Measurement 95 (2017) 8–20

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Optimal PMU placement for power system observability using Taguchi

binary bat algorithm
Basetti Vedik ⇑, Ashwani K. Chandel
Electrical Engineering Department, NIT-Hamirpur, H.P, India

a r t i c l e i n f o a b s t r a c t

Article history: Phasor measurement units (PMUs) are considered as a promising tool in wide area monitoring, protec-
Received 1 December 2015 tion, and control of power system networks. In this paper, a novel technique based on Taguchi binary
Received in revised form 12 September bat algorithm (TBBA) is proposed to determine the optimal number and placement locations of PMUs
2016
such that power system is completely observable. The proposed TBBA combines the systematic reasoning
Accepted 16 September 2016
Available online 17 September 2016
ability of the Taguchi method with the traditional binary bat algorithm thereby enhances the initial pop-
ulation, and subsequently, improves the computational efficacy of the solution. The effectiveness of the
proposed method is demonstrated on a number of power system test systems under different operating
Keywords:
Binary bat algorithm
conditions. Results thus obtained are compared with the existing techniques published in the literature.
Observability From the results it has been found that the proposed technique not only yields multiple global optimal
Optimal placement solutions in single experimental run but also provides solution with lesser computational time.
Phasor measurement unit Further, the proposed technique provides similar or better optimal solutions in comparison with the
Taguchi method other techniques proposed in the literature.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction operation of power system by utilising the synchronised voltage

and current phasor measurements. In wide-area interconnected
Modern power industry has been witnessing a major transi- system, these measurements are synchronised by utilising the time
tional phase during the last decade. With the integration of renew- stamp provided by the Global Positioning System (GPS) [3]. The
able energy resources and deregulation of power system network, distinctive ability of PMUs to provide synchronised voltage and
the power flow pattern has become less predictable. Further, with current phasor measurements makes it to be one of the most
the ever-increasing power demand the present day electrical grid essential measuring devices for monitoring and control of power
is operated under intemperately stressed conditions. In such cir- systems. Once the PMU is installed at a bus, it provides voltage
cumstances, so as to operate the power system in reliable, secure, phasor of the bus and current phasors of some or all the branches
and stable manner, the energy management system (EMS) has incident to the installed bus. Subsequently, by employing Kirch-
been equipped with a tool called state estimator (SE). The main hoff’s laws, voltage phasors of the neighbouring buses can be com-
function of SE is to provide the real time operating condition of puted. Thus, a PMU with sufficient number of channel capacity
the power system by utilising the network topology and the infor- installed at a bus can not only make the installed bus observable
mation obtained from various measurement devices and monitor- but also the adjacent buses which are incident to it. As a result,
ing systems [1]. These measurement devices are judiciously it is not essential to install the PMUs at all the buses for complete
dispersed across the power system such that the power system power system observability. Moreover, it is not economical to
network is completely observable [2]. Complete power system place PMUs at every bus of the power system due to their high
observability determines the ability of SE to provide unique solu- installation cost and not possible due to lack of communication
tion from the available number of measurements. infrastructure [4]. Consequently, a suitable method is required to
Phasor measurement units (PMUs) are intelligent electronic determine the optimal placement and locations of PMUs. There-
devices, which provide the coherent picture of the real-time fore, the optimal PMU placement (OPP) problem is formulated as
an optimisation problem by the researchers. The main objective
of OPP problem is to minimise the number of PMUs required to
⇑ Corresponding author. be deployed in the power system while maintaining complete
E-mail address: [email protected] (V. Basetti). system observability [5]. In order to make an N-bus system

http://dx.doi.org/10.1016/j.measurement.2016.09.031
0263-2241/Ó 2016 Elsevier Ltd. All rights reserved.
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 9

observable, there exists a solution space of 2N possible combina- improved fruit fly (IFF) optimisation algorithm is utilised to deter-
tions. Therefore, this OPP problem is considered as a NP-hard com- mine the minimum number of PMUs. However, it is stated in
binatorial optimisation problem and has been an area of research [1,7,14,15] that the time taken to provide the solution using
for many engineers and researchers [1–15]. meta-heuristic techniques is high and the solution thus obtained
As yet, several approaches have been employed in the technical may not always be optimum [1,15] for higher order systems, i.e.
literature to solve this NP-hard combinatorial optimisation prob- IEEE 57-bus, IEEE 118-bus, etc. To overcome the shortcomings
lem [5–13]. These methods can be broadly divided into two main of the individual algorithms, several researchers are now
classes of algorithms, viz. numerical and topological. The numeri- paying attention to hybridise different optimisation techniques
cal observability method is based on the measurement gain or [16–20].
Jacobian matrix. If the gain matrix is of full rank, then the power The Taguchi method (TM) is an optimisation technique pro-
system network is said to be numerically observable. This method posed by Taguchi [20] based on the concept of statistical experi-
is iterative in nature and requires huge matrix handling and is mental design. It is used to obtain the robust solutions for
computationally expensive [7]. On the other hand, topological evaluating and implementing enhancements in manufacturing
observability method is the commonly used technique which is design problems [20]. TM provides a systematic and efficient
based on the concept of spanning tree. Some of the important con- approach with reduced number of experiments to obtain optimal
tributions concerning the state-of-the-art optimisation techniques values of complex systems. Thus, this method has been applied
in the area of OPP problem can be found in [8]. to various optimisation problems in the electric power system
In [9], a dual search algorithm is developed to obtain the mini- [16–21]. The important tools used in the Taguchi method for
mal number of PMUs. This dual search algorithm utilises the mod- searching the optimal solution are (i) orthogonal arrays (OAs)
ified bisecting search algorithm and the simulated annealing and (ii) signal-to-noise ratio (SNR). OA is used to analyse the deci-
method to obtain the minimal number and minimal placement sion variables concomitantly, on the other hand SNR is used to
locations of PMUs respectively. However, this method suffers from measure the quality of the solution [16]. The distinctive feature
high computational burden. Further, it has been reported in [10] of the Taguchi method is that it has the ability to select optimal
that the dual search technique may not yield optimal solutions number of design experiments for testing from all the possible
for larger test systems. An evolutionary technique based on genetic number of combinations [20]. Hence, TM requires less number of
algorithm is formulated in [6] for OPP by considering zero injection testing scenarios to provide optimum solution for NP-hard prob-
buses (ZIB). However, this approach [6] is studied on small test sys- lems. Due to this reason, TM provides optimal solution with
tems when compared to those exist in practice. Further, the impli- shorter computational time. Although, the Taguchi method is suf-
cation aspect such as execution time has not provided. In [11], ficient to deal with single-objective NP-hard problems, it can be
Tabu search algorithm is suggested to find the optimum number hybridised with another optimisation method in order to obtain
of PMUs by reducing the search space in obtaining the optimal multiple solutions and further to enhance the initial population
solution. The shortcoming of this methodology is that it may not [19].
result in global optimal solution that can make the system com- This paper proposes a novel hybrid Taguchi binary bat algo-
pletely observable [7]. In [7], a meta-heuristic optimisation tech- rithm (TBBA) for OPP problem. Its silent feature is to incorporate
nique based on immunity genetic algorithm (IGA) is presented to the Taguchi method through generation of initial population for
overcome the computational burden in obtaining the optimum binary bat algorithm (BBA). The main aim of this hybrid technique
solution. Although, the IGA approach provides optimal solution is to obtain multiple optimal solutions in single experimental run
with minimal execution time, single line outage and loss of PMU and to improve the computational time by decreasing the feasible
contingencies are not considered. The authors in [4] formulated search locations (solution space). In real-time application many
OPP problem using binary particle swarm optimisation (BPSO) factors such as installation cost, communication infrastructure,
technique by simultaneously considering multiple objectives, viz. zero-injection buses, etc., may affect the optimal number of PMUs.
minimising the installation cost of PMUs and maximising the mea- Consequently, a good PMU placement algorithm must not only find
surement redundancy. However, this approach is highly sensitive the minimum number of PMUs under normal conditions but also
to control parameters and requires high tuning of parameters in during the loss of PMUs, lack of communication infrastructure at
obtaining the optimal solution. substations, channel limitations of PMUs, unequal PMU cost, in
In [12], a modified binary particle swarm optimisation (MPSO) the presence of conventional measurements, and by considering
technique is employed by considering both PMU loss and single the effect of ZIBs so as to make the power system completely
loss of branch conditions. Further, a new rule based on observabil- observable. The major contribution of this paper is threefold. (i)
ity analysis of ZIB is suggested to obtain the minimum number of To propose a novel Taguchi binary bat algorithm to solve OPP
PMUs. However, it has been observed that the execution time in problem. (ii) To obtain multiple optimal solutions in single exper-
obtaining the optimal solution using MPSO is quite high. The imental run. (iii) To improve the computational speed and reliabil-
authors in [13] presented a flexible approach based on cellular ity in obtaining optimum number of PMUs under various
genetic algorithm (CGA) in finding the optimal number of PMUs constraints.
by considering channel limitations. However, the effect of unequal The proposed TBBA has been implemented on IEEE 14-bus, IEEE
PMU cost is not incorporated in the problem formulation. A new 30-bus, IEEE 57-bus, IEEE 118-bus, and Polish 2383-bus test sys-
technique based on binary imperialistic competition algorithm tems. The results thus obtained using the proposed technique is
(BICA) is employed for OPP in [5]. Although the BICA method con- compared with other evolutionary and conventional techniques
siders various contingency conditions, yet channel limitation and reported in the literature. The obtained results show the efficiency
unequal PMU cost are not considered in the objective function. In of the proposed technique both in computational speed and accu-
[3], the multi-objective OPP is suggested using cellular learning racy. The paper is organised as follows. Section 2 briefly describes
automata (CLA) by simultaneously determining the minimum the optimal PMU placement problem formulation of each model.
number of PMUs as well as maximum measurement redundancy. Taguchi method is introduced in Section 3. The proposed hybrid
In [10], a two-step approach is developed to minimise the size of Taguchi binary bat algorithm approach is presented in Section 4.
the PMU location set. In the first step, the best configuration of Section 5 briefly outlines the implementation of the proposed TBBA
PMUs for complete system observability is obtained. Thereafter, for OPP problem with flow chart. Simulation results are discussed
in the second step a novel meta-heuristic technique known as in Section 6. Finally, conclusions are provided in Section 7.
10 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

2. Problem formulation of the optimal PMU placement problem nel, the cost of PMU has been increased with a decimal fraction of
0.1 p.u. as suggested in [23].
The main objective of the PMU placement problem is to min-
imise the number of PMUs required to make the power system 2.3. Effect of channel limitation
completely observable with minimal cost [7,22]. For a N bus sys-
tem, the optimal PMU problem is formulated as: In this subsection, the effect of channel limitation on the num-
ber and location of PMUs required to make the power system com-
minx JðxÞ ¼ W t X
ð1Þ pletely observable has been considered. As stated in Section 2.2,
^
s:t: f ðxÞ P 1 each PMU comes with channel limitation and the cost of each
where, JðxÞ represents the objective function, f ðxÞ denotes the vec- PMU mainly depends on the number of measurement channels.
tor function, W signifies the installation cost of PMUs at different In practice, there are numerous types of PMUs that can be
employed with different measurement channels. Consequently,
buses given by W ¼ ½w1 ; w2 ; . . . ; wN t , X indicates PMU location vec-
in this work a more realistic OPP problem has been formulated
tor given by X ¼ ½x1 ; x2 ; . . . ; xN t , xi denotes a binary variable, either 0
according to the procedure given in [26] by considering the PMUs
or 1, which represents the absent or present of a PMU at bus i, and 1 ^
with limited channel capacity. Thus, the PMU installed at a bus can
indicates unit vector. measure only the limited number of current phasors. Thus, a PMU
Installation costs of PMUs in (1) at all the buses are assumed to can only make a limited number of buses observable among all the
have equal cost. Hence, the installation cost vector W is a unit vec- available adjacent buses [13,27]. The number of ways to make this
tor. The observability constraint f i for the ith bus is given by limited number of buses observable increases with increased num-
X
N ber of branches connected to the installed bus. For a N bus system,
fi ¼ Aij xj P 1 ð2Þ the optimal PMU problem with channel limitation is formulated
j¼1 as:
where, A represents the connectivity matrix of the power system minx JðyÞ ¼ W t Y þ aNUO
network and is defined by ð6Þ
^
s:t: f ðYÞ ¼ BT Y P 1

1; if i ¼ j or i is adjacent to j
Aij ¼ ð3Þ where, JðyÞ represents the objective function, f ðYÞ denotes the vec-
0; otherwise
tor function, W signifies the installation cost of PMUs at different
This constraint optimisation problem has been subsequently buses W ¼ ½w1 ; w2 ; . . . ; wN t , and Y indicates ðs  1Þ PMU location
transferred to an unconstrained problem by adding the penalty binary integer vector. Matrix B is formed according to connectivity
factors in the objective function. As a result, the modified objective matrix A according to (3) and number of rows in B for each bus is
function of the OPP problem is defined according to (4) obtained according to (7).
8 
JðxÞ ¼ W t X þ aNUO ð4Þ >
< n if n P m
where, NUO indicates the number of unobservable buses associated rk ¼ m ð7Þ
>
:
with individual i and a represents the penalty factor. 1 if n < m

where, n indicates the number of buses adjacent to bus i, m denotes

2.1. Effect of PMU loss
the number of measurement channels. For n P m the number of
Optimal number of PMUs obtained using (1) and (2) guarantees rows for bus i in matrix B is the number of m combinations of n that
makes the adjacent buses observable. However, for n < m the num-
that the power system is observable under normal operating con-
dition. However, the constraint in (2) does not ensure complete ber of available measurement channels can make all the adjacent
buses observable. Hence, the corresponding row in connectivity
system observability during the loss of single PMU or loss of any
communication link [2]. Therefore, it is necessary to enhance the matrix A is unaltered.
system reliability by installing the additional PMUs in order to pro-
tect the system against the loss of single PMU or any communica- 2.4. Effect of lack of communication infrastructure
tion link [23–25]. These additional PMUs should be installed in
such a way that each bus is observed by at-least two PMUs. This In wide area monitoring system, communication infrastructure
can be accomplished by multiplying the right hand side of the (i.e. communication link) is required to transmit the data collected
inequality given in (2) by 2. by the measurement devices to the control centre. The installation
of these communication links and its upgrade are the two main
X
N
factors affecting the installation cost of PMUs [28]. It is specified
fi ¼ Aij xj P 2 ð5Þ
j¼1 in [28] that owing to the absence of sufficient existing communica-
tion link and thereby up-gradation of the communication infras-
tructure increase the installation cost of PMUs by a factor of
2.2. Effect of unequal cost seven. Further, it is also stated in [28], the communications cost
for installing additional PMUs is comparatively low once a commu-
The number of current phasors that could be measured depends nication network is installed. Hence, lack of communication infras-
on the channel capacity of the PMUs. Moreover, as reported in the tructure at substations should be considered to determine its effect
literature the cost of these PMUs varies from $30 K to $40 K on number and location of PMUs to make the system fully observ-
depending upon the number of measuring channels [23]. There- able. In the present study, this is considered by imposing a higher
fore, in order to determine the effect of unequal cost on the num- cost at substations without communication infrastructure. Thus,
ber and position of PMUs, the installation cost given in (1) has to be the installation cost W should be modified in the objective function
altered. In the present paper, the cost of a PMU that can measure which is to be minimised according to (8).
one current phasor measurement in the incident branch connected
to the installed bus is set to 1 p.u. Further, for each additional chan- minx JðxÞ ¼ W t X þ aNUO ð8Þ
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 11

Here, the lack of communication infrastructure at a particular bus i 3. Taguchi method

is indicated with high installation cost (wi ¼ 27  104 ) and remain-
ing buses are assigned with an installation cost of wi ¼ 1 as sug- The Taguchi method is an optimisation technique based on the
gested in [28]. concept of statistical experimental design. This method has an abil-
ity to optimise quickly by changing the factors (variables) to obtain
a desired outcome. The important tools used in the Taguchi
2.5. Effect of flow measurements
method for searching the optimal solution are (i) orthogonal arrays
(OAs) and (ii) signal-to-noise ratio (SNR). OA is used to analyse the
In this subsection, the OPP problem is discussed in the presence
decision variables concomitantly whereas SNR is used to measure
of few existing conventional flow measurements. This is because
the quality of the solution [16].
PMU technology is comparatively new and most of the existing
power systems already consist of considerable number of tradi-
tional measurements to monitor the power system. Present work
3.1. Orthogonal array
also takes into account the situation where the power system con-
sists of a few observable and unobservable islands [22,25]. This
In exhaustive search (ES) technique, experiments are per-
situation may arise in practice due to the decision of replacing
formed using all factorial combinations to achieve the optimal
some of the aging or malfunctioning of existing conventional
solution. In the Taguchi method, OA is utilised to obtain the
measurement units [22,25]. In the present paper, a topological
desired outcome by performing a subset of experiments of all
method proposed in [5,22] has been considered for OPP problem
possible combinations of the values of all the factors (variables).
in the presence of conventional measurements. With the existing
An OA is a fractional factorial matrix that guarantees an equal
conventional measurements, the power system can be made com-
comparison of levels of any variable or interaction of variables.
pletely observable with fewer numbers of PMUs. This is because,
This fractional matrix consists of numbers arranged in rows
the presence of power flow measurement on a branch can measure
and columns. In this matrix each row indicates the level of the
both active and reactive power flows of the corresponding branch
variable and each column indicates a particular variable that
say i  j. Hence, if the voltage phasor of bus i is known either
can be changed in each experimental run [30]. The experiment
directly or indirectly, the voltage phasor at other end of the branch,
combinations are selected such that adequate information is
i.e. bus j can be calculated using the branch parameters. Thus,
provided to determine the effects of parameter values of each
the constraints allied with the bus i and bus j (terminal buses of
variable on the solution. Hence, this decreases the computational
the measured branch) can be merged into one constraint [4]. The
time compared to the ES technique in attaining the optimal solu-
OPP problem in the presence of conventional flow measurements
tion [21].
is to obtain the minimal number of PMUs to make the power
This is explained with an example. Let us consider an
system completely observable as a single island during normal
optimisation problem that consists of seven factors, say,
operating conditions [25].
U i ; i ¼ 1; 2; . . . ; 6; 7 with each factor consist of two levels, say, A
and B. In this example, the ES technique requires a total of
2.6. Effect of zero injection buses
27 ¼ 128 experiments to obtain the optimal combination of these
factors. When there are F number of factors (variables) with Q
In the present work, a method suggested in [29] has been
number of levels for each factor, then a total number of Q F exper-
employed for modelling the zero injection buses that exists in
iments (possible combinations) are required to be performed by ES
the power system network. A bus is said to be zero injection bus
technique in order to obtain optimal combinations of the factors
if the bus has neither a load nor a generator, i.e. net power injection
[30]. Hence, it is very difficult to perform ES technique when these
at this bus is zero. These ZIBs can be used as pseudo measurements
values are large. On the other hand, for the above example an
to make the power system network observable with less number of
orthogonal array requires only 32 experiments to obtain the opti-
PMUs due to the following conditions:
mal solution. An OA with F factors with each having Q levels is
(a) If the ZIBs together with all its incident buses are observable symbolised as LM ðQ F Þ, where, L indicates a Latin square, M repre-
except one, then the unobservable bus can also be sents the number of experiments. In general, M is much smaller
made observable by applying Kirchhoff’s current law (KCL) than the total number of combinations, i.e. Q F [21]. For example,
at ZIB. an orthogonal array with seven factors and two levels is symbol-
(b) If all the adjacent buses to the ZIB are observable then the ised as L32 ð27 Þ: This L32 ð27 Þ orthogonal array is given in Appendix.
ZIB can be made observable by performing the nodal analy- The way the OAs are constructed and some of the readily available
sis at ZIB. OAs can be found in [31].
(c) If there exist a group of adjacent ZIBs which are unobserv-
able and all the buses adjacent to ZIBs are observable then
these ZIBs can also be made observable by executing the 3.2. Signal-to-noise ratio
nodal analysis at these buses.
The second important tool in the Taguchi method is the signal-
As a result of above three conditions, the effect of ZIBs can be to-noise ratio. The main function of SNR is to measure the quality
modelled as: of the solution and to find the suitable level for each variable that
improves the quality of the solution. Depending on the types of
X
N X
k
characteristics, various types of SNRs are available, namely, (i)
fi ¼ Aij xj þ Ai;zl si;zl P 1 ð9Þ
j¼1 l¼1
continuous or discrete, (ii) smaller-the-better (STB) or larger-
the-better (LTB), and (iii) nominal-is-best. For both STB and LTB
Pn Pn
where, si;zl represents the only unobservable bus among the ZIB characteristics, the SNR is calculated using i¼1 1=J i and i¼1 J i
along with its incident buses with value one. It is to be noted that respectively. Here, J i is the fitness value of the ith level in an exper-
similar procedure can be implemented for OPP problem by consid- iment run and n indicates the number of variables. The value with
ering power injection measurements. large SNR is the most favourable solution [18].
12 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

4. Proposed Taguchi binary bat algorithm Table A.1. Now, let us consider the following two positional vectors
are selected randomly from the initial population, each consisting
The binary bat algorithm (BBA) is a new heuristic optimisation of seven factors which correspond to ui ði ¼ 1; 2; . . . ; 6; 7Þ for per-
technique based on the echolocation behaviour of bats [32,33]. forming the various matrix experiments of an OA. The fitness of
This echolocation works on the principle of natural sonar which these positional vectors has been found to be 4 and 3 respectively.
is used by the bats for navigating and hunting purposes. The three
important characteristics of bats when detecting prey have been ulev el1 ¼ ½1; 1; 0; 0; 1; 1; 0 ð13Þ
employed in designing the basic structure of bat algorithm
[34,35]. These characteristics are listed as follows. ulev el2 ¼ ½0; 1; 0; 0; 1; 0; 1 ð14Þ
Subsequently, Table 1 is developed by allocating the level 1 val-
(i) All the bats utilises the echolocation behaviour to sense the
ues with the values given in (13) and the level 2 values with the
distance and to differentiate between the prey and the back-
values given in (14), i.e. A and B in Table A.1 are replaced with
ground barriers.
the corresponding value of ulev el1 and ulev el2 for each factor
(ii) All the bats fly randomly with velocity v i at position xi . These
respectively. The values of the experiment #4 in Table A.1, i.e.
bats emit pulses with a fixed frequency f min , varying wave-
[A A A B B B B] are replaced with the values [1 1 0 0 1 0 1] respec-
length k, and loudness A0 . The wavelength of the emitted
tively for each factor shown in Table 1. Then, for each experiment
pulses and the rate of pulse emission r 2 ½0; 1 are adjusted
in Table 1, the fitness is calculated according to (4) and SNR is
automatically depending upon the proximity of their target.
calculated as the inverse of the objective function value [18,36].
(iii) The loudness A0 is assumed to be varied from a large positive
In experiment #4 the objective function value and SNR is found
value to a minimum constant value Amin .
to be 4 and 0.2500 respectively and shown in Table 1. Once all the
experiments are performed, the effect of each factor f on the objec-
In the present paper, a novel optimisation technique is pro-
posed by applying Taguchi method to improve the initial popula- tive function for each level l is computed in the following way:
tion of BBA and to obtain the multiple optimal solutions. The X
application of proposed TBBA technique to OPP problem is dis- Sfl ¼ SNRlfi experiment i
i
cussed in the following subsections.
where, l signifies the level number, f represents the factor name, i
4.1. Generation of initial population denotes the experiment number, and Sf 1 and Sf 2 indicate the appro-
priateness of level-1 or level-2 for a factor respectively. For exam-
Alike all other meta-heuristic techniques, BBA also begins by ple, in order to select suitable level for the factor u1 then the
randomly generating an initial population P with a certain number corresponding S11 and S12 are calculated as follows:
of individuals called position vectors np with dimension size D.
0 1 S11 ¼ ½0:2500 þ 0:2000 þ 0:3333 þ 0:2500 þ 0:2500 þ 0:3333
x11 ... x1D
B . .. .. C þ 0:2000 þ 0:2500 þ 0:2500 þ . . .
P¼B
@ .
. . . C A ð10Þ
. . . 0:3333 þ 0:2000 þ 0:2500 þ 0:2500 þ 0:2000
xnp 1    xn p D
þ 0:3333 þ 0:2500 ¼ 4:1333
In the present paper, number of buses represents the dimension
size of the problem, i.e. for an N bus system the size of each posi- S12 ¼ ½0:3333 þ 0:5000 þ 0:2500 þ 0:3333 þ 0:3333 þ 0:2500
tion vector will be N. Therefore, the dimension size of P is np  N.
þ 0:5000 þ 0:3333 þ 0:3333 þ . . .
. . . 0:2500 þ 0:5000 þ 0:3333 þ 0:3333 þ 0:5000
4.2. Enhancement of initial population using Taguchi method
þ 0:2500 þ 0:3333 ¼ 5:6667
After initialising, Taguchi method is performed on the initial For any factor, if Sf 1 is larger than Sf 2 , then level-1 is selected as
population to exploit the area of potential solutions. Consequently, the optimal level for that particular factor and vice versa. For
the positional vector of BBA has an optimal solution or a near opti- example, the optimal level for factor u1 is level-2, i.e. ‘B’ because
mal solution. Thus, the proposed TBBA method enhances the effi- S11 < S12 . Similarly the optimal level for all the factors are com-
cacy (in terms of convergence speed, reliability, and robustness) puted and shown in Table 1. The obtained suitable level for each
of the conventional BBA technique. factor based on maximum SNR is [B A A A A B A]. Then correspond-
After obtaining the initial population, Taguchi method is per- ing best positional vector is [0 1 0 0 1 0 0]. This positional vector
formed using two level OA and SNR. To perform two-level OA, with objective function value 2 is the optimal solution for the cor-
for each positional vector another positional vector is randomly responding positional vector. Here, it can be observed that the
selected from the population. Let these two chromosomes be indi- objective function value after performing the Taguchi method is
cated as ulev el1 and ulev el2 respectively [18,36]. As stated previously, better than the objective function values of original positional vec-
the number of buses represents the number of factors (variables) of tors. Thus, the Taguchi method is an efficient approach in obtaining
OA. Hence, number of factors for an N bus system is N. the best or near best solution by executing less number of experi-
ulev el1 ¼ ½u1 ; . . . ; uN  ð11Þ ments. This process is repeated for each positional vector. The
resultant population known as Taguchi population will act as
ulev el2 ¼ ½u1 ; . . . ; uN  ð12Þ initial population for binary bat algorithm [18,36].

In order to execute various set of experiments a two-level OA

with N factors has been selected to obtain the best level or near 4.3. Evaluation
best level for each factor. This can be better explained with an
example. Consider a seven bus system [4] to explain the applica- The fitness of each positional vector is calculated using the
tion of Taguchi method to OPP problem. For which, a two-level objective function given in (4) on the current population. The cur-
OA (L32) with seven factors has been selected which is given in rent best solution is stored and updated [32,33].
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 13

Table 1
Generation of better positional vector from two positional vectors using the Taguchi method.

Exp no Factors Objective function value Signal-to-noise ratio

U1 U2 U3 U4 U5 U6 U7
1 1 1 0 0 1 1 0 4 0.2500
2 1 1 0 0 1 1 1 5 0.2000
3 1 1 0 0 1 0 0 3 0.3333
4 1 1 0 0 1 0 1 4 0.2500
5 1 1 0 0 1 0 1 4 0.2500
6 1 1 0 0 1 0 0 3 0.3333
7 1 1 0 0 1 1 1 5 0.2000
8 1 1 0 0 1 1 0 4 0.2500
9 1 1 0 0 1 0 1 4 0.2500
10 1 1 0 0 1 0 0 3 0.3333
11 1 1 0 0 1 1 1 5 0.2000
12 1 1 0 0 1 1 0 4 0.2500
13 1 1 0 0 1 1 0 4 0.2500
14 1 1 0 0 1 1 1 5 0.2000
15 1 1 0 0 1 0 0 3 0.3333
16 1 1 0 0 1 0 1 4 0.2500
17 0 1 0 0 1 0 1 3 0.3333
18 0 1 0 0 1 0 0 2 0.5000
19 0 1 0 0 1 1 1 4 0.2500
20 0 1 0 0 1 1 0 3 0.3333
21 0 1 0 0 1 1 0 3 0.3333
22 0 1 0 0 1 1 1 4 0.2500
23 0 1 0 0 1 0 0 2 0.5000
24 0 1 0 0 1 0 1 3 0.3333
25 0 1 0 0 1 1 0 3 0.3333
26 0 1 0 0 1 1 1 4 0.2500
27 0 1 0 0 1 0 0 2 0.5000
28 0 1 0 0 1 0 1 3 0.3333
29 0 1 0 0 1 0 1 3 0.3333
30 0 1 0 0 1 0 0 2 0.5000
31 0 1 0 0 1 1 1 4 0.2500
32 0 1 0 0 1 1 0 3 0.3333
Sf1 4.1333 4.9000 4.9000 4.9000 4.9000 4.1333 5.6667
Sf2 5.6667 4.9000 4.9000 4.9000 4.9000 5.6667 4.1333
Optimal contribution values 0 1 0 0 1 0 0 2

(  
if rand < V v ki ðt þ 1Þ
1
4.4. Parameter updating ðxki ðtÞÞ
xki ðt þ 1Þ ¼   ð18Þ
xki ðtÞ rand P V v ki ðt þ 1Þ
As pointed in previous subsection, the frequency and velocity
parameters of BBA are iteratively updated as
F i ¼ F min þ bðF max  F min Þ ð15Þ where, xki ðtÞ and v ki ðtÞ indicate the position and velocity of ith par-
 1
ticle at iteration t in kth dimension, xki ðtÞ denotes the comple-
V i ðt þ 1Þ ¼ V i ðtÞ þ F i ðX i ðtÞ  GbestÞ ð16Þ ment of xki ðtÞ, and rand signifies a uniform distributed random
where, F i and V i ðt þ 1Þ indicate the frequency and the velocity of bat variable generated within the interval [0,1].
i respectively, F min and F max denote the minimum and maximum
range of frequency with values 0 and 2 respectively, b signifies
the uniform distributed random variable generated within in the 4.6. Population updating
interval [0,1], X i symbolises the position vector of bat i, and Gbest
indicates the best solution obtained so far that gives the optimal fit- After updating the position of each bat, the fitness of each posi-
ness value [32,33]. tion vector is evaluated. Now, the fitness of each position vector is
compared with the fitness of position vector previously generated
4.5. Generation of candidate solution [32,33]. The most fitted position vectors will act as population for
next generation based on the loudness of the bat. This is repre-
In the BA, each artificial bat updates its position so as to find its sented mathematically as:
prey/food by adding the velocities to its positions. However, in BBA
updating the position means flipping the positions from ‘‘0” to ‘‘1” (
_ _
or vice versa based on the velocity of the positional vectors (bats). x i ; if fit i < fiti and rand < A
Hence, the authors in [32] have suggested a v-shaped transfer xi ¼ ð19Þ
xi ; otherwise
function to change the position of the bats based on the probability
of their velocity. This v-shaped transfer function offers a high prob-
ability of changing the position for a large absolute velocity and where, A is loudness of the bat. This control variable varies from 0 to
vice versa [33]. This is mathematically formulated as 1 and controls the acceptability of the improved position vectors.
     This process is iteratively repeated until the number of iterations
V v ki ðtÞ ¼ 2=p  tan1 p=2  v ki ðtÞ  ð17Þ
reaches the maximum number of generations.
14 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

5. Implementation of the proposed TBBA for optimal placement Step 6: Repeat steps 3–5 is repeated until number of positional
of PMUs vectors in Taguchi population is equal to the population size.
Step 7: In this step, frequency is adjusted and the velocity of
The complete flow of the proposed TBBA technique is given in each individual in the population is updated.
Fig. 1. The various steps in the procedure are enumerated below. Step 8: Calculate the transfer function value and update the
position of each individual.
Step 1: Initialise the control parameters of bat algorithm such Step 9: Select global best solution among the best solutions if
as initial population, pulse rate, and loudness. the random number generated is greater than the pulse rate.
Step 2: Select suitable two-level orthogonal array and SNR ratio Further, obtain the local solution around the global best
for enhancing the initial population using Taguchi optimisation solution.
method. Step 10: Generate the new solution by flying randomly.
Step 3: Select two position vectors from the initial population Step 11: Evaluate fitness of each individual.
and calculate the fitness for various set of experiments in Step 12: In this step, new solutions are accepted. Pulse rate and
orthogonal array. loudness are updated if the random number generated is less
Step 4: Perform SNR ratio on objective function value obtained than the loudness and fitness of new solutions is less than the
using OA. global best solution.
Step 5: Now the effect of each factor on objective function for Step 13: Ranking of individual solutions based upon their fit-
each level is calculated and a new position vector is obtained. ness value is carried out in this step and the global best solution
is selected accordingly.

Fig. 1. Flow chart of optimal PMU placement problem using TBBA technique.
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 15

Step 14: Steps 7–13 are repeated until number of iterations solutions in a single experimental run. For instance, in case of IEEE
reaches maximum value. 14-bus system the proposed technique provides same number of
PMUs, i.e. 4 in comparison with the methods proposed in
6. Results and discussion [1,2,25,38] so as to obtain complete system observability. On the
other hand, the main advantage of the proposed method compared
The proposed TBBA technique determines the optimal number to [1,2,25,38] is that the proposed approach provides multiple
of strategic PMU bus locations for complete observability of the solutions with same number of PMUs, however, with different
power system [2]. The proposed method has been applied on five placement locations, i.e. [2, 6, 7, 9], [2, 6, 8, 9], and [2, 7, 10, 13].
different standard systems, viz. IEEE 14-bus, IEEE 30-bus, IEEE This shows the higher reliability of the proposed technique.
57-bus, IEEE 118-bus, and Polish 2383-bus systems [37]. In the
present work, for all the test systems the initial population, pulse 6.2. Test case 2. Loss of single PMU
rate, and loudness control parameters values are initialised as 60,
0.025, and 8 respectively. The performance of the proposed algo- In this case study, the effect of loss of single PMU has been
rithm has been compared with the other techniques proposed in investigated. The loss of PMU measurements not only makes the
the literature. The comparison is performed with respect to the bus unobservable at which PMU is installed, but also the
computational time and optimal number of PMUs required for neighbouring buses (the buses that are indirectly observable)
complete observability. To validate the efficiency and accuracy of unobservable. For example, consider a PMU placement location
the proposed method, six different scenarios have been considered set {2, 7, 10, 13} of IEEE 14-bus system under normal operating
to solve the OPP problem. condition given in Table 2. This PMU placement location set can
measure 12 current phasor measurements in the branches con-
6.1. Test case 1. Normal operating condition nected to these buses. From this placement location set, let us
assume an unexpected failure of a PMU has occurred on bus 7.
In this case study, observability analysis has been performed The loss of PMU at bus 7 not only makes bus 7 to be unobservable,
under normal operating condition. The optimal number of PMUs but also its adjacent buses {4, 8, 9} unobservable. However, in the
and corresponding placement locations are obtained using the pro- present case only buses 7 and 8 are unobservable. This is because
posed method for different test systems. Results thus obtained are the buses 4 and 9 are observed using PMUs 2 and 10 respectively.
given in Table 2. From Table 2 it can be observed that the present In order to make these two buses, i.e. buses 7 and 8 to be observ-
method provides multiple placement locations with the same able an additional PMU is required to be placed at bus 8. Hence, in
number of PMUs. Optimal placement location can be selected order to make the system completely observable under loss of any
based upon the applications such as state estimation accuracy single PMU additional PMUs are needed to be placed in the power
and monitoring of tie-lines. Table 3 provides the comparative anal- system. For the test systems under consideration placement loca-
ysis of the test results obtained using the proposed technique with tions are given in Table 4. From Table 4, it can be seen that similar
other evolutionary and conventional methods suggested in to case 1 the proposed technique provides multiple solutions with
[1,2,4,22,24,25,38]. It can be observed that all the methods provide same number of PMUs. Table 5 shows the comparative results
the same minimal number of PMUs for complete observability. obtained by other techniques suggested in [22,24,25]. It can be
However, in comparison the proposed technique provides multiple observed that the proposed method provides similar or better

Table 2
Optimal placement of PMUs under normal condition.

System Optimal number of PMUs Optimal locations of PMUs

IEEE 14-bus 4 2, 6, 7, 9
2, 6, 8, 9
2, 7, 10, 13
IEEE 30-bus 10 3, 6, 7, 10, 11, 12, 15, 20, 25, 27
2, 3,6, 9, 10, 12, 18, 23, 26, 30
1, 5, 10, 11, 12, 18, 24, 26, 27, 28
1, 6, 7, 10, 11, 12, 18, 23, 25, 29
IEEE 57-bus 17 1, 5, 9, 15, 19, 22, 26, 29, 30, 32, 36, 38, 41, 47, 51, 53, 57
1, 6, 12, 15, 19, 22, 25, 27, 32, 36, 41, 44, 47, 50, 52, 55, 57
1, 4, 6, 13, 20, 22, 25, 27, 29, 32, 36, 39, 41, 45, 47, 51, 54
1, 4, 9, 20, 23, 25, 26, 29, 32, 36, 39, 41, 45, 46, 48, 50, 54
IEEE 118-bus 32 1, 6, 10, 11, 12, 17, 21, 23, 28, 30, 34, 37, 42, 45, 49, 53, 56, 62, 64, 71, 75, 77, 80, 85, 87, 90, 94, 101, 105, 110, 114, 116
1, 5, 10, 12, 15, 17, 21, 23, 28, 30, 36, 40, 44, 46, 50, 51, 54, 62, 64, 71, 75, 77, 80, 85, 87, 91, 94, 101, 105, 110, 114, 116
Polish 2383- 746 –
bus⁄
⁄
Optimal PMU locations of Polish 2383-bus system is not provided due to large number.

Table 3
Comparison of optimal number of PMUs of the proposed method with other methods under normal condition.

Method IEEE 14-bus IEEE 30-bus IEEE 57-bus IEEE 118-bus

[1] 4 10 – –
[2] 4 10 17 32
[25] 4 10 17 32
[38] 4 10 17 32
Proposed 4 10 17 32
16 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

Table 4
Optimal placement of PMUs for single PMU loss condition.

System Optimal Optimal locations of PMUs

number
of PMUs
IEEE 14-bus 9 2, 4, 5, 6, 7, 8, 9, 11, 13
1, 2, 4, 6, 7, 8, 9, 10, 13
2, 3, 5, 6, 7, 8, 9, 11, 13
1, 2, 4, 6, 7, 8, 9, 11, 13
IEEE 30-bus 21 2, 3, 4, 5, 6, 9, 10, 11, 12, 13, 15, 17, 18, 19, 21, 24, 25, 26, 27, 28, 30
1, 3, 5, 6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 18, 19, 21, 24, 25, 26, 29, 30
1, 2, 4, 6, 7, 9, 10, 11, 12, 13, 15, 16, 18, 19, 21, 24, 25, 26, 28, 29, 30
1, 2, 3, 6, 7, 8, 9, 10, 11, 12, 13, 15, 16, 18, 19, 21, 23, 25, 26, 27, 30
IEEE 57-bus 33 1, 2, 4, 6, 9, 12, 15, 19, 20, 22, 24, 25, 27, 28, 29, 31, 32, 33, 35, 36, 37, 38, 39, 41, 43, 45, 46, 47, 50, 51, 53, 54, 56
1, 3, 4, 6, 9, 12, 15, 19, 20, 22, 24, 25, 26, 28, 29, 31, 32, 33, 35, 36, 37, 38, 41, 43, 44, 46, 47, 50, 51, 53, 54, 56, 57
1, 3, 4, 6, 9, 11, 12, 15, 19, 20, 22, 24, 25, 26, 28, 29, 30, 32, 33, 35, 36, 37, 38, 41, 45, 46, 47, 50, 51, 53, 54, 56, 57
1, 2, 4, 6, 9, 12, 15, 19, 20, 22, 24, 26, 28, 29, 30, 31, 32, 33, 35, 36, 38, 39, 41, 43, 45, 46, 47, 50, 51, 53, 54, 56, 57
IEEE 118 bus 68 2, 3, 5, 6, 9, 10, 11, 12, 15, 17, 19, 21, 22, 26, 27, 28, 29, 30, 32, 34, 35, 37, 40, 41, 44, 45, 46, 49, 52, 53, 56, 57, 58, 59, 62, 63, 65,
67, 68, 70, 71, 72, 73, 75, 77, 78, 80, 83, 85, 86, 87, 89, 91, 92, 94, 96, 100, 101, 105, 106, 108, 110, 111, 112, 114, 116, 117, 118
2, 3, 5, 6, 9, 10, 11, 12, 15, 17, 19, 21, 22, 25, 27, 28, 30, 31, 32, 34, 35, 37, 40, 42, 43, 45, 46, 49, 50, 51, 52, 54, 56, 59, 61, 64, 66, 67,
68, 70, 71, 72, 73, 75, 76, 77, 78, 80, 84, 85, 86, 87, 89, 91, 92, 94, 96, 100, 101, 105, 106, 108, 110, 111, 112, 115, 116, 117
1, 2, 5, 7, 9, 10, 11, 12, 15, 17, 19, 21, 22, 25, 26, 27, 28, 29, 32, 34, 36, 37, 40, 42, 44, 45, 46, 49, 50, 51, 52, 54, 56, 59, 62, 63, 65, 67,
68, 70, 71, 72, 73, 75, 76, 77, 78, 80, 83, 85, 86, 87, 89, 91, 92, 94, 96, 100, 101, 105, 106, 108, 110, 111, 112, 115, 116, 117
Polish 2383-bus⁄ 1681 –
⁄
Optimal PMU locations of Polish 2383-bus system is not provided due to large number.

Table 5
Requirement of optimal number of PMUs in the proposed method as compared to the other methods for single PMU loss condition.

Method IEEE 14-bus IEEE 30-bus IEEE 57-bus IEEE 118-bus

[22] 9 22 35 72
[24] 9 – 33 68
[25] 9 21 33 68
Proposed 9 21 33 68

results in comparison with the other techniques. For example, in 6.4. Test case 4. Lack of communication infrastructure
case of IEEE 118-bus system the proposed technique along with
[24,25] require 68 PMUs to make complete system observability Optimal placement locations and minimal number of PMUs
as compared to 72 in case of ILP technique suggested in [22]. required for power system observability by considering the effect
of lack of communication infrastructure at buses (substations) is
6.3. Test case 3. Unequal cost of PMUs shown in Table 8. There are four columns in Table 8. Second col-
umn of the table indicates the buses with lack of communication
In this case study, the effect of unequal costs of PMUs on facility for different test systems under consideration. The optimal
optimal location and number has been considered. As stated in number of PMUs and their locations are given in columns 3 and 4
Section 2, for each additional channel the cost of each PMU is of Table 8 respectively. The results thus obtained by applying the
increased by 0.1. Results for different test systems considering proposed TBBA method illustrate that the optimal number of PMUs
the effect of unequal PMU cost are given in Table 6. The compara- required to make the power system completely observable, is
tive results for IEEE 14-bus, IEEE 30-bus, and IEEE 57-bus systems greater or equal to the optimal number of PMUs achieved by ignor-
under equal and unequal cost of PMUs are shown in Table 7. From ing the effect of lack of communication infrastructure [39]. In order
Table 7 it can be seen that the optimal number of PMUs for both to validate the efficacy of the proposed method the results thus
equal and unequal PMU costs is identical, however, the placement obtained are compared with the results provided in [39]. Table 9
locations are different. It can also be observed from Table 7 column shows the comparative results of the proposed and SQP method.
6 that due to the difference in number of measuring channels over- It can be observed from Table 8 and from [39] that both the pro-
all installation cost of PMUs is affected (decreased) by considering posed and sequential quadratic programming (SQP) methods pro-
unequal costs. For example, by considering unequal cost of PMUs vide multiple placement locations with same number of PMUs.
the overall installation cost in case of IEEE 57-bus system results However, the main advantage of the proposed method compared
in 20.1 p.u., in contrast the installation cost of PMUs by considering to SQP method is that the proposed method being the population
equal cost results in 20.5 p.u. This shows the effect of unequal cost based technique provides multiple placement locations in single
of PMUs on optimal placement locations. experimental run while the SQP requires multiple experimental

Table 6
Optimal placement of PMUs considering unequal cost.

System Optimal number of PMUs Optimal locations of PMUs

IEEE 14-bus 4 2, 8, 10, 13
IEEE 30-bus 10 3, 5, 8, 10, 11, 12, 18, 23, 26, 30
IEEE 57-bus 17 1, 4, 7, 9, 13, 20, 23, 27, 30, 32, 36, 39, 41, 44, 47, 51, 53
IEEE 118 bus 32 1, 5, 9, 12, 15, 17, 20, 23, 28, 30, 36, 40, 43, 46, 50, 51, 54, 62, 64, 68, 71, 75, 77, 80, 85, 86, 90, 94, 102, 105, 110, 114
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 17

Table 7
Comparison of optimal number and locations of PMUs of the proposed method under equal and unequal costs of PMUs.

System Optimal Optimal location of PMUs Number of Installation

number of measuring cost (p.u.)
PMUs channels
IEEE 14-bus Equal cost 4 2, 6, 7, 9 15 5.1
Unequal cost 4 2, 8, 10, 13 10 4.6
IEEE 30-bus Equal cost 10 3, 6, 7, 10, 11, 12, 15, 20, 25, 27 36 12.6
Unequal cost 10 3, 5, 8, 10, 11, 12, 18, 23, 26, 30 25 11.5
IEEE 57-bus Equal cost 17 1, 5, 9, 15, 19, 22, 26, 29, 30, 32, 36, 38, 41, 47, 51, 53, 57 52 20.5
Unequal cost 17 1, 4, 7, 9, 13, 20, 23, 27, 30, 32, 36, 39, 41, 44, 47, 51, 53 48 20.1
IEEE 118 bus Equal cost 32 1, 6, 10, 11, 12, 17, 21, 23, 28, 30, 34, 37, 42, 45, 49, 53, 56, 62, 64, 71, 75, 77, 117 40.5
80, 85, 87, 90, 94, 101, 105, 110, 114, 116
Unequal cost 32 1, 5, 9, 12, 15, 17, 20, 23, 28, 30, 36, 40, 43, 46, 50, 51, 54, 62, 64, 68, 71, 75, 114 40.2
77, 80, 85, 86, 90, 94, 102, 105, 110, 114

Table 8
Optimal placement of PMUs considering lack of communication infrastructure.

System Lack of CI bus Optimal Optimal locations of PMUs

number of
PMUs
IEEE 14-bus 2, 9 5 1, 3, 8, 10, 13
4, 5, 8, 11, 13
1, 4, 7, 10, 13
4, 5, 7, 11, 13
IEEE 30-bus 10, 14, 30 11 3, 5, 11, 12, 17, 19, 21, 23, 25, 28, 29
3, 5, 9, 12, 15, 16, 20, 22, 25, 28, 29
2, 3, 6, 11, 12, 15, 17, 20, 22, 26, 29
2, 4, 6, 11, 12, 16, 19, 22, 23, 26, 29
IEEE 57-bus 1, 9, 25, 38, 54 18 2, 6, 10, 12, 19, 22, 24, 28, 31, 32, 36, 39, 41, 45, 46, 49, 52, 55
2, 6, 12, 15, 19, 22, 24, 28, 31, 32, 36, 39, 41, 44, 47, 50, 52, 55
2, 6, 12, 15, 19, 22, 26, 29, 30, 32, 36, 39, 41, 44, 47, 50, 53, 55
2, 6, 12, 19, 22, 26, 27, 30, 32, 36, 41, 45, 46, 48, 50, 52, 55, 57
IEEE 118 bus 3, 7, 9, 11, 12, 17 35 2, 5, 6, 10, 15, 16, 19, 22, 26, 27, 31, 32, 34, 37, 42, 45, 49, 52, 56, 62, 63, 68, 70,
71, 76, 79, 84, 86, 89, 92, 96, 100, 105, 110, 117
1, 5, 6, 10, 15, 16, 19, 22, 23, 26, 27, 28, 32, 34, 37, 41, 45, 49, 53, 56, 62, 64, 68,
71, 75, 77, 80, 85, 86, 91, 94, 102, 105, 110, 117
1, 5, 6, 10, 15, 16, 18, 21, 27, 29, 30, 32, 34, 36, 40, 45, 49, 52, 56, 62, 63, 68, 70,
71, 76, 79, 85, 87, 89, 92, 96, 100, 105, 110, 117
1, 4, 6, 10, 15, 16, 19, 22, 27, 30, 31, 32, 36, 40, 43, 46, 50, 51, 54, 62, 64, 68, 70,
71, 78, 84, 86, 89, 92, 96, 100, 105, 110, 117, 118
Polish 2383-bus⁄ 6, 16, 90, 195, 568, 978, 1017, 1564, 1756, 752 –
1939, 2167, 2296, 2374, 2380
⁄
Optimal PMU locations of Polish 2383-bus system is not provided due to large number.

Table 9
Comparison of optimal number of PMUs of the proposed method with other method for lack of communication infrastructure condition.

Method IEEE 14-bus IEEE 30-bus IEEE 57-bus IEEE 118-bus

[39] 5 – – 35
Proposed 5 11 18 35

runs with different initial points. This shows the effectiveness of umns 3 and 4 of Table 10 respectively. It can be seen from
the proposed method compared to SQP method. Table 10 column 3 that in the presence of conventional power flow
measurements the optimal number of PMUs required to make
6.5. Test case 5. Inclusion of conventional measurements complete system observability decreases in comparison with the
absence of conventional measurements.
The presence of traditional power flow measurements is consid-
ered in the power system network in this case. Table 10 presents 6.6. Test case 6. Channel limitation
the results obtained using the proposed technique for different test
systems. Table 10 consists of four columns. Second column of the In this case study, the OPP problem has been solved by consid-
table indicates the placement locations of conventional measure- ering the effect of channel limitation. Test case is solved by assum-
ments that already exist in the power system network. These con- ing predefined number of channel capacity of PMUs, i.e. 1, 2, 3, and
ventional measurements are randomly distributed such that there 4. Given a PMU with the channel capacity one, the PMU installed at
exist some unobservable islands. Hence, the proposed method is a bus can only measure one voltage phasor and one current phasor
executed to ensure complete system observability under normal measurements in any one of the branches connected to the
operating condition. The optimal number of PMUs and their loca- installed bus. For example, a PMU installed at bus 3 of IEEE 14-
tions thus obtained using the proposed method are given in col- bus system can measure voltage phasor at bus 3 and current
18 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

Table 10
Optimal placement of PMUs considering the effect of power flow measurements.

System Flow measurements Optimal Optimal locations of PMUs

number of
PMUs
IEEE 14-bus 1–2, 7–8, 9–10, 9–14 2 4, 6
IEEE 30-bus 1–3, 6–7, 6–8, 9–11, 12–13, 16–17, 18–19, 24–25, 25–26 4 2, 10, 15, 27
IEEE 57-bus 6–7, 15–45, 22–38, 24–25, 30–31, 34–35, 36–40, 37–38, 38–44, 39–57, 41–43, 50–51, 53–54 11 1, 4, 9, 20, 24, 27, 32, 46, 49, 52, 56
IEEE 118-bus 1–3, 5–6, 11–13, 16–17, 20–21, 22–23, 23–25, 27–28, 29–31, 34–43, 35–36, 41–42, 44–45, 18 9, 10, 11, 19, 26, 32, 37, 49, 59, 61, 70,
46–48, 50–57, 51–52, 53–54, 56–58, 60–62, 65–66, 66–67, 68–81, 71–73, 75–118, 76–77, 71, 80, 85, 92, 105, 110, 116
77–82, 78–79, 86–87, 90–91, 95–96, 100–101, 114–115

phasor either in branch {3–2} or in branch {3–4}. Thus, a PMU with of zero injection buses has been tabulated in Table 13. There are
channel capacity of one can only make two buses observable, i.e. three columns in Table 13. Second and third column of Table 13
either buses 3 and 2 are observable or buses 3 and 4 are observable indicates the optimal number of PMUs and their placement loca-
in this case. Hence, under single channel capacity condition the tions respectively. The results thus obtained by applying the pro-
IEEE 14-bus, IEEE 30-bus, and IEEE 57-bus systems require 7, 15, posed TBBA method illustrate that the optimal number of PMUs
and 29 PMUs in order to make the system completely observable. required to make the power system completely observable, is
The results obtained for the IEEE 14-bus, IEEE 30-bus, and IEEE 57- found to be less than the optimal number of PMUs obtained by
bus systems are shown in Table 11. It can be observed from ignoring the effect of ZIBs. Further, it can be observed from Table 13
Table 11 that as the channel capacity of PMUs increases the opti- that the proposed method provides multiple placement locations
mal number of PMUs required to be installed in the power system with same number of PMUs. In order to validate the efficacy of
network decreases. Furthermore, the results thus obtained by the the proposed method the results thus obtained are compared with
proposed technique have been compared with the results obtained the results provided in the literature in Table 14. It can be seen
in [26,27]. It can be observed from Table 12 that all the methods from Table 14 column 4 that in case of IEEE 118 bus system the
provide same minimal number of PMUs for complete system proposed method requires only 28 PMUs to make the system com-
observability. pletely observable while other methods [3,4,7,9,23,24,39] require

6.7. Test case 7. Effect of zero injection buses Table 14

Comparison of optimal number and locations of PMUs by considering the effect of
Optimal placement locations and minimal number of PMUs ZIBs.
required for power system observability by considering the effect Method IEEE 14-bus IEEE 30-bus IEEE 118-bus

Table 11 Proposed 3 7 28
Optimal placement of PMUs considering the channel limitation condition. [3] 3 7 29
[4] 3 7 29
Channel limit IEEE 14-bus IEEE 30-bus IEEE 57-bus [7] 3 7 29
[9] 3 – 29
1 7 15 29
[22] 3 8 28
2 5 11 19
[23] – 7 29
3 4 10 17
[24] 3 – 29
4 4 10 17
[39] 3 – 29

Table 12
Comparison of optimal number of PMUs of proposed method with other methods considering the channel limitation condition.

Channel limit IEEE 14-bus IEEE 30-bus IEEE 57-bus

Proposed method [26] [27] Proposed method [26] [27] Proposed method [26] [27]
1 7 7 7 15 15 15 29 29 29
2 5 5 5 11 11 11 19 19 19
3 4 4 4 10 10 10 17 17 17
4 4 4 4 10 10 10 17 17 17

Table 13
Optimal placement of PMUs considering the effect of ZIBs.

System Optimal number of PMUs Optimal locations of PMUs

IEEE 14-bus 3 2, 6, 9
IEEE 30-bus 7 1, 2, 10, 12, 15, 20, 27
2, 4, 10, 12, 18, 24, 29
2, 4, 10, 12, 19, 23, 27
3, 7, 10, 12, 18, 24, 29
IEEE 118-bus 28 3, 6, 8, 12, 15, 17, 21, 25, 29, 34, 40, 45, 49, 53, 56, 62, 72, 75, 77, 80, 85, 86, 90, 94,101,105, 110, 114
1, 8, 11, 12, 17, 21, 25, 28, 34, 37, 41, 45, 49, 53, 56, 62, 72, 75, 77, 80, 86, 87, 91, 94,102,105, 110, 114
2, 10, 11, 12, 17, 21, 25, 28, 34, 36, 40, 45, 49, 52, 56, 62, 72, 75, 77, 80, 85, 86, 90, 94,101,105, 110, 114
2, 6, 9, 12, 15, 17, 21, 25, 29, 34, 40, 45, 49, 53, 56, 62, 72, 75, 77, 80, 85, 87, 91, 94,101,105, 110, 114
Polish 2383-bus⁄ 553 –
⁄
Optimal PMU locations of Polish 2383-bus system is not provided due to large number.
 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20 19

Table 15
Comparison of computational time(s) for obtaining optimal number of PMUs using the proposed method with other methods under normal condition.

Technique IEEE 14-bus IEEE 30-bus IEEE 57-bus IEEE 118-bus

[2] 1.3 s 2.6 h – –
[7] 0.444 s 1.483 s 15.462 s 75.219 s
[12] 1 min 6 min 43 min 85 min
[13] 0.162 s 1.121 s 10.705 s 53.122 s
[40] 0.05 s – 30.41 s 1625.30 s
[41] 0.298 s 18.43 s 345.04 s –
[42] 1.114 s 11.366 s 128.388 s 572.610 s
[43] 0.205 s 2.205 s 21.402 s 187.144 s
Proposed 0.032 s 0.0706 s 1.3103 s 20.5334 s

29 PMUs. This shows the effectiveness of the proposed method in and Polish 2383-bus test systems. The performance of the pro-
obtaining the optimal number of PMUs. posed technique has been validated under different conditions,
Further, computational time taken by the proposed method in namely, normal condition, loss of single PMU, effect of unequal
obtaining the optimal solution under normal condition is given PMU cost, lack of communication infrastructure, presence of con-
in Table 15 for different systems under consideration. From ventional power flow measurements, effect of channel limitation,
Table 15, it can be seen that the computational time taken to pro- and effect of zero injection buses. Comparison of results obtained
vide minimal number of PMUs by the proposed method (shown in using the proposed method has been performed with conventional
bold in the table) is minimal in comparison with other evolution- and evolutionary techniques. Unlike conventional techniques, the
ary and conventional techniques [2,7,13,40–43]. For instance, in proposed method works with set of solutions. This distinctive fea-
case of IEEE 57-bus system the proposed technique gives a compu- ture of the proposed approach alleviates the necessity of multiple
tation time of 1.3103 s as compared to 15.462 s, 10.705 s, 30.41 s, executions of OPP with different starting points to obtain multiple
345.04 s, 128.388 s, and 21.402 s in case of [7,13,40–43] respec- global optimal solutions with same minimal number of PMUs.
tively. This shows the computational superiority of the proposed Another potential advantage of the proposed TBBA method is that
method compared to other techniques in obtaining the optimal it provides these multiple global optimal solutions with lower
number of PMUs. computational time in comparison with other evolutionary and
conventional approaches. Therefore, the proposed TBBA approach
can be utilised in various applications such as state estimation,
7. Conclusions and identification of fault location to obtain minimum number of
PMUs under various conditions. Further, investigations in this area
In this paper, a new hybrid Taguchi binary bat algorithm has can be accomplished by considering the effect of loss of multiple
been proposed for optimal placement of PMUs by ensuring com- PMUs. This shall be the future work of the authors.
plete observability of the power system. The proposed approach
consists of two steps. In the first step, systematic reasoning capa-
bility of the Taguchi method has been utilised to enhance the ini-
tial population by providing optimal or near optimal solution. In Appendix A
the second step, binary bat algorithm has been employed to obtain
multiple optimal solutions by utilising the enhanced population An orthogonal array with seven factors and each factor consist-
provided by the Taguchi method. The efficiency of the proposed ing of 2 levels is symbolised as L32 ð27 Þ: This L32 ð27 Þ orthogonal
TBBA approach has been implemented on different IEEE standard array is given in Table A.1.

Table A.1
Orthogonal array L32(27).

Exp no Factors Objective function value Signal-to-noise ratio

U1 U2 U3 U4 U5 U6 U7
1 A A A A A A A J1 1/J1
2 A A A A B A B J2 1/J2
3 A A A B A B A J3 1/J3
4 A A A B B B B J4 1/J4
5 A A B A A B B J5 1/J5
6 A A B A B B A J6 1/J6
7 A A B B A A B J7 1/J7
8 A A B B B A A J8 1/J8
9 A B A A A B B J9 1/J9
10 A B A A B B A J10 1/J10
11 A B A B A A B J11 1/J11
12 A B A B B A A J12 1/J12
13 A B B A A A A J13 1/J13
14 A B B A B A B J14 1/J14
15 A B B B A B A J15 1/J15
16 A B B B B B B J16 1/J16
17 B A A A A B B J17 1/J17
18 B A A A B B A J18 1/J18
19 B A A B A A B J19 1/J19
20 B A A B B A A J20 1/J20

(continued on next page)

20 V. Basetti, A.K. Chandel / Measurement 95 (2017) 8–20

Table A.1 (continued)

Exp no Factors Objective function value Signal-to-noise ratio

U1 U2 U3 U4 U5 U6 U7
21 B A B A A A A J21 1/J21
22 B A B A B A B J22 1/J22
23 B A B B A B A J23 1/J23
24 B A B B B B B J24 1/J24
25 B B A A A A A J25 1/J25
26 B B A A B A B J26 1/J26
27 B B A B A B A J27 1/J27
28 B B A B B B B J28 1/J28
29 B B B A A B B J29 1/J29
30 B B B A B B A J30 1/J30
31 B B B B A A B J31 1/J31
32 B B B B B A A J32 1/J32
(A) (A) (A) (A) (A) (A) (A)
Sf1 V1 V2 V3 V4 V5 V6 V7
(B) (B) (B) (B) (B) (B) (B)
Sf2 V1 V2 V3 V4 V5 V6 V7
Optimal contribution values B A A B A B A J

References [21] D. Liu, Y. Cai, Taguchi method for solving the economic dispatch problem with
nonsmooth cost functions, IEEE Trans. Power Syst. 20 (2005) 2006–2014.
[22] Optimal Placement of Phasor Measurement Units for State Estimation Power
[1] S. Chakrabarti, E. Kyriakides, Optimal placement of phasor measurement units
Systems, http://www.pserc.org, assessed May 2014.
for power system observability, IEEE Trans. Power Syst. 23 (2008) 1433–1440.
[23] N.H. Abbasy, H.M. Ismail, A unified approach for the optimal PMU location for
[2] B.K. Saha Roy, A.K. Sinha, A.K. Pradhan, An optimal PMU placement technique
power system state estimation, IEEE Trans. Power Syst. 24 (2009) 806–813.
for power system observability, Electr. Power Energy Syst. 42 (2012) 71–77.
[24] D. Dua, S. Dambhare, R.K. Gajbhiye, S.A. Soman, Optimal multistage scheduling
[3] S.M. Mazhari, H. Monsef, H. Lesani, A. Fereidunian, A multi-objective PMU
of PMU placement: an ILP approach, IEEE Trans. Power Delivery 23 (2008)
placement method considering measurement redundancy and observability
1812–1820.
value under contingencies, IEEE Trans. Power Syst. 28 (2013) 2136–2146.
[25] S. Chakrabarti, E. Kyriakides, D.G. Eliades, Placement of synchronized
[4] A. Ahmadi, Y. Alinejad-Beromi, M. Moradi, Optimal PMU placement for power
measurements for power system observability, IEEE Trans. Power Delivery
system observability using binary particle swarm optimization and
24 (2009) 12–19.
considering measurement redundancy, Expert Syst. Appl. 38 (2011) 7263–
[26] M. Korkali, A. Abur, Placement of PMUs with channel limits, in: Proc. IEEE
7269.
Power & Energy Society General Meeting, 2009, pp. 1–4.
[5] A. Mahari, H. Seyedi, Optimal PMU placement for power system observability
[27] R. Kumar, V.S. Rao, Optimal placement of PMUs with limited number of
using BICA, considering measurement redundancy, Electr. Power Syst. Res. 103
channels, in: Proc. IEEE North American Power Symposium (NAPS), 2011, pp.
(2013) 78–85.
1–7.
[6] F.J. Marin, F. Garcia-Lagos, G. Joya, F. Sandoval, Genetic algorithms for optimal
[28] Factors Affecting PMU Installation Costs, https://www.smartgrid.gov, assessed
placement of phasor measurement units in electrical networks, Electron. Lett.
May 2014.
39 (2003) 1–2.
[29] L. Huang, Y. Sun, J. Xu, W. Gao, J. Zhang, Z. Wu, Optimal PMU placement
[7] F. Aminifar, C. Lucas, A. Khodaei, M. Fotuhi-Firuzabad, Optimal placement of
considering controlled islanding of power system, IEEE Trans. Power Syst. 29
phasor measurement units using immunity genetic algorithm, IEEE Trans.
(2014) 742–755.
Power Delivery 24 (2009) 1014–1020.
[30] M. Tsai, H. Gowb, W. Lin, Hybrid Taguchi-Immune algorithm for the thermal
[8] N.M. Manousakis, G.N. Korres, P.S. Georgilakis, Taxonomy of PMU placement
unit commitment, Electr. Power Energy Syst. 33 (2011) 1062–1069.
methodologies, IEEE Trans. Power Syst. 27 (2012) 1070–1077.
[31] W.Y. Fowlkes, C.M. Creveling, Engineering Methods for Robust Product Design:
[9] T.L. Baldwin, L. Mili, M.B. Boisen, R. Adapa, Power system observability with
Using Taguchi Methods in Technology and Product Development, first ed.,
minimal phasor measurement placement, IEEE Trans. Power Syst. 8 (1993)
Prentice Hall, New York, 2012.
707–715.
[32] S. Mirjalili, S.M. Mirjalili, X. Yang, Binary bat algorithm, Neural Comput. Applic.
[10] B. Ramachandran, G.T. Bellarmine, Improving observability using optimal
25 (2014) 663–681.
placement of phasor measurement units, Electr. Power Energy Syst. 56 (2014)
[33] M. Kang, J. Kim, J. Kim, Reliable fault diagnosis for incipient low-speed
55–63.
bearings using fault feature analysis based on a binary bat algorithm, Inf. Sci.
[11] J. Peng, Y. Sun, H.F. Wang, Optimal PMU placement for full network
294 (2015) 423–438.
observability using Tabu search algorithm, Electr. Power Energy Syst. 28
[34] T. Pan, T. Dao, T. Nguyen, S. Chu, Hybrid particle swarm optimization with bat
(2006) 223–231.
algorithm, in: Proc. Int. Conf. on Genetic and Evolutionary Computing, 2014,
[12] M. Hajian, A.M. Ranjbar, T. Amraee, B. Mozafari, Optimal placement of PMUs to
pp. 37–47.
maintain network observability using a modified BPSO algorithm, Electr.
[35] R.Y.M. Nakamura, L.A.M. Pereira, K.A. Costa, D.J. Rodrigues, P. Papa, X.S. Yang,
Power Energy Syst. 33 (2011) 28–34.
BBA: a binary bat algorithm for feature selection, in: Proc. IEEE Int. Conf. on
[13] Z. Miljanic, I. Djurovic, I. Vujosevic, Optimal placement of PMUs with limited
Graphics, Patterns and Images (SIBGRAPI), 2012, pp. 291–297.
number of channels, Electr. Power Syst. Res. 90 (2012) 93–98.
[36] J. Tsai, T. Liu, J. Chou, Hybrid Taguchi-genetic algorithm for global numerical
[14] A. Enshaee, R.A. Hooshmand, F.H. Fesharaki, A new method for optimal
optimization, IEEE Trans. Evol. Comput. 8 (2004) 365–377.
placement of phasor measurement units to maintain full network
[37] Power System Test Archive, http://www.ee.washington.edu/research/pstca,
observability under various contingencies, Electr. Power Syst. Res. 89 (2012)
assessed May 2014.
1–10.
[38] N.M. Manousakis, G.N. Korres, A weighted least squares algorithm for optimal
[15] F. Aminifar, A. Khodaei, M. Fotuhi-Firuzabad, M. Shahidehpour, Contingency-
PMU placement, IEEE Trans. Power Syst. 28 (2013) 3499–3500.
constrained PMU placement in power networks, IEEE Trans. Power Syst. 25
[39] N.P. Theodorakatos, N.M. Manousakis, G.N. Korres, Optimal placement of
(2010) 516–523.
phasor measurement units with linear and non-linear models, Electr. Power
[16] W. Lin, H. Gowa, M. Tsai, An efficient hybrid Taguchi-immune algorithm for
Compon. Syst. 43 (2015) 357–373.
the unit commitment problem, Expert Syst. Appl. 38 (2011) 13662–13669.
[40] R. Kavasseri, S.K. Srinivasan, Joint placement of phasor and power flow
[17] A.R. Yildiz, Hybrid Taguchi-differential evolution algorithm for optimization of
measurements for observability of power systems, IEEE Trans. Power Syst. 26
multi-pass turning operations, Appl. Soft Comput. 13 (2013) 1433–1439.
(2011) 1929–1936.
[18] P. Subbaraj, R. Rengaraj, S. Salivahanan, Enhancement of self-adaptive real-
[41] Q. Li, T. Cui, Y. Weng, R. Negi, F. Franchetti, M.D. Ilic, An information-theoretic
coded genetic algorithm using Taguchi method for economic dispatch
approach to PMU placement in electric power systems, IEEE Trans. Smart Grid
problem, Appl. Soft Comput. 11 (2011) 83–92.
4 (2013) 446–456.
[19] H.M. Hasanien, Design optimization of PID controller in automatic voltage
[42] M. Hajian, A.M. Ranjbar, T. Amraee, A.R. Shirani, Optimal placement of phasor
regulator system using Taguchi combined genetic algorithm method, IEEE
measurement units: particle swarm optimization approach, in: Proc. Int. Conf.
Syst. J. 7 (2013) 825–831.
on Intelligent Systems Application to Power Systems, 2007, pp. 1–6.
[20] H. Yu, C.Y. Chung, K.P. Wong, Robust transmission network expansion
[43] Z. Miljanic, I. Djurovic, I. Vujosevic, Optimal PMU placement using ant colony
planning method with Taguchi’s orthogonal array testing, IEEE Trans. Power
optimization approach, in: Proc. Int. Conf. on UPEC, 2011, 1-D-05.
Syst. 26 (2011) 1573–1580.