552
1, JANUARY 2017
A PMU Placement Scheme Considering Realistic
Costs and Modern Trends in Relaying
Anamitra Pal, Member, IEEE, Anil Kumar S. Vullikanti, Member, IEEE, and S. S. Ravi, Senior Member, IEEE
Abstract—Synchrophasor deployment costs have evolved over
time. The cost of upgrading a substation, which is much larger
than the cost of an individual device, has emerged as the primary
constituent of the total expenditure. Given these circumstances, the
optimal phasor measurement unit placement formulation needs to
consider not only the number of devices that must be placed at the
substations, but also the number of substations that must be up-
graded to support those devices. This paper presents an integer lin-
ear programming methodology for such a placement scheme while
considering realistic costs and practical constraints. The IEEE 30
bus system is used to illustrate the proposed concept, while the
IEEE 118, IEEE 300, and Polish 2383 bus systems are used to show
the performance of the method under different test environments.
Index Terms—Critical-bus concept, dual-use line relay (DULR),
integer linear programming (ILP), N–t contingency, observability,
optimal cost, phasor measurement unit (PMU).
I. INTRODUCTION
HASOR measurement unit (PMU) based state estimation
P was proposed for the first time in 1985–1986 [1]. Since
then a great deal of research has been carried out to develop
efficient state estimation algorithms (hybrid/linear) [2]–[5] and
compute optimal locations for placing PMU devices in the grid.
The latter studies were considered to be mostly “theoretical”
or “academic” until, courtesy of Federal funding, many utili-
ties began placing these devices in bulk in their networks [6].
While doing so, they realized that factors other than the price
of the device itself were primarily responsible for the high cost
of synchrophasor deployment. These factors included commu-
nication infrastructure costs, security (cybersecurity) costs, and
labor costs [7]. Under such circumstances, it became essential
to reexamine the formulation of the optimal PMU placement
(OPP) problem.
In the traditional formulation, the goal is to find the minimum
number of PMU devices that, when placed in the power net-
work, ensures the attainment of some pre-defined objective(s).
Manuscript received October 4, 2015; revised February 18, 2016; accepted
April 3, 2016. Date of publication April 7, 2016; date of current version Decem-
ber 20, 2016. This work was supported in part by the following grants: Defense
Threat Reduction Agency Grant HDTRA1-11-1-0016, DTRA Comprehensive
National Incident Management System Contract HDTRA1-11-D-0016-0001,
and National Science Foundation Network Science and Engineering Grant CNS-
1011769. Paper no. TPWRS-01393-2015.
A. Pal is with the Network Dynamics and Simulation Science Laboratory,
Biocomplexity Institute, Virginia Tech, Blacksburg, VA 24061 USA (e-mail:
anam86@vbi.vt.edu).
A. K. S. Vullikanti is with the Department of Computer Science and the
Biocomplexity Institute, Virginia Tech, Blacksburg, VA 24061 USA (e-mail:
akumar@vbi.vt.edu).
S. S. Ravi is with the Department of Computer Science, University of
Albany—State University of New York, Albany, NY 12222 USA (e-mail:
sravi@albany.edu).
Digital Object Identifier 10.1109/TPWRS.2016.2551320
0885-8950 © 2016 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.
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 32, NO.
Over the past two decades, extensive research has been done
to solve this problem. The popular techniques that have been
employed for achieving optimal results include simulated an-
nealing, integer programming, exhaustive search, weighted least
squares (WLS), Tabu search, (immunity) genetic algorithm, par-
ticle swarm optimization, etc. For the purposes of brevity, the
following literature survey is limited to some of the archival
papers that have been published recently. For a more detailed
description of PMU placement methodologies, [8] and [9] are
suggested.
In [10], PMU deployment was approached based on a state
estimation theoretic criterion. The optimization was performed
via convex relaxation and semidefinite programming. In [11],
constraints such as transmission line outages and measurement
channel limitations were incorporated into the OPP problem for-
mulation through an integer linear programming (ILP) model.
ILP was also used in [12] and [13] for solving the OPP problem
under controlled islanding conditions and detecting bad data
in a hybrid state estimator, respectively. An enumeration-based
ILP was used in [14] for enhancing the reliability of subsequent
PMU additions. A mixed-ILP based multi-objective probabilis-
tic model for PMU placement was developed in [15]. Fuzzy
logic was used in [16] for creating a PMU placement frame-
work that enhances wide-area situational awareness. In [17],
PMU placement was done using a greedy algorithm for min-
imizing state estimation errors. The OPP problem was solved
in [18] by maximizing the determinant of the empirical observ-
ability Gramian for performing robust dynamic state estimation.
A WLS algorithm with continuous decision variables was used
to solve the OPP problem in [19], while sequential quadratic
programming was used in [20] to attain the dual objectives of
minimizing the total number of PMU devices and maximizing
measurement redundancy of all buses.
Some inferences drawn on the basis of the above literature
survey are summarized as follows:
1) ILP is the most popular choice for solving OPP problems
[11]–[15]: The reason for this is that unlike meta-heuristic
approaches, ILP always gives an optimal solution. More-
over, with powerful processors and efficient optimizers
(such as Gurobi and CPLEX), even for large systems,
the computational time required for finding the optimal
locations for PMU placement is not substantial.
2) Transformer tap ratios are not taken into account
[8]–[20]: The tap settings of the transformers are either
assumed to be known or ignored altogether.
3) Most of the PMUs considered in the literature are of the
“bus” type [10]–[20]: That is, when placed at a bus of
the system, the PMU observes the voltage of that bus and
PAL et al.: PMU PLACEMENT SCHEME CONSIDERING REALISTIC COSTS AND MODERN TRENDS IN RELAYING
some or all of the branch currents (depending on whether
measurement channel limitations are considered or not)
that emerge from that bus.
4) Those solutions that satisfy all pre-defined system con-
straints and require the least number of PMUs for doing
so are considered the best solutions [10]–[20]: This is be-
cause the cost of the device is assumed to be the primary
reason for not placing PMUs at all buses.
In this paper, these inferences will be analyzed using realistic
costs obtained from publicly available reports, as well as by
considering trends in relaying technologies that are currently in
vogue.
The rest of the paper is structured as follows. Section II de-
scribes through illustrations the need for a new PMU placement
scheme. The mathematical formulation of the problem and the
practical constraints that are handled by the proposed approach
are described in Section III. The results obtained by applying this
approach to standard IEEE systems as well as a 2383 bus Polish
system are summarized in Section IV. Section V concludes the
paper.
II. NEED FOR A NEW PMU PLACEMENT SCHEME
As can be seen from the literature survey in Section I, the
basic objective of any OPP scheme is observability. From a PMU
placement perspective, observability is defined as the ability to
measure, either directly or indirectly, some or all of the complex
voltages (known as states) of a network. Topologically, a power
network is a graph whose nodes are buses, one or more of which
are placed inside a substation, and whose edges are transmission
lines or transformers. However, topological observability is not
the only criterion for a robust PMU placement scheme. Flows
in a power system are governed by Kirchhoff’s laws and a
variety of transient and dynamic stability constraints must be
satisfied for the power system to remain resilient to failures. A
big challenge with respect to the resilience requirement is the
successful integration of large-scale renewable energy sources
(especially massive wind farms) with the high-voltage network.
PMUs have been found to play important roles in these contexts
as they provide real-time knowledge of the operation state [21],
[22]. Thus, these criteria must be incorporated while choosing
locations for placing PMUs in the grid.
The second factor to be considered regarding PMU placement
is that although voltages of the buses are meant to be observed,
PMUs can only be placed in substations. A bus is a single node of
the network having a particular voltage value, whereas a substa-
tion is a collection of one or more buses. When a PMU is placed
at a substation, that substation has to often undergo disruption
of its normal service in order to set up the communication and
(cyber)-security infrastructures. Although utilities prefer doing
these upgrades during scheduled outages of a substation or dur-
ing the off-peak load period, this may not always be possible in a
liberalized electricity market [23]. Similarly, a report published
by the U.S. Department of Energy (DOE) stated that the cost
of a new installation can be almost 20 times the cost of a PMU
device [7]. Therefore, the cost of upgrading a substation is an
important component of the synchrophasor installation cost and
must be considered in the OPP formulation. In [24], the task
553
of placing bus-type PMUs was done substation-wise. However,
the intent of that paper was to minimize the total number of
PMU devices. As pointed in [7] and [23], device cost is not
a major portion of the PMU placement cost. Therefore, from a
cost minimization perspective such a formulation is not optimal.
Mishra et al. [25] also came to a similar conclusion. However,
Mishra et al. [25] used binary particle swarm optimization for
attaining its objectives which is not guaranteed to always give
optimal solutions [11], [12], [26].
The third factor that influences modern PMU placement is the
availability of dual-use line relays (DULRs) acting as PMUs.
Also called branch PMUs in some publications (see [27], [28]),
they are designed to monitor a single edge by measuring the
voltage and current phasors at one end of the monitored edge.
Unlike traditional PMUs (TPMUs) which are stand-alone and
whose single function is measuring synchrophasors, DULRs
are transmission line and transformer protection relays (digi-
tal relays) that while performing their protection functions also
report phasor data. The advantage of using DULRs is that by
enabling synchrophasor functionality inside digital relays, addi-
tional substation panels are not required. This reduces the total
cost and saves precious control house space. DULRs also pro-
vide added benefits such as uniform distribution in the network,
adaptability to deployment in multiple stages, higher reliability,
and time-tagged breaker statuses which are practically a ne-
cessity for linear state estimators [5]. Although Gomez et al.
[27] highlighted the use of these devices for enhancing reliabil-
ity of power networks, to the best of our knowledge no PMU
placement model has exploited their use from a cost-reduction
perspective.
To summarize, a new PMU placement scheme is needed to
achieve the following objectives:
1) Different buses have different priorities [23]: Since some
buses are more important than others from system stabil-
ity/security perspectives, they must be given higher pref-
erence when placing PMUs.
2) Substations are different from buses: The new scheme
must minimize the number of substations where PMUs
are placed for observing all the buses of the network,
while being subject to practical constraints.
3) DULRs are to be used for ensuring observability and
redundancy: A DULR can measure voltage of both ends
of an edge. Now, if one node has two incident edges, both
of which have DULRs on either of their ends, then that
node satisfies the N − 1 contingency criterion (which is
the industry norm). It will be shown in the next section
that the methodology developed here can provide N − t
redundancy, where t ≥ 1.
One way to attain these objectives is to treat substations as
super-nodes and grouping buses that are inside a substation into
a single node to find the optimal number of such super-nodes
where PMUs should be placed. However, this approach will not
ensure complete bus observability. This is explained through
Fig. 1, which shows a multi-level voltage system where circles
indicate substations and colored dots represent buses. Consider-
ing the most general case, it is assumed that all buses/substations
have non-zero injections and that the transformer tap settings
between the different voltage levels inside a substation are not
554
Fig. 1. Example system illustrating the inability of super-node based observ-
ability to ensure complete bus observability.
known. The latter implies that knowing the voltage of a colored
dot inside a substation does not translate to knowing the voltage
of a differently colored dot inside the same substation. An ILP
formulation for only minimizing the number of circles (super-
nodes or substations) gives the optimal substation locations for
PMUs as 1, 2, and 10. However, from the figure it becomes clear
that the 500 kV bus in substation 7 is not observed by this solu-
tion. Hence, an OPP scheme whose goal is observability of all
buses and minimization of number of substation installations
cannot be formulated through super-node based observability
alone.
Another approach might be to imagine the power network
as being composed of a number of independent sub-networks
each of which is comprised of buses that are at a particular
voltage level. In that case, the example system depicted in Fig. 1
would consist of not one but three different networks—red, blue,
and green, respectively. A possible solution obtained from an
independent ILP formulation for each network is as follows:
PMUs need to be placed on the red bus of substation 2; blue
buses of substations 5, 7, and 11; and green buses of substations
1 and 2 for complete observability of the red, blue, and green
networks, respectively. From this solution it becomes clear that
if the sub-networks are treated as completely independent of
one another, PMUs would need to be placed in five substations.
Considering the objectives outlined in this paper, an optimal
solution for the system shown in Fig. 1 will be one where PMUs
are placed in four substations, namely, 1, 2, 7 and 10. In terms of
number and locations of DULRs, a possible solution set could
be: Three DULRs at substation 1 monitoring edges 1-5, 1-6, and
1-11; three DULRs at substation 2 monitoring edges 2-1, 2-3,
and 2-4; two DULRs at substation 7 monitoring edges 7-1, and
7-8; and two DULRs at substation 10 monitoring edge 10-5, and
10-9. A formulation which will give such a solution for practical
power system networks is described in the next section.
III. MATHEMATICAL FORMULATION
AND PRACTICAL CONSIDERATIONS
From Fig. 1 it is clear that maximizing bus observ-
ability and minimizing substation disruptions are conflicting
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 32, NO. 1, JANUARY 2017
objectives. The proposed methodology reconciles the two ob-
jectives through a cost-based formulation. It places DULRs on
edges for observing all the buses while minimizing the total
cost, which is the sum of the cost of disrupting substations and
the cost of all DULRs placed.
A. Nomenclature Used
Let the power network be represented by an undirected graph
G (V, E) where V is the set of nodes (buses) and E is the set of
edges (transmission lines or transformers). Further, the node set
V is partitioned into k ≥ 2 blocks B1 , B2 , . . . , Bk , where each
block represents a substation. Let ri denote |Bi | for 1 ≤ i ≤ k.
Since the tap ratios of the transformers are not known, there
is no edge between any pair of nodes inside the same block.
Thus, the nodes in each block form an independent set, thereby
implying that G is a k–partite graph.
Now, each node of V will be denoted by a pair of integers
(x, y) where x ∈ {1, 2, . . . , k} is the block number and y ∈
{1, 2, . . . , rx } is an index number within block Bx . Next, a
lexicographic ordering amongst the nodes is introduced. That
is, given two nodes v1 = (x1 , y1 ) and v2 = (x2 , y2 ) from two
different blocks Bx 1 and Bx 2, we define v1 ≺ v2 if x1 < x2 .
Each edge e = {v1 , v2 } ∈ E joins two nodes v1 = (x1 , y1 ) and
v2 = (x2 , y2 ) from two different blocks. When specifying an
edge e as {v1 , v2 }, it will be assumed that v1 ≺ v2 ; thus, v1 and
v2 can be referred to as the low end and the high end of edge e,
respectively.
For any node v ∈ V, the neighborhood of v, denoted by Nv ,
contains the node v itself and all nodes that are adjacent to v. As
a simple extension of this definition, the neighborhood of a set
of nodes X, denoted by NX , is defined by NX = ∪v∈X Nv . Also,
a DULR must be placed at either end of an edge. When a DULR
is placed on an edge e = {v1 , v2 }, it can observe both the nodes
v1 and v2 . If it is placed on the v1 = (x1 , y1 ) end, which is the
low end of e, then the block x1 must be disrupted. Likewise, if
it is placed on the v2 = (x2 , y2 ) end, which is the high end of
e, then the block x2 must be disrupted. Finally, let ci be the cost
of disrupting block Bi for 1 ≤ i ≤ k and let Δ denote the cost
of each DULR. A formulation based on the above terminology
is described below.
B. Basic DULR Placement Problem (DULRPP)
The basic version of DULRPP assumes that the given sys-
tem does not have any traditional bus PMU or DULR already
installed. Its goal is to place DULRs on some of the edges so
that the following conditions are satisfied: (a) all the nodes are
observed; and (b) the total cost, which is the sum of the cost of
disrupting substations and the cost of the DULRs added, is min-
imized. An ILP-based formulation of this problem is developed
as follows.
For each substation Bi for 1 ≤ i ≤ k, there is a {0, 1} valued
variable yi such that

1, if substation Bi is disrupted
yi = (1)
0, otherwise.
PAL et al.: PMU PLACEMENT SCHEME CONSIDERING REALISTIC COSTS AND MODERN TRENDS IN RELAYING
For each edge e, there are two {0, 1} valued variables wel and
wehsuch that
 is to ensure that each bus v ∈ C is observed by at least one
1, if DULR is placed at the low end of edge e
wel = (2)
0, otherwise
 outage of the line on which the DULR is placed, this con-
1, if DULR is placed at the high end of edge e
weh = (3)
0, otherwise.
Using (1)–(3), the basic objective function can be defined as
 k  by at least t + 1 DULRs. This is done by replacing the
   constraint specified by (5) with the one given in
h l
Minimize ci yi + Δ we + we . (4)
i=1 e∈E
Three sets of constraints that must be imposed on the variables
used in this basic objective function are defined as follows. For
any node v, let Lv denote the set of edges incident on v such
that for each edge in Lv , v is the low end of that edge. Similarly,
for any node v, let Hv denote the set of edges incident on v
such that for each edge in Hv , v is the high end of that edge.
Let Ev = Lv ∪ Hv denote the set of all edges incident on node
v. Then, for each node v ∈ V
 
weh + wel ≥ 1. (5)
e∈E v
Eq. (5) ensures that each node v is observed by placing a
DULR on at least one of the edges incident on v.
For each node v = (i, j) and each edge e ∈ Lv
yi ≥ wel . (6)
Eq. (6) ensures that if a DULR is placed on an edge for which
v = (i, j) is the low end, then Bi must be disrupted.
Similarly, for each node v = (i, j) and each edge e ∈ Hv
yi ≥ weh . (7)
Eq. (7) ensures that if a DULR is placed on an edge for
which v = (i, j) is the high end, then Bi must be disrupted. This
completes the formulation of the basic DULRPP. Similar to the
traditional OPP problem, DULRPP is also NP-complete and this
is proved in the Appendix.
C. Additional Constraints
Practical constraints that can be further imposed on the pro-
posed objective function are as follows:
1) Redundancy to critical buses: Measurement redundancy
under N − 1 contingency has been a common theme
in many papers on PMU placement ([12], [15], [20],
[23], etc.). However, a closer inspection of the results
reveals that a large number of substations (>50%) must
be disrupted in order to provide redundancy to all pha-
sor measurements. A more practical scheme is to provide
redundancy in measurements for only the most important
buses of the system. A methodology to identify such crit-
ical buses has already been proposed in [29]. To provide
redundancy in measurement for only the critical buses of
the network, the proposed formulation is modified in the
following way. When a DULR on an edge e fails, it cannot
observe any end point of e. Consider a given set C ⊆ V
555
of critical buses and an integer t ≥ 1 that represents the
maximum number of DULRs that may fail. Then, the goal
DULR even when any subset of t or fewer DULRs fail.
Since the failures can be due to the device itself and/or
straint ensures observability under N − t contingency. In
the proposed ILP formulation, this requirement is accom-
modated by ensuring that each critical bus is observed
 
weh + wel ≥ t + 1 ∀v ∈ C. (8)
e∈E v
2) Handling of prohibited substations: In the field, there ex-
ist some substations where installations cannot be made in
the planning horizon established for synchrophasor place-
ment. This is a practical constraint and is addressed as fol-
lows: Suppose S ⊆ {B1 , B2 , . . . , Bk } is the set of substa-
tions that cannot be disrupted. Then, for each substation
Bi ∈ S, the constraint yi = 0 must be added to the ILP
formulation described in (4)–(7).
3) Handling of existing PMUs: Similar to Item 2 above, there
might be some substations where PMUs may already be
present. Such existing PMU locations can be integrated
into the proposed formulation in the following manner. Let
P ⊆ V be the set of nodes at which PMUs have already
been placed. Also, let the PMU placed at the ith node
of P observe the nodes in Mi where Mi ⊆ Ni . Then, the
placement of DULRs needs to observe only the nodes in
V − MP , where

MP = Mi . (9)
1≤i≤|P|
In (9), it is assumed that the union operator eliminates du-
plicate entries. Therefore, to account for P, the constraint
specified by (5) needs to only be applied to each node
v ∈ V − MP .
4) Handling unknown transformer tap ratios: The tap-
settings of transformers are not known in real-life. If not
considered in the PMU placement formulation, the un-
known tap ratios may reduce the accuracy of the resulting
state estimates [30], [31]. The proposed formulation ac-
counts for unknown tap ratios in the following manner:
a) At the start of the optimization, any edge that joins
two nodes (buses) inside a block (substation) is elim-
inated.
b) If the elimination of edges causes a node to be dis-
connected from the rest of the system, then that
disconnected node is placed inside a dummy block
and is connected to the same node(s) of the real
block by dummy line(s). If the number of dummy
lines added to the node v of the real block is d,
where d ≥ 1, then (5) for node v becomes,
 
weh + wel ≥ d + 1. (10)
e∈E v
556
Fig. 2. Handling ZI buses in absence/presence of unknown tap ratios.
c) DULR-based observability of the multi-partite
i = Ni − Ni,j , that is, N
i is the set of elements that are in Ni
graph is computed next. When a DULR is placed
j = Nj − Ni,j . Construct Ri,j
at the end of a line joining a dummy block to a real
block, the cost of the DULR and the cost of dis-
i , add the 2-element
rupting the real block are retained but the cost of
disrupting the dummy block is ignored.
j , add the 2-element
The solution obtained by following this methodology ensures
that buses located at either end of a transformer are observed by
different DULRs. Therefore, even if the transformer tap ratios
are not known, voltages of all the buses of the network can be
i × N
j × Ni,j .
accurately estimated.
5) Presence of zero-injection (ZI) buses: Practical power sys-
i , q ∈ N
j andr ∈ Ni,j . For each triple
tem networks have ZI buses. These are buses through
which power neither enters nor leaves the system. Many
papers on OPP have exploited the presence of ZI buses
for reducing the total number of PMUs required ([12],
[15], [32], etc.). However, with unknown tap ratios, treat-
ing all ZI buses in the same way can lead to scenarios
where topological observability does not guarantee nu-
merical observability [33]. An example of such a scenario
is shown in Fig. 2.
In Fig. 2, solid circles denote buses, dotted circles/ovals de-
note substations, numbers refer to bus numbers, letters A to G
refer to substations, arrows depict injections, and solid rectan-
gles depict DULRs. From the figure it becomes clear that buses
2 and 6 are ZI buses while the other buses have non-zero injec-
tions. Buses 6 and 8 inside substation F are at different voltage
levels (indicated by the colors red and blue) and have a trans-
former between them whose tap ratio is unknown. In Fig. 2(a),
the voltage of bus 4 can be estimated using the DULRs located
on edge 1-2 and 3-2. In Fig. 2(b), due to the presence of a trans-
former between buses 6 and 8, the DULRs placed at 5-6 and 7-6
will not be able to estimate the voltage of bus 8. Since a purely
topological approach does not differentiate between transform-
ers and transmission lines, it cannot account for this failure in
estimation. The proposed approach addresses this concern in
the following manner. Let TZI denote the set of all ZI buses
present in the system. To construct observability constraints, it
is necessary to consider a subset of TZI, denoted by PZI. A
formal definition of PZI is as follows. PZI = {x : x ∈ TZI and
for every y ∈ Nx , x and y are at the same voltage level}. For
convenience, we refer to the condition specified in the above def-
IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 32, NO. 1, JANUARY 2017
inition of PZI as the iso–voltage zero injection bus(IvZIB)
condition. Thus, all elements of PZI must satisfy the IvZIB
condition. Once PZI is formed, the methodology proposed in
[34] can be extended to the DULRPP for reformulating (5) for
buses belonging to PZI as shown below.
Let |PZI| = k. Depending on the value of k, we explain
below how to construct a set R of objects from which all the
observability constraints can be generated.
For k = 1: Let i be the single element of PZI and Ni be the
neighborhood of i. For each pair of elements p and q in Ni , R
contains the 2-element set {p, q}.
For k ≥ 2: Let i and j be two elements of PZI. Define Ni and
Nj as the neighborhoods of i and j, respectively. Note that Ni
and Nj may have elements in common. Let Ni,j = Ni ∩ Nj , that
is, Ni,j is the set of elements that occur in both Ni and Nj . Let
N
but not in Nj . Similarly, define N
as follows:
a) For each pair of elements p and q in N
set {p, q} to Ri,j .
b) For each pair of elements p and q in N
set {p, q} to Ri,j .
c) For each pair of elements p and q in Ni,j , add the 2-element
set {p, q} to Ri,j .
d) Construct the cross-product set Qi,j = N
It is to be noted that Qi,j contains all triples (p, q, r)
such thatp ∈ N
(p, q, r) ∈ Qi,j , add the 3-element set {p, q, r} to Ri,j .
For the elements i and j of PZI, every 2-element or 3-element
set in Ri,j leads to an observability constraint. Now, consider
each pair of elements i and j in PZI and generate the set Ri,j
for that pair using steps a–d given above. The collection R
from which observability constraints can be generated for all
the k ≥ 2 elements of PZI is given by

R= Ri,j . (11)
1≤i≤j≤k
In (11), it is assumed that the union operator eliminates du-
plicate entries. Finally, for every 2-element set {p, q} and 3-
element set {p, q, r} in R, the modified observability constraints
are given in
   h 
weh + wel + we + wel ≥ 1 (12a)
e ∈E p e ∈E q
   h   h 
weh + wel + we + wel + we + wel ≥ 1.
e ∈E p e ∈E q e ∈E r
(12b)
6) Absence of measurement channel limitation: As each
DULR only monitors one edge, the problem of measure-
ment channel limitation does not arise.
IV. SIMULATION AND RESULTS
A. Realistic Costs
One of the salient points of this paper is that realistic costs
have been considered in specifying the cost objective. Federal
PAL et al.: PMU PLACEMENT SCHEME CONSIDERING REALISTIC COSTS AND MODERN TRENDS IN RELAYING
Fig. 3. The substation synchrophasor architecture design that was the basis of
the proposed DULRPP formulation.
agencies such as DOE have collected data on how their invest-
ment has been used by different utilities in purchasing equip-
ment and upgrading infrastructure. Reports published by these
agencies [6], [7] highlight the following facts:
1) The median cost for installing a single PMU at a new
substation was approximately USD43 400. This cost was
comprised of the cost of the device itself as well as de-
sign and engineering costs, labor and material costs for
installation/construction, and overhead [6].
2) There were no economies of scale; that is, projects in-
stalling a greater number of PMUs did not have lower
average costs per device [6].
3) The key drivers of the cost of PMU installations in de-
creasing order of cost, were: communication, security
(mission-critical systems and mission-support systems),
labor (specialized crew or decentralized crews) and cost
of equipment (that is, PMU hardware cost) [7].
4) PMU hardware cost was less than 5% of the total cost for
the installed synchrophasor system. The cost of a phasor
data concentrator (PDC) was found to be higher than that
of a PMU but much less than the cost of the other drivers
mentioned in Item 3 above [7].
5) Once the other infrastructures were in place, the cost
of adding more PMUs to a substation was miniscule.
Similarly, the cost of ancillary equipment such as global
positioning system (GPS) clocks, network switches and
routers, etc., were found to be low relative to the cost of
upgrading substations [7].
Taking these facts into consideration, for the study done here,
the following costs were associated with the different compo-
nents of a synchrophasor installation:
1) Cost of disrupting a substation = USD 40 000.
2) Cost of a single DULR device = USD 2000.
3) Cost of a single PDC device = USD 8000 [35], [36].
Fig. 3 shows the typical design of a substation synchropha-
sor architecture that is the basis of the proposed study. TPMUs
(if present) are assumed to already be in place (shown by dot-
ted black lines). The proposed formulation is designed to add
more DULRs to the substations to provide the desired number
of phasor measurements required for complete observability. A
stand-alone PDC is located at every substation where a PMU is
placed. Some researchers have suggested optimizing the loca-
tions of PDCs to reduce the cost of communication infrastructure
[37]–[39]. However, the stand-alone PDC serves two purposes:
(a) the synchrophasor system becomes independent of (equip-
557
Fig. 4. DULRs placed in IEEE 30 bus system to satisfy practical constraints.
ment) failures in the SCADA system, and (b) the PDC inside the
substation provides a back-up archive of high-resolution PMU
data in case data loss happens in downstream applications.
B. IEEE 30 Bus System
The IEEE 30 bus system was modified by grouping buses into
substations. This was done using the rule that buses which were
connected by transformers belonged to the same substation. The
remaining buses which were not connected to any transformer
formed their own independent substations. As such, the IEEE
30 bus system became a 26-substation system as seen in Fig. 4.
In Fig. 4, the black circle/ovals denote substations where buses
of different voltage levels are present, while the colors red, blue,
and green denote the respective voltage networks.
The PMU placement results for basic observability of this
system are shown in Table I. For creating this table, ZI buses
were treated as normal buses. When specified numerically, en-
tries i − j in the third column imply that the PMUs monitor i
end of line i − j, where i and j are bus numbers. The results for
[28] and the proposed approach were obtained using DULRs.
The number of DULRs required was equal to the number of
lines monitored. The results for [12], [37], [40]–[43] were ob-
tained using TPMUs. When a TPMU was placed on a bus and
the lines monitored by it were not mentioned, then all the lines
that emerged from that bus were assumed to be observed by that
TPMU. That is, unless stated explicitly, measurement channel
limitation of TPMUs was not considered. The cost of monitoring
a line using TPMU was set equal to the cost of a DULR device.
Also, PDCs were assumed to be placed at those substations
where PMUs were placed.
558 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 32, NO. 1, JANUARY 2017

TABLE I
COST COMPARISON OF PROPOSED TECHNIQUE WITH OTHER APPROACHES FOR THE IEEE 30 BUS SYSTEM

∗
Refs [42] and [43] had two optimal solutions for the IEEE 30 bus system.

TABLE II
COST COMPARISON OF PROPOSED TECHNIQUE WITH OTHER APPROACHES FOR THE THREE LARGE TEST SYSTEMS

Test System Method #TPMU #Lines Monitored #Substations Disrupted System Observable if Tap Ratios are Unknown Total Cost (in USD)

IEEE 118 Bus Ref. [12] 32 132 32 No 1.800 M

Ref. [40] 32 132 31 No 1.752 M
Ref. [41] 32 119 32 No 1.774 M
Ref. [42]∗ 32 126 32 No 1.788 M
32 125 31 No 1.738 M
Ref. [43]∗ 44 74 41 No 2.116 M
38 80 35 No 1.840 M
Proposed – 84 31 Yes 1.656 M
IEEE 300 Bus Ref. [41] 87 337 76 No 4.322 M
Ref. [48] 95 382 80 No 4.604 M
Proposed – 200 75 Yes 4.000 M
Polish 2383 Bus Ref. [41] 746 2503 717 No 39.422 M
Ref. [48] 799 2802 753 No 41.748 M
Proposed – 1572 704 Yes 36.936 M

∗
Refs [42] and [43] had two different optimal solutions for the IEEE 118 bus system.

In the next set of simulations, the following constraints are TABLE III
COST COMPARISON OF PROPOSED TECHNIQUE WITH [29] WHILE CONSIDERING
added to the optimization: (a) Substation 21 is unsuitable for HIGH-VOLTAGE NETWORKS OF THE THREE TEST SYSTEMS AS CRITICAL
synchrophasor deployment; (b) Buses 8 and 13 are critical for
system stability purposes and need N − 1 redundancy; and (c)
Test #Lines #Substations Total Cost
Bus 30 has a TPMU on it which monitors its voltage as well as Syst. Method #TPMU Monitored Disrupted (in USD)
the currents that flow in lines 30–27 and 30–29. Section III-C
IEEE 118 Bus Ref. [29] 39 135 36 1.998 M
explains how the proposed methodology handles these practical Proposed – 88 34 1.808 M
constraints. The results are shown in Fig. 4, where the thick IEEE 300 Bus Ref. [29] 96 356 81 4.600 M
black lines indicate buses observed by the TPMU at bus 30. The Proposed – 206 77 4.108 M
Polish 2383 Bus Ref. [29] 782 2601 737 40.494 M
abbreviations CRIT BUS and PROH SUB denote critical bus Proposed – 1599 712 37.374 M
and prohibited substation, respectively. The colored blue dots
indicate locations where DULRs must be added. From the figure
it becomes clear that all buses are observed by the combination
of the TPMU at bus 30 and the 19 DULRs that were added in 8 The total cost for this DULR placement scheme comes out to be
substations (S2, S4, S6, S8, S10, S16, S20, and S22). Moreover, USD 422 000 = ((40 000 × 8) + (2000 × 19) + (8000 × 8)).
no DULR is placed at the prohibited substation (S21) while the This example highlights the ability of the proposed approach to
critical buses (8 and 13) are observed by independent DULRs. simultaneously handle different practical constraints.
PAL et al.: PMU PLACEMENT SCHEME CONSIDERING REALISTIC COSTS AND MODERN TRENDS IN RELAYING 559

TABLE IV
EFFECT OF ZI BUSES ON RESULTS OBTAINED USING PROPOSED APPROACH∗

Test #Lines #Substations Total Cost CPU

System Scenario #TZI #PZI Monitored Disrupted (in USD) Time (s)

IEEE 118 Bus Basic Observability – – 84 31 1.656 M 0.089763

Basic Observability considering HV buses as Critical – – 88 34 1.808 M 0.066501
Basic Observability considering ZI Buses 10 2 82 30 1.604 M 0.187890
Basic Observability considering ZI Buses and HV buses as Critical 10 2 88 33 1.760 M 0.062640
IEEE 300 Bus Basic Observability – – 200 75 4.000 M 0.038525
Basic Observability considering HV buses as Critical – – 206 77 4.108 M 0.041276
Basic Observability considering ZI Buses 67 20 193 68 3.650 M 0.051294
Basic Observability considering ZI Buses and HV buses as Critical 67 20 199 71 3.806 M 0.041554
Polish 2383 Bus Basic Observability – – 1572 704 36.936 M 1.314919
Basic Observability considering HV buses as Critical – – 1599 712 37.374 M 1.179524
Basic Observability considering ZI Buses 553 376 1350 567 29.916 M 6.892090
Basic Observability considering ZI Buses and HV buses as Critical 553 376 1380 577 30.456 M 6.787720

∗
For the three large test systems, information about the substations which were disrupted and the lines on which the DULRs were placed can be found in [47].

C. Application to Large Systems
The proposed approach was next applied to IEEE 118 bus,
IEEE 300 bus, and Polish 2383 bus systems. The data for the test
systems was obtained from MATPOWER [44] toolbox of MAT-
LAB [45]. The ILP optimization was performed using GUROBI
[46]. The computations were performed on an Intel (R) Core i7
Processor having a speed of 2.60 GHz and an installed memory
(RAM) of 16 GB. By combining buses into substations (based
on the locations of the transformers), the three test systems be-
came a 107-substation, a 184-substation, and a 2215-substation
system, respectively [47]. The results obtained by comparing
the proposed approach with other techniques is shown in Ta-
ble II. The comparison was made while treating ZI buses as
normal buses. Also, unless specified otherwise, all the lines that
emerged out of the bus where the TPMU was placed were as-
sumed to be observed by it. Finally, the number of PDCs were
assumed to be equal to the number of disrupted substations.
From the table it becomes clear that minimizing the number of
buses where PMUs must be placed does not result in a signif-
icant decrease in cost of synchrophasor deployment. However,
by reducing the number of substations where installations must
be made, the overall cost can be substantially reduced. Further-
more, among the compared techniques, the proposed approach
is the only one that ensures system observability when the tap
ratios are not known. Therefore, given that no single technique
can be universally superior to all others, the proposed approach
does provide unique techno-economic benefits.
In the next set of simulations, the highest voltage networks
of the three test systems were assumed to be critical, requiring
N − 1 redundancy. This is a valid assumption because utilities
tend to initially place PMUs on their highest voltage level buses
[49], [50]. The number of high-voltage buses in the three test
systems are 11, 14 and 50, respectively [44], [45]. Table III
compares the results obtained using the proposed method and
the technique developed in [29]. The number of PDCs required
were assumed to be equal to the number of substations disrupted.
From Table III it becomes clear that the proposed method is able
to achieve the desired objective for all the three test systems at
a lower total cost. A comparison of Tables II and III shows that
the proposed approach is able to guarantee real-time protection
of the critical portions of the system with only a slight increase
in total investment.
In the simulations done so far, all buses were assumed to have
non-zero injections. However, ZI buses do exist in real systems.
In order to account for their presence, those ZI buses of the
three large test systems which satisfied the IvZIB condition
described in Section III-C point 5 were treated as ZI buses, and
the optimizations described in Tables II and III were repeated.
The results obtained are shown in Table IV. From the table it is
realized that when properly handled, ZI buses can significantly
reduce the cost of synchrophasor deployment without compro-
mising the accuracy of state observability. Table IV also gives
the CPU time required for performing the optimizations based
on the proposed approach.
V. CONCLUSION
Now that PMUs are being added in bulk in power networks,
new formulations of the OPP problem must be considered. Due
to the emergence of DULRs, the device cost is no longer the
primary impediment to synchrophasor deployment. Moreover,
it is neither necessary nor practical to place these devices at all
available locations. Keeping this in mind, this paper proposes
an ILP-based DULR placement scheme that minimizes overall
cost while ensuring adequate redundancy under device failures
and/or line outages to the most critical assets. The novelty of this
paper lies in using a realistic cost objective that considers both
the cost of disrupting substations and the cost of DULRs. Thus,
communication infrastructure cost, (cyber)-security infrastruc-
ture cost, labor cost, as well as device cost are simultaneously
optimized. At the same time, practical constraints imposed on
the basic objective (of observability) ensure that the results ob-
tained have real-world applications. Therefore, the proposed
scheme provides an optimal techno-economic balance to the
OPP problem.
APPENDIX
Let Γ denote the total cost budget for DULRPP. Thus, the
goal of DULRPP is to place DULRs for complete observability
while ensuring the total cost is ≤ Γ. Before we establish the
560 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 32, NO. 1, JANUARY 2017

complexity of DULRPP, we provide a necessary graph theoretic Laboratory at Virginia Tech have contributed greatly towards
definition. Given an undirected graph H (VH , EH ), a dominating this research.
set of H is a subset V
⊆ VH of nodes such that every node
v ∈ VH − V
is adjacent to some node w ∈ V
; that is, the edge
{v, w} is in EH . In such a case, we say that v is dominated by w. REFERENCES
A dominating set of minimum cardinality is called a minimum [1] A. G. Phadke, J. S. Thorp, and K. J. Karimi, “State estimation with phasor
dominating set (MDS). It is well known that any dominating set measurements,” IEEE Trans. Power Syst., vol. 1, no. 1, pp. 233–238,
Feb. 1986.
V
of a connected undirected graph H can be modified without [2] M. Gol and A. Abur, “A hybrid state estimator for systems with limited
increasing |V
| so that each node w ∈ V
has an edge to at least number of PMUs,” IEEE Trans. Power Syst., vol. 30, no. 3, pp. 1511–1517,
one node v ∈ VH − V
. May 2015.
[3] Y. Chakhchoukh, V. Vittal, and G. T. Heydt, “PMU based state estima-
Theorem 1: DULRPP is NP-complete. tion by integrating correlation,” IEEE Trans. Power Syst., vol. 29, no. 2,
Proof: It is easy to see that DULRPP is in NP, since one pp. 617–626, Mar. 2014.
can guess a placement of DULRs and verify efficiently that it [4] M. Gol and A. Abur, “A robust PMU based three-phase state estima-
tor using modal decoupling,” IEEE Trans. Power Syst., vol. 29, no. 5,
provides full observability and that its cost is at most Γ. pp. 2292–2299, Sep. 2014.
We establish the NP-hardness of DULRPP through a reduc- [5] K. D. Jones, J. S. Thorp, and R. M. Gardner, “Three-phase linear state
tion from the MDS problem defined as follows: given a con- estimation using phasor measurements,” in Proc. IEEE Power Eng. Soc.
General Meeting, Vancouver, BC, Canada, Jul. 21–25, 2013, pp. 1–5.
nected undirected graph H (VH , EH ) and an integer Q ≤ |VH |, [6] U.S. Department of Energy, Electricity Delivery and Reliability, “Syn-
does H have a dominating set of size at most Q? MDS is known chrophasor technologies and their deployment in the recovery act smart
to be NP-complete [51]. grid programs,” Aug. 2013. [Online]. Available: http://energy.gov/sites/
prod/files/2013/08/f2/SynchrophasorRptAug2013.pdf
Given an instance of the MDS problem consisting of the graph [7] U.S. Department of Energy, Office of Electricity Delivery and Energy
H (VH , EH ) and integer Q, we construct an instance of DULRPP Reliability, “Factors affecting PMU installation costs,” Oct. 2014. [On-
as follows. Let |VH | = n and |EH | = m. The graph G (V, E) line]. Available: https://www.smartgrid.gov/files/PMU-cost-study-final-
10162014_1.pdf
of the DULRPP instance is identical to H; that is V = VH and [8] N. M. Manousakis, G. N. Korres, and P. S. Georgilakis, “Taxonomy of
E = EH . Further, each node of V is a block by itself; that is, PMU placement methodologies,” IEEE Trans. Power Syst., vol. 27, no. 2,
there are n = |V| blocks. The cost ci of disrupting any block is pp. 1070–1077, May 2012.
[9] F. Aminifar, M. Fotuhi-Firuzabad, A. Safdarian, A. Davoudi, and
chosen as m + 1 where 1 ≤ i ≤ n. The cost of each DULR is M. Shahidehpour, “Synchrophasor measurement technology in power sys-
chosen as 1. The cost budget Γ is chosen as Q (m + 1) + m. tems: Panorama and state-of-the-art,” IEEE Access, vol. 2, pp. 1607–1628,
This completes the construction. We now prove that there is a Jan. 2015.
[10] V. Kekatos, G. B. Giannakis, and B. Wollenberg, “Optimal placement
solution to DULRPP if and only if there is a solution to the MDS of phasor measurement units via convex relaxation,” IEEE Trans. Power
problem. Syst., vol. 27, no. 3, pp. 1521–1530, Aug. 2012.
Suppose H has a dominating set V
with |V
| ≤ Q. As stated [11] O. Gomez, M. A. Rios, and G. Anders, “Reliability-based phasor mea-
surement unit placement in power systems considering transmission
earlier, we may assume that each node of V
has an edge to at line outages and channel limits,” IET Gener., Transm., Distrib., vol. 8,
least one node of V − V
. We choose the nodes of V
as the no. 1, pp. 121–130, Jan. 2014.
substations to be disrupted. We place a DULR on each edge [12] L. Huang, Y. Sun, J. Xu, W. Gao, J. Zhang, and Z. Wu, “Optimal PMU
placement considering controlled islanding of power system,” IEEE Trans.
{w, v} that joins a node w ∈ V
to a node v ∈ V − V
. This Power Syst., vol. 29, no. 2, pp. 742–755, Mar. 2014.
DULR is placed at the end corresponding to w (since w ∈ V
is [13] B. Gou and R. G. Kavasseri, “Unified PMU placement for observability
a substation that will be disrupted). Using the facts (a) V
is a and bad data detection in state estimation,” IEEE Trans. Power Syst.,
vol. 29, no. 6, pp. 2573–2580, Nov. 2014.
dominating set, (b) each node w ∈ V
has an edge to at least one [14] Y. Wang, C. Wang, W. Li, J. Li, and F. Lin, “Reliability-based incremental
node in V − V
and, (c) number of DULRs placed is at most m PMU placement,” IEEE Trans. Power Syst., vol. 29, no. 6, pp. 2744–2752,
(the number of edges in G); it is easy to verify that this DULR Nov. 2014.
[15] J. Aghaei, A. Baharvandi, A. Rabiee, and M. A. Akbari, “Probabilistic
placement observes all the nodes and the total cost is at most Γ. PMU placement in electric power networks: An MILP-based multiobjec-
In other words, there is a solution to DULRPP. tive model,” IEEE Trans. Ind. Informat., vol. 11, no. 2, pp. 332–341, Apr.
For the converse, suppose there is a solution to DULRPP. In 2015.
[16] R. Sodhi and M. I. Sharieff, “Phasor measurement unit placement frame-
this solution, if Q + 1 or more substations are disrupted, then work for enhanced wide-area situational awareness,” IET Gener., Transm.,
the total cost will be at least (Q + 1) (m + 1), which exceeds Distrib., vol. 9, no. 2, pp. 172–182, Jan. 2015.
the budget Γ = Q (m + 1). Hence, the number of substations [17] P. Yang, Z. Tan, A. Wiesel, and A. Nehorai, “Placement of PMUs con-
sidering measurement phase-angle mismatch,” IEEE Trans. Power Del.,
disrupted is at most Q. Let V
be set of substations that are vol. 30, no. 2, pp. 914–922, Apr. 2015.
disrupted. Using the fact that we have a valid DULR placement, [18] J. Qi, K. Sun, and W. Kang, “Optimal PMU placement for power system
it is easy to verify that V
forms a dominating set for G. Further, dynamic state estimation by using empirical observability Gramian,” IEEE
Trans. Power Syst., vol. 30, no. 4, pp. 2041–2054, Jul. 2015.
since |V
| ≤ Q, we have a solution to MDS, and this completes [19] N. M. Manousakis and G. N. Korres, “A weighted least squares algorithm
the proof. for optimal PMU placement,” IEEE Trans. Power Syst., vol. 28, no. 3, pp.
3499–3500, Aug. 2013.
[20] N. P. Theodorakatos, N. M. Manousakis, and G. N. Korres, “A sequential
quadratic programming method for contingency constrained phasor mea-
ACKNOWLEDGMENT surement unit placement,” Int. Trans. Electr. Energy Syst., vol. 25, no. 12,
The authors would like to thank especially Prof. J. S. Thorp pp. 3185–3211, Dec. 2015.
[21] A. E. Leon and J. A. Solsona, “Power oscillation damping improvement
for his valuable inputs. Many faculty and students of the Power by adding multiple wind farms to wide-area coordinating controls,” IEEE
Laboratory and the Network Dynamics and Simulation Science Trans. Power Syst., vol. 29, no. 3, pp. 1356–1364, May 2014.
PAL et al.: PMU PLACEMENT SCHEME CONSIDERING REALISTIC COSTS AND MODERN TRENDS IN RELAYING
[22] H. Morais, P. Vancraeyveld, A. H. B. Pedersen, M. Lind, H. Johannsson,
and J. Ostergaard, “SOSPO-SP: Secure operation of sustainable power
systems simulation platform for real-time system state evaluation and
control,” IEEE Trans. Ind. Informat., vol. 10, no. 4, pp. 2318–2329, Nov.
2014.
[23] Z. H. Rather, Z. Chen, P. Thogerson, P. Lund, and B. Kirby, “Realistic
approach for phasor measurement unit placement: consideration of prac-
tical hidden costs,” IEEE Trans. Power Del., vol. 30, no. 1, pp. 3–15, Feb.
2015.
[24] W. Bao, R. Guo, Z. Han, L. Chen, and M. Lu, “A substation oriented
approach to optimal phasor measurement units placement,” J. Electr. Eng.
Technol,, vol. 9, pp. 742–753, Sep. 2014.
[25] C. Mishra, K. D. Jones, A. Pal, and V. A. Centeno, “A binary PSO-based
optimal substation coverage algorithm for PMU installations in practical
systems,” IET Gener., Transm., Distrib., vol. 10, no. 2, pp. 555–562, Feb.
2016.
[26] A. Pal, G. A. Sanchez-Ayala, J. S. Thorp, and V. A. Centeno, “A
community-based partitioning approach for PMU placement in large sys-
tems,” Electr. Power Compon. Syst., to be published.
[27] O. Gomez, C. Portilla, and M. A. Rios, “Reliability analysis of substation
monitoring systems based on branch PMUs,” IEEE Trans. Power Syst.,
vol. 30, no. 2, pp. 962–969, Mar. 2015.
[28] R. Emami and A. Abur, “Robust measurement design by placing synchro-
nized phasor measurements on network branches,” IEEE Trans. Power
Syst., vol. 25, no. 1, pp. 38–43, Feb. 2010.
[29] A. Pal, G. A. Sanchez, V. A. Centeno, and J. S. Thorp, “A PMU placement
scheme ensuring real-time monitoring of critical buses of the network,”
IEEE Trans. Power Del., vol. 29, no. 2, pp. 510–517, Apr. 2014.
[30] G. N. Korres, P. J. Katsikas, and G. C. Contaxis, “Transformer tap setting
observability in state estimation,” IEEE Trans. Power Syst., vol. 19, no. 2,
pp. 699–706, May 2004.
[31] M. Shiroie and S. H. Hosseini, “Observability and estimation of trans-
former tap setting with minimal PMU placement,” in Proc. IEEE Power
Eng. Soc. General Meeting-Conv. Del. Electr. Energy 21st Century, Pitts-
burgh, PA, USA, Jul. 20–24, 2008, pp. 1–4.
[32] G. N. Korres, N. M. Manousakis, T. C. Xygkis, and J. Löfberg, “Op-
timal phasor measurement unit placement for numerical observability
in the presence of conventional measurements using semidefinite pro-
gramming,” IET Gener., Transm., Distrib., vol. 9, no. 15, pp. 2427–2436,
Nov. 2015.
[33] N. C. Koutsoukis, N. M. Manousakis, P. S. Georgilakis, and G. N. Korres,
“Numerical observability method for optimal phasor measurement units
placement using recursive Tabu search method,” IET Gener., Transm.,
Distrib., vol. 7, no. 4, pp. 347–356, Apr. 2013.
[34] K. G. Khajeh, E. Bashar, A. M. Rad, and G. B. Gharehpetian, “Integrated
model considering effects of zero injection buses and conventional mea-
surements on optimal PMU placement,” IEEE Trans. Smart Grid, to be
published.
[35] G. E. Multilin P30 Phasor Data Concentrator. (2016). [Online] Available:
https://www.gedigitalenergy.com/multilin/catalog/P30.htm
[36] SEL-3373 Station Phasor Data Concentrator. (2016). [Online] Available:
https://www.selinc.com/SEL-3373/
[37] M. Shahraeini, M. S. Ghazizadeh, and M. H. Javidi, “Co-optimal place-
ment of measurement devices and their related communication infrastruc-
ture in wide area measurement systems,” IEEE Trans. Smart Grid, vol. 3,
no. 2, pp. 684–691, Jun. 2012.
[38] D. Georges, “Optimal PMU-based monitoring architecture design for
power systems,” Control Eng. Practice, vol. 30, pp. 150–159, Sep. 2014.
[39] M. B. Mohammadi, R. A. Hooshmand, and F. H. Fesharaki, “A new ap-
proach for optimal placement of PMUs and their required communication
infrastructure in order to minimize the cost of the WAMS,” IEEE Trans.
Smart Grid, vol. 7, no. 1, pp. 84–93, Jan. 2016.
[40] S. Chakrabarti, E. Kyriakides, and D. G. Eliades, “Placement of synchro-
nized measurements for power system observability,” IEEE Trans. Power
Del., vol. 24, no. 1, pp. 12–19, Jan. 2009.
[41] J. De La Ree, V. Centeno, J. S. Thorp, and A. G. Phadke, “Synchronized
phasor measurement applications in power systems,” IEEE Trans. Smart
Grid, vol. 1, no. 1, pp. 20–27, Jun. 2010.
[42] M. H. F. Wen, J. Xu, and V. O. K. Li, “Optimal multistage PMU place-
ment for wide-area monitoring,” IEEE Trans. Power Syst., vol. 28, no. 4,
pp. 4134–4143, Nov. 2013.
[43] V. Rokkam and R. Bhimasingu, “A novel approach for optimal PMU
placement considering channel limit,” in Proc. IEEE Int. Conf. Power
Syst. Technol., Chengdu, China, Oct. 20–22, 2014, pp. 1164–1171.
561
[44] R. D. Zimmerman, C. E. Murillo-Sánchez, and R. J. Thomas, “MAT-
POWER steady-state operations, planning and analysis tools for power
systems research and education,” IEEE Trans. Power Syst., vol. 26, no. 1,
pp. 12–19, Feb. 2011.
[45] MATLAB (Computer Software) Release 2015a, The MathWorks, Inc.,
Natick, MA, USA, Mar. 2015.
[46] GUROBI Optimization, Inc., “GUROBI Optimizer Reference Manual,”
Version 6.0.2, Houston, TX, USA, Feb. 2015. [Online]. Available:
http://www.gurobi.com
[47] Results for the Three Large Test Systems. Results_IEEE118_IEEE300_
Polish2383.pdf. (2016). [Online]. Available: http://staff.vbi.vt.edu/
anam86/Results_IEEE118_IEEE300_Polish2383.pdf
[48] N. Xie, F. Torelli, E. Bompard, and A. Vaccaro, “A graph theory based
methodology for optimal PMUs placement and multiarea power system
state estimation,” Electr. Power Syst. Res., vol. 119, pp. 25–33, Feb. 2015.
[49] K. D. Jones, A. Pal, and J. S. Thorp, “Methodology for performing syn-
chrophasor data conditioning and validation,” IEEE Trans. Power Syst.,
vol. 30, no. 3, pp. 1121–1130, May 2015.
[50] A. Pal, P. Chatterjee, J. S. Thorp, and V. A. Centeno, “On-line calibration
of voltage transformers using synchrophasor measurements,” IEEE Trans.
Power Del., vol. 31, no. 1, pp. 370–380, Feb. 2016.
[51] M. R. Garey, and D. S. Johnson, Computers and Intractability: A Guide
to the Theory of NP-Completeness. San Francisco, CA, USA: Freeman,
1979.
Anamitra Pal (S’10–M’14) received the B.E. de-
gree (summa cum laude) in electrical and electronics
engineering from the Birla Institute of Technology,
Mesra, India, in 2008, and the M.S. and Ph.D. degrees
in electrical engineering from Virginia Tech, Blacks-
burg, VA, USA, in 2012 and 2014, respectively. From
2008 to 2010, he was a Manager with Tata Steel, Ltd.,
Jamshedpur, India. He was also a summer intern with
Electric Power Group, LLC, Pasadena, CA, USA, in
2012 and 2013. He is currently an Applied Electrical
and Computer Scientist in the Network Dynamics and
Simulation Science Laboratory at the Biocomplexity Institute of Virginia Tech,
Blacksburg. His research interests include power system modeling, transient
and dynamic stability analysis, and wide area measurements-based protection,
monitoring, and control.
Anil Kumar S. Vullikanti (M’11) received the Bach-
elor of Technology degree in computer science and
engineering from the Indian Institute of Technology,
Kanpur, India, in 1993, and the Ph.D. degree in com-
puter science from the Indian Institute of Science,
Bangalore, India, in 1999.
He is an Associate Professor at the Department of
Computer Science and the Biocomplexity Institute of
Virginia Tech, Blacksburg, VA, USA. Prior to this, he
was a Technical Staff Member with the Los Alamos
National Laboratory, Los Alamos, NM, USA. His
research interests include approximation algorithms, mobile computing, com-
binatorial optimization, and simulation of large sociotechnical systems.
S. S. Ravi (M’85–SM’09) received the B.E. and M.E.
degrees from the Indian Institute of Science, Banga-
lore, India, in 1977 and 1979, respectively, and the
Ph.D. degree in computer science from the Univer-
sity of Pittsburgh, Pittsburgh, PA, USA, in August
1984. Since September 1984, he has been with the
Computer Science Department at the University at
Albany—State University of New York, Albany, NY,
USA, where he is currently a Distinguished Teaching
Professor. His areas of interest include design and
analysis of algorithms, social network analysis, data
mining, fault-tolerant computing, wireless networks, and discrete dynamical
systems.