDFs Anr and the transmission
Using these linearizations the MPEC is converted to an MILP
capacity scarcity premiums (max min ).
t t model, which consists of:
Regarding the computational complexity of the two formula-
tions, we note the following: For every hour t, the Nodal formulation
comprises N nodal power balance equality constraints (6), (L + N) (i) The Producer units constraints (1)(3), which constitute the
two-sided inequality constraints for the enforcement of branch upper-level problem.
 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149 145

Table 1 set equal to 95% of their respective maximum power ow observed

Location of strategic producer units.
during the 24-h scheduling period in the Base model solution. This
Test cases Strategic producer units: type (#bus number) results in the congestion of branches #19 and #27 during several
24.A U76 (#1), U76 (#1), U155 (#23), U155 (#23), U100 (#7), U100 hours and the total congestion-hours are equal to 6 or 7, depending
(#7) on whether the UC modeling is taken into account or not.
24.B, 24.C U76 (#1), U76 (#1), U76 (#2), U76 (#2) In the case that UC is ignored, Table 2 shows that all three models
118 U300 (#12), U350 (#26), U100 (#36), U100 (#92), U250 (#49), that take into account the network constraints provide identical
U80 (#70), U30 (#72)
results, but the execution time of the Nodal model is signicantly
higher compared to the other two network models (PTDF Reduced
and PTDF Full).
(ii) The KKT optimality conditions (13) and (14) of the lower-level
When the UC modeling is considered, Table 2 shows that the
problem, which are analyzed in detail in Appendix.
Nodal model was not able to provide a proven optimal solution
(iii) The linearizations of the complementarity conditions (27)(42),
even after a long execution time (the optimality gap is 3.49% after
which are analyzed in Appendix.
12 h). Both other network models provide proven optimal solutions
(iv) The linear expression (26) for the term ni t qift , which is calcu-
(i.e. optimality gap = 0%) in considerably lower execution times
lated in detail in Appendix.
(Table 2).
(v) Eq. (12) which determines LMP of bus n in hour t, which is com-
These results demonstrate that the PTDF formulation is far more
puted by post-processing of the second level problem solution.
efcient than the respective Nodal formulation when UC model-
ing is considered. This is due to the fact that the numerous binary
3. Test results variables used in the Nodal formulation for the linearization of the
complementarity constraints of the two sided inequalities (8) on
The proposed model has been tested on the IEEE 24-bus system the voltage phase angles [17] are not needed in the PTDF formu-
and a modied IEEE 118-bus system. Network, unit and demand lation, making the latter far more efcient. In addition, the PTDF
data are given for each test system in [35,36], respectively. Three Reduced model is considerably faster than the PTDF Full model.
different test cases have been executed for the IEEE 24-bus system, In case 24.A (without UC) the Producer can exploit the network
proving the efciency of the PTDF formulation as compared to the constraints to his benet and increases his prot from $923,723.8 to
Nodal formulation, especially when the size of the system and the $925,109.9. The modeling of UC constraints however, limits the Pro-
number of the congested branches increase. An additional test on ducers exibility to exploit the network constraints and reduces
the IEEE 118-bus system, investigates the efciency of the proposed the prot of the Producer relative to the Base case from $909,793.8
model on larger systems. to $905,241.7. However, the incorporation of the network con-
The techno-economic data of the generating units of the 24-bus straints in the Producer offering strategy model limits the negative
system are calculated based on the unit data given in [35] and the effect of congestion on his prot. In this sense, Table 3 compares
publicly available fuel prices by the U.S. Energy Information Admin- the Producer unit outputs and prot for two cases: (a) the pro-
istration (EIA) [37], except for the nuclear fuel price, which is taken ducer considers the network constraints in his optimal offering
from the World Nuclear Association [38]. The techno-economic strategy problem (PTDF Reduced) and (b) the Producer neglects
data of the generating units of the 118-bus system are based on the network constrains in his optimal bidding strategy problem
the unit data of the 24-bus system. (Base offers & ISO-OPF). In both cases the ISO is assumed to clear
All tests were performed on a 3.2 GHz Intel Core i7 CPU with 24 the market by solving and optimal power ow (OPF) problem, i.e.
GB of RAM. The models were implemented in GAMS 23.7 using the respecting transmission constraints. In the latter case (Base offers
CPLEX 12.0 solver [39]. & ISO-OPF), the Producer daily prot is reduced by 1.95% as com-
For all test cases, four different models were executed, as fol- pared to the PTDF Reduced model demonstrating the benet of
lows: including the network constraints in the Producer bidding strategy
problem.
(a) The Base model, which ignores all network constraints.
(b) The PTDF Reduced model, which solves the network con-
strained producer bidding strategy problem in two steps: In 3.1.2. Test case 24.B
the rst step a Base model is solved, ignoring network con- In the second case, it is considered that the Producer owns 4
straints. In the second step all branches that were overloaded generating units located on nodes #1 and #2 (Table 1). A single
or nearly overloaded (e.g. loaded above 95% of their rating) in case with UC modeling of the Producer units is now considered.
the Base model solution are included in the Producer prot The rating of branches #20, #26, #30 and #33 is set equal to 95%
maximization problem using the PTDF formulation (10) and of their respective maximum power ow observed during the 24-h
(11). scheduling period in the Base model solution. This results in the
(c) The PTDF Full model, which includes the branch ow con- congestion of branches #26, #30 and #33. The total congestion-
straints of all branches, using the PTDF formulation (10)-(11). hours are 28. The large number of congestion-hours makes the
(d) The Nodal model, which includes the branch ow constraints problem difcult to solve, resulting in a signicant increase of the
of all branches, using the Nodal formulation (6)(9). execution time with respect to Case 24.A, especially in the PTDF
Full model (Table 2). The scope of this case is to show the ef-
3.1. IEEE 24-bus system ciency of the PTDF Reduced model as compared to both the PTDF
Full and the Nodal models, when the size of the problem (number
3.1.1. Test case 24.A of congestion-hours) increases.
In the rst test case, the Producer is supposed to solve his Similarly to the previous case, Table 4 shows that the incorpo-
prot maximization problem under two different considerations ration of the network constraints in the Producer offering strategy
regarding the unit commitment (UC) modeling of his own units: (a) model gives him the opportunity to increase his daily prot by
without UC constraints and (b) with UC constraints. The Producer 1.07% with respect to the case where the Producer offering strat-
owns 6 generating units located at different nodes of the electric- egy model neglects transmission constraints, yet the ISO market
ity network (Table 1). The rating of branches #19, #21, and #27 is clearing takes them into account (Base offers & ISO-OPF).
146 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149

Table 2
Summary test results.

Test case Model Total daily Prot ($) Execution Test case Model Total daily Prot ($) Execution
production time (s) production time (s)
(MW) (MW)

24.A without Base 7714.05 923,723.8 5.8 24.B with UC Base 4149.60 527,082.1 8.9
UC PTDF Reduced 7570.79 925,109.9 8.2 PTDF Reduced 4351.32 553,297.7 39.8
PTDF Full 7570.79 925,109.9 12.4 PTDF Full 4351.32 553,297.7 94 min
Nodal 7570.79 925,109.9 73.1 Nodal 4351.32 553,297.7 659 min
24.A with UC Base 7985.70 909,793.8 10.9 24.C with UC Base 4149.60 527,082.1 8.9
PTDF Reduced 7833.56 905,241.7 81.6 PTDF Reduced 4102.79 559,555.8 13.8
PTDF Full 7833.56 905,241.7 693.2 PTDF Full 4102.79 559,555.8 12.3
Nodal 7640.76 899,842.9 720 min(3.49% gap) Nodal 4102.79 559,555.8 302

Table 3
Case A - test results.

Test case Model U76 (#1) U76 (#1) U155 (#23) U155 (#23) U100 (#7) U100 (#7) Total Total prot
(MW) (MW) (MW) (MW) (MW) (MW) production ($)
(MW)

24.A with UC PTDF reduced 1086.26 1042.21 2865.34 2839.75 0.00 0.00 7833.56 905,241.7
Base offers & ISO-OPF 978.55 976.55 2901.50 2787.51 0.00 0.00 7644.11 887,508.9
Difference 107.71 65.66 36.16 52.24 0.00 0.00 189.45 17,732.8 (1.95%)

Table 4
Cases B & C test results.

Test Case Model U76 (#1) U76 (#1) U76 (#2) U76 (#2) Total Total prot
(MW) (MW) (MW) (MW) production ($)
(MW)

24.B with UC PTDF reduced 1048.80 1067.60 1094.92 1140.00 4351.32 553,297.7
Base offers & ISO-OPF 1068.80 1114.19 1081.20 912.00 4176.19 547,363.1
Difference 20.00 46.59 13.72 228.00 175.13 5,934.6 (1.07%)
24.C with UC PTDF reduced 1075.20 1094.40 1068.06 865.13 4102.79 559,555.8
Base offers & ISO-OPF 1046.00 1110.40 1081.20 912.00 4149.60 527,082.1
Difference 29.20 16.00 13.14 46.87 46.81 32,473.8 (5.80%)

3.1.3. Test case 24.C In this test case, it is considered that the Producer owns 7 gen-
In the third case, the same Producer with case 24.B is consid- erating units located on different nodes throughout the electricity
ered. In this case, the capacity of branch #10 is set slightly above its network (Table 1).
respective maximum power ow observed in the Base model solu- A single case, with UC modeling of the Producer units, is con-
tion, i.e. equal to 99 MW, while the capacity of all other branches sidered. The capacity of branches #54, #88, #133, #153, #167 and
remains unchanged. This case shows how the Producer can re- #176 is set equal to the 95% of their respective maximum power
dispatch his units so as to create congestion on branch #10 and ow observed during the 24-h scheduling period in the Base model
increase his prots. solution. The solution of the problem results in the congestion of
In this case, the incorporation of the network constraints in all these branches and the total congestion-hours are equal to 16.
the Producer offering strategy model gives him the opportunity to Table 5 shows that the solution times to reach a proven optimal
exercise locational market power. Similarly to the previous cases, solution of both the Base model and the PTDF Reduced model on
Table 4 shows that under the proposed strategy (PTDF Reduced) the the 118-bus system are considerably higher than the respective
Producer manages to cause congestion on branch #10 and increase
his daily prot by 5.8% with respect to the case where the Pro-
ducer offering strategy model neglects transmission constraints,
yet the ISO market clearing takes them into account (Base offers &
ISO-OPF).
Fig. 1 illustrates (a) the resulting SMP of the Base model, which is
identical for all buses, since congestion is neglected (network con-
straints are ignored) and (b) the LMP of buses #1 and #2, where the
strategic producers units are located, in the PTDF Reduced model.
It is evident from Fig. 1 that by congesting branch #10 during hours
19, 20, 21, the producer succeeds to increase the LMP of buses #1
and #2, where his units are located.

3.2. IEEE 118-bus system

Finally, the IEEE 118-bus system was used in order to examine

the computational efciency of the four different models on a larger
test system. Fig. 1. SMP/LMP in the Base model and the PTDF reduced model (case 24.C).
 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149 147

Table 5
IEEE 118-bus system summary test results.

Test case Model Total daily production (MW) Prot ($) Execution time (s)

118 with UC Base 17,650.00 489,761.9 68.6

PTDF reduced 17,789.74 529,390.4 13 min

ones on the 24-bus system. Both the other two network models
examined (PTDF Full and Nodal) failed to converge to an optimal  
solution, which shows their inability to deal with large test systems. qift Ldt = 0 t T (14)
iI, f F i dD

4. Discussion conclusion
max
0 qift ift 0 i I, f F i , t T (15)
This paper presented an MILP model for the solution of the
optimal bidding strategy problem for a strategic producer who max
0 Qift qift ift 0 i I, f F i , t T (16)
participates in a transmission-constrained day-ahead electricity
market. The transmission network was modeled under two dif-

ferent approaches, based on the Nodal and PTDF formulation,
  
respectively. 0 Flmax Anr qift Ldt max 0
Test results on the IEEE 24-bus system and a modied IEEE 118- t
nN in , f F i dn
bus system demonstrate the benets of incorporating the network
constraints in the Producer offering strategy model. In addition, the  L, t T (17)
proposed PTDF-based formulation of the transmission constraints
has proved computationally far more efcient than the Nodal for-
mulation, especially when the size of the system and the number

of congested branches increase.   
0 Flmax + Anr qift Ldt min 0
Regarding the applicability of the proposed method, it is rec- t
nN in , f F i dn
ognized that it suffers from the standard modeling limitations of
similar methods addressing the complex topic of developing opti-  L, t T (18)
mal bidding strategies for electricity producers. The main modeling
limitations are:
As already mentioned in Section 3.2, the resulting MPEC is a
non-linear and non-convex optimization problem. The nonlineari-
(a) The second-level ISO market clearing problem ignores the non-
ties of the MPEC are restricted to: (a) the existence of the product
convex unit commitment constraints. This modeling does not
of variables ni t qift in the objective function (1) and (b) the com-
allow for the exact formulation and solution of the ISO day-
plementarity conditions (15)(18).
ahead market clearing followed by most US ISOs/RTOs.
Based on the KKT conditions and the strong duality theory, a
(b) The ISO local market power mitigation (LMPM) procedures
linear expression for the term ni t qift can be produced [17].
incorporated in the day-ahead electricity markets of most US
According to the strong duality theory, the objective functions
ISOs/RTOs are ignored.
of the primal and dual problems have the same value in the opti-
(c) It is assumed that the producer has full knowledge of the trans-
mal solution, when the problem is convex. Therefore, the following
mission system as well as perfect information regarding his
equation arises:
rivals energy offers.

pift qift
Overcoming the above limitations will be the focus of our future
iI, f F i , tT
research.
  
max
= ift Qift max
t
(Flmax + And r Ldt )
Acknowledgements iI, f F i , tT L, tT dD
  
This work was nancially supported by the State Scholarships min
t
(Flmax And r Ldt ) + rt Ldt (19)
Foundation of Greece in the context of the IKY Fellowships of L, tT dD dD
Excellence for Postgraduate studies in Greece - Siemens Program.
By multiplying both sides of Eq. (13) with qift and summing for
all i IS , f Fi , t T, the following condition arises:
Appendix.

pift qift
The KKT rst-order optimality conditions of the ISO market
iI S , f F i , tT
clearing problem are presented in the following Eqs. (13)(18) only
for the PTDF Formulation, since a detailed representation of the
 
max min
= rt qift (ift ift ) qift
resulting MPEC model for the Nodal Formulation is presented in
[17]. iI S , f F i , tT iI S , f F i , tT

  
max min
rt ift + ift Ani r (max min ) pift =0
t t (max min ) Ani r qift (20)
L t t

L, tT iI S , f F i
i I, f F i , t T (13)
148 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149

By processing the complementarity condition (15), the follow- min

ift vmin
ift
M i I, f F i , t T (30)
ing equation arises:

min
ift qift = 0 i I S , f F, t T min
ift qift = 0 (21) Qift qift 0 i I, f F i , t T (31)
iI S , f F i , tT
max
ift 0 i I, f F i , t T (32)
Similarly, the complementarity condition (16) leads to:
max
ift max
qift = ift Qift i I S , f F i , t T (22) Qift qift (1 vmax ) MQ i I, f F i , t T (33)
ift

thus
  max
ift vmax M i I, f F i , t T (34)
max max ift
ift qift = ift Qift (23)
iI S , f F i , tT iI S , f F i , tT
  
By processing (20), (21), (23), we get: Anr qift Ldt + Flmax 0  L, t T

nN in , f F i dn
pift qift
(35)
iI S , f F i , tT
 
max
= rt qift ift Qift
min
t
0 t T,  L (36)
iI S , f F i , tT iI S , f F i , tT

    
(max min ) Ani r qift (24) Anr qift Ldt + Flmax (1 vt
L min
) ML
t t

L, tT iI S , f F i nN in , f F i dn

By combining (19) and (24), the following equation results:  L, t T (37)


rt qift
iI S , f F i , tT min
t
vLtmin M L t T,  L (38)
 
max
= pift qift ift Qift
  
iI R , f F i , tT iI R , f F i , tT Amr qift Ldt Flmax 0  L, t T
 
nN in , f F i dn
max
t
Flmax min
t
Flmax
(39)
L, tT L, tT
 
 
(max min ) And r Ldt max
t
0 t T,  L (40)
t t
L, tT dD
  
   Anr qift Ldt + Flmax (1 vt
L max
) ML
+ (max min ) Ani r qift + rt Ldt (25)
t t
nN in ,f F i dn
L, tT iI S , f F i dD
 L, t T (41)
Finally, by processing (25) with (12), which determines the loca-
tional marginalprice (LMP) of bus n in hour t, we get the nal linear
expression of  qift
iI S , f F i ni t max L max
vt M L  L, t T (42)
t

ni t qift In the above linear expressions, the binary variables
vmin
ift
, vmax
ift
L min , vL max {0, 1} and the large constants MQ , M ,
, vt t
iI S , f F i
   ML , ML are used. They are all related to the power productions
= pift qift max
ift Qift max Flmax and the power ows and have been appropriately chosen.
t
iI R , f F i iI R , f F i F
  References
min
t
Flmax + nd t Ldt t T (26)
[1] J.M. Arroyo, A.J. Conejo, Optimal response of a thermal unit to an electricity
F dD spot market, IEEE Trans. Power Syst. 15 (2000) 10981104.
[2] S. Bisanovic, M. Hajro, M. Dlakic, Hydrothermal self-scheduling in a day-ahead
Regarding the non-linearities due to the complementarity con- electricity market, Electr. Power Syst. Res. 78 (2008) 15791596.
[3] H. Haghighat, H. Sei, A.R. Kian, On the self-scheduling of a power producer in
ditions, they can be linearized based on the linear expressions
uncertain trading environments, Electr. Power Syst. Res. 78 (2008) 311317.
proposed in [34]. The resulting linearizations are the follows: [4] J.P.S. Catalo, S.J.P.S. Mariano, V.M.F. Fendes, L.A.F.M. Ferreira, A practical
approach for prot-based unit commitment with emission limitations, Int. J.
qift 0 i I, f F i , t T (27) Electr. Power Energy Syst. 32 (2010) 218224.
[5] C.K. Simoglou, P.N. Biskas, A.G. Bakirtzis, Optimal self-scheduling of a thermal
producer in short-term electricity markets by MILP, IEEE Trans. Power Syst. 25
min
ift 0 i I, f F i , t T (28) (2010) 19651977.
[6] J.I. Perez-Diaz, J.R. Wilhelmi, J.A. Sanchez-Fernandez, Short-term operation
scheduling of a hydropower plant in the day-ahead electricity market, Electr.
qift (1 vmin
ift
) MQ i I, f F i , t T (29) Power Syst. Res. 80 (2010) 15351542.
 E.G. Kardakos et al. / Electric Power Systems Research 109 (2014) 141149 149

[7] F.J. Diaz, J. Contreras, J.I. Munoz, Optimal scheduling of a price-taker cascaded [22] C.K. Simoglou, P.N. Biskas, A.G. Bakirtzis, Optimal self-scheduling of a domi-
reservoir system in a pool-based electricity market, IEEE Trans. Power Syst. 26 nant power company in electricity markets, Int. J. Electr. Power Energy Syst. 43
(2011) 604615. (2013) 640649.
[8] H.M.I. Pousinho, V.M.F. Mendes, J.P.S. Catalao, A stochastic programming [23] B.F. Hobbs, C.B. Metzler, J.S. Pang, Strategic gaming analysis for electric power
approach for the development of offering strategies for a wind power producer, systems: an MPEC approach, IEEE Trans. Power Syst. 15 (2010) 638645.
Electr. Power Syst. Res. 89 (2012) 4553. [24] C.J. Day, B.F. Hobbs, J.S. Pang, Oligopolistic competition in power networks: a
[9] S. Nojavan, K. Zare, M.R. Feyzi, Optimal bidding strategy of generation station conjectured supply function equilibrium approach, IEEE Trans. Power Syst. 17
in power market using information gap decision theory (IGDT), Electr. Power (2002) 597607.
Syst. Res. 96 (2013) 5663. [25] V.P. Gountis, A.G. Bakirtzis, Efcient determination of Cournot equilibria in
[10] J. Garcia-Gonzalez, J. Barquin, Self-unit commitment of thermal units in a electricity markets, IEEE Trans. Power Syst. 19 (2004) 18371844.
competitive electricity market, in: Proceedings of the IEEE Power Engineering [26] L.A. Barroso, R.D. Carneiro, S. Granville, M.V. Pereira, M.H.C. Fampa, Nash equi-
Society Summer Meeting, 2000. librium in strategic bidding: a binary expansion approach, IEEE Trans. Power
[11] S. De La Torre, J.M. Arroyo, A.J. Conejo, Price maker self-scheduling in a pool- Syst. 21 (2006) 629638.
based electricity market: a mixed-integer LP approach, IEEE Trans. Power Syst. [27] C.A. Diaz, J. Villar, Electricity market equilibrium based on conjectural varia-
17 (2002) 10371042. tions, Electr. Power Syst. Res. 80 (2010) 15721579.
[12] V.P. Gountis, A.G. Bakirtzis, Bidding strategies for electricity producers in [28] C. Ruiz, A.J. Conejo, Y. Smeers, Equilibria in an oligopolistic electricity pool with
a competitive electricity marketplace, IEEE Trans. Power Syst. 19 (2004) stepwise offer curves, IEEE Trans. Power Syst. 27 (2012) 752761.
356365. [29] E.G. Kardakos, C.K. Simoglou, A.G. Bakirtzis, Short-term electricity market sim-
[13] A. Baillo, M. Ventosa, M. Rivier, A. Ramos, Optimal offering strategies for gener- ulation for pool-based multi-period auctions, IEEE Trans. Power Syst. 28 (2013)
ation companies operating in electricity spot markets, IEEE Trans. Power Syst. 25262535.
9 (2004) 745753. [30] S.O. Kimbrough, F.H. Murphy, Strategic bidding of offer curves: an agent-based
[14] M.V. Pereira, S. Granville, M.H.C. Fampa, R. Dix, L.A. Barroso, Strategic bidding approach to exploring supply curve equilibria, Eur. J. Oper. Res. 229 (2013)
under uncertainty: a binary expansion approach, IEEE Trans. Power Syst. 20 165178.
(2005) 180188. [31] S.J. Wang, S.M. Shahidehpour, D.S. Kirschen, et al., Short-term generation
[15] A.G. Bakirtzis, C.K. Simoglou, N.P. Ziogos, A.C. Tellidou, G.A. Bakirtzis, Electricity scheduling with transmission and environmental constraints using an aug-
producer offering strategies in day-ahead energy markets, in: Proceedings of mented lagrangian-relaxation, in: Proceedings of the 1994 IEEE/PES Summer
the 2007 iREP Symposium Bulk Power System Dynamics and Control VII, Meeting, San Francisco, CA, July, 1994.
Revitalizing Operational Reliability, August, 2007. [32] R.D. Christie, B.F. Wollenberg, I. Wangensteen, Transmission management in
[16] T. Li, M. Shahidehpour, Strategic bidding of transmission-constrained GEN- the deregulated environment, Proc. IEEE 88 (2000) 170195.
COs with incomplete information, IEEE Trans. Power Syst. 20 (2005) [33] E. Litvinov, Design and operation of the locational marginal prices-based elec-
437447. tricity markets, IET Gen. Transm. Distrib. 4 (2010) 315323.
[17] C. Ruiz, A.J. Conejo, Pool strategy of a producer with endogenous formation of [34] J. Fortuny-Amat, B. McCarl, A representation and economic interpretation of a
locational marginal prices, IEEE Trans. Power Syst. 24 (2009) 18551866. two-level programming problem, J. Oper. Res. Soc. 32 (1981) 783792.
[18] L. Xu, R. Baldick, Y. Sutjandra, Bidding into electricity markets: a transmission- [35] Reliability System Task Force, The IEEE reliability test system-1996: a report
constrained residual demand derivative approach, IEEE Trans. Power Syst. 26 prepared by the reliability test system task force of the application of proba-
(2011) 13801388. bility methods subcommittee, IEEE Trans. Power Syst. 14 (1999) 10101020.
[19] A.G. Bakirtzis, N.P. Ziogos, A.C. Tellidou, G.A. Bakirtzis, Electricity producer [36] Electrical and Computer Engineering Department Illinois Institute of Tech-
offering strategies in day-ahead energy market with step-wise offers, IEEE nology. Network Data for the IEEE 118-Bus System [Online]. Available:
Trans. Power Syst. 22 (2007) 18041818. http://motor.ece.iit.edu/data/IEEE118bus data gure.xls
[20] L.A. Barroso, A. Street, M.V. Pereira, S. Granville, Offering strategies and simu- [37] U.S. Energy Information Administration (EIA). Fuel Prices [Online]. Available:
lation of multi-item iterative auctions of energy contracts, in: Proceedings of http://www.eia.gov/forecasts/steo/tables/pdf/2tab.pdf
the General Meeting of the IEEE Power and Energy Society, San Diego, CA, July, [38] World Nuclear Association. Nuclear Fuel Prices [Online]. Available:
2012. http://www.world-nuclear.org/info/inf02.html
[21] H.M.I. Pousinho, J. Contreras, A.G. Bakirtzis, J.P.S. Catalao, Risk-constrained [39] A. Brooke, D. Kendrick, A. Meeraus, GAMS Users Guide, The Scientic Press,
scheduling and offering strategies of a price-maker hydro producer under Redwood City, CA, 1990 [Online] Available: http://www.gams.com/docs/
uncertainty, IEEE Trans. Power Syst. 28 (2012) 18791887. document.htm