1

Abstract In this paper a method for optimal placement of

Static Var Compensator (SVC) to maximize power system
loadability is presented. The method is based on mixed integer
linear programming (MILP) and line flow based equations are
used in the formulation. This proposed method has the ability to
not only determine the location and settings of SVCs, but to
minimize the required compensation as well. The method is
tested on the IEEE 9 bus system as well as a 30 bus test system
and successfully determines the number of devices, along with
their locations and device settings. As part of ongoing
investigations the authors are aiming to verify this proposed
method with a realistic large scale Australian HV electricity
transmission network.
I. INTRODUCTION
N recent years electricity transmission networks around the
world have been put under increasing strain to meet the
increased load demands of consumers. The move to
deregulation has also led to networks being operated ever
closer to their limits. Increased strain in the system leads to
increased risk of facing instability issues such as voltage
collapse or power oscillations.
One method to reduce the strain on transmission systems is
to construct new transmission lines. This approach however is
costly and often too difficult to obtain right-of-way for new
lines. Another method which has come into its own over the
last decade or so is the use of Flexible AC Transmission
System (FACTS) devices [1]. These devices allow system
parameters such as line impedance, bus voltage and angles to
be controlled which in turn allows for better utilization of the
transmission network through directing power away from
heavily loaded lines and into lightly loaded lines [2]. They are
based on power electronics and can be switched in and out
much more dynamically compared to traditional fixed devices.
Because of this they are also capable of providing damping
contributions to reduce system oscillations.
The Queensland HV transmission network is an example of
such a network which utilizes FACTS devices. This network
employs SVCs to provide reactive power compensation, thus
improving transfer capacity and increasing voltage stability


This work was supported by the Australian Research Council Linkage
Project. R. W. Chang ([email protected],au) and Tapan K Saha
([email protected]) are with the School of Information Technology &
Electrical Engineering, The University of Queensland, St. Lucia, QLD-4072,
Australia.
margins. A few of these SVCs are also fitted with
supplementary controllers to contribute to the damping of
inter-area power oscillations.
To maximize the benefits provided by FACTS devices
several aspects must be carefully considered when employing
this technology:
What type of FACTS device should be used?
How many devices should be installed?
Where should these devices be located?
What should be their initial settings?
These questions can be answered according to a variety of
objectives, such as improving power transmission capacity or
improving system stability margins. To answer these
questions many approaches have been proposed in the
literature in recent years [3-9].
Due to the non-linear, non-convex nature of these
placement problems the majority of research focus has been
placed on evolutionary algorithm (EA) based solutions.
Gerbex et al. [3] applies genetic algorithms (GA) to find
optimal numbers and locations of multi-type FACTS devices
to maximize loadability. The formulation however, does not
consider investment costs and computation times are not
specified. Ongsakul et al. [5] uses improved evolutionary
programming to locate multi-type FACTS devices for
maximizing total transfer capacity and minimize system real
power loss. The simulation time for this approach is reported
to be over an hour for a 58 bus system. This highlights a
common disadvantage of EA based techniques their
relatively high computation requirements which make them
less suited for handling larger real world power systems
consisting of hundreds of buses.
One alternative technique is to formulate the problem using
mixed integer linear programming (MILP) where possible
locations are represented by binary variables. The efficiency
of this approach has improved significantly since advances in
branch and bound algorithms were made in recent years [10].
Lima et al. [11] utilizes MILP for TCPST allocation to
provide a fast solution for complex systems. However, the
system is modeled with the DC load flow model which
ignores reactive power balance, tap dependence of transformer
reactances as well as reactive power losses [12]. Yang et al.
[13] proposes a MILP approach for TCSC placement using
line flow based (LFB) power flow equations. The approach
overcomes the shortcomings of DC load flow models and the
authors have extended the work in [9] using a quadratic
objective function. The proposed technique however is only
Maximizing Power System Loadability by
Optimal Allocation of SVC using Mixed Integer
Linear Programming
R. W. Chang, Student Member, IEEE, T. K. Saha, Senior Member, IEEE
I
978-1-4244-6551-4/10/$26.00 2010 IEEE
2
applied to TCSC placement and the problem must be
reformulated to consider alternative device types.
This paper presents a method for SVC allocation using
MILP. The approach of linearizing LFB power flow equations
first presented by Yang et al. [13] has been adopted here and
an additional term is introduced to the objective function to
provide a better indication of optimal device number. The
objective of this proposed method is to maximize loadability
by simultaneously determining device locations and initial
settings and minimizing compensation requirements. This
approach provides a better solution in terms of finding the
ideal number of devices.
This paper is organized as follows: section II presents a
summary of mixed integer linear programming along with line
flow based equations. In section III the formulation of the
SVC allocation problem is described. Preliminary case studies
are performed on the IEEE 9 bus system and a 30 bus test
system which are described in section IV and the results are
presented in section V. Finally, conclusions and future work
are summarized in section VI.
II. SUMMARY OF MILP AND LFB EQUATIONS
The SVC allocation problem is formulated as a mixed
integer linear programming problem where device locations
are designated by binary integer variables. The constraint
equations for this formulation consists of steady state power
flow constraints as described by Ping et al. [14]. Summaries of
MILP and the LFB equations are provided below.
A. Mixed-Integer Linear Programming
A MILP problem is one where certain variables are
restricted to having only integer values. MILP has been
applied to location problems in the past [11, 13] and has the
advantage of being computationally more efficient compared
to EA based techniques. The mathematical form of a
generalised mixed integer programming problem is [15]:


( ) x c Qx x x f
T T
x
+ =
2
1
min
(1)
subject to
max min
x x x
(2)

max min
b Ax b
(3)

i
x integer I i
(4)

where x , c ,
min
x ,
max
x
n
, Q
n n
, A
n m
,
min
b ,
max
b
m
and I is the set of integer variable indices. For non-zero
matrices of Q , a mixed integer quadratic programming
(MIQP) problem is formed, while with Q as zero the problem
becomes a MILP one.
B. Line Flow Based Equations
To model power flow constraints LFB equations introduced
by Ping et al. [14] are used here. As mentioned previously this
representation provides better accuracy compared to the
common DC load flow models which do not take into
consideration reactive power balance or real and reactive
power losses.
For the remainder of this paper the following notations will
be adopted:
bus
n will refer to the number of buses present in
the system while
br
n and
gen
n will denote the number of
branches and generators respectively.
The LFB power flow equations are presented below:


GL
P l A p A = +
(5)

GL
Q V H m A q A = +
2
(6)

k V A A T q X p R
PQ
T
PQ
T
PQ
+

+ +
+ 2 2
2 2
2 2
PV
T
PV
V A T = (7)

where
A Bus incidence matrix
A Modified A matrix all elements of -1 set to zero
PV
A Bus incidence matrix corresponding to PV buses
+
PQ
A Modified A matrix corresponding to PQ buses all
elements of -1 set to zero

PQ
A Modified A matrix corresponding to PQ buses all
elements of +1 set to zero
R Diagonal matrix of line resistance
X Diagonal matrix of line reactance
H Diagonal matrix of bus charging and compensating
susceptance
2
T Diagonal matrix of line transformer tap values squared
GL
P Vector of bus active power injection (generation minus
consumption)
GL
Q Vector of bus reactive power injection
2
V Vector of bus voltage magnitude squared
2
PV
V Vector of PV bus voltage magnitude squared
2
PQ
V Vector of PQ bus voltage magnitude squared
p Vector of active power flow at line receiving end
q Vector of reactive power flow at line receiving end
l Vector of active power loss in each line
m Vector of reactive power loss in each line
k Vector of composite variables, one for each line

The bus incidence matrix A (not to be confused with the
constraint matrix in (3)) is a matrix of size
br bus
n n which
describes branch connections in the system. Each branch in
the system has a corresponding index and it is assumed that
each branch also has an associated direction which is
arbitrarily set. For example, an entry in the A matrix could be
as follows: for a given system where line number five leaves
bus number three and enters bus number six
35
A would have a
value of 1 and
65
A will have a value of -1. Elements not
associated with a bus or branch remain as zero so that the
incidence matrix is almost always sparse.
Equations (5) and (6) above describe the real and reactive
power balance in a system with respect to line power flows
and losses and power generated at each bus. Equation (7)
relates to the line voltage drop equations in a system. In a
3
given system there will be exactly
bus
n equations in each of
(5) and (6) and
br
n equations in (7). The formulas for
calculating l , mand k are given below:


ij
j
ij ij
ij
r
v
q p
l
2
2 2
+
=
(8)

ij
j
ij ij
ij
x
v
q p
m
2
2 2
+
=
(9)

( )
2 2
2
2 2
ij ij
j
ij ij
ij
x r
v
q p
k +
+
=
(10)

where subscripts i and j indicate sending end and receiving
end buses respectively.
III. SVC ALLOCATION PROBLEM FORMULATION
A. Steady State SVC Model
The focus of this paper will be on the optimal placement of
SVCs. For the problem formulation an SVC is modelled in
steady state as a variable shunt susceptance. This model is
shown in the two bus transmission line diagram of Fig 1.
Modeling the SVC as a voltage regulating reactive power
source will be done in the next step of this ongoing work.


Fig. 1. Ideal model of SVC installed at bus j

The variable vectors and with lengths of
bus
n are
used in the formulation. For the purpose of SVC placement,
j
is a binary integer variable which when set to 1 indicates
an SVC located at bus j . The variable
j
is the compensating
susceptance setting at bus j . For the rest of this paper the
notation will represent the element-wise multiplication of
these two vectors.
B. MILP Constraint Equations
Equations (11)-(13) describe the line power flow and
voltage equations as equality constraints. As in [13] a load
multiplier or loadability factor denoted by is introduced,
which describes the increase in active and reactive loads. The
objective of the optimization is to maximize the load which
can be supplied, or in other words to maximize .

0
2
= + + +
L G
P P GM l A V S p A (11)
0
2
= + +
L G
Q P GM m A V H q A (12)
0
2
1
2
2
1
= + + k V CM q X p R (13)

where
G G
Q P , Vectors of generator active and reactive power
generations
L L
Q P , Vectors of bus active and reactive power
consumptions
GM Coefficient matrix to adjust dimensions of the
G
P
and
G
Q vectors
CM Coefficient matrix combining the effects of terms
(
T
PQ
T
PQ
A A T
+
+
2
) and (
T
PV
A T
2
) in equation (7)
Due to the linear nature of the formulation non-linear
equations (8)-(10) cannot be represented as they are. Through
algebraic manipulations linear relationships can be obtained
from these equations:

0 = m R l X (14)
0 = + k m X l R (15)

To introduce an SVC at a given bus in the form of a
variable shunt impedance, alterations must be made to the H
matrix of equation (12). This H matrix is diagonal and each
diagonal element
j
h corresponds to the sum of line charging
susceptances
2
ij
y
of every branch connected to bus j . With
the addition of an SVC the compensating susceptance must
also be added to form a new element
j
h . Effectively the bus
shunt susceptance value becomes:

j j j j
h h + =
(16)

Substituting (16) into (12) gives:

0
2 2
= + + V Q P GM m A V H q A
L G
(17)

The term
2
V represents the element-wise multiplication
of
2
V and . This quadratic product must be converted to
linear relationships for a MILP formulation. This is done by
using the limits of
2
V to obtain two linear inequality
constraints:

0
2
max
2
+ + V Q P GM m A V H q A
L G
(18)
0
2
min
2
+ + V Q P GM m A V H q A
L G
(19)

For this approximated relationship it is assumed that only
buses lacking reactive power in steady state require
compensation so that only capacitive compensation is required
4
and the value of is always positive.
The non-linear element-wise product of and integer
vector can be replaced by four linear constraints following
the binary variables theorem [16]. The product of is
represented by the variable z and the constraints are as
follows:


0
max
z
(20)

0
min
+ z
(21)

min min
z
(22)

max max
+ + z
(23)

Equations (11), (13)-(15) and (18)-(23) represent the linear
constraints represented by equation (3) of the MILP problem.
The vector x contains variables as follows:

{ } z k m l Q P V q p x
G G
T
, , , , , , , , , , ,
2
=

This vector of variables is subject to the bounds in (2). The
vector is a vector of binary integer variables.
It should be noted that additional constraints on the limits of
p and q exist:


2 2
max
2
max T
R q p + (24)

where
T
R is the transmission line thermal ratings. This non-
linear constraint is implemented here by searching for
solutions iteratively. Each iteration,
max
p and
max
q are
reduced for lines which violate this constraint.
Additional constraints may be added from a practical
perspective:

max
N A , { }
bus
n N ,..., 2 , 1 , 0
max
=
(25)

max
C z A
(26)

where
max
N is the maximum allowed device number and
max
C is the maximum investment cost. Vector is the
associated investment costs at each bus. These constraints,
(25) and (26) limit the device numbers and total investment
costs respectively. For this initial study the investment
constraints are not yet considered and will be investigated in
ongoing work.
C. Objective Function
The objective function in this optimization problem is to
maximize , the loadability factor. This objective function is
defined as:


( )

= z x f max
(27)

It has been shown that for a given system loadability does not
improve beyond a certain limit produced by a number of well
located devices [3, 13]. This limit was found in [13] by
iteratively adjusting the device number limits and observing
the point where loadability begins to level off. By introducing
the additional term

z in the objective function (which

minimizes the total compensating reactance) the same can be
achieved with a single run of the algorithm. The factor
must be set sufficiently small so as not to adversely affect the
optimal solution of .
It should be noted that due to the approximate nature of this
problem formulation the proposed method is more suitable for
initial analysis in power system design and results need to be
further verified by a full AC model, which takes security and
stability issues into consideration.
IV. CASE STUDIES
Using the above formulation test studies were performed on
the IEEE 9 bus system and a 30 bus test system to determine
optimal SVC allocations. Data for these two test systems are
taken from the MATPOWER toolbox [17]. The 9 bus system
consists of nine branches, three generators and three load
buses with total real and reactive consumption of 315 MW
and 115 MVAR respectively. The 30 bus system consists of
41 branches, six generators and 20 load buses with a total
consumption of 189.2 MW and 107.2 MVAR. Reactive
compensation is also present in bus 5 (19 MVAR) and bus 24
(4 MVAR). Single line diagrams of the 9 bus and 30 bus
systems are shown in Fig. 2 and Fig. 3 respectively.
It would be worthwhile to perform case studies on larger
systems as in practical terms these two test systems have
relatively few buses. The authors intend to demonstrate and
verify the proposed technique on the Queensland HV
electricity network, a system with over 600 buses. However,
at this stage the study has not yet been completed but is part of
ongoing research. Results from this Queensland case study
will be presented at a later stage.
The MILP problems were solved using TOMLAB/CPLEX
12.1 [18, 19] and results are presented in the next section. In
finding these results the allowable range of reactive
compensation
i
for each SVC was set to between 0.5 MVAR
and 50 MVAR.


Fig. 2. Single-line diagram of 9 bus test system

5

Fig. 3. Single line diagram of 30 bus test system

V. RESULTS AND VERIFICATION
A. Simulation Results
For the 9 bus system a total of four SVCs were allocated for
a maximum loadability factor of 2.4108, with a combined
reactive compensation of 165 MVAR (based on 1 pu voltage
calculations). The 30 bus system reached maximum
loadability of 1.7648 with five devices and a total
compensation of 44.87 MVAR. Fig. 4 and Fig. 5 illustrate the
optimal device distribution on these two test systems.


Fig. 4. IEEE 9 bus system illustrating four optimal SVC locations


Fig. 5. 30 bus test system illustrating five optimal SVC locations



Solutions were then found while varying the maximum
number of devices allowed from zero up to the optimal
number. Plots of versus maximum device number are
shown in Fig. 6 and Fig. 7 and key results are presented in
Table I and II for 9 bus and 30 bus test systems respectively. It
can be seen in the 9 bus case, for a device limit of three that
only two SVCs are required to provide maximum loadability.
It can also be observed that in the 30 bus case loadability
appears to be equal between three and five devices. However
compensation levels in Table II show a decrease in the total
reactive compensation from 59.9 MVAR to 44.87 MVAR as
the number of SVCs increase. This result is interesting as it
suggests that a spread of reactive compensation may be more
effective in improving loadability compared to higher levels
of compensation focused at a single location.


Fig. 6. versus maximum device number for 9 bus system


Fig. 7. versus maximum device number for 30 bus system

TABLE I
KEY OPTIMIZATION RESULTS FOR THE 9 BUS TEST SYSTEM
FOR DIFFERENT DEVICE NUMBER MAXIMUMS

Maximum
Device
Number
Loadability ()
SVC Locations
(Buses)
Total System
Compensation
(MVAR)
0 1.9378 - 0
1 2.2185 9 50
2 2.3231 9, 5 100
3 2.3231 9, 5 100
4 2.4108 9, 5, 7, 4 164.90


6
TABLE II
KEY OPTIMIZATION RESULTS FOR THE 30 BUS TEST SYSTEM
FOR DIFFERENT DEVICE NUMBER MAXIMUMS

Maximum
Device
Number
Loadability ()
SVC Locations
(Buses)
Total System
Compensation
(MVAR)
0 1.0727 - 0
1 1.5806 8 34
2 1.6172 8, 30 53.83
3 1.7648 8, 30, 10 59.91
4 1.7648 8, 30, 19, 21 46.32
5 1.7648 8, 30, 19, 21, 26 44.87

Practically speaking, the level of compensation required is
only one of the aspects to consider in the planning process
with other considerations such as device installation costs also
playing a role. In the 30 bus case a drop in required
compensation of 22.7% is gained by increasing the number of
devices to four. Installing five however only provides an
additional 3.1% reduction in compensation levels and may not
justify the additional cost of an SVC installation.

B. Comparison of Results via PV Analysis
Using the continuation power flow technique [20] PV
curves for the test systems are obtained to compare the system
voltage stability margins with and without SVC devices
installed. The plots presented below are prepared with PSAT,
a MATLAB toolbox for electric power system analysis and
control [18]. In Fig. 8 the PV curves for the 30 bus system
without SVCs (base case) are shown. Fig. 9 shows curves of
the same buses after the five optimal SVCs are installed. The
loading parameter represents the multiple of total system real
power consumption. The three buses with lowest voltages at
base case are chosen to illustrate the loadability improvement.
With the addition of these five SVCs an additional 35.5%
of total real load or 67.2 MW, can be tolerated before the
voltage at bus 8 the lowest in the system falls to 0.9 pu. These
curves show that the proposed MILP approach for SVC
placement is successful in providing locations and settings
which improve system loadability significantly.


Fig. 8. PV curve of 30 bus system without SVCs


Fig. 9. PV curve of 30 bus system with SVCs installed

VI. CONCLUSION
This paper presents the technique of MILP combined with
LFB equations to optimally allocate SVCs for improved
loadability. The additional term in the objective function
allows compensation levels to be minimized and removes
solutions with unnecessary compensation. The proposed
method is tested on the IEEE 9 bus system and a 30 bus test
system, simulation results show that device numbers can be
optimized successfully along with locations and initial settings
Research presented in this paper is ongoing and the authors
plan to verify the technique with the Queensland HV
transmission network as a case study. In the next stage effects
of a weighted loadability factor w will also be investigated
and investment costs will be taken into consideration.
It should be stressed that due to the approximations made in
the problem formulation, this method is only suited for initial
analysis and results require further verification by a full AC
model with security and stability considerations.
VII. REFERENCES

[1] K. R. Padiyar, Facts controllers in power
transmission and distribution. New Delhi :: New Age
International, 2007.
[2] N. G. Hingorani and L. Gyugyi, Understanding
FACTS : concepts and technology of flexible AC
transmission systems. Delhi :: IEEE Press, 2001.
[3] S. Gerbex, R. Cherkaoui, and A. J. Germond,
"Optimal location of multi-type FACTS devices in a
power system by means of genetic algorithms,"
Power Systems, IEEE Transactions on, vol. 16, pp.
537-544, 2001.
[4] G. I. Rashed, H. I. Shaheen, X. Z. Duan, and S. J.
Cheng, "Evolutionary optimization techniques for
optimal location and parameter setting of TCSC
under single line contingency," Applied Mathematics
and Computation, vol. 205, pp. 133-147, 2008.
[5] W. Ongsakul and P. Jirapong, "Optimal placement of
multi-type FACTS devices to enhance total transfer
capability using improved evolutionary
7
programming," International Journal of Energy
Technology and Policy, vol. 7, pp. 180-200, 2009.
[6] M. Eghbal, N. Yorino, Y. Zoka, and E. E. El-Araby,
"Application and comparison of metaheuristic
techniques to reactive power planning problem,"
IEEJ Transactions on Electrical and Electronic
Engineering, vol. 3, pp. 721-730, 2008.
[7] N. K. Sharma, A. Ghosh, and R. K. Varma, "A novel
placement strategy for FACTS controllers," Power
Delivery, IEEE Transactions on, vol. 18, pp. 982-
987, 2003.
[8] M. Kowsalya, K. K. Ray, U. Shipurkar, and
Saranathan, "Voltage stability enhancement by
optimal placement of UPFC," Journal of Electrical
Engineering and Technology, vol. 4, pp. 310-314,
2009.
[9] G. Yang, G. Hovland, R. Majumder, and Z. Dong,
"TCSC allocation based on line flow based equations
via mixed-integer programming," Pittsburgh, PA,
USA, 2008, p. 1 pp.
[10] R. E. Bixby, M. Fenelon, G. Zonghao, and E.
Rothberg, "MIP: theory and practice-closing the
gap," Cambridge, UK, 2000, pp. 19-49.
[11] F. G. M. Lima, F. D. Galiana, I. Kockar, and J.
Munoz, "Phase shifter placement in large-scale
systems via mixed integer linear programming,"
Power Systems, IEEE Transactions on, vol. 18, pp.
1029-1034, 2003.
[12] T. J. Overbye, X. Cheng, and Y. Sun, "A comparison
of the AC and DC power flow models for LMP
calculations," Big Island, HI., United states, 2004,
pp. 725-734.
[13] G. Y. Yang, G. Hovland, R. Majumder, and Z. Y.
Dong, "A novel TCSC planning model based on line
flow equations via MILP," Tampa, FL, United States,
2007, p. 4275507.
[14] Y. Ping and A. Sekar, "Analysis of radial distribution
systems with embedded series FACTS devices using
a fast line flow-based algorithm," IEEE Transactions
on Power Systems, vol. 20, pp. 1775-82, 2005.
[15] "TOMLAB / CPLEX v12.1 User's Guide," Tomblab
Optimization Inc., Vasteras2009.
[16] G. Sierksma, Linear and integer programming :
theory and practice, 2nd ed. ed. New York :: Marcel
Dekker, 2002.
[17] R. D. Zimmerman, C. E. Murillo-Snchez, and D.
Gan, "MATPOWER version 3.2," 2007.
[18] F. Milano, "PSAT version 2.1.5," 2009.
[19] "CPLEX version 12.1," Gentilly: ILOG Inc., 2009.
[20] V. Ajjarapu and C. Christy, "The continuation power
flow: a tool for steady state voltage stability
analysis," in Power Industry Computer Application
Conference, 1991. Conference Proceedings, 1991,
pp. 304-311.




VIII. BIOGRAPHIES

Robert Wan-Fu Chang was born in Taipei, Taiwan
in 1987. He immigrated to New Zealand in 1993, and
then to Australia in 1999. In 2008 he received his
B.E. (Hons) degree in electrical engineering from the
University of Queensland, Brisbane, Australia, where
he is currently pursuing a Ph.D. degree in the School
of Information Technology and Electrical
Engineering. His research interests include power
system optimisation, stability analysis and system
identification.


Tapan Kumar Saha (M93, SM97) was born in
Bangladesh in 1959 and immigrated to Australia in
1989. He received his B. Sc Engineering (electrical
and electronic) in 1982 from the Bangladesh
University of Engineering & Technology, Dhaka,
Bangladesh, M. Tech (electrical engineering) in1985
from the Indian Institute of Technology, New Delhi,
India and PhD in 1994 from the University of
Queensland, Brisbane, Australia. Tapan is currently
Professor of Electrical Engineering in the School of
Information Technology and Electrical Engineering,
University of Queensland, Australia. Previously he has had visiting appointments
for a semester at both the Royal Institute of Technology (KTH), Stockholm,
Sweden and at the University of Newcastle (Australia). He is a Fellow of the
Institution of Engineers, Australia. .His research interests include condition
monitoring of electrical plants, power systems and power quality.