Electrical Power and Energy Systems 124 (2021) 106332

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

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

A complete state estimation algorithm for a three-phase four-wire low T

voltage distribution system with high penetration of solar PV☆
Wele Gedara Chaminda Bandaraa, , Dilini Almeidaa, Roshan Indika Godaliyaddaa,
⁎

Mervyn Parakrama Ekanayakea, Janaka Ekanayakea,b

a
Department of Electrical and Electronic Engineering, Faculty of Engineering, University of Peradeniya, Peradeniya 20400, Sri Lanka
b
School of Engineering, Cardiff University, The Parade, Cardiff CF24 3AA, United Kingdom

ARTICLE INFO ABSTRACT

Keywords: Low Voltage Distribution Grids (LVDGs) become highly unbalanced due to the advent of single-phase solar PV
State estimation plants. As a result, the voltage and current levels of the neutral conductor show a significant increase. Therefore,
Admittance matrix monitoring of the entire state of the network is essential. However, the existing state estimation algorithms
Solar PV estimate voltage states of the phase conductors while ignoring the state of the neutral conductor. This paper
Energy management system
presents a novel approach to estimate the complete state of the LVDGs. A novel state reduction method was
Low voltage distribution grids
introduced to model the three-phase four-wire feeder line using a 3 × 3 admittance matrix, which incorporates
the neutral coupling effect on phase conductors. Next, the reduced admittance matrix together with the linear
approximations of active and reactive power functions were combined to formulate the Low Voltage-Linear State
Estimation (LV-LSE) algorithm. Finally, the performance of LV-LSE algorithm was analyzed for different mea­
surement uncertainties, scales of line lengths of the network, and data-loss conditions. Results show that, for all
the cases, LV-LSE algorithm together with the proposed reduction method can estimate voltage states with an
average maximum voltage magnitude error of less than 4.32 × 10 3 pu and current states with an average
maximum current magnitude error of less than 1.81 × 10 3 pu.

1. Introduction PV systems, bidirectional power flows can be seen on the LVDGs, thus
causing adverse consequences. Some of the possible impacts are voltage
It is now recognized that one of the main contributors to climate rise, voltage fluctuations and unbalances, thermal overloads, higher
change is the higher percentage of emissions from conventional power levels of harmonics etc. The severity of these impacts depends on the
generating sources. This has encouraged the governments around the configuration of the distribution network, penetration level, and the
world to constitute several targets for the integration of Renewable location of the PV node in the distribution network.
Energy Sources (RESs) into the grid. Out of different RESs, the capital
cost of solar photovoltaic (PV) continues to fall at a considerably higher 1.1. Impact of high intake of solar PV
rate. According to a recent report by the regulator of Australia, AEMO,
it has been forecasted that solar PV would be the main contributor to A study on investigating the impact of high intake of solar PV was
electricity generation by 2050 [1]. reported in [2]. A Monte Carlo based methodology was developed and
During the last decade, the number of PV plants connected to LVDGs applied for a typical unbalanced residential network (Fig. 1) in Sri
have shown a rapid increase. However, the integration of scattered PV Lanka using a three-phase, four wire LV network model.
plants instigates additional challenges to the existing LVDGs and those Under the Monte Carlo implementation, pre-defined number of PV
challenges become aggravated with the increasing level of PV pene­ panels having randomly generated PV capacity between 2 and 7 kW
tration. was connected to randomly generated nodes and phases of the dis­
Traditionally, power systems have been designed to operate on tribution system. One hundred power flow simulations were performed
unidirectional power flows. However, due to the high proliferation of by varying the PV capacity and location, and several critical parameters

This research was supported by the National Science Foundation (NSF), Sri Lanka (grant number: RG/2018/EA & ICT/01).
☆

Corresponding author.
⁎

E-mail addresses: chaminda.bandara@eng.pdn.ac.lk (W.G. Chaminda Bandara), dilinialmeida2@gmail.com (D. Almeida),

roshangodd@ee.pdn.ac.lk (R.I. Godaliyadda), mpb.ekanayake@ee.pdn.ac.lk (M.P. Ekanayake), ekanayakej@cardiff.ac.uk (J. Ekanayake).

https://doi.org/10.1016/j.ijepes.2020.106332
Received 9 November 2019; Received in revised form 27 May 2020; Accepted 27 June 2020
0142-0615/ © 2020 Elsevier Ltd. All rights reserved.
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 1. Single line diagram of the residential network. More details about the network can be obtained from [2] and Appendix A.

were extracted from the load flow results. Then the impacts were Fig. 2 illustrates the maximum value of the variation throughout the
analyzed using voltage unbalances, neutral currents, voltage and one hundred simulations (arranged in ascending order) for neutral
thermal limits. A more comprehensive study on this is accepted for current and voltage unbalance factor for the worst-case scenario, i.e.
publication [3] and only some preliminary results are shown here to minimum loading and maximum PV generation. The variation of
highlight the importance of determining the state of the neutral con­ maximum voltage with the total installed PV capacity is shown in Fig. 3
ductor. for the same case.

Fig. 2. Variation of (a) maximum neutral current and (b) maximum voltage unbalance factor for hundred simulations.

2
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 3. Variation of maximum voltage with total installed PV capacity.

Results from the impact assessment revealed that under high pe­ estimation capabilities is presented. In [21], a two-step procedure has
netration of solar PV, depending on their location and phase, some of been introduced for multi-area state estimation in distribution systems.
the critical parameters such as voltage and voltage unbalance factor Here, the first step local estimations are refined through a WLS based
violated the statutory limits. Also, it was observed that the highly un­ second step which integrates the information available from neigh­
balanced allocation of single-phase PV units can create significantly boring zones. In [22], a three-phase admittance matrix-based DSSE
large neutral currents. model has been proposed. The proposed DSSE model tackles issues
In order to mitigate the aforementioned impacts of solar PV and related to consistency, zero-injections, and inclusion of voltage mea­
thereafter to ensure optimal, secure, reliable and coordinated operation surements. The real and imaginary parts of the bus voltages are utilized
of LVDGs, an Energy Management System (EMS) is crucial. State as state variables while power and voltage measurements are converted
Estimation (SE) is considered as an essential element of the EMS as it is into equivalent currents and voltages. Reference [23] proposed a DSSE
not possible to perform real-time monitoring and control without model using particle swarm optimization to estimate on-load tap
knowing the state of the entire LVDG [4]. changing for voltage control. The proposed method utilized both dis­
crete and continuous variables to estimate transformer tap positions.
However, due to the different grounding configurations, the asym­
1.2. Distribution System State Estimation (DSSE)
metrical nature of loads, distributed energy sources, and the number of
available measurements, will limit the universal applicability of the
Distribution System State Estimation (DSSE) algorithms process raw
aforementioned DSSE models proposed in the existing literature.
measurements (including voltage and current magnitudes, active and
References [20–23] can be utilized for LVDGs with a multi-grounded
reactive power measurements) and additional pseudo values to esti­
neutral configuration where neutral voltages are zero and can be
mate the actual state of a power system. DSSE enables real-time dis­
naturally excluded. Successively, an original admittance matrix can be
tribution grid monitoring and provides the initial condition of the
reduced to 3 × 3 using the Kron’s reduction. However, rather than
power system for a variety of applications including voltage control,
multi-grounded neutral, in some countries, the neutral conductor is
feeder reconfiguration and Demand Side Management (DSM).
only grounded at the secondary side of the MV-LV transformer, which
Algorithms developed for DSSE were widely adapted from estima­
renders non-zero potential across the neutral conductor. Furthermore,
tion techniques developed for Transmission System State Estimation
the recent deployment of proliferated distributed energy sources re­
(TSSE) that have been developed for over six decades [5]. When con­
sulted in high levels of neutral currents and voltages as studied in this
sidering TSSE, the Weighted Least Squares (WLS) approach is the most
paper. This warrants increased interest in three-phase DSSE models that
popular SE algorithm among the many alternatives proposed [6–11].
incorporates the neural conductor states. However, incorporating the
The common choices for state variables are node voltage or branch
neutral conductor states in a DSSE model increases the number of states
current [12,13]. There is always an assumption of a “balanced system”
to be estimated and increases the size of the Jacobian matrix, thus in­
at transmission-level. This allows for the decoupling of the three-phase
troducing a significant computational burden. Nevertheless [19] con­
system, making in per-phase positive sequence network data being
sidered a three-phase DSSE algorithm with explicit neutral configura­
sufficient for analysis [14–18].
tion. In this reference, a multi-phase DSSE model has been proposed for
In contrast, distribution grids under radial and weakly-meshed op­
distribution systems with a three-phase four-wire configuration. The
erations have highly unbalanced three-phase branches due to numerous
proposed DSSE model utilized two types of state reduction strategies to
combinations of unequal single-phase, two-phase, three-phase loads
improve its computational performance. The first is the use of the KCL
and small-scale distributed generators. In distribution networks, real-
theorem at neutral to represent neutral voltage as a linear combination
time measurements are limited and network observability is not
of non-neutral phases. The second is the elimination of voltages at zero
achieved unless pseudo values and network reduction techniques are
injection phases by linearly combining voltages of non-zero neutral and
used. These distinct features of distribution systems prohibit the direct
non-zero injection phases. Consequently, only voltages of non-neutral
application of mature state estimation methods developed for TSSE
and non-zero injection phases are used as state variables to reduce the
[19,5].
scale of the DSSE model. Moreover, different types of measurements
In addition to the state estimators proposed for transmission sys­
such as phase-to-ground voltages, phase-to-neutral voltages, neutral
tems, three-phase DSSE that accommodates unsymmetrical character­
currents, and non-neutral currents are utilized to achieve full network
istics of distribution systems are reported in the literature. In [20], a
observability. However, the aforementioned DSSE model has not been
multi-phase DSSE that calculates system states by utilizing a con­
generalized as an admittance matrix reduction technique that can be
strained WLS method while including standard three-phase state

3
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

used for any three-phase four-wire DSSE model or any load flow algo­ • An efficient sequential LV-LSE algorithm: The existing WLS based
rithm to eliminate neutral conductor states. Also, it does not estimate SE algorithms require a larger amount of measurements compared
the voltage states of neutral and zero-injection phases. It should be to the number of state variables that are required to be estimated by
stressed that, the estimation of complete state estimation is essential for a factor of 1.7 to 2.2. In most of the practical situations, available
the Distribution System Operators (DSOs) for highly unbalanced dis­ measurements are limited due to the limited number of smart meters
tribution systems. In this paper, solutions to all these above issues are installed, unavailability of smart meters, communication errors at a
addressed. given time, and possibility of data loss during transmission. The
With the recent deployment of smart meters and distribution system linearized active and reactive power functions, the assignment of
measuring infrastructure, the visibility of LVDGs was enhanced. smart pseudo values for loads, voltage magnitudes and angles, and the
meters are installed at the connection points of some customers’ that are sequential nature of estimating voltage states allow the proposed
capable of monitoring the real-time active and reactive power flows and LV-LSE algorithm to operate effectively, even with 60% availability
additional information required by the DSOs such as voltage and cur­ of smart meter data.
rent [24–26]. Different types of DSSE techniques that utilized such
smart meter data have been proposed in recent literature. In [25] the
technical feasibility of using smart meters and their measurements for 2. Materials and methods
LV network observability and controllability through SE techniques
have been analyzed. In [26,27], LV SE through real-time measurements 2.1. LV feeder model
of smart meters have been proposed. In [28], the impact of different
uncertainty sources such as smart meters, non-synchronized measure­ 2.1.1. Representation of three-phase feeder lines as a 4 × 4 admittance
ments and synchronized measurements provided by Phasor Measure­ matrix
ment Units (PMSs) on three-phase SE has been analyzed. In [29], a Most of the distribution lines in LVDGs consist of four conductors
smart control system for LVDGs has been designed using an automatic (thee phase conductors and one neutral conductor). The mutual cou­
evaluation of critical voltage nodes based on power snapshot data pling between each conductor is significant since the conductors are
collected from smart meters. A three-phase SE algorithm which utilized arranged in geometrically close distances and carry significant amount
customers’ energy bills to calculate average demands and a three-phase of current through them. Fig. 4 shows the series and mutual impedances
load flow algorithm to generate pseudo-measurements of voltage of a three-phase distribution line. The method developed by Carson
magnitudes, active and reactive power injections which ensures com­ [34] can be used to accurately compute series and mutual impedances
plete observability and a low investment cost for application in a typical for both overhead and underground distribution lines. The application
distribution system has been proposed in [30]. In [31], a three-phase of Carsons’s equation has become the standard for computation of line
four-wire state estimation algorithm has been proposed by utilizing impedances. Because Carson’s equation has resulted in an infinite
(4 × 4) admittance matrix. However, to avoid numerical instabilities in series, a modified version of Carson’s equations have been proposed in
this method, the sum of currents are assumed to be zero at all nodes of [35]. The modified Carson’s equations compute series and mutual im­
the LV network, even at the non-injection nodes. This is not a realistic pedance values with an error less than 0.3% compared with the values
assumption when LVDGs are highly unbalanced. obtained using the original Carson’s equations. Therefore, the modified
Smart meters can provide real-time measurements unless there are Carson’s equations were used to calculate series and mutual impedance
no communication errors [32]. Ideally, an LVDG with 100% smart values. The modified Carson’s equations for a overhead cable assuming
meter coverage, where all smart meters can upload their measurements a frequency of 50 Hz and a ground resistivity of 100 m are:
in real-time, can directly estimate the total load of the LVDG. However,
in reality, due to the fact that a large number of smart meters are 0.3048
z ii = 0.0493 + ri + j0.0628 ln + 8.0251 /km
sparsely distributed in an LVDG, only a small fraction of smart meters GMRi (1)
can report their voltage and power values at the same time. In addition,
the information collection procedure of smart meters is subjected to
data loss because smart meters use wireless signals to transmit data 0.3048
z ij = 0.0493 + j0.0628 ln + 8.0251 /km
from smart meters to an Access Point (AP) [33]. Due to these practical Dij (2)
problems with smart meters, the amount of measurements in an LVDG
is far from enough to guarantee observability. Thus, the DSSE model where, z ii is the self impedance of the i -th conductor ( / km ), z ij is the
based on linearized active and reactive power functions proposed in mutual impedance between i -th conductor and j -th conductor ( / km ),
this paper, which provides complete state estimation of the network, ri is the ac resistance of the i -th conductor ( ), GMRi is the geometric
even under 60% availability of smart meters tackles the limited mea­ mean radius of the i -th conductor (m) and Dij is the distance between
surement availability problem more effectively compared to existing SE i -th conductor and j -th conductor (m) . Therefore, the impedance matrix
techniques. Further, the performance of the proposed LV-LSE algorithm of the unbalanced three-phase four wire system was formulated as,
was analyzed under different levels of smart meter data loss to de­
monstrate its robustness under such adverse conditions.

1.3. Objective and contribution of the paper

The contributions of this paper are summarized as follows:

• An accurate reduction technique to model three-phase four-wire

system by incorporating the neutral coupling effect on phase con­
ductors. The use of Kron’s reduction to simplify LV feeder lines leads
to inaccurate results in the monitoring of LVDGs, especially when
the system is highly unbalanced. Therefore, the proposed reduction
method is a more generalized approach to simplify the three-phase
four-wire system and significantly improves the accuracy of the LV-
LSE algorithm. Fig. 4. Impedance of LV three-phase feeder section [17,36].

4
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

z aa z ab z ac z an
Z abcn =
z ba
z ca
z bb
z cb
z bc
z cc
z bn
z cn
(1 + ) yan
ynn
yan
ynn
yan
ynn
IaAB
zna znb znc znn (3)
ybn
ynn (1 + ) ybn
ynn
ybn
ynn
IbAB
IcAB
where, the diagonal elements in Z abcn represents the self-impedance (z ii )
of each conductor and the off-diagonal elements represents the mutual-
ycn
ynn
ycn
ynn (1 + ) ynn
ycn

impedance (z ij ) between the two conductors. Z abcn matrix relates the

voltage drop between each conductor and the phase currents. From the
(y aa
yan yna
ynn ) (y ) (y
ab
yan ynb
ynn ac
yan ync
ynn )
Z abcn, Y abcn was computed by taking the matrix inverse of Z abcn . = (y ba
ybn yna
ynn ) (y ) (y
bb
ybn ynb
ynn bc
ybn ync
ynn )
2.1.2. Representation of three-phase feeder line as a 3 × 3 admittance (y ca
ycn yna
ynn ) (y ) (y
cb
ycn ynb
ynn cc
ycn ync
ynn )
matrix
VaAB
Most of the SE algorithms then reduce 4 × 4 admittance matrix to a
simpler form of 3 × 3 under several assumptions. The most common VbAB
reduction method is to use Kron’s reduction. Kron’s reduction [37] can VcAB (10)
be used to simplify 4 × 4 admittance matrix to a 3 × 3 admittance
matrix by decoupling the neutral conductor from phase conductors. Let’s denote Eq. (10) as,
Kron’s reduction achieves this by assuming the potential of the neutral AB
L ·Iabc AB
= Y abc· Vabc (11)
conductor is zero. However, this assumption is not valid in most of the
LVDGs. The primary objective of this section is to develop an accurate where, = AB
Iabc [IaAB IbAB IcAB ]T ,
=[ AB
Vabc L is a VaAB VbAB VcAB ]T ,
reduction method to simplify 4 × 4 admittance matrix into a 3 × 3 ad­ complex valued scalar matrix and Y abc is exactly the same as in the
mittance matrix by accounting for the possible non-zero potential of the Kron’s reduction (Iabc
AB AB
= Y abc Vabc ) . As can be noted, the introduction of
neutral conductor. the L matrix is the innovation of the proposed method and it in­
The 4 × 4 admittance matrix relates the phase currents and voltage corporates the neutral conductor coupling effects on the phase con­
drop of the line section as, ductors. In addition, it can be clearly seen that the matrix L has a full
rank, hence the inverse of L always exists. Therefore, the proposed
IaAB yaa yab yac yan VaAB reduction method was written as,
IbAB yba ybb ybc ybn VbAB
= yca ycb ycc ycn
AB
Iabc AB
= (L 1Y abc )· Vabc AB
= Y abc· Vabc (12)
IcAB VcAB
yna ynb ync ynn
InAB VnAB (4) The matrix: Y abc
=L defined in (12) is the reduced version of the
1Y abc

Y abcn in (4), while including the effects of the neutral conductor on

where, IaAB , IbAB , IcAB and InAB denote the phase currents and phase conductors. Let’s denote the elements of the Y abc as,
VaAB, VbAB, VcAB and VnAB denote the voltage drop across each phase
between A and B points (see Fig. 4). Therefore, current through the y¯aa y¯ab y¯ac
phase-m {a , b , c, n} was written as, Y abc = y¯ba y¯bb y¯bc
y¯ca y¯cb y¯cc (13)
ImAB = ymk · VkAB
k = a, b, c, n (5)

where, (5) generates a set of four equations for phases-a, b, c , and n . 2.1.3. Kron’s reduction
The neutral current (InAB ) was approximated as, Kron’s reduction assumes there is no significant current through the
neutral conductor (i.e. a balanced three-phase system). Hence, voltage
InAB I AB
(6) drop across the neutral conductor can be approximated as zero. For this
= a, b, c
reason, the conventional SE algorithms assume that the neutral con­
The negative sign in the (6) indicates that the direction of neutral ductor is grounded at both ends of the distribution line (see Fig. 5) and
current flow is opposite to the current flow of phase conductors. By that is represented as a zero potential across the neutral conductor of
equating (5) when m = n and (6), the whole system. This assumption allows the Kron’s reduction to
eliminate the neutral conductor and reduce 4 × 4 admittance matrix
InAB I AB = ynk · VkAB (Y abcn ) to a simpler form of 3 × 3 (Y abc ) [37–39],
= a, b, c k = a, b, c , n (7)

Therefore,

I AB ynk · VkAB
= a, b, c k = a, b, c
VnAB =
ynn (8)

By substituting to VnAB in (5) from (8) and after simplification for

m {a , b , c} , generate three set of equations as,

ymn AB ymn AB ymn ynq

1+ Im + Ip = ymq VqAB
ynn p = m1, m2
ynn q = a, b, c
ynn (9)

where, in the equation for phase m , variables m1 and m2 denotes the

other two phases. For example, if m = b then m1, m2 = a, c . Thereafter,
by rearranging the set of three equations generated in (9) for
m {a , b , c} in matrix form, Fig. 5. A four-wire multi-grounded distribution line.

5
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

sections.
(y aa
yan yna
ynn ) (y
ab
yan ynb
ynn ) (y
ac
yan ync
ynn )
Y abc = (y ba
ybn yna
ynn ) (y
bb
ybn ynb
ynn ) (y
bc
ybn ync
ynn ) 2.2.3. Linearization of the measurement function
Generally, the measurement function (h (·) ) is nonlinear for all
(y ca
ycn yna
ynn ) (y
cb
ycn ynb
ynn ) (y
cc
ycn ync
ynn ) (14)
measurements (active and reactive power flows, current magnitudes,
etc.) except for voltage magnitudes. In this section, the nonlinear active
where the symbols have the same meanings as defined previously. Due and reactive power functions were linearized using the Taylor series
to the “balanced” assumption made in the Kron’s reduction, the accu­ approximations of sin(·) and cos(·) terms.
racy of the SE results significantly decreases specially when the system In Section 2.1.2, three-phase, four-wire line was modeled using a
is highly unbalanced. In contrast, the proposed reduction technique 4 × 4 admittance matrix and then reduced to a 3 × 3 admittance matrix
(Y abc ) incorporates the neutral coupling effects on phase conductors (Y ) while accounting for the effects of the neutral conductor. Moreover,
and thereby allows SE algorithms to perform well under severely un­ Y matrix relates the phase currents and voltage drop across the line
balance situations. segment AB using the relationship,
AB
Iabc AB
= Y abc · Vabc (18)
2.2. State estimation algorithm
The complex power flow (SmAB ) through phase-m is given by the equa­
2.2.1. The SE problem tion for m {a , b , c} , as,
The objective of the SE is to calculate consistent values for state
variables, based on a set of real-time measurements available at the SmAB = VmA × (ImAB ) = VmA y¯m V AB
EMS from the smart meters and other LV measuring infrastructure. The = a, b, c (19)
WLS SE problem was formulated as a minimization of objective func­
where, * denotes the complex conjugation.
tion ( J (x ) ) as in [40,41],
Let’s denote, mA is the voltage angle for node- A and phase-m , |VmA| is
m
the voltage magnitude for node- A and phase-m , g¯m and b̄m are the
J (x ) = wi ei2
i=1 (15) conductance and the susceptance of admittance element ȳm of the
admittance matrix Y . Therefore, (19) was expanded as,
subjected to the constraint,
z = h (x ) + e (16) SmAB = |VmA | A
m [(g¯m + jb¯m )(|V A | A
|V B | B
)]
= a, b, c
where, wi is the weighting factor for respective measurement, e is the m
dimensional measurement error vector, z is the m dimensional mea­ (20)
surement vector, consists of active and reactive power flows and bus Let’s denote,
voltage magnitudes, x is the n dimensional state vector, which consists AB A B
of all the bus voltage magnitudes and angles and h (·) is the measure­ m = m for m , {a , b , c}
ment function which relates x to z . cosmAB=cos( AB
m )
AB
and sinm =sin( AB
m ) (21)
using the above notations in (21) the active power and reactive (PmAB )
2.2.2. Measurement redundancy ( )
power (QmAB ) flows for phase-m can be expressed as (see Appendix C for
A necessary condition for the application of WLS based SE is a po­
the derivation),
sitive measurement redundancy ( > 0 ) [42–44]. In other words, the
number of available measurements (m) must be greater than the PmAB
number of state variables (n) . The above definition of the measurement
= |VmA | {|V A |[g¯m ·cosmAA + b¯m ·sinm
AA
]
redundancy can be expressed in a mathematical form as [44], = a, b, c
AB ¯
m |V B|·[g¯m ·cosm AB
+ bm ·sinm ]} (22)
= 1, n = Nphases· 2N 1
n (17)
QmAB
where, Nphases is the number of phases considered in the SE algorithm AA ¯
= |VmA | {|V A |[g¯m ·sinm AA
bm ·cosm ]
and N is the total number of nodes in a network. For example: = a, b, c
Nphases = 1 for per-phase SE in transmission networks, Nphases = 3 for
three-phase SE in distribution networks and Nphases = 4 for SE in LVDGs |V B|·[g¯m ·sinmAB b¯m ·cosm
AB
]} (23)
including the neutral conductor. The Taylor series approximations of sin(·) and cos(·) were defined as
In most of the practical cases is in the range of 0.7 to 1.2 [42]. As (only linear terms were considered),
mentioned in the introduction, achieving a positive measurement re­
dundancy is difficult in LVDGs because not each and every household is
AA AA AA AA AA
sinml =sin( ml ) sin( ml, r ) + cos( ml, r )· ml
equipped with smart meters - even the available smart meters are not AA
cosml =cos( AA AA AA AA
ml ) cos( ml, r ) sin( ml, r )· ml
highly reliable and only measures voltage magnitudes and complex AB AB AB AB AB
power flows. Therefore, in this research it was assumed that about 60% sinml =sin( ml ) sin( ml, r ) + cos( ml, r )· ml

of households are equipped with smart meters which measure voltage AB

cosml =cos( AB
ml ) cos( AB
ml, r ) sin( AB
ml, r )·
AB
ml (24)
magnitudes and complex power flows at the location where they are
installed. As can be noted from Eq. (17), the inclusion of the neutral The assignment of operating points (points where the functions are
conductor states in the SE algorithm also reduces the measurement linearized) for the voltage angles were conducted as,
redundancy in LVDGs. Hence, these practical limitations present in the 0, if (m )=0
LVDGs prohibit the use of WLS based SE algorithms to effectively de­ 2
termine the complete states of network. Therefore, a sequential LV-LSE
AA AB
= 3
, if (m )= 1 2
m ,r m ,r
was developed by utilizing the novel reduction technique as outlined in 2
, if (m )= 2 1
3 (25)
section 2.1.2 to achieve a positive measurement redundancy. The pro­
posed sequential LV-LSE algorithm is described in the following under the assignment of phase a = 1, phase b = 2 and phase c = 3.

6
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 6. Flow chart of the proposed sequential LV-LSE algorithm.

The operating points (points where the functions are linearized) for where, U is the diagonal matrix with respective measurement variances
voltage magnitudes were calculated locally for each line segment by of measuring instruments given by,
utilizing the available smart meter data to improve SE accuracy. Let’s
U = diag ( 12, 2
2,
2
3, …, 2
m) (31)
assume q number of smart meter voltage measurements (|Vi |) were re­
ceived for line segment AB. Then the local operating point for voltage
magnitude |Vr| was calculated as, 2.3. Overall procedure of LV-LSE algorithm
q
1
|Vr| |Vi | The proposed sequential LV-LSE algorithm consists of five steps,
q (26)
i=1 which is shown in Fig. 6. It also shows the location of the smart meters
Assuming the operating points for voltage angles as given by (25) and the type of measurements available to the LV-LSE algorithm for a
and voltage magnitudes as given by (26), and applying Taylor series selected Monte-Carlo simulation.
approximations on (22) and (23), the linearized approximations for the
active (P mAB ) and reactive (Qm ) power functions for phase-m were ex­
AB STEP: 1 Collect smart meter data and assign pseudo values for un­
pressed after simplifications as (see Appendix D for the derivation), known measurements: An LVDG network having single-phase
PV plants and single-phase households were considered for this
P mAB = |Vr|· P ·(|V
A
| |V B|) |Vr |2 · P ·(
A B
) study. As mentioned previously, it was assumed that all the PV
= a, b, c = a, b, c (27) plants and about 60% of single-phase customers are connected
to the LVDG through smart meters. The location, phase, and
QmAB = |Vr|· Q ·(|V
A
| |V B|) |Vr |2 · Q·(
A B
)
(28) availability of smart meters of single-phase customers and PV
= a, b, c = a, b, c
plants are shown in Fig. 6 and summarized in Table 1. In ad­
where, dition to the smart meters, three-phase power meters were
distributed among the LVDG as shown in Fig. 6 (denoted by
P =g¯m, cos( AB
m ,r ) + b¯m, sin( AB
m ,r )
the triangles).
P= g¯m, sin( AB
m ,r ) + b¯m, cos( AB
m ,r ) For the unknown single-phase loads, load pseudo values were
Q = P=g¯m, sin( AB
m ,r) b¯m, cos( AB
m ,r ) generated by utilizing available active and reactive power
measurements collected from households, PV-plants, and
Q = P =g¯m, cos(
AB
m ,r ) + b¯m, sin( AB
m ,r )
three-phase power meters. As shown in Fig. 6, the considered
After linearizing the active and reactive power flow functions LVDG network was divided into eleven regions
through phase conductors, the measurement function (h (·)) given in (denoted by Rr where r = 1, , 11) for the generation of load
(16) was represented as a linear equation for feeder section AB as, pseudo values. The net power consumption (PRNet r ,m
) in a parti­
cular region-(Rr ) for phase-m a, b , c was calculated from
z = H·x + e (29)
three-phase power meter measurements by applying the power
where, z and e have the same meanings as previously described in (16) balance equation (i.e. PRNet
r,m
= Power coming into the Rr -
and H is the linearized measurement matrix. With the application of Power going out from the Rr - Power loss in Rr ). Then, the load
WLS, the estimated state vector x was calculated with the equation pseudo values for unknown loads in the region-Rr were ran­
given by, domly generated subjected to the constraints,

x = [(H T U 1H ) 1H T U 1]· z (30)

7
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Table 1 Pmpseudo
,i = PRNet
r ,m
+ Pmpv, j Pmload
,k
Detailed description of single-phase loads and PV panels connected to the LVDG i Rr j Rr k Rr (32)
network, and the availability of smart meters at the connection points.
and,
load
Pmin Pmpseudo
,i
load
Pmax (33)
where, i Rr
Pmpseudo
,i is the sum of load pseudo values of phase-
m in region-Rr , PRNet
r, m
is the net power consumption of phase-m
in region-Rr , j Rr m, j is the total power generated by the PV
P pv

plants connected to phase-m in the region-Rr , P load is

k Rr m , k
the total power consumption of households with a smart me­
ters connected to phase-m in the region-Rr and, Pmax load
and Pminload

are the maximum and minimum active power consumption

values of a household, respectively. The maximum and
minimum possible values of single-phase loads were obtained
from prior knowledge about the LVDG. In addition, the power
factor values of single-phase households for the considered
network are generally around 0.90. Hence, the power factor
values of known loads were randomly generated between 0.85
and 0.95.
STEP: 2 Backward step: Computation of complex power flows through
the phase conductors: Once the active and reactive power
values of each household and PV plant were known, the
complex power flow (SmAB ) through each branch was computed
in the backward direction starting from the last branch and
proceeding backwards towards the root node at each step (see
the direction of blue arrows in Fig. 7).
STEP: 3 Forward step: Estimation of voltage states of phase con­
ductors: In this step, the voltage states of all the nodes were
evaluated in sequential manner starting from the secondary
side of the MV-LV transformer and then proceeding towards
the end of network at each step (see the direction of red arrows
in Fig. 7). The sequential nature of the SE algorithm allows to
achieve the requirement of positive measurement redundancy
( > 0 ) despite the low availability of smart meter data. Con­
sider the first branch of the MV-LV transformer. Let it be the
line segment AB as shown in Fig. 4. Therefore, the node-A is
equivalent to the secondary side of the transformer. The active
and reactive power flows through phase conductors
(Pa, Qa, Pb, Qb, Pc , and Qc ) are available from step 2. In addi­
tion, the phase voltage magnitudes of node A
(|VaA|, |VbA|, and |VcA|) are known to the SE algorithm from the
measuring instrument placed at the secondary side of the
transformer. The pseudo values for voltage angles for three-
phases at node-A ( aA , bA , and cA ) were assumed to be
0o, 120o and 240 o . Thus, the measurement vector z was
formulated as,
z = [|VaA|, |VbA|, |VcA|, A
a ,
A
b ,
A
c , Pa, Qa, Pb, Qb, Pc , Qc ]T (34)
by assuming, no measurements were available from node-B. If
any measurement is taken from the node-B, it should be in­
cluded in (34) to enrich the accuracy and the measurement
redundancy ( ) of LV-LSE algorithm. Then, the state vector
(x = [ VaA , aA , VbA , bA , VcA , cA , VaB , aB , VbB , bB , VcB , cB ]T ) for dis­
tribution line-AB was calculated by utilizing Eq. (30). It is
important to note that Eq. (30) can always provide the optimal
solution under the linearized condition when the length of the
measurement vector (z ) is equal to or greater than the length
of the state vector ( x ) at any given time. To ensure the above
condition, the estimated voltage states of the previous branch
will be utilized as pseudo values when formulating the mea­
surement vector for the current branch. For example, when
calculating the state vector for distribution line BC, the voltage
states determined in the previous step of node B
(|VaB|, |VbB|, |VcB|, aB , bB , cB ) were used as the pseudo-values
for unknown voltage magnitudes and angles in this step, so

8
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 7. Calculation routes for the proposed LV-LSE algorithm.

Table 2 node-B (InAB ) was calculated by the relationship,

Maximum measurement uncertainties of smart meters.
InAB = I AB
Voltage Active and Reactive Power = a, b, c (36)

Full-Scale Uncertainty Full-Scale Uncertainty

STEP: 5 Forward step: Estimation of voltage states of the neutral
Case I 300 V 0.5% 7 kW 1.0%
Case II 300 V 0.5% 7 kW 1.5%
conductor
Case III 300 V 1.0% 7 kW 1.5% In most of the LVDGs, the neutral connection is firmly
Case IV 300 V 1.0% 7 kW 2.0% grounded on the secondary side of the MV-LV transformer.
Therefore, the potential of the neutral conductor is approxi­
mately zero on the secondary side of the MV-LV transformer.
that a positive measurement redundancy can be guaranteed. Thus the voltage states of the neutral conductor was de­
By repeating the process for each proceeding line segment in termined for each node using the relation given by,
the forward direction as shown in Fig. 7 - red arrows, voltage
states of all phase conductors at all nodes were evaluated. VnB = VnA zn · I AB
STEP: 4 Forward step: Estimation of current through the neutral = a, b, c, n (37)
conductor: Next, the complex current flow of each branch for
phase-m {a , b , c} was calculated using,

SmAB PmAB + jQmAB For optimum performance, it is important to follow the calculation
ImAB = = routes defined in Fig. 7 and the steps described above.
VmA VmA (35)
where, is the current flow of phase-m from node-A to node-
ImAB 2.4. Application of proposed LV-LSE algorithm for three-phase three-wire
B, SmAB = PmAB + jQmAB is the complex power flow of phase-m systems
from node-A to node-B and VmA is the phase voltage at node A
for phase m . Then, the neutral current flow from node-A to The proposed analysis on the assumption that, the neutral conductor

9
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 8. Performance of LV-LSE algorithm with Kron’s reduction - Y abc (represented by dotted lines and unfilled markers) and with proposed reduction - Y abc
(represented by solid lines and filled markers) for different measurement uncertainties of smart meters (for the four cases given in Table 2) with transformer loading
(a) AMVME (pu) (b) SDVME (pu) (c) AMCME (pu) and (d) SDCME (pu).

and its associated state variables are present in all distribution lines. yaa yab yac
However, the proposed LV-LSE algorithm can be easily adapted for the Y3abcph,3 wire = yba ybb ybc
state estimation of three-wire systems with a simple modification. yca ycb ycc (38)
Three-wire distribution lines can be represented using 3 × 3 ad­
mittance matrices considering self-impedances of phases a, b, and c as where the symbols have the same meanings as defined previously. Since
the diagonal elements and the mutual coupling among them as off-di­ there is no neutral coupling effect on phase conductors, these ad­
agonal elements. Following this model, the admittance matrix mittance values can be directly utilized in the LV-LSE algorithm without
(Y3abcph,3 wire ) of a three-wire distribution line was written as, any further modification. Also, STEP 5 in the LV-LSE algorithm de­
scribed in Section 2.3 can be discarded as there were no neutral voltage

10
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

avg
Table 3 1
MC
i
Comparison of performance metrics of the proposed LV-LSE algorithm with AMCME = I = max I j I ij pu
MC 1 j NI
(41)
Kron’s reduction and proposed reduction under different voltage, active and i=1

reactive power measurement uncertainties of smart meters (refer Table 2) for

full transformer loading condition.
4. Standard Deviation of Current Magnitude Error - SDCME (pu)
Case No. Error Metric Error of LV-LSE algorithm with Percentage avg
reductionof error NI 2
i
Kron’s Proposed Ij I ij
reduction - Y abc reduction - Y abc 1
MC
j=1
SDCME = I = pu
Case I AMVME (pu) 30.60% MC NI
5.35 × 10 3 3.71 × 10 3 i=1
SDVME (pu) 2.84 × 10 3 2.12 × 10 3 25.45%
AMCME (pu) 2.72 × 10 3 1.22 × 10 3 55.21% 2 (42)
SDCME (pu) 1.30 × 10 3 0.62 × 10 3 52.40%

Case II AMVME (pu) 5.35 × 10 3 3.70 × 10 3 30.78%

SDVME (pu) 2.83 × 10 3 2.12 × 10 3 25.12% where, MC is the number of Monte Carlo simulation performed for each
AMCME (pu) 2.73 × 10 3 1.23 × 10 3 55.13% case (MC = 500), N V is the total number of voltage states in the net­
SDCME (pu) 1.29 × 10 3 0.62 × 10 3 51.92% work (N V = 63 nodes × 4 [voltage states/node] = 252 voltage states),
i
V j is the estimated voltage magnitude (pu) of j -th state variable at the
Case III AMVME (pu) 5.43 × 10 3 3.93 × 10 3 27.71%
SDVME (pu) 17.24% i -th Monte Carlo simulation, V ij is the true voltage magnitude of the j -th
3.01 × 10 3 2.49 × 10 3

AMCME (pu) 36.90% state variable at the i -th Monte Carlo simulation (obtained from the
2.73 × 10 3 1.72 × 10 3
i
SDCME (pu) 1.32 × 10 3 0.88 × 10 3 33.85% load flow analysis), I j is the estimated current magnitude (pu) of j -th
branch current at the i -th Monte Carlo simulation, I ij is the true current
Case IV AMVME (pu) 5.43 × 10 3 3.93 × 10 3 27.63% magnitude (pu) of the j -th branch current at the i -th Monte Carlo si­
SDVME (pu) 3.00 × 10 3 2.48 × 10 3 17.15% mulation (obtained from the load flow analysis), and N I is the total
AMCME (pu) 2.77 × 10 1.72 × 10 37.87%
number of branch currents (including the neutral current flows) in the
3 3

SDCME (pu) 1.33 × 10 3 0.87 × 10 3 34.61%

network (N I = 62 feeder sections × 4 [current states/feeder section] =
248 current states).
states to be estimated in the three-wire system.
3. Results and discussion
2.5. Ground truth generation and performance metrics
The network shown in Fig. 1 was used to implement the proposed
LV-LSE algorithm. The network consists of 92 households and 19
2.5.1. Ground truth generation
rooftop PV plants, with a total installed PV capacity of approximately
A Monte Carlo based simulation scheme for the proposed LV-LSE
60 kW. The capacity of loads and rooftop PV plants for each simulation
was implemented on the Matlab-OpenDSS hybrid environment. Under
were randomly assigned. For the simulations, about 60% of households
the Monte Carlo implementation, load power values, load power factor
and all the rooftop PV plants were assumed to be connected to the
values, capacities of PV plants and the location of smart meters were
LVDG through smart meters. In addition, there is a measuring device at
randomly assigned for each simulation run. Next, a load flow analysis
the secondary side of the MV-LV transformer to measure aggregated
was performed on the LV network shown in Fig. 1, and the corre­
power flows into the LVDG and node voltages. For this network, it was
sponding voltage and current values (magnitudes and angles) were used
assumed that Pmin
load
= 0.2 kW and Pmax
load
= 7 kW .
as the ground truth for the SE algorithm.
The performance of the proposed LV-LSE algorithm was analyzed
under three scenarios: (A) for different measurement uncertainties of
2.5.2. Performance metrics
smart meters (B) for different scales of the original network and (C) for
The following performance metrics were used to analyze the per­
different data loss conditions of smart meter data.
formance of the proposed LV-LSE algorithm.
3.1. Performance under different measurement uncertainties of smart
1. Average Maximum Voltage Magnitude Error - AMVME (pu)
meters
MC
1 i
AMVME = || V ||avg = max Vj V ij pu Smart meters with various precision levels are currently available in
MC 1 j NV
(39)
the market to measure voltage, current, active and reactive power
i=1

flows. High-performance smart meters measure these voltages, cur­

2. Standard Deviation of Voltage Magnitude Error - SDVME (pu) rents, and power levels with great accuracy and do not require ad­
avg vanced SE algorithms to determine the true states of the network.
2
NV
i However, high-accuracy smart meters are expensive and are not eco­
Vj V ij nomical for large-scale deployment. In contrast, low-accuracy smart
MC
1 j=1
meters are cheaper to use in LVDGs, however they add some un­
SDVME = V = pu
MC i=1
NV certainty to the original measurement. This allows more burden on SE
algorithms to accurately estimate the actual states of the network. In
2 (40) this case study, the influence of smart meter measurement uncertainty
on the accuracy of the proposed LV-LSE algorithm was explored. The
measured voltage magnitudes, active and reactive power values (zmeas )
3. Average Maximum Current Magnitude Error - AMCME (pu)
for the LV-LSE algorithm were calculated using,

11
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 9. Performance of LV-LSE algorithm with Kron’s reduction - Y abc (represented by dotted lines and unfilled markers) and with proposed reduction - Y abc
(represented by solid lines and filled markers) for different scales of the original network (L = 0.5 × L 0 , L = 1.0 × L0 , and L = 1.5 × L 0 ) with transformer loading. (a)
AMVME (b) SDVME (c) AMCME (d) SDCME.

z meas = z true + FS· N (0, P , Q, V ) (43) in Table 2 by 1.96 to ensure 95% of confidence level, and N (0, P, Q, V )
is the normally distributed random variable with zero mean and P2, Q, V
where, z true is the true value of the measurement which was obtained variance.
from the load-flow analysis, FS is the full-scale meter reading related to The performance of the proposed LV-LSE algorithm was analyzed
each type of measurement, and P, Q, V is the standard measurement for both Kron’s reduction and the proposed reduction method under
uncertainty related to the each type of measurement, which was de­ different measurement uncertainties of smart meters (as listed in
termined from dividing the maximum measurement error values given Table 2) with transformer loading from 10% to 100% as shown in

12
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Table 4 uncertainty of 0.5% and power uncertainty of 1% (case I in the previous

Comparison of performance metrics of the proposed LV-LSE algorithm with case study).
Kron’s reduction and proposed reduction under different scales of line lengths In general, most LVDGs have a total length of about 2 km. The total
of the network (L = 0.5 × L 0 , L = 1.0 × L0 and L = 1.5 × L 0 ) shown in Fig. 1 for length of the network shown in Fig. 1 is about 1.3 km. Therefore, there
full transformer loading condition
is a possibility of expanding the original network by a factor of 2 km/
Case Error Metric Error of LV-LSE algorithm with Percentage 1.3 km = 1.5 in maximum. Thus, the performance metrics for LV-LSE
reductionof were calculated for two different scales of original line lengths (i.e., 0.5
Kron’s Proposed error
and 1.5) for Kron’s reduction and proposed reduction methods, as
reduction - reduction -
shown in Fig. 9. The results shown in Fig. 9 revealed that the perfor­
Y abc Y abc
mance metrics are sensitive to the variation of line length of the net­
L = 0.5 × L0 AMVME (pu) 3.79 × 10 3 2.55 × 10 3 32.73% work.
SDVME (pu) 1.86 × 10 3 1.29 × 10 3 30.99% From the results shown in Fig. 9, the proposed LV-LSE algorithm
AMCME (pu) 1.97 × 10 3 0.57 × 10 3 77.66% with the proposed reduction method performs better as compared to
SDCME (pu) 0.86 × 10 3 0.27 × 10 3 68.66% the proposed LV-LSE algorithm with Kron’s reduction for all of the
uncertainty cases. Besides, as we observed in the previous case study,
L = 1.0 × L0 AMVME (pu) 5.21 × 10 3 3.60 × 10 3 30.85%
the use of the proposed reduction method in the LV-LSE algorithm
SDVME (pu) 2.57 × 10 3 1.79 × 10 3 30.28%
(represented by continuous lines with filled markers) resulted in the
AMCME (pu) 2.67 × 10 3 0.65 × 10 3 75.56%
SDCME (pu) 73.71%
degradation of all error metrics as compared to the use of Kron’s re­
1.18 × 10 3 0.31 × 10 3
duction (represented by dotted lines with unfilled markers). In order to
L = 1.5 × L0 AMVME (pu) 6.22 × 10 3 4.28 × 10 3 31.17% highlight this observation, a comparison of the error metrics (AMVME,
SDVME (pu) 3.05 × 10 3 2.10 × 10 3 31.16% SDVME, AMCME, and SDCME) under full transformer loading condi­
AMCME (pu) 3.07 × 10 3 0.77 × 10 3 74.80% tions for different scales of the original network is listed in Table 4 for
SDCME (pu) 1.36 × 10 3 0.37 × 10 3 72.98% LV-LSE algorithm with proposed reduction and LV-LSE algorithm with
Kron’s reduction.

Fig. 8. For each of the cases, the performance metrics given in Section 3.3. Performance under different data loss conditions
2.5.2 were calculated by performing Monte Carlo simulations.
From the results shown in Fig. 8, the proposed LV-LSE algorithm As discussed earlier, it was assumed that about 60% of households
with the proposed reduction method outperforms the proposed LV-LSE in the network shown in Fig. 1 are equipped with smart meters.
algorithm with Kron’s reduction for all of the uncertainty cases. For Moreover, pseudo values for unknown measurements were assigned as
example, let us consider the case I where voltage uncertainty of 0.5% give by the Eq. (32) and (33) to obtain a positive measurement re­
and active and reactive power uncertainty of 1.0%. The variation of the dundancy.
performance metrics of the LV-LSE algorithm with Kron’s reduction and Various communication technologies have been developed to
proposed reduction method against the transformer loading is shown transmit smart meter data to EMS (or DSOs). The most common tech­
using dotted red lines with unfilled markers and continuous red lines nique is to use the Wide Area Network (WAN). However, not all of these
with filled markers, respectively. As can be noted from Fig. 8 - (a), (b), communication techniques are 100% reliable and are subject to data
(c) and (d), all the error metrics (AMVME, SDVME, AMCME, and loss. Therefore, this section analyzed and presented the performance of
SDCME) gradually increase with transformer loading; however the LV-LSE under data loss conditions.
error values of the proposed LV-LSE algorithm with the proposed re­ It is essential to mention the underlying meaning behind the com­
duction method is always considerably small as compared to LV-LSE munication/transmission data loss. For example, “5% data loss” implies
algorithm with Kron’s reduction. that 5% of the voltage, active, and reactive power measurements, which
It can also be observed that, the accuracy of the SE algorithm is are previously known to the SE algorithm (from the installed smart
more sensitive to the uncertainties in voltage measurements than to meters), are now unknown due to losses in the communication network.
uncertainties in active and reactive power measurements. This phe­ Therefore, for the vacated measurements, pseudo values were assigned
nomenon can be distinguished by comparing the variation of the error in the same manner as described in Section 2.3.
metrics for the cases I and II with cases III and IV. In cases I (Fig. 8 - red Fig. 10 depicts the variation of error metrics with transformer
lines) and II (Fig. 8 - black lines), when only the uncertainty of active loading for different data loss conditions from 0% to 25% in steps of 5%
and reactive power measurements are different; however, the variation for the LV-LSE algorithm with Kron’s reduction and with the proposed
of the error metrics almost coincide with each other throughout all the reduction. According to the figures, voltage and current magnitude
transformer loading conditions. In contrast, when considering cases II errors of the LV-LSE algorithm gradually increase with the data loss, as
(Fig. 8 - black lines) and III (Fig. 8- blue lines), when only the un­ expected. Also, it is clear that the values of error metrics are con­
certainty of voltage measurements is different (voltage uncertainty of siderably low for the LV-LSE algorithm with the proposed reduction
case III is higher than the voltage uncertainty of case II), leading to a method (illustrated in continuous lines filled markers in Fig. 10) as
significant increase in error metrics in case III compared to case II. compared to the LV-LSE with Kron’s reduction (illustrated in dotted
A comparison of the error metrics (AMVME, SDVME, AMCME, and lines an unfilled markers in Fig. 10). Moreover, the comparison of error
SDCME) under full transformer loading conditions for each of the un­ metrics under the full transformer loading condition for the LV-LSE
certainty cases are listed in Table 3 for LV-LSE algorithm with proposed algorithm with Kron’s reduction and LV-LSE algorithm with the pro­
reduction and LV-LSE algorithm with Kron’s reduction. posed reduction method is given in Table 5.

3.2. Performance under different line lengths 3.4. Computational performance of the LV-LSE algorithm

In this case study, the performance metrics of the proposed LV-LSE The proposed LV-LSE algorithm was written on Matlab® (version:
algorithm were analyzed for different scales of line lengths of the ori­ R2016a) - Open DSS (version: 8.4.1.1) hybrid environment. When this
ginal LVDG shown in Fig. 1, assuming all smart meters have voltage algorithm was executed on a processor with Intel Core i7-7700HQ with

13
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Fig. 10. Performance of LV-LSE algorithm with Kron’s reduction - Y abc (represented by dotted lines and unfilled markers) and with proposed method - Y abc (re­
presented by solid lines and filled markers) under different data loss conditions (0 %, 5 %, 10%, 15%, 20%, and 25%) with transformer loading. (a) AMVME (b)
SDVME (c) AMCME (d) SDCME.

32 GB RAM running at 2.8 GHz, the average value of time to process the WLS SE was recorded as 1.0340 s (on the processor with Intel Core i7-
LV-LSE was 0.0928 s. When implemented on a regular low-cost pro­ 7700HQ) which exceeds the average execution time of the proposed
cessor unit with Intel Core i5-8250U and an 8 GB RAM running at LV-LSE algorithm (0.0928 s). Hence, the proposed LV-LSE algorithm is
2.53 GHz, the average execution time was 0.1022 s. Moreover, the computationally efficient when compared to the iterative WLS SE al­
number of Floating-Point Operations (FLOPS) of the proposed LV-LSE gorithm.
algorithm was calculated to be 16368, using a Matlab® program de­
veloped by Hang Qian [45]. The average execution time of the iterative

14
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Table 5 proposed LV-LSE algorithm has been simulated on a typical unbalanced

Comparison of performance metrics of the proposed LV-LSE algorithm with residential network that consists of 92 single-phase and three-phase
Kron’s reduction and proposed reduction under different data loss conditions of loads, and 19 rooftop PV plants.
the network (0%, 5%, 10%, 15%, 20%, and 25%) shown in Fig. 1 for full The proposed LV-LSE algorithm uses a novel method for simplifying
transformer loading condition
the 4 × 4 admittance matrix of the three-phase four-wire feeder line
Percentage Error Metric Error of LV-LSE algorithm with Percentage into a 3 × 3 admittance matrix, while incorporating the effect of the
DataLoss reduction of neutral states on phase conductor states. The LV-LSE algorithm devel­
Kron’s reduction Proposed error
oped based on this simplified feeder model and the Taylor series ap­
reduction
proximations of active and reactive power functions, was able to pro­
0% AMVME (pu) 3.81 × 10 3 3.61 × 10 3 05.09% vide accurate and fast results even under low availability of smart meter
SDVME (pu) 2.01 × 10 3 1.79 × 10 3 11.24% data on the network.
AMCME (pu) 2.73 × 10 3 0.63 × 10 3 77.08% Above all, the proposed LV-LSE algorithm has been validated on an
SDCME (pu) 1.12 × 10 3 0.29 × 10 3 74.31% OpenDSS-Matlab hybrid simulation environment for three different
practical scenarios, namely for different measurement uncertainties of
5% AMVME (pu) 4.07 × 10 3 3.64 × 10 3 10.49%
smart meters, different scales of feeder lengths and different measure­
SDVME (pu) 2.11 × 10 3 1.81 × 10 3 14.27%
ment data loss conditions. Results of these case studies reveal that the
AMCME (pu) 2.96 × 10 3 0.88 × 10 3 70.25%
proposed state estimation algorithm has been able to estimate voltage
SDCME (pu) 1.23 × 10 3 0.41 × 10 3 66.76%
states with the average maximum voltage error of 4.48 × 10 3 pu and
10% AMVME (pu) 4.55 × 10 3 3.79 × 10 3 16.75% average voltage error of about 2.54 × 10 3 pu, under 1% measurement
SDVME (pu) 2.31 × 10 3 1.90 × 10 3 11.75% uncertainty for voltage measurements and 0.5% measurement un­
AMCME (pu) 3.16 × 10 3 1.17 × 10 3 62.97% certainty for power measurements at 100% transformer loading and 60%
SDCME (pu) 1.34 × 10 3 0.53 × 10 3 60.33% availability of smart meters. As the average value of time to process the
LV-LSE is approximately 0.1 s, a commonly available smart meter with
AMVME (pu) 4.94 × 10 3.98 × 10 19.54%
15%
accuracy type 1% and sampling rate 1 s is adequate to implement the
3 3

SDVME (pu) 2.48 × 10 3 1.97 × 10 3 20.45%

proposed LV-LSE.
AMCME (pu) 3.40 × 10 3 1.47 × 10 3 56.81%
Moreover, it has been observed that the accuracy of the state esti­
SDCME (pu) 1.46 × 10 3 0.66 × 10 3 54.45%
mation algorithm can be improved by increasing the number of mea­
20% AMVME (pu) 5.36 × 10 3 4.10 × 10 3 23.56% suring nodes in the distribution system’s infrastructure. The estimation
SDVME (pu) 2.63 × 10 3 2.02 × 10 3 23.14% of the optimum number of instruments required and optimal placement
AMCME (pu) 3.48 × 10 3 1.61 × 10 3 53.78% of measuring nodes will be explored as future work.
SDCME (pu) 1.51 × 10 3 0.72 × 10 3 52.30%

25% AMVME (pu) 5.68 × 10 3 4.32 × 10 3 23.91% Declaration of Competing Interest

SDVME (pu) 2.80 × 10 3 2.12 × 10 3 24.17%
AMCME (pu) 3.48 × 10 3 1.81 × 10 3 49.96%
The authors declare that they have no known competing financial
SDCME (pu) 1.59 × 10 3 0.81 × 10 3 48.90%
interests or personal relationships that could have appeared to influ­
ence the work reported in this paper.

4. Conclusion
Acknowledgment
This work outlines a complete state estimation algorithm, developed
to estimate and monitor the overall states of the low voltage distribu­ We would like to acknowledge the financial support provided by the
tion network under high PV penetration. The proposed algorithm per­ National Science Foundation (NSF), Sri Lanka (Research Grant No: RG/
forms accurately, even under severely unbalanced conditions. The 2018/EA & ICT/01).

Appendix A. Details of the residential network

Details of the residential network used for the analysis is given in Table A.6.

Table A.6
Details of the LVDG network shown in Fig. 1
Property Description

Distribution transformer specifications 11 kV/415 V, 400 kVA

Distribution transformer vector group Dy11
Transformer earthing type Solid-earthed neutral
No. of distribution feeders 4
Number of customers 94 (single phase)
Number of rooftop PV systems 18 (single phase)
Total installed PV capacity 93.4 kW
Total length of the network 1.3 km
Medium of electricity distribution Overhead cable, type ABC-70

15
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

Appendix B. Technical information of the overhead line in the residential network

The technical information of the overhead cables used in the LVDG network is given in Table B.7.

Table B.7
Technical information of the overhead line (ABC-70) used in LVDG network
shown in Fig. 1
Property Value/ (units)

Nominal area of the phase conductors 70 mm2

Nominal area of the neutral conductor 1 54.6 mm2
Insulation material XLPE
Conductor shape Circular
DC resistance of phase conductors 0.443 /km
DC resistance of neutral conductor 0.630 /km
Insulation thickness of phase conductors 1.8 mm
Insulation thickness of neutral conductor 1.6 mm

Appendix C. Derivation of non-linear active (PmAB ) and reactive power (QmAB ) flow functions

SmAB = |VmA | A
m [(g¯m + jb¯m )(|V A | A
|V B | B
)]
= a, b, c (C.1)

SmAB = |VmA | A
m [g¯m |V | A A
g¯m |V |B B
+ jb¯m |V | A A
jb¯m |V | B B
]
= a, b, c (C.2)

SmAB = |VmA | A
m [g¯m |V A | A
g¯m |V B | B
jb¯m |V A | B
+ jb¯m |V B | B
]
= a, b, c (C.3)
Using the notations given in (21), the above equation was simplified as,

SmAB = |VmA | [g¯m |V A | AA

ml g¯m |V B | AB
ml jb¯m |V A | AA
ml + jb¯m |V B | AB
ml ]
= a, b, c (C.4)
Let us separate the above equation to it’s real and imaginary components as,
AA ¯
SmAB = |VmA | {|V A |[g¯m ·cosm AA
+ bm ·sinm ] |V B |[g¯m ·cosmAB + b¯m ·sinm
AB
]}
= a, b, c
AA ¯
+ j |VmA | {|V A |[g¯m ·sinm AA
bm ·cosm ] |V B |[g¯m ·sinmAB +b¯m ·cosm
AB
]}
= a, b, c (C.5)
The real part of (C.5) gives the active power flow through phase-m {a , b , c, } between busbars A and B (PmAB ) as,
AA ¯
PmAB = |VmA | {|V A
|[g¯m ·cosm + bm AA
·sinm ] |V B
|[g¯m ·cosmAB + b¯m AB
·sinm ]}
= a, b, c (C.6)
The imaginary part of (C.5) gives the reactive power flow through phase-m {a , b , c, } between bus-bars A and B (QmAB ) as,

QmAB = |VmA | {|V A AA

|[g¯m ·sinm b¯m ·cosmAA] |V B AB ¯
|[g¯m ·sinm + bm AB
·cosm ]}
= a, b, c (C.7)

Appendix D. Derivation of linearized active (P mAB ) and reactive power (Qm ) flow functions
AB

Let us consider the non-linear active power function (P mAB ) for phase-m {a, b, c} given by,
AA ¯ AB ¯
PmAB = |VmA | {|V A |[g¯m ·cosm AA
+ bm ·sinm ] |V B|·[g¯m ·cosm + bm ·sinmAB]}
= a, b, c (D.1)
The nonlinear sin(·) and cos(·) functions were substituted with their Taylor series approximations (24) as,

P mAB = |VmA | {|V A |[g¯m ·(cos( AA

ml, r ) sin( AA
ml, r )·
AA
ml )
= a, b, c

+ b¯m ·(sin( AA
ml, r ) + cos( AA
ml, r )·
AA
ml )]

|V B |[g¯m ·(cos( AB
ml, r ) sin( ml AB
, r )·
AB
ml )

+ b¯m ·(sin( AB
ml, r ) + cos( AB
ml, r )·
AB
ml )]} (D.2)
It was assumed that because the distance between two adjacent basbars in LVDGs are very small as compared to MVDGs (see Fig. 1).
AA
m ,r
AB
m ,r
With this assumption, above equation was rewritten as,

16
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

P mAB = |VmA | {|V A |[g¯m ·(cos( AB

ml, r ) sin( AB
ml, r )·
AA
ml )
= a, b, c

+ b¯m ·(sin( AB
ml, r ) + cos( AB
ml, r )·
AA
ml )]
B AB AB AB
|V |[g¯m ·(cos( ml, r ) sin( ml, r )· ml )
+ b¯m ·(sin( AB
ml, r ) + cos( AB
ml, r )· AB
ml )]} (D.3)
Eq. (D.3) was rearranged as,

P mAB = |VmA | {(g¯m cos( AB

ml, r ) + b¯m sin( AB A
ml, r ))·(|V | |V B|)
= a, b, c

+ ( g¯m sin( AB
ml, r ) + b¯m cos( AB A
ml, r ))·(|V |·
AA
ml |V B|· AB
ml )} (D.4)
The local operating points for the voltage magnitudes were assigned (as given in (26)) to (D.4) as,

P mAB = |Vr | {(g¯m cos( AB

ml, r ) + b¯m sin( AB A
ml, r ))·(|V | |V B|)
= a, b, c

+ ( g¯m sin( AB
ml, r ) + b¯m cos( AB
ml, r ))·(|Vr|·
AA
ml |Vr |· AB
ml )} (D.5)
Eq. (D.5) was simplified as,

P mAB = {|Vr |(g¯m cos( AB

ml, r ) + b¯m sin( AB A
ml, r ))·(|V | |V B|)
= a, b, c

+ |Vr |2 ·( g¯m sin( AB

ml, r ) + b¯m cos( AB
ml, r ))·(
AA
ml
AB
ml )} (D.6)
As previously defined in (21), it is known that = l and = AA
m
A
m
A AB
m
A
m l
B
. Therefore, AA
ml
AB
ml = ( A B
) . By substituting this
to (C.5), the linearized active power function for phase-m (P mAB ) was obtained as,

P mAB = {|Vr |(g¯m cos( AB

ml, r ) + b¯m sin( AB A
ml, r ))·(|V | |V B|)
= a, b, c

|Vr |2 ·( g¯m sin( AB

ml, r ) + b¯m cos( AB
ml, r ))·(
A B
)} (D.7)
Similarly, the linearized reactive power function for phase-m (QmAB ) was obtained as,
AB
Qm = {|Vr |(g¯m sin( AB
ml, r ) b¯m cos( AB A
ml, r ))·(|V |
B
|V |)
= a, b, c

|Vr |2 ·(g¯m cos( AB

ml, r ) + b¯m sin( AB
ml, r ))·(
A B
)} (D.8)

Appendix E. Supplementary material

Supplementary data associated with this article can be found, in the online version, at https://doi.org/10.1016/j.ijepes.2020.106332.

References [12] Primadianto A, Lu C. A review on distribution system state estimation. IEEE Trans
Power Syst 2017;32(5):3875–83. https://doi.org/10.1109/TPWRS.2016.2632156.
[13] Geisler KI. Ampere magnitude line measurements for power systems state estima­
[1] Graham PW, Hayward J, Foster J, Story O, Havas L. GenCost 2018 Updated pro­ tion. IEEE Trans Power Apparatus Syst 1984;PAS-103(8):1962–9. https://doi.org/
jections of electricity generation technology costs. CSIRO 2018:1–63. 10.1109/TPAS.1984.318500.
[2] Almeida DW, Abeysinghe AHMSMS, Ekanayake JB. Analysis of rooftop solar im­ [14] Chen T, Chen M, Inoue T, Kotas P, Chebli EA. Three-phase cogenerator and trans­
pacts on distribution networks. Ceylon J Sci 2020;48(2):103–12. https://doi.org/ former models for distribution system analysis. IEEE Trans. Power Delivery
10.4038/cjs.v48i2.7614. 1991;6(4):1671–81. https://doi.org/10.1109/61.97706.
[3] Almeida D, Abeysinghe S, Ekanayake MP, Godaliyadda RI, Ekanayake J, Pasupuleti [15] Sakis Meliopoulos AP, Zhang Fan. Multiphase power flow and state estimation for
J. Generalized approach to assess and characterise the impact of solar pv on lv power distribution systems. IEEE Trans Power Syst 1996;11(2):939–46. https://doi.
networks. Int J Electr Power Energy Syst 2020;121:106058. https://doi.org/10. org/10.1109/59.496178.
1016/j.ijepes.2020.106058. http://www.sciencedirect.com/science/article/pii/ [16] Vempati N, Shoults RR, Chen MS, Schwobel L. Simplified feeder modeling for
S0142061519335173. loadflow calculations. IEEE Trans Power Syst 1987;2(1):168–74. https://doi.org/
[4] Liacco TD. The role and implementation of state estimation in an energy manage­ 10.1109/TPWRS.1987.4335094.
ment system. Int J Electr Power Energy Syst 1990;12(2):75–9. https://doi.org/10. [17] Chen T, Chen M, Hwang K, Kotas P, Chebli EA. Distribution system power flow
1016/0142-0615(90)90002-S. analysis-a rigid approach. IEEE Trans Power Delivery 1991;6(3):1146–52. https://
[5] Primadianto A, Lu C-N. A review on distribution system state estimation. IEEE Trans doi.org/10.1109/61.85860.
Power Syst 2016;32(5):3875–83. [18] Sun DIH, Abe S, Shoults RR, Chen MS, Eichenberger P, Farris D. Calculation of
[6] Alsac O, Vempati N, Stott B, Monticelli A. Generalized state estimation. IEEE Trans energy losses in a distribution system. IEEE Trans Power Apparatus Syst 1980;PAS-
Power Syst 1998;13(3):1069–75. https://doi.org/10.1109/59.709101. 99(4):1347–56. https://doi.org/10.1109/TPAS.1980.319557.
[7] Li Ke. State estimation for power distribution system and measurement impacts. [19] Liu Y, Li J, Wu L. State estimation of three-phase four-conductor distribution sys­
IEEE Trans Power Syst 1996;11(2):911–6. https://doi.org/10.1109/59.496174. tems with real-time data from selective smart meters. IEEE Trans Power Syst
[8] Alsac O, Vempati N, Stott B, Monticelli A. Generalized state estimation. IEEE Trans 2019;34(4):2632–43. https://doi.org/10.1109/TPWRS.2019.2892726.
Power Syst 1998;13(3):1069–75. https://doi.org/10.1109/59.709101. [20] Džafic I, Jabr RA, Huseinagic I, Pal BC. Multi-phase state estimation featuring in­
[9] Bose A, Clements KA. Real-time modeling of power networks. Proc IEEE dustrial-grade distribution network models. IEEE Trans Smart Grid
1987;75(12):1607–22. https://doi.org/10.1109/PROC.1987.13930. 2017;8(2):609–18.
[10] Schweppe FC, Wildes J. Power system static-state estimation, part i: Exact model. [21] Pau M, Ponci F, Monti A, Sulis S, Muscas C, Pegoraro PA. An efficient and accurate
IEEE Trans Power Apparatus Syst 1970;PAS-89(1):120–5. https://doi.org/10.1109/ solution for distribution system state estimation with multiarea architecture. IEEE
TPAS.1970.292678. Trans Instrum Meas 2017;66(5):910–9.
[11] Brandalik R, Waeresch D, Wellssow WH, Tu J. Linear three-phase state estimation [22] de Almeida MC, Ochoa LF. An improved three-phase amb distribution system state
for lv grids using pseudo-measurements based on approximate power distributions. estimator. IEEE Trans Power Syst 2017;32(2):1463–73.
CIRED Open Access Proc J 2017;2017(1):1871–4. https://doi.org/10.1049/oap- [23] Nanchian S, Majumdar A, Pal BC. Three-phase state estimation using hybrid particle
cired.2017.0071. swarm optimization. IEEE Trans Smart Grid 2015;8(3):1035–45.

17
W.G. Chaminda Bandara, et al. Electrical Power and Energy Systems 124 (2021) 106332

[24] Dedé A, Della Giustina D, Rinaldi S, Ferrari P, Flammini A, Vezzoli A. Smart meters analysis of distribution feeders. 2011 IEEE/PES power systems conference and ex­
as part of a sensor network for monitoring the low voltage grid. 2015 IEEE sensors position 2011. p. 1–6. https://doi.org/10.1109/PSCE.2011.5772579.
applications symposium (SAS) 2015. p. 1–6. https://doi.org/10.1109/SAS.2015. [36] Sun DIH, Abe S, Shoults RR, Chen MS, Eichenberger P, Farris D. Calculation of
7133616. energy losses in a distribution system. IEEE Trans Power Apparatus Syst 1980;PAS-
[25] Abdel-Majeed A, Braun M. Low voltage system state estimation using smart meters. 99(4):1347–56. https://doi.org/10.1109/TPAS.1980.319557.
In: 2012 47th international universities power engineering conference (UPEC); [37] Dorfler F, Bullo F. Kron reduction of graphs with applications to electrical networks.
2012. p. 1–6. doi:10.1109/UPEC.2012.6398598. IEEE Trans Circuits Syst I Regul Pap 2013;60(1):150–63. https://doi.org/10.1109/
[26] Leite JB, Mantovani JRS. State estimation of distribution networks through the real- TCSI.2012.2215780.
time measurements of the smart meters. 2013 IEEE grenoble conference 2013. p. [38] Kettner AM, Paolone M. Performance assessment of kron reduction in the numerical
1–6. https://doi.org/10.1109/PTC.2013.6652170. analysis of polyphase power systems. 2019 IEEE Milan PowerTech. 2019. p. 1–6.
[27] Alimardani A, Therrien F, Atanackovic D, Jatskevich J, Vaahedi E. Distribution [39] Lu CN, Teng JH, Liu WE. Distribution system state estimation. IEEE Trans Power
system state estimation based on nonsynchronized smart meters. IEEE Trans Smart Syst 1995;10(1):229–40. https://doi.org/10.1109/59.373946.
Grid 2015;6(6):2919–28. https://doi.org/10.1109/TSG.2015.2429640. [40] Nanchian S, Majumdar A, Pal B. Three-phase state estimation using hybrid particle
[28] Muscas C, Sulis S, Angioni A, Ponci F, Monti A. Impact of different uncertainty swarm optimization. 2016 IEEE power and energy society general meeting
sources on a three-phase state estimator for distribution networks. IEEE Trans (PESGM) 2016. p. 1. https://doi.org/10.1109/PESGM.2016.7741159.
Instrum Meas 2014;63(9):2200–9. https://doi.org/10.1109/TIM.2014.2308352. [41] Rousseaux P, Mallieu D, Cutsem TV, Ribbens-Pavella M. Dynamic state prediction
[29] Dedé A, Giustina DD, Rinaldi S, Ferrari P, Flammini A, Vezzoli A. Smart Meters as and hierarchical filtering for power system state estimation. Automatica
Part of a Sensor Network for Monitoring the Low Voltage Grid 00416. 1988;24(5):595–618. https://doi.org/10.1016/0005-1098(88)90108-2.
[30] Soares TM, Bezerra UH, Tostes MEdL. Full-observable three-phase state estimation [42] Toyoshima D, Castillo MRC, Fantin CA, London JBA. Observability and measure­
algorithm applied to electric distribution grids. Energies 2019;12(7):1327. ment redundancy analysis on three-phase state estimation. 2012 IEEE power and
[31] Alam MJE, Muttaqi KM, Sutanto D. A three-phase power flow approach for in­ energy society general meeting 2012. p. 1–8. https://doi.org/10.1109/PESGM.
tegrated 3-wire mv and 4-wire multigrounded lv networks with rooftop solar pv. 2012.6345507.
IEEE Trans Power Syst 2013;28(2):1728–37. [43] Fantin CA, Massignan JAD, Castillo MR, London JBA. Observability, redundancy
[32] Li H, Mao R, Lai L, Qiu RC. Compressed meter reading for delay-sensitive and secure and gross error processing in state estimation using scada and synchronized phasor
load report in smart grid. 2010 First IEEE international conference on smart grid measurements. 2015 IEEE Eindhoven PowerTech 2015. p. 1–5. https://doi.org/10.
communications. IEEE; 2010. p. 114–9. 1109/PTC.2015.7232434.
[33] Chen Q, Kaleshi D, Armour S, Fan Z. Reconsidering the smart metering data col­ [44] Waeresch D, Brandalik R, Wellssow WH, Jordan J, Bischler R, Schneider N. Linear
lection frequency for distribution state estimation. 2014 IEEE international con­ state estimation in low voltage grids based on smart meter data. 2015 IEEE
ference on smart grid communications (SmartGridComm). IEEE; 2014. p. 517–22. Eindhoven PowerTech 2015. p. 1–6. https://doi.org/10.1109/PTC.2015.7232343.
[34] Carson JR. Wave propagation in overhead wires with ground return. Bell Syst Tech [45] Qian H. Counting the floating point operations (flops), Matlab central file exchange
J 1926;5(4):539–54. https://doi.org/10.1002/j.1538-7305.1926.tb00122.x. 23. https://in.mathworks.com/matlabcentral/fileexchange/50608-counting-the-
[35] Kersting WH, Green RK. The application of carson’s equation to the steady-state floating-point-operations-flops.

18