Electric Power Systems Research 89 (2012) 110

Contents lists available at SciVerse ScienceDirect

Electric Power Systems Research

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

A new method for optimal placement of phasor measurement units to maintain full network observability under various contingencies
Ali Enshaee, Rahmat Allah Hooshmand , Fariborz Haghighatdar Fesharaki
Department of Electrical Engineering, University of Isfahan, Isfahan, Iran

a r t i c l e

i n f o

a b s t r a c t
The application of phasor measurement units (PMUs) in power systems is increasing because of their advantages such as the capability for online state estimation and improvements in the speed of control, and protection systems. In this paper, we propose a new method using binary integer linear programming for the optimal placement of PMUs to guarantee full observability of a power system as well as maximizing the measurement redundancy. Moreover, the problem of the optimal placement of these units in the case of a single PMU loss or single line outage is investigated. A practical limitation is also considered on the maximum number of PMU channels, in the proposed formulation. In all of the investigations, the effect of zero-injection buses in the power system was considered. The efciency of the proposed method was demonstrated in different conditions. The method was applied to several IEEE standard test systems, i.e., the 14-, 30-, 39-, 57-, and 118-bus systems, and in two very large-scale systems, i.e., 2383- and 2746-bus systems. The simulation results veried the acceptable performance of the proposed method. 2012 Elsevier B.V. All rights reserved.

Article history: Received 4 July 2011 Received in revised form 29 January 2012 Accepted 30 January 2012 Available online 6 March 2012 Keywords: Phasor measurement unit (PMU) Optimal placement Power system observability Measurement security Integer programming

1. Introduction Currently, rapidly increasing power demands, disproportionate growth of power generation and transmission systems, power system restructuring, and other factors have overloaded the existing electrical networks and subsequently decreased the stability margin of these networks [1,2]. In such circumstances, to ensure the stable and proper operation of the system, a precise measurement and monitoring of the system states are required. This monitoring was conventionally performed by utilizing the supervisory control and data acquisition (SCADA) system, in which state estimation is derived based on measurements that are not usually synchronized. Therefore, a steady state or, in the most optimistic situation, a quasi-steady state will be obtained for the power system. Hence, the system operator has no access to the dynamic state of the system, which is required to maintain the system in the normal condition. To overcome this limitation in the SCADA, the wide-area monitoring, protection and control (WAMPAC) system has been employed, in which phasor measurement units (PMUs) are considered the basic components [2]. These units, which are synchronized with clock signals from global positioning system (GPS) satellites, are able to provide synchronized measurements [1,2]. When these units are installed on a system bus, the phasor of the bus voltage

Corresponding author. E-mail addresses: ali.enshaee@gmail.com (A. Enshaee), Hooshmand r@eng.ui.ac.ir (R.A. Hooshmand), fr haghighat@yahoo.com (F.H. Fesharaki). 0378-7796/$ see front matter 2012 Elsevier B.V. All rights reserved. doi:10.1016/j.epsr.2012.01.020

can be measured as well as the phasor of the line currents emanating from that bus. Hence, the voltage phasor of adjacent buses can be calculated using Kirchhoffs laws in the steady-state condition. Therefore, it is not necessary to install these units on all of the system buses to control or estimate the system states [2,3]. Furthermore, installation of the PMUs on all of the system buses is impossible because of their high cost and the lack of communication facilities [3]. Thus, one of the important issues is to nd the optimal number and location of PMUs according to the desired objectives. The main goal is therefore to minimize the number of required PMUs to be installed in the power system while maintaining full observability of the system. To date, several methods have been employed to solve this optimization problem. These methods can generally be divided into conventional methods and heuristic optimization algorithms. In conventional methods, the optimal placement of PMUs is expressed as an integer linear programming problem [410]. Hence, the proper denition of constraints that guarantee the satisfaction of the desired objective plays a key role in achieving the optimal solution. Therefore, the differences between these methods lie in the formulation of the required constraints. In [4], with the aim of full network observability, the constraints of the optimal placement problem are formulated as a set of nonlinear inequalities. In [5], after changing the system topology, the constraints are derived as a set of linear inequalities for the recongured system. In the formulations presented in [610], the constraints are introduced as a set of linear inequalities, without any reconguration in the system topology. Despite all of these differences,

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

the common feature of these methods is the small solution time for large-scale systems. We note that each of these methods can achieve an optimal solution for the PMU placement problem to serve a specic purpose. For example, among the prior studies, the minimum number of PMUs needed to ensure that the power system is fully observable was obtained in [9,10]. However, in [6], a smaller number of PMUs are presented to guarantee the full observability of the power system in the case of a single PMU loss. Therefore, there remains the need for a generalized method for determining the optimal number and locations of PMUs to satisfy these goals in various cases. Among the heuristic optimization algorithms, simulated annealing [3], binary search [11], Tabu search [12], genetic algorithm [13], particle swarm optimization [1416], ant colony optimization [17], immunity genetic algorithm [18] and immunity particle swarm optimization [19] have been developed. Unlike conventional methods, in these methods, there is no need to dene a set of constraints to achieve the desired goal. A special property of these methods is that the algorithm parameters can be tuned exibly so as to obtain the best solution. Using these methods for the PMU placement problem, the number of unobservable buses is considered a component of the objective function. Hence, by minimizing this term, full observability of the power system is guaranteed [3,1315,1719]. Among these previous works, the minimum number of required PMUs to achieve the full observability of the power system is presented in [14,15,18], and this number is equal to that obtained in [9,10] using the conventional methods. In comparison to conventional methods, the utilization of heuristic methods causes increase in solution time for very large-scale power systems [18]. In this paper, a new method is presented for the optimal placement of PMUs utilizing an integer programming technique. This method is able to determine the minimal number and optimal locations of PMUs in order to provide the full network observability as well as maximizing the measurement redundancy in normal operation and also in the case of a single PMU loss or a single line outage. The proper objective function and the required constraints to ensure the achievement of these objectives are described without changing the system topology. In addition, the proposed formulation is extended to consider a practical limitation on the maximum number of PMU channels. In the next section, the formulation of the optimal placement of PMUs with the aim of full network observability is presented. In Section 3, the necessary constraints to maintain full network observability given the occurrence of a single PMU or line outage are described. Finally, the performance of the presented formulation is assessed using several IEEE standard systems. The simulation results show the ability of the proposed method to yield the optimal placement of these units under different conditions.

The discrete nature of the optimal PMU placement problem requires the vector X to be dened as follows [410,16]: [X ] i = xi = 1, 0, if a PMU is installed at bus i otherwise (2)

where each entry of this vector shows the status of the installation of a PMU on each bus. Because the aim of solving the optimal placement of PMUs is to nd the minimum number of buses on which these units must be installed, the objective function of the problem can be written as follows [79]:
N

min
i=1

xi

(3)

In the above objective function, the cost of all PMUs is assumed to be equal. The cost of a PMU depends on several factors; such as the number of measuring channels, CT and PT connections, power connection, ground connection, and the GPS receiver. However, what really distinguishes between different PMU costs is the number of channels, because the remainder items are the same for all PMUs [6]. It is obvious that a PMU with more channels is costlier than that of with less number of channels. To consider unequal cost for PMUs, xi in (3) should be replaced by ci xi , where ci is the cost of installed PMU at the ith system bus [46,9,10]. The cost of a PMU which is installed on a bus with only one incident line can be set to 1 per-unit. For each additional incident line, the cost will be increased with an incremental factor . A reasonable selection for is 0.1 [6]. Minimization of the number of used PMUs by the objective function (3) may leads to various arrangements of PMUs with the same numbers. The question is this: which arrangement is the best one to be installed? In this paper, maximizing the measurement redundancy throughout the power system is considered as the objective in the problem of optimal measurement placement. In this way, the following objective function is selected:
N N N

min
i=1

ci xi

+w
i=1

mi
i=1

fi

(4)

2. Problem formulation in normal operation of the power system To nd the optimal locations of PMUs, it is sufcient to know the system topology and the type of system buses. The system bus connections are displayed using a system connectivity matrix. This matrix shows the interconnection of buses by transmission lines. Here, denoting A as the connectivity matrix and N as the number of system buses, A forms an N N matrix with entries dened as follows [410,16]:

where w is a weighting factor which is selected such that the two components of the objective function could be comparable in terms of magnitude, mi represents the maximum number of times that the ith bus can be observed ideally (i.e., the number of its incident lines plus one), and fi represents the number of times that the ith bus is observed by the set of installed PMUs obtained in X. The second term in (4) indicates the difference between the sum of ideal and actual number of times that each bus of the system is observed. The maximum redundancy will be gained by minimizing this difference, subsequently. Proper denition of the constraints that ensure the full observability of the system is the key to the solution of the optimal PMU placement problem. This denition changed according to the existing conditions of the system such as the inclusion or exclusion of zero-injection buses. Hence, in the following, we develop a method for determining the constraints of the optimization problem for each case. To facilitate this discussion, a 7-bus system is used as an example, as illustrated in Fig. 1. Here, the solid circles are system buses and zero-injection buses are distinguished from other buses by dots beside them (here, buses 3 and 5). 2.1. Problem constraints without considering zero-injection buses

[A] ij = aij =

1, if i = j
1, if buses i and j are connected 0, otherwise (1)

In the proposed method, we assume that a PMU installed on a bus has a sufcient number of input channels to measure both the voltage phasor of the bus and the current phasors of all lines

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

buses 2, 3, 4 and 6 are replaced by this constraint. In other words, the set of constraints related to this system is converted into:

f1 = x1 + x2 1

f7 = x2 + x4 + x7 1

f2 + f3 + f4 + f6 = x1 + 3x2 + 4x3 + 2x4 + x5 + 3x6 + 2x7 3

f4 + f5 = x3 + 2x4 + 2x5 + x7 1

(8)

Fig. 1. Graph of example 7-bus power system.

emanating from the bus. Thus, the voltage phasor of all buses connected to that bus can be calculated. In such cases, a bus is considered observable when a PMU is installed on it or on one of its adjacent buses. Thus, the observability constraint of the ith bus in the system can be dened as follows [410]:
N

f1 = x1 + x2 1

On the basis of the method presented in Ref. [7], the above constraints can be simplied in two ways. In the rst procedure, for the observability of the ve buses 2, 3, 4, 5 and 6, just three of the four buses 2, 3, 4 and 6 must be observable. Hence, the set of inequalities (8) would be rewritten as follows:

f7 = x2 + x4 + x7 1 f2 + f3 + f4 + f6 = x1 + 3x2 + 4x3 + 2x4 + x5 + 3x6 + 2x7 3

(9)

fi =
j=1

aij xj 1,

i = 1, 2, . . . , N

(5)

For example, the required constraints for full observability in the 7-bus system of Fig. 1 according to expression (5) are as follows:

f1 = x1 + x2 1

In the second procedure, for the observability of these ve buses, just one of buses 4 and 5 and two of the three buses 2, 3 and 6 must be observable. Therefore, the set of inequalities (8) can be written as:

f7 = x2 + x4 + x7 1

f1 = x1 + x2 1 f2 = x1 + x2 + x3 + x6 + x7 1 f3 = x2 + x3 + x4 + x6 1 f5 = x4 + x5 1 f6 = x2 + x3 + x6 1
f7 = x2 + x4 + x7 1 f4 = x3 + x4 + x5 + x7 1 (6)

f2 + f3 + f6 = x1 + 3x2 + 3x3 + x4 + 3x6 + x7 2

f4 + f5 = x3 + 2x4 + 2x5 + x7 1

(10)

In these inequalities, a + sign is used to represent an algebraic sum. Moreover, the value of 1 on the right-hand sides of these inequalities ensures that at least one of the variables appearing in the summations on the left-hand sides will not be zero. For example, constraint f1 1 shows that a PMU should be installed on bus 1 or 2 (or both) to make bus 1 observable [4,5]. 2.2. Problem constraints when considering zero-injection buses In this case, buses 3 and 5 are considered to be zero-injection buses. For example, if the voltage phasor of three buses in the set of buses including 2, 3, 4 and 6 were known, the voltage phasor of the fourth bus can be obtained by applying Kirchhoffs Current Law (KCL) at bus 3, in which the net injection is equal to zero. Therefore, the constraints associated with this set of buses must be modied. In the following, we describe this modication of the mentioned constraints using the methods presented in two previous works [6,7]. After dening the related drawbacks, we describe how the proposed method is able to modify the problem constraints in such a way that all of the drawbacks are resolved. 2.2.1. Modication of constraints in Refs. [6,7] According to the previously proposed method, for example, the constraint required to provide observability for three of the four buses 2, 3, 4 and 6 can be stated by inequality (7): f2 + f3 + f4 + f6 3 (7)

In [7], the simplication of the set of constraints is not stated in the case of connecting a system bus to two or more zero-injection buses. Therefore, it is not clear which set of inequalities, i.e., (9) or (10), should be selected as a set of constraints to solve the optimization problem. In Fig. 1, the 7-bus system is made fully observable by placing a PMU on bus no. 2. The optimal solution can be achieved by considering the inequality (9) as the set of constraints for this optimization problem. Thus, we can conclude that the formulation presented in [6] cannot lead to an optimal solution for all power systems. 2.2.2. Modication of constraints in the proposed method The idea presented in the above-referenced works indicates that a good model for inclusion of zero-injection buses would be to determine the constraints of the described optimization problem, but it has the two following drawbacks. First, if a bus is connected to two or more zero-injection buses, there is no need for the corresponding fi to appear in all of the inequalities corresponding to those zero-injection buses. Second, it is unclear for which of the zero-injection buses that the corresponding fi must be considered in the corresponding inequality. To overcome these drawbacks, it appears that a simplication of constraints should be considered implicitly in the set of constraints of this optimization problem. Hence, the following set of constraints is proposed for the implementation of this task: fi 1 +
j ZIB

aij ui sij ui ,

i = 1, 2, . . . , N

(11)

aij yij = ui ,
j ZIB

i = 1, 2, . . . , N

(12)

aij sij = i ZIBk j = ZIBk sij yij , i ZIBk j = ZIBk

aij yij 1,

k = 1, 2, . . . , Nzib

(13)

In the above inequality, the + sign is applied as an algebraic sum and the necessary constraints fullling the observability of

i Z,

j ZIB

(14)

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

In these expressions, yij and sij are binary decision variables (i.e., with values of 0 or 1). ZIB is the set of zero-injection buses in the power system, ZIBk is the kth member of this set and Nzib represents the number of members in this set. Buses connected to a bus ZIBk along with this bus form the set ZIBk . The set Z and the variable ui are dened by expressions (15) and (16), respectively:
Nzib

Inequality (11) also ensures that the observability of buses by PMUs installed on them or their adjacent buses must be provided. According to the above stipulations, we can say that the proposed set of constraints resolves the drawbacks appearing in other methods and thus ensures an optimal solution of the PMU placement problem in all power systems. The accuracy of this claim was then investigated, as described in Section 4, wherein the simulation results are presented. 2.3. Problem constraints when considering a practical restriction on the number of PMU channels Since the number of PMU channels is limited in practice, a PMU installed on a bus could not measure the current phasors of all its incident lines. Thus, the aim of this subsection is to modify the formulation proposed in the previous section to consider the effect of this measurement limitation. This modication can be done by changing the introduced fi in expression (11) into:
N

Z=
k=1

ZIBk

(15)

ui = 0 ui = 1

if i /Z if i Z

(16)

ZIB = {3, 5} Nzib = 2

ZIB1 ZIB2

For example, in the 7-bus system depicted in Fig. 1, the sets dened in expressions (11) to (14) are as follows:

= {2, 3, 4, 6} = {4, 5}

(17)

Z = {2, 3, 4, 5, 6} fi =

f1 = x1 + x2 1 f2 = x1 + x2 + x3 + x6 + x7 s2,3 f 3 = x2 + x3 + x4 + x6 s3,3 f5 = x4 + x5 s5,5 f6 = x2 + x3 + x6 s6,3

f7 = x2 + x4 + x7 1

and expressions (11)(14) for this system are written as (18)(21):

aij dji ,
j=1

i = 1, 2, . . . , N

(22)

as well as adding the following two constraints to the previous set:

N max aij dij di , j=1

f4 = x3 + x4 + x5 + x7 s4,3 + s4,5

(18)

i = 1, 2, . . . , N

(23)

dij xi ,

i, j = 1, 2, . . . , N

(24)

y2,3 = 1, y3,3 = 1, y4,3 + y4,5 = 1, y5,5 = 1, y6,3 = 1 s2,3 + s3,3 + s4,3 + s6,3 = (y2,3 + y3,3 + y4,3 + y6,3 ) 1 s4,5 + s5,5 = (y4,5 + y5,5 ) 1

(19) (20)

s2,3 y2,3 , s3,3 y3,3 , s4,3 y4,3 , s4,5 y4,5 , s5,5 y5,5 , s6,3 y6,3 (21) To satisfy this set of constraints, y2,3 , y3,3 , y4,3 , y5,5 , y6,3 , s2,3 , s3,3 , s6,3 , and x2 must be equal to 1 and other variables must be set to 0. In this way, the optimal solution X = [0 1 0 0 0 0 0]t will be achieved. The meaning of y4,3 = 1 and y4,5 = 0 is that f4 must be considered in the inequality corresponded to zero-injection of bus no. 3 and there is no need for f4 to appear in the inequality corresponded to zero-injection of bus no. 5. In other words, the set of constraints (8) must be simplied using the rst procedure, and thus, the set of inequalities (9) should be selected as the set of constraints. Also, The meaning of s2,3 , s3,3 , and s6,3 = 1 is that f2 , f3 , and f6 must be greater than 1 to satisfy the third inequality of (9). According to the above explanations, the followings can be concluded about the proposed formulation: Expression (12) guarantees that if the ith bus of the system is connected to two or more zero-injection buses, fi is only considered in the inequality corresponding to one of these zero-injection buses. Thus, the drawback associated with the method proposed in [6] is resolved. Expression (13) stipulates that if buses connected to a zeroinjection bus accompany this bus to form an n-member set, for this set to be observable, only n 1 members must be observable. In addition to the recent case, expressions (13) and (14) indicate the buses which will be made observable by PMUs through a set of connected buses to a zero-injection bus.

where dij is a binary variable which will be set to 1 to indicate that the jth bus is observable through the measured phasor current of the line connected between the ith and the jth buses, by the PMU max is the maximum number installed on the ith bus. Moreover, di of phasors which can be measured by the PMU installed on the ith system bus. Hence, if it sets to 1, it indicates that the PMU only measures the voltage phasor of the ith bus, or if it sets to 2, it implies that the voltage phasor of the ith bus in accompany with the current phasor of the only one its incident lines can be measured, as well. For example, in the 7-bus system, by considering the presented denition for fi as (22), the expression (11) can be written as follows:

f1 = d1,1 + d2,1 1 f2 = d1,2 + d2,2 + d3,2 + d6,2 + d7,2 s2,3 f3 = d2,3 + d3,3 + d4,3 + d6,3 s3,3 f5 = d4,5 + d5,5 s5,5 f6 = d2,6 + d3,6 + d6,6 s6,3
f7 = d2,7 + d4,7 + d7,7 1 f4 = d3,4 + d4,4 + d5,4 + d7,4 s4,3 + s4,5

(25)

Moreover, the expressions (23) and (24) can be written as (26) and (27), respectively.

max d1,1 + d1,2 d1 max d2,1 + d2,2 + d2,3 + d2,6 + d2,7 d2 max d3,2 + d3,3 + d3,4 + d3,6 d3 max d5,4 + d5,5 d5 max d6,2 + d6,3 + d6,6 d6
max d7,2 + d7,4 + d7,7 d7 max d4,3 + d4,4 + d4,5 + d4,7 d4

(26)

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

d1,1 , d1,2 x1 d2,1 , d2,2 , d2,3 , d2,6 , d2,7 x2 d3,2 , d3,3 , d3,4 , d3,6 x3 d5,4 , d5,5 x5 d6,2 , d6,3 , d6,6 x6
d7,2 , d7,4 , d7,7 x7 d4,3 , d4,4 , d4,5 , d4,7 x4 (27)

For example, in the 7-bus system, the expressions (28)(30) are written as (31)(33):

f1 = x1 + x2 2 f2 = x1 + x2 + x3 + x6 + x7 s2,3 + t2,3 f3 = x2 + x3 + x4 + x6 s3,3 + t3,3 f5 = x4 + x5 s5,5 + t5,5 f6 = x2 + x3 + x6 s6,3 + t6,3

f7 = x2 + x4 + x7 2 f4 = x3 + x4 + x5 + x7 (s4,3 + s4,5 ) + (t4,3 + t4,5 ) (31)

max = 2 (for i = 1, 2,. . ., 7), to satisfy the above conAssuming di straints along with the presented constraints in (19)(21), y2,3 , y3,3 , y6,3 , y4,5 , y5,5 , s2,3 , s6,3 , s4,5 , d1,1 , d1,2 , d6,6 , d7,4 , d7,7 , x1 , x6 , and x7 must be equal to 1 and the other variables must be set to 0. This implies that buses 1, 6, and 7 are made observable, directly, buses 2 as well as 4 are made observable by the installed PMUs at buses 1 and 7, respectively, and buses 3 as well as 5 are made observable by applying the KCL at them. Comparing this result with the presented one in Section 2.2.2 indicates that the optimal number of required PMUs for full observability of the power system will be increased by considering the limitation on the number of PMU channels.

t2,3 + t3,3 + t4,3 + t6,3 = (y2,3 + y3,3 + y4,3 + y6,3 ) 1 t4,5 + t5,5 = (y4,5 + y5,5 ) 1

(32)

t2,3 y2,3 , t3,3 y3,3 , t4,3 y4,3 , t4,5 y4,5 , t5,5 y5,5 , t6,3 y6,3 (33)

3. Proposed formulation in the case of a single PMU loss or a single line outage PMU malfunction and system line outages are probable occurrences, as are failures of other equipment used in power systems. Thus, continuous state estimation in a power system requires that the placement of these units be performed such that full observability of the system is maintained in either condition. However, the high cost of PMUs restricts the number of units that can be installed in a power system. Therefore, they should be placed optimally so as to reduce the number of units required to as few as possible.

To satisfy the above constraints and the constraints presented in (19)(21), y2,3 , y3,3 , y4,5 , y5,5 , y6,3 , s2,3 , s3,3 , s4,5 , t2,3 , t5,5 , t6,3 , x1 , x2 , and x4 must be equal to 1 and other variables must be set to 0. Therefore, this system will be fully observable in the case of a single PMU loss by placing PMUs on the 1st, the 2nd, and the 4th system bus. 3.2. Problem constraints in the case of a single line outage To derive the problem constraints in this case, in the rst step one of the system lines is eliminated, and the set of constraints presented in expressions (11)(14) is then obtained for the new system. This task should be repeated for the next line outage, and this process continued until the outages of all lines have been evaluated. It should be noted that in each step, the new system differs from the main system only in the absence of the line that is removed in that step. The denition of constraints that must be changed depends on the location of the eliminated line in each corresponding step. In other words, if the lth line (i.e., that between ith and jth buses) is eliminated, the symbols and variables used in expressions (11)(14) are changed to one of the two following forms: i and j / ZIB aij al ij fi fil aij al ij fi fil ,
ZIBk

3.1. Problem constraints in the case of a single PMU loss To guarantee the full observability of the power system when a single PMU loss occurs, we must dene the problem constraints so that the voltage phasors of all of the system buses are accessible in at least two ways. In other words, each system bus must be observable twice. To accomplish this condition, among the set of constraints presented in expressions (11)(14), the inequality (11) should be replaced by the following inequality:

fi 2 +
j ZIB

aij ui sij ui +
j ZIB

aij ui tij ui ,

(34)

i or j ZIB (28)

l ZIBk

i = 1, 2, . . . , N

Z Zl

l yij yij l sij sij

(35)

In this inequality, tij , as for sij , is a binary decision variable (the number 0 or 1). Hence, it is necessary that the same constraints be applied to these variables as described for the variables sij . In other words, the two inequalities below must be added to the previous set of constraints: aij tij = i ZIBk j = ZIBk tij yij , i ZIBk j = ZIBk aij yij 1, k = 1, 2, . . . , Nzib (29)

In these expressions, the symbols and variables changed because of the lth line outage are specied by the superscript l. In explaining these expressions, we note that the change of each symbol, for example, ZIBk or Z, that represents a set will cause the members of this set to change. Additionally, if the symbol represents a variable (like yij or sij ) changes, a new variable must be dened. Thus, for al we now write: ij al = ij 0 aij if line l is between buses i and j otherwise (36)

i Z, j ZIB

(30)

The reason for converting fi into fil is justied by inspection of expressions (5) and (36).

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

For example, by eliminating the line between the 1st and the 2nd buses of the 7-bus system, a1,2 and a2,1 will be equal to 0. Thus, only the rst two constraints of (18) will be changed into: f1 = x1 1 f2 = x2 + x3 + x6 + x7 s2,3 (37)

and other constraints presented in expressions (18)(21) will be left without any modication. However, if the line between the 2nd, and the 3rd system bus eliminates, not only a2,3 and a3,2 will be equal to 0 but also the following sets will be changed:
ZIB1

= {3, 4, 6}

Z = {3, 4, 5, 6}

(38)

Therefore, the new variables yij and sij should be dened. If these variables are denoted by the superscript l23 , then the expressions (11)(14) for this new system will be written as follows:

f1 = x1 + x2 1 f2 = x1 + x2 + x6 + x7 1 l23 f = x3 + x4 + x6 s3 ,3 3 l23 f5 = x4 + x5 s5 ,5 l23 f = x + x + x s6 6 2 3 6 ,3

f7 = x2 + x4 + x7 1

l23 + sl23 f4 = x3 + x4 + x5 + x7 s4 ,3 4,5

(39)

l23 l23 l23 l23 l23 y3 ,3 = 1, y4,3 + y4,5 = 1, y5,5 = 1, y6,3 = 1 l23 + sl23 + sl23 = (yl23 + yl23 + yl23 ) 1 s3 ,3 4,3 6,3 3,3 4,3 6,3 l23 + sl23 = (yl23 + yl23 ) 1 s4 5,5 5,5 ,5 4,5 l23 l23 l23 l23 l23 l23 l23 l23 l23 l23 s3 ,3 y3,3 , s4,3 y4,3 , s4,5 y4,5 , s5,5 y5,5 , s6,3 y6,3

(40)

Fig. 2. Proposed owchart for determining the problem constraints in the event of one of the discussed contingencies.

(41)

4. Simulation results To evaluate the ability of the presented formulation, optimal PMU placement for the 14-, 30-, 39-, 57- and 118-bus IEEE standard systems was performed so as to meet the following four objectives:

(42) Case 1. Full network observability in the case of normal system operation Case 2. Full network observability in the case of a single PMU loss Case 3. Full network observability in the case of a single line outage Case 4. Full network observability in the case of the occurrence of one contingency among the two previous ones. Furthermore, to assess the implementation feasibility of the proposed method in a very large-scale system, the optimal PMU placement problem in Cases 1 and 2 was solved for 2383- and 2746bus systems. The specications of these systems are given in Table 1 [20,21]. It should be noted that to solve the problem in each case, a program was written in GAMS software using the CPLEX solver. This particular solver was used because of its high capability for nding the optimal solution in the minimum possible time for all studied cases. In each of these programs, the relative termination criterion was set at zero to achieve the global optimal solution. Because all variables in our proposed formulation are binary and also the objective function and all constraints are dened as linear equations and inequalities, the problem was solved using Mixed Integer Programming (MIP). The computer used had an Intel Core 2 Duo CPU E7500 running at 2.93 GHz with 2.0GB of RAM. Finally, it should be pointed out that in all of the case studies, the effect of zero-injection buses was considered and the expression (4) was used as an objective function with w = 0.1. Also, to compare the proposed method in this paper with the methods presented in the corresponding references,

By comparing the above constraints with those presented in (18)(21), it can be seen that only the rst and the last inequalities of (18) and (39) are the same. Hence, there is no need to consider these inequalities again in (39). 3.3. Problem constraints in the occurrence of one contingency In this case, it should be noted that the discussed contingency can be only one of the two contingencies, i.e., the loss of a PMU or the outage of a system line. Hence, the set of constraints that were dened in the two previous subsections should be considered simultaneously. For this purpose, two inequalities (29) and (30) must rst be added to expressions (11)(14), and inequality (11) is replaced by inequality (28). Next, a set of constraints is determined using the presented expressions. One of the system lines is then eliminated, and the set of constraints is obtained according to expressions (11)(14) for the new system. The eliminated line is then returned to service, and this procedure is repeated by the outage of another system line. The sets of constraints obtained in the successive iterations are nally joined. It is clear that some of obtained constraints in various iterations are the same, and there is no need to consider them again. Furthermore, it is clear that the number of iterations will be equal to the number of system lines. Thus, if we consider Nbranch as the number of system lines, then the owchart presented in Fig. 2 can be used to determine the set of constraints in this case.

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110 Table 1 Specications of IEEE standard systems [20,21]. IEEE system 14-Bus 30-Bus 39-Bus 57-Bus 118-Bus 2383-Bus 2746-Bus Number of lines 20 41 46 80 186 2896 3514 Number of zero-injection buses 1 6 12 15 10 552 704 Locations of zero-injection buses 7 6, 9, 22, 25, 27, 28 1, 2, 5, 6, 9, 10, 11, 13, 14, 17, 19, 22 4, 7, 11, 21, 22, 24, 26, 34, 36, 37, 39, 40, 45, 46, 48 5, 9, 30, 37, 38, 63, 64, 68, 71, 81 Buses without load and generator Buses without load and generator Number of radial buses 1 3 9 1 7 504 320 Locations of radial buses 8 11, 13, 26 30, 31, 32, 33, 34, 35, 36, 37, 38 33 10, 73, 87, 111, 112, 116, 117

the PMU cost at all buses was assumed to be the same. Thus, all coefcients ci (i = 1, 2,. . ., N) were chosen equal to 1 in (4). Case 1. Full network observability in the case of normal system operation. In this case, the optimal numbers and locations of the PMUs in the systems under consideration are shown in Table 2. Here, the PMU locations in the 2383- and 2746-bus systems were omitted for brevitys sake. According to the results in Table 2, we concluded that by equipping approximately 22% of system buses with PMUs on average, their full observability will be ensured. In Table 3, a comparison between the results presented in Table 2 and other studies is provided. As shown here, the number of PMUs obtained using the proposed method required for full observability of the 14-, 30-, 39-, 57- and 118-bus IEEE systems are equal to the numbers reported in [9,15]. Thus, the accuracy of the claim in Section 2.2.2 is veried. In addition, Tables 2 and 3 show the superiority of the proposed method over evolutionary optimization methods for solving the optimal PMU placement problem to provide full observability of very large-scale systems, because in [14,15], solutions for large-scale systems are not presented, and in [18], this problem is considered for a 2746-bus system, but the number of PMUs and the time for required calculation are much higher than our results. Among the referenced works, maximizing the measurement redundancy is only addressed in [8,1012,16]. To compare the proposed method in this paper with the methods presented in the mentioned references, the System Measurement Redundancy N Index (SMRI) is dened as f added by the number of zeroi=1 i entries of AX (i.e., the number of system buses which are made observable by applying KCLs). It is clear that having the more PMUs is installed in the power system, the greater SMRI will be yielded. Hence, the SMRI is calculated for each IEEE test system which: (1) is studied in aforementioned references; (2) its optimal number of PMUs is equal to what is obtained using the proposed method; and (3) its optimal location of PMUs is reported. Table 4 shows the results of this calculation. The SMRI is compared in this Table for various IEEE systems using the proposed formulation with what is calculated for some aforementioned references. It is observed that the SMRI for these references is lower than SMRI obtained in this paper, which shows the superiority of the proposed method in terms of measurement redundancy. In the referenced works, measurement channel limitation is only considered in [9]. The proposed formulation presented in Section 2.3 will lead to the same results reported in this reference. Therefore, these results are not presented here, again. However, it can be seen that the proposed formulation in this paper is very simpler in comparison with the one presented in [9]. Case 2. Full network observability in the case of a single PMU loss. Table 5 shows the optimal numbers and locations of PMUs in this case. As for the previous case, presentation of the PMU locations for the 2383- and 2746-bus systems was omitted. Table 5 shows

that for full observability of the system in the case of a single PMU loss, approximately 47% of the buses should be equipped with these units on average. Table 6 provides a comparison of the proposed method and other studies in which the optimal number of PMUs was calculated for this case. Accordingly, we found out that full observability can be achieved with fewer PMUs using the proposed method than reported in prior studies. Case 3. Full network observability in the case of a single line outage. The optimal numbers and locations of PMUs for this case are shown in Table 7. The results presented in this table show that to guarantee full observability of the system in the case of single line outage, approximately 42% of the buses should be equipped with PMUs on average. In the referenced works, this case was only considered in [9,11]. In [11], unlike the [9], the possibility of outages in lines connected to buses with only one incident line was ignored. This assumption does not seem very accurate, because the possibility of outage always exists for any line in the system, and we cannot consider this possibility equal to zero for a certain number of lines. However, on an outage of such a line, the incident bus will be isolated from the system. Therefore, to measure the voltage phasor, we cannot disregard the possibility of an outage in the line connected to it. Hence, if the possibility of outages in lines connected to these buses is ignored, the number of required PMUs will be certainly lower. However, according to any of the assumptions discussed in [9,11], the number of PMUs calculated using the proposed method will be equal to the reported by those authors. Case 4. Full network observability in the case of the occurrence of one contingency among the two previous contingencies. Table 8 shows the optimal numbers and locations of PMUs in this case. The results presented in this table indicate that full observability of the system requires the installation of these units on approximately 50% of the buses on average. Among the referenced works, this case was only considered in [9,10,15]. However, in [10,15], as in [11], the possibility of outages for lines connected to buses with only one incident line was again ignored. Under this assumption, it is expected that these would calculate a lower number of PMUs than those reported in [9] and in Table 8 of the present study. To compare the proposed method with the method presented in [10,15], simulations for this case were also performed using the assumptions made by those authors. The number of PMUs calculated under this assumption using our method is presented in Table 9 along with that reported in [10,15]; utilizing the same assumptions, the proposed method is superior to that presented in [10,15]. Table 10 shows a comparison between the results of the proposed method and the method presented in [9] regarding the number of PMUs. Here, the proposed method is considerably superior to the previously presented method for achieving the optimal solution for this case.

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110

Table 2 The results of applying the proposed method for the rst case of simulations. IEEE system Number of PMUs 3 7 8 11 28 553 594 Locations of PMUs Percentage between the number of PMUs and the number of system buses 21.43 23.33 20.51 19.3 23.73 23.21 21.63 Execution time (s) <1 <1 <1 <1 <1 15.28 40.63 SMRI

14-Bus 30-Bus 39-Bus 57-Bus 118-Bus 2383-Bus 2746-Bus

2, 6, 9 2, 4, 10, 12, 15, 18, 27 3, 8, 13, 16, 20, 23, 25, 29 1, 4, 13, 20, 25, 29, 32, 38, 51, 54, 56 3, 8, 11, 12, 17, 21, 27, 31, 32, 34, 37, 40, 45, 49, 53, 56, 62, 72, 75, 77, 80, 85, 86, 90, 94, 102, 105, 110

16 41 43 59 156 2788 3216

Table 3 Comparison of the number of PMUs listed in Table 2 with the numbers presented in prior studies. IEEE system Proposed method Ref. [3] Ref. [4,5] Ref. [6,16] Ref. [7] Ref. [8] Ref. [9] Ref. [10] Ref. [11] Ref. [12] Ref. [13] Ref. [14] Ref. [15] Ref. [18] Ref. [19] 14-Bus 3 3 3 3 3 3 3 3 3 3 3 3 3 3 30-Bus 7 7 7 7 7 7 7 7 7 7 39-Bus 8 8 8 8 10 8 8 57-Bus 11 12 13 11 14 11 11 13 12 11 11 11 118-Bus 28 29 29 29 29 28 28 29 28 28 28 2383-Bus 553 553 2746-Bus 594 609

Table 4 Comparison of the SMRI listed in Table 2 with those calculated for prior studies. IEEE system Proposed method Ref. [10] Ref. [11] Ref. [12] Ref. [16] 14-Bus 16 16 16 16 30-Bus 41 41 39 34 39-Bus 43 43 57-Bus 59 59 118-Bus 156 149 2383-Bus 2788 2746-Bus 3216

Table 5 The results of applying the proposed method for the second case of simulations. IEEE system Number of PMUs 7 14 17 22 61 Locations of PMUs Percentage between the number of PMUs and the number of system buses 50 46.67 43.59 38.6 51.69 Execution time (s) <1 <1 <1 <1 <1 SMRI

14-Bus 30-Bus 39-Bus 57-Bus 118-Bus

2383-Bus 2746-Bus

1224 1299

2, 4, 5, 6, 9, 11, 13 1, 3, 5, 7, 10, 12, 13, 15, 17, 18, 20, 24, 27, 29 2, 6, 8, 13, 14, 16, 18, 20, 21, 23, 25, 26, 29, 34, 36, 37, 38 1, 3, 6, 9, 12, 15, 19, 20, 25, 28, 29, 30, 32, 33, 38, 41, 46, 50, 51, 53, 54, 56 1, 3, 7, 8, 9, 11, 12, 15, 17, 19, 21, 22, 23, 24, 27, 29, 31, 32, 34, 35, 40, 42, 44, 45, 46, 49, 51, 52, 54, 56, 57, 59, 62, 66, 68, 70, 71, 75, 76, 77, 78, 80, 83, 85, 86, 87, 89, 91, 92, 94, 96, 100, 101, 105, 106, 108, 110, 111, 112, 115, 117

34 56 66 94 284

51.36 47.31

44.12 202.86

4634 5107

Table 6 Comparison of the number of PMUs listed in Table 5 with the numbers presented in prior studies. IEEE system Proposed method Ref. [5] Ref. [6] Ref. [8] Ref. [9] 14-Bus 7 7 7 7 7 30-Bus 14 16 15 39-Bus 17 18 57-Bus 22 26 25 29 26 118-Bus 61 66 61 64 63 2383-Bus 1224 2746-Bus 1299

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110 Table 7 The results of applying the proposed method for the third case of simulations. IEEE system Number of PMUs 7 13 15 19 53 Locations of PMUs Percentage between the number of PMUs and the number of system buses 50 43.33 38.46 33.33 44.92 Execution time (s) <1 <1 1.52 3.24 1.48 SMRI

14-Bus 30-Bus 39-Bus 57-Bus 118-Bus

1, 3, 6, 8, 9, 10, 13 1, 4, 5, 10, 11, 12, 13, 15, 17, 19, 24, 26, 30 3, 8, 16, 23, 26, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38 1, 3, 6, 12, 14, 15, 19, 27, 29, 30, 32, 33, 38, 41, 49, 51, 53, 55, 56 1, 6, 10, 11, 12, 15, 17, 19, 21, 23, 24, 25, 27, 29, 32, 34, 35, 40, 42, 44, 46, 49, 51, 53, 56, 57, 59, 62, 66, 70, 73, 75, 76, 78, 80, 83, 85, 87, 89, 91, 92, 94, 96, 100, 102, 105, 106, 109, 111, 112, 115, 116, 117

25 50 51 86 242

Table 8 The results of applying the proposed method for the fourth case of simulations. IEEE system Number of PMUs 8 16 18 22 62 Locations of PMUs Percentage between the number of PMUs and the number of system buses 57.14 53.33 46.15 38.6 52.54 Execution time (s) <1 <1 4.76 3.37 1.65 SMRI

14-Bus 30-Bus 39-Bus 57-Bus 118-Bus

2, 4, 5, 6, 8, 9, 11, 13 1, 2, 4, 5, 10, 11, 12, 13, 15, 17, 18, 19, 24, 26, 27, 30 4, 8, 16, 18, 20, 23, 25, 26, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38 1, 3, 6, 9, 12, 15, 19, 20, 25, 28, 29, 30, 32, 33, 38, 41, 46, 50, 51, 53, 54, 56 1, 3, 6, 8, 10, 11, 12, 15, 17, 19, 21, 22, 24, 25, 27, 28, 31, 32, 34, 36, 40, 42, 44, 45, 46, 49, 51, 53, 54, 56, 57, 59, 62, 66, 70, 71, 73, 75, 77, 78, 80, 83, 85, 86, 87, 89, 91, 92, 94, 96, 100, 102, 105, 106, 108, 110, 111, 112, 115, 116, 117, 118

35 62 60 94 280

Table 9 Comparison of the number of PMUs calculated using the proposed method with the numbers presented in Refs. [10,15]. IEEE system Proposed method Ref. [10] Ref. [15] 14-Bus 7 7 30-Bus 14 15 15 39-Bus 17 17 57-Bus 22 25 22 118-Bus 61 61 62

Table 10 Comparison of the number of PMUs listed in Table 8 with the numbers presented in Ref. [9]. IEEE system Proposed method Ref. [9] 14-Bus 8 8 30-Bus 16 17 39-Bus 18 22 57-Bus 22 26 118-Bus 62 65

5. Conclusions In this paper, a new method for optimal placement of PMUs was proposed with the aim of full network observability as well as maximizing the measurement redundancy. This optimization problem was introduced in the form of a binary integer programming problem and associated sets of constraints consisting of linear equations and inequalities. In addition, the required constraints to consider the measurement channel limitation on these units were included in the proposed formulation. The optimal placement of these units was determined in the contingencies of a single PMU loss and a single line outage. GAMS software was used to solve the optimization problems. The results of applying the proposed method to the solution of all of the optimization problems for the different systems under study showed that this method is capable of nding a lower number of PMUs than most previous methods for each of the problems. In addition, this method can be run through the GAMS software, in which the calculation time is reduced substantially in comparison with the previous methods, which are based on evolutionary optimization algorithms. References
[1] J. Bertsch, C. Carnal, D. Karlsson, J. McDaniel, K. Vu, Wide-area protection and power system utilization, Proceedings of the IEEE 93 (May (5)) (2005) 9971003.

[2] A. Mao, J. Yu, Z. Guo, PMU placement and data processing in WAMS that complements SCADA, in: IEEE in Power Engineering Society General Meeting, vol. 1, 1216 June 2005, 2005, pp. 780783. [3] T.L. Baldwin, L. Mili, M.B. Boisen, R. Adapa, Power system observability with minimal phasor measurement placement, IEEE Transactions on Power Systems 8 (May (2)) (1993) 707715. [4] B. Xu, A. Abur, Observability analysis and measurement placement for systems with PMUs, in: IEEE/PES Power Systems Conference and Exposition, vol. 2, 1013 October 2004, 2004, pp. 943946. [5] B. Xu, Y.J. Yoon, A. Abur, Optimal placement and utilization of phasor measurements for state estimation, in: Power Systems Computation Conference, Liege, Belgium, August 2005, 2005. [6] N.H. Abbasy, H.M. Ismail, A unied approach for the optimal PMU location for power system state estimation, IEEE Transactions on Power Systems 24 (May (2)) (2009) 806813. [7] B. Gou, Generalized integer linear programming formulation for optimal PMU placement, IEEE Transactions on Power Systems 23 (August (3)) (2008) 10991104. [8] D. Dua, S. Dambhare, R.K. Gajbhiye, S.A. Soman, Optimal multistage scheduling of PMU placement: an ILP approach, IEEE Transactions on Power Delivery 23 (October (4)) (2008) 18121820. [9] F. Aminifar, A. Khodaei, M. Fotuhi-Firuzabad, M. Shahidehpour, Contingencyconstrained PMU placement in power networks, IEEE Transactions on Power Systems 25 (February (1)) (2010) 516523. [10] S.M. Mahaei, M. Tarafdar Hagh, Minimizing the number of PMUs and their optimal placement in power systems, Electric Power Systems Research 83 (February (1)) (2012) 6672. [11] S. Chakrabarti, E. Kyriakides, Optimal placement of phasor measurement units for power system observability, IEEE Transactions on Power Systems 23 (August (3)) (2008) 14331440. [12] J. Peng, Y. Sun, H.F. Wang, Optimal PMU placement for full network observability using Tabu search algorithm, International Journal of Electrical Power and Energy Systems 28 (May (4)) (2006) 223231.

10

A. Enshaee et al. / Electric Power Systems Research 89 (2012) 110 [17] Bo Wang, Dichen Liu, Li Xiong, An improved ant colony system in optimizing power system PMU placement problem, in: Asia-Pacic Power and Energy Engineering Conference, 2731 March 2009, 2009, pp. 13. [18] F. Aminifar, C. Lucas, A. Khodaei, M. Fotuhi-Firuzabad, Optimal placement of phasor measurement units using immunity genetic algorithm, IEEE Transactions on Power Delivery 24 (July (3)) (2009) 10141020. [19] Chunhua Peng, Xuesong Xu, A hybrid algorithm based on BPSO and immune mechanism for PMU optimization placement, in: 7th World Congress on Intelligent Control and Automation, 2527 June 2008, 2008, pp. 70367040. [20] Power Systems Test Case Archive. http://www.ee.washington.edu/research/ pstca/. [21] MATPOWER Version 3.2. http://www.pserc.cornell.edu/matpower/.

[13] F.J. Marin, F. Garcia-Lagos, G. Joya, F. Sandoval, Genetic algorithms for optimal placement of phasor measurement units in electrical networks, Electronics Letters 39 (September (19)) (2003) 14031405. [14] M. Hajian, A.M. Ranjbar, T. Amraee, A.R. Shirani, Optimal placement of phasor measurement units: particle swarm optimization approach, in: International Conference on Intelligent Systems Applications to Power Systems, 58 November 2007, 2007, pp. 16. [15] M. Hajian, A.M. Ranjbar, T. Amraee, B. Mozafari, Optimal placement of PMUs to maintain network observability using a modied BPSO algorithm, International Journal of Electrical Power and Energy Systems 33 (January (1)) (2011) 2834. [16] A. Ahmadi, Y. Alinejad-Beromi, M. Moradi, Optimal PMU placement for power system observability using binary particle swarm optimization and considering measurement redundancy, Expert Systems with Applications 38 (June (6)) (2011) 72637269.