IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO.

6, JUNE 2020 3451

Placement of Distribution-Level Phasor

Measurements for Topological Observability and
Monitoring of Active Distribution Networks
Kapil Chauhan , Student Member, IEEE, and Ranjana Sodhi, Member, IEEE

Abstract— This article presents a new method for the place- Lately, there has been a major rollout of smart meters (SMs)
ment of distribution-level phasor measurement units (D-PMUs) in the various DNs across the globe, which has somewhat
in the active distribution networks (ADNs). The new objective helped in easing out the lack of real measurement problem
function and constraints are proposed to ensure the minimization
of capital cost as well the maximization of system observability, at the DNs. Yet, the proliferation of the distributed generators
in the presence of distributed generations (DGs), zero injection (DGs) at the distribution level of the grid is causing more
buses (ZIBs), and tie-switches. The integer linear programming variability, complexity, and uncertainty in the system condi-
(ILP) is used to solve the framed objective function and the tions. Under such dynamic scenarios, the reliable operation
constraints. Furthermore, the accuracy of ADN state estima- of active distribution networks (ADNs) is becoming more and
tion (ADN-SE), in various system conditions, such as network
reconfiguration, change in load demand, and measurement error more challenging. This, in turn, has recently necessitated the
degradation, is analyzed to check the impact of the proposed development of distribution-level phasor measurement units
placement. The uncertainties caused by D-PMUs, DGs, and (D-PMUs) (or μ-PMUs) [4], [5], which can accurately mea-
pseudomeasurements are considered well in ADN-SE using sure not only the voltage magnitudes but also the voltage phase
Monte Carlo simulations. The IEEE 69, 123, Indian 85, and angles of the fundamental as well as harmonic components.
UKGDS 95 bus distribution systems are used in the case studies.
Such measurement devices (MDs) are expected to augment
Index Terms— Active distribution network (ADN), the existing measurement infrastructure at the distribution
distribution-level phasor measurement unit (D-PMU), integer grid with the most accurate and real-time measurements,
linear programming (ILP), state estimation (SE), topological
observability. and serve to various applications [4], [6]. The first and the
foremost step toward the integration of real-time monitoring
into the future distribution management systems (DMSs) is
I. I NTRODUCTION the judicious placement of such devices as these sensors are

S INCE a practical power distribution network (DN) is

comprised of a large number of low-voltage (LV) substa-
tions/nodes, it becomes economically nonviable to equip the
costly. To this end, the proposed work can serve as a metering
system planning tool for determining the number and location
of distribution-level phasor measurements while considering
entire network with sensors to monitor it completely. As a the key characteristics of the ADNs.
result of this limitation, the reliable monitoring, operation, A study reveals that there exist a large number of meter
and control of DNs heavily rely on the DN state-estimation placement methods in the literature. References [7]–[9] present
(DN-SE) program [1]. At present, in order to make the an extensive review report, specifically on the available PMU
system completely observable and to assess the system states placement techniques. The survey in [8] reveals that most of
under all possible configurations with acceptable accuracy, the the existing articles carry out the optimal placement of PMUs
DN-SE makes use of a very few actual measurements and the in the transmission system. On the contrary, the placement of
remaining are pseudomeasurements, derived from historical phasor measurements in the ADNs has been limitedly carried
data. These pseudomeasurements, however, have nearly 50% out in the literature. Moreover, since the ADNs differ from the
mean error, and therefore adversely affect the accuracy of the transmission-level systems in many peculiar ways, the already
DN-SE [2], [3]. existing methods can just not be applied as such. As an
example, the following exists.
Manuscript received June 10, 2019; revised August 3, 2019; accepted
August 19, 2019. Date of publication September 9, 2019; date of current 1) ADNs have a limited number of measurements owing to
version May 12, 2020. This work was supported in part by the Fund for
Improvement of Science and Technology Infrastructure (FIST) Grant by usually less number of MDs being installed in the DNs.
the Department of Science and Technology (DST), New Delhi, through This, in turn, reduces the measurement redundancy and,
the IIT Ropar Project, under Grant ETA-422/2016. The Associate Editor hence, the accuracy of the ADN-SE [10].
coordinating the review process was Dr. Lihui Peng. (Corresponding author:
Kapil Chauhan.)
2) A transmission system is meshed in the formation,
The authors are with the Department of Electrical Engineering, while the distribution system is mostly radial or weakly
IIT Ropar, Rupnagar 140001, India (e-mail: [email protected]; meshed in nature.
[email protected]).
Color versions of one or more of the figures in this article are available
3) The topology of the distribution system is highly
online at http://ieeexplore.ieee.org. dynamic in nature. The system may become unobserv-
Digital Object Identifier 10.1109/TIM.2019.2939951 able after reconfiguration.
0018-9456 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 3452 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO. 6, JUNE 2020

4) There might be active island formation in the distribution 4) A methodical framework is presented to demonstrate the
system, which may split a network into two or more efficacy of the D-PMU measurements placement on the
networks. The separated islands/networks so formed ADN-SE.
may be unobservable. 5) A better SE performance is established in terms of the
Thus, it is very much desirable in the ADN scenarios accuracy and convergence characteristics based on the
to maximize the number of measurements with a minimum proposed D-PMU placement strategy.
number of MDs. The placement of MDs in the ADNs can This article is organized in five sections. Section II
be divided in two ways: 1) having a fixed number of MDs describes the proposed optimization formulation for the
and placing them to get a minimum SE error, such as D-PMU placement to make the ADN topologically observ-
in [11]–[17] and 2) finding the optimal (or minimum) number able. In Section III, the distribution-level phasor measurement
and location of MDs to get the SE error below a preset placement-based real-time monitoring architecture of the ADN
threshold, as in [18]–[23]. is presented. The D-PMU placement results are discussed
Recently, some articles are reported for the topologi- for various test systems in Section IV-A, along with a
cal observability of the ADN while focusing on specific detailed demonstration of proposed method (PM) on the IEEE
issues [24], [25]. An interesting two-stage optimal μPMU and 69-bus system. Section IV-B demonstrates the impact of the
SM placement strategy is proposed in [24], while minimizing proposed D-PMU placement on ADN-SE accuracy. Finally,
the overall cost of SMs and μPMUs including contingency the conclusion is drawn in Section V of this article.
constraints. A bender decomposition technique is used to
divide the decision variables into two subproblems and solved II. D ISTRIBUTION -L EVEL P HASOR M EASUREMENT
using mixed-integer linear programming (ILP). The algorithm P LACEMENT FOR T OPOLOGICAL O BSERVABILITY
process is interesting, but the required number of devices is A. Conventional ILP-Based PMU Placement
large. Moreover, the article suggests different sets of location In the graph-theoretical sense, a power system network
for different topologies. The topology change is considered can be viewed as a graph, where substations are represented
in [25] while minimizing the cost of μPMU and dual-use by the nodes and transmission lines are represented by the
line relay (DULR) for load loss minimization. However, these branches of the graph. Thus, every power system network
methods do not ensure the maximum total system observabil- has a certain topology/configuration, and the system is said
ity (TSO) of the ADN. to be topologically observable if all the nodes of the graph
Thus, the main objective of the present work is to address are directly or indirectly traversed/observed by at least one
the issue of economic placement of distribution-level pha- PMU [26]. To achieve a complete topological observability in
sor measurements for both increasing the system observ- the transmission networks, the PMU placement problem can be
ability (SO) and enhancing the ADN-SE accuracy. To this modeled as an ILP [27], as briefly described in the following.
end, a new objective function as the minimization of the 1) Objective Function: Minimization of the total cost of the
marginal cost is formulated. The set of new constraints are PMUs is formulated as the objective function such that
designed after analyzing the properties of the ADNs, for
topological observability. The optimization problem so formed 
N
f 1 = Min ci x i , x i ∈ R, Z (1)
is solved using ILP, which is simple and computationally
i=1
light, and subsequently, the effect of various events as network
reconfiguration, DG uncertainty, DG penetration, measure- where N is the number of buses in the network, ci is the
ment redundancy, measurement uncertainty, and so on, on cost of placing a PMU at bus-i , and x i is the binary decision
ADN-SE considering the proposed D-PMU location is ana- variable such that
lyzed. The results of the proposed D-PMU placement and x i = 1 if bus-i is equipped with a PMU, 0 otherwise. (2)
its impact on the ADN-SE are demonstrated on the modified
IEEE 69, 123, Indian 85, and UKGDS 95 node distribution 2) Basic Constraints: The set of observability constraints
systems, revealing the following key contributions of the are defined so as to observe each i th-bus with bi number
proposed work. of PMUs, either directly or indirectly. Direct observability is
defined as when a bus is equipped with a PMU, and indirect
1) A novel objective function is proposed in this arti- observability is defined when a bus is directly connected to a
cle, which effectively embeds two contradictory objec- bus, having a PMU. Thus, the constraints are defined as
tives, viz., minimization of the total number of MDs
and maximization of the TSO, in a single objective 
N
yi = Ai j x j ≥ bi ; i = 1, 2, ..., N. (3)
function.
j =1
2) A simple yet automatic, marginal cost-based weight
assignment scheme is developed for various D-PMU The value of bi is considered as unity for all the buses. Ai j
locations. is the element of the connectivity matrix such that

3) A set of new observability constraints are formulated 1, if i = j or if node i and j are connected
for the topological observability of an ADN so as Ai j = (4)
0, otherwise.
to minimize the effect of various factors like mea-
surement uncertainties, DG penetration, and network The optimization problem (1)–(3), termed conventional ILP
reconfiguration. (CILP), is finally be solved for x i using ILP.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 CHAUHAN AND SODHI: PLACEMENT OF DISTRIBUTION-LEVEL PHASOR MEASUREMENTS 3453

B. Proposed Method where Cii is a diagonal matrix containing a nonzero value

Minimization of f 1 with constraint (3) may result in mul- for specific bus(es), as discussed in the following. In addition
tiple solutions with the same number of PMUs and/or the to equality constraint (9), the corresponding inequality con-
same total cost. When these solutions are assessed with respect straints, Ai j , are also modified, as described in the following.
to the degree of observability, they result in different TSOs. 1) Substation Bus: The substation provides a reference
The TSO is calculated by summing up the individual bus angle for the complete ADN, and the angle measurement
observability index (BOI) [28], where the BOI of a bus-i is for the same must be accurately available. To this end,
calculated as how many buses can directly and/or indirectly the substation is assumed to be equipped with a D-PMU.
be observed by placing a PMU at bus-i . Ideally, it is desirable Thus, if s is a substation bus and j is the incident bus
to minimize the total cost of the PMUs while achieving the to s
maximum SO. These two contradictory objectives are modeled
Cs,s = 1, ds = 1 (10)
in a single objective function as follows.
1) Objective Function: The PM models the objective func- As,i = 0, bs = 0 ∀(i ∈ N) (11)
tion as the minimization of the total marginal cost of PMU A j,i = 0, b j = 0 ∀(i ∈ N). (12)
installation such that
2) DG Bus: The output power of DGs is stochastic in

N
nature, and substantially influences the SE results. Apart
f2 = Min λi x i (5) from achieving a good SE accuracy, detecting any
i=1
islanding scenarios also necessitates the real-time mon-
where λi is the marginal cost of placing one D-PMU at node-i . itoring of the DG buses. Real-time DG bus monitoring
The marginal cost is calculated as the ratio of change in can be achieved by installing a D-PMU at the DG bus-k
the capital cost (CC) and change in the SO by placing an as
additional D-PMU at bus-i , that is,
Ck,k = 1, dk = 1; Ak,i = 0, bk = 0 ∀(i ∈ N).
(CC)i
λi = (6) (13)
(SO)i
and Consequently, there is no need for direct or indirect
observability of the DG connected bus(es), that is
CCi − CC0 SOi − SO0
(CC)i = ; (SO)i = .
CCmax SOmax Ak+l,i = 0; bk+l = 0; ∀ l directly connected to k
(7) (14)
In (7), CC0 is the initial CC, CCi is the cost by placing an where k ∈ {1, 2, ..., K }, with K as total number of DG
additional D-PMU at i th bus, SO0 is the initial SO, and SOi buses.
is the SO by placing the additional D-PMU at the i th bus. The 3) Tie-Switches: Tie-switches are open in a normal operat-
values of CC0 and SO0 are 0, as there is no D-PMU in the ing condition. To achieve full observability even in the
network initially, and CCi = 1, ∀i , assuming that placing a case when the tie-switches are operated and the ADN
D-PMU at any bus costs the same. SOi = Ni + 1, with Ni topology gets changed, one of the tie-switch buses must
as the number of incidenting lines at bus-i . CCmax is the be equipped with a D-PMU, that is
maximum CC incurred if D-PMUs are placed at all the buses,
and SOmax is the maximum SO achieved by having D-PMUs A N+ p,t 1  = 1, A N+ p,t 2  = 1; b N+ p = 1 (15)
p p
at all the buses. They are used to convert the change in cost
(CCi − CC0 ) and the change in SO (SOi − SO0 ) in per unit ∀ p ∈ {1, 2, ..., P}, where P is the total pair of tie-
quantities, respectively. It can be noted that unlike CILP-based switches. t p is a pair of tie switch, containing (t 1p , t 2p )
PMU placement where every bus is assigned the same weight and t p ⊂ N.
ci in (1), the objective function f2 in the PM automatically 4) Zero-Injection Buses (ZIBs): ZIBs are the ones with no
assigns different weights to every bus. The bus-i , having generation or load, and can be considered as a bus hav-
maximum SO, results in the minimum marginal cost (λi ) and ing very much accurate power measurement. In a radial
gets priority to be equipped with a D-PMU. Consequently, system, there may be many concatenated ZIBs, and may
the proposed formulation effectively results in a minimum have a degree of 2 or more than 2, where the degree
overall D-PMU placement cost as well as a maximum overall of ZIB is defined as the number of lines connected
SO via single objective function. to it. Based on the degree and concatenation of ZIBs,
ADN Constraints: Full topological observability of the the following observability constraints are formulated.
ADNs is ensured through the following set of equality and a) ZIBs having degree more than 2 are treated as
inequality constraints: normal buses.
b) For ZIBs having degree less than or equal to 2,

N
at first, ensure that not to keep a D-PMU at ZIB.
Inequality Constraints : Ai j x j ≥ bi ; i = 1:N (8)
Thus,
j =1
Equality Constraints : Cii x i = di ; i = 1:N (9) C z,z = 1, dz = 0; A z,i = 0, bz = 0 (16)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 3454 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO. 6, JUNE 2020

TABLE I
M EASUREMENT U NCERTAINTIES

Fig. 1. Basic real-time monitoring architecture of the ADN.

and the system states E can be estimated by minimizing the
WLS error in the measurements such that
∀ z ∈ [1, 2, ..., Z ], ∀ (i ∈ N) Z is total number
of ZIBs having degree less than or equal to 2, and Ê = arg minx [Zm − h(E)]T [R−1 ][Zm − h(E)] (19)
Z ⊂ N. Thereafter, the following holds.
i) If there are two or less than two ZIBs in con- where the covariance matrix [R] = diag(σ12 , σ22 , ..., σn2 ), with
catenation, the adjacent buses to ZIBs should a standard deviation (SD) of the i th measurement as σi . At
be directly/indirectly observable. It will be ful- the k + 1th iteration
filled by constraint (8).
ii) If there are more than two ZIBs in concate- Ê(k+1) = Êk + Ek (20)
nation, one of the adjacent buses to the ZIB
group should be directly observable. Thus, add where Ek = [G(Ek )]−1 HT R−1 [Zm − h(Ek )], [G] =
equality constraint [H]T [R]−1 [H], [H] = δh(E)/δE. The iterative process of (20)
is stopped, once Ek reaches below a threshold of approxi-
C ze,ze = 1, dze = 1 (17) mate to 10−11 .
The various measurements considered in this article are
where ze is the non-ZIB connected to the listed in Table I, with their respective ME and error distribu-
extreme corner ZIB of all connected ZIBs. tion (ED). The measurement uncertainties, used in this article,
Then, remove ze constraints from (8) have been adopted from [12], where the errors in different
types of measurements are chosen based on the type of the
A ze,i = 0; bze = 0; ∀(i ∈ N). (18)
measuring instrument. The pseudomeasurements are collected
Finally, the solution of the optimization problem, f 2 , with from the historical data, thus having high variance. The volt-
ADN constraints (8)–(18) is obtained using ILP. age/current phasors are collected from D-PMUs, thus having
a low ME of 1%–3% as per the IEEE C37.118.1a-2014 [30].
The substation measurements are assumed to have an ME
III. M ONITORING OF ACTIVE D ISTRIBUTION N ETWORK of 1%. The ZIBs are considered as virtual measurements, with
The basic real-time monitoring architecture of the ADN is very low variance. Furthermore, the impact of uncertainties
shown in Fig. 1, which majorly contains three parts: 1) new caused by measurements and DGs is simulated via Monte
metering devices, i.e., D-PMUs; 2) phasor data concentrator Carlo simulation (MCS), as shown in Fig. 2. In each MCS,
(PDC), which collects the real-time D-PMU measurements measurement vector Zm is generated by adding a randomly
from a number of locations; and 3) ADN-SE program, which generated error in the true value Ztrue , obtained from load flow
runs in the DMS using various field measurements. The analysis. The randomly generated ME is assumed to follow the
accuracy, numerical stability, and robustness of the ADN-SE Gaussian distribution, i.e., SD = ME/3. The output of the DGs
against large measurement error (ME) depend on the number, is modeled as the Gaussian mixture model (GMM) [19] such
type, and location of the MDs, as analyzed in the following. that the non-Gaussian distribution FY can be expressed into a
GMM as

A. Impact of Distribution-Level Phasor Measurements on 

M

ADN State Estimation FY (k) = wi FM(μi ,σi ) (k) (21)

i=1
The full observability of the ADNs using only phasor
measurements is sought in Section II-B to have a converged where M is the number of Gaussian component and μi ,
and more accurate solution of the weighted least square σi , and wi are the mean, SD, and weight of the i th
(WLS)-based SE [29]. Thus, after placing the D-PMUs Gaussian component, respectively. The FY is obtained from
in the ADNs, at the locations resulted from Section II-B, the yearly data of wind generation and solar generation.
a D-PMU-assisted ADN SE is implemented, as briefly dis- The value of wi , u i , and σi is calculated
M using expec-
cussed in the following. tation maximization (EM), having w = 1. Finally,
 i i
In a DN, the field measurement vector Zm can be related the DGs’ power is generated as iM wi nor mr nd(μi , σi ) in
to the state vector, E, as Zm = h(E) + e, where e is the MEs, each MCS.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 CHAUHAN AND SODHI: PLACEMENT OF DISTRIBUTION-LEVEL PHASOR MEASUREMENTS 3455

TABLE II
S YSTEM D ESCRIPTION

 N D−PMU
i=1 BOIi + NZIB , where N D−PMU and NZIB are the
number of buses having D-PMUs and the number of ZIBs not
having D-PMUs at their adjacent buses, respectively. The NZIB
is added in the TSO, because the ZIBs are made observable
by using adjacent bus phasors, as explained in (29)–(36).
1) Superiority of the Proposed Objective Function: To
demonstrate the effectiveness of the proposed objective func-
tion, f 1 , which automatically assigns different weights to each
bus, the results of the MILP1 are compared with that of
the CILP, and the results are listed in Table III. It can be
observed that for the same number of D-PMUs, the proposed
objective function, f 2 , yields high TSO as compared with the
CILP. Just by including the proposed objective function in the
CILP scheme, an enhancement of 8.69%, 25.78%, 18.75%,
and 21.35% is observed in the SO of the IEEE 69, IEEE
123, IN 85 bus, and UK 95-bus system, respectively. Table III
also indicates the CPU time taken by the two methods.
As expected, MILP1 takes slightly more time than the CILP
because of the calculation of weight (λ).
Table IV shows the comparison between various objective
functions, available in the recent literature, with the same
observability constraints in (3). All the methods use ILP
to solve the optimization problem except [25] that uses the
genetic algorithm together with ILP. Reference [25] is framed
to minimize the cost of branch PMUs (ql ) in addition to the
minimization of the number of D-PMUs. Although the resulted
number of D-PMUs by MILP1 and that in [25] comes out to be
equal, but the resulted TSO in [25] is significantly low because
Fig. 2. Validation process of ADN-SE.
of its different objective function formulation. On the other
IV. R ESULTS AND D ISCUSSION hand, the aim of the objective functions in [26] and [31] is
almost similar to that of MILP1, i.e., minimization of cost and
A. Number and Location of D-PMUs
maximization of measurement redundancy. However, it can be
The efficacy of the PM is demonstrated on four test systems, observed from the results of Table IV that the optimization
viz., IEEE 69-bus [22], IEEE 123-bus [32], Indian 85-bus [23], results in [26] and [31] heavily depend on the tuning of the
and UK 95-bus [33], as given in Table II. The results of the weight parameter, w, unlike the MILP1. A comparison of the
following four variants of the D-PMU placement schemes are PM with recent literature reveals that the proposed objective
compared and discussed in Sections IV-A1 and IV-A2. function is different from the existing methods—simple, free
1) CILP: Conventional objective function of (1) with basic from any weight tuning, and results in maximum TSO with a
constraints in (3). minimum number of D-PMUs.
2) Modified CILP (MILP1): New proposed objective func- 2) Demonstration of the PM on IEEE 69-Bus System:
tion of (5) with basic constraints in (3). The total number of decision variables in this test system
3) Modified CILP2 (MILP2): Conventional objective func- is 69, and the equality constraints as formulated in (8)–(18) in
tion of (1) with new ADN constraints in (8)–(18). Section II-B are constructed as follows.
4) PM: Proposed new objective function of (5) with new 1) For substation bus, s = 1 [see Fig. 3(a)], the constraints
ADN constraints in (8)–(18). are
All the four methods are implemented in MATLAB
C1,1 = 1, d1 = 1; A1,1 = 0, A1,2 = 0, b1 = 0
R2016a environment, on an Intel Core i7-6700 CPU
at a 3.40-GHz, 16.0-GB RAM personal computer. Once (22)
the solution is obtained, the TSO is calculated as A2,1 = 0, A2,2 = 0, A2,3 = 0, b2 = 0. (23)

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 3456 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO. 6, JUNE 2020

TABLE III
D-PMU R ESULT P ERFORMANCE OF CILP AND MILP1

TABLE IV
C OMPARISON OF VARIOUS O BJECTIVE F UNCTIONS
FOR IEEE 69 B US S YSTEM

Fig. 3. Substation and DG cases. (a) Substation with D-PMU. (b) DG with
D-PMU.

2) DG bus constraints for bus-16, shown in Fig. 3(b), are

C16,16 = 1, d16 = 1; A16,i = 0, b16 = 0 (24)

A15,i = 0; b15 = 0; A17,i = 0; b17 = 0. (25)

3) The system is having five pairs of tie-switches, and Fig. 4. ZIB cases (black circles are ZIBs). (a) One ZIB. (b) Two ZIBs.
(c) Three ZIBs.
the equality constraint (15) for each such pair, as listed
in Table II, results in five additional constraints to (8)
as follows:
by applying Kirchhoff’s current law (KCL) at 23,
y70 = x 12 + x 44 ≥ 1; y71 = x 26 + x 47 ≥ 1 (26) as shown in Fig. 4(a)
y72 = x 14 + x 22 ≥ 1; y73 = x 66 + x 28 ≥ 1 (27) V22 − V23 V23 − V24
= (31)
y74 = x 51 + x 60 ≥ 1. (28) Z 22−23 Z 23−24
4) Likewise, the system is having 16 ZIBs (listed where V22 is a voltage phasor at bus-22 and Z 22−23
in Table II), with degree less than or equal to 2. First, is the line impedance between bus-22 and bus-23.
it is ensured not to place a D-PMU at these buses Suppose there are two ZIBs together, as shown in
by adding equality constraints according to (16) for all Fig. 4(b), then bus-22 and bus-25 will be indirectly
ZIBs, i.e., by keeping observable (V22 and V25 are known) in the worst
case by constraint (8). The voltage phasor at bus-
C15,15 = 1, d15 = 0; C63,63 = 1, d63 = 0; (29) 23 and bus-24 can be calculated by applying KCL
A15,i = 0, b15 = 0; A63,i = 0, b63 = 0. (30) at bus-23 and bus-24
V22 − V23 V23 − V24 V24 − V25
Thereafter, = = . (32)
Z 22−23 Z 23−24 Z 24−25
a) If there are less than two or two ZIBs together,
constraint (8) ensures their observability. For exam- However, in the case of three ZIBs together,
ple, for a single ZIB as z = 23, constraint (29) as shown in Fig. 4(c), the KCL rule at bus-56, bus-
ensures not to keep a D-PMU at 23 and constraint 57, and bus-58 does not solve the purpose.
(8) ensures indirect observability of 22 and 24 in b) Therefore, if there are more than two ZIBs
the worst case (V22 and V24 are known). Thus, together, the one extreme bus (non-ZIB at the
the voltage phasor at bus-23 can be calculated end of ZIBs) must have a D-PMU. Constraints

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 CHAUHAN AND SODHI: PLACEMENT OF DISTRIBUTION-LEVEL PHASOR MEASUREMENTS 3457

TABLE V
D-PMU R ESULT P ERFORMANCE OF MILP2 AND PM

(17) and (18) can be written as follows, for ze = 59

C59,59 = 1, d59 = 1; A59,i = 0, b59 = 0 .
(33)
In this case, a D-PMU at bus-59 will give V59 and
I58−59 . Applying KCL
I58−59 = I57−58 = I56−57 = I55−56. (34)
Thus, the voltage phasor at 56, 57, and 58 can be
calculated as
V58 = V59 + Z 58−59 I58−59
V57 = V58 + Z 57−58 I57−58 (35)
V56 = V57 + Z 56−57 I56−57 . (36)
Likewise, having incorporated all the ADN constraints
in (8)–(18) in the IEEE 69-bus system, 26 D-PMU locations
are resulted by the PM. These locations are indicated with red
colored stars in the single line diagram of the system in Fig. 5.
The D-PMU results of the CILP, MILP1, and MILP2 are
also indicated in Fig. 5 with yellow, blue, and green stars,
respectively.
Finally, Table V shows the results of the overall PM as
compared with the results of MILP2. Table V reveals that the
minimum number of D-PMUs required making systems fully
observable with ADN constraints is 26, 31, 34, and 47 for
the IEEE 69, IN 85, UK 95, and IEEE 123, respectively.
Again, with the same number of D-PMUs, PM yields more
TSO than the MILP2. Thus, when the problem is formulated
with the same set of observability constraints (basic or ADN),
Fig. 5. Single line diagram of 69-bus system. Black circle: ZIBs. Green circle:
the proposed objective function results in an enhanced TSO, DG. Dashed line: tie-switches. Solid line: sectionalizing switches. Yellow star:
while minimizing the number of D-PMUs needed to com- D-PMU location given by CILP. Red star: D-PMU location given by PM. Blue
pletely observe the system. star: D-PMU location given by MILP1. Green star: D-PMU location given by
MILP2.
In addition, it is observed that in the case of UK 95-bus sys-
tem, the inclusion of ADN constraints leads to reduction in the
total number of D-PMUs required to make the system observ- the substation bus get equipped with a D-PMU, some amount
able, i.e., MILP1 results in 35 and PM results in 34 numbers of TSO gets automatically fixed based on the total number
of D-PMUs. Since there are no tie-switches in the IN 85 and of connected buses to the DG as well as substation buses.
UK 95-bus systems and all the ZIBs have degree greater Thereafter, when the overall problem is solved, it led to an
than 2 in IN 85, there will be no effect on ZIB constraints. overall less number of D-PMUs, i.e., 34, and consequently,
However, position is reallocated because of the substation and a marginally reduced TSO, i.e., 123. However, the PM always
the DG constraints. In UK 95, the ZIBs having degree 2 or less results in a TSO higher than CILP.
than 2 are 2,11,12,44,45,70,71,75,77,78,81,83,85,88,91,93. It The D-PMU placement results also reveal that the sensi-
is observed that TSO is reduced because of the forced/equality tivity of the TSO with respect to the number of D-PMUs is
constraints of (10)–(14) on the DG and substation bus observ- highly dependent on system topology, i.e., how each node is
ability. Since the PM ensures that the DG buses as well as connected to other nodes and the type of bus, i.e., whether a

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 3458 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO. 6, JUNE 2020

TABLE VI
E FFECT OF R EMOVING D-PMU S ON N UMERICAL O BSERVABILITY AND A CCURACY OF ADN-SE

given bus is a DG bus, or ZIB, or a simple load bus, and so on. and two substation measurements (one active and one reac-
Consequently, different amount of increment/decrement in D- tive power-flow). In this case, it is found that removal of
PMUs has resulted in random amount of increment/decrement D-PMUs will not change the numerical observability of the
in TSO in different test systems. system, as the rank of H remains 137 (see Table VI, case b).
However, 20% removal of D-PMUs increases the SE error
B. Analysis of ADN State-Estimation Accuracy Under from 0.0020 to 0.0207 pu and 1.970e−4 to 1.90e−2 rad
Proposed Placement Scheme in the voltage magnitude and the voltage angle, respec-
tively. The ADN-SE is found numerically observable up to a
The accuracy of the ADN-SE under different uncertainty
D-PMU removal of approximately 80%. Thereafter, removal
conditions is assessed in the presence of the proposed
of any single D-PMU leads the system to be numerically
D-PMUs placement, adopting the process of validation,
unobservable. It happens because of high quantitative and low
as shown in Fig. 2. The true state values (Vtrue and φtrue )
accurate pseudomeasurements.
are obtained from load flow under different topologies and
Finally, it can be concluded that to estimate the states
loading conditions, and stored a priori. Thereafter, the absolute
of an ADN accurately, a full topological observable sys-
error of all the MCS and the maximum of absolute error for
tem is required. Numerical observability can be achieved by
k = 1, 2, ..., N bus is calculated in both the voltage magnitude
φ pseudomeasurements in the case of less number of D-PMUs,
(kv ) as well as phase angle (εk ), as follows:
but it costs the SE accuracy.
 
v
εi,k = |(Vˆi,k − Vtrue,k )|/Vtrue,k ; εmax
v
= max εkv (37) Second, the impact of various uncertainties on SE is also
φ  φ analyzed to check the robustness of the proposed placement
ε = |φˆi,k − φtrue,k |; εmax = max ε .
i,k
φ
k (38)
scenario. The effect of an individual event is considered at
The IEEE 69-bus system having a base load of 3.803 MW, a time, keeping rest of the quantities under normal operating
2.694 MVAR on a base value of 10 MW, and 12.66 kV is used conditions.
to demonstrate the ADN-SE results. The number of MCS is Case 2 (Effect of Network Reconfiguration): The reconfig-
set to be 500. uration of the system is performed using fixed number of tie-
Case 1 (Assessment of Numerical Observability): First, switches (see Table II) and N − 1 number of sectionalizer
the numerical observability of the system is ensured in this switches. The performance is analyzed based on the depth of
post-D-PMU placement stage, where all the placed D-PMU unobservability (DU) [22]. The DU of a node is calculated as
measurements are used to calculate the measurement Jacobian the maximum node distance from another node having indirect
matrix, H. For an N-bus system to be numerical observable, observability. The four network topologies are analyzed to
H should be a full-ranked matrix, which is also a prerequisite ensure the radial topology of the network . It is observed that
for the successful convergence of the ADN-SE program. SE accuracy degrades in the case of loss of observability. The
v ), max( ε φ ) is found to be 2e−3 pu and 2.03e−4 rad,
max(εmax
While running the ADN-SE program with D-PMU placement, max
the measurement Jacobian matrix is found to be full ranked respectively, in the case of full topology. The max DU is found
for all the test system, thereby ensuring the numerical observ- as the one during reconfiguration. The maximum estimation
ability too. error in some topologies having DU equal to 1 is noticed as
For the IEEE 69-bus system, the rank of H is found as full 0.0027 pu and 0.0021 rad.
as 137, in the presence of all D-PMUs. However, it can be Case 3 (Effect of Measurement Uncertainties): The effect of
noted from Table VI that removing a single D-PMU decreases pseudo-ME (PME) on SE is shown in Table VII in two differ-
the rank of H to 135 and the system becomes numerically ent cases. The PME is varied from 50% to 70%. It is observed
unobservable. It happens because of low measurement redun- that SE accuracy remains approximately constant with respect
dancy. to increment in PME, in the case of full observability. The
After ensuring the numerical observability of the system increment in the PME creates a significant error in the SE
with only D-PMU measurements, the measurement redun- not having full topological observability. For example, in the
dancy is further increased by including 138 pseudomeasure- case of removing 20% D-PMU, the max error increases from
ments (69 active and 69 reactive power-flow measurements) 0.081 to 0.1174 pu by increasing pseudo-ME up to 70%.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 CHAUHAN AND SODHI: PLACEMENT OF DISTRIBUTION-LEVEL PHASOR MEASUREMENTS 3459

TABLE VII
E FFECT OF P SEUDOMEASUREMENT U NCERTAINTY

TABLE VIII
E FFECT OF D-PMU S ’ M EASUREMENT U NCERTAINTY

TABLE IX
S UMMARY OF C OMPARATIVE A NALYSIS

Fig. 6. Mean, minimum, and SD of estimation error with (black line)/without

(gray line) direct observability of DG bus.

The impact of phasor MEs on the SE accuracy in the

case of fully observable system and 20% D-PMU removed Finally, Table IX presents a comparative performance
is shown in Table VIII. As per the synchrophasor std. analysis of the PM with some of the existing
IEEE C37.118.1a-2014, for static studies, 1% total vector error distribution-level MD placement methods, including PMUs.
(TVE) is selected, and under dynamic conditions, 3% TVE The methods in [16] and [18] minimize the SE error using
is considered. It is observed that the increment in the TVE fixed MDs, while the cost of the devices is minimized using
increases the SE error. However, the effect is found to be less the SE error as a constraint in [22] and [23], and the cost of
in the case of a full observable system than that of 20% D- MDs ensuring full SO is carried out in [24] and the PM.
PMU removed. Unlike [16], [18], [22], and [23], the PM and binary integer
Case 4 (Effect of DG Penetration): The impact of direct linear programming (BILP) [24] do not use SE accuracy in
monitoring of a DG bus via D-PMU is also examined. The problem formulation, thereby resulting in a rather simple,
wind energy-based DG of a 7% capacity of total load is fixed computationally efficient topological observability formula-
at bus-16, and its uncertainty is modeled as GMM [18]. The tion. This, in turn, demands a higher number of MDs to make
effect of with (solid line) and without (dashed line) direct the system fully observable than other methods. Moreover,
monitoring of DG is shown in Fig. 6. It is noted that by it increases the measurement redundancy and results in low
direct monitoring of a DG bus, the estimation error diminishes error in SE. Although, as compared with BILP, the PM uses
approximately ten times. Moreover, having a D-PMU on DG only D-PMU placement that results in a higher number of
bus can help in phasor-based islanding detection. PMUs but reduces the requirement of other MDs. It can be
Finally, it can be concluded that various factors such as concluded that considering the full SO results in a slightly
measurement inaccuracies, load change, network reconfigura- more number of D-PMUs (more than N/3), however, ensures
tion, and DG uncertainty affect the SE accuracy significantly. a high measurement redundancy and robust ADN-SE.
However, in the case of fully observable ADNs, the impact of
these factors on ADN-SE becomes insignificant, which clearly V. C ONCLUSION
attributes to the low ME and the high measurement redundancy This article presents a simple approach for the placement of
provided by the proposed D-PMU locations. distribution-level phasor measurements in the ADNs. To this

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.
 3460 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 69, NO. 6, JUNE 2020

end, a novel objective function, along with simple equality [12] R. Singh, B. C. Pal, and R. B. Vinter, “Measurement placement in
and inequality constraints, is formulated while considering distribution system state estimation,” IEEE Trans. Power Syst., vol. 24,
no. 2, pp. 668–675, May 2009.
the effect of DGs, ZIBs, and tie-switches of the ADNs. [13] R. Singh, B. C. Pal, R. A. Jabr, and R. B. Vinter, “Meter place-
The impact of various measurement uncertainties on D-PMU- ment for distribution system state estimation: An ordinal optimization
assisted SE is thereafter analyzed, leading to the following approach,” IEEE Trans. Power Syst., vol. 26, no. 4, pp. 2328–2335,
Nov. 2011.
salient conclusions. [14] T. C. Xygkis, G. N. Korres, and N. M. Manousakis, “Fisher information-
1) The proposed D-PMUs placement strategy simultane- based meter placement in distribution grids via the D-optimal experi-
mental design,” IEEE Trans. Smart Grid, vol. 9, no. 2, pp. 1452–1461,
ously ensures minimum cost as well as maximum SO Mar. 2018.
of the ADN. [15] T. C. Xygkis and G. N. Korres, “Optimized measurement allocation
2) Due to the radial nature and less degree of a particular for power distribution systems using mixed integer SDP,” IEEE Trans.
Instrum. Meas., vol. 66, no. 11, pp. 2967–2976, Nov. 2017.
node in the ADNs, the number of D-PMUs to make a [16] Y. Yao, X. Liu, and Z. Li, “Robust measurement placement for distri-
DN fully topological observable is found to be lying in bution system state estimation,” IEEE Trans. Sustain. Energy, vol. 10,
the range of ceil(N/3) to floor(N/2.7), for the given five no. 1, pp. 364–374, Jan. 2019.
[17] M. G. Damavandi, V. Krishnamurthy, and J. R. Marti, “Robust meter
test systems viz., IEEE 33, IEEE 69, IN 85, UK 95, and placement for state estimation in active distribution systems,” IEEE
IEEE 123. Trans. Smart Grid, vol. 6, no. 4, pp. 1972–1982, Jul. 2015.
3) The maximum depth of unobservability is found as 1 in [18] H. Wang, W. Zhang, and Y. Liu, “A robust measurement placement
method for active distribution system state estimation considering
reconfiguration cases. network reconfiguration,” IEEE Trans. Smart Grid, vol. 9, no. 3,
4) SE results show that locations of D-PMUs, as resulted pp. 2108–2117, May 2018.
by the PM, play a crucial role in obtaining accurate [19] R. Singh, B. C. Pal, and R. A. Jabr, “Distribution system state estimation
through Gaussian mixture model of the load as pseudo-measurement,”
ADN-SE. IET Generat., Transmiss., Distrib., vol. 4, no. 1, pp. 50–59, 2010.
5) A D-PMU-based topologically observable ADN is neg- [20] J. Liu, J. Tang, F. Ponci, A. Monti, C. Muscas, and P. A. Pegoraro,
ligibly affected by various measurement uncertainties, “Trade-offs in PMU deployment for state estimation in active distri-
bution grids,” IEEE Trans. Smart Grid, vol. 3, no. 2, pp. 915–924,
DG penetration, and network reconfiguration. Jun. 2012.
[21] J. Liu, F. Ponci, A. Monti, C. Muscas, P. A. Pegoraro, and S. Sulis,
R EFERENCES “Optimal meter placement for robust measurement systems in active
distribution grids,” IEEE Trans. Instrum. Meas., vol. 63, no. 5,
[1] A. Angioni, J. Shang, F. Ponci, and A. Monti, “Real-time monitoring pp. 1096–1105, May 2014.
of distribution system based on state estimation,” IEEE Trans. Instrum. [22] S. Prasad and D. M. V. Kumar, “Robust meter placement for active
Meas., vol. 65, no. 10, pp. 2234–2243, Oct. 2016. distribution state estimation using a new multi-objective optimisation
[2] A. Angioni, T. Schlösser, F. Ponci, and A. Monti, “Impact of pseudo- model,” IET Sci., Meas. Technol., vol. 12, no. 8, pp. 1047–1057,
measurements from new power profiles on state estimation in low- 2018.
voltage grids,” IEEE Trans. Instrum. Meas., vol. 65, no. 1, pp. 70–77, [23] S. Prasad and D. M. V. Kumar, “Trade-offs in PMU and IED deployment
Jan. 2016. for active distribution state estimation using multi-objective evolutionary
[3] C. Muscas, S. Sulis, A. Angioni, F. Ponci, and A. Monti, “Impact algorithm,” IEEE Trans. Instrum. Meas., vol. 67, no. 6, pp. 1298–1307,
of different uncertainty sources on a three-phase state estimator for Jun. 2018.
distribution networks,” IEEE Trans. Instrum. Meas., vol. 63, no. 9, [24] S. Teimourzadeh, F. Aminifar, and M. Shahidehpour, “Contingency-
pp. 2200–2209, Sep. 2014. constrained optimal placement of micro-PMUs and smart meters in
[4] A. von Meier, E. Stewart, A. McEachern, M. Andersen, and microgrids,” IEEE Trans. Smart Grid, vol. 10, no. 2, pp. 1889–1897,
L. Mehrmanesh, “Precision micro-synchrophasors for distribution sys- Mar. 2019.
tems: A summary of applications,” IEEE Trans. Smart Grid, vol. 8, [25] Z. Wu et al., “Optimal PMU placement considering load loss and relay-
no. 6, pp. 2926–2936, Nov. 2017. ing in distribution networks,” IEEE Access, vol. 6, pp. 33645–33653,
[5] K. Chauhan, M. V. Reddy, and R. Sodhi, “A novel distribution- 2018.
level phasor estimation algorithm using empirical wavelet trans- [26] L. Huang, Y. Sun, J. Xu, W. Gao, J. Zhang, and Z. Wu, “Optimal PMU
form,” IEEE Trans. Ind. Electron., vol. 65, no. 10, pp. 7984–7995, placement considering controlled islanding of power system,” IEEE
Oct. 2018. Trans. Power Syst., vol. 29, no. 2, pp. 742–755, Mar. 2014.
[6] P. A. Pegoraro, A. Meloni, L. Atzori, S. Sulis, and P. Castello, [27] B. Gou, “Optimal placement of PMUs by integer linear programming,”
“PMU-based distribution system state estimation with adaptive accuracy IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1525–1526, Aug. 2008.
exploiting local decision metrics and IoT paradigm,” IEEE Trans. [28] R. F. Nuqui and A. G. Phadke, “Phasor measurement unit placement
Instrum. Meas., vol. 66, no. 4, pp. 704–714, Apr. 2017. techniques for complete and incomplete observability,” IEEE Trans.
[7] N. M. Manousakis, G. N. Korres, and P. S. Georgilakis, “Taxonomy Power Del., vol. 20, no. 4, pp. 2381–2388, Oct. 2005.
of PMU placement methodologies,” IEEE Trans. Power Syst., vol. 27, [29] R. Singh, B. C. Pal, and R. A. Jabr, “Choice of estimator for distribution
no. 2, pp. 1070–1077, May 2012. system state estimation,” IET Gener., Transmiss. Distrib., vol. 3, no. 7,
[8] F. Aminifar, M. Fotuhi-Firuzabad, A. Safdarian, A. Davoudi, and pp. 666–678, Jul. 2009.
M. Shahidehpour, “Synchrophasor measurement technology in power [30] IEEE Standard for Synchrophasor Measurements for Power
systems: Panorama and state-of-the-art,” IEEE Access, vol. 2, Systems—Amendment 1: Modification of Selected Performance
pp. 1607–1628, 2014. Requirements, IEEE Standard C37.118.1a-2014, Apr. 2014, pp. 1–25.
[9] D. K. Mohanta, C. Murthy, and D. S. Roy, “A brief review of phasor [31] A. Enshaee, R. A. Hooshmand, and F. H. Fesharaki, “A new
measurement units as sensors for smart grid,” Electr. Power Compon. method for optimal placement of phasor measurement units to main-
Syst., vol. 44, no. 4, pp. 411–425, 2016. tain full network observability under various contingencies,” Electric
[10] D. D. Giustina, M. Pau, P. A. Pegoraro, F. Ponci, and S. Sulis, Power Syst. Res., vol. 89, pp. 1–10, Aug. 2012. [Online]. Available:
“Electrical distribution system state estimation: Measurement issues and http://www.sciencedirect.com/science/article/pii/S0378779612000405
challenges,” IEEE Instrum. Meas. Mag., vol. 17, no. 6, pp. 36–42, [32] K. P. Schneider et al., “Analytic considerations and design basis for the
Dec. 2014. IEEE distribution test feeders,” IEEE Trans. Power Syst., vol. 33, no. 3,
[11] A. Shafiu, N. Jenkins, and G. Strbac, “Measurement location for pp. 3181–3188, May 2018.
state estimation of distribution networks with generation,” IEE [33] C. Muscas, M. Pau, P. A. Pegoraro, and S. Sulis, “Uncertainty of voltage
Proc.—Generat., Transmiss. Distrib., vol. 152, no. 2, pp. 240–246, profile in PMU-based distribution system state estimation,” IEEE Trans.
Mar. 2005. Instrum. Meas., vol. 65, no. 5, pp. 988–998, May 2016.

Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY SILCHAR. Downloaded on December 23,2020 at 20:32:18 UTC from IEEE Xplore. Restrictions apply.