742 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO.

2, MARCH 2014
Optimal PMU Placement Considering
Controlled Islanding of Power System
Lei Huang, Student Member, IEEE, Yuanzhang Sun, Senior Member, IEEE, Jian Xu, Member, IEEE,
Wenzhong Gao, Senior Member, IEEE, Jun Zhang, Member, IEEE, and Ziping Wu, Student Member, IEEE
AbstractThis paper proposes an optimal phasor measurement
unit (PMU) placement model considering power system con-
trolled islanding so that the power network remains observable
under controlled islanding condition as well as normal operation
condition. The optimization objectives of proposed model are
to minimize the number of installed PMUs and to maximize the
measurement redundancy. These two objectives are combined
together with a weighting variable so that the optimal solution
with minimum PMU number and maximum measurement redun-
dancy would be obtained fromthe model. To reduce the number of
required PMUs, the effect of zero-injection bus is considered and
incorporated into the model. Furthermore, additional constraints
for maintaining observability following single PMU failure or line
loss are also derived. At last, several IEEE standard systems and
the Polish 2383-bus system are employed to test the presented
model. Results are presented to demonstrate the effectiveness of
the proposed method.
Index TermsControlled islanding, integer linear program-
ming, measurement redundancy, optimal phasor measurement
unit (PMU) placement, state estimation.
I. INTRODUCTION
S
TATE estimator plays an important role in the security of
power system operation. Its main purpose is to precisely
estimate the voltage phasors of all system buses based on a
set of acquired measurements [1]. When a measurement set al-
lows a unique solution of the state estimation (SE) problem, the
power system is said to be observable [2]. Recently phasor mea-
surement units (PMUs) have been used as measurement device
for SE, which will advance the traditional supervisory control
and data acquisition (SCADA) system. PMU provides voltage
phasor of the bus where it is installed and current phasors of all
branches incident to that bus [3]. The PMU measurements from
different buses, which are synchronized by the common clock
signal from global positioning system (GPS), can help simplify
Manuscript received February 16, 2013; revised June 04, 2013 and September
07, 2013; accepted October 06, 2013. Date of publication November 04, 2013;
date of current version February 14, 2014. This work was supported in part by
the National Natural Science Foundation of China under Grant 51007067 and
the Ministry of Science and Technology of China under Grant 2012AA050218.
Paper no. TPWRS-00198-2013.
L. Huang, Y. Sun, and J. Xu (corresponding author) are with the
School of Electrical Engineering, Wuhan University, Wuhan, Hubei
430072, China (e-mail: huanglei1987@whu.edu.cn; yzsun@whu.edu.cn;
xujian@whu.edu.cn).
W. Gao, J. Zhang, and Z. Wu are with the Department of Electrical and
Computer Engineering, University of Denver, Denver, CO 80210 USA (e-mail:
Wezhong.Gao@du.edu; jun.zhang@du.edu; wzpbsb@gmail.com).
Color versions of one or more of the gures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identier 10.1109/TPWRS.2013.2285578
the process of state estimation and improve the accuracy of es-
timation results. If we place PMUs in all busses of a network,
all the voltage phasors can be directly measured without run-
ning any state estimator [4]. However, PMU and its associated
communication facilities are costly. Furthermore, the voltage
phasor of the bus incident to the bus with PMU installed can
be computed with branch parameter and branch current phasor
measurement [5]. So it is neither economical nor necessary to
install PMUs at all system buses. Thus, one of the important is-
sues is to nd the optimal number and placement of PMUs for
power system state estimation.
Optimal PMU placement (OPP) is rstly attempted in [6],
formulating as a combinatorial optimization problem of min-
imizing the PMU number for system observability. In recent
literatures, the OPP model has been generalized to include
additional constraints or contingencies. In [7], an integer pro-
gramming formulation of OPP problem is proposed with the
presence of conventional measurements. A generalized integer
linear programming (ILP) formulation for OPP is presented
in [8], considering effects of zero-injection buses (ZIBs) and
conventional measurements. The model proposed in [9] takes
into account PMU channel limitations. Contingency conditions
of line outage or PMU loss are considered separately or si-
multaneously in the OPP model proposed in [10]. Generally,
the existing OPP models concerns about the determination of
minimum number and optimal location set of PMUs, ensuring
that the entire power system remains a single observable island
[1]. In another word, these models can only handle the cases
in which the power system is operated as a single and inte-
grated network. However, some severe faults may lead parts
of the network to angle, frequency or voltage instability. In
that case, trying to maintain system integrity and operate the
system entirely interconnected is very difcult and may cause
propagation of local weaknesses to other parts of the system
[11]. As a solution, controlled islanding (CI) is employed by
system operators, in which the interconnected power system
is separated into several planned islands prior to catastrophic
events [12], [13]. After system splitting, wide area blackout
can be avoided because the local instability is isolated and
prevented from further spreading [14]. In order to operate
each island with power balancing and stability after controlled
islanding, it is essential to provide an OPP scheme which can
keep the network observable for the post-islanding condition as
well as normal condition.
After having determined the model of OPP, many investiga-
tors have presented different methods to solve this optimization
problem. These methods can be generally divided into math-
ematical and heuristic algorithms [1]. The representative
mathematical algorithm is ILP [7][10], which is capable of
solving large-scale problems in a short time and achieving the
globally optimal solution. Exhaustive search [15] is another
0885-8950 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 743
mathematical method used for OPP. However, it is not suitable
for large-scale systems with huge search space. Among the
heuristic optimization algorithms, genetic algorithm [16], Tabu
search [17], simulated annealing [6], differential evolution
[18], particle swarm optimization [19], immunity algorithm
[20], iterated local search [21], spanning tree search [22], and
greedy algorithm [23] have been developed. In spite of some
advantages, the major disadvantage associated with intelligent
search-based methods is that they do not guarantee that a
globally optimal solution is found [15].
In this paper, an ILP model of OPP considering controlled is-
landing (OPP-CI) is proposed. This model is able to determine
the minimal number and optimal location set of PMUs in order
to provide the full network observability in normal operation as
well as in controlled islanding scenario. To distinguish multiple
optimal solutions, measurement redundancy is incorporated into
the optimization objective. Meanwhile the effect of ZIBs is con-
sidered for further reducing the number of required PMUs. In
addition, to enhance the robustness of OPP scheme, the contin-
gencies of single PMU or line outage are also incorporated into
the proposed OPP model.
The rest of this paper is structured as follows. In Section II,
the concepts and rules of network observability are introduced
and the basic OPP formulation is described. The derivation of
OPP model considering controlled islanding is introduced with
details in Section III. In Section IV, contingencies of single
PMU or line outage are incorporated into OPP-CI model indi-
vidually or simultaneously. In Section V the performance of the
proposed new model is assessed using several IEEE standard
systems and a practical 2383-bus system. Finally, conclusions
are given in Section VI.
II. BASIC FORMULATION OF OPTIMAL PMU PLACEMENT
A. Observability Analysis Based on PMUs
The observability of a system is the precondition for its state
estimation. Power system observability analysis is usually car-
ried out in two different ways, namely, numerical observability
analysis and topological observability analysis [24]. Numerical
observability algorithm makes use of the gain matrix of state
estimation in a system. When the gain matrix is of full rank,
the system is said to be numerically observable. However, due
to high computation burden of verifying the rank of the gain
matrix, this approach would not be preferred for practical ap-
plications [19]. On the other hand, graph concept is utilized in
topological observability methods. The network is considered
to be topologically observable if a spanning tree of full rank
can be found in the graph [25]. This tree connects all nodes and
branches observed by direct measurements or calculations.
Observability analysis with phasor measurements has been
studied in a number of literatures. A method to combine phasor
measurements and conventional measurements in observability
analysis is rstly proposed in [26] and then improved in [27]. It
is able to handle not only systems including only phasor mea-
surements, but also the ones with both phasor measurements
and conventional measurements. Additionally, current phasor
measurements which lead to multiple-solutions can be detected
using the proposed method. In [28], a direct numerical algorithm
is presented to determine observable islands and restore observ-
ability for power systems. The adoption of reduced network
model makes the proposed method computationally attractive.
Based on PMU measurements as well as conventional measure-
ments, a hybrid topological/numerical method for power system
observability analysis is provided in [29], which shows good
performance in simulations.
In this paper the concept of topological observability is
adopted and the following simple rules have been applied [19].
If voltage phasor and current phasor at one end of a branch
are known, voltage phasor at the other end of that branch
can be obtained using Ohms law.
If voltage phasors at both ends of a branch are known, the
current phasor through this branch can be calculated.
The measurements such as bus voltage phasors and branch
current phasors, directly obtained from PMUs, are referred to
as direct measurements; measurements derived by employing
the above two rules are referred to as indirect measurements,
or pseudo measurements as in [30]. When the voltage phasor
at a bus can be obtained either from direct measurements or
indirect measurements, this bus is identied as observable. In
an observable network, each and every bus must be observed at
least once by using direct or indirect measurement.
B. Basic Optimal PMU Placement
In a power system network, the PMU placement at a bus can
be seen as a binary decision variable dened as
if
otherwise.
(1)
For a system with buses, therefore, the optimal PMU place-
ment problem can be formulated as an integer linear program-
ming problem as follows:
(2)
subject to constraints
(3)
where
is the cost of installing a PMU at bus . Without loss of
generality, cost of PMU installation at each bus is assumed
to be equal to 1 per unit in the present study.
refers to the number of times that the th bus is observed
through PMU measurements.
is the th entry of network connectivity matrix
dened as
if
otherwise.
(4)
For example, with (2), minimizing the number of PMUs
for the IEEE 14-bus system (Fig. 1) can be formulated as
follows:
(5)
where are PMU placement variables that are generated
by binary integer programming. Therefore, (5) represents the
minimum number of PMUs.
744 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
Fig. 1. IEEE 14-bus system.
The observability constraints (3) are listed explicitly as
follows:
(6)
The inequality constraints in (6) represent the requirements
of observed times for each of the 14 buses. For example, if at
least one PMU is installed on bus 1, 2 or 5, constraint is
satised, and bus 1 will be observable. In Fig. 1, the 14-bus
system is made completely observable by placing 4 PMUs on
buses 2, 6, 7, and 9 [31], although this is not the only optimal
solution.
III. OPTIMAL PMU PLACEMENT CONSIDERING
CONTROLLED ISLANDING
A. Controlled Islanding
Cascading failures are the most signicant threats for power
systemsecurity. Cascading failures together with additional line
tripping can lead the system to uncontrolled splitting [11]. For-
mation of uncontrolled islands with signicant power imbalance
is the main reason for system blackouts. In order to avoid cat-
astrophic wide area blackouts due to cascading failures, con-
trolled islanding has been considered as an effective defense
strategy. The main advantages of controlled islanding of power
systems can be listed as follows [11]:
It can separate weak and vulnerable areas from other stable
parts of the system.
Compared to the whole system, small subsystems are
easier to be handled and controlled under dynamic and
emergency conditions.
To date, a lot of investigations have been conducted on this
topic and various methods for controlled islanding [11][14],
[32][35] have been proposed; for example, a method of con-
trolled islanding with constraint of observability is presented in
[36]. Since our research is focused on the OPP problem, this
paper does not study methods of controlled islanding in detail
but only uses the controlled islanding results of several IEEE
standard systems presented in [13], [34], and [35], and assumes
a suitable controlled islanding scheme for the Polish 2383-bus
system. However, the proposed OPP method can be applied to
any other controlled islanding schemes.
After establishment of planned islands, there exist some
factors which may threat the stability and integrity of each
island, such as power imbalance, line overloading, voltage,
angle and frequency instabilities, etc. [11]. Therefore, to main-
tain static and dynamic stability, necessary load shedding and
other control actions may be needed in each island, which
always require real-time information throughout the island. In
addition, real-time measurements in different islands should
be collected and analyzed together to determine whether and
how the power system can be restored to normal operation. To
ensure the effectiveness of all the above actions, it is essential
to keep each island totally observable through properly placed
PMUs. In other words, the optimal placement of PMUs should
be carried out in such a manner that the network remains ob-
servable under controlled islanding condition as well as normal
operation condition.
Compared to (3), the observability constraints of OPP-CI
model are modied as follows:
(7)
where is the binary entry in the connectivity matrix for
post-islanding network, which is dened as
if
otherwise.
(8)
For instance, assuming that the controlled islanding is in ef-
fect for the IEEE 14-bus system following a cascading fault,
as shown in [34] and Fig. 1, the whole system is separated
into two subsystems and several lines are opened during the is-
landing process. According to (7), the observability constraints
for OPP-CI can be written explicitly as follows:
(9)
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 745
Comparing (6) and (9), it is concluded that the observability
constraint shown in (7) is stricter than that in (3). Therefore, an
OPP scheme subject to (7) can keep the power network com-
pletely observable both for normal operation scenario and con-
trolled islanding condition.
B. Dealing With Multiple Optimal Solutions
PMUs placement through the objective function (2) and in-
equality constraints (7) may lead to multiple optimal solutions
with the same minimum number of PMUs. For the 14-bus
system in Fig. 1, installations of 5 PMUs in {1, 2, 6, 8, 9}, {1,
4, 6, 7, 9}, {2, 5, 6, 8, 9}, {4, 5, 6, 7, 9}, and {4, 5, 6, 8, 9} can
all satisfy the constraints (7) and lead the system to complete
observability in both normal operation condition and controlled
islanding scenario.
In this study, thus, maximizing the measurement redundancy
is considered as an additional objective to pick out the most suit-
able OPP scheme for power systems. Conventionally, measure-
ment redundancy is dened as the ratio of the number of mea-
surements (including direct measurements and indirect mea-
surements) to the number of states [37]. Considering that the
most important state variables in state estimation are bus voltage
phasors, the measurement redundancy can be redened as the
ratio of the number of voltage measurements to the number of
systembuses. Moreover, the measurement redundancy under is-
landing operation scenario as well as normal operation should
be considered.
To keep consistency with (2) which is a minimization
problem, the objective function of maximizing measurement
redundancy is formulated as a minimization problem as well:
(10)
where is the total number of system buses; constant is
the maximum number of times that the th bus can be observed
in normal operation, which equals to the number of its incident
lines plus one; variable represents the number of times that
the th bus is observed by the solved OPP scheme in normal
operation; and refer to the corresponding constant and
variable in islanding operation condition, respectively; and
are weighting factors assigned to the two components
of the objective function. Since there is greater probability for
a power system to be operated in normal condition than in is-
landing condition, in this study and are set at 0.7
and 0.3, respectively.
Therefore, for the 14-bus system, set {4, 5, 6, 7, 9} is the most
suitable solution because it has smaller value of than other
ones, as shown in Table I.
C. OPP Model Considering Controlled Islanding
In this part, the problem of optimal PMU placement con-
sidering controlled islanding is modeled. The objective of
OPP-CI is to minimize the number of installed PMUs and to
maximize the measurement redundancy with the full network
observability for normal operation and controlled islanding
scenarios as the constraints.
1) OPP-CI Ignoring the Effect of Zero-Injection Bus: : The
ZIBs refer to the network nodes without generation or load con-
nected. A ZIB together with all its incident buses can be dened
TABLE I
COMPARISON ON MEASUREMENT REDUNDANCY OF DIFFERENT
OPP SOLUTIONS FOR IEEE 14-BUS SYSTEM
as a zero injection cluster (ZIC). For a power network, if the in-
uence of ZIBs is ignored or the network does not contain any
ZIBs at all, its OPP-CI model can be formed as
(11)
subject to observability constraints (7), where is the
weighting factor.
Here the two components of the objective function, and
, stand for the considerations of PMU number (2) and mea-
surement redundancy (10), respectively. The weighting factor
is used to determine which factor is more dominant than the
other one in the OPP procedure. In this study, reducing PMU
number is selected as the more important objective. Note that
is the number of incident lines to the th bus; let
and . This way
of specifying can ensure the value of to be less than
1, which guarantees the priority of minimizing PMU number
in OPP-CI problem. As a result, the globally optimal solution
with minimum PMU devices installed and maximum measure-
ment redundancy can be found out.
2) OPP-CI Considering the Effect of Zero-Injection Bus: If
the effect of ZIBs is considered, the total number of PMUs in
OPP problem will be reduced due to the following rules [20]:
In a ZIC, if the zero-injection bus is observable and its ad-
jacent buses are all observable except one, then the unob-
servable bus will be identied as observable by applying
Kirchhoffs current law (KCL) at ZIB.
In a ZIC, if all the buses are observable except the zero-in-
jection one, then the zero-injection bus can be also identi-
ed as observable by using nodal equations.
Combining these two cases can lead to the conclusion that a
ZIC is observable when it has at most one unobservable bus.
Since this unobservable bus could become observable nally
by means of the properties of ZIC, in this paper, it is dened as
pseudo unobservable bus.
Assuming that is the index of the th zero-injection bus
and is the th zero injection cluster, the auxiliary binary
variable is dened so that implies that bus is
the pseudo unobservable bus in . Then the observability
constraint of can be mathematically formulated as
(12)
Notice that the power system could be operated in both
normal operation condition and controlled islanding condition,
and the elements of a ZIC may change due to the line tripping in
746 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
the process of islanding. Thus the incorporation of ZIB effects
into OPP-CI problem should be simultaneously considered for
these two operation scenarios of power system.
For an -bus power network with zero-injection buses,
the observability constraints of OPP-CI considering the effect
of ZIBs are listed as follows:
(13)
Here, and indicate the numbers of times that bus
is observed by means of the installed PMUs and the proper-
ties of ZIC, in normal operation condition and controlled is-
landing condition, respectively. On the other hand, the second
and fourth equations in (13) indicate that the number of pseudo
unobservable buses in a ZIC must be less than or equal to 1,
for both normal scenario and islanding scenario. Therefore, the
constraints in (13) simultaneously ensure that one of the buses
in a ZIC will be observable when it can be reached by PMUs,
or it is the only pseudo unobservable bus in this ZIC.
IV. OPP-CI FORMULATION AGAINST SINGLE
PMU LOSS OR SINGLE LINE OUTAGE
The previous OPP-CI formulation ensures complete observ-
ability of the network assuming a xed network topology after
controlled islanding as well as absolutely reliable measurement
devices. However, PMUs may fail to work due to loss of GPS
signal, failure of measurement instruments or loss of communi-
cation channels. Furthermore, transmission line outages may re-
sult in loss of complete observability. Thus, operators may plan
to have a reliable monitoring system by installing extra PMUs
in the network.
In this section, constraints associated with OPP-CI against
single PMU or single line loss are formulated. Meanwhile, the
objective function still remains the same as (11). Therefore, for
OPP-CI considering each contingency, it is sufcient to replace
the previous constraints with the following related constraints.
A. OPP-CI Formulation Considering Single PMU Outage
Outage of a PMU at bus , denoted as , can be con-
sidered into the previous OPP-CI model by setting the corre-
sponding decision variable to zero. To facilitate the formu-
lation of the optimization problem, a parameter, , is dened
as follows:
if
otherwise
(14)
then, the associated constraints to OPP-CI situation considering
single PMU outage are as follows.
For network without ZIBs:
(15)
For network with ZIBs:
(16)
In these expressions, and are binary auxil-
iary variables whose values are equal to 1, if bus is the pseudo
unobservable bus for and (the th zero injection
cluster in controlled islanding condition), respectively.
B. OPP-CI Formulation Considering Single Line Outage
Outage of a line may cause the loss of observability for one of
its terminal buses which would otherwise be observable using
current phasor of that line [15].
For a power network with lines, single line contingen-
cies can be dened. The connectivity matrices for the network
in normal operation condition and controlled islanding condi-
tion change in each of such dened contingencies. Let param-
eters and be dened as th entries of the
connectivity matrices for networks in normal and islanding sce-
narios, respectively, where superscript represents loss of
line . Thus, related constraints of OPP-CI considering single
line outage are as follows:
For network without ZIBs:
(17)
For network with ZIBs:
(18)
Similarly, and are binary auxiliary variables
whose values are equal to 1, if bus is pseudo unobservable bus,
when line is out, for and , respectively.
C. OPP-CI Formulation Considering Single Contingencies
In this case, the assumed contingency for power system could
be either one of the two types, i.e., the single PMU loss or the
single line outage. Hence, the set of constraints dened in the
two previous subsections should be considered simultaneously.
It can be expected that more PMUs are required in order to main-
tain network observability in this case than in cases with only
single contingency.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 747
TABLE II
SPECIFICATIONS OF TEST SYSTEMS
TABLE III
CONTROLLED ISLANDING SCHEMES FOR DIFFERENT SYSTEMS
TABLE IV
COMPARISON OF OPP RESULTS WITH OR WITHOUT CONSIDERATION OF CONTROLLED ISLANDING (IGNORING THE EFFECT OF ZIB)
V. CASE STUDIES AND RESULTS
The proposed OPP-CI model was tested on the IEEE 14-, 30-,
39-, 118-bus systems, and the Polish 2383-bus system, whose
detailed information can be found in Table II. The case studies
were performed in two parts. First, for all the test systems, OPPs
considering controlled islanding, with and without inclusion of
the effect of zero-injection bus, are carried out and the results
are compared with those neglecting controlled islanding. Next,
single line and single PMU contingencies are taken into account
and their inuence on the OPP-CI solution is studied. The simu-
lation results with respect to IEEE standard systems and Polish
2383-bus system are given in the following. All simulations are
executed in a laptop having a 2.60-GHz dual-core CPU and 4
GB of RAM. The OPP problem is modeled in MATLAB and
solved by CPLEX Toolbox for MATLAB.
A. Case Results for IEEE Standard Test Systems
1) OPP Considering Controlled Islanding: At the rst part,
optimal PMU placement is carried out so as to achieve totally
observability of network under both normal operation condition
and controlled islanding condition. To perform the OPP-CI pro-
cedure, the controlled islanding plans for different IEEE sys-
tems should be known a priori. In this paper, these controlled
islanding schemes are extracted from [34], [35], and [13]. For
clarity, they are listed again in Table III. As for IEEE 14-bus
system, two islands with 6 buses and 8 buses in each island,
respectively, are included in the islanding scheme. However, as
mentioned in Section III, the proposed OPP-CI model is not just
suitable to the above controlled islanding cases but also can be
applied to any other controlled islanding schemes.
Table IV provides the comparison of the number and loca-
tions of required PMUs resulting fromOPP with or without con-
sideration of controlled islanding. The effect of zero-injection
bus is incorporated into the comparison in Table V. the PMUs
installation percentage in the two tables refers to the ratio be-
tween the number of PMUs and the number of system buses.
Results of Tables IV and V reveal that generally more PMUs
are required by power network to maintain observability for
both controlled islanding scenario as well as normal condition.
748 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
TABLE V
COMPARISON OF OPP RESULTS WITH OR WITHOUT CONSIDERATION OF CONTROLLED ISLANDING (INCLUDING THE EFFECT OF ZIB)
TABLE VI
OPP-CI RESULTS WITH CONSIDERATION OF SINGLE CONTINGENCIES (IGNORING THE EFFECT OF ZIB)
TABLE VII
OPP-CI RESULTS WITH CONSIDERATION OF SINGLE CONTINGENCIES (INCLUDING THE EFFECT OF ZIB)
Comparing Tables IV and V , it is also noticed that considering
the inuence of zero-injection bus reduces the number of re-
quired PMUs in all cases.
2) OPP-CI Considering Contingencies of Single Line and
PMU Outages: In this part, OPP-CI is implemented for IEEE
14-, 30-, 39- and 118-bus systems considering contingencies
of line or PMU outages. Corresponding to the constraints
(15)(16), (17)(18) and the combination of (15)(18), three
different cases are considered, i.e., outage of a single line, loss
of a single PMU, and a single contingency of line outage or
PMU loss. The inuence of zero-injection bus is neglected in
Table VI, while it is considered in Table VII.
As expected, a robust measurement system against single
PMUor line outages needs more PMUs than the case neglecting
contingencies. Additionally, in comparison with single line
outage, single PMU loss has more adverse impact on the net-
work observability, which can be concluded from the required
number of PMUs in the relevant cases.
Table VIII shows the CPU computation times for solving
OPP-CI problems. For each IEEE test system, only the time
needed for the most complex calculation, i.e., OPP-CI consid-
ering the single contingency and the effect of ZIB, is listed.
In all the previous calculations in the paper, the uncertain-
ties of PMUmeasurements and network parameters are ignored,
and the values of weighting factors and are xed. How-
ever, two appendixes are added at the end of this paper: in Ap-
pendix A, the uncertainties associated with the voltage phasors
measured or computed by the proposed OPP-CI schemes are
assessed, while the variances of OPP-CI results with different
values of and are shown in Appendix B.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 749
TABLE VIII
CPU COMPUTATION TIMES FOR CALCULATIONS OF OPP-CI CONSIDERING SINGLE CONTINGENCY AND EFFECT OF ZIB
TABLE IX
OPP RESULTS AND THEIR ASSOCIATED CPU COMPUTATION TIMES FOR POLISH 2383-BUS SYSTEM
TABLE X
STANDARD UNCERTAINTIES IN THE INDIRECT-MEASUREMENTS
OF MAGNITUDES AND PHASOR ANGLES OF VOLTAGE
PHASORS FOR IEEE 14-BUS SYSTEM
B. Case Results for Polish 2383-Bus System
The details about the Polish 2383-bus system can be obtained
from [38] which indicates that all the buses in this system are
divided into different areas. Based on this, a CI scheme for the
Polish 2383-bus system is assumed, in which the partition of
controlled islands is roughly consistent with that of sub-areas
determined in [38]. However, a few buses are repartitioned in
the CI scheme to avoid isolated bus or areas in each island.
Finally, in the CI scenario the whole Polish 2383-bus system
will be divided into 5 islands with 369 buses, 281 buses, 880
buses, 560 buses, 293 buses, respectively. The OPP-CI results
with different considerations and their associated CPU compu-
tation times are shown in Table IX.
VI. CONCLUSION
An effective OPP scheme should ensure complete observ-
ability of a power network under various operation conditions.
To avoid wide-area blackout following cascading failures,
power system might be operated in controlled islanding mode.
TABLE XI
STANDARD UNCERTAINTIES IN THE MEASUREMENTS OF VOLTAGE
PHASORS AT PSEUDO UNOBSERVABLE BUS FOR IEEE 14-BUS SYSTEM
In this paper, an OPP model considering controlled islanding
of power system is proposed. The proposed model guarantees
complete observability of power network for normal condition
as well as controlled islanding condition, with or without
considering the effect of zero-injection bus. By introducing
the measurement redundancy into the optimization objective,
our OPP-CI model can nd the globally optimal solution with
the minimum number of PMUs and maximum measurement
redundancy. Furthermore, single PMU or line loss is also in-
corporated into the model. At last, case studies on several IEEE
standard test systems and a large-scale practical system provide
verication of the effectiveness of the presented OPP models.
APPENDIX A
EVALUATION OF MEASUREMENT UNCERTAINTY
FOR IEEE 14-BUS AND 118-BUS SYSTEMS
The uncertainties associated with the voltage phasors mea-
sured or computed by the proposed OPP-CI congurations are
evaluated in this appendix. For each IEEEtest system, the uncer-
tainty evaluation is performed only for the OPP-CI scheme con-
sidering the effect of ZIB, which implies the minimum number
of installed PMUs and consequently the worst performance on
uncertainty.
Detailed formulas of uncertainty calculation for different
types of measurements are derived in the following.
Direct Measurements: The uncertainties of direct measure-
ments are calculated from the manufacturers specications.
Assuming that the probability of measurement uncertainty is
of uniform distribution and the is the bounding limits
of the measurement of , the standard uncertainty in the
750 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
TABLE XII
OPP-CI RESULTS WITH REGARDS TO DIFFERENT VALUES OF FOR IEEE 30-BUS SYSTEM
measurement can be expressed as [39]
(19)
where is usually specied by PMU manufacturers [40].
Indirect-Measurements: The uncertainties for indi-
rect-measurements are evaluated by using the classical uncer-
tainty propagation theory [39].
Let there be a PMU installed at bus , with bus connected to
bus through line . With the help of PMU measurements,
i.e., the voltage phasor at bus and the current
phasor through the line , the voltage at bus
can be expressed as
(20)
where and are line resistance and reactance.
By decoupling (20) into two equations in the real-imaginary
coordinate system and solving the new equations, the magni-
tude and the phasor angle of the voltage phasor can be
obtained, which are functions of the magnitude and phase angle
of and , and the parameters of line :
(21)
Assuming that the input quantities in (21) are uncorrelated
(similar assumptions are made for the following derivations),
the combined standard uncertainty of the voltage magnitude
and the phase angle , according to [39], can be given by
(22)
where is the partial derivative, , , , , ,
, and is the standard uncertainty in the measurement
.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 751
TABLE XIII
OPP-CI RESULTS WITH REGARDS TO DIFFERENT VALUES OF FOR IEEE 39-BUS SYSTEM
Measurements Calculated With the Properties of ZIC: For
this type of measurements, i.e., the voltage phasors at the pseudo
unobservable buses, their uncertainties can also be obtained by
the classical uncertainty propagation theory.
If the pseudo unobservable bus is exactly the ZIB of one
ZIC, it implies that there is no bus in the ZIC with PMU
installed. The voltage phasor of bus is obtained from the
following KCL equation:
(23)
where is the number of buses in this ZIC, is the number of
lines connecting bus and the zero-injection bus . and
are resistance and reactance, respectively, of th line between
bus and bus .
The voltage magnitude and phasor angle of bus can then be
expressed as
(24)
where and are sets of voltage magnitudes and phasor
angles, respectively, of buses in the ZIC except the bus .
and are parameter sets of lines incident to the bus .
Thus, the standard uncertainty of and can be given by
(25)
with , , , and is the total number of variables
in .
For the situation that the pseudo unobservable bus is not
the zero-injection bus (assuming that bus is the zero-injec-
tion bus), there may exist some PMUs having been placed at
the buses incident to bus . Another equation, thus, should be
formulated to obtain the bus s voltage phasor:
(26)
here, is the set of PMU buses in the ZIC and refers to the
set of buses without PMU installed. is the current phasor
through the th line from bus to bus .
752 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
TABLE XIV
OPP-CI RESULTS WITH REGARDS TO DIFFERENT VALUES OF FOR IEEE 30-BUS SYSTEM
Similarly to (24), there exists a new pair of equations for the
magnitude and phasor angle of voltage phasor :
(27)
where and are sets of magnitudes and phasor angles, re-
spectively, of the currents from buses in to bus . and
are sets of voltage magnitudes and phasor angles of buses in
except the bus , respectively. and are parameter sets of
lines from buses in to bus .
Then the standard uncertainties can be obtained with the fol-
lowing formula:
(28)
Table X shows the standard uncertainties in the voltage
phasors corresponding to the indirect-measurements, while
Table XI gives the standard uncertainties in the voltage phasors
of pseudo unobservable buses. The uncertainties of the PMU
buses are excluded, since they can be directly computed from
(19). The typical values of maximum uncertainties in PMU
measurements are specied by the manufacturer in [40], where
the maximum uncertainties for voltage and current magnitude
Fig. 2. Standard uncertainties in calculated voltage phasors for the IEEE
14-bus system.
are 0.02% and 0.03% of the actual values, respectively; and
the maximum error in the measurement of phase angle is
0.01 degrees. The actual values of phasor measurements are
determined by performing power ow for the power system.
Additionally, a 5% uncertainty is assumed to all transmission
line parameters.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 753
TABLE XV
OPP-CI RESULTS WITH REGARDS TO DIFFERENT VALUES OF FOR IEEE 39-BUS SYSTEM
It is noted that under the proposed OPP-CI conguration a
bus may be observed by more than one PMU, such as bus 7
in Table X. In that case, a pair of measurement uncertainties,
i.e., uncertainty in voltage magnitude and uncertainty in voltage
phasor angle, can be obtained with regard to each connected
PMU. The minimum value of magnitude uncertainty and min-
imum value of angle uncertainty should be chosen and treated
as the nal uncertainties for that bus. Additionally, there are
some buses located on the boundary of islands. In other words,
these buses have lines incident to other islands, such as bus 2 in
IEEE 14-bus system. As shown in Table X, the CI process may
cause the loss of the observations on these buses from the PMUs
located in other islands. Therefore, their measurement uncer-
tainties in CI condition may be different from that in normal
condition.
The standard uncertainties in calculated voltage phasors for
IEEE 14-bus system are shown in Fig. 2. The order in which
the buses are selected to depict the measurement uncertainties
corresponds to the bus indices. In other words, the order for
IEEE 14-bus system is 1-3, 7-8, and 10-14. In Fig. 2, the solid
circle and x-mark refer to the measurement uncertainties for
normal operation condition, while the dashed circle and x-mark
are used to display the measurement uncertainties in CI opera-
tion condition.
Fig. 3 shows the standard uncertainties associated with cal-
culated voltage phasors for IEEE 118-bus system. There are 4
buses having different uncertainties for normal condition and
CI condition. Among them buses 26 and 65 are the boundary
buses with tie lines incident to other islands, and buses 26 and
65 are pseudo unobservable buses. It is noted that buses 26 and
65 have small measurement uncertainties in CI condition than
that in normal condition. The reason is that the bus numbers
of the ZICs which they belonged to are reduced due to the CI
process. This leads to the reduction of uncertainty sources in the
calculation equations of their voltage phasors, and consequently
results in smaller measurement uncertainties.
APPENDIX B
OPP-CI RESULTS WITH DIFFERENT AND
FOR IEEE 30-BUS AND 39-BUS SYSTEMS
Results With Different Weighting Factor : Tables XII
and XIII show the OPP-CI results with regards to different
values of for IEEE 30-bus system and 39-bus system,
754 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 29, NO. 2, MARCH 2014
Fig. 3. Standard uncertainties in calculated voltage phasors for the IEEE
118-bus system.
respectively. For all of the calculations, the values of are set
the same as before, i.e., .
In these two tables, corresponds to the results cal-
culated in Section V. refers to the case in which only
the measurement redundancy of normal operation condition is
considered, while indicates that only the measurement
redundancy of islanding condition is activated. The cells with
the same OPP result are joined together.
Since the priority of minimizing PMU number is ensured in
(11) by the selected value of , all the OPP results for a given
OPP-CI strategy are with the same minimum number of PMU,
as shown in the tables. However, the PMU locations may vary
with the value of , especially for the cases associated with
the IEEE 30-bus system. It is because that the weighted propor-
tions of the two components in (10), i.e., the measurement re-
dundancy differences in normal condition and islanding condi-
tion, are changed due to the varying . Consequently, for those
buses with tie lines connected to other islands, their weighted
measurement redundancy differences will be changed.
Results With Different Weighting Factor : The variances
of OPP-CI solutions with the weighting factor in (11) are
listed in Tables XIV and XV. All the calculations are accom-
plished under the value of . Similarly, for each table
the cells with the same result are combined together.
The values of calculated from are
0.1111 for IEEE 30-bus system and 0.1429 for 39-bus system.
For the cases of , the component in (11) is
an integer and the component has the value between 0 and
1. This guarantees that minimizing PMU number is more domi-
nant than maximizing measurement redundancy in the OPP pro-
cedure. Thus with a given , a robust OPP solution can be ob-
tained for each OPP-CI strategy. On the other hand, the value
of will increase with and may exceed 1. In that case,
additional PMU may be needed due to the trade-off between the
objective of minimizing PMUnumber and the objective of max-
imizing measurement redundancy.
REFERENCES
[1] N. M. Manousakis, G. N. Korres, and P. S. Georgilakis, Taxonomy
of PMU placement methodologies, IEEE Trans. Power Syst., vol. 27,
no. 2, pp. 10701077, May 2012.
[2] A. Abur and A. G. Exposito, Power System State Estimation: Theory
and Implementation. New York, NY, USA: Mercel Dekker, 2004.
[3] B. Milosevic and M. Begovic, Nondominated sorting genetic algo-
rithmfor optimal phasor measurement placement, IEEE Trans. Power
Syst., vol. 18, no. 1, pp. 6975, Feb. 2003.
[4] A. Enshaee, R. A. Hooshmand, and F. H. Fesharaki, Anewmethod for
optimal placement of phasor measurement units to maintain full net-
work ob-servability under various contingencies, Elect. Power Syst.
Res., vol. 89, no. 1, pp. 110, Aug. 2012.
[5] B. K. S. Roy, A. K. Sinha, and A. K. Pradhan, An optimal PMU place-
ment technique for power system observability, Int. J. Elect. Power
Energy Syst., vol. 42, no. 1, pp. 7177, Nov. 2012.
[6] T. L. Baldwin, L. Mili, M. B. Boisen, and R. Adapa, Power systemob-
servability with minimal phasor measurement placement, IEEETrans.
Power Syst., vol. 8, no. 2, pp. 707715, May 1993.
[7] B. Xu and A. Abur, Observability analysis and measurement place-
ment for systems with PMUS, in Proc. IEEE Power Eng. Soc. Power
Systems Conf. Expo., Oct. 2004, pp. 943946.
[8] B. Gou, Generalized integer linear programming formulation for op-
timal PMU placement, IEEE Trans. Power Syst., vol. 23, no. 3, pp.
10991104, Aug. 2008.
[9] M. Korkali and A. Abur, Placement of PMUs with channel limits, in
Proc. IEEE Power Eng. Soc. General Meeting, 2009.
[10] F. Aminifar, A. Khodaei, M. Fotuhi-Firuzabad, and M. Shahidehpour,
Contingency-constrained PMU placement in power networks, IEEE
Trans. Power Syst., vol. 25, no. 1, pp. 516523, Feb. 2010.
[11] M. R. Aghamohammadi and A. Shahmohammadi, Intentional
islanding using a new algorithm based on ant search mechanism,
Int. J. Elect. Power Energy Syst., vol. 35, no. 1, pp. 138147, Feb.
2012.
[12] G. Xu, V. Vittal, A. Meklin, and J. E. Thalman, Controlled islanding
demonstrations on the WECC system, IEEE Trans. Power Syst., vol.
26, no. 1, pp. 334343, Feb. 2011.
[13] L. Ding, F. M. Gonzalez-Longatt, P. Wall, and V. Terzija, Two-step
spectral clustering controlled islanding algorithm, IEEE Trans. Power
Syst., vol. 28, no. 1, pp. 7584, Feb. 2013.
[14] S. S. Ahmed, N. C. Sarker, A. B. Khairuddin, M. R. B. A. Ghani, and
H. Ahmad, A scheme for controlled islanding to prevent subsequent
blackout, IEEE Trans. Power Syst., vol. 18, no. 1, pp. 136143, Feb.
2003.
[15] S. Azizi, A. S. Dobakhshari, S. A. N. Sarmadi, and A. M. Ranjbar,
Optimal PMU placement by an equivalent linear formulation for ex-
haustive search, IEEE Trans. Smart Grid, vol. 3, no. 1, pp. 174182,
Mar. 2012.
[16] B. Milosevic and M. Begovic, Nondominated sorting genetic algo-
rithmfor optimal phasor measurement placement, IEEE Trans. Power
Syst., vol. 18, no. 1, pp. 6975, Feb. 2003.
[17] J. Peng, Y. Sun, and H. F. Wang, Optimal PMU placement for full
network observability using tabu search algorithm, Int. J. Elect. Power
Energy Syst., vol. 28, no. 4, pp. 223231, May 2006.
[18] C. Peng, H. Sun, and J. Guo, Multi-objective optimal PMU placement
using a non-dominated sorting differential evolution algorithm, Int. J.
Elect. Power Energy Syst., vol. 32, no. 8, pp. 886892, Oct. 2010.
[19] M. Hajian, A. M. Ranjbar, T. Amraee, and B. Mozafari, Optimal
placement of PMUs to maintain network observability using a modi-
ed BPSO algorithm, Int. J. Elect. Power Energy Syst., vol. 33, no.
1, pp. 2834, Jan. 2011.
[20] F. Aminifar, C. Lucas, A. Khodaei, and M. Fotuhi-Firuzabad, Optimal
placement of phasor measurement units using immunity genetic algo-
rithm, IEEE Trans. Power Del., vol. 24, no. 3, pp. 10141020, Jul.
2009.
[21] M. Hurtgen and J.-C. Maun, Optimal PMU placement using iterated
local search, Int. J. Elect. Power Energy Syst., vol. 32, no. 8, pp.
857860, Oct. 2010.
HUANG et al.: OPTIMAL PMU PLACEMENT CONSIDERING CONTROLLED ISLANDING OF POWER SYSTEM 755
[22] R. F. Nuqui and A. G. Phadke, Phasor measurement unit placement
techniques for complete and incomplete observability, IEEE Trans.
Power Del., vol. 20, no. 4, pp. 23812388, Oct. 2005.
[23] I. Kamwa and R. Grondin, PMU conguration for system dynamic
performance measurement in large multiarea power systems, IEEE
Trans. Power Syst., vol. 17, no. 2, pp. 385394, May 2002.
[24] B. Gou and A. Abur, A direct numerical method for observability
analysis, IEEE Trans. Power Syst., vol. 15, no. 2, pp. 625630, May
2000.
[25] G. N. Korres, P. J. Katsikas, K. A. Clements, and P. W. Davis, Nu-
merical observability analysis based on network graph theory, IEEE
Trans. Power Syst., vol. 18, no. 3, pp. 10351045, Aug. 2003.
[26] M. Gl and A. Abur, Observability analysis of systems containing
phasor measurements, in Proc. 2012 IEEE Power Eng. Soc. General
Meeting, pp. 16.
[27] M. Gl and A. Abur, Observability and Criticality Analyses for
Power Systems Measured by Phasor Measurements, IEEE Trans.
Power Syst., to be published.
[28] G. N. Korres and N. M. Manousakis, Observability analysis and
restoration for systems with conventional and phasor measurements,
Int. Trans. Elect. Energy Syst., DOI: 10.1002/etep.1684., 2012.
[29] G. N. Korres and N. M. Manousakis, Observability analysis and
restoration for state estimation using SCADA and PMU data, in Proc.
2012 IEEE Power Eng. Soc. General Meeting, pp. 18.
[30] R. Sodhi, S. C. Srivastava, and S. N. Singh, Optimal PMU placement
method for complete topological and numerical observability of power
system, Elect. Power Syst. Res., vol. 80, no. 9, pp. 11541159, Sep.
2010.
[31] B. Gou, Optimal placement of PMUs by integer linear programming,
IEEE Trans. Power Syst., vol. 23, no. 3, pp. 15251526, Aug. 2008.
[32] K. Sun, D.-Z. Zheng, and Q. Lu, Splitting strategies for islanding op-
eration of large-scale power systems using OBDD-based methods,
IEEE Trans. Power Syst., vol. 18, no. 2, pp. 912923, May 2003.
[33] P. A. Trodden, W. A. Bukhsh, A. Grothey, and K. I. M. McKinnon,
MILP formulation for controlled islanding of power networks, Int.
J. Elect. Power Energy Syst., vol. 45, no. 1, pp. 501508, Feb. 2013.
[34] S. A. Nezam-Sarmadi, S. Nouri-Zadeh, A. M. Ranjbar, and M. R.
Pishvaie, An islanding algorithm to restore a PMU installed power
system, in Proc. 2010 IEEE Asia-Pacic Power and Energy Eng.
Conf., pp. 14.
[35] B. Pradhan, K. H. Reddy, D. S. Roy, and D. K. Mohanta, Intentional
islanding of electric power systems in a grid computing framework: a
graph-theoretic approach, in Proc. 2011 Int. Conf. Recent Trends in
Inf. Syst., pp. 156160.
[36] M. Rihan, M. Ahmad, M. S. Beg, and A. Anees, Optimal PMU place-
ment on network branches for intentional islanding to prevent black-
outs, Int. J. Emerg. Technol. Adv. Eng., vol. 3, no. 3, Mar. 2013.
[37] A. Monticelli, Electric power system state estimation, Proc. IEEE,
vol. 88, no. 2, pp. 262282, Feb. 2000.
[38] R. D. Zimmerman, C. E. Murillo-Snchez, and D. Gan, MAT-
POWER: a MATLAB Power System Simulation Package, Cornell
Univ.. Ithaca, NY, USA. [Online]. Available: http://www.pserc.cor-
nell.edu/matpower.
[39] JCGM 100:2008: Evaluation of Measurement DataGuide to the Ex-
pression of Uncertainty in Measurement, 2008.
[40] Model 1133A GPS-Synchronized Power Quality/Revenue Standard,
Operation Manual, Arbiter Systems. Paso Robles, CA, USA.
Lei Huang (S13) was born in Hunan, China, in
1987. He received the B.E. degree in electrical
engineering from Wuhan University, Wuhan, China,
in 2009. Currently he is pursuing the Ph.D. degree
in the School of Electrical Engineering, Wuhan Uni-
versity. He is also a visiting scholar in the University
of Denver, Denver, CO, USA.
His research interests include optimal PMU place-
ment, voltage stability analysis and wind power.
Yuanzhang Sun (SM01) received the Ph.D. degree
in electrical engineering from Tsinghua University,
Beijing, China, in 1988.
He is currently a Professor of the School of Electric
Engineering, Wuhan University, Wuhan, China. He is
also an adjunct Professor at Tsinghua University. His
main research interests are power system dynamics
and control, voltage stability, and renewable energy.
Jian Xu (M08) received the B.E. and Ph.D. degrees
in electrical engineering from Wuhan University,
Wuhan, China, in 2002 and 2007, respectively.
Currently he is an Associate Professor in the
School of Electric Engineering, Wuhan University.
Also, he is a visiting scholar in Washington State
University, Pullman, WA, USA. His research inter-
ests are PMU application, power system operation,
and voltage stability.
Wenzhong (David) Gao (SM03) received the M.S.
and Ph.D. degrees in electric power engineering from
Georgia Institute of Technology, Atlanta, GA, USA,
in 1999 and 2002, respectively.
He is currently an Associate Professor in the De-
partment of Electrical and Computer Engineering,
University of Denver, Denver, CO, USA. His re-
search interests are renewable energy, smart grid,
and power system analysis.
Jun Zhang (M09) received the Ph.D. degree in
electric engineering from Arizona State University,
Tempe, AZ, USA, in 2008.
He is currently an Assistant Professor in the De-
partment of Electrical and Computer Engineering,
University of Denver, Denver, CO, USA. His re-
search interests are smart grid and statistical signal
processing.
Ziping Wu (S12) received the B.E. degree in
thermal power engineering and the M.S. degree
in electrical power engineering from North China
Electric Power University, Beijing, China, in 2006
and 2009, respectively. Now he is pursuing the Ph.D.
degree in the Department of Electrical and Computer
Engineering, University of Denver, Denver, CO,
USA.
His research interests include wind power genera-
tion, renewable energy, and smart grid.