8th International Conference on Probabilistic Methods Applied to Power Systems, Iowa State University, Ames, Iowa,

September 12-16, 2004

Slack Bus Treatment in Load Flow Solutions

with Uncertain Nodal Powers
Aleksandar Dimitrovski and Kevin Tomsovic

Abstract — This paper addresses the problem introduced by same definition of the problem as the traditional deterministic
the slack bus in load flow solutions with uncertain nodal powers. approach. That is, load buses are defined as PQ buses,
While balancing powers in the system the slack bus will also generator buses as PV buses, and one bus is assumed to be a
absorb all uncertainty. The results obtained are of no practical slack bus to balance the active and reactive power in the
interest unless realistic constraints are imposed on slack power
system. The ‘slack’ bus (or ‘swing’ bus) is defined as Vθ bus.
production/consumption. Two methods of dealing with these
constraints are investigated suitable for implementation within While this definition of the load flow problem is appropriate
the recently developed boundary load flow. for a deterministic solution (although it may still be helpful to
define a distributed ‘slack’ among several buses), it has an
Index Terms — Fuzzy sets, load flow analysis, power system inherent drawback when dealing with uncertain input
planning, slack bus, uncertainty.
variables: the slack bus must absorb all uncertainties arising
from the solution and thus, will have the widest nodal power
I. INTRODUCTION possibility (probability) distributions in the system. If even

T he most common formulation of the load flow problem

requires all input variables (PQ at loads, PV at generators)
to be specified as deterministic (‘crisp’) values. Each set of
moderate amounts of uncertainty are allowed in a large
system, the resulting distributions will frequently contain
values well beyond the generating margins of the slack
specified values corresponds to one system state, which is generator.
deemed representative for some set of system conditions. This problem has been neglected so far in the literature
Thus, when the input conditions are uncertain, as is except for the case of a linearized fuzzy DC load flow [7]. In
predominantly the case in planning, there is a need for that work, three approaches, conceptually the same, use an
numerous scenarios to be analyzed. A load flow approach that iterative corrective procedure in order to satisfy constraints
could directly incorporate uncertainty into the solution process imposed on the slack bus. Recently, the authors have
has been long recognized as useful. The results from such developed a methodology that enables an accurate solution
from a non-linear AC fuzzy load flow [8]. It follows the
analysis would be expected to give solutions over the range of
concept of boundary load flow (BLF) solutions, where
the uncertainties, i.e., solutions that are sets of values or
solutions are based on an optimization procedure for
regions instead of single operating points.
implicitly defined vector functions. Numerical results obtained
To date, two families of uncertain load flow algorithms from test systems have shown the feasibility of this approach,
have evolved. The first one is the probabilistic load flow but they also have shown the problems associated with the
(PLF), which considers loads and generations as random inappropriate definition of the slack bus.
variables with some probability distributions (e.g., [1] - [4]). This paper extends the previous work and investigates
The results of the load flow, i.e., voltages, power flows, and different ways of incorporating the constraints imposed on the
so on, are also random variables with resultant probability dis- slack bus in the framework of boundary load flow solutions.
tributions obtained using probabilistic techniques. The second Two methods of dealing with this problem are considered: 1)
is the fuzzy load flow family of algorithms where input vari- slack bus to PV bus and PV bus to slack bus conversion, and
ables are represented as fuzzy numbers (e. g., [5] - [7]). Fuzzy 2) distributed slack bus modeling. The results obtained from
numbers are described by possibility distributions and can be different test systems as well as the specifics in different
considered to be intervals with indistinct boundaries. The approaches are discussed and compared.
results obtained are also fuzzy numbers with resultant
possibility distributions. The authors have recently extended
II. BOUNDARY LOAD FLOW SOLUTIONS
these concepts to the so-called boundary load flow [8].
Both families of uncertain load flow algorithms use the The BLF was presented for the first time in [2] within the
context of PLF. In that paper, an approximate solution for the
ranges of values for state and output variables, given the
This work was supported in part by the National Science Foundation ranges of values of input variables from their probability
(NSF) under Grants No. DGE-0108076 and No. EEC 02-24810
A. Dimitrovski is with the School of EE, University “Sv. Kiril i Metodij”, distributions, was found. The ranges of variables were then
Skopje, 1000, Macedonia, currently on leave at Washington State University, used to determine multiple points of linearization for the load
Pullman, WA 99164, USA (e-mail: [email protected]). flow equations in order to improve the accuracy of the PLF
K. Tomsovic is with the School of EECS, Washington State University,
Pullman, WA 99164, USA (e-mail: [email protected]). solutions, particularly for the tail regions of the probability
distributions.

Copyright Iowa State University, 2004

 The authors have developed a methodology where an Only the signs of the partial derivatives that comprise the
accurate solution for a non-statistical interval load flow is gradient are used in the solution since our experience has
obtainable [8]. In the following, a brief explanation of this shown that the values of the partials are not useful for
methodology is given. efficiently determining the updates. Further, a procedure is
The load flow problem is defined by two sets of nonlinear needed to maintain feasibility of the solution, i.e., ensure the
equations: input variables remain within the constraints for all iterations.
The iterative procedure is reviewed in the following.
Y = g(X) (1)
Suppose that the minimum value of Xi is sought. If Kij is
and positive (negative), then decrease (increase) the value of Yj by
some fixed step size. After repeating for all Yj we obtain a new
Z = h(X), (2)
point of Y from which a new X from (1) can be found. From
where: this new point, the above steps are repeated until one of the
following is true for all input variables:
X is the vector of unknown state variables (voltage
magnitudes and angles at PQ buses; and voltage angles • the partial derivative is positive and the associated
and reactive power outputs at PV buses), variable is at a minimum;
Y is the vector of predefined input variables (real and
reactive injected nodal powers at PQ buses; and voltage • the partial derivative is negative and the associated
magnitudes and real power outputs at PV buses), variable is at a maximum;
Z is the vector of unknown output variables (real and • the partial derivative is zero.
reactive power flows in the network elements), and
If the final condition does not hold for any variable, then
g, h are the load flow vector functions.
the solution is clearly a local constrained extremum. Because
The boundary values are the extreme points found by of the nonlinearity of (1) and (2), this point may not be the
allowing the inputs to vary over their range. In our notation, only extrema. In practice, we have found the physical nature
we want to find the extreme values for the elements of X and of the load flow problem leads to either a unique solution or a
Z implicitly expressed in (1) and (2), in terms of the elements relatively small number of extrema.
of Y which, in turn, are constrained. Thus, finding the When one or more of the partial derivates are zero, the
boundary values in a load flow problem is a process of solution point lies somewhere on the boundary surface. Such a
locating the constrained extrema of implicitly defined vector point is either a local constrained extremum or a saddle point.
functions of vector arguments. Though it is unlikely that by preceding in a downhill direction
Because X cannot be explicitly expressed in terms of Y, the one will end up trapped in a local maximum or a saddle point,
solution of the system of equations (1) is found by an iterative theoretically such a possibility exists. Here, previous values of
process. Given an initial trial solution, X', the error is Xi are recorded at each step and if Xi fails to decrease, then the
calculated as: step length is modified.
∆Y = Y –Y' = Y – g(X'). (3) Finally, in the special case when all the partial derivatives
are zero, a solution cannot be obtained due to the singularity
If a Newton-Raphson (N-R) based scheme is used, (1) is of the Jacobian. Such a point typically indicates infeasibility
linearized around X' and an update for the new solution is of the load flow and a loading limit for the system considered.
found as: A singularity of the Jacobian may also occur even if not all of
∆X = K⋅∆Y, (4) the partial derivatives are zero. In such cases, the ranges of
values of the input variables are too great and one must repeat
where K is the inverse of the Jacobian of g evaluated at X'. the calculations with reduced variations for some or all of the
The element Kij of this matrix is the partial derivative of Xi variables. Note, the procedure described here must be repeated
with respect to Yj. Similarly, if we linearize (2) and substitute for each state and output variable considered, and therefore, is
for ∆X from (4) we will obtain: computationally intensive.
∆Z = S⋅∆X = L⋅∆Y, (5)
III. SLACK BUS TREATMENT
where S is the Jacobian of h at the given point of linearization. The concept of slack bus, as is well-known, is a
The matrix L = S⋅K is a sensitivity coefficient matrix and the mathematical necessity but has no physical relationship to any
element Lij is the partial derivative of Zi with respect to Yj. generator bus. Exception arises when a small system is linked
Each row of K and L represents the gradient vector of the to a much bigger system via a single tie line (single bus). In
corresponding state and output variable Xi and Zi, respectively. this case, one can represent the large system with an
Similar to derivative based optimization procedures, by equivalent generator, which can hold the voltage constant and
iteratively following the direction of the gradient, extreme generate as much power as needed, i.e. the slack bus
points (possibly local) of the state or output variable can be characteristics. Similarly in a distribution network fed by a
found.
substation, the transmission network acts as a slack bus with  ∂PSlack ∂PSlack 
respect to the distribution network.  ∂θT ∂VPQT 
The slack bus allows the solution of the nonlinear set of  PV + PQ 
equations (1) to be feasible. Since the power losses in the  ∆PSlack   ∂PPV −1 ∂PPV −1 
network are not known in advance, its role is to pick up the  ∆P   ∂θT ∂VPQT 
 PV −1  =  PV + PQ   ∆θ PV + PQ 
‘slack’ and balance the active and reactive power in the ⋅ (6)
system. This usually does not represent a problem in a well  ∆PPQ   ∂PPQ ∂PPQ   ∆VPQ 
   T T

defined deterministic load flow problem. However, in the case  ∆Q PQ   ∂θ PV + PQ ∂VPQ 
with uncertain nodal powers, the slack bus also must absorb  
all the resulting uncertainties from the solution. As a result, it  ∂Q PQ ∂Q PQ 
has the widest nodal power possibility (probability)  ∂θTPV ∂VPQT 
 
distributions in the system. This will frequently result in
operating points well beyond its generating margins. This also where:
defeats the purpose behind the study of uncertainties, which is PV is the set of all PV buses,
to investigate the impact on practical operating scenarios. In PV-1 is the set of all PV buses without the one with relaxed
the following, two ways of satisfying the constraints imposed real power,
on the slack bus are explained. PQ is the set of all PQ buses,
PV+PQ is the set of all PV and PQ buses,
A. Slack Bus - PV Bus Conversion
P, Q are the real and reactive nodal power vector functions,
This method is analog to that of PV bus to PQ bus V, θ are the vectors of unknown state variables (voltage
conversion for PV buses with reactive power limits. During magnitudes and angles), and
the course of solution of a load flow, when a PV bus’s ∂/∂(⋅)T denotes Jacobian of the corresponding vector function.
produced (or consumed) reactive power extends beyond its
limits, it is fixed at the violated limit and its voltage magnitude The problem formulation as in (6) keeps the reference
is relaxed. Thus, the PV bus has been converted to a PQ bus, angle at the slack bus (usually 0º). Another approach will be
bus with specified active and reactive power. Later, during the to relax the voltage angle of the slack bus and declare the
solution, if the bus voltage shows tendency to return and the voltage angle of the PV bus with relaxed real power as the
reactive power again falls within the limits, the bus will be reference (i.e. known). This can simply be done by replacing
converted back from PQ to PV. it in (6) with the now unknown angle at the slack and
Following the same approach as in PV bus to PQ bus retaining its current value. This will result in a complete slack
conversion above, if the slack bus real power generation (or, to PV bus and PV to slack bus conversion. In this case the
theoretically, consumption) extends beyond its predefined system of equations has the usual symmetry, with the slack
limits, it is fixed at the violated limit. Some other PV bus’s bus completely swapped.
active power generation (or consumption) then must be In the second approach, the original slack will change its
relaxed in order to be able to solve the load flow problem. The voltage angle from the initial value during the course of
PV bus to choose seems to be a matter of preference, but it is solution. However, since angles are relative to each other, we
logical to pick the one that has the highest margin from the can force it back to the initial value if desired, by subtracting
current production (consumption) to either its lower or upper that difference from each voltage angle obtained from the
limit, depending on which limit was violated at the slack bus. solution. In this way, we will obtain exactly the same solution
With the choice of a PV bus to relax, it is now possible to as with the previous formulation.
redefine the load flow problem in (1) by swapping only the Regardless of the treatment of the reference angle, the new
equation for the real power at the chosen PV bus with the slack bus takes over the balancing of power and, initially, its
equation for the slack bus real power, without changing the production (consumption) will be either decreased or
unknown state variables. In other words, the slack bus increased, depending on the limit violation at the previous
becomes a PVθ bus and the PV bus becomes just a V bus. We slack bus. During the course of solution, the production of the
still have a system of n equations with n unknowns, only the new slack bus will change and it is possible that one of its
known and unknown variables have changed and Jacobian limits gets violated also. In this case, the procedure is repeated
loses some symmetry. In this case, the system of equations with some other PV bus capable of taking over the slack. If
corresponding to (4) will have the following form: there is no such bus available, i.e., all PV buses are on their
limits, the problem is infeasible.
B. Distributed Slack Bus
Instead of assigning the excess load (or, generation) to only
one PV bus as in the previous method, we can also choose a
number of PV buses that will share it in a predetermined
manner. Two methods of sharing are: 1) proportional to the
current injections, and 2) proportional to the margin between
the current injections and the lower or upper limits,
accordingly. Of course, there are many other combinations
that may be used if deemed appropriate for some particular
application. In any case, there is no bus type conversion with
this method. If the slack bus production (consumption)
extends beyond its limits, it is relieved by redistributing the
excess load (generation) to the other PV buses. The reference
angle remains the same during the load flow solution process.
It was noted previously that in order to maintain the
feasibility of the problem, the available generation should
always match the load requirement. Cases when this is not
always true are not considered here. For example, a case with
excess generation (if each generator has some minimum limit
and their sum is bigger than the total load) requires a different
unit commitment. A case with too little generation requires a
procedure for load shedding and/or some kind of adequacy
assessment. Fig. 1. IEEE/AEP 14-bus test system.

IV. CASE STUDIES

TABLE I BOUNDARY VALUES [P.U.] FOR THE IEEE/AEP 14-BUS
Let us now apply the described methods for slack bus SYSTEM VOLTAGE MAGNITUDES – UNCONSTRAINED CASE
treatment in finding boundary load flow solutions of the small
IEEE 14-bus test system shown in Fig. 1. The system data and bus nodal power variation
voltage [50% - 150%]
the base case descriptions can be found elsewhere (for
Vmin V100% Vmax
example, [10]).
V1 1.0600 1.0600 1.0600
Table I presents the results for boundary values of voltage V2 1.0105 1.0450 1.0450
magnitudes when all specified nodal powers in the network V3 0.9645 1.0100 1.0100
vary in the range [50% - 150%] of the base case values. V4 0.9583 1.0186 1.0330
Shown are columns with minimal, base case, and maximal V5 0.9649 1.0203 1.0328
voltages. The slack bus real power generation was not V6 1.0094 1.0700 1.0700
V7 0.9904 1.0620 1.0762
restricted in this case and was found to vary in the range
V8 1.0314 1.0900 1.0900
[0.8372 - 4.0599] p.u. However, this range is outside the limit V9 0.9920 1.0563 1.0797
of the slack bus generator which is [0.1 - 2.5] p.u. V10 1.0013 1.0513 1.0746
Let us now include constraints for the slack bus and use V11 1.0189 1.0571 1.0709
Slack bus – PV bus conversion when they are violated. The V12 0.9843 1.0552 1.0638
new range of values for the slack bus generator is now V13 0.9902 1.0504 1.0628
restricted to [0.8372 - 2.5] and the new results for voltage V14 0.9581 1.0358 1.0633
magnitudes are shown in Table II.
Table III and Table IV present real and reactive power
flows in all elements of the system for the constrained and TABLE II BOUNDARY VALUES [P.U.] FOR THE IEEE/AEP 14-BUS
SYSTEM VOLTAGE MAGNITUDES – CONSTRAINED CASE
unconstrained case, respectively.
The results show that the biggest differences between the bus nodal power variation
[50% - 150%]
two cases occur when the system is heavily loaded, as voltage
Vmin V100% Vmax
expected. In other words, the biggest differences occur in the
V1 1.0600 1.0600 1.0600
values of minimal voltages and maximal power flows, which V2 1.0441 1.0450 1.0450
are usually the most interesting results. The reason for this is V3 1.0100 1.0100 1.0100
that in the most stressed scenario the BLF, in its search for an V4 0.9969 1.0186 1.0330
optimum, tries to supply almost all of the power from the V5 1.0007 1.0203 1.0328
slack. Thus, when its limits are respected the system is less V6 1.0507 1.0700 1.0700
stressed and conditions in the system improved. For example, V7 1.0327 1.0620 1.0762
V8 1.0721 1.0900 1.0900
the real power flow in the most heavily loaded branches
V9 1.0249 1.0563 1.0797
between buses 1 and 2 fell from 2.826 to 1.905 p.u. V10 1.0248 1.0513 1.0746
V11 1.0414 1.0571 1.0709
V12 1.0268 1.0552 1.0638
V13 1.0261 1.0504 1.0628
V14 0.9959 1.0358 1.0633
 It should be noted here that simple arithmetic calculations TABLE III BOUNDARY VALUES [P.U.] FOR THE IEEE/AEP 14-BUS
REAL AND REACTIVE POWER FLOWS – UNCONSTRAINED CASE
for finding the boundary values of power flows can not be
applied, due to the nonlinearity of the problem. For example, power nodal power variation
in the constrained case the maximal real power production flow [50% - 150%]
from the slack is constrained to the value of 2.5 p.u. This Smin S100% Smax
S1-2 0.4189 + j 0.1439 1.5683 + j-0.2039 2.8260 + j 0.1056
value is less than the sum of the maximal real power flows in
S1-5 0.3015 + j-0.0094 0.7555 + j 0.0350 1.2372 + j 0.1138
the branches incident to the slack, i.e., branches 1-2 and 1-5.
S2-3 0.3472 + j 0.0658 0.7319 + j 0.0357 1.1395 + j 0.0881
These results correspond to different conditions and different
S2-4 0.2660 + j-0.0716 0.5614 + j-0.0229 0.8681 + j 0.0496
load flow solutions and can not be simply lumped together.
S2-5 0.1891 + j-0.0300 0.4151 + j 0.0076 0.6524 + j 0.0688
It is interesting to note how the slack bus changed during
S3-4 -0.5526 + j-0.1079 -0.2333 + j 0.0281 0.0929 + j 0.1721
the course of solution for the minimal values of voltage
S4-5 -0.9871 + j 0.0323 -0.6122 + j 0.1567 -0.2374 + j 0.2676
magnitudes. In all the cases for the buses with unspecified
S4-7 0.1115 + j-0.1583 0.2809 + j-0.0942 0.4507 + j-0.0436
voltages (PQ buses), the initial slack bus 1 was swapped with
S4-9 0.0637 + j-0.0409 0.1609 + j-0.0032 0.2585 + j 0.0367
bus 3, then 3 was swapped with 2, and 2 was finally swapped
S5-6 0.2068 + j 0.0888 0.4406 + j 0.1282 0.6781 + j 0.1727
with 1 again. This is just a result of this particular system
S6-11 -0.0212 + j-0.0222 0.0734 + j 0.0347 0.1720 + j 0.0905
structure and the problem solution approach and does not
S6-12 0.0331 + j 0.0039 0.0778 + j 0.0249 0.1240 + j 0.0472
represent a general pattern.
S6-13 0.0716 + j 0.0133 0.1774 + j 0.0717 0.2877 + j 0.1310
Similar results are obtained when distributed slack bus
S7-8 0.0000 + j-0.2305 -0.0000 + j-0.1691 -0.0000 + j-0.0844
modeling approach is used. However, they are appear to be
S7-9 0.1115 + j-0.0371 0.2809 + j 0.0580 0.4507 + j 0.1481
slightly more optimistic. The most extreme power flows in the
S9-10 -0.0606 + j-0.0228 0.0524 + j 0.0431 0.1648 + j 0.1085
branches tend to be smaller and minimal voltages at the buses
S9-14 -0.0128 + j-0.0101 0.0944 + j 0.0367 0.2023 + j 0.0838
with the smallest values tend to be higher. This can be -0.1312 + j-0.0703 -0.0377 + j-0.0153 0.0550 + j 0.0415
S10-11
attributed to the BLF algorithm and not to the treatment of the -0.0236 + j-0.0120 0.0161 + j 0.0074 0.0570 + j 0.0279
S12-13
slack bus. Namely, it is more difficult to locate the exact -0.0341 + j-0.0269 0.0563 + j 0.0169 0.1517 + j 0.0634
S13-14
extremum when several variables simultaneously vary than
when only one or few vary. (Although Fig. 1 shows only one
generator at bus 2 besides the slack generator at 1, according TABLE IV BOUNDARY VALUES [P.U.] FOR THE IEEE/AEP 14-BUS
to the data file, the other PV buses: 3, 5, and 8, also have some REAL AND REACTIVE POWER FLOWS – CONSTRAINED CASE
real power generating limits.) nodal power variation
power
flow [50% - 150%]
The results from the analysis of bigger test systems show Smin S100% Smax
similar differences in uncertainty, only scaled to the system S1-2 0.4189 + j-0.2203 1.5683 + j-0.2039 1.9055 + j 0.1056
size. For example using the IEEE 118-bus test system, the S1-5 0.3015 + j-0.0094 0.7555 + j 0.0350 0.9042 + j 0.1138
unconstrained slack bus real power generation has maximal S2-3 0.3472 + j 0.0163 0.7319 + j 0.0357 1.2907 + j 0.0881
value of 13.98 p.u., for specified nodal powers variation in the S2-4 0.2660 + j-0.0716 0.5614 + j-0.0229 0.7572 + j 0.0589
range [90% - 110%] of the corresponding base case values. S2-5 0.1891 + j-0.0300 0.4151 + j 0.0076 0.5965 + j 0.0746
This is far away from its limit of 8.05 p.u. So, when S3-4 -0.5526 + j-0.1079 -0.2333 + j 0.0281 0.0929 + j 0.1812
constrained with either of the two methods described, its value S4-5 -0.8735 + j 0.0323 -0.6122 + j 0.1567 -0.2374 + j 0.2483
is held to 8.05 p.u. The branch with the biggest power flow in S4-7 0.1111 + j-0.1424 0.2809 + j-0.0942 0.4507 + j-0.0436
the unconstrained case is, not surprisingly, one connected to S4-9 0.0635 + j-0.0405 0.1609 + j-0.0032 0.2585 + j 0.0367
the slack bus (branch 69 – 68). Its real power flow is 5.44 p.u. S5-6 0.2068 + j 0.0879 0.4406 + j 0.1282 0.6605 + j 0.1727
In the constrained case this value is much smaller, 2.67 p.u. S6-11 -0.0212 + j-0.0216 0.0734 + j 0.0347 0.1646 + j 0.0905
The branch with the highest power flow in this case is not S6-12 0.0331 + j 0.0038 0.0778 + j 0.0249 0.1227 + j 0.0472
connected directly to the slack bus (branch 9-10) and its real S6-13 0.0716 + j 0.0133 0.1774 + j 0.0717 0.2824 + j 0.1310
power flow is 4.89 p.u. Also, the minimal voltages in the S7-8 0.0000 + j-0.2312 -0.0000 + j-0.1691 0.0000 + j-0.0844
constrained case are higher or at least equal to the minimal S7-9 0.1111 + j-0.0371 0.2809 + j 0.0580 0.4507 + j 0.1552
voltages in the unconstrained case. S9-10 -0.0543 + j-0.0228 0.0524 + j 0.0431 0.1648 + j 0.1080
S9-14 -0.0086 + j-0.0101 0.0944 + j 0.0367 0.2023 + j 0.0835
S10-11 -0.1245 + j-0.0703 -0.0377 + j-0.0153 0.0550 + j 0.0410
S12-13 -0.0236 + j-0.0120 0.0161 + j 0.0074 0.0559 + j 0.0279
S13-14 -0.0341 + j-0.0269 0.0563 + j 0.0169 0.1451 + j 0.0634
 V. CONCLUSIONS [5] V. Miranda, M.A. Matos and J.T. Saraiva, “Fuzzy Load Flow – New
Algorithms Incorporating Uncertain Generation and Load
The necessary inclusion of slack bus in the load flow Representation”, in Proc. 10th Power Systems Computation Conference,
problem definition has an inherent drawback when dealing Graz, Austria, 1990, pp. 621-627.
with uncertain nodal powers. While serving its purpose of [6] V. Miranda and J.T. Saraiva, “Fuzzy Modelling of Power System
Optimal Power Flow”, IEEE Trans. on PWRS, Vol. 7, No. 2, May 1992,
balancing powers in the system, it also absorbs all pp. 843-849.
uncertainties. The result is a solution that is usually of no [7] M.E. El-Hawary, editor, Electric Power Applications of Fuzzy System,
practical interest. To overcome this problem, we have New York: IEEE Press, 1988, pp. 232-234.
investigated two ways of treating the slack bus so that the [8] A. Dimitrovski and K. Tomsovic, “Boundary Load Flow Solutions”,
IEEE Transactions on Power Systems, Vol. 19, No. 1, Feb. 2004, pp.
solution obtained also satisfies its’ constraints. In the 348-355.
boundary load flow context both methods should give [9] A. Dimitrovski and K. Tomsovic, "Risk Assessment using Boundary
approximately the same results since the objective is that of Load Flow Solutions," Proceedings of the 2003 International
the same global constrained optimum. Still, this very much Conference on Intelligent System Applications to Power Systems,
Lemnos, Greece, Sept. 2003, pp. 1-6.
depends of the actual implementation within the BLF [10] http://www.ee.washington.edu/research/pstca/
algorithm as the already difficult task is further complicated
with an inclusion of yet another constraint. We are VII. BIOGRAPHIES
investigating further methods to improve the robustness of
these methods and make them applicable to planning practical Aleksandar Dimitrovski received his B.Sc. and Ph.D. in Power Engineering
large scale systems. from University "Sv. Kiril i Metodij" in Skopje, Macedonia, and M.Sc. in
Computer Science Application from University of Zagreb, Croatia. He is an
Assistant Professor in power systems at the University "Sv. Kiril i Metodij",
currently on leave at Washington State University. His subjects of interests
VI. REFERENCES include advanced computing techniques in power system analysis.
[1] R.N. Allan, A.M. Leite da Silva and R.C. Burchett, “Evaluation Methods
and Accuracy in Probabilistic Load Flow Solutions”, IEEE Trans. on Kevin Tomsovic received the BS from Michigan Tech. University, Houghton,
PAS, Vol. PAS-100, No. 5, May 1981, pp. 2539-2546. in 1982, and the MS and Ph.D. degrees from University of Washington,
[2] R.N. Allan and A.M. Leite da Silva, “Probabilistic load flow using Seattle, in 1984 and 1987, respectively, all in Electrical Engineering. He is
multilinearisations”, Proc. IEE, Vol. 128, Pt. C, No. 5, Sep. 1981, pp. currently a Professor in the School of Electrical Engineering and Computer
280-287. Science at Washington State University. Visiting university positions have
[3] A.P. Meliopoulos, G.J. Cokkinides and X.Y. Chao, “A New included Boston University, National Cheng Kung University, National Sun
Probabilistic Power Analysis Method”, IEEE Trans. on PWRS, Vol. 5, Yat-Sen University and the Royal Institute of Technology in Stockholm. He
No. 1, Feb. 1990, pp. 182-190. held the Advanced Technology for Electrical Energy Chair at Kumamoto
[4] A. Dimitrovski, “Probabilistic Load Flow in Weakly Meshed University in Japan from 1999-2000.
Distribution Networks”, PMAPS '97 Conference, Vancouver, Canada,
September 21-25, 1997, pp. 521-527