Você está na página 1de 5

Computation of Voltage Stability Margins of

Distribution Systems
C.H. Fujisawa and C.A. Castro, Senior Member, IEEE

Abstract- A robust and efficient tool for computing the voltage


stability margin of distribution systems is proposed in this
paper. This margin corresponds to the distance from the current
operating point to the maximum loading point in load parameter
space along a predefined load increase direction. This tool is based
on computing the voltage stability index of a two-bus equivalent
circuit for predicting load increases and a specialized, robust
Newton's load flow method. It may also be useful for evaluating
the impacts of the insertion of distributed generation on the
system operation and voltage stability conditions. The proposed
methodology has shown to be suited for environments subjected
to severe computation time constraints, such as distribution
management systems.
Index Terms- Distributed Generation, Distribution Management Systems, Distribution Systems, Maximum Loading, Voltage
Stability.
I. INTRODUCTION

OR many years distribution systems (DSs) delivered


power with little or no analysis. As a result, DSs were
typically overdesigned [1]. Lately, there has been a remarkable
improvement of models, analysis methods and operational
techniques for DSs as it has ocurred in the transmission
systems (TSs). A direct consequence is the possibility of
operating DSs at their maximum capacities. In other words, the
continously growing demand can be supplied by the system
through the better use of existing equipment, postponing
expansion investments. Power systems restructuring and deregulation have transformed this possibility into a necessity. There
has been a shift in business goals, including the maintenance
of quality standards while maximizing the use and life of
assets through improved monitoring and analysis [2]. Current
Distribution Management Systems (DMSs) involve the use of
advanced technologies in terms of equipment and software.
The goal of this paper is to present a contribution to the
development of efficient software for DMSs.
Operating DSs close to their maximum capacities, as it already happens in TSs, implies in increased risk of instabilities,
including voltage instabilities [3]. Currently, most DSs operate
with comfortable voltage stability margins (VSM). However,
this situation tends to change as load demand increases and
DS equipment are stressed. Actually, voltage instability events
have already been observed in industrial areas under critical
loading conditions [3], [4].
Methods specifically developed for computing the VSM
according to DS characteristics can be found in the literature,
C.H. Fujisawa and C.A.Castro are with the University of Campinas,
Brazil. The authors thank the financial support from Brazilian funding agency
FAPESP. Emails:{cassiofujisawa@gmail.com. ccastro@ieee.org}

978-1-4244-2218-0/08/$25.00 2008 IEEE.

as in [4], [6]. In [4] a voltage stability index for identifying


the critical bus with respect to voltage collapse was proposed.
In [6] a method for computing the maximum admissible
loading of DSs based on a voltage stability index and an
equivalent two-bus system was proposed. This method shows
very interesting characteristics in terms of efficiency and has
great potencial for practical use in DMSs, as it will be shown
in the following section.
The need for robust and efficient methods to assess the
voltage stability condition of DSs is recognized. In this paper,
a method is proposed that combines the interesting ideas of
[6] and the robustness and efficiency of the load flow method
with step size optimization (LFSSO) proposed in [9], [10].
The simulation results show that the proposed method
achieves the requirements of robustness and efficiency. Good
results were also obtained for DSs with the insertion of
distributed generation, situation that has become very common
lately.
II. BASIC IDEAS

A. Method I: Computing the maximum loading point of DSs


The method proposed in [6], hereafter referred to as method
I, has the objective of computing the maximum loading point
of a DS. The idea is to obtain the maximum load factor that
multiplies the current real and reactive load powers such that
the operation of the system is still stable. Fig. 1 shows a simple
two-bus example system, where the substation bus s, whose
voltage is Vs , supplies the load bus e through a distribution
line whose impedance is r+ jx. The real and reactive demand
powers at the load bus are Pe and Qe, respectively. For the
current operating conditions, that will be referred to as the
base case, the load factor A is equal to one.

o
Fig. 1.

Simple two-bus example system.

Considering the base case (,x == 1), the voltage stability


index SIe associated to load bus e is computed in [6] as

SIe == Vs4

4Vs2 (rPe + xQe) - 4 (xPe - rQe)2 .

(1)

Index SIe decreases as the loading increases (by increasing


load factor 'x). The goal is to obtain the largest value of ,x,

represented by A*, that still allows a stable operation of the


system. At this point the voltage stability index S Ie is equal
to zero. It is shown in [6] that A* can be computed directly as
A*

{ -rPe - xQe

= v2

+ J[(r 2 + x 2 ) (P; + Q~)]}

[2 (xP

e -

rQe)2]

(2)

For other operating points (A =I 1), Eqs. (1) and (2) are
used by replacing Pe and Qe respectively by APe and AQe.
It was also shown in [6] that it is possible to obtain a
two-bus equivalent system from any radial DS, by keeping
the substation bus s and some particular load bus k. The
impedance of the equivalent line that connects bus s to k is
given by
(3)
where Z sk is the equivalent impedance between the substation
bus s and the bus k, f is the set of branches of the path that
connects bus s to bus k, Jk is the current injected at bus k, and
Zi and Ji are the impedance and the current through branch
i.
In method I, A* is iteratively obtained for any radial DS.
The procedure can be summarized as follows.
(a) Compute the voltage stability index for all load buses
using (1).
(b) Obtain the two-bus equivalent system containing the
substation and the weakest bus (which has the smallest
voltage stability index), using (3).
(c) Compute the maximum load factor for the equivalent
system using (2).
(d) Perform steps (a)-(c) until the maximum load factor
converges to A*.
For the sake of illustrating the performance of method I, Fig.
2 shows the evolution of the iterative process for the 33-bus
distribution system [5].
380,----------.------.---...---------r------.------.-----r-------.----.

370

~360

"'"
B

u 350

~
""0

...J 340

I
I

330

II
I

- - -

Maximum loadability
Method I

320 L---L_--I.---=:::i::::===::L:::=:::::==:t:===::::::::t:===:t====:::L==::J
1
5
6
8
10
1

Fig. 2.

Iteration
Method I - 33-bus distribution system.

After exhaustive simulations using method I for several DSs,


the following comments can be made.
For DSs without distributed generation, the iterative process always converges to a load factor smaller than the

real value. Fig. 2 shows an example of such a situation,


where the real maximum loading is 370.77% and method
I converges to 361.36%, with a final error of 2.54%.
Even though these errors are small for most cases, there
is no control over them. In other words, there may
be cases, for certain conditions, where this error could
be unacceptable. The accuracy of method I could be
improved if there were some control over the final error.
The penetration of distributed generation has been increasing significantly [7]. For DSs containing independent generators with voltage regulation, method I may
provide load factors larger than the real value, resulting in
infeasible operating points and divergence of the conventionalload flow methods. This aspect was not considered
in [6]. An example of such a situation will be shown
later. The robustness of method I could be improved if the
infeasibility situations were dealt with in an appropriate
manner.
This paper has the purpose of making a contribution to
solving the two points discussed above.

B. Load.flow method with step size optimization (LFSSO)


In general terms, the load flow method with step size
optimization (LFSSO) consists of controlling the step size of
the Newton's iterative method by updating the voltages using
a scalar, usually referred to as the optimal multiplier. This
idea was initially proposed in [8], where the nodal powers
and the voltages were represented in rectangular coordinates.
The LFSSO in which the voltages are represented in polar
coordinates, being easily incorporated to production grade load
flow programs, was proposed in [9]. Recently, the efficiency
of LFSSO was validated in [10].
The basic equation of LFSSO at iteration n is
1

D.() ] n+ == [ D.() ] n
.
D.V
+M
[ D.V

-1 n

(J)

D.P ] n+
D.Q
' (4)

where D.() and D. V are the voltage phase angle and magnitude
corrections, D.P and D.Q are the real and reactive power mismatches, and J is the Jacobian matrix. The optimal multiplier
J1 is added to the conventional load flow formulation, being
computed to minimize a quadratic function based on the power
mismatches. The derivation of M is shown in [9], where it is
clear that second order information is used to improve the
performance of the load flow method.
For well-conditioned systems, M is close to one, and does
not affect the iterative process significantly. For ill-conditioned
systems, M assumes different values to minimize the illconditioning effects, providing the correct solution for the
problem. For infeasible systems (those for which there are
no solutions for the load flow equations) M tends to zero in
a few iterations, indicating that the power mismatches cannot
be decreased further and the best possible solution has been
reached. Also, Overbye [11] showed that, for infeasible cases,
LFSSO leads to a point on the feasibility boundary rather than
to simply diverge. These features are of particular interest for
voltage stability.

The behavior of LFSSO can be better understood with the


help of Fig. 3, where a two-bus example system is shown. Fig.
3 also shows the parameter (load) space for this system, where
points A, Band C correspond respectively to the base case, an
infeasible point and the operating point provided by LFSSO in
case point B were specified. The dashed line which connects
points A and B corresponds to a pre-determined direction of
load increase that could be, for instance, related to maintaining
a constant power factor (the real and reactive powers increase
proportionally). ~ is the boundary that separates the feasible
(internal) and infeasible (external) regions. Note that point C
provided by LFSSO, in case point B were specified, is on the
feasibility boundary but not necessarily on the dashed line that
indicates the direction of the load increase. Experience shows
that point C is always within the close vicinity of the dashed
line.

1------1---

Substation

CD

Load

P2

+ jQ2

A.-'

Fig. 3. Example of two-bus system and its respective parameter (load) space.

Incorporating LFSSO into method I provides the robustness


and efficiency discussed earlier in the Sec. II-A, as shown in
the next section.

C. Proposed method
The proposed method comprises three basic steps: load
increase, load shedding and binary search.
In the first step, method I [6] is used to obtain the values
of the load factor ,\ corresponding to the load increase. Since
LFSSO is very robust and deals properly with infeasible cases,
the idea is to use its powerful characteristics to increase the
load very quickly, intentionally seeking a load factor larger
than the real maximum loading (in the infeasible region), but
not too far from it. In order to achieve this goal, method I is
slightly modified, by multiplying the computed load factor in
(2) by an arbitrarily chosen scalar. In this paper, this scalar was
set to 1.1 (in other words, the load increase will be 10% larger
than the one determined by method I). This step is carried out
iteratively until an infeasible point is found, for which LFSSO
does not converge (M tends to zero), but provides an operating
point on the feasibility boundary.

In the second step, the best solution given by LFSSO, which


is an operating point on the feasibility boundary, is used as
a starting point for a load shedding procedure. Then, each
computed power (on the boundary) is divided by its respective
base case power, providing a load factor very close to the
real maximum loading. However, a more representative value
for the whole system is obtained by computing the average
of these computed values for all load buses. Three kinds of
values that are inadequate are eliminated from this average
calculation, namely the values corresponding to computed
powers that are larger than the specified values (in module),
since they indicate that they are not on the feasibility boundary
but in the infeasible region; positive values, since they indicate
that the respective buses contain generation rather than load;
and values for which the specified powers are equal to zero,
since it results in division by zero. As the idea is to obtain a
value of the load factor in the feasible region, but very close
to the feasibility boundary, the computed average is multiplied
by an arbitrary scalar (in this paper this scalar is equal to
0.9, corresponding to a 10% reduction). This step is carried
out iteratively until a load factor within the feasible region is
obtained.
In the third step, the smallest load factor corresponding to
an infeasible point (obtained in the first step) and the largest
load factor corresponding to a feasible point (obtained in
the second step) are used to start a binary search procedure,
which consists of computing a load factor halfway from these
two initial points. Using LFSSO, it is verified whether this
new value is within or without the feasible region. In case
this point is feasible, it replaces the old feasible value. If it
is infeasible, it replaces the old infeasible value. This step
is carried out iteratively until the value of the load factor
converges according with a desired precision. In this paper, the
convergence criteria of the binary search process was based on
the difference between the last two load factors. The process
is interrupted in case this difference becomes smaller than 1%.
Note that there is now a control over the final error.
After computing the maximum load factor, the voltage
stability margin is easily computed as the difference from this
recently obtained value and the load factor of the base case.
The proposed method is described below.
(a) Compute the voltage stability index for all load buses,
using (1).
(b) Obtain a two-bus equivalent system containing the substation and the weakest bus (which has the smallest
voltage stability index), using (3).
(c) Compute the maximum load factor of the equivalent
system with (2) and multiply it by a scalar to force the
loading even further. In this paper, this scalar was set to
1.1 (loading forced 10% further).
(d) Perform steps (a) to (c) until the new operating point is
infeasible (LFSSO provides J1 == 0). Then, proceed to
step (e).
(e) From the power mismatches provided by LFSSO, decrease the load factor (curtail load), and multiply it by
a scalar to force the loading to fall into the feasibility
region. In this paper, the scalar was set to 0.9. Repeat
this step until a feasible point is obtained.

(f) From the most recent feasible point and the last infeasible point, perform binary search until a point on the
feasibility boundary is obtained. A threshold is defined
to stop the binary search.

Step (e) corresponds to computing a load factor from the


power mismatches given by LFSSO in an infeasible situation.
The procedure is detailed below.
(1) Initialize () == 0 and n == O.
(2) For each load bus of the system whose consumption is

not null do
(2.1) Divide the real and reactive powers by their respective base case values, obtaining an estimate for the
load factor.
(2.2) Add the computed values to () and increment n,
except
(i) if the computed real (and/or reactive) power is
larger than its respective specified power, or
(ii) if the real (and/or reactive) power is null or
generated.
(3) Compute the average of the n estimated valid load
factors, obtaining .xest == () In.
III.

TESTS AND RESULTS

Simulation tests have been carried out for several distribution systems. In this paper the results for a 33-bus system and
a 202-bus system will be shown only, due to space constraints.
Also, the results for other systems lead to the same conclusions
as far as the performance of the proposed method is concerned.
The data of the 33-bus system can be found in [5]. The 202bus system is a realistic primary distribution system in the
State of Sao Paulo, Brazil.
Simulations with the 33-bus system considered two scenarios, namely (a) considering the original data, without
distributed generation, and (b) with the insertion of an independent generator of 2MW, with regulated voltage at 1.0pu,
and reactive power limits of -O.8MVAr and 1.0MVAr at bus
12. The evolution of the load factors obtained by applying
method I [6] can be observed in Fig. 2. Figs. 4 and 5 illustrate
the results provided by the proposed method, respectively
without and with distributed generation. Tables I and II show
the same results in detail.
From Figs. 4 and 5, note that a few iterations are necessary
(six and eight) for reaching a load factor very close to the
expected one with errors smaller than 1%. The values in
italics in Tables I and II indicate load factors larger than
the expected one. The presence of distributed generation with
voltage regulation leads method I to estimate large, infeasible
load factors. The proposed method provides an appropriate
way to manage this situation, and an accurate solution is found.
The 202-bus system was also tested considering two scenarios, namely (a) without distributed generation, and (b) with the
insertion of an independent generator of 14MW, with regulated
voltage at 1.0pu, and reactive power limits of -5.5MVAr and
6.8MVAr at bus 202. The evolution of the load factors for the
two scenarios provided by the proposed method can be seen
in the Fig. 6. Table III shows the same results in more detail.

395,--------.------,-------,.---------.--------,
...
1\

390

\
\

385

~380

....

1:) 375
~

'.

"'0

"-

~ 370

"-

I
I

365

"-

"-

"-

360

355

Fig. 4.

Maximum loadabI1lty
Proposed Method

1=-=-=

Iteration

Proposed method: 33-bus system without distributed generation.

520
\
\

500 \
\

~480
.s

~ 460

\
\

"'0

~ 440

i\

\
\\

~---.-\-----,...------:\--~-..::.--,--__=__----<'1-~______=4

420

400

\/ ./

\/ 1/

\/

Maximum loadablhty
- ..- - Proposed Method

380 L-._L_---L-===::L:::::::::::=:L::===:L====~==::J
1

Fig. 5.

Iteration

Proposed method: 33-bus system with distributed generation.


TABLE I

COMPARISON OF METHODS FOR THE 33-BUS SYSTEM WITHOUT


DISTRIBUTED GENERATION.

M
'
I oa d'IDe = 37077%0
aXlmum

Iteration
1
2
3
4
5
6
7
8
9
10

Method I
Error
A
322.94%
-12.95%
-06.730/0
345.80%
-04.85%
352.790/0
-03.96%
356.10%
357.82,/0
-03.490/0
-03.200/0
358.900/0
-02.91 %
359.990/0
360.450/0
-02.780/0
-02.660/0
360.91%
361.36%
-02.540/0

Proposed Method
Error
A
-4.19%
355.230/0
393.08%
6.020/0
374.15%
0.910/0
364.690/0
-1.640/0
369.420/0
-0.360/0
371.79%
0.280/0

From Fig. 6 and Table III it is possible to see that the


proposed method converges in a few iterations (six) to reach
the maximum load factor with good precision (error smaller
than 0.5%). Besides, it converges to the real value of maximum
loading. Specially in this case, the insertion of distributed
generation does not affect the performance of the method in
a significant way, since the maximum load factors for both
cases are very close to each other.

TABLE II
COMPARISON OF METHODS FOR THE 33-BUS SYSTEM WITH DISTRIBUTED
GENERATION.

Iteration
1
2
3
4
5
6
7
8

V.

Maximum loadin2 = 427490;.0


Proposed Method
Method I
Error
Error
.x
.x
21.92%
10.83%
521.18%
473.80%
-11.09%
380.070/0
5.41%
450.63%
-2.84%
415.35%
1.29%
432.99%
424.17%
-0.78%
0.25%
428.58%
426.37%
-0.260/0

690~------r-----r----.----,------.,

680
670

~660

B
650
u

r.a 640

"'0

ro
o

630
620
610

-e-

Maximum loadability
Proposed Method
Maximum loadability (DG)
Proposed Method (DG)

600'---------'----...L..-----'------'----~

Iteration

Fig. 6.
Proposed method: 202-bus system without and with distributed
generation (DG).
TABLE III
EVOLUTION OF THE PROPOSED METHOD FOR THE 202-BUS SYSTEM.
Without distributed
generation

With distributed
generation

A*=632.72%

A*=634.730/0

Iteration
.x
Error
1
623.52%
-1.450/0
686.10%
8.440/0
2
654.81%
3.49%
3
4
639.17%
1.02%
631.340/0
-0.22%
5
6
635.25%
0.400/0
A* IS the maXImum loadmg.

efficient procedure.

.x

625.24%
688.03%
656.63%
640.94%
633.090/0
637.01%

Error
-1.50%
8.400/0
3.450/0
0.98%
-0.26%
0.36%

IV. CONCLUSION

A robust and efficient method for computing the maximum


loadability of distribution systems was proposed in this paper.
The proposed method presents a good performance in terms
of computational effort and precision, since the number of
iterations was low and the errors were smaller than 1%. It
is important to point out that the proposed method allows
the control over the final error. Its robustness was demonstrated through simulation results considering the presence of
distributed generation. The utilization of LFSSO along with
method I was crucial, since situations of infeasibility in the
load flow were dealt with in an appropriate manner. Actually,
the use of LFSSO allowed the intentional search for infeasible
points during the computational process, resulting in a very

REFERENCES

[1] W.H. Kersting, Distribution System Modeling and Analysis, CRC Press,
2nd ed., 2007.
[2] 1. Northcote-Green and R. Wilson, Control and Automation of Electrical
Power Distribution Systems, Taylor & Francis, 2007.
[3] G. Brownell and H. Clark, Analysis and Solutions for Bulk System
Voltage Instability, IEEE Computer Applications in Power, 1989.
[4] M. Chakravorty and D. Das, Voltage stability analysis of radial distribution networks, Electrical Power & Energy Systems, vol.23, 2001.
[5] D. Das, Maximum Loading and Cost of Energy Loss of Radial Distribution Feeders, Electrical Power & Energy Systems, vol.26, 2004.
[6] A. Augugliaro, L. Dusonchet, S. Favuzza, M.G. Ippolito and E.R. Sanseverino, A Simple Method to Assess Loadability of Radial Distribution
Networks, IEEE Powertech, S1. Petersburg, Russia, 2005.
[7] T. Ackermann, G. Andersson and L. Soder, Distributed generation: a
definition, Electric Power Systems Research, vol.57, 2001.
[8] S. Iwamoto and Y. Tamura, A Load Flow Calculation Method for 111Conditioned Power Systems, IEEE Trans. on Power Apparatus and Syst.,
vol. PAS-I00, 1981.
[9] L.M.C. Braz, C.A. Castro and C.A.F. Murari, A critical evaluation of
step size optimization based load flow methods, IEEE Trans. Power Syst.,
vol. 15, 2000.
[10] J.E. Tate and TJ. Overbye, A comparison of the optimal multiplier in
polar and rectangular coordinates, IEEE Trans. Power Syst., vol.20, 2005.
[11] T.J. Overbye, A power flow measure for unsolvable cases, IEEE Trans.
Power Syst., vol.9, 1994.

Você também pode gostar