This article has been accepted for publication in a future issue of this journal, but has not been

fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

Optimal PMU-Communication Link Placement for

Smart Grid Wide-Area Measurement Systems
Xingzheng Zhu, Miles H. F. Wen, Victor O. K. Li, Fellow, IEEE,
and Ka-Cheong Leung, Member, IEEE

Abstract—Due to the relatively high costs of phasor mea- a hierarchy of at least three main tiers: local area networks
surement units (PMUs), the optimal PMU placement problem, (LAN), field area networks (FAN), and wide area networks
which focuses on minimizing the number of PMUs for full (WAN) [13]. In order to fulfill the complicated transmission
system observability, has long been a popular research topic.
However, PMUs alone will not make a system observable. An requirements of different applications, a variety of commu-
appropriate communication network that transmits all PMU nication technologies has been adopted in the design of the
data to the phasor data concentrator plays a key role. This communication network in smart grid. Early communication
paper proposes the optimal PMU-communication link placement technologies, like power line carrier and microwave communi-
(OPLP) problem that investigates the placement of PMU and cation, have limitations in reliability, scalability and robustness
communication links (CLs) for full observability in a power
system. Besides the location of PMUs and CLs, to ensure the [14]. However, the adoption of optical fiber communication
reliable and timely transmission of PMU data, the communication in power systems renders the end-to-end data transmission
capacity needed on every CL is also captured by the OPLP latency low enough to meet the communication needs [13].
problem. We carry out numerical studies on the IEEE 30-bus, Thus, in this work, we investigate the use of optical fibers in
57-bus, 118-bus and 300-bus systems for the proposed models. smart grid.
The results show that, compared with the traditional optimal
PMU placement model, OPLP can reduce the total installation In many existing works, such as [15] and [16], commu-
cost significantly. nication network availability is considered as a constraint in
choosing the optimal places to install PMUs. In other words,
Index Terms—Optimal PMU placement, communication link,
bandwidth assignment. these proposals consider a bus as a candidate to install a
PMU only if it has a communication link (CL) connect-
ing to the PDC. The OPP problem assumes the existence
I. I NTRODUCTION of a communication network with relatively good coverage
Wide area measurement system (WAMS) involves the use of and sufficient bandwidth. If this assumption does not hold,
system-wide information and the communication of selected new CLs must be installed. Relatively few studies, such
local information to a remote data center [1]. Synchronized as [17]–[20], considered the placement of CLs jointly with
measurement technology is an important enabler of WAMS. the PMUs. The co-optimal placement problem of PMU and
By providing synchronized real-time information, including their communication infrastructure was formulated and solved
voltage and current phasors, the phasor measurement units by Imperialistic Competition Algorithm (ICA) and Genetic
(PMUs) are important parts of WAMS [2]. The optimal Algorithm (GA) in [17] and [20], respectively. In [18], Rather
PMU placement (OPP) problem, which investigates optimal et al. took the hidden cost into consideration, and, jointly
installation strategies that make the entire system observable solved the placement problem of PMU and ancillary facil-
with the minimum number of PMUs, has been extensively ities, including communication facilities by Particle Swarm
studied in the literature [3]–[10]. Optimization (PSO). The authors of [19] proposed a solution
However, PMUs alone will not make a power system to place a minimum number of PMUs and CLs in the power
observable. An appropriate communication infrastructure must system so that the steady-state availability of synchrophasor
also be established to support timely data transmission be- data at each bus meets a prescribed level. In [21], Pal et
tween PMUs and the phasor data concentrator (PDC). The al. presented an integer linear programming methodology for
topological design of the smart grid communication networks the PMU placement problem while considering the communi-
is becoming important. Not only do the applications in smart cation infrastructure upgrading cost. However, these studies
grid require communication to be real-time, reliable, scalable did not take the PMU data transmission requirements into
and manageable, but they also require communications to consideration. They all assumed that the network capacity is
be interoperable, secure, future-proof, and cost-effective [11]. sufficient. However, network capacity can be a very expensive
Although there are no industrial standards, the communication resource in large-scale power systems. Appropriate allocation
infrastructure in smart power system is envisioned as a col- of the transmission capacity of every CL is very important in
lection of interconnected networks that will be structured into the design of WAMS.
The main objective of this paper is to extend the traditional
Xingzheng Zhu, Miles H. F. Wen, Victor O. K. Li, and Ka-Cheong OPP problem to the optimal PMU-CL placement (OPLP)
Leung are with the Department of Electrical and Electronic Engineering,
the University of Hong Kong, HKSAR, China (e-mail: [email protected],hk; problem to investigate the optimal placement of PMUs and
[email protected]; [email protected]; [email protected]). CLs jointly. Both the effect of zero-injection buses (ZIBs)

1
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

and PMU measurement capacity limitation are considered in We consider an N -bus power system and denote N =
the OPLP problem. Furthermore, this paper investigates the {1, 2, . . . , N } as the set of buses. We use M to represent
assignment of link capacity of every CL to fulfill the commu- the connectivity matrix. Its (i,j) entry, mi,j , satisfies:
nication requirements. A linear CL cost model and the shortest 
path routing scheme are used. We fit the OPLP problem 
1 if bus i and bus j are connected
into the mixed integer quadratically constrained programming mi,j = 1 if i = j (1)


(MIQCP) framework, which can be solved efficiently using 0 otherwise
commercial numerical solvers. We also provide numerical
analysis for OPLP on the IEEE 30-bus, 57-bus, 118-bus, and and µ = [µ1 , µ2 , · · · , µN ] is defined as the PMU installation
300-bus systems. From the numerical results, we have the vector, where
following particular findings: {
1 if PMU is installed on bus i
• The optimal placement of CL relies on the installation µi = (2)
0 otherwise
of PMUs and the selection of data routing paths. The
different choices on transmission paths for PMU data can In this work, we assume that there is one PDC in the WAMS
affect the CL cost. Specifically, the cost is additionally and the location of the PDC is fixed. In order to transmit
reduced by the optimization of data routing path. the PMU data reliably and in a timely fashion, sufficient link
• Since the bandwidth cost constitutes a significant propor- capacities should be assigned to these CLs. In this paper,
tion of the total system installation cost, the shortest CL we use the term bandwidth inter-changeably with channel
does not necessarily indicate minimum total cost. In some capacity. We use matrix B to denote the power system
cases, the CL will be located at the branch which is not bandwidth requirement, with its element bi,j indicating the
the shortest one for reducing the overall bandwidth cost. bandwidth requirement on the CL placed at branch i-j. In
• Since the existence of ZIBs reduces the number of general, bi,j ≥ 0. Moreover, bi,j = 0 indicates that there is
buses that are needed to be measured, the number of no data transmitted via branch i-j, which means that there is
PMUs for full observability and the total data traffic in no CL placed at branch i-j. Furthermore, the requirements on
the WAMS reduce. Furthermore, ZIBs would lessen the the CL capacity depend on the PMU data traffic and the data
optimal number of PMUs as well as the cost on CLs. routing scheme. In the following, we introduce the WAMS
The rest of the paper is organized as follows. We first model traffic model and data routing scheme in detail.
WAMS in Section II. Then, we show the OPLP model and give
some brief extensions of the proposed model in Section III, A. WAMS Traffic Model
followed by the MIQCP formulation of OPLP. In Section IV,
we carry out numerical study and provide numerical results In WAMS, the PMUs generate phasor data to make the
before we conclude in Section V. power system observable. However, the amount of data traffic
generated by different PMUs varies. In [23], the bandwidth
requirements of different smart grid applications are studied.
II. W IDE A REA M ONITORING S YSTEM (WAMS) M ODEL In most cases, 100 Kbps per node are required. For the wide
area situational awareness, the bandwidth requirement ranges
In this section, the basic model of WAMS consisting of three from 600 Kbps to 1500 Kbps. According to [24], the required
major components, PMUs, a PDC and a communication net- bandwidth for synchrophasor data transfer is dedicated based
work, is introduced. All PMUs can be placed on the buses, and on the reporting rate and the message size. In this section,
CLs can be placed on the branches that connect the buses. In we introduce the WMAS traffic model based on the linear
this way, the communication network topology overlaps with traffic model proposed in [25]. The amount of data generated
the power network topology. As summarized in [11], the over- by PMU i per second can be formulated as a linear function
lapping of networks can reduce the maintenance cost as well of the number of buses that PMU i measured as follows,
as the construction fees of WAMS. Besides, the coincidence di = (Wi p + α) · Fs (3)
of power network and the communication network would
allow the smart grid to support multiple applications [12], e.g., where Wi is the number of buses that PMU i measures
wide area measurement, substation automation, etc. Also, the (including bus i), p is the size of the data portion in a
implementations of some communication technologies, such single phasor data frame, α is the size of the frame overhead
as power-line communications that carry data on power lines, generated by this PMU, and Fs is the configured phasor data
require the overlapping of power lines and communication frame reporting frequency. In fact, α is very small compared
lines. Therefore, the assumption that only communication lines with the size of the phasor data, and α can be dropped from
can be built between two adjacent buses is widely accepted Di , resulting in,
in the optimal PMU and CL placement problem [19]. In the d i = Wi · d (4)
overlapped system, the PMU data may not be transmitted to
the PDC via a direct CL. Thus, as a key component of the where d = pFs . We assume that the data size and measurement
multi-hop data transmission, the routing scheme of the PMU frequency for all PMUs are the same. Thus, d is a positive
data will be introduced in this section. constant in this paper.

2
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

B. Data Routing Scheme be the sum of all PMU requirements. Therefore, B can be
For a certain data routing scheme, one or several paths calculated as:
connecting a PMU and the PDC will be adopted. Furthermore, ∑
N ∑
Ci

the bandwidth requirement on every CL in the adopted path B= di µi · ( Γki P ki ) (8)

i=1 k=1
is determined based on the data traffic of all PMUs in the
system. Thus, when the data routing paths are determined, which is a quadratic function of µ and Γi .
the bandwidth requirement matrix B is fixed for a certain We take the standard IEEE 5-bus system [26], as shown in
placement of PMU. Similarly, the choice of data routing Fig. 1, as an example to explain the concepts mentioned above.
paths influences the bandwidth requirements even for a fixed Suppose that two PMUs are located at bus 2 and bus 3 to make
placement of PMUs. the system observable. The PDC is located at bus 5. The PMU
Let Pi , i ∈ N , be the set of all simple paths between bus i placement here may not be the optimal one, and we just take it
and the PDC. We denote the kth element of Pi by an N × N as an example to explain the relationship between the system
path matrix, P ki . Its (j,t) entry, where j, t ∈ N , is given by: bandwidth requirement and the routing path selection. Then,
{ P2 = {bus 2 → bus 5, bus 2 → bus 4 → bus 5, bus 2 →
i,k 1 if P ki has branch j-t bus 3 → bus 4 → bus 5, bus 2 → bus 1 → bus 3 →
pj,t = (5)
0 otherwise bus 4 → bus 5} represents all the simple paths from bus
2 to the PDC. As defined in (5), the four elements in P2 can
The number of hops of the routing path P ki is denoted be represented by four 5 × 5 path matrices correspondingly.
by |P ki |. Given that there may be multiple paths with the Here, we take the third element of P2 , i.e. P23 , as an example.
same number of hops, we constrain the elements of Pi by The corresponding path matrix is shown as below:
the following rule,  
0 0 0 0 0
|P ki | ≤ |P k+1 |, ∀k ∈ N (6)  0 0 1 0 0 
i  
P2 = 
3
 0 1 0 1 0 
 (9)
For simplicity, we assume the shortest path routing scheme  0 0 1 0 1 
is adopted in this work. The shortest path means the path
0 0 0 1 0
with the minimum number of hops instead of the smallest
link length. We assume that the PMU data is routed over the Suppose that there are three buses, including bus 1, 2 and
shortest path without traffic splitting. We use P̃i to represent 5, assigned to PMU 2 to measure. According to (4), d2 = 3d
the set of all shortest paths from bus i to the PDC, and only the holds. Since we adopt the shortest path routing scheme in this
shortest path can be chosen as the data routing path. Denote work, P21 is selected as the data path and it should fulfill
the number of shortest paths connecting bus i to the PDC as the bandwidth requirement. As the green slash line shown
Ci . Thus, P̃i = {P 1i , · · · , P C
i }.
i in Fig. 1, Γ12 is set as one and the bandwidth of the CL at
The shortest paths, Pi , ∀k ≤ Ci , can be calculated by
k branch 2-5 should be at least 3d.
algorithms, such as Dijkstra Algorithm. We assume that the The use of PMU 2 alone cannot yield the full observability.
matrix P̃i and the value of Ci are known for all i. We define a Another PMU is thus placed at bus 3 for observing buses 3
set of path selection variables, Γi = (Γ1i , Γ2i . . . , ΓC and 4. The set of shortest paths from bus 3 to the PDC is
i ), where
i

i ∈ N , to indicate which path(s) will be adopted by the PMU represented by P̃3 = {bus 3 → bus 2 → bus 5, bus 3 →
installed on bus i. Γki = 1 if the k th path connecting bus i and bus 4 → bus 5}. The selections between two shortest path
the PDC is used, and Γki = 0 otherwise. Assume that a PMU result in two different bandwidth required by PMU 3. If Γ13 =
installed on bus i requires hi paths to reliably communicate 1 and PMU 3 is assigned to measure the phasor data at buses 3
with the PDC, hi ≥ 1. hi ∈ Z, we must have: and 4, the bandwidth requirements of PMU 3 on CLs located
on branches 3-2 and 2-5 are 2d. Moreover, both PMUs 2 and
∑
Ci
3 transmit data via the CL on branch 2-5. The bandwidth
Γki = hi µi (7)
requirement on CL 2-5 should be 5d. Hence, by summing the
k=1
bandwidth requirements of PMU 2 and PMU 3, we can obtain
For simplicity, we set hi = 1 in this paper. the system bandwidth matrix B as follows:
For the PMU located on bus i, it generates di amount of  
0 0 0 0 0
data per second. In order to ensure timely transmission, the  0 0 2d 0 5d 
chosen simple paths from bus i to the PDC should satisfy the  
B=  0 2d 0 0 0 
 (10)
communication bandwidth requirement, di . In other words, the  0 0 0 0 0 
bandwidth of all CLs belonging to the chosen paths should be
0 5d 0 0 0
at least di .
For the transmission of all PMU data, a CL may belong Similarly, if Γ23 = 1, we can calculate B in a similar way.
to multiple data transmission paths. We assume that the time- We thus omit the calculation here. As shown above, we can see
division multiple access (TDMA) scheme is adopted to avoid how the data routing selection influences the system bandwidth
the collision of the data transmission of numerous PMUs over requirement which affects the installation cost of CL directly.
a CL. In order to fulfill the transmission requirements of all Thus, the data routing selection can have a significant impact
PMUs in the system, the system bandwidth matrix B should on the optimal PMU-CL placement.

3
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

PMU 3 OPLP can be modeled as:

宅宸家季孶 宅宸家季孷
∑
N ∑
N ∑
N
宅宸家季孴
Minimize K µj + f (bi,j , di,j ) (13)
j=1 i=1 j>i
Subject to µM ≥ k (14)
2d bi,j ≥ bi,j ∀i, j ∈ N , (15)

where K is the installation cost of a single PMU, including

宅宸家季孵 宅宸家季學
the PMU device cost and other labour costs, k = [k, k, · · · , k].
3d 2d
PDC Equation (14) ensures that the system is fully observable.
When k = 1, (14) models the case without consideration of
PMU 2
any equipment outage. Here we consider a more general for-
mulation (k ≥ 1) which takes PMU outage into consideration.
Fig. 1: Standard IEEE 5-Bus System [26].
For example, when k = 2, the constraint in (14) means that
each bus should be measured by at least two PMUs, that is
(N − 1) redundancy [16]. For simplicity, we do not consider
C. CL Construction Cost Model
the outage of branch or CL in this work. When an equipment
In this section, we introduce a general construction cost outage occurs, we assume here that the loss of observability
model of CLs. The cost of communication infrastructure is can be resolved by referring the relevant historical data about
calculated as the sum of two components, passive cost and the captioned equipment [27].
active cost [20]. Active devices of a wired communication The following discussions in Section III-A and III-B will
network include switches and routers, whose costs depend expand the above formulation by incorporating specific char-
mainly on the communication capacity provided [17]. Here acteristics of power networks, including ZIBs and PMU
we formulate the cost of active devices as a linear function measurement limitations. These extensions can be taken into
of bandwidth. On the other hand, the passive devices include consideration simultaneously or separately. In Section III-C,
optical fiber cables and fiber path panel [17]. Besides the we give some instructions on how to solve OPLP collaborating
devices cost, the passive cost includes the labour cost, e.g. with the effect of ZIBs and PMU measurement limitations.
installation and procurement cost of the fiber, as well. Thus,
the passive cost mainly depends on the length of CL. In
summary, the construction cost of a CL is calculated based A. Effect of ZIBs
on its bandwidth and length as shown in (11). ZIBs are those buses without current injection. Therefore,
ZIBs correspond to the transshipment nodes in the system,
Costcomm = Eb · Bandwidth + El · Length (11) which brings the following new rule for assessing the observ-
ability. Among a ZIB and its adjacent buses, a single bus can
where Eb is the bandwidth price and El is passive CL price
be made observable by making the others observable [15].
per kilometer.
Thus, for a ZIB, the only one "credit of observability" can be
More generally, there may exist some CLs in the power assigned to itself or its adjacent buses. We use yi,j ∈ {0, 1}
system before we design the placement of CLs. For branch to indicate the assignment of the "credit" from bus j to bus i.
i-j, if there is an existing CL, αi,j = 1 and its preassigned When yi,j = 1, the "credit" provided by bus j is assigned to
bandwidth is denoted as qi,j , otherwise, αi,j = 0. Denote bus i, and yi,j = 0, otherwise.
the cost of building a CL placed on branch i-j with designed
To account for ZIB in the OPLP model, we replace (14) by
bandwidth bi,j and link length di,j as the function f (bi,j , di,j ).
the following constraint:
Then, we have f (bi,j , di,j ) as shown in (12).
 gi ≥ k, ∀j ∈ N (16)

Eb bi,j +El di,j if bi,j > 0 and αi,j = 0
f (bi,j , di,j ) = Eb (bi,j −qi,j ) if bi,j > qi,j > 0 where


0 otherwise ∑ ∑
(12) gi = mi,j µj + mi,j zj yi,j , ∀i ∈ N (17)
j∈N j∈N
∑
mi,j yi,j = zj , ∀j ∈ N (18)
III. O PTIMAL J OINT PMU-CL P LACEMENT (OPLP) i∈N
P ROBLEM
where zj equals one when bus j is ZIB, and zero, otherwise.
In this section, we present the OPLP problem that models As shown in (18), if zj equals one, bus j has a total number
how to optimally install PMUs jointly with the CLs in the of one "credit", which can be assigned to itself or its adjacent
proposed power system. We will first give the basic formu- buses. For bus i, its observability can be calculated as shown
lation of the OPLP problem. Then, we will extend the OPLP in (17), the sum of the measurements provided by PMUs and
problem to make it more general. the "credit".

4
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

B. PMU Measurement Limitations To realize OPLP, there may exist the installation of new
Usually, a PMU with multiple measurement channels is CLs as well as provision more capacities to some existing CLs.
installed at the bus with several neighboring buses. There exist Based on this fact, we can understand the physical meaning of
an upper bound on how many buses one PMU can measure, li,j and ηi,j in (23) easily. The binary variable li,j indicates
and this upper bound is called the measurement capacity if there is a CL placed on branch i-j. When there is a CL
limitation of this PMU. We consider the effect of PMU mea- installed on branch i-j, li,j = 1 holds, and otherwise, li,j =
surement capacity 0. When there is an existing CL on branch i-j, αi,j = 1
∑ limitations on the∑ PMU placement problem
holds. Then, according to the first part in g(bi,j , di,j ), even
by substituting j∈N mi,j µj with j∈N mi,j ωi,j µj in (16).
Thus, if li,j = 1, the passive cost on the corresponding CL equals
∑ ∑ to zero. The binary variable, ηi,j , indicates if the CL needs
gi = mi,j ωi,j µj + mi,j zj yi,j , ∀i ∈ N (19) a bandwidth upgrade. If there is an existing CL on brand i-
j∈N j∈N j and the preassigned bandwidth is needed to be upgraded,
In addition, we introduce the following constraints: ηi,j = 1, and otherwise, ηi,j = 0. The constraint (24) ensures
∑ that the placement of CL can only happen when the bandwidth
mi,j ωi,j ≤ ωjmax , ∀j ∈ N (20) requirement on the CL is larger than d. Here we formulate
i∈N
the lower bound on the bandwidth requirement as d instead of
ωi,j ≤ µj , ∀i, j ∈ N (21) zero. Based on the linear PMU traffic model, d is the minimum
where ωi,j ∈ {0, 1} represents the measurement at bus i with unit of data traffic that a PMU can generate. The constraint
the PMU located at bus j. If ωi,j = 1, bus i is measured (25) ensures that there will be an upgrade on the bandwidth,
by the PMU located at bus j, and ωi,j = 0, otherwise. if and only if the existing CL bandwidth, qi,j cannot meet the
The two constraints mean that the total number of buses a bandwidth requirement, i.e. bi,j .
PMU measures should be less than its measurement capacity With the transformation, the OPLP problem can be modeled
limitation, ωjmax . as (26). Since (24) and (25) are quadratic constraints, OPLP is
a Mixed Integer Quadratic Constraint programming (MIQCP)
C. Solving OPLP problem. Although the MIQCP for OPLP is an NP-hard
problem, many commercial solvers, such as, Cplex, Gurobi,
In this section, we will introduce a general way to solve
and Mosek, can provide approximate solution efficiently [28].
OPLP. All through this section, we take both the ZIB effect
Cplex utilies McCormick Branch and Bound approximation to
and PMU measurement limitation into consideration.
solve MIQCP problem. Branch and bound algorithms provide
Constraint (15) specifies the bandwidth requirement on each
a systematic enumeration of candidate solutions by means
CL. Since the CL cost function, f (·), as defined in (12), is
of state space search. On each search point, a second order
monotonic increasing with bi,j , the optimal solution must lie
cone programming (SOCP) should be solved. CPLEX use
on the boundary of constraint (15). In other words, OPLP is
McCormick method to provide a linear programming approx-
equivalent to:
imation of the SOCP problem. According to [29], [30], Cplex
∑
N ∑
N ∑
N maintains a provable upper and lower bounds on the (globally)
Minimize K µj + f (bi,j , di,j ) (22) optimal objective value. They can converge to the ϵ-suboptimal
j=1 i=1 j>i point in a finite number of steps.
Subject to µM ≥ k
∑
Ci
Γki = hi µi IV. N UMERICAL A NALYSIS
k=1 In this section, we evaluate the proposed OPLP model on the
∑
N ∑
Ci
IEEE 30-bus, 57-bus, 118-bus, and 300-bus test systems [31].
B= di µi ( Γki P ki ) Cplex optimization package1 is used for our study. We test all
i=1 k=1 systems under two initial conditions, with and without existing
where X = {µi ∈ ∈ {0, 1}, bi,j ≥ 0|∀i ∈ N , ∀k ≤
{0, 1}, Γki CLs. We name the cases with existing CLs OPLPeL and the
Ci } denotes the decision space of (22). cases without existing CLs OPLPnL. By comparing the results
In (22), the objective function has some piecewise linear of OPLPnL and OPLPeL, we try to find how the existing links
functions, f (bi,j , di,j ). Thus, the optimization problem cannot affect the placement of PMU and CLs. In this section, we first
be solved by any commercial solver. To tackle the issue, compare and discuss the performance results of OPP, OPLPnL
we introduce two sets of binary variables, li,j ∈ {0, 1} and and OPLPeL. Then we examine the optimal placements of CLs
ηi,j ∈ {0, 1}, where i, j ∈ N , to model the piecewise function and PMUs as well as the bandwidth requirements for CLs.
and transform the problem into a mixed integer programming Next, we show the results of OPLPnL under different PMU
problem. Then, we can transform f (bi,j , di,j ) into g(bi,j , di,j ) measurement limitations. In the end, we will test OPLPnL on
by adding constraints (24) and (25). different pricing scenarios of PMU and CLs.
g(bi,j , di,j ) = (1−αi,j )(El di,j )li,j + Eb (bi,j −qi,j )ηi,j(23) 1 Cplex Optimizer, Inc., “IBM ILOG CPLEX Optimization Studio.”

d li,j ≤ bi,j ≤ li,j bi,j (24) [Online] http://https://www.ibm.com/support/knowledgecenter/en/SSSA5P_

12.7.0/ilog.odms.studio.help/Optimization_Studio/topics/COS_home.html,
(1−ηi,j )(bi,j −qi,j ) ≤ bi,j −qi,j ≤ ηi,j (bi,j −qi,j ) (25) Accessed, Mar. 2018.

5
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

∑
N ∑
N −1 ∑
N
Minize K µj + g(bi,j , di,j )
j=1 i=1 j=i+1
j̸=i
 ∑ ∑

 mi,j ωi,j µj + mi,j zj yi,j ≥ k, ∀i ∈ N



 ∑
j∈N j∈N

 , ∀j ∈ N

 ∑i∈N mi,j yi,j = zjmax

 m ω ≤ ω , ∀j ∈ N

 i∈N i,j i,j j

 ωi,j ≤ µj , ∀i, j ∈ N



 g(bi,j , di,j ) = (1 − αi,j )(El di,j )li,j + Eb (bi,j − qi,j )ηi,j



 d li,j ≤ bi,j ≤ li,j bi,j



 (1 − ηi,j )(bi,j − qi,j ) ≤ bi,j − qi,j ≤ ηi,j (bi,j − qi,j )
 ∑Ci k
Subject to k=1 Γi = hi µi (26)

 ∑N ∑Ci

 Γki P ki )

 B= di µi (




i=1 k=1

 li,j ∈ {0, 1} ∀i, j ∈ N , i ̸= j



 ηi,j ∈ {0, 1} ∀i, j ∈ N , i ̸= j



 µi ∈ {0, 1} ∀i ∈ N



 ωi,j ∈ {0, 1}, ∀i, j ∈ N



 yi,j ∈ {0, 1}, ∀i, j ∈ N

Γki ∈ {0, 1}, ∀i ∈ N , ∀k ≤ Ci

To solve OPLP, we need to define the values of K, Eb , TABLE I: Existing CLs locations and bandwidths.
and El , where K is the cost of a PMU, Eb is the CL IEEE 30-bus System
bandwidth price per kbit/s, and, El is the passive CL price per Locations of CLs 4-6 16-17 10-17
Bandwidth (Kbps) 4d 3d 4d
kilometer. These three values vary significantly for different IEEE 57-bus System
power systems. According to [20], the price from Chinese Locations of CLs 36-37 1-16 12-16
telecommunication compancy2 and U.S-based vendor3 , we Bandwidth (Kbps) 4d 5d 3d
take K = 40000, Eb = 120, and El = 1500 in units IEEE 118-bus System
Locations of CLs 5-8 114-115 82-96
of USD in our numerical studies. In [20], the installation Bandwidth (Kbps) 5d 5d 3d
cost of optical fiber communication line is about $10000 per IEEE 300-bus System
kilometer, and 85% of the price is digging fee. Since we Locations of CLs 9-11 100-102 198-209
consider the model where CL overlaps with the power line, Bandwidth(Kbops) 5d 1d 5d
the digging fee is saved. Then, we set El = 1500. Besides,
these values represent the estimated prices of a standard
at bus 10. Before solving the MIQCP model proposed in (26),
PMU offered by a U.S.-based vendor3 and the Ethernet lines
we need to pre-calculate P i , where i ∈ N . As mentioned
provided by a Chinese telecommunication company2 . To be
before, in this paper, we evaluate data transmissions from any
able to approximate the distance matrix of each IEEE test
potential PMU bus to the PDC by considering the shortest
network, we have assumed that all transmission lines have
path only.
the same conductors with the same configurations. Thus, the
relative distances between system buses can be extracted from
the system admittance matrix [20]. We assume that the total A. Performance Comparisons
length of transmission lines in IEEE 30-bus, 57-bus, 118-bus,
First, the standard IEEE 30-bus, 57-bus, 118-bus, and 300-
and 300-bus test networks equal to 3000, 5712, 9884 and
bus test systems are investigated by the proposed OPLPeL and
25128 kilometers, respectively.
OPLPnL models. Table II shows the PMU placement results
For OPLPnL, since there are no existing CLs, we set all of OPP [15], OPLPnL and OPLPeL. Three different cases are
αi,j equals to zero. For OPLPeL, we randomly select three considered, namely, the case without ZIBs or PMU outage,
branches with pre-installed CLs, each having a bandwidth with PMU outage or (N − 1) redundancy, and, with ZIBs.
between d to 5d. The detailed configuration is shown in Columns 2-4 of Table II show the minimum PMU number
Table I. The bandwidth associated with these links, however, achieved by OPP under these three cases. Columns 5-16 show
are assigned at random. the minimum PMU number and CLs achieved by OPLPnL and
In all systems we studied, we assume that PDC is located OPLPeL under three different cases. By comparing the results
of Case 1 and 3, we can see that the existence of ZIBs reduces
2 [Online] http://www.chinatelecomglobal.com/tariffinformation/iepliplc, the number of PMUs. Besides, almost double the number of
Accessed, Sep 2017.
3 [Online] https://www.selinc.com/synchrophasors/products. Accessed, May PMUs are needed to achieve (N − 1) redundancy. Besides
2016. ZIBs, the existing of conventional measurement (CM) sensors,

6
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

TABLE II: Results of PMU Placement for IEEE standard test systems in different cases.
OPP OPLPnL OPLPeL
Test System Case 14 Case 25 Case 36 Case 14 Case 25 Case 36
Case 14 Case 25 Case 36
PMU Link PMU Link PMU Link PMU Link PMU Link PMU Link
IEEE 30-bus 10 17 7 10 13 17 18 7 9 10 15 17 19 7 11
IEEE 57-bus 17 29 11 18 33 29 38 14 27 18 33 32 59 14 28
IEEE 118-bus 32 59 28 35 60 71 85 32 59 35 59 71 85 32 59
IEEE 300-bus 87 162 68 90 167 205 303 69 141 90 165 205 303 69 141
4 The case without zero-injection buses or device outage.
5 (N − 1) redundancy case without consideration of zero-injection buses.
6 The case with zero-injection buses and no device outage.

4d 4d
5d

5d
PDC
OPLPnL Optimal PMU
OPLPnL CL Location &Bandwidth 8d

OPLPeL Optimal PMU 4d

OPLPeL CL Location&Bandwidth
Existing CLs Location&Bandwidth
8d 8d
5d 5d

3d
3d

8d
8d
6d
3d 3d 4d
6d
6d 3d

6d

d
19

25d
3d 4d 9d
3d 3d

3d

3d
Fig. 2: Cost Comparison Between OPLP and OPP.

e.g. power flow measurements and voltage measurement, will Fig. 3: OPLP Results for IEEE 30-bus system [31].
reduce the number of required PMUs according to [3]. Since
we do not consider the effect of CMs in OPLP, there would
be observability redundancies provided by the CM sensors. 118-bus, and 300-bus systems, although OPLP adopts more
From another point of view, the observability redundancies PMUs than the minimal number, it could reduce the total cost
enhance the resilience against potential cyberattack [32] and significantly by reducing the CL cost. This is because the costs
PMU failure [3]. of CLs will outweigh the costs of PMUs in large-scale power
Based on the results of Table II, existing CLs seem to systems. In this case, placing more PMUs can help to reduce
have little influence on the number of PMUs and CLs when the cost on CLs.
we compare OPLPnL and OPLPeL results. Next, we show
the locations of PMUs, locations and bandwidth of CLs of B. IEEE Standard Cases
OPLPnL and OPLPeL. All the results shown in Fig. 3, Table We show results of OPLPeL and OPLPnL in Fig. 3 based
III and IV are tested under the assumption of Case 1. on the IEEE 30-bus diagram from [31]. The solution to the
By comparing the OPLP results with those of OPP, we can OPLPnL problem suggests that, to make the system fully
see that, for the case of the IEEE 30-bus system, OPLP and observable, a total of ten PMUs are to be installed on buses
OPP require the same number of PMUs for system observ- 3, 6, 7, 9, 10, 12, 19, 24, 25, and 27. To successfully transmit
ability. However, OPLP would install more PMUs compared all PMU data to the PDC, a total of 13 CLs with properly
with OPP in 57-bus, 118-bus, and 300-bus systems. provisioned bandwidths is required. The symbol attached to
We show the cost comparison between OPLP and OPP of each CL denotes its required bandwidth of the CL. OPLPnL
IEEE 30-bus, 57-bus, 118-bus, and 300-bus systems in Fig. 2. calculates the total cost to be $1,523,320.
The bars of OPLP-CL and OPLP-PMU represent the cost of If three CLs with bandwidths 4d, 3d, and 4d were previously
CLs and PMUs acheieved by OPLP. The bars of OPP-CL and installed in the system on branches 4-6, 16-17, and 10-17,
OPP-PMU represent the cost of CLs and PMUs achieved by respectively, the solution to the OPLPeL problem suggests the
OPP, respectively. All the results are obtained from the tests same installations of PMUs are adopted as shown in Fig 3.
under the condition that there are no existing CLs. From Fig. However, the installations of CLs are different for the two
2, we can see that compared with OPP, OPLP can reduce the cases. By comparing the CL installations, we can easily find
total cost by about 25%. For small-scale systems, i.e., IEEE that a new CL is placed at branch 12 → 16 to build a new
30-bus system, OPLP and OPP pay the same amount of money transmission path, i.e., 12 → 16 → 17 → 10. Thus, the existing
on PMUs. OPLP would save the cost on CLs by adopting CLs can be fully utilized. Furthermore, the data of the PMU
the optimal locations of PMUs and selecting the optimal data located at bus 12 is routed via the new path instead of the old
routing paths. For the large-scale systems, i.e., IEEE 57-bus, path, so the bandwidth requirement on CL 4 → 6 reduces from

7
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

9d to 3d. A reduction of bandwidth requirement on CL 6 →

10 from 25d to 19d is obtained for the same reason. Since the
solution to OPLPeL utilizes the existing CLs effectively, which TABLE IV: OPLP results for IEEE 118-bus system.
reduces the cost of CL bandwidth, it achieves a reduction of
OPLPnL OPLPeL
39840 dollars in the total cost. In this case, OPLPeL calculates
Locations of P- 2 5 10 11 12 17 20 23 2 5 10 11 12 17 20 23
the total cost of enabling this PMU network to be $1,483,480. MU 25 29 34 37 40 45 49 25 29 34 37 40 45 49
50 51 52 59 65 66 71 50 51 52 59 65 66 71
TABLE III: OPLP results for IEEE 57-bus system. 75 77 80 85 87 91 94 75 77 80 85 87 91 94
101 105 110 114 116 101 105 110 114 116
OPLPnL OPLPeL Total Cost $4,047,520.0 $4,046,320.0
Locations of P- 3 4 8 12 15 20 24 1 4 8 10 20 21 24 Branch Bandwidth Branch Bandwidth
MU 28 31 32 36 38 41 28 31 32 36 41 44 2-12 3d 2-12 3d
5-8 22d ∗ 5-8 22d
46 51 52 55 57 46 49 52 55 57
Total Cost $2,801,940.0 $2,711,520.0 5-11 16d 5-11 16d
Branch Bandwidth Branch Bandwidth 8-9 161d 8-9 161d
3-15 4d ∗ 1-16 5d 8-30 139d 8-30 139d
4-6 5d 4-6 5d 9-10 161d 11-12 11d
6-8 5d 6-8 5d 11-12 11d 15-17 3d
7-8 6d 7-8 7d 15-17 3d 15-19 3d
7-29 6d 7-29 7d 15-19 3d 17-30 16d
8-9 15d 8-9 15d 17-30 16d 17-31 3d
9-10 26d 9-10 26d 17-31 3d 17-113 3d
9-11 8d 9-11 8d 17-113 3d 19-20 3d
9-55 3d 9-55 3d 19-20 3d 23-25 5d
10-12 43d 10-12 37d Link Placement 23-25 5d 25-26 9d
10-51 3d 11-41 8d and Bandwidth 25-26 9d 26-30 9d
11-41 8d 12-13 32d (Kbps) 26-30 9d 29-31 3d
12-13 37d ∗ 12-16 4d 29-31 3d 30-38 114d
13-14 3d 13-14 3d 30-38 114d 32-113 3d
CL Placement 32-113 3d 32-114 3d
13-15 10d 13-49 29d
and Bandwidth 32-114 3d 34-37 5d
13-49 24d 14-46 3d
(Kbps) 34-37 5d 37-38 17d
14-46 3d 20-21 3d
20-21 3d 21-22 6d 37-38 17d 37-40 5d
21-22 3d 22-23 4d 37-40 5d 38-65 97d
22-23 4d 22-38 10d 38-65 97d 45-49 4d
22-38 7d 23-24 4d 45-49 4d 49-50 3d
23-24 4d 28-29 3d 49-50 3d 49-51 7d
28-29 3d 29-52 3d 49-51 7d 49-66 4d
29-52 3d 31-32 3d 49-66 4d 49-69 28d
31-32 3d 32-34 7d 49-69 28d 51-52 3d
32-34 7d 34-35 7d 51-52 3d 59-63 7d
34-35 7c 35-36 7d 59-63 7d 63-64 7d
35-36 7d ∗ 36-37 11d 63-64 7d 64-65 7d
36-37 11d 37-38 11d 64-65 7d 65-68 86d
37-38 11d 38-44 3d 65-68 86d 68-69 53d
38-49 24d 38-49 24d 68-69 53d 68-81 31d
41-56 3d 41-56 3d 68-81 31d 68-116 2d
56-57 3d 56-57 3d 68-116 2d 69-70 4d
* 69-70 4d 69-75 6d
Link bandwidth upgrades required.
69-75 6d 69-77 15d
69-77 15d 70-71 4d
Compared with IEEE 30-bus system, 57-bus and 118-bus 70-71 4d 77-82 8d
77-82 8d 80-81 31d
systems have complex network topologies. Thus, showing their 80-81 31d 80-96 6d
results in a graphic manner as Fig. 3 is not readable. Next 80-96 6d 80-98 5d
we show the results of IEEE 57-bus and 118-bus system in 80-98 17d 80-99 12d
82-83 8d 82-83 8d
Table III and Table IV. The readers can refer to the system 83-85 8d 83-85 8d
graphics in [31] for the understanding of the results. 85-86 2d 85-86 2d
For the IEEE 57-bus system with no existing CLs, a total 86-87 2d 86-87 2d
91-92 3d 91-92 3d
of 18 PMUs, placed on buses 3, 4, 8, 12, 15, 20, 24, 28, 92-100 3d 92-100 3d
31, 32, 36, 38, 41, 46, 51, 52, 55 and 57, so as to make 94-96 6d 94-96 6d
98-100 17d ∗ 98-100 5d
the system fully observable is suggested. In order to support
100-101 3d 99-100 12d
timely data delivery of these PMUs, 33 CLs with bandwidths 100-103 11d 100-101 3d
properly provisioned are required for the system, as shown in 103-105 6d 100-103 11d
Table III. 103-110 5d 103-105 6d
103-110 5d
With three CLs in the system, OPLPeL will offer a different 9-10 161d
PMU installation strategy, as shown in Table III. On the one * Link bandwidth upgrades required.
hand, it requires different locations for PMU installations. On
the other hand, there is a dramatic reduction of $90,420 in the
total installation cost as compared with OPLPnL.

8
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

Eb=12 Eb=120 Eb=1200

The Number of PMUs

40

20

0
0.15 1.5 15 150 1500 15000 150000
El
2000

Length of CLs
The Total
1000

0
0.15 1.5 15 150 1500 15000 150000
El
1500

Bandwidth of CLs
1000

The Total
500

0
0.15 1.5 15 150 1500 15000 150000
El

Fig. 4: Results of OPLP for IEEE standard test systems Fig. 5: Results of OPLP under different pricing scenarios.
considering measurement limitation.

IEEE 118-bus system. The price of PMU, K, and, the price of

For the 118-bus sytem, OPLPnL and OPLPeL adopt the CLs, including the passive CL price, El , and, the bandwidth
same installation strategy of PMUs but different CL installa- price, Eb , may influence the placement of PMU and CL. In
tion schemes. By changing the data routing paths, OPLPeL order to determine how the results will be influenced, we do
employs the existing CLs. Consequently, the total installation the following tests. First, we keep the PMU price K = 40000
cost achieved by OPLPeL is $1,200 cheaper than the cost of fixed. Then, we change the passive CL price El from 0.15 to
OPLPnL. 150000 under different bandwidth prices, including Eb = 12,
The results on these case studies suggest that existing CLs Eb = 120 and, Eb = 1200. The results are shown in Figure
in a power system may alter the data routing paths used 5. The blue, yellow and red bars show the results when Eb
by the installed PMUs in order to utilize the existing CLs. is 12, 120 and 1200, respectively. As shown in the figure, the
Then, OPLPeL may result in a different PMU installation increase of El results in the decrease of total CL length. As El
strategy and different CL installations. In summary, OPLPeL increases, both the number of PMUs and total CL bandwidth
takes advantage of the existing CLs so that the total cost of increase.
establishing the WAMS is reduced. Then, we have a look at each cluster of bars in Figure 5 and
take El = 150 as an example. As Eb increases, although the
C. Measurement Capacity Limitation Effects number of PMUs remains the same, the total CL bandwidth
Assuming that all PMUs come with the same measurement decreases and the total CL length increases. This is because the
capacity limitation, we test nine different cases on OPLPnL locations of PMUs change to reduce the CL cost, although the
model in IEEE 30-bus, 57-bus, 118-bus, and 300-bus systems. total number of PMUs remains unchanged. However, with a
Fig. 4 shows the minimum numbers of PMUs and CLs large value of El , El ≥ 1500, for example, the changes of CL
achieved by OPLPnL. The figure shows that as PMU mea- placement under different bandwidth prices become tiny. We
surement capacity limitation increases, the numbers of PMUs can observe similar findings in Fig. 6. When El ≤ 1500, due
and CLs decrease. However, when the PMU measurement to the relative small value of the total cost, the differences of
capacity is larger than four, even though it increases, the the total cost under different bandwidth prices are relatively
optimal number of PMUs and CLs do not decrease. The reason apparent. When El > 1500, the differences of bars within
is that when the PMU measurement capacity is large, the one cluster are tiny. This is because when El > 1500, the
PMU measurement capacity cannot be fully utilized due to passive CL cost takes a large proportion of the total cost and
the power system topology. For bus n, since it has a limited the bandwidth cost becomes negligible.
number of adjacent buses, the maximum number of buses In summary, the placements of PMUs and CLs change a lot
that need to be measured by PMU n is limited. For example, under different pricing scenarios. The cost on CLs, including
even though a PMU may have a measurement capacity of 100 the passive cost and active cost, is non-negligible in the OPP
buses, at least a measurement capacity of 94 buses would be problem, which shows the importance of OPLP problem .
unutilized when it is located at the bus with only 5 adjacent
buses. Consequently, more PMUs must be placed to measure V. C ONCLUSIONS
the buses that are not connected to the PMU. Therefore, as In this paper, we extend the OPP problem by formulating an
the PMU measurement capacity exceeds a certain value, the OPLP problem. The proposed model not only minimizes the
increase of PMU measurement capacity will not make the total cost of PMUs and CLs, but also takes the power system
required number of PMUs and CLs reduce further. observability, and data transmission bandwidth requirements
into consideration. Some extensions, including ZIBs, PMU
D. OPLP Results under Different Pricing Scenarios measurement capacity limitation, are considered. By carrying
In this section, we show the results of OPLP under different out numerical studies on the IEEE standard systems, we
pricing scenarios in Fig. 5 and Fig. 6. The test is conducted in discovered that, when communication costs are considered,

9
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

[11] J. Gao, Y. Xiao, J. Liu, W. Liang and C. L. P. Chen, “A survey of

communication/networking in smart grids," Future Generation Comput-
er Systems, Vol. 28, No. 2, pp, 391-404, Feb. 2012.
[12] N. E. Wu and M. Ruschmann, “Toward a highly available modern
grid,” Chapter 10 of Control and Optimization Theories for Electric
Smart Grids, A. Chakrabortty and M. Ilic, Eds. Berlin, Germany:
SpringerVerlag, 2011.
[13] M. Kuzlu, M. Pipattanasomporn, and S. Rahman, “Communication
Network Requirements for Major Smart Grid Applications in HAN,
NAN and WAN,” Computer Networks, Vol. 67, pp. 74–88, Jul. 2014.
[14] M. Shahraeini, M. H. Javidi, and M. S. Ghazizadeh, “Comparison be-
tween Communication Infrastructures of Centralized and Decentralized
Wide Area Measurement Systems,” IEEE Transactions on Smart Grid,
Vol. 2, No. 1, pp. 206–211, Mar. 2011.
[15] F. Aminifar, A. Khodaei, M. Fotuhi-Firuzabad, and M. Shahidehpour,
“Contingency-Constrained PMU Placement in Power Networks,” IEEE
Fig. 6: Total cost of OPLP under different pricing scenarios. Transactions on Power Systems, Vol. 25, No. 1, pp. 516-523, Feb. 2010.
[16] R. F. Nuqui and A. G. Phadke, “Phasor measurement unit placement
techniques for complete and incomplete observability,” IEEE Transac-
tions on Power Delivery, Vol. 20, No. 4, pp. 2381-2388, Oct. 2005.
more PMUs may be required to minimize the total cost of [17] M. B. Mohammadi, R. A. Hooshmand and F. H. Fesharaki, “A New
building WAMS. Moreover, by comparing the cases with and Approach for Optimal Placement of PMUs and Their Required Com-
munication Infrastructure in Order to Minimize the Cost of the WAMS,”
without existing CLs, we have found that the presence of IEEE Transactions on Smart Grid, Vol. 7, No. 1, pp. 84-93, Jan. 2016.
the existing CLs may result in the relocation of PMUs and [18] Z. H. Rather, Z. Chen, P. Thøgersen, P. Lund, and B. Kirby, “Realistic
different transmission paths for PMU data. Finally, we found Approach for Phasor Measurement Unit Placement: Consideration of
Practical Hidden Costs,”IEEE Transactions on Power Delivery, Vol. 30,
that a tiny change on PMU or CL price influences optimal No. 1, pp. 3-15, Feb. 2015.
results significantly, which proves the importance of OPLP in [19] J. Huang, N. E. Wu and M. C. Ruschmann, “Data-Availability-
saving cost. Constrained Placement of PMUs and Communication Links in a Power
System,"" IEEE Systems Journal, Vol. 8, No. 2, pp. 483-492, Jun. 2014.
[20] M. Shahraeini, M. S. Ghazizadeh, and M. H. Javidi, “Co-optimal
Placement of Measurement Devices and Their Related Communication
VI. ACKNOWLEDGMENT Infrastructure in Wide Area Measurement Systems,”IEEE Transactions
on Smart Grid, Vol. 3, No. 2, pp. 684-691, Jun. 2012.
This work is supported in part by the Theme-based Research [21] Pal, A. K. S. Vullikanti, and S. S. Ravi, “A PMU Placement Scheme
Scheme of the Research Grants Council, Hong Kong Special Considering Realistic Costs and Modern Trends in Relaying,” IEEE
Administrative Region, China, under Grant No. T23-701/14- Transactions on Power Systems, Vol. 32, No. 1, pp. 552-561, Jan. 2017.
[22] B. Yang, K. V. Katsaros, W. K. Chai and G. Pavlou, “Cost-Efficient Low
N. Latency Communication Infrastructure for Synchrophasor Applications
in Smart Grids,” IEEE Systems Journal, No. 99, pp. 01-11
[23] P. Ritosa et al..“Traffic Modelling, Communication Requirements
R EFERENCES and Candidate Network Solutions for Real-time Smart Grid Control,”
[Online]. Avaiable:http://sunseed-fp7.eu/wp-content/uploads/2015/04/
[1] V. Terzija et al., “Wide-Area Monitoring, Protection, and Control of SUNSEED-WP3-D31-V20-03072015.pdf,2017. Accessed: Mar. 2018.
Future Electric Power Networks,” Proceedings of the IEEE, Vol. 99, [24] T. Babnik, K. Görner, B. Mahkovec, “Wide Area Monitoring System,”
No. 1, pp. 80-93, Jan. 2011. Monitoring, Control and Protection of Interconnected Power Systems
[2] A. G. Phadke and J. S. Thorp, Synchronized Phasor Measurements and (ed. U. Häger, C. Rehtanz, N. Voropai), Springer, 2014.
Their Applications. Springer Science & Business Media, 2008. [25] M. H. F. Wen and V. O. K. Li, “Optimal Phasor Data Concentrator
[3] K. G. Khajeh, E. Bashar, A. M. Rad and G B. Gharehpetian, “Integrated Installation for Traffic Reduction in Smart Grid Wide-Area Monitoring
Model Considering Effects of Zero Injection Buses and Conventional Systems,” in Proceedings of IEEE GLOBECOM 2013, pp. 2622-2627,
Measurements on Optimal PMU Placement,” IEEE Transactions on Dec. 2013.
Smart Grid, Vol. 8, No. 2, pp. 1006-1013, Mar. 2017. [26] G. W. Stagg, and A. H. El-Abiad, Computer Methods in Power System
[4] M. Esmaili, “Inclusive multi-objective PMU placement in power systems Analysis, McGraw-Hill Publication, 1968.
considering conventional measurements and contingencies,” Internation- [27] A. A. Saleh, A. S. Adail and A. A. Wadoud, “Optimal phasor
al Transactions on Electrical Energy Systems, Vol. 26, No. 3, pp. 609- measurement units placement for full observability of power sys-
626, Mar. 2016. tem using improved particle swarm optimisation," IET Generation,
[5] D. J. Brueni and L. S. Heath, “The PMU Placement Problem,” Transmission&Distribution, Vol. 11, No. 7, pp. 1794-1800, May 2017.
SIAM Journal on Discrete Mathematics, Vol. 19, No. 3, pp. 943-946, [28] H. Miittelmann,“The State-of-the-Art in Conic Optimization Software,”
Dec. 2005. in Handbook of Semidefinite, Conic and Polynomial Optimization, ser.
[6] C. Rakpenthai, S. Premrudeepreechacharn, S. Uatrongjit, and International Series in Operations Research and Management Science,
N. R. Waston, “An optimal PMU placement method against M. F. Anjos and J. B. Lasserre, Eds. USA: Springer Science+Business
measurement loss and branch outage,” IEEE Transactions on Power Media, LLC, 2012, Vol. 166, pp. 671–686.
Delivery, Vol. 22, No. 1, pp. 101-107, Jan. 2007. [29] N. Boland, S. S. Dey, T. Kalinowski, M. Molinaro, and F. Rigterink.
[7] N. M. Manousakis, and G. N. Korres, “Optimal PMU Placement “Bounding the gap between the McCormick relaxation and the con-
for Numerical Observability Considering Fixed Channel CapacityA vex hull for bilinear functions,”Mathematical Programming, No. 162,
Semidefinite Programming Approach,” IEEE Transactions on Power pp. 523535, 2017
Systems, Vol. 31, No. 4, pp. 3328-3329, Jul. 2016. [30] S. Boyd, “Branch-and-bound methods,” [Online] http://www.stanford.
[8] M. H. F. Wen, X. Jin, and V. O. K. Li, “Optimal Multistage PMU edu/class/ee364b/lectures/bbslides.pdf, 2007, Accessed: Mar. 2018.
Placement for Wide-Area Monitoring,” IEEE Transactions on Power [31] University of Washington, Power Systems Test Case Archive,
Systems, Vol. 28, No. 4, pp. 4134-4143, Nov. 2013. [Online] http://www.ee.washington.edu/research/pstca/. Accessed: Mar.
[9] J. Xu, M. H. F. Wen, V. O. K. Li, and K.-C. Leung, “Optimal 2018.
PMU Placement for Wide-Area Monitoring Using Chemical Reaction [32] M. Jin, J. Lavaei and K. Johansson, “A semidefinite programming
Optimization,” in Proceedings of IEEE ISGT 2013, Feb. 2013. relaxation under false data injection attacks against power grid AC state
[10] C. X. Zhang, Y. W. Jia, and X. Zhao, “Optimal PMU Placement estimation,” 2017 55th Annual Allerton Conference on Communication,
for Voltage Control,” in Proceedings of IEEE SmartGridComm 2016, Control, and Computing (Allerton), Monticello, IL, pp. 236-243, 2017.
pp. 747-751, Dec. 2016.

10
1949-3053 (c) 2018 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.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TSG.2018.2860622, IEEE
Transactions on Smart Grid

Xingzheng Zhu received the B.S. degree in automa- Ka-Cheong Leung (S’95–M’01) received the
tion from Harbin Institute of Technology, Harbin, B.Eng. degree in Computer Science from the Hong
China, in 2011 and the M.S. degree in control Kong University of Science and Technology, Hong
science and control engineering from Shanghai Jiao Kong, in 1994, the M.Sc. degree in Electrical Engi-
Tong University, Shanghai, China, in 2014. She is neering (Computer Networks) and the Ph.D. degree
currently working toward the Ph.D. degree at the in Computer Engineering from the University of
University of Hong Kong under supervision of Dr. Southern California, Los Angeles, California, USA,
K.-C. Leung. Her research interests are in the opti- in 1997 and 2000, respectively. He worked as Senior
mization of communication systems and smart grids, Research Engineer at Nokia Research Center, Nokia
including optimal PMU placement problem, energy Inc., Irving, Texas, USA from 2001 to 2002. He was
management in microgrids, demand response, ener- Assistant Professor at the Department of Computer
gy storage, and integration of renewable energy sources. Science at Texas Tech University, Lubbock, Texas, USA, between 2002 and
2005. Since June 2005 he has been with the University of Hong Kong, Hong
Kong, where he is currently Assistant Professor at the Department of Electrical
and Electronic Engineering. His research interests include smart grid, vehicle-
to-grid (V2G), power flow routing, smart microgrids, transport layer protocol
design, congestion control, and wireless packet scheduling.

Miles H. F. Wen received his Bachelor degree

in Computer Engineering with First Class Honors
and minor in Finance from HKU in 2011. He then
obtained his PhD degree in Electrical & Electronic
Engineering under Prof. Victor Li from HKU in
2015.
Dr. Wen is an alumnus of Lee Hysan Hall, for-
mer resident tutor at New College, former Found-
ing Vice-President of Eta Kappa Nu (Lambda Iota
Chapter) at HKU, and former Fulbright scholar at
Department of Electrical Engineering & Computer
Science, UC Berkeley. He is an expert in power system, information and
communication technologies, optimization algorithms, big data, machine
learning. He is the author of over 10 academic papers and the inventor of one
patent. In 2015, he co-founded Fano Labs (formerly known as "Accosys"), a
high-tech startup backed by HKU, Hong Kong Science & Technology Park,
angel investors, and Horizons Ventures. He is currently the CEO of Fano Labs.
Since early 2016, he has been appointed as an Honorary Assistant Professor
(Electrical and Electronic Engineering) at HKU.

Victor O. K. Li (S’80-M’81-SM’86-F’92) received

SB, SM, EE and ScD degrees in Electrical Engi-
neering and Computer Science from MIT. Prof. Li
is Chair of Information Engineering and Cheng Yu-
Tung Professor in Sustainable Development at the
Department of Electrical & Electronic Engineering
(EEE) at the University of Hong Kong. He is the
Director of the HKU-Cambridge Clean Energy and
Environment Research Platform, an interdisciplinary
collaboration with Cambridge. He was the Head of
EEE, Assoc. Dean (Research) of Engineering and
Managing Director of Versitech Ltd. He serves on the board of Sunevision
Holdings Ltd., listed on the Hong Kong Stock Exchange and co-founded
Fano Labs Ltd., an artificial intelligence (AI) company with his PhD student.
Previously, he was Professor of Electrical Engineering at the University of
Southern California (USC), Los Angeles, California, USA, and Director of
the USC Communication Sciences Institute. His research interests include
big data, AI, optimization techniques, and interdisciplinary clean energy and
environment studies. In Jan 2018, he was awarded a USD 6.3M RGC Theme-
based Research Project to develop deep learning techniques for personalized
and smart air pollution monitoring and health management. Sought by gov-
ernment, industry, and academic organizations, he has lectured and consulted
extensively internationally. He has received numerous awards, including the
PRC Ministry of Education Changjiang Chair Professorship at Tsinghua
University, the UK Royal Academy of Engineering Senior Visiting Fellowship
in Communications, the Croucher Foundation Senior Research Fellowship,
and the Order of the Bronze Bauhinia Star, Government of the HKSAR. He
is a Fellow of the Hong Kong Academy of Engineering Sciences, the IEEE,
the IAE, and the HKIE.

11
1949-3053 (c) 2018 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.