Vous êtes sur la page 1sur 9

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO.

2, MAY 2004 987

Slack Bus Selection to Minimize the System


Power Imbalance in Load-Flow Studies
Antonio Gmez Expsito, Senior Member, IEEE, Jos Luis Martnez Ramos, Member, IEEE, and
Jess Riquelme Santos

AbstractThis paper reconsiders the notion of slack bus in arbitrarily proposed as slack in absence of better criteria, which
load-flow studies. Instead of determining a priori which bus plays is a good choice in case the total imbalance (i.e., the loss esti-
the role of slack bus, it is selected on the fly during the load-flow mation error) is relatively large.
iterative process in such a way that the system power imbalance is
minimized. The problem of selecting the best slack bus, or buses,
Earlier load-flow algorithms were mainly concerned with
is first formulated as a nonlinear optimization problem. Then, the convergence problems when selecting the slack bus. The first
results obtained are justified on the basis of the involved equality systematic study about this issue was presented in [6], where
constraint being a quasilinear function, leading to an LP problem it is concluded that the optimal choice from this point of
with trivial solution. It turns out that the optimal solution can view is the bus with largest short-circuit current (i.e., the one
be easily found from the results of a conventional load flow at a whose diagonal element is smallest). Those concerns
moderate cost. The proposed heuristic procedure is tested on the
IEEE test systems. virtually vanished when the NewtonRaphson method was
fully developed, as it proved to be much more tolerant of slack
Index TermsIncremental transmission losses, linear program-
bus location than older methods [7].
ming, load flow, slack bus.
Other suggested criteria for single slack bus selection are [6]:
a) have a large number of lines connected to it; b) have a voltage
I. INTRODUCTION leading all other voltages of the system.

S OLVING the load-flow problem requires that total gen-


erated power matches the total demand plus transmission
losses. However, as such losses cannot be determined before-
B. Distributed Slack Bus
The distributed slack bus concept is implemented in some
hand, total generation needed to supply a known demand cannot power flow programs, particularly those used in EMS applica-
be exactly specified a priori. In consequence, it is necessary to tions. Under this approach, total system imbalance is frequently
have at least one bus (the slack bus) whose real power genera- interpreted as a deviation of the load-frequency control mecha-
tion can be rescheduled to supply the difference between total nism. Therefore, it is assumed that the set of generators involved
system load plus losses and the sum of active powers specified in the automatic generation control (AGC) contribute to balance
at generation buses [1]. Following [9] and [10], this difference the system in proportion to the so-called participation factors
will be named system power imbalance. Furthermore, as phase [8], [9]. Such coefficients can be determined based on [8]: a) ma-
angles of bus voltage phasors must be referred to some arbitrary chine inertias; b) governor droop characteristics; c) frequency
reference, the voltage phasor of the slack bus is usually taken as control participation factors; d) economic dispatch [9], [10].
reference and, hence, its phase angle becomes zero [1], [2]. However, the units used for steady-state loss compensation
Existing approaches for selecting the slack bus can be broadly need not be the same as those participating in AGC [9].
classified as follows: In [11], for instance, the vector of participation factors is
made colinear with the specified generation vector (i.e., each
A. Single Slack Bus
factor is proportional to the respective generation scheduling).
Most textbooks fall within this category [3], [4]. The slack bus For localized power-flow solutions, a geographical criterion
is considered a mathematical artifact created by the load-flow is suggested in [12], where it is shown that distributing the
analyst, without any direct link with the physical system [5]. Ac- system imbalance among several nearby generators improves
tive power is specified at generation buses, most likely including the performance with respect to the use of a single, probably
an estimation of ohmic losses. Hence, the difference between remote slack.
computed and specified active power at the slack bus represents When a distributed slack bus is adopted, all active power
the error in the prior estimate of system losses. Only in rare, mismatches are retained in the unknown vector while the
typically small tutorial cases does the system imbalance repre- state vector gets augmented with the system power imbalance
sent total power losses, which may exceed the rating of certain in order to compensate for the missing phase reference [9],
generators for realistic systems. Usually, the largest generator is [10], [13]. Nonzero elements of the extra Jacobian column are
simply the respective participation factors.
Manuscript received July 31, 2003. This work was supported in part by the In addition to the stand-alone load-flow problem, partici-
Spanish MCYT and in part by Junta de Andaluca under Grants DPI2001-2612 pation factors frequently arise in other applications where the
and ACC-1021-TIC-2002, respectively. load-flow problem constitutes the core, like economic dispatch
The authors are with the Department of Electrical Engineering, University of
Seville, Seville, Spain (e-mail: age@us.es; jlmr@esi.us.es; jsantos@us.es). [10], [11], deregulated markets [13], [14], sensitivity analysis
Digital Object Identifier 10.1109/TPWRS.2004.825871 [15], etc.
0885-8950/04$20.00 2004 IEEE
988 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 2, MAY 2004

It is worth noting that, by definition, participation factors are power losses, will be affected to a certain extent by the slack bus
all positive, which means that all involved generators will in- selection.
crease (decrease) its scheduled power when the net power im- Consider, as an example, the three-bus network shown in
balance is positive (negative). Fig. 1, where generation scheduling is determined in such a way
Somewhat between both schemes, the multiple slack bus con- that the system imbalance exactly matches ohmic losses. When
cept can be used to achieve net interchange adjustments be- bus 1 is the slack bus, total losses amount to 2.385 MW, while
tween control areas in multiarea load-flow algorithms [16]. In these losses are 2.412 if bus 3 is selected as slack bus. There-
this case, each control area may have one of its generating units fore, it is preferable, from the point of view of reducing losses,
designated as area slack, and net interchange is adjusted by to select bus 1 as slack bus in this case. Note that further reduc-
changing the power output of all area slacks except the system tion of losses could be achieved by increasing (decreasing) the
slack bus, which must account for total system imbalance. This net power injected at bus 1 (bus 3), as a consequence of branch
scheme can be also considered a particular case of distributed 12 resistance being much smaller than the other, but this goes
slack bus in which participation factors are determined on the beyond the load-flow scope in which generation powers are ex-
fly to match the desired interchanges. ternally determined and the only degree of freedom is the slack
It may be argued that a distributed slack bus better resembles bus choice.
the way power systems are operated, provided participation fac- In this paper, it will be assumed, like in the former example,
tors are properly chosen. However, this depends on the context that active power at all generation buses is specified. This is the
in which the load flow is used (EMS, planning, deregulated mar- case also when a distributed slack bus is adopted. Total specified
kets, etc.). generation may, or may not, as in the example, take into account
In this paper, no assumption is made a priori about the slack an estimation of network losses, leading to positive or negative
bus being unique or distributed. Any generation bus, or com- system imbalances. The important thing, as shown above, is that
bination of buses, can play the slack bus role but, instead of the resulting imbalance depends upon which particular bus is
resorting to a given set of participation factors, total power in- chosen as slack. Consequently, there is a margin to obtain im-
jected by each generator is decomposed into a major constant proved load-flow solutions by carefully selecting the slack bus.
term, specified beforehand in the input file, and a slack power Furthermore, there is no reason a priori to believe that choosing
representing its unknown contribution to the net power imbal- a single slack bus is better than distributing the total imbalance
ance. As the need for the slack bus is a direct consequence of among several generators. In fact, based mainly on intuition, it is
the existence of the power imbalance, it makes sense to select generally believed that distributing the system imbalance among
as slack, the bus or set of buses that minimize such an imbal- several generators leads to lower losses [9]. The above observa-
ance, using the slack powers as control variables. This criterion tions provide the foundation for the work reported in this paper.
can be shown to be equivalent to that of minimizing active power
losses, even though the net system imbalance will differ, in gen- III. PROBLEM FORMULATION
eral, from actual losses. Conditions leading to the choice of a
single or distributed slack are also presented. Mathematically, the choice of a slack bus to minimize the
The ideas presented in this paper might find application in system power imbalance, including the possibility for it to
some of the market-related issues still open, like that of loss al- be distributed, can be exactly formulated as an optimization
location, but this is out of the scope of the paper which is focused problem.
on ordinary load flows. The proposed criterion to select the slack Let denote the contribution of generator to match the
bus(es) is original and easy to implement in existing load flows. system imbalance (in the sequel, will be referred to as the
Unlike most slack bus selection criteria, which require the in- slack power of bus ). Then, the load-flow equations can be
tervention of a skilled user, familiarized with the power system written as follows:
being solved, the proposed scheme provides an automatic pro-
cedure that may be useful in those cases where the user has no (1)
preliminary idea about the right candidate for slack bus. (2)
The paper is organized as follows: First, the problem of (3)
selecting the slack generator(s) in load-flow studies so as to
minimize the power imbalance is presented, and the math- where, assuming bus 1 is the reference bus, the augmented state
ematical formulation as an optimization problem is derived. vector is given by
Then, some test results are presented and discussed. Finally,
a heuristic approach to automatically find the optimal slack (4)
bus from the results of a conventional load flow is presented
(5)
and tested on some IEEE systems.
(6)
II. MOTIVATION Note that the sum of slack powers must be equal to the system
In load-flow studies, it is customary to choose as slack bus power imbalance, that is
the same bus whose phase angle is arbitrarily set to zero. How-
ever, while it is irrelevant for the load-flow solution which bus (7)
is taken as phase origin, the total system imbalance, and hence,
EXPSITO et al.: SLACK BUS SELECTION TO MINIMIZE THE SYSTEM POWER IMBALANCE IN LOAD-FLOW STUDIES 989

three-bus system of Fig. 1, the unconstrained minimum of


is achieved when and .
This is not a desirable feature for a conventional load flow,
which should not be allowed to redistribute power in this
way when selecting the slack bus. Therefore, the slack
powers should be further constrained, in addition to
the upper limits given by (11). In the same way as partic-
ipation factors are all positive when a distributed slack is
adopted, it is reasonable to force each to be positive
Fig. 1. Three-bus system and associated data (S = 100 MVA).
if the system imbalance is positive, in order that no gen-
erator reduces its scheduled power when there is a power
In compact form, the set (1)(3) can be expressed as deficit. Similarly, all s should be negative when there
(8) is a surplus of generated power. The problem is that there
is no way to determine a priori which is the case, unless
The above system comprises equations in a load-flow solution exists. Considering that each term
unknowns, which means that there are degrees of (or ) is null only when
freedom. Conventionally, values of coefficients are (or ), the requirement that all s are simulta-
set to zero, and that of the slack bus is determined after the load neously positive or negative can be fulfilled by enforcing
flow is solved, but there are other possibilities to obtain such the following constraint:
coefficients keeping in mind a certain merit function.
The objective function proposed in this paper is to minimize
(12)
the net power imbalance

(9) Consequently, ignoring for simplicity the upper bounds (11),


the problem of choosing the slack powers in an optimal way
can be formulated as follows:
which, according to (7), amounts to minimizing the power
losses.
min
In case each slack power is assigned a certain cost , the slack
bus(es) could be alternatively selected in such a way that the
scalar

(10)
(13)
is minimized. Minimizing the cost associated with the system
In order to remove the operator from the model, it is
imbalance is just a matter of scaling the slack powers, and the
customary to split each as follows:
procedure developed below for is directly applicable to .
Two types of inequality constraints could be enforced when (14)
solving the equality-constrained optimization problem formu-
lated above, as follows. where and are both positive. This way, the equivalent
1) Equipment physical limits, such as maximum power problem that must be solved in practice is
flows, generator ratings, etc. However, except for re-
active power limits, the remaining constraints are out min
of the scope of a conventional load flow and should be
handled by ad hoc congestion management procedures.
Furthermore, as the system imbalance is usually quite
small, it is very unlikely that the slack bus choice drives
any power flow beyond the respective branch rating.
Anyway, simple constraints like (15)
(11) It should be noted that the solution to this problem minimizes
the extra power that generators must deliver when the power
could be easily added as a safeguard against large
imbalance is positive, or maximizes the reduction of generated
deviations.
power when the imbalance is negative.
2) If no constraints are imposed on the slack variables ,
it may well happen that certain generators strategically
located significantly increase their power share, while IV. TEST RESULTS
others get their scheduled power virtually cancelled by To begin with, the five-bus system of Fig. 2 will be tested. Ac-
a negative (i.e., ). For instance, in the cording to the data presented therein, total generation is equal
990 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 2, MAY 2004

TABLE I
TEST RESULTS FOR THE 5-BUS SYSTEM (POWERS IN MEGAWATTS)

TABLE II
TEST RESULTS FOR THE 57-BUS SYSTEM (POWERS IN MEGAWATTS)

Fig. 2. Five-bus system and associated data.

in this case to total demand, which means that the system im-
balance amounts to power losses.
Table I collects the results of solving the optimization Losses using bus 1 as slack bus.
problem (15) in three cases. Case 0 refers to the base case;
case 1 differs from case 0 in the way generators 4 and 5 share provided negative reschedulings are not allowed when there is
the total load, while in case 2, the load at bus 1 is significantly a power deficit (and vice versa). This section is devoted to jus-
increased and the generation scheduling is also changed. The tify this conclusion and to discuss possible situations in which
main conclusion is that, in all cases, the optimum is achieved a distributed slack may arise.
with a single, rather than distributed, slack bus. The results also As stated in Section III, when the slack powers are
show that the optimal slack bus location is a function of both added to the load-flow model, degrees of freedom arise.
load level and generator scheduling, contrary to the systematic This means that, if it was possible for the conventional state
choice of the same slack bus conventionally performed in variables to be eliminated from (1)(3), then all would be
load-flow studies. linked by a single nonlinear equation such as
The IEEE 57-bus system, comprising four generators, has (16)
been also tested. Table II presents the results obtained in four
cases. Case 0 refers to the standard base case, in case 1, the total Of course, because of the nonlinearity of the load-flow equa-
load is equally split among the generators, case 2 differs from tions, there is no way to explicitly write (16). However, it is easy
the base case in the fact that the scheduled powers for generators to obtain its linearized counterpart around a load-flow solution
3 and 8 are exchanged, and the power of bus 12 is increased in point ( , , ). Let denote the column vector
case 3 so that the system imbalance is negative. The lowest row when the reference bus is omitted. Then, linearization of (1)(3)
shows, for comparison, the losses that would take place by run- around ( , , ) yields
ning the load flow with the standard slack bus (bus 1). For this
system, this particular slack bus is actually the worst in most
cases, which shows the importance of carefully choosing the (17)
best slack bus in accordance with network topology, load level,
etc. This is clearly seen in case 3, where there is an excess of
generated power and bus 1 is selected to maximize the power re- (18)
duction. Like in the five-bus system, it turns out that the system
imbalance (power losses except for case 3) is optimally provided where is the conventional load-flow Jacobian, is the Jaco-
by a single bus, which means that losses can be further reduced bian row corresponding to the reference bus (missing in ordinary
only if the slack powers s are allowed to become positive load flows), and the column vector has been split for conve-
and negative simultaneously. nience into its generator and demand components and ,
respectively. From (17) and (18), the following relationship is
V. EXPLANATION OF RESULTS obtained:
The results presented above, and many more obtained in other
experiments, lead to the conclusion that the system power im- (19)
balance is minimized when a single generator accounts for it,
EXPSITO et al.: SLACK BUS SELECTION TO MINIMIZE THE SYSTEM POWER IMBALANCE IN LOAD-FLOW STUDIES 991

or, in more compact form TABLE III


HYPERPLANE COEFFICIENTS AT DIFFERENT SOLUTION POINTS
(20)

where refers to the columns of corresponding to .


Note that the row vector can be obtained from the LU
factors of , without computing explicit inverses.
Therefore, for small deviations around the load-flow solution,
(16) can be linearized as

(21)

Note that the right-hand side of the above equation reduces to


in conventional load-flow solutions, where the slack and
reference buses are the same. Rearranging terms in (21), the
hyperplane that approximates (16) around ( , , ) can
be written in standard form as

(22)

It is important to realize that the coefficients depend on


the solution point. The more constant these coefficients are, the
more linear (16) is, and vice versa. Table III presents these co- Original slack bus.
efficients for the three networks tested above (case 0), when the
load-flow equations are linearized around every possible solu-
tion with single slack bus.
The degree of parallelism between two different hyperplanes
can be assessed by performing the scalar product of the orthog-
onal vectors associated with them, whose coordinates are deter-
mined by the coefficients . This leads to the angles shown
in the bottom rows of Table III, the largest one being 0.22 .
Therefore, it can be concluded that (16) is almost a linear func-
Fig. 3. Relationship among the coefficients P for two- and
tion if the slack powers are restricted to the region three-dimensional spaces.
(or when the imbalance is negative). Fig. 3 represents
the subspace (16) for the three-bus and five-bus networks. As TABLE IV
the visual representation suggests, both subspaces fit pretty well HYPERPLANES OBTAINED FOR THE 5-BUS SYSTEM BY LINEAR
REGRESSION ON 35 POINTS
the respective linear variety (the curvature has been exagger-
ated). For the five-bus system, for instance, a set of 35 load flows
was run by choosing different slack powers within the feasible
triangular region. A linear regression provides the best hyper-
plane coefficients shown in Table IV, along with the upper and
lower hyperplanes defining the 95% confidence interval. These
figures confirm that the subspace (16) is nearly a linear variety. When the coefficients s are negative (i.e., negative imbal-
Assuming the coefficients are positive and ignoring the non- ance), the inequality constraints to be enforced are but
linearity of (16), the problem of choosing the slack bus that min- (24) still applies (the slack bus corresponds in this case to that
imizes the system imbalance can then be formulated as follows: with largest ).
For very large system imbalances (positive or negative),
however, the validity of the linear model (23) could be ques-
min tioned, and the LP theory can no longer be applied to conclude
that the optimum lies at a vertex. Fig. 4 shows, in addition to
the linear problem, two nonlinear functions with increasing
curvature for the two-dimensional case (the thin parallel lines
(23) represent constant values of the objective function). As theo-
retically predicted, in the linear case, the solution point
It is easy to show (see, for instance, [17]) that the above LP lies at the vertex with smallest coefficient ( in the
problem converges to the vertex where example). This can also be the case for slightly nonlinear func-
tions, like in Fig. 4(b), but not necessarily, like in Fig. 4(c),
min (24) where lies at an intermediate point.
992 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 2, MAY 2004

Based on the above findings, the following heuristic proce-


dure is proposed that should lead, except for very pathological
cases, to the same solution as if (15) was actually solved.
1) Obtain the load-flow solution by provisionally using the
reference bus as slack bus. At this stage, there is no need
to adopt a very tight convergence criterion.
2) Compute the coefficients of the tangent hyperplane (22)
based on the LU factors of the last Jacobian.
Fig. 4. Two-dimensional optimization problem: (a) linear; (b) and (c) 3) Select as slack bus the one whose coefficient satisfies
nonlinear equality constraint.
(24).
4) Using the new slack, perform extra iterations starting
As proved in the Appendix, for every pair of coordinates from the solution of step 1, until full convergence. This
and , the following two conditions must be simultaneously requires that the rows corresponding to the old and new
satisfied for the optimum to lie at an intermediate point: slack buses be exchanged (phase angles should be shifted
as well so that the new slack bus is also the reference
(25) bus).
This procedure is based on the empirical observation that
the bus whose hyperplane coefficient is the smallest seldom
where refers to the coefficient of in the tangent sub-
changes with the slack bus, because of the parallelism of the
space (22) computed at the vertex given by and different hyperplanes. In case of doubt, step 2 can be performed
for . again to check that the slack bus remains the same.
Since the function (16) is so linear, it is very unlikely that Furthermore, let be the chosen slack bus and the bus
the two conditions (25) are satisfied, which explains the test re- whose coefficient is closest to that of . In case ,
sults presented in Section IV. However, such a possibility exists, the possibility of a distributed slack can be confirmed or dis-
particularly when the weighted function (10) is adopted in order carded by obtaining a new load flow with as slack bus and
to minimize the cost rather than the system imbalance. Defining then checking (25). This extra work is seldom justified because,
the scaled variables , it is straightforward to prove even if the system imbalance is distributed between and , the
that (25) is equivalent to power reduction achieved will be very small.
The need for a distributed slack bus may arise, however, when
(26) upper limits are imposed upon slack powers. For instance, if
, then and
For instance, applying (26) to the three-bus system yields

is the optimal solution provided .


which is indeed a very narrow interval. The intermediate point
and is reached when
. VII. RESULTS PROVIDED BY THE HEURISTIC APPROACH
The Appendix also shows that the hyperplane coefficients The procedure presented in the former section has been
are related with the incremental transmission loss coefficients applied to the IEEE 118-bus system (data taken from [18]).
through Table V presents the three buses , , and with smallest
tangent hyperplane coefficients. The first row corresponds to
the standard slack bus of the test case, the second refers to the
slack bus deemed as optimal , and the third is obtained with
the best competitor of the chosen slack bus .
which means that a given slack bus is optimal, from the point of The following comments are in order:
view of reducing the system imbalance, when the resulting ITL
As happened with the 57-bus system, the original slack
coefficients are all positive. Otherwise, the bus with lowest ITL
bus is not the best choice to reduce the system imbalance
should be chosen as slack.
(in this case, it is third in the ranking).
The ranking of candidate slack buses is not altered by the
VI. HEURISTIC SELECTION OF THE SLACK BUS particular slack bus adopted to obtain the tangent hyper-
From the above discussion, it can be concluded that there is plane. This may not be true in all cases.
no need to worry in practice about the possibility for the system The hyperplane coefficients are negative this time,
imbalance to be distributed among several generating buses, un- which means that the generated power must be reduced.
less it is very large and the two smallest hyperplane coefficients Choosing bus 89 as slack, instead of bus 69, leads to an
are very similar. Furthermore, it would be quite awkward to re- extra 8.6% reduction, which is not negligible.
sort to an optimization package every time a load flow must be Computing the coefficients when bus 87 is the slack bus
solved. is not actually needed. They are included in Table V just
EXPSITO et al.: SLACK BUS SELECTION TO MINIMIZE THE SYSTEM POWER IMBALANCE IN LOAD-FLOW STUDIES 993

TABLE V system is solved twice, first with a single slack bus and then with
SMALLEST HYPERPLANE COEFFICIENTS AT DIFFERENT VERTICES a distributed one. The authors claim that, in the second scenario,
(118-BUS SYSTEM)
power losses are less than in the case of a single slack bus, as
expected. However, [9, Tables 1 and 6] actually show that the
opposite is true, confirming the theory presented in this paper.

IX. CONCLUSION
In this paper, it is assumed that when solving the load-flow
to check that the two conditions (25) are not satisfied and, problem, all generated powers are specified and that any bus,
consequently, to confirm that a single slack bus is the op- or combination of buses, can play the role of the slack bus. For
timal choice. this purpose, a slack power is introduced at each generation bus
whose value is determined in such a way that the system power
imbalance is minimized. When the slack powers are all con-
VIII. DISCUSSION AND REMARKS strained to be positive (or negative if there is a power surplus),
The following issues deserve further clarification. the resulting optimization problem becomes, in practice, an LP
OPF versus load flow. Selecting one among several slack problem with a single equality constraint, whose solution is triv-
bus candidates, keeping in mind a certain merit function, ially obtained at one of the vertices (single slack bus). Condi-
constitutes an optimization problem and, as such, can be tions for the unlikely possibility of a distributed slack bus to
considered a particular case of the so-called OPF. One of arise are also deduced from linearization around a load-flow so-
the goals of the paper is to show that, for this purpose, lution. A simple modification to existing load-flow tools is pro-
there is no need to resort to such a complex tool, as the re- posed that should detect in nearly all cases the slack bus which
quired information is a byproduct of the load-flow iterative minimizes the net power imbalance. It is also shown how the
process. The proposed procedure can be easily embedded proposed method can be easily adapted to the case in which the
within existing load flows, and the user will simply notice cost of the system imbalance is to be minimized.
that the selected slack bus sometimes differ from that of Major contributions of this paper are: 1) an original and
the input file when this option is activated. simple criterion to select the slack bus in load-flow studies
On the other hand, the OPF problem formulated in this without user intervention; (2) a theoretical development pro-
paper to theoretically found the proposed method is orig- viding further insight into the implications of choosing a single
inal in the way control variables, objective function, and or distributed slack bus from the point of view of ohmic losses;
constraints are combined. Conventionally, active powers and (3) an algebraic and geometrical justification about the
are used as control variables when cost is minimized, remote possibility of a distributed slack bus being preferable to
while reactive powers are rescheduled to minimize losses a single slack when the power imbalance is a concern.
[19]. In this paper, a preliminary loss minimization is
achieved by carefully selecting the bus that takes care of APPENDIX A
the system imbalance. REQUIREMENTS FOR DISTRIBUTED SLACK BUS
Constraining the slack power signs. From the OPF per- Let and be a couple of candidate slack buses. When the
spective, constraining the slack powers to be all positive remaining generators are discarded, the objective function (9)
(negative) when the power imbalance is positive (negative) becomes
might seem artificial. However, this is required to prevent
the generation scheduling, determined beforehand, from (28)
being redistributed merely because of a small power im-
balance. The reader should be aware that this constraint is Also, in the resulting two-dimensional subspace, the tangent
also implicit when a distributed slack, based on conven- hyperplanes computed at the vertices and are
tional participation factors, is employed. This is clearly
seen from the expression (29)

(27)
(30)
where is the system imbalance and the respective
participation factor [10]. As by definition, it fol- For small perturbations, the objective function can be linearized
lows that the sign of is the same as that of the power around both vertices as follows:
imbalance.
Note that this constraint is crucial to reach one of the main (31)
conclusions of the paper, namely the fact that the system imbal-
ance is minimized virtually in all cases when the slack power
is provided by a single generator, which is somewhat contrary (32)
to intuition and general belief. For instance, in [9] a five-bus
994 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19, NO. 2, MAY 2004

The optimal solution will lie within the two vertices when the
objective function decreases for perturbations in that direction,
that is, when the following two conditions are satisfied:

and

or, equivalently

(33)
Fig. 5. Geometrical interpretation of conditions for distributed slack bus.

Otherwise, the optimal solution will be the vertex


However, according to (36), this is only true if
when (and vice versa). Geometrically, (33) can be
interpreted as both angles and in Fig. 5 being larger than
45 . (38)

APPENDIX B which is not exactly the case, as shown by Tables III and V (the
RELATIONSHIP OF TANGENT HYPERPLANE COEFFICIENTS WITH hyperplanes are not perfectly parallel).
INCREMENTAL TRANSMISSION LOSSES
In this section, it will be assumed that the system imbalance REFERENCES
amounts to the power losses. [1] I. J. Nagrath and D. P. Khotari, Modern Power System Analysis. New
For a given slack bus , the tangent hyperplane is defined by York: McGraw-Hill, 1980.
[2] R. N. Dhar, Computer Aided Power System Operations & Anal-
ysis. New York: McGraw-Hill, 1982.
[3] A. Bergen, Power Systems Analysis. Englewood Cliffs, NJ: Prentice-
(34) Hall, 1986.
[4] J. J. Grainger and W. D. Stevenson, Power System Analysis. New York:
McGraw-Hill, 1994.
Assuming this linear variety is a good approximation of the [5] B. Stott, Review of load-flow calculation methods, Proc. IEEE, vol.
62, pp. 916929, July 1974.
nonlinear function (16), losses can be expressed as follows: [6] L. L. Freris and M. Sasson, Investigation of the load-flow problem,
Proc. Inst. Elect. Eng., vol. 115, no. 10, pp. 14591470, Oct. 1968.
[7] W. F. Tinney and C. E. Hart, Power flow solution by Newtons method,
(35) IEEE Trans. Power App. Syst., vol. PAS-86, pp. 14491460, Nov. 1967.
[8] P. H. Haley and M. Ayres, Super decoupled load flow with distributed
slack bus, IEEE Trans. Power App. Syst., vol. PAS-104, pp. 104113,
Jan. 1985.
where the variable has been eliminated by means of (34). [9] A. Zobian and M. D. Ilic, Unbundling of transmission and ancillary
services. Part I: Technical issues, IEEE Trans. Power Syst., vol. 12, pp.
Consequently, the incremental transmission loss coefficient of 539548, May 1997.
bus , , can be obtained from [10] X. Guoyu, F. D. Galiana, and S. Low, Decoupled economic dispatch
using the participation factors load flow, IEEE Trans. Power App. Syst.,
vol. PAS-104, pp. 13771384, June 1985.
[11] J. Meisel, System incremental cost calculations using the participa-
tion factor load-flow formulation, IEEE Trans. Power Syst., vol. 8, pp.
357363, Feb. 1993.
(36) [12] R. Bacher and W. F. Tinney, Faster local power flow solutions: The zero
mismatch approach, IEEE Trans. Power Syst., vol. 4, pp. 13451354,
Oct. 1989.
This shows that, as expected, the ITL information is embedded [13] P. Yan, Modified distributed slack bus load flow algorithm for deter-
in the hyperplane definition. The opposite is not true, as each mining economic dispatch in deregulated power systems, in Proc. IEEE
Power Eng. Soc. Winter Meeting, 2001, pp. 12261231.
individual cannot be obtained from the ITL coefficients [14] G. M. Huang and H. Zhang, Transmission loss allocations and pricing
unless is specified. From the load-flow solution ( , via bilateral energy transactions, in Proc. IEEE Power Eng. Soc.
Summer Meeting, 1999, pp. 720725.
), this coefficient can be computed as . Note [15] W. R. Barcelo and W. W. Lemmon, Standardized sensitivity coeffi-
that is the optimal slack bus if and only if for all cients for power system networks, IEEE Trans. Power Syst., vol. 3, pp.
15911599, Nov. 1988.
. [16] G. L. Kusic, Computer Aided Power System Analysis. Englewood
The following expression, relating the ITL coefficients for Cliffs, NJ: Prentice-Hall, 1986.
two different slack buses and can be found in the literature [17] D. G. Luenberger, Linear and Nonlinear Programming. Reading, MA:
Addison-Wesley, 1984.
[20] [18] [Online]. Available: http://www.ee.washington.edu/research/pstca/
[19] M. E. El-Hawary, Optimal Power Flow: Solution Techniques, Require-
ments and Challenges, in Proc. IEEE Tutorial Course, 1996, TP 111-0.
(37) [20] A. J. Wood and B. F. Wollenberg, Power Generation, Operation and
Control. New York: Wiley, 1996.
EXPSITO et al.: SLACK BUS SELECTION TO MINIMIZE THE SYSTEM POWER IMBALANCE IN LOAD-FLOW STUDIES 995

Antonio Gmez Expsito (SM95) was born in Jess Riquelme Santos was born in Las Palmas de
Spain in 1957. He received the electrical engineering Gran Canarias, Spain, in 1967. He received the Ph.D.
and doctor engineering degrees from the University degree in electrical engineering from the University
of Seville, Seville, Spain. of Seville, Seville, Spain.
Currently, he is a Professor and Head of the Currently, he is an Associate Professor with the
Department of Electrical Engineering, University Department of Electrical Engineering at the Univer-
of Seville, where has been since 1982. His primary sity of Seville, where he has been since 1994. His
areas of interest are state estimation and optimization primary areas of interest are active power optimiza-
techniques. tion and control, power system analysis, and power
quality.

Jos Luis Martnez Ramos (M99) was born in Dos


Hermanas, Spain, in 1964. He received the Ph.D. de-
gree in electrical engineering from the University of
Seville, Seville, Spain.
Currently, he is an Associate Professor with the
Department of Electrical Engineering at the Univer-
sity of Seville, where he has been since 1990. His pri-
mary areas of interest are active and reactive power
optimization and control, power system analysis, and
power quality.

Vous aimerez peut-être aussi