Electrical Power and Energy Systems 54 (2014) 154–162

Contents lists available at SciVerse ScienceDirect

Electrical Power and Energy Systems

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

A three-phase state estimation in active distribution networks

Aleksandar Ranković a,⇑, Branko M. Maksimović b, Andrija T. Sarić a,c
a
Faculty of Technical Sciences, University of Kragujevac, Čačak, Serbia
b
Technical High School, Čačak, Serbia
c
Faculty of Technical Sciences, University of Novi Sad, Novi Sad, Serbia

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

Article history: This paper presents the improved Weighted Least Square (WLS) based three-phase state estimation algo-
Received 28 May 2012 rithm for active distribution networks. The three-phase component models are developed for character-
Received in revised form 4 July 2013 istic types of transformers (respecting their winding connections), lines (three and four wires), loads
Accepted 5 July 2013
(voltage dependent grounded-Y, ungrounded-Y and ungrounded-D) and distributed generation (DG)
units (synchronous generator, induction generator and three-phase (three- and four-wire) electroni-
cally-coupled based DG units). The unmonitored (or partially monitored) powers of loads and DG units
Keywords:
are initially estimated from normalized daily load profiles, or from external inputs (historical data-
Distribution network
State estimation
base/weather forecast), such as wind, sun and water inflow forecasts, respectively. These measurements
Three-phase algorithm are introduced with lower weights (treated as the pseudo-measurements in the proposed state estima-
Weighted Least Square (WLS) algorithm tion algorithm). The results and practical aspects of the proposed methodology are demonstrated on two
Distributed generation characteristic examples: modified IEEE 13-bus test system and 322-bus real-world distribution network.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction impractical to telemeter all points of the network. However, a prac-

tical distribution networks should have the following features and
State estimation (SE) is mainly used to filter the noise (random types of measurements that are available [6–10]:
error) of redundant measurements and other data and provide a
reliable SE of the transmission or distribution network [1]. The – Substation measurements: Several types of real-time measure-
SE in distribution networks is usually based on a limited number ments are available at the source or distribution substations
of real-time measurements. The distribution networks consist (bus real/reactive power injections/feeder flows, bus voltage
mainly of radial, short and un-transposed feeders with high R/X ra- magnitudes, bus current injection/feeder flow magnitudes,
tios and with single-, two- or three-phase laterals, and loads along and circuit breaker statuses).
the feeders that usually can be single- and two-phase for residen- – Few measurements on critical points of the network: Any afford-
tial customers, or three-phase for both commercial and industrial able measurements need to be placed on the critical points in
customers. Therefore, the distribution networks are unbalanced the network to reduce the standard deviation of the computed
in general. Furthermore, even if the distribution network is bal- state variables or provide the network observability.
anced, the asymmetrical faults additionally introduce an imbal- – Customer load estimates: They are usually available for large
ance. This means that the proposed three-phase SE model is number of unmeasured buses remote from the source substa-
interesting for both normal and emergency conditions. To avoid tion. In general, the observability of the distribution network
the significant error arising from the inherent network imbalance, is achieved using these pseudo-measurements.
the detailed component models are required for SE in distribution – DG unit measurements/estimates: They are usually provided with
networks in both normal and contingency operation conditions. bus voltage and real power injection real-time measurements.
Note that by respecting the above mentioned characteristics of The unavailable real-time measurements at unmonitored DG
the distribution networks, the decoupled Weighted Least Square units can be replaced with pseudo-measurements obtained
(WLS) based algorithms are not suitable for their SE [2–5]. from the external inputs (hystorical database and/or weather
Since the number of consumers and distributed generators forecast).
(DGs) connected to a distribution network may be large, it is
Based on previously published references we can draw a con-
clusion that for unbalanced transmission and distribution power
⇑ Corresponding author. Tel.: +381 32 302 761; fax: +381 32 342 101. networks the three-phase power flow [11–16] is more frequently
E-mail address: [email protected] (A. Ranković). explored problem than the three-phase SE [17–19].

0142-0615/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijepes.2013.07.001
 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162 155

Only few previously proposed three-phase SE algorithms can 2. Component three-phase models
be adopted for unbalanced distribution networks [2–5,15]. A
common feature of the previously proposed distribution SE algo- The matrix equation that is used to describe the three-phase
rithms is that they are based on the modification of the WLS- distribution network in admittance form is
based single-phase SE algorithms originaly developed for trans-
I ¼ Y BUS V; ð1Þ
mission networks, or minimally adopted for radial (meshed) dis-
tribution networks. However, in previously proposed algorithms where I, V is the (3N)-dimensional vectors of bus phase currents
all the problems which pertain to the SE in modern distribution and voltages, respectively (N is total number of buses in distribution
networks are not resolved in general, respecting the following network) and Y BUS is the (3N)  (3N)-dimensional bus admittance
shortcomings: matrix.
Characteristic types of the distribution network elements in the
 Due to an imbalance of the distribution system operating condi- current–voltage model (1), such as transformers, lines, loads and
tions, the complete and accurate three-phase models are desir- DG units, are derived in the following subsections.
able for distribution and inline transformers of various core and
winding configurations. 2.1. Transformers
 Detailed models of (un-)grounded Y- and D-connected nonlin-
ear loads are usually replaced with the simple constant phase By eliminating the neutral point with Kron reduction, the
PQ injection bus model. resulting transformer bus admittance matrix (Y T ) can be reduced
 The DG units are usually modeled by PQ (PV) bus injection to (6  6)-dimensional [11,12,19]. The general three-phase trans-
model, without detailed modeling of the DG units. It is impor- former model is shown on Fig. 1.
tant to note that determining a suitable model for each DG unit By using the transformer bus admittance matrix Y T in Fig. 1, the
in SE of the distribution networks requires the knowledge of the three-phase transformer current–voltage relationship between
DG operation and the type of its connection to the grid (direct or transformer primary (p) and secondary (s) sides can be expressed
indirect connection). as
    
Ip Y pp Y ps þ Y m Vp
Based on the previous considerations, the improved three- ¼ ; ð2Þ
phase SE algorithm is proposed in this paper with the following Is Y sp Y ss þ Y m Vs
features included: where Y pp , Y ps , Y sp , Y ss is the (3  3)-dimensional submatrices de-
fined in Appendix A and Y m is the (3  3)-dimensional magnetizing
 Detailed transformer models including copper and core admittance matrix, calculated as [19]
losses, winding connection and phase-shifting between 2 1 3
primary and secondary windings. An individual phase repre- 2Y m þ Y 0m Y 1m þ Y 0m Y 1m þ Y 0m
16 7
sentation is used, as opposed to a balanced three-phase Ym ¼ AY 012
m A
1
¼ 4 Y 1m þ Y 0m 2Y 1m þ Y 0m Y 1m þ Y 0m 5; ð3Þ
3
model. Y 1m þ Y 0m Y 1m þ Y 0m 2Y 1m þ Y 0m
 Line models for both coupled and uncoupled feeders.
 Loads can be a combination of constant current, constant where A is the Fortescue transformation matrix;
power, and constant admittance models. They may be either Y 012 0 1 2 1 2
m ¼ diagfY m ; Y m ; Y m g; Y m ¼ Y m the positive (equal with negative)

(un-)grounded Y- or D-connected. sequence transformer magnetizing admittance; and Y 0m is the zero

 The significance of the DG model presented here stems sequence transformer magnetizing admittance (different from zero
from the inclusion of the inherent phase imbalance caused only for grounded-Y on secondary transformer side, where this
by the distribution network operating imbalances. The admittance formally is assigned - see Fig. 1).
DGs can be either directly connected synchronous or induc-
tion generator, as well as three-phase electronically-coupled 2.2. Lines
DG units, and can be on either primary or secondary
systems. In this paper, the lines are modeled as the multi-grounded
 Different types of monitored, partially monitored or unmoni- three-phase four-wire line sections. After applying Kron reduction
tored DG units and loads are included in the SE algorithm. technique [19], the general three-phase line model is shown in
The initial internal powers for unmonitored (partially moni- Fig. 2. From this figure, the line three-phase current–voltage rela-
tored) DG units are calculated on the basis of external inputs, tionship can be expressed as
such as wind, sun and water inflow forecasts (depending on
p s
the type of DG unit), or from the historical database and from Ip Is
the normalized daily load profiles for loads. Y pp Y ps
YT =
Vp Y sp Y ss Vs
Ym
The proposed methodology was verified on two characteristic
test power examples: (1) modified IEEE 13-bus test system and 0 0
(2) 322-bus real-world distribution network.
This paper is organized as follows. Section 2 presents the three- Fig. 1. General three-phase transformer model.
phase models of transformers, lines, loads and DG units. Section 3
explains the interaction among state variables, bus (or full net-
work) observability and adopted pseudo-measurements in order 1 2
to provide the full network observability. Section 4 describes in I1 −1 I2
Y =Z
more details the WLS-based three-phase SE algorithm in active dis- V1 Y sh Y sh
V2
tribution networks. The final part of the paper presents simulation 2 2
results (in Section 5) and main derived conclusions (in Section 6). 0 0
The details about used three-phase transformer models are pro-
vided Appendix A. Fig. 2. General three-phase line model.
156 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162

    
I1 Y ‘ þ Y ‘sh =2 Y ‘ V1 Table 1
¼ ; ð4Þ Types of the DG units and a way for their connection to the grid.
I2 Y ‘ Y ‘ þ Y ‘sh =2 V2
DG unit Way of connection Section
where
Small- and medium- Synchronous generator (SG) with 2.4.2
2 3 2 31 scale hydro turbines salient-poles (directly connected to the
Y aa Y ab Y ac Z aa Z ab Z ac
6 7 6 7 grid)
Y ‘ ¼ 4 Y ab Y bb Y bc 5 ¼ Z 1
‘ ¼ 4 Z ab Z bb Z bc 5 ; ð5Þ Micro-scale hydro Induction generator (IG) (directly 2.4.3
Y ac Y bc Y cc Z ac Z bc Z cc turbines connected to the grid)
Wind turbines with Doubly fed IG (SG) (connected to the 2.4.4
Y ‘ (Z ‘ ), Y ‘sh is the (3  3)-dimensional three-phase series admit- variable speed grid by voltage-sourced converter
(VSC))
tance (impedance) and shunt admittance matrices, respectively;
Wind turbines with Squirrel cage IG (directly connected to 2.4.3
for the calculation of particular elements, see [19-Sections 4, 5]. fixed speed the grid)
Photovoltaic systems VSC 2.4.4
2.3. Loads Internal combustion SG with round-rotor, or IG (directly 2.4.1,
engines connected to the grid) or 2.4.3
Fuel cells VSC 2.4.4
Loads can be represented as a combination of constant power, Gas turbines SG with round-rotor (directly connected 2.4.1
constant current, and constant admittance models. Those same to the grid)
loads may be either Y- or D-connected. It is assumed that for load Micro-turbines High-speed permanent magnet SG 2.4.4
T (connected to the grid by VSC)
(L) V L is the three-phase voltage (V L ¼ ½ V La V Lb V Lc  ), I L is the
three-phase load current (I L ¼ ½ ILa ILb ILc T ) and Y L is (3  3)-
dimensional admittance matrix. Therefore, there are the following based PV bus model, which is represented by the specified injected
specific load models [20]: real power and bus voltage where the DG is connected to the grid.
The Thevenin equivalent sequence circuits can be converted to
 Grounded-Y: All grounded-Y connected loads can be modeled Norton equivalent sequence circuits, as shown in Fig. 3, where the
as a combination of the following models: current source is used only in positive sequence circuit (I1I ¼ II ) in-
stead of the internal source voltage (EI ), while zero, positive and
ð1Þ Constant power : S L ¼ ½ SLa SLb SLc T ¼ I L  V L ; ð6Þ negative sequence admittances are respectively:

Y 0g ¼ 1=ðZ 0g þ 3Z n Þ; Y 1g ¼ 1=Z 1g ; Y 2g ¼ 1=Z 2g ; ð11Þ

ð2Þ Constant current : I L ¼ ½ ILa ILb ILc T ; ð7Þ
where Z n is impedance between neutral and ground (SGs are usu-
ð3Þ Constant admittance : I L ¼ Y L V L ; ð8Þ ally directly grounded-Y, or Z n ¼ 0) and Z 0g (Z 2g ) is zero (negative) se-
quence impedance.
where  in (6) denotes element-wise operation. The sequence models of the SG from Fig. 3 can be represented as
The model of real/reactive loads for phases a, b and c is a com-
bination of constant power (S), constant current (I) and constant I 012
g ¼ ½ 0 II 0 T  Y 012 012
g Vg ; ð12Þ
admittance (Y) models, respectively:
where
 2
V Lx V Lx T
SLx ¼ kPSx PLxn þ kPIx PLxn þ kPYx PLxn I 012
g ¼ ½ I0g I1g I2g  ;
V Ln V Ln
"  2 # Y 012 ¼ diagfY 0g ; Y 1g ; Y 2g g;
g
V Lx V Lx  T
þ j kQSx Q Lxn þ kQIx Q Lxn þ kQYx Q Lxn ; x ¼ a; b; c; ð9Þ V 012 ¼ V 0g V 1g V 2g :
V Ln V Ln g

where P Lxn (Q Lxn ) denote the real (reactive) load on the rated volt- The three-phase SG model can be obtained by transforming the
age, while the participation factors for real and reactive powers sequence model (12) to the phase coordinates using the transfor-
respectively satisfy the following conditions: mation matrix (A)

kPSx þ kPIx þ kPYx ¼ 1:0; kQSx þ kQIx þ kQYx ¼ 1:0: ð10Þ AI 012
g ¼ A½ 0 II 0 T  AY 012 1 012
g A AV g ; ð13Þ

Both ungrounded-Y and ungrounded-D connected loads can or

be modeled in a similar fashion as above described grounded-Y T
I g ¼ ½ 1 a2 a  II  Y g V g ; ð14Þ
configuration [20].
where
2.4. Distributed generation (DG) units
I g ¼ ½ Iga Igb Igc T ¼ AI 012
g ;

Different types of the DG units and a way for their connection to

the grid are summarized in Table 1 (including sections in which the Y g ¼ AY 012 1
g A ;
models are described in more details) [21,22].
Based on the presented overview of the DG unit types (Table 1),
the following models are adopted.
1 1 1
2.4.1. Synchronous generator (SG) with round-rotor [11,23] I 0g I 1g I 2g
The SGs that are driven by steam turbines (turbo-generators)
Y 0g V 0g II Y 1g V 1g Y 2g V 2g
have cylindrical (round) rotors with slots in which distributed field
windings are placed. The SGs can be modeled by Thevenin equiva- 0 0 0
lent circuit with an internal source voltage (EI ) behind the genera-
Zero sequence Positive sequence Negative sequence
tor’s impedance in d-axis only in a positive sequence circuit
(Z 1g ¼ Z d ). This model is different from the traditional power flow Fig. 3. The Norton equivalent sequence circuits of the SG with round-rotor.
 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162 157

Table 2
State variables and necessary measurements for DG bus observability.

Unit State variables Bus measurements Internal power measurements

# Type # Type Description (for SL = 0) # Description and type

Synchronous generator (SG) 8 Vg, II ¼ I1I 6 rt Sg 2 PT (rt), Q T (rt)

p Sg, Eq. (15) PT (p), Q T (p) (Sections 3.1 and 3.2)
v S = 0, Eq. (17)
Induction generator (IG) 7 Vg, s 6 rt Sg 1 P T (rt)
p S g ; Eq. (20) P T (p) (Sections 3.1 and 3.2)
v S ¼ 0; Eq. (17)
Three-wire VSC-coupled DG unit 10 V g , II ¼ I1I ; I2I 6 rt Sg 4 P T ¼ P 1T (rt), Q 1T ¼ 0 (v) P 2T ¼ Q 2T ¼ 0 (v)
p S g ; From Eqs. (22)–(25) P T ¼ P 1T (p), Q 1T ¼ 0 (v) (Sections 3.1 and 3.2) P 2T ¼ Q 2T ¼ 0(v)
v S ¼ 0; Eq. (17)
Four-wire VSC-coupled DG unit 8 V g , II ¼ I1I 6 rt Sg 2 P T (rt), Q T ¼ 0 (v)
p S g ; From Eq. (25) P T (p), Q T ¼ 0 (v) (Sections 3.1 and 3.2)
v S ¼ 0; Eq. (17)

T
V g ¼ ½ V ga V gb V gc  ¼ AV 012
g :
Y 012
g ¼ diagf0; Y 1g ; Y 2g g;

By multiplying the conjugated phase current from vector (14) T

on both sides by the phase voltages from vector V g , the three- V 012
g ¼ ½ 0 V 1g V 2g  ;
phase apparent powers injected to the grid can be derived
1 1
Y 1g ¼ ; Y 2g ¼ ;
S g ¼ ½ Sga Sgb Sgc T ¼ ½ ST ST ST T =3  ðY g V g Þ  V g ð15Þ Z s þ ðZ r þ R1r ÞkZ m Z s þ ðZ r þ R2r ÞkZ m

and complex equation for internal SG apparent power is Z s ; Z r ¼ ðRr þ jX r Þ, Z m is the stator, rotor and magnetizing imped-
ances, respectively.
ST ¼ PT þ jQ T ¼ ð½ 1 a a 2 II Þ  Vg: ð16Þ The three-phase IG model can be derived by transforming the
Finally, the SG is modeled with the three-phase power balance sequence model (18) to the phase coordinates using the transfor-
equation at the bus where SG is connected to the grid mation matrix (A)

S ¼ S g  S L ¼ ðY BUS;g V g Þ  V g ; ð17Þ AI 012

g ¼ AY 012 1 012
g A AV g ; or I g ¼ Y g V g : ð19Þ

where S is vector of injected three-phase apparent powers. where I g , V g and Y g are defined similarly as in (14).
In the case where there is no load connected to the generator’s Similarly as in (15), by multiplying the conjugated phase cur-
bus (S L ¼ 0 in (17)), the three-phase equality constraints (17) are rent from vector (19) on both sides by the phase voltages from vec-
treated as the virtual measurements in the SE algorithm (see tor V g , the three-phase apparent powers injected to the grid can be
Table 2). derived
Based on (16) and (17) three additional complex (or six real)
S g ¼ ½ Sga Sgb Sgc T ¼ I g  V g ¼ ðY g V g Þ  V g : ð20Þ
phase voltage state variables (complex vector V g ) for SG are intro-
duced, at the bus where SG is connected to the grid and two addi- Finally, similarly as in (17), the IG is modeled with three-phase
tional internal state variables (complex current of Norton power balance equation in bus where IG is connected to the grid.
equivalent, II in Fig. 3). This means that for observability of the Internal real power of the IG is
bus where the unmonitored SG is connected two additional (usu- 2 2
ally pseudo) measurements of real and reactive powers are needed PT ¼ P 1T þ P2T ¼ 3R1r ðI1r Þ  3R2r ðI2r Þ ; ð21Þ
(see Section 3 and Table 2).
where

2.4.2. SG with salient-poles Zm

I1ð2Þ
r ¼ V g1ð2Þ ;
The SGs that are driven by water wheels in hydraulic turbines ðR1ð2Þ
r þ Z r ÞðZ m þ Z s Þ þ Z s Z m
(hydro-generators) have laminated salient-pole rotors with con-
centric field windings and, usually, a large number of poles. This 1s ð1  sÞ
R1r ¼ Rr ; R2r ¼ Rr ;
type of electric machines has different electric and magnetic prop- s 2s
erties on two axes 90° apart: (1) the direct (d) axis (axis of symme-
try of a field pole) and (2) the quadrature (q) axis (axis of symmetry 1 1
V 1g ¼ ðV a þ aV b þ a2 V c Þ; V 2g ¼ ðV a þ a2 V b þ aV c Þ:
midway between two field poles). For effects of saliency to the 3 3
parameters in the three-phase model from Fig. 3, see Ref. [23-
Sec. 6].

2.4.3. Induction generator (IG) [11,15,23]

The sequence model of the IG from sequence equivalent circuits 1
I 1r,2 Z r = Rr + jX r I 1s,2 = I 1g,2 Z s
in Fig. 4 is
I 1m,2
I 012 ¼ Y 012 012 V 1g,2 P1,2
g g Vg ; ð18Þ PT1,2 Rr1,2 Zm g

where QT = 0
1,2
Q1g,2
0
T
I 012
g ¼ ½ 0 I1g I2g  ;
Fig. 4. The IG positive and negative sequence equivalent circuits [23-Sec. 6].
158 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162

Based on (20) and (21) three complex (or six real) phase voltage I1I ¼ II ; I0;2
I ¼ 0; Y 0g ¼ 1=ðZ f þ 3Z n Þ; Y 2g ¼ 1=Z f ; ð25Þ
state variables (complex vector V g ) for IG are introduced, at the bus
where IG is connected to the grid and one additional internal state where Z n is impedance between neutral and ground.
variable (slip of the IG). This means that for observability of the bus Similarly as for SG (12)–(17), the three-phase voltage-current
where the unmonitored IG is connected only one additional (usu- model for three-phase electronically-coupled DG unit, can be de-
ally pseudo) measurement of real power (21) is requested (see Sec- rived from the sequence models in Fig. 5 and (22)–(25).
tion 3 and Table 2).
3. State variables, bus observability and additional (pseudo)
2.4.4. Three-phase electronically-coupled DG unit measurements
Zero, positive and negative sequence Norton equivalent circuits
of the three-phase electronically-coupled DG unit are interfaced Different DG models (described in Section 2) introduce a new
via a three-phase, three-wire (four-wire) voltage-sourced con- set of state variables and additional model based equations, as ex-
verter (VSC), are shown in Fig. 5. It is assumed that the VSC tightly plained in Table 2. Sometimes, the DG units are unmonitored (or
regulates the DC-link voltage, and thus the primary source does partially monitored), and therefore the bus observability is not sat-
not affect the steady-state model of the unit with respect to its isfied. To provide the full DG bus observability, the additional set of
point of connection to the grid. real-time (rt), pseudo (p), or virtual (v) measurements in network
buses with DGs installed, as well as and real-time and pseudo
2.4.4.1. Three-wire VSC-coupled DG unit [15 and Refs. therein]. If a internal power measurements are requested. The internal active
three-wire interface-VSC is controlled only based on the positive power is determined by the type of the DG unit (Section 3.1), while
sequence d–q current control method, the DG unit also exchanges the internal reactive power is determined by the applied DG con-
negative sequence current with the network when the network is nection to the grid (Section 3.2).
not balanced (i.e., when it includes negative sequence compo-
nents). In this case, the positive, zero and negative sequence com- 3.1. Internal real power of the DG units [24 and Refs. therein]
ponents of the model are specified as
For micro, small and medium hydro turbines without a reser-
I1I ¼ II ; I0;2
I ¼ 0; Y 0g ¼ 0; Y 2g ¼ 1=Z f ; ð22Þ voir the internal real power is extracted from the water flow as
where Z f is the equivalent series impedance of the VSC output filter Ph ¼ PT ¼ gh qh gqh hh ðkWÞ; ð26Þ
between the point of common coupling and the short-circuited VSC
terminals. where gh is the efficiency of the DG unit; qh is the density of water
If a three-wire interface-VSC is equipped with both positive and (1000 kg/m3); g is the standard (Earth) gravity (9.81 m/s2); qh is
negative sequence current control schemes, and the negative se- the flow of watercourse (m3/s); and hh is the water head (m).
quence current control is assigned to prevent negative sequence For wind turbine the internal real power is extracted from the
current exchange with the network, then the positive, zero and airflow as
negative sequence models are defined as PWT ¼ PT ¼ 0:5qa AWT cp ðkÞv 3 ðkWÞ; ð27Þ
I1I ¼ II ; I0;2
I ¼ 0; Y 0;2
g ¼ 0: ð23Þ 3
where qa is the density of air (1.225 kg/m ); AWT is the rotor area
For some applications, the negative sequence controller of the of wind turbine (m2); cp ðkÞ is the power coefficient, defined by a set
interface-VSC is used to inject a pre-specified negative sequence of curves [25], and approximated by the following polynomial
current to the grid, for example, for islanding detection. In this X
n

case, the zero, positive and negative sequence models of the inter- cp ðkÞ ¼ ak kk ; ð28Þ
k¼0
face-VSC are
with coefficients ak obtained by a criterion of the minimum least-
I0I ¼ 0; I1I ¼ II ; I2I ¼ INSCI =/NSCI ; Y 0;2
g ¼ 0; ð24Þ
square deviations; k is the tip speed ratio of wind turbine, defined as
where INSCI and /NSCI are the magnitude and phase angle of the pre- k ¼ xR R=v ; ð29Þ
specified negative sequence current, respectively.
For a three-wire interface-VSC, the parameters of the zero se- xR is the angular speed of wind turbine rotor (rad/s); R is the radius
quence model are always zero, since there is no path for zero se- of rotor blades at wind turbine (m); m is the wind speed at the tur-
quence current flow. bine blades (m/s).
For photovoltaic based DG unit the internal real power can be
2.4.4.2. Four-wire VSC-coupled DG unit [15 and Refs. therein]. The calculated as
split-capacitor and the four-leg interface-VSC configurations do
PPV ¼ PT ¼ gPV RcoscAPV ðkWÞ; ð30Þ
enable neutral connection and establish a four-wire VSC system.
If a four-wire VSC only injects a controlled positive sequence cur- where gPV is the efficiency of photovoltaic panel; R is the solar radi-
rent in the grid and permits the system to determine the ex- ation (W/m2); c is the angle of incidence of photovoltaic panel; and
changed positive, zero and negative sequence current APV is the area of photovoltaic panel (m2).
components, then the corresponding models are For internal combustion engines, fuel cells, gas turbines and mi-
cro turbines (Table 1) the internal real power (PT) and the engine
1 1 1 speed are controlled by power electronics interface that optimizes
I 0g I 1g I 2g the fuel usage. In our simulations for these units, it is assumed that
I 0I Y 0g V 0g I 1I Y 1g V 1g I 2I Y 2g V 2g the internal real power is equal to the rated power.
0 0 0
3.2. Internal reactive power of the DG units
Zero sequence Positive sequence Negative sequence

Fig. 5. The Norton equivalent circuits for three-phase electronically-coupled DG For different connections of the DG unit to the grid, the internal
unit. reactive power (QT) can be calculated as:
 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162 159

SG based DG unit: In this case, Q T can be adopted from P-Q diagram, Step 1a:- Set iteration count to (k) = 0.
respecting the internal real power (PT) and the SG unit limits [26- – Calculate the initial state vector (xðkÞ ) by symmetri-
Part I, Sec. 3]. cal power flow.
IG based DG unit: In this case, the reactive power is taken from the – Define the subvector of real-time measurements
grid (Qg) to cover the reactive losses in equivalent circuit from (zrt ).
Fig. 4, or Q T ¼ 0. Usually this configuration uses a capacitor bank – Identification of buses with zero injections (virtual
for the reactive power compensation, where the compensation measurements), that determines the set of equality
level can be calculated from SE results. constraints (cðxÞ ¼ 0).
Electronically-coupled DG based unit: Similarly as for IG based DG Step 1b: Load allocation (for real power with assumed constant
units, in this case it is assumed that Q T ¼ 0 [15]. power factor) for load buses from normalized daily
load profiles (LP), respecting load, day and season type
4. The WLS-based three-phase SE algorithm [24,28], and update these elements in vector of
pseudo-measurements (zp ).
The WLS-based algorithm for three-phase SE in active distribu- – Real/Reactive power allocation for buses with DG
tion networks is applied to minimize the objective function (real- units connected from external inputs (Section 3),
time and pseudo-measurements based model) [27] and update these elements in vector of pseudo-mea-
n o surements (zp ).
T
min JðxÞ ¼ ½z  hðxÞ R1 ½z  hðxÞ ; ð31Þ Step 2: Calculate the increments of all measurements by
x

subject to equality constraints (virtual measurements)

DzðkÞ ¼ zðkÞ  hðxðkÞ Þ; ð33Þ
cðxÞ ¼ 0; ð32Þ
Step 3: Calculate Jacobian [HðxÞ ¼ @hðxÞ=@x; CðxÞ ¼ @cðxÞ=@x],
where x is the state vector composed from the components of com- and information [GðxÞ ¼ HðxÞT R1 HðxÞ] matrices in point
plex voltage (phasor magnitudes and angles) and additional state x ¼ xðkÞ , as well as the increment of state vector (DxðkÞ ) from
variables depending on the type of DG unit (see Section 3 and Ta-
" #" # " #
ble 2); z is the measurement vector; and hðxÞ is the measure- T
DxðkÞ T
GðxðkÞ Þ CðxðkÞ Þ HðxðkÞ Þ R1 DzðkÞ
ments-state variables model (note that in hðxÞ are included in the ¼ ; ð34Þ
CðxðkÞ Þ 0 kðkÞ cðxðkÞ Þ
additional equations and state variables for different applied DG
unit models, as described in Section 2). Step 4: Check the convergence criterion

R ¼ diag r21 ; r22 ; . . . r2M , where r is standard deviation and M is maxjDxxðkÞ j 6 e; k ¼ 1; 2; . . . ; K; ð35Þ
a total number of real-time and pseudo-measurements.
The WLS-based three-phase SE algorithm in active distribution where K is a total number of state variables and e is a convergence
network can be summarized in the following steps (Fig. 6): criterion.

Fig. 6. Flow-chart of the proposed algorithm for three-phase SE algorithm in active distribution networks.
160 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162

^ ¼ xðkÞ , and Go to STOP, Else

If criterion (34) is satisfied, then x (0)
Continue.
Source substation
Step 5: Update the state vector 650 (1)

xðkþ1Þ ¼ xðkÞ þ DxðkÞ ; ð36Þ

646 (6) 645 (3) 632 (2) 633 (5) 634 (10)
Step 6: ðk þ 1Þ ! ðkÞ and Go to Step 2.
In general, the three-phase SE algorithm for unbalanced active
distribution networks has about three times more state variables,
compared with the single-phase SE algorithm (depending on net- PV
work topology and participation of the single and two phase later-
als), but it does not have significant influence to the convergence 611 (11) 684 (7) 671 (4) 692 (9) 675 (13)
(number of iterations) of the proposed algorithm (see Section 5).
SG

5. Application
652 (12) 680 (8)
The proposed three-phase SE algorithm in unbalanced active
distribution networks is tested on two characteristic examples. IG
The first example is the IEEE 13-bus test system, for which all Bus/branch real-time measurements
the simulation details with an analysis of observed physical laws Bus virtual measurements (zero injections)
are provided. For the 322-bus 110/35/10/0.4 kV real-world distri- Bus load pseudo measurements
bution network, only performance results of the proposed SE algo- (obtained from normalized daily load profiles)
rithm are provided, with paying attention to the applicability to Bus generation pseudo measurements
large-scale distribution network. (obtained from weather forecast data)
Explained algorithm does not perform corrections of assumed
Fig. 7. Three-phase topology of the modified IEEE 13-bus test system.
Gaussian distribution of pseudo-measurements (mean and vari-
ance), respecting the calculated state variables (operating condi-
tion) of the distribution networks (load/generation reallocation
procedure). Sometimes the reallocation procedure can filter the
inconsistencies between real-time and pseudo-measurements, Table 3
Bus load/generation and power flow real-time (rt) and pseudo (p) measurements.
and consequently improve the estimations results [24]. This proce-
dure, originally proposed for single-phase SE can be included in the Bus/ Measurement Phase a Phase b Phase c
proposed three-phase SE without theoretical and/or practical branch type
(kW) (kV (kW) (kV (kW) (kV
constraints. Ar) Ar) Ar)
2 rt 8.5 19.0 5.0 58.5 33.0 34.0
5.1. Modified IEEE 13-bus test system [29] 3 rt – – 170.0 125.0 – –
4 rt 393.5 239.0 390.0 278.5 418.0 254.0
5 p (SG) 10.0 0.0 10.0 0.0 10.0 0.0
This is an unbalanced 4.16 kV radial distribution network with 6 p (LP) – – 115.0 66.0 115.0 66.0
minimally modified input bus data, where three DGs were added 8 p (IG) 16.7 3.3 16.7 3.3 16.7 3.3
in buses 5, 8 and 13, as shown in the three-phase diagram in 9 p (LP) 86.7 44.0 86.7 44.0 86.7 44.0
Fig. 7. The modified bus load/generation and power flow real- 10 rt 160.0 110.0 120.0 90.0 120.0 90.0
11 p (LP)  – – – 170.0 80.0
time/pseudo-measurements are presented in Table 3. 12 p (LP) 128.0 86.0 – – – –
Standard deviation (r) of real-time/pseudo measurement is cal- 13 p (PV) 30.0 10.0 30.0 10.0 30.0 10.0
culated from measurement mean value and accuracy (Ac) as 0–1 rt 810.0 490.0 910.0 660.0 970.0 510.0
[22,24].

r ¼ Mean  ðAc=300Þ; ð37Þ

where assumed accuracies in simulation tests are: Acrt = 3% and

Acp = 20%.
For load condition given in Table 3 and feeder denoted by dot-
ted line in Fig. 7, the initial and the estimated three-phase real
powers in load/generation buses (with real-time or pseudo-mea-
surements) are shown in Fig. 8. Estimated bus voltage phasors
and phase voltage deviations (from the rated phase voltage) are
shown in Fig. 9. The convergence criterion in SE algorithm
(e ¼ 1  104 ) was reached after three iterations.
In this case, the single-phase lines/loads significantly increases
the average voltage imbalance factors, where average values are
V2/V1 = 26.95% and V0/V1 = 27.15% (see Section 5.2).

5.2. 322-Bus real-world distribution network [24]

This is an unbalanced 110/35/10/0.4 kV real-world distribution Fig. 8. Initial and estimated three-phase real power in load/generation buses with
network. real-time and pseudo-measurements.
 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162 161

(a) Voltage phasors.

(b) Phase voltage deviations (from the rated phase voltage (Vn = 1 p.u.), calculated by
Fig. 9. Estimated voltage phasors in buses.

of three-phase transformers, lines, loads and distribution genera-

Table 4 tion (DG) units.
Maximum and average voltage imbalance factors for different random variations of
By conducting the extensive simulation studies, with different
loads.
number, type and locations of real-time, virtual and pseudo-mea-
Maximum and Random load variation (%) surements, we can conclude that the algorithm is robust and effi-
average imbalance
Base case ±10 ±20 ±30 cient for application in Distribution Management System.
factors (%)
2 1
Performed and presented numerical tests show that asymmetry
Max {V /V } 3:06  107 2:43  102 5:47  102 6:16  102 of the distribution network conditions can be significant in the
Max {V0/V1} 9:48  107 2:66  102 4:56  102 7:48  102 real-world operation, disregarding the application of single-phase
Average {V2/V1} 2:29  107 5:31  103 1:57  102 1:37  102 state estimation algorithms. In addition, the detailed modeling of
Average {V0/V1} 1:25  107 7:18  103 1:09  102 1:55  102
different DG types (including their connection to the grid) is very
important for the estimation of losses and asymmetry between
The total number of available real-time measurement places the internal and output variables (voltages and real/reactive
(with paired real/reactive power injections or power flow mea- powers).
surements) is 17, distributed on:
Acknowledgement
 2 Substations 110/36.75 kV and 14 substations 35/10.5 kV.
 1 Load bus (industry complex). The authors thank to the Ministry of Education and Science of
the Republic of Serbia for its support of this work provided within
Initial system’s total real load is 8.80 p.u., where all p.u. values Projects III-42004 and III-42009.
are referenced to the base power of 3 MVA.
The available unmonitored DGs are: Appendix A. Three-phase transformer model [11,12]

 Five wind generators with the same technical characteris- Table A1 shows the transformer submatrices from Fig. 1 for the
tics and initial total internal real power output 0.17 p.u. (ini- common step-down three-phase transformer connections (to sim-
tial estimation based on the weather forecast). plify the presentation, the neutral transformer tap changer position
 One photovoltaic unit with assumed the internal real power is assumed) where particular submatrices in Table A1 are defined as
output 0.01 p.u. (initial estimation based on the weather 2 32 3 2 3
forecast). 1 0 0 Y sc 2 1 1
6 76 7 16 7
 One gas turbine with assumed the internal (rated) power Y I ¼ 4 0 1 0 54 Y sc 5; Y II ¼ 4 1 2 1 5Y sc ;
3
0.2 p.u. 0 0 1 Y scþn 1 1 2
 The convergence criterion (e ¼ 1  104 ) is reached after 2 3 ðA1Þ
1 1 0
seven iterations. 1 6 7
Y III ¼ pffiffiffi 4 0 1 1 5Y sc ;
3
Influence on different random variations in three-phase unbal- 1 0 1
anced loads (from balanced base case) on the voltage imbalance
factors is presented in Table 4. Table A1
Transformer submatrices (YT) for common three phase step-down transformer
connections [11,12].

Connection Y pp Y ps Y sp Y ss
6. Conclusion
YNyn YI Y I Y I YI
YNy, Yyn, Yy, Dd Y II Y II Y II Y II
In this paper, the Weighted Least Square (WLS) based state
YNd YI Y TIII Y III Y II
estimation algorithm for unbalanced active distribution networks
Yd, Dy Y II Y TIII Y III Y II
was proposed and verified on two characteristic test examples. Dyn Y II Y TIII Y III YI
The proposed model accommodates the different configurations
162 A. Ranković et al. / Electrical Power and Energy Systems 54 (2014) 154–162

where [11] Chen TH, Chen MS, Inoue T, Kotas P, Chebli EA. Three-phase cogenerator and
transformer models for distribution system analysis. IEEE Trans Power Deliv
Y sc ¼ 1=Z sc ; Y scþn ¼ 1=ðZ sc þ 3Z n Þ; 1991;6:1671–81.
[12] Xiao P, Yu DC, Yan W. A unified three-phase transformer model for distribution
Z sc is the short circuit admittance and Z n is the impedance be- load flow calculations. IEEE Trans Power Syst 2006;21:153–9.
tween neutral and ground (for directly grounded transformer is [13] Singh S, Ghose T. Improved radial load flow method. Int J Electr Power Energy
Syst 2013;44:721–7.
Z n ¼ 0). [14] Strezoski VC, Trpezanovski Lj. Three-phase asymmetrical load-flow. Int J Electr
Matrix Y I is regular, while Y II and Y III are singular matrices. Fur- Power Energy Syst 2000;22:511–20.
thermore, based on Table A1, depending on connection type, the [15] Kamh MZ, Iravani R. A unified three-phase power-flow analysis model for
electronically coupled distributed energy resources. IEEE Trans Power Deliv
matrices Y pp , Y ps , Y sp and Y ss can be regular or singular, where sin- 2011;26:899–909.
gular matrices are not a constraint for the application of proposed [16] Araujo LR, Penido DRR, Vieira FA. A multiphase optimal power flow algorithm
SE algorithm, because the symmetrical initial phase voltages for for unbalanced distribution systems. Int J Electr Power Energy Syst
2013;53:632–42.
calculation of matrices H and C (Step 3 in Section 4) are assumed.
[17] Abbasi AR, Seifi AR. A new coordinated approach to state estimation in
integrated power systems. Int J Electr Power Energy Syst 2013;45:152–8.
References [18] Hansen CW, Debs AS. Power system state estimation using three-phase
models. IEEE Trans Power Syst 1995;10:818–24.
[1] Abur A, Exposito AG. Power system state estimation: theory and [19] Kersting WH. Distribution system modeling and analysis. CRC Press; 2002.
implementation. Marcel Dekker; 2004. [20] Zhu Y, Tomsovic K. Adaptive power flow method for distribution systems with
[2] Singh R, Pal BC, Jabr RA. Choice of estimator for distribution system state dispersed generation. IEEE Trans Power Deliv 2002;17:822–7.
estimation. IET Gener Transm Distrib 2009;152:240–6. [21] Tafreshi SMM, Mashhour E. Distributed generation modeling for power flow
[3] Farantatos E, Huang R, Cokkinides GJ, Meliopoulos AP. Implementation of a 3- studies and a three-phase unbalanced power flow solution for radial
phase state estimation tool suitable for advanced distribution management distribution systems considering distributed generation. Electr Power Syst
systems. IEEE/PES Power Syst Conf Expos (PSCE) 2011;1:1–8. Res 2009;79:680–6.
[4] Lu CN, Teng JH, Liu W-HE. Distribution system state estimation. IEEE Trans [22] Cobelo I, Shafiu A, Jenkins N, Strbac G. State estimation of networks with
Power Syst 1995;10:229–40. distributed generation. Eur Trans Electr Power 2007;17:21–36.
[5] Meliopoulos AP, Cokkinides GJ, Stefopoulos GK. Numerical experiments for [23] Anderson PM. Analysis of faulted power systems. John Wiley and Sons; 1995.
three-phase state estimation performance and evaluation. Power Tech IEEE [24] Sarić AT, Ranković A. Load reallocation based algorithm for state estimation in
Russ 2005;1:1–7. distribution networks with distributed generators. Electr Power Syst Res
[6] Wu W, Ju Y, Zhang B, Sun H. A distribution system state estimator 2012;84:77–82.
accommodating large number of ampere measurements. Int J Electr Power [25] Fox B, Flynn D, Bryans L, Jenkins N. Wind power integration: connection and
Energy Syst 2012;43:839–48. system operational aspects. London, UK: IEE; 2007.
[7] Thornley V, Jenkins N, White S. State estimation applied to active distribution [26] Machowski J, Bialek JW, Bumby JR. Power systems dynamics and stability. 2nd
networks with minimal measurements. In: 15th PSCC conference, Liege, vol. 1; ed. Wiley; 2008.
2005. p. 1–7. [27] Korres GN. A robust method for equality constrained state estimation. IEEE
[8] Li K. State estimation for power distribution system and measurement Trans Power Syst 2002;17:305–14.
impacts. IEEE Trans Power Syst 1996;11:911–6. [28] Sarić AT, Ćirić RM. Integrated fuzzy state estimation and load flow analysis in
[9] Haibin W, Schulz NN. A revised branch current-based distribution system state distribution networks. IEEE Trans Power Deliv 2003;18:571–8.
estimation algorithm and meter placement impact. IEEE Trans Power Syst [29] http://ewh.ieee.org/soc/pes/dsacom/testfeeders/index.html; 2013 [accessed
2004;19:207–13. 01.07.13].
[10] Lubkeman DL, Jianzhong Z, Ghosh AK, Jones RH. Field results for a distribution
circuit state estimator implementation. IEEE Trans Power Deliv
2000;15:399–406.