Electrical Power and Energy Systems 73 (2015) 601607

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

Voltage stability constrained dynamic optimal reactive power ow

based on branch-bound and primaldual interior point method
Jinquan Zhao a,, Lijie Ju a, Zemei Dai b, Gang Chen a
a
b

College of Energy and Electrical Engineering, Hohai University, Nanjing 210098, China
State Grid Electric Power Research Institute, Nanjing 210003, China

a r t i c l e

i n f o

Article history:
Received 17 March 2015
Accepted 9 May 2015

Keywords:
Dynamic optimal reactive power ow
Voltage stability
Reactive voltage network partition
VQ curves
Dynamic reactive power reserves
Branch-bound and primaldual interior
point method

a b s t r a c t
A day-ahead voltage stability constrained dynamic optimal reactive power ow (VSC-DORPF) model is
proposed in this paper. The amount of dynamic reactive power reserves (DRPR) is used as a measure
of voltage stability of power system. The effective dynamic reactive power reserves (EDRPR) of reactive
power sources are calculated to obtain DRPR of each area and the maximum variations in reactive power
generation under contingency are taken as the required minimal DRPR for each area. Then the DRPR are
introduced into the VSC-DORPF model as one of multiple objective functions and constraints in order to
enhance the voltage stability of power system. A hybrid method, integrated by branch-bound method
and primaldual interior point (PDIP) method, is proposed to solve this VSC-DORPF problem. The discrete
control variables and the time coupled constraints are handled by the proposed branching and pruning
principles. As a result, the VSC-DORPF problem is decomposed into a series of optimal reactive power
ow (ORPF) problems with continuous control variables only. Numerical tests with IEEE 30-bus system
and IEEE 118-bus system show that the proposed model and method are effective.
2015 Elsevier Ltd. All rights reserved.

Introduction
Dynamic optimal reactive power ow (DORPF) determines the
proper settings of reactive power control devices in next day based
on the day-ahead load forecast and active power scheduling plan in
order to reduce the daily network losses, enhance voltage prole
and avoid excessive operation.
As voltage stability has not been taken into account in the general DORPF model, the scheduling results cannot respond to the
impact which acute load uctuation bring to power system. A
fuzzy membership function of bus voltage was taken as one of
optimization objectives to increase voltage quality in [1].
However, keeping bus voltages within qualied ranges simply
cannot maintain voltage stability. Thus it is necessary to carry
out further researches on DORPF considering voltage stability.
Dynamic reactive power reserves (DRPR) have always been linked
with voltage stability as they have a signicant effect on the reliable operation of power system [2]. In [3] an optimal reactive
power ow (ORPF) model with DRPR of power system being one
of objective functions was proposed. It is worth noting that since
each reactive power source gives a different impact on the entire

Corresponding author. Tel.: +86 13584073152.

E-mail address: jqzhao2@tom.com (J. Zhao).
http://dx.doi.org/10.1016/j.ijepes.2015.05.038
0142-0615/ 2015 Elsevier Ltd. All rights reserved.

system, DRPR of large system cannot be obtained by merely summing up individual reserves. Thus the network was partitioned
into several areas and the reactive power sources were assigned
weighting factors based on the reactive power load margin of each
area in [3]. But it is unreasonable to give the same factors to the
reactive power sources in an area. Moreover, it cannot be guaranteed that there are sufcient DRPR in each area to maintain voltage
stability merely by the weighted sum of individual reserves in
objective functions without any explicit constraints.
On the other hand, DORPF problem is essentially a large scale
mixed integer nonlinear programming problem. The presence of
a large number of discrete control variables and time coupled constraints makes it difcult to solve. Different methods have been
proposed and they can be classied basically into four categories.
(1) Simultaneous solution method [4]. The operation limits of control devices are described by the analytic mathematic expressions
of their control variables. The DORPF problem is solved as a whole
and the discrete control variables achieve their discrete values by
an embedded algorithm. Although this method usually shows good
performance on small test systems, its application on larger power
system will be hard. (2) Modern intelligent algorithm [1,5]. The
control variables in the whole day are encoded into an individual
and a modern intelligent algorithm is adopted to solve the problem. This kind of algorithm cannot be put into practical application
because of its stochastic nature. (3) Decomposition coordination

602

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

method [6,7]. The entire problem is decomposed into two

sub-problems with only continuous control variables or discrete
control variables respectively which interact through a coordination technology. But the independent solution of continuous and
discrete variables will lead to a deviation in the search path of
solution. (4) Heuristic algorithm [8,9]. The dynamic problem is
converted to a series of static ones by determining an operation
sequence of each control device by some heuristic rules.
However, it is difcult to achieve reasonable settings of control
devices by the assumed operation time.
In view of the above, a new DORPF model considering voltage
stability is presented in this paper. The DRPR are taken as an index
of voltage stability and they appeared in both optimization objective functions and constraints of the proposed model. It can reduce
daily network losses, improve voltage quality and enhance voltage
stability of power system. On the other hand, considering the two
difculties in solving DORPF problem are both related to discrete
control variables, a hybrid method combined by branch-bound
method and primaldual interior point (PDIP) method is adopted
to solve the problem. Branch-bound method [10] is used to solve
integer programming problem and PDIP method [11] is used to
solve nonlinear programming problem. The two methods are often
integrated together to solve mixed integer nonlinear programming
problem, such as ORPF [12,13] and unit commitment [14]. During
the solution process of DORPF problem, the discrete control
variables achieve discrete values via a branch-bound tree and the
time coupled constraints are met by reasonable branching and
pruning principles. Numerical tests with IEEE 30 and 118-bus
system show that the proposed model and method are effective.

2. Operation constraints

V i;min 6 V ti 6 V i;max

i 1; . . . ; NB ; t 1; . . . ; N T

3. The constraints of DRPR for each area

NG;k 
X


Q tg;j;eff  Q tg;j P Q trs;k;min

k 1; . . . ; Narea t 1; . . . ; NT

j1

4
4. The constraints of control variables

Q g;i;min 6 Q tg;i 6 Q g;i;max

K i;min 6 K ti 6 K i;max

i 1; . . . ; NG ; t 1; . . . ; NT

i 1; . . . ; NK ; t 1; . . . ; NT

Q c;i;min 6 Q tc;i 6 Q c;i;max

i 1; . . . ; Nc ; t 1; . . . ; NT

In this paper, daily active power losses, voltage deviation and

DRPR of power system are taken as objective functions to reduce
network losses, improve voltage quality and voltage stability. The
DRPR of each area are kept larger than their required values in
constraints for the purpose of mitigating voltage collapse. The
proposed VSC-DORPF model can be presented as follows.
Objective functions

3
2
PN G  t
2
t
PNB  t
NT
t
X
V

V
j1 Q g;j;eff  Q g;j
P
i;set
i
i1
loss
4x1
5
min
 x3
t x2
t
t
f1
f2
f3
t1
1
where NT is the number of intervals, N B and NG are the number of
buses and generators, x1 , x2 and x3 are the weighting factors of
optimization objectives, the rst objective component is daily network losses, P tloss is the active power losses at interval t, the second
component is voltage deviation, V ti is the voltage magnitude of bus i
at interval t, V i;set is the expected voltage magnitude of bus i, the

Constrains of maximum allowable action range between

successive intervals:

jK ti  K t1
j 6 Sk;i;D K i;step
i

i 1; . . . ; NK ; t 1; . . . ; NT

jQ tc;i  Q t1
c;i j 6 SQc;i;D Q c;i;step

i 1; . . . ; Nc ; t 1; . . . ; NT

8
9

Constrains of maximum allowable action number in a day:

NT 
X


K ti  K it1 6 Sk;i;max

i 1; . . . ; NK

NT 
X


Q tc;i  Q t1
6 SQc;i;max
c;i

10

i 1; . . . ; Nc

where V i;max and V i;min are voltage limits of bus i, Q g;i;max and
Q g;i;min are reactive power limits of generator i, N K and N C are
the number of transformers and compensators, Q tc;i , Q c;i;max ,
Q c;i;min and Q c;i;step are the reactive power compensation, its
limits and step size of compensator i at interval t, K ti , K i;max ,
K i;min and K i;step are the ratio, its limits and step size of transformer i at interval t, xt is the vector of control variables and
state variables at interval t, Narea is the number of areas, NG;k
is the number of generators in area k, Q trs;k;min is the required
DRPR for area k at interval t, Sk;i;D , SQc;i;D , Sk;i;max and SQc;i;max are
maximum allowable action range between successive intervals and maximum allowable action number in a day of
transformer i and compensator i respectively.

Constraints
1. Power ow equations

11

t1

interval t, f 1 , f 2 and f 3 are the optimal value of each optimization

objective when optimized only at interval t respectively.

g t xt 0 t 1; . . . ; NT

5. Time coupled constraints

third component is DRPR of power system, Q tg;j and Q tg;j;eff are the
reactive power output and its effective limit of generator j at
t

t1

The problem formulation of VSC-DORPF

Fig. 1. VQ curve.

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

603

where m is the number of controlled buses, n is the number of reactive power source buses.
Thus the spatial electrical distance between bus i and bus j can
be calculated using (15).

Mij

Fig. 2. The VQ curve in contingency.

The computation of DRPR

The amount of DRPR is taken as a measure of voltage stability in
the proposed VSC-DORPF model. The DRPR and its required
minimal value for each area of each interval should be computed
in advance to establish the mathematic model.
Network partition based on a mapping partitioning algorithm
Because of the local equilibrium of reactive power, it is effective
to partition network into several areas in reactive power-voltage
control. When traditional clustering algorithms are used in
network partition, there are shortcomings in scalability and it is
difcult to guarantee that the controlled buses and the corresponding reactive power source buses are partitioned into the same area.
Thus a mapping-based partitioning algorithm for reactive
power-voltage control is proposed in this paper.
First of all, the sensitivity of reactive power source buses to
controlled buses is calculated.

aij



DV j DV j DV i 1

DV i DQ i DQ i

12

where aij is the sensitivity of reactive power source bus i to

controlled bus j, DV i is the variation in voltage magnitude of bus i,
DQ i is the variation in reactive power injection of bus i.
To amplify the differences between sensitivities, the logarithmic form is taken as follows.

dij  lg jaij j

13

Then each controlled bus is mapped into the multi-dimension

space as a spatial point, the dimension of which is equal to the
number of reactive power source buses. In this control space, the
sensitivity dij is dened as the component of the coordinate for
controlled bus. The electrical distance matrix D is thereby
obtained.

d11

d12



d1n

6d
6 21
D6
6 ..
4 .

d22
..
.


..
.

d2n
..
.

7
7
7
7
5

dm1

dm2

   dmn

Fig. 3. Sketch diagram of branching.

q
2
2
di1  dj1    din  djn

15

Based on the electrical distance, a hierarchical clustering algorithm [15] is adopted to obtain the optimal number of areas and
the areas of reactive power source buses. In order to make sure that
controlled buses are partitioned into the same area with their
corresponding reactive power source buses, a mapping partition
algorithm is used to determine the areas of controlled buses. The
electrical distance between the controlled bus and each reactive
power source bus is calculated and the controlled bus is classied
into the same area with its electrically closest reactive power
source.
The computation of DRPR for each area based on VQ-curve method
Reactive power reserves can be classied into two types: static
reactive power reserves and DRPR. Static reactive power reserves
cannot respond to unforeseen events effectively as they do not
have the ability to assist voltage control continuously while
dynamic reactive power sources can increase their production
quickly to maintain the controlled voltages at constant values in
a contingency. Thus DRPR are the major study objects in this paper.
The most straightforward denition of DRPR is the difference
between the maximum reactive power generation that determined
by the physical factors and its current value. However, the above
denition may not represent the useful quantity of DRPR since at
the voltage collapse point not all the amount of DRPR can be
utilized. Thus the concept of effective dynamic reactive power
reserves (EDRPR) of reactive power sources is introduced in this
paper which is dened as the difference between the reactive
power output at the voltage collapse point and current operation
point [16,17].
In order to make the process of voltage collapse insensitive to
load increasing modes, VQ-curve method is used to compute the
voltage collapse point [18,19]. First, the pilot bus of each area is
determined based on the electrical distance obtained in partition
process. The average electrical distance from controlled bus to all
the reactive power source buses of the area is given by formula
(16). Choose the bus with closest average electrical distance to
be the pilot bus. The obtained pilot bus is the electrical central
bus of the area with moderate electrical distance to each reactive
power source and it can give a comprehensive consideration to
all the reactive power sources in the area.

j
D

X
2
d
i2G ij

16

14

Fig. 4. Sketch diagram of principle (4).

604

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

 j is the average electrical distance from controlled bus j to

where D
reactive power source buses of area k, Gk is the set of reactive power
source buses of area k.
Then a ctitious reactive power support Q f is connected to the
pilot bus and a series of power ows are simulated with the change
of the voltage magnitudes V f at the bus. A curve of bus voltage V f
versus ctitious reactive power output Q f is thereby generated as
shown in Fig. 1. The minima of the VQ curve A is the voltage
collapse point and the reactive power outputs of reactive power
sources at this point are called effective reactive power limits.
The DRPR of each area can be calculated by summing up the
EDRPR of reactive power sources as follows.

Q rs;k

N G;k
X
Q g;j;eff  Q g;j k 1; . . . ; Narea

17

j1

where Q rs;k is the DRPR of area k.

The above DRPR can take into account the network topology,
load characteristics and the generation capability of all units
effectively.

The computation of the required DRPR for each area

In order to mitigate voltage collapse, each area should keep a
certain amount of DRPR. Since the network topology and load
conditions of each area are different, the minimal reactive power
reserves one area should maintain vary from each other.
Considering that the DRPR of each area at base case should be
larger than the maximum variations in reactive power generation
of the area under contingency. The most stressed line outage
contingency is taken as an example in this paper to calculate the
required minimal DRPR. The VQ curve at pilot bus under contingency is computed again and it is shown in the dotted line in Fig. 2.
The required DRPR for each area can be calculated as follows.

Q rs;k;min

N G;k 

X
Q g;j;eff  Q g;j
k 1; . . . ; Narea

18

Branching process
Considering the inuences of different control variables on
objective function are different, a proper ranking of discrete control
variables before branching can improve the calculation efciency.
As compensators change the quantity of reactive power directly
while transformers only effect its distribution, the compensators
are given a high priority. Besides, another priority list according
to the difference between the non-discrete value of discrete control
variable and its nearest discrete value is also used here.
The branching method used in this paper is called dichotomy.
If the discrete control variable does not achieve its discrete
value, bound constraints are introduced to construct two new
sub-problems. As constraints (8) and (9) should be taken into
account during the solution process, it is need to judge whether
the sub-problem is at the same interval with its parent problem.
If so, the bound constraints are introduced by the method shown
in Fig. 3. In the gure, ~
xi is the continuous solution of the discrete
control variable xi between [xi;min ; xi;max ], xi;step is its step size, Ii is a
xi 6 Ii xi;step .
discrete value of discrete control variable xi and Ii 6 ~
If not, the bound constraints should be calculated using formula
(19) and (20).

T1

~xi  Si;D xi;step ; Ii \ xi;min ; Ii 

19

Ii xi;step ; ~xT1
Si;D xi;step \ Ii xi;step ; xi;max 
i

20

where ~xT1
is the value of control device xi at the previous interval,
i
Si;D is the maximum allowable action range between successive
intervals.
Solution of sub-problems by PDIP method
As the discrete control variables are considered as continuous
variables within their bounds in sub-problems, each sub-problem
becomes essentially an ORPF problem with only continuous
control variables as follows. It can be solved by PDIP method [11]
directly.

j1

where Q g;j;eff represents the reactive power outputs at the voltage

collapse point A , Q g;j represents the reactive power outputs at the
operation point B .

The solution algorithm based on branch-bound and PDIP

method
The fundamental of the solution algorithm of VSC-DORPF problem based on branch-bound and PDIP method is that the discrete
control variables approach their discrete values progressively
through a branch-bound tree and constraints (8) and (9) are taken
into account during the branching process. Then the sub-problem
which is an ORPF problem with only continuous control variables
is obtained. It can be solved by PDIP method directly. And the
sub-problem which does not meet constraints (10) and (11) is
removed during the pruning process.

Table 1
The data of compensators in IEEE 30-bus system (MVar).
Bus

Lower limits

Upper limits

Hierarchy capacity

10
24

0
0

20.0
10.0

4.0
2.0

Table 2
The network losses of IEEE 30-bus system (MW).
Interval Load rate
(p.u.)

Initial
values

Continuous
method

Upper
bound

This
Paper

Ref.
[1]

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

3.1731
3.1542
3.0841
3.0249
3.0394
3.2190
3.6636
3.8800
4.2756
4.8910
5.0724
5.0930
4.2588
4.2264
4.7725
4.8531
4.8901
4.9512
4.8099
5.3058
5.4403
4.9945
4.0193
3.3708

2.6732
2.6561
2.5922
2.5382
2.5515
2.7149
3.1167
3.3109
3.6968
4.1018
4.2727
4.2920
3.6815
3.6519
3.9894
4.0659
4.1009
4.1587
4.0249
4.4906
4.6151
4.1995
3.4619
2.8525

2.6735
2.6563
2.5926
2.5386
2.5519
2.7152
3.1175
3.3124
3.7002
4.1081
4.2804
4.2999
3.6848
3.6550
3.9947
4.0719
4.1072
4.1654
4.0305
4.5003
4.6262
4.2066
3.4640
2.8529

2.6735
2.6563
2.5925
2.5386
2.5518
2.7153
3.1173
3.3111
3.6973
4.1042
4.2764
4.2959
3.6820
3.6523
3.9911
4.0681
4.1033
4.1615
4.0268
4.4949
4.6192
4.2026
3.4621
2.8556

2.6736
2.6563
2.5926
2.5386
2.5518
2.7157
3.1175
3.3120
3.6975
4.1073
4.2766
4.2958
3.6824
3.6529
3.9919
4.0680
4.1063
4.1647
4.0298
4.4977
4.6251
4.2066
3.4624
2.8534

0.7677
0.7653
0.7563
0.7486
0.7505
0.7735
0.8271
0.8517
0.8984
0.9447
0.9635
0.9656
0.8966
0.8931
0.9321
0.9407
0.9446
0.9510
0.9361
0.9869
1.0000
0.9555
0.8703
0.7923

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

Table 3
The action number of discrete control devices of IEEE 30-bus system.
The discrete variables

This paper

Ref. [1]

Bc10
Bc24
T11
T12
T15
T36

2
2
1
1
1
1

4
2
1
1
1
1

Ploss
min x1
x2
f1
s:t: gx 0

PN B

i1 V i

 V i;set 2

f2

V i;min 6 V i 6 V i;max

PN G
 x3

j1 Q g;j;eff

 Q g;j

f3

i 1; . . . ; NB

Q g;i;min 6 Q g;i 6 Q g;i;max

i 1; . . . ; N G

K 0i;min 6 K i 6
Q 0c;i;min 6 Q c;i

1; . . . ; N K

K 0i;max i
6 Q 0c;i;max

21

i 1; . . . ; Nc

NG;k

Q g;j;eff  Q g;j P Q rs;k;min

k 1; . . . ; N area

j1

where K 0i;min , K 0i;max , Q 0c;i;min and Q 0c;i;max are the ratio limits of transformer i and compensation limits of bus i after branching
respectively.

605

method to get an integer feasible solution of VSC-DORPF problem.

As discrete control devices cannot be adjusted during the day, the
obtained solution must be inferior to the optimal one and it can be
taken as the upper bound of original problem. Then the pruning
principle (4) is judged by the method shown in Fig. 4. The white
dot in the gure represents the sub-problem that has not obtained
integer solution, the black dot represents the sub-problem that has
obtained integer solution of its interval already, ZBt is its
objective function value, UBt is the objective function value at
the corresponding interval in upper bound, the grid shaped dot
represents the sub-problem to be handled, T is its interval, objt
is its objective function value.
The sub-problem with greater value in formula (22) is removed.
In this way, the sub-problems with bad optimization results can be
eliminated as early as possible. Moreover, as all the control
devices can be adjusted in early intervals, if a good optimization
result cannot be obtained at that time, the possibility of the
sub-problem to get the optimal solution in the later intervals is
small. Consequently, the above strategy can help ensure the overall
optimality of algorithm.
T1
T
X
X
ZBt objT v s
UBt
t1

22

t1

Solution procedure
Pruning process
Once the sub-problem meets pruning principles in the solution
process, it should be removed. Four pruning principles of
VSC-DORPF problem are listed. (1). The sub-problem is infeasible;
(2). All the discrete control variables have achieved discrete values
in the sub-problem; (3). The discrete control variables in the
sub-problem do not meet constraints (10) and (11); (4) The objective function value of sub-problem is greater than or equal to the
upper bound.
Brief instructions of pruning principles (4) are given as follows.
As the VSC-DORPF problem is decomposed into a series of static
ORPF problems, the objective function value of the entire problem
cannot be obtained until all the intervals in a day have been solved.
Thus, a heuristic strategy is adopted in pruning principle (4). First a
xed upper bound is determined in advance. Based on the periodicity between interval 0 and interval 24, the values of discrete control variables at interval 24 can be taken as the initial values at
interval 0 in the next day. Give discrete control variables at each
interval the initial values and solve the ORPF problems by PDIP

Fig. 5. Reactive power compensation of capacitor 10.

The steps in VSC-DORPF problem based on the proposed

method are described as follows.
1. Partition the network into several areas.
2. Compute the DRPR and its required minimal value for each
area of each interval according to formula (17) and (18)
respectively.
3. Establish the VSC-DORPF model using (1)(11).
4. Determine the upper bound of original problem.
5. Set T 1. Relax the ORPF problem of interval 1 and solve it by
PDIP method. Then store it in the list BP for further analysis.
6. Branch the sub-problems in list BP in turn and solve them by
PDIP method. Then add them into the list XP.
7. Handle sub-problems with different solutions as follows. (a)
If only the branching variable achieve discrete value, record
its action number. (b) If all the variables at current interval
achieve discrete value, record the action number and set
T T 1.
8. Prune the sub-problems in list XP according to pruning principles (1)(4).

Fig. 6. Ratio of transformer 36.

606

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

9. Update the list BP with the pruned list XP.

10. Identify whether the list BP is empty. If so, choose the smallest one among the obtained integer feasible solutions to be
the optimal solution. If not, go to step 6.

Table 6
Dynamic reactive power reserves of IEEE 118-bus system (MVar).
Interval

Area

Before
optimization

Network losses
minimized only

The proposed
model

Total
2
3

4300.26
749.84
301.74

4684.22
767.13
192.86

4894.38
872.56
309.78

10

Total
2
3

4260.79
790.26
303.27

4688.82
824.96
174.01

4989.62
919.76
306.09

20

Total
2
3

4214.23
783.45
303.05

4652.01
812.08
170.21

4979.11
921.61
304.73

Simulations
The proposed VSC-DORPF is applied to IEEE 30 and 118-bus
system to test its effectiveness. The algorithm has been coded in
Matlab R2010b on a computer with Pentium R 3.0 GHz CPU and
3 GB memory.
IEEE 30-bus system

The mathematic model using (1)(11) is established in which

The adjustable range of compensator is shown in Table 1. The
adjustable range of transformer is 1  4  0:0125. The maximum
allowable action range between successive intervals is 1. The maximum allowable action number of compensator and transformer in
a day are 4 and 1 respectively.
In the model of IEEE 30-bus system DRPR are not included.
Branch-bound and PDIP method is used to solve the problem.
The results are compared to [1], shown in Tables 2 and 3 and
Figs. 5 and 6. It can be seen that the result of continuous optimization is best. But the discrete control devices are adjusted continuously which is inconsistent with the actual situation of power
system. In upper bound, the discrete control variables are xed
on initial values and only continuous control variables can be
adjusted, thus the obtained solution is worst. In this paper and
[1], both the discreteness and operation limits of discrete control
variables are taken into account, therefore the results meet the
actual demand of power system. But the results in this paper are
better and more stable. Moreover, the proposed pruning strategy
can avoid a lot of unnecessary calculation and the algorithm in this
paper takes only 171.94s. The method in [1] takes 16150.99s which
is inefcient. In conclusion, the proposed method in this paper is
reasonable and effective.
IEEE 118-bus system
The adjustable range of compensator is shown in Table 4. The
other data are the same with those of IEEE 30-bus system.

x1 0:5503; x2 0:0002; x3 0:4495. Branch-bound and PDIP

method is used to solve the problem. The results are compared
to the general DORPF which optimizes daily network losses only,
shown in Tables 5 and 6. It can be concluded that although the general DORPF can cut down the daily network losses, it will lead to an
increase in voltage deviation and a reduction of DRPR in some
areas. However, the VSC-DORPF in this paper can avoid an uneven
distribution of DRPR by making the DRPR of each area larger than
the required values. It can reduce daily network losses, improve
voltage quality and enhance the voltage stability of power system
at the same time.

Conclusion
As voltage stability is not taken into account in general DORPF
model, the amount of DRPR is taken as a measure of voltage stability and a VSC-DORPF model is presented in this paper. It can reduce
daily network losses, improve voltage quality and enhance voltage
stability of power system. A hybrid method combined by
branch-bound method and PDIP method is adopted to solve the
VSC-DORPF problem. The discrete control variables and time coupled constraints can be handled accurately by the proposed
branching and pruning principles. Numerical tests with IEEE 30
and 118-bus system show that the proposed model and method
are effective.

Acknowledgements
Table 4
The data of compensators in IEEE 118-bus system (MVar).
Bus

Lower limits

Upper limits

Hierarchy capacity

5
17, 107, 110
34, 4446, 48, 74, 83
37
79, 82, 105

50.0
0
0
30.0
0

0
10.0
20.0
5.0
30.0

10.0
2.0
5.0
5.0
5.0

This work is a part of the projects National Natural Science

Foundation of China (51077042), the National High Technology
Research and Development Program of China (863 Program)
(2012AA050207), Research Fund for the Doctoral Program of
Higher Education of China (20120094110008). The authors would
like to thank for the supports from the National Natural Science
Foundation of China, National High Technology Research and
Development Program of China and Research Fund for the
Doctoral Program of Higher Education of China.

Table 5
The optimization results of IEEE 118-bus system.
Interval

Before
optimization

Network losses
minimized
only

Time (s)

The
proposed
model

10894.39

12603.44

Network
losses
(MW)

5
10
20

76.01
118.64
129.23

58.96
94.22
103.10

64.55
99.88
109.10

Voltage
deviation
(p.u.)

5
10
20

0.03
0.04
0.04

0.38
0.36
0.35

0.02
0.03
0.03

References
[1] Zhang A, Yang H. Pareto-set based multi-objective dynamic reactive power and
voltage control. In: International conference on sustainable power generation
and supply; 2009. p. 16.
[2] Leonardi B, Ajjarapu V. Investigation of various generator reactive power
reserve (GRPR) denitions for online voltage stability/security assessment. In:
IEEE power and energy society general meeting, 2008. p. 17.
[3] Dong F, Chowdhury BH, Crow ML, et al. Improving voltage stability by reactive
power reserve management. IEEE Trans Power Syst 2005;20:33845.
[4] Liu M, Canizares C, Huang W. Reactive power and voltage control in
distribution systems with limited switching operations. IEEE Trans Power
Syst 2009;24:88999.

J. Zhao et al. / Electrical Power and Energy Systems 73 (2015) 601607

[5] Shu J, Zhang L, Liu Y, et al. An improved genetic algorithm for dynamic reactive
power optimization in electricity market. In: International conference on
power system technology; 2004. p. 150812.
[6] Hong YY, Liao CM. Short-term scheduling of reactive power controllers. IEEE
Trans Power Syst 1995;10:8608.
[7] Liu F, Chung CY, Wong KP, et al. Hybrid immune genetic method for dynamic
reactive power optimization. In: International conference on power system
technology; 2006. p. 16.
[8] Sharif SS, Taylor JH. Dynamic optimal reactive power ow. In: American
control conference; 1998. p. 34104.
[9] Deng Y, Ren X, Zhao C, et al. A heuristic and algorithmic combined approach for
reactive power optimization with time-varying load demand in distribution
system. IEEE Trans Power Syst 2002;17:106872.
[10] Leblanc LJ. Algorithm for the discrete network design problem. Trans Sci
1975;9:18399.
[11] Granville S. Optimal reactive dispatch through interior point methods. IEEE
Trans Power Syst 1994;9:13646.
[12] Yang H, Gu Y, Zhang Y, et al. Reactive power optimization of power system
based on interior point method and branch-bound method. In: International
conference on power electronics and intelligent transportation system; 2009.
p. 58.

607

[13] Estevam CRN, Rider MJ, Amorim E, et al. Reactive power dispatch and planning
using a non-linear branch-and-bound algorithm. IET Gener Transm Distrib
2010;4:96373.
[14] Xie Y, Chiang H. A novel solution methodology for solving large-scale
thermal unit commitment problems. Electr Power Compon Syst 2010;38:
161534.
[15] Lagonotte P, Sabonnadiere JC, Leost JY, et al. Structural analysis of the electrical
system: application to secondary voltage control in France. IEEE Trans Power
Syst 1989;4:47986.
[16] Capitanescu F, Van Cutsem T. Evaluation of reactive power reserves with
respect to contingencies. In: Bulk power system dynamics and control V; 2001.
p. 19.
[17] Song H, Lee B, Kwon SH, et al. Reactive reserved-based contingency
constrained optimal power ow (RCCOPF) for enhancement of voltage
stability margins. IEEE Trans Power Syst 2003;18:153846.
[18] Chowdhury BH, Taylor CW. Voltage stability analysis: VQ power ow
simulation versus dynamic simulation. IEEE Trans Power Syst 2000;15:
13549.
[19] Mousavi OA, Bozorg M, Ahmadi-Khatir A, et al. Reactive power reserve
management: preventive countermeasure for improving voltage stability
margin. In: IEEE power and energy society general meeting; 2012. p. 17.