2012 3rd IEEE PES Innovative Smart Grid Technologies Europe (ISGT Europe), Berlin

State Estimation of Active Distribution

Networks: Comparison Between WLS and
Iterated Kalman-Filter Algorithm
Integrating PMUs
S. Sarri, Student Member, IEEE, M. Paolone, Senior Member, IEEE, R. Cherkaoui, Senior Member,
IEEE, A. Borghetti, Senior Member, IEEE, F. Napolitano, Member, IEEE, C.A. Nucci, Fellow, IEEE

to active systems, it is expected that in the future the use of

Abstract—One of the challenging tasks related to the real- PMUs will also facilitate improved operational practices of
time control of Active Distribution Networks (ADNs) is these networks.
represented by the development of fast (i.e. sub-second) state Within the above context, one of the main challenging
estimation (SE) processes. As known, the problem of SE of
tasks in the operation of Active Distribution Networks
power networks links the measurements performed in the
network with a set of non-linear equations representing the (ADNs) is represented by their SE (e.g. [10]–[12]) in view
links between the network node voltage phasors (i.e. the system of the fact that the inclusion of data gathered by PMUs
states) and measured quantities. The calculation of these certainly allows to simplify and improve the SE of ADNs
voltages is accomplished by the solution of a minimization (e.g. [12], [13])1.
problem by using, for instance, Weighted Least Squares (WLS) As known, classic methods used for the solution of the SE
or Kalman filter (KF) methods. The availability of phasor
problem are based on the use of Weighted Least Squares
measurement units (PMUs), characterized by high accuracy
and able to directly measure node voltage phasors, allows, in (WLS) method (e.g. [5]). More recently, methods based on
principle, a simplification of the SE problem. Within this the use of the Kalman filter (KF) [15] have been proposed
framework, the paper has two aims. The first is to propose a too. In particular, Blood et al. [16] evaluated three different
procedure based on the use of the Iterated KF (IKF) aiming at algorithms in order to study the performance of a pseudo–
making achievable, in a straightforward manner, the SE of dynamic system, whereas in [17] they combined the basic
ADNs integrating PMU measurements. The second goal is to
power flow equations with the load forecast to create a
present a sensitivity analysis of the performances of WLS vs
IKF methods as a function of the measurements and process discrete–time dynamic model for SE. Gelagaev [18] made a
covariance matrices. comparison between the WLS and the Extended Kalman
Filter (EKF) methods and proposed a first analysis of the
Index Terms—Active distribution networks (ADNs), Kalman relative importance of process and measurement covariance
filter (KF), iterated Kalman filter (IKF), phasor measurement matrices. Methods used for dynamic SE have been also
units (PMUs), state estimation (SE), smart grids. proposed in [19]–[25]. A further version of the KF is
represented by the Iterated Kalman Filter (IKF) which is
I. INTRODUCTION
characterized by an iterative application of the EKF for the

T HE increasing penetration of distributed energy

resources in power distribution networks will involve a
substantial evolution in the operational practice of this part
case of non-linear processes (e.g. [26] –[28]).
Within this context, the paper has two aims. The first is to
propose a procedure based on the use of the IKF aiming at
of the power network infrastructure (i.e., [1], [2]). The making achievable, in a straightforward manner, the SE of
above issues will certainly call for a massive use of ADNs integrating PMUs. The second one is to discuss the
advanced and smarter monitoring tools capable of sensitivity of the WLS vs the KF performances as a function
increasing, in principle, the operation reliability of these of the measurements and process covariance matrices. This
networks. analysis aims at showing that the behaviors of these two
In the last decade, the large use of the so–called Phasor methods are largely dependent on the selection of these
Measurement Units (PMUs) for the monitoring of the matrices and, therefore, on the availability of specific
transmission networks, has influenced the operational measurement systems.
practices of these systems concerning the following aspects: The structure of the paper is the following. Section II
i) system stability; ii) state estimation (SE); iii) system illustrates the theoretical background as well as the
reliability; and iv) protections schemes (e.g. [3]–[9]). In analytical aspects related to the proposed IKF–based SE
view of the evolution of distribution networks from passive

A. Borghetti, F. Napolitano and C.A. Nucci are with the University of

Bologna, Italy, (e-mail: [email protected],
[email protected], [email protected]).
1
S. Sarri, R. Cherkaoui and M. Paolone are with the École Polytechnique As discussed in [14], it is technically possible to develop specific
Fédérale de Lausanne EPFL, Lausanne, Switzerland, (e-mail: PMUs for ADN applications by making use of hardware architectures used
[email protected], [email protected], [email protected]). in standard control/monitoring environments.

978-1-4673-2597-4/12/$31.00 ©2012 IEEE

 2

algorithm including PMU measurements. Section III independent from w k (clearly, v k ∈ ℜ m as z k does).
presents an application example with the relevant WLS vs Equations (1) and (2) are those adopted for the general
IKF sensitivity analysis. Section IV concludes the paper formulation of the KF problem [15], [29].
with some final remarks about the obtained results and As known, the state of an electrical power network is
future developments. associated to the identification of the phase–to–ground
voltage phasors in correspondence of the N network buses
II. INTEGRATING PMU DATA IN IKF STATE ESTIMATION
(e.g. [5]). Assuming to define the first bus of the network as
This section aims at formalizing the analytical aspects the slack one ( δ 1 = 0 ), the state vector x is composed by
related to inclusion of PMU measurements into IKF-based n = 2 N − 1 unknowns being the angles of the other network
power systems SE.
buses expressed as a function of the angle of the slack bus3:
As known, in general KF is a method aiming at
formulating the SE problem of a general system defined by T
a set of independent variables x ∈ ℜ n (where ℜn is the n – x = ⎡⎣δ *,2 ,..., δ *, N ,V 1 ,...,V N ⎤⎦ (3)
dimensional field of real numbers) whose discrete time–
evolution is described by the following set of stochastic where δ *,i ( i = 2,...N ) represents the relative phase of the
equations:
voltage phasor of i th bus as a function of the angle of the
x k = Ax k −1 + Bu k −1 + w k −1 (1) slack bus δ 1 ( δ *,i = δ i − δ 1 = δ i ). As discussed in [30], it is
worth noting that when synchrophasors measurements are
available, the selection of the bus voltage phasor angle
where:
reference in the SE problem can be suitably modified
ƒ x k and x k −1 represent the state of the system in compared to the traditional approach here used. It is also
correspondence of discrete time steps k and k − 1 important to mention that, if PMU measurements are
respectively; available, at least one of these should be located in
ƒ u k −1 ∈ ℜ h represents a set of h control variables of the correspondence of the slack bus in order to properly define
equation (3).
system at time step k − 1 ;
We can express the power injections u as
ƒ w k −1 represents the system process noise assumed
white and with a normal probability distribution T
u = ⎡⎣ P1 ,..., P N , Q1 ,..., Q N ⎤⎦ (4)
(clearly, w ∈ ℜn );
ƒ A is a n × n matrix that links that state of the system
at time step k − 1 with the one of the current time step where P i and Q i ( i = 1,...N ) represent the active and
k for the case of null active injections and process reactive power injections (produced by generation systems,
noise; absorbed by loads or exchanged by storage systems) in
ƒ B is a n × h matrix that links the time evolution of the correspondence of the i th network bus.
state of the system with the h injections at current Let now present a form of (1) applicable to the specific
time step k − 1 for the case of null process noise. case of power networks. In particular, let make use of the
The SE is based on the availability of a set of known formulation of the load flow problem in the implicit
measurements z ∈ ℜm that are, in general, non–linearly form.
N
linked with the system state by means of the following set of f Pi ( x, u ) = P i − V i ∑V h[G ih cos (δ i − δ h ) +
non–linear equations2: h =1

+ B sin (δ − δ
ih i h
)] = 0
z k = h( x k ) + v k (2) N
(5)
f Qi ( x, u ) = Q i − V i ∑V h[G ih sin (δ i − δ h ) +
where: h =1

ƒ z k represents the set of the available measurements in − B cos (δ − δ

ih i h
)] = 0
correspondence of the current time step k ; By linearizing (5) as a function of the system state x and
ƒ h(x k ) is a measurement function that links the set of power injections u and considering that f Pi ( x, u ) and
measurements at time k with the system state, at the
f Qi ( x, u ) are implicit functions, we obtain:
same time step k , for the case of null measurement
noise ( h(x k ) ∈ ℜm as z k does);
ƒ v k represents the measurement noise at the same time
step k ; it is assumed white and with a normal equations. We have preferred to state the problem in this way as we are
probability distribution. v k is also assumed making use of the IKF.
3
It is worth mentioning that, although distribution networks are
characterized by unbalanced lines and unsymmetrical loads with
consequent presence of different sequence components into the three–phase
voltage and current phasors, we here consider, as a first approximation, the
2
The link between the system state, xk, and the available measurements, case of balanced networks. Therefore, we shall make reference to the direct
zk, provided by (2) could appear in the literature as a linearized set of sequence only. Moreover, the topology of the network is assumed known.
 3

∂f ( x, u ) ∂f ( x, u ) elements appearing in (11).

Δx + Δu + e = 0 (6)
∂x ∂u
The term
where

γ k = z k − h ( x k ,i ) (12)
f ( x, u ) = ⎡⎣f P ( x, u ) , fQ ( x, u ) ⎤⎦ (7)

is called measurement innovation (e.g. [29], [31]) and it

Equation (6) can be also written as:
represents the new information brought by the latest
observation.

−J

Now, we should express (2), or, in other terms (12), for
⎡ ∂f P ( x, u ) ∂f P ( x, u ) ⎤ 
I

⎢− − ⎥ the problem of interest. In order to include the PMU data in
∂δ ∂V ⎥ ⎡ I 0⎤
⎢ Δx + ⎢ ⎥ Δu + e = 0 (8). our algorithm, the set of measurements is assumed to be
⎢ ∂fQ ( x, u ) ∂fQ ( x, u ) ⎥ ⎣0 I ⎦ composed as follows:
⎢− − ⎥
⎣ ∂δ ∂V ⎦ ƒ type– d nodes where we assume to measure phase–to–
ground voltage phasors by means of PMU devices;
Equations (6) or (8) appear closer to (1). Indeed, we can ƒ type– s nodes where we assume to measure active and
rewrite (8) as follows: reactive power injections so that d + s ≥ N .
The relevant array z of measurements defined by (2) is,
x k = x k −1 + J −k 1 ( u k − u k −1 ) + J −k 1e (9). therefore, the following:

T
⎡ ⎤
d s
The term J −k 1e of (9) corresponds to the system process z
  z

noise w k of (1) and, in our case, is due to the error of the z = δ ,..., δ ,V ,...,V , P ,..., P , Q1 ,..., Q s ⎥
⎢ 2 d 1 d 1 s
(13).
⎢ ⎥
linearization process of (5). ⎣ ⎦
Equation (9) may be adopted as a pseudo–dynamic model
of the network with matrix A of (1) equal to the identity As the definition of the blending factor, or Kalman gain
matrix. However, in KF–based SE methods this equation is K, is based on a linearized link between the system state x
only used to provide the so–called a–priori estimation of the and the measurements z, we need to derive a linearized
system state at a given time step. In principle, the proximity version of (2) which can be written as follows:
of the a–priori state predicted by (9) to the real one has a
direct influence on the number of iterations that the KF has z k = H k x k + v 'k (14)
to perform to converge. As this aspect deserves specific
investigations, we are here assuming for the a-priori where H k represents the above-cited linear dependency and
estimated state the so–called ‘flat–start’ composed by v 'k a combined measurement and h(xk)–linearization errors.
voltage angles are equal to zero and voltage magnitudes
equal to one in per–unit: Let define a ( 2 d − 1) × ( 2 N − 1) matrix T composed by
the rows of the ( 2 N − 1) identity matrix corresponding to
T
⎡P N −1 P N
⎤ the type– d nodes where phase–to–ground voltage phasors
x k ,0 = ⎢ 0,...,0,1,...,1⎥ (10)
⎣⎢ ⎦⎥ are measured by PMUs. Matrix T allows linking the first
part of the measurement array z dk to the system state
where x k ,0 indicates the a-priori state in correspondence of variables.
the generic k-th time step4.
After the definition of x k ,0 , the IKF algorithm starts an z dk = Tx k + v dk (15)

iterative process that, at the generic time step k , aims at

minimizing the a–posteriori error by means of a non–linear where, the term v dk represents the array of the uncertainties
combination of an a–priori state estimate and a weighted of the state variables measured by the PMUs.
difference between the current measurement z k and a We should now find a linear expression that allows to link
measurement prediction h(x k ,i ) [29]. the measured injected power into type– s nodes, namely z sk ,
to the state of the system. For this purpose, we can use (8):

x k ,i +1 = x k ,i + K k ,i ⎡⎣ z k − h ( x k ,i ) ⎤⎦ (11) ⎡ ∂f Ps ( x, u ) ∂f Ps ( x, u ) ⎤
⎢− − ⎥
⎢ ∂δ ∂V ⎥ ( x k − x k −1 ) +
where K is the so–called KF blending factor or Kalman ⎢ ∂fQs ( x, u ) ∂fQs ( x, u ) ⎥ (16)
gain. In what follows we will define in detail all the ⎢− − ⎥
⎣ ∂δ ∂V ⎦
4
As it will be clarified next, subscript ‘0’ of (12) corresponds to the KF
+ ( z sk − z sk −1 ) + e = 0
process is at iteration zero.
 4

∂ ⋅s where vk , jj are the elements of the array v k , and Pk ,i −1 of

where terms refer to the rows of the Jacobian matrix J
∂⋅ (21) is the a–priori estimate covariance matrix defined as:
relevant to type– s nodes and z sk , z ks −1 to the measured
injected active and reactive powers in correspondence to Pk ,i −1 = APk −1AT + Q k −1 (23).
type– s nodes.
From (16), we can then infer z sk : In our case, matrix A is the identity matrix I and Q k −1
represents the process covariance matrix, namely the
⎡ ∂f Ps ( x, u ) ∂f Ps ( x, u ) ⎤ approximations introduced into the linearization of the
⎢ ⎥ process. It is worth observing that in the literature the value
∂δ ∂V
z ks = ⎢ s ⎥ x k + v sk (17) of matrix Q k −1 is usually selected arbitrarily although, in
⎢ ∂fQ ( x, u ) ∂fQs ( x, u ) ⎥
⎢ ⎥ principle, it could be computed if the process is known.
⎣ ∂δ ∂V ⎦ As the relative values of Q k −1 and R k directly influence
the behavior of the IKF, they will be the subject of the
where the term v ks is given by: sensitivity analysis reported in the next section. In our case
we assume the values of Q k −1 known and constant for
⎡ ∂f Ps ( x, u ) ∂f Ps ( x, u ) ⎤ different time steps k (i.e. Q k −1 = Q ).
⎢ ⎥
∂δ ∂V
v ks = − ⎢ s ⎥ x sk −1 + z ks −1 − e (18).
⎢ ∂fQ ( x, u ) ∂fQs ( x, u ) ⎥ The iterative process is, then, the following.
⎢ ⎥ i) Iteration i = 0:
⎣ ∂δ ∂V ⎦
ƒ construction of the state x k ,0 by using (10);
The term v ks is clearly not constant with the time step k ƒ projection of the error covariance ahead:
and represents an equivalent uncertainty that combines the
one of the type–s measurements, with the one related to the Pk ,0 = Pk −1 + Q (24).
approximation of (8).
Therefore, (2) could be expressed as follows: ii) Iteration i > 0:
ƒ Computation of the blending factor by using (21);
Hk
  ƒ update of the state:
⎡ ⎤
⎢ T ⎥
⎢ s ⎥ x k ,i = x k ,i −1 + K k ,i ⎡⎣ z k − h k ( x k ,i −1 ) ⎤⎦ (25).
⎢ ∂f P ( x, u ) ∂f P ( x, u ) ⎥
s
zk = ⎢ ⎥ xk + vk (19)
∂δ ∂V ƒ update of the error covariance matrix:
⎢ s ⎥
⎢ ∂fQ ( x, u ) ∂fQs ( x, u ) ⎥
⎢⎣ ∂δ ∂V ⎥⎦ Pk ,i = ( I − K k ,i H k ) Pk ,i −1 (26).

where The stopping criterion applies to the absolute values of

differences between two consecutive iterations of phase–to–
⎡ v dk ⎤ ground voltages angles and magnitudes. In particular, the
vk = ⎢ s ⎥ (20).
⎢⎣ v k ⎥⎦ following criteria have to be satisfied simultaneously. The
values of ε δ and εV are suitably chosen as a function of the
The only matrix that remains to be defined is the blending values of phase–to–ground voltages angles and magnitudes.
factor K . As defined in [29], [32], and [33], the typical
form to express this matrix in order to make (11) a (
max δ k ,i − δ k ,i −1 < ε δ )
converging process is the following.
( )<ε
(27).
max V −V V
k ,i k ,i −1
K k ,i = Pk ,i −1HTk ( H k Pk ,i −1HTk + R k )
−1
(21)
III. APPLICATION EXAMPLE AND SENSITIVITY ANALYSIS
where R k is the diagonal matrix associated to the Fig. 1 shows the IEEE 13–bus distribution test feeder [34]
uncertainties of the measurements: that has been used to obtain the results reported in this
section. Bus 1 has been assumed to be the slack of the
⎡ vk ,11 " 0 ⎤ system as it represents the connection to the sub–
⎢ ⎥ transmission network. The test feeder has been considered
Rk = ⎢ # % # ⎥ (22) to be balanced, therefore, as mentioned in note 1 of Section
⎢ 0 " vk , mm ⎥ II, we have made reference to the direct sequence only.
⎣ ⎦
 5

Slack: 4.16kVR
LF
Phase:0
LF1
The procedure adopted to evaluate the accuracy of the
state estimation is described below.
BUS_1
1) For each time step (i.e. every 15 min), a load flow is

Line1_2

PI
BUS6 BUS_5 BUS_3 BUS_4 computed to determine the true state of the system.

+
BUS_2
2) The measured quantities are provided by modifying
+
Line5_6
+
Line2_5
+
Line2_3
+
Line3_4
their true values obtained from the previous step 1) and
PI PI PI PI

adding randomly–generated errors which are function

Load6 LF

LF
Load4
Load3
LF
Load2
of assumed standard deviations of the measurements
Load5
LF

LF

error distributions.
Line2_7

PI

3) On the basis of the set of measurements provided by the

+

BUS_11 BUS_10 LF BUS_8 BUS_9

BUS_7 Load7 previous step 2), the estimation of the system state is
Line10_11 Line7_10 Line7_8 Line8_9 achieved by using two different computational methods,
+ + + +
PI PI PI PI namely the IKF algorithm described in Section II and

Load9
Load11 LF

LF
the WLS (e.g. [5]).
Line7_12
Load10
LF

PI
Line10_13

LF
Load8
PI

The comparison between the two algorithms is performed

+

for different measurements ( R ) and process covariance

Load13

Load12

BUS_13 BUS_12

( Q ) matrices [29]. Such an analysis is of importance as

LF

LF

Fig. 1. The simulated IEEE 13–bus distribution test feeder [34].

these two quantities are largely influencing the
performances of the KF process. In particular, R is
The configuration of the measurement points, namely,
weighting how much the KF trusts the measurements,
measured phasors and active–reactive power injections, is
whereas Q weights how much the KF trusts the estimated
reported in Table I. The daily total active and reactive load
profiles are shown in Fig. 2. They are separated by 15 values. Tuning the SE-KF involves achieving suitable
minutes time intervals. The loads are not equally distributed values for R and Q matrices, so that the best estimation
in the system and their repartition is a function of a–priori accuracy is obtained [35]. For general matrices we need to
assumed rated power of secondary substations. compare the eigenvalues of Q and R . In case Q and R
are diagonal with all equal terms say Q and R
TABLE I
TYPE OF MEASUREMENTS IN THE SIMULATED NETWORK
( Q = QI, R = RI ), we can define a ‘tuning ratio’ Q / R . As
Bus R gets larger and Q gets smaller, the tuning ratio becomes
PMU measurements PQ measurements smaller, whereas as R gets smaller and Q gets larger, it
Number
1 9 9 becomes larger. Therefore, two different cases have been
2 9 analyzed for each of the above-mentioned conditions. Table
3 9 9
II reports the specific values adopted for R and Q
4 9
5 9 9 concerning Cases I and II. With reference to Case I, Fig. 3
6 9 shows the true and estimated magnitudes and phases of
7 9 9 direct-sequence voltage in correspondence of bus 7 of Fig. 1
8 9 9
obtained by using WLS and IKF methods. As in this case
9 9
10 9 9 the tuning ratio is small, the IKF estimations are
11 9 characterized by a lower quality, as expected, compared to
12 9 those provided by WLS. On the other hand, Fig. 4 shows the
13 9 comparative performance of the two methods for a larger
6
value of the tuning ratio (see Table II). In this case, the IKF
3
x 10
gives results characterized by the same level of accuracy as
WLS. These results might be influenced by the type and
Active power
Reactive power

2.5
location of available measurements and will be the subject
of further investigations.

2
TABLE II
SELECTED VALUES OF THE MEASUREMENTS ( R ) AND PROCESS ( Q )
Total load power

1.5
COVARIANCE MATRICES FOR FIG. 3 AND 4

Quantity Case I Case II

-5
1
PMU RMS variance [p.u.] 5.778·10 5.778·10-4
PMU phase variance [rad] 6.561·10-5 6.561·10-4
R Active power variance [p.u.] 4.000·10-5 4.000·10-4
0.5 Reactive power variance [p.u.] 4.000·10-5 4.000·10-4
Q 10-7 10-1
Tuning ratio Q / R 10-2 103
0
0 5 10 15 20 25
Time (h)

Fig. 2. Adopted daily total active and reactive load profiles.

 6

1 1

V o l ta g e ma g n i tu d e (p .u .)
V o l ta g e m a g n i tu d e (p .u .)

0.99 0.99

0.98 0.98

0.97 0.97

True value True value

0.96 WLS estimation 0.96 WLS estimation
IKF estimation IKF estimation
0.95 0.95
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90
Ti m e s te p s Ti m e s te p s

-0.01
0

Vo l ta g e p h a se (ra d i a n s)
-0.01 -0.02
V o l ta g e p h a s e (r a d i a n s )

-0.02 -0.03

-0.03
-0.04
-0.04 True value
True value -0.05 WLS estimation
-0.05 WLS estimation
IKF estimation
IKF estimation
-0.06
-0.06 10 20 30 40 50 60 70 80 90
10 20 30 40 50 60 70 80 90 Ti m e s te p s
Ti m e s te p s

a) a)
-3
-3 x 10
x 10 1.5
10
IKF relative error
IKF relative error
8 1 WLS relative error
WLS relative error

re l ati ve e rror (p .u .)
rel ati ve e rro r (p.u.)

Vo lta g e ma gn i tu d e
Vol ta ge mag ni tu de

6 0.5

4 0

2 -0.5

0 -1

-2 -1.5
10 20 30 40 50 60 70 80 90 10 20 30 40 50 60 70 80 90
Ti m e s te p s Ti m e s te p s

-4 -4
x 10 x 10
8
IKF relative error 2 IKF relative error
6
re la ti ve error (rad ia ns)

WLS relative error

re l a ti ve erro r (ra di a n s)

WLS relative error

4 1
Vol tag e ph ase

Vo l tag e p h a se

2
0
0

-2 -1
-4
-2
-6
10 20 30 40 50 60 70 80 90
Ti m e s te p s 10 20 30 40 50 60 70 80 90
Ti m e s te p s
b) b)
Fig. 3. True and estimated values of the direct-sequence voltage phasor in Fig. 4. True and the estimated values of the direct-sequence voltage phasor
correspondence of node 7 of the IEEE 13 nodes test distribution feeder of in correspondence of node 7 of the IEEE 13 nodes test distribution feeder
Fig. 1 for values of R and Q corresponding to Case I of Table II: a) absolute of Fig. 1 for values of R and Q corresponding to Case II of Table II: a)
magnitude and phase of voltage phasor; b) relative errors of magnitude and absolute magnitude and phase of voltage phasor; b) relative errors of
phase of voltage phasor. magnitude and phase of voltage phasor.

TABLE III
Fig. 5 shows the norm variation of the relative errors
SELECTED VALUES OF THE MEASUREMENTS ( R k ) AND PROCESS ( Q )
related to all bus–voltage magnitudes and phases for a
COVARIANCE MATRICES FOR FIG. 5
specific time step (the rest of the time steps provide similar
Quantity Value behavior). These results are obtained for a constant value of
PMU RMS variance [p.u.] 5.778·10-6
R and different values of Q as reported in Table III. Fig. 6
PMU phase variance [rad] 6.561·10-6 shows the complementary results of Fig. 5 but with a
R Active power variance [p.u.] 4.000·10-6 constant value of Q and different values of R, as reported in
Reactive power variance [p.u.] 4.000·10-6 Table IV.
Q [10-3÷10-7]
As it can be observed from Fig. 5 and Fig. 6, there could
be a trade–off in the performance of IKF and WLS as a
TABLE IV
function of the values assumed by Q and R .
SELECTED VALUES OF THE MEASUREMENTS ( R k ) AND PROCESS ( Q )
This can be justified, by taking a more thorough look to
COVARIANCE MATRICES FOR FIG. 6
(21). The following term, that is a component of the Kalman
Quantity Value Gain K, is called measurement prediction covariance
RMS var. [p.u.] 5.778·[10-3÷10-7]
matrix:
phase var. [rad] 6.561·[10-3÷10-7]
R Act. power var. [p.u.] 4.000·[10-3÷10-7] S k ,i = H k Pk ,i −1HTk + R k (28).
React. power var. [p.u.] 4.000·[10-3÷10-7]
Q 10-4
Assuming increasing values of R k , in general, elements
of matrix S increase, but, due to the use of the inverse S −1
 7

in the definition of K in (21), the overall weight of K from PMUs and nodal power injections. The algorithm has
decreases, which leads to an overall decrease of the been conceived in order to specifically take advantage of
contribution of the measurement innovation γ in the update PMUs measurements.
equation (25). On the other hand, increasing values of Q By making reference to the IEEE 13–bus test feeder, the
and, as a consequence of P , might lead to higher values for paper has shown a comparative analysis of the developed
S . However, it is worth noting that P is also adopted for IKF algorithm in comparison to the traditional WLS one. In
particular, this analysis has been focused on the evaluation
the calculation of K where S −1 is used as follows: of measurements and process covariance error matrices on
the performances of the two SE methods. In particular, the
K k ,i = Pk ,i −1HTk S k ,i −1 (29). paper has shown that the performances of WLS and IKF SE
methods are largely dependent on the process and
In (29) compared to S , the term Pk ,i−1 plays a more measurement covariance matrices and that their evaluation
plays an important role in the selection of the proper SE
important role in the calculation of K as Pk ,i−1 is directly
algorithm.
derived from Q (see equation (23)). Therefore, increasing Further research will focus on the deterministic
values of Q results in increasing K , which leads to an calculation of Q, on the influence of type and location of
increased contribution of the measurement innovation γ in available measurements and, finally, on the
the update equation (25). computational/performance improvements of the proposed
IKF algorithm.
-2
10
N o rm o f th e vo l ta g e ma g n i tu d e

V. REFERENCES
IKF norm
-3 WLS norm
r e l a ti ve e rr o r (p .u .)

10
[1] N. Jenkins, R. Allan, P. Crossley, D. Kirschen, and G. Strbac,
10
-4
Embedded Generation. London, U.K.: Inst. Elect. Eng., 2000.
[2] W. Allen, “Effects of wide-area control on the protection and
10
-5
operation of distribution networks,” in Power Systems Conference,
2009, PSC 2009, Clemson, South Carolina, USA, pp. 1–10.
10
-6
-7 -6 -5 -4 -3
[3] S. E. Stanton, C. Slivinsky, K. Martin, and J. Nordstrom,
10 10 10
Q
10 10
“Application of phasor measurements and partial energy analysis in
stabilizing large disturbances,” IEEE Trans. on Power Systems, vol.
10
-3 10, issue 1, pp: 297 – 306, Feb. 1995.
IKF norm [4] A. G. Phadke and J. S. Thorp, Synchronized Phasor Measurements
N o r m o f th e vo l ta g e p h a se

WLS norm
and Their Application. New York: Springer, 2008.
re l a ti ve e r ro r ( ra d i a n s )

-4
10
[5] A. Abur and A. G. Exposito, Power System State Estimation—Theory
10
-5
and Implementation. New York: CRC Press, M. Dekker, 2004.
[6] A. Ishibashi, M. Imai, K. Omata, S. Sato, T. Takagi, Y. Nakachi, and
10
-6
S. Ogawa, “New type of islanding detection system for distributed
generation based on voltage angle difference between utility network
10
-7
-7 -6 -5 -4 -3
and distributed generation site,” in Proc. 8th IEE Int. Conf. Develop.
10 10 10
Q
10 10
Power Syst. Protection, 2004, vol. 2, pp. 542–545.
Fig. 5. Variation of the norms of the voltage phase and magnitude relative [7] D. M. Laverty, D. J. Morrow, R. J. Best, and P. A. Crossley,
errors, for a specific time step (time step 53) and for a constant value of R “Differential ROCOF relay for loss-of-mains protection of renewable
(shown in Table III), as a function of Q. generation using phasor measurement over internet protocol,” in Proc.
10
-2 CIGRE/IEEE Power Energy Soc. Joint Symp. Integr. Wide-Scale
Renewable Resources Power Del. Syst., Calgary, AB, Canada, Jul.
N o rm o f th e vol ta g e m a g n i tu d e

-3 29–31, 2009, pp. 1–7.

re l a ti ve e rro r (p .u .)

10
[8] O. Samuelsson, M. Hemmingsson, A. H. Nielsen, K. O. H. Pedersen,
10
-4 and J. Rasmussen, “Monitoring of power system events at
transmission and distribution level,” IEEE Trans. Power Syst., vol.
10
-5 IKF norm 21, no. 2, pp. 1007–1008, 2006.
WLS norm [9] J. Tlusty, A. Kasembe, Z. Muller, J. Svec, T. Sykora, A. Popelka, E.
10
-6 V. Mgaya, and O. Diallo, “The monitoring of power system events on
10
-7 -6
10 10
-5
10
-4 -3
10
-2
10 transmission and distribution level by the use of phasor measurements
R
units (PMU),” in Proc. 20th Int. Conf. Exhib. Elect. Distrib. (CIRED),
-2
Prague, Czech Republic, Jun. 8–11, 2009.
10 [10] A. G. Exposito, A. de la Jaen, and J. L. R. Izaga, “An alternative state
estimation formulation for radial distribution networks,” in Proc. of
N o rm o f th e vo l ta g e p h a se
re l a ti ve e rro r (ra d i a n s)

10
-4 the 2007 IEEE Lausanne PowerTech, 2007, pp. 396–400.
[11] M. M. Nordman, and M. Lehtonen, “Distributed agent-based state
estimation for electrical distribution networks,” IEEE Trans. on
10
-6
PWRS, vol. 20, no. 2, pp. 652–658, 2005.
IKF norm
WLS norm
[12] M. Powalko, K. Rudion, P. Komarnicki, and J. Blum-schein,
“Observability of the distribution system,” in Proc. 20th Int. Conf.
-8
10
10
-7 -6
10 10
-5
10
-4 -3
10
-2
10
Exhib. Elect. Distrib. (CIRED), Prague, Czech Republic, Jun. 8–11,
R 2009.
Fig. 6. Variation of the norms of the voltage phase and magnitude relative [13] A. P. S. Meliopoulos, G. J. Cokkinides, F. Galvan, and B. Farnadesh,
errors, for a specific time step (time step 53) and for a constant value of Q “Distributed state estimator–Advances and demostration,” in Proc. of
(shown in Table IV), as a function of R. the 41st Hawaii International Conference on System Sciences – 2008,
Waikoloa, Hawaii, USA, Jan. 7–10, 2008.
[14] M. Paolone, A. Borghetti, and C. A. Nucci, “A synchrophasor
IV. CONCLUSIONS estimation algorithm for the monitoring of active distribution
The paper has shown the derivation of a SE algorithm networks in steady state and transient conditions,” in Proc. of the 17th
Power Systems Computation Conference (PSCC 2011), Aug. 22-26,
based on the use of IKF integrating measurements coming 2011, Stockholm, Sweden, vol. 1, pp. 213–220.
 8

[15] R. E. Kalman, “A new approach to linear filtering and prediction Switzerland, where she is pursuing the Ph.D. degree in electrical
problems,” Transaction of the ASME–Journal of Basic Engineering, engineering. Her current research interests include distribution network
pp. 33–45, 1960. monitoring and state estimation and real-time operation of electrical grids
[16] E. A. Blood, M. D. Ilic, J. Ilic, and B. H. Krogh, “A Kalman filter with special reference to power distribution networks.
approach to quasi-static state estimation in electric power systems,” in
Proc. of the 38th North American Power Symposium, 2006 (NAPS Mario Paolone (M’07–SM’10) was born in Campobasso, Italy, in 1973.
2006), 17-19 Sept. 2006, Carbondale, IL, USA, pp. 417–422. He received the M.Sc. degree (with honors) in electrical engineering and
[17] E. A. Blood, B. H. Krogh, and M. D. Ilic, “Electric power system the Ph.D. degree from the University of Bologna, Italy, in 1998 and 2002,
static state estimation through Kalman filtering and load forecasting,” respectively.
in Proc. of the IEEE PES General Meeting–Conversion and Delivery In 2005, he was appointed Researcher in Electric Power Systems at the
of Electrical Energy in the 21st Century, Pittsburgh, PA, USA, University of Bologna where he was with the Power Systems Laboratory
Hawaii, USA, July 20-24, 2008, pp. 1–6. until 2011. In 2010, he received Associate Professor eligibility from the
[18] R. Gelagaev, P. Vermeyen, and J. Driesen, “State estimation in Politecnico di Milano, Italy. Currently, he is Associate Professor at the
distribution grids,” in Proc. of the 13th International Conference on Swiss Federal Institute of Technology, Lausanne, Switzerland, where he
Harmonics and Quality of Power 2008, Wollongong, Australia, Sept. accepted the EOS Holding Chair of Distributed Electrical Systems
28–Oct. 1, 2008, pp. 1–6. Laboratory. He is secretary and member of several IEEE and Cigré
[19] G. Durgaprasad, and S. S. Thakur, “Robust dynamic state estimation Working Groups. He was cochairperson of the technical committee of the
of power systems based on m–estimation and realistic modeling of ninth edition of the International Conference of Power Systems Transients.
system dynamics,” IEEE Trans. Power Syst., vol. 13, no. 4, pp. 1331– His research interests are in the area of smart grids, with particular
1336, November 1998. reference to real-time monitoring and operation, power system protections,
[20] A. K. Sinha, and J. K. Mondal, “Dynamic state estimator using ANN power systems dynamics, and power system transients with particular
based bus load prediction,” IEEE Trans. Power Syst., vol. 14, no. 4, reference to LEMP-interaction with electrical networks.
pp. 1219–1225, November 1999.
[21] K.–R. Shih, and S.–J. Huang, “Application of a robust algorithm for Rachid Cherkaoui (M’05-SM’07) received both the M.Sc. and Ph.D.
dynamic state estimation of a power system,” IEEE Trans. Power degrees in electrical engineering in 1983 and 1992, respectively, from
Syst., vol. 17, no. 1, pp. 141–147, February 2002. École Polytechnique Fédérale de Lausanne (EPFL), Switzerland. Since
[22] Z. Huang, K. Schneider, and J. Nieplocha, “Feasibility studies of 1992, he has been senior scientist at EPFL. His research and teaching
applying kalman filter techniques to power system dynamic state interests are in electricity market deregulation, in distributed generation and
estimation,” in Proc. of the International Power Engineering storage and in power system vulnerability. He is senior member of IEEE,
Conference 2007 (IPEC 2007), Dec. 3-6, 2007, Singapore, pp. 376– member of technical program committees of various conferences, member
382. of CIGRE task forces C5-2, and former IEEE Swiss Chapter officer (2005-
[23] E. Ghahremani, and I. Kamwa, “Dynamic state estimation in power 2011). He serves regularly as reviewer for different journals and
system by applying the extended kalman filter with unknown inputs conferences. Dr. Cherkaoui is the author or co-author of more than 80
to phasor measurements,” IEEE Trans. Power Syst., vol. 26, no. 4, pp. scientific publications.
2556–2566, November 2011.
[24] J. A. de la O Serna, and J. R.–Maldonado, “Instantaneous oscillating Alberto Borghetti (M’97–SM’03) was born in Cesena, Italy, in 1967. He
phasor estimates with TaylorK–Kalman filters,” IEEE Trans. Power received the laurea degree (with honors) in electrical engineering from the
Syst., vol. 26, no. 4, pp. 2336–2344, November 2011. Universityof Bologna, Italy, in 1992.
[25] J. Zhang, G. Welch, G. Bishop, and Z. Huang, “Reduced Since then, he has been working with the power system group at the
measurement–space dynamic state estimation (ReMeDySE) for power University of Bologna, initially as a researcher and since 2004 as Associate
systems,” in Proc. of the 2011 IEEE Trondheim PowerTech, 2011, Professor of Electric Power Systems. His main research interests concern
pp. 1–7. power system analysis and optimization, power system restoration after
[26] A. S. Debs, and R. E. Larson, “A dynamic estimator for tracking the blackout, electromagnetic transients due to lightning, distribution system
state of a power system,” IEEE Trans. on Power Apparatus and operation, and microgrids. He is Associate Editor of IEEE
Systems, vol. PAS–89, no. 7, pp. 1670–1678, Sept.–Oct. 1970. TRANSACTIONS ON SMART GRID.
[27] B.M. Bell, F.W. Cathey, “The Iterated Kalman filter update as a
Gauss-Newton method”, IEEE Transactions on Automatic Control, Fabio Napolitano was born in Lucera, Italy, in 1975. He received the
Vol. 38(2), Feb 1993. M.Sc. degree (with honors) in electrical engineering, in 2003, and the Ph.D.
[28] N. Xin–liang, Z. Guo–qing, L. Yuan–hua, and C. Hong, “An degree from the University of Bologna, Bologna, Italy, in 2009.
improvement on the iterated Kalman filter,” in Proc. of the Radar He is currently in the Power System Group, Department of Electrical
Conference, 2009 IET International, Guillin, China, April 20–22, Engineering, University of Bologna. His research interests include power
2009, pp. 1–4. system transients, lightning electromagnetic pulse interaction with electrical
[29] G. Welch and G. Bishop, An introduction to the Kalman Filter. TR networks.
95-041, Dep. of Computer Science, University of North Carolina,
USA, July 2006. Carlo Alberto Nucci (M’91–SM’02–F’07) is full professor and head of the
[30] Jun Zhu, A. Abur, “Effect of Phasor Measurements on the Choice of Power Systems Laboratory of the Department of Electrical Engineering of
Reference Bus for State Estimation”, Proc. of the 2007 IEEE PES the Faculty of Engineering of Bologna, where he is now serving as Deputy
General Meeting, 2007. Dean. Prof. Nucci’s research interests cover Power systems transients, with
[31] M. I. Ribeiro, Kalman and Extended Kalman Filters: Concept, particular reference to those originated by lightning; Dynamics of electric
Derivation and Properties, Institute for Systems and Robotics, power plants and of power systems, with particular reference to power
Instituto Superior Técnico, Lisboa, Portugal, February 2004. system restoration after blackouts and to the influence of load modeling on
[32] R. G. Brown, and P. Y. C. Hwang, Introduction to Random Signals voltage collapse simulations; Distribution Networks/Smart Grid
and Applied Kalman Filtering. Second Edition, John Wiley & Sons, management in the presence of small scale generating plants; Fault location
Inc. 1992. in distribution networks; Phasor Measurement Units for Distribution
[33] O. L. R. Jacobs, Introduction to Control Theory.Second Edition, Networks; Characterization of innovative batteries for electrical vehicles
Oxford University Press, 1993. applications; Power Systems Capacitors (medium and low voltage); Waste
[34] IEEE Distribution Planning Working Group, “Radial distribution test treatment based on Joule-effect in-situ verification. He is author or co-
feeders,” IEEE Trans. Power Syst., vol. 6, pp. 975–985, 1991. author of over 230 scientific papers published on peer-reviewed journals or
[35] R. N. Banavar, and A. Nath, Notes on Kalman Filter, Systems & on proceedings of international conferences, of five book chapters edited by
Control Eng., Indian Institute of Technology, Bombay, India, March IEE (two), Kluwer, Rumanian Academy of Science and WIT press and of a
2004. couple of IEEE Standards and some CIGRE technical brochures. He is a
Fellow of the IEEE and of the IET and has received some best
paper/technical international awards, including the CIGRE Technical
VI. BIOGRAPHIES Committee Award. Since January 2006 he is serving as Chairman of Cigré
Styliani Sarri was born in Larisa, Greece, in 1986. In 2011 she graduated Study Committee C4 ‘System Technical Performance’. Since January 2012
in electrical engineering from the Aristotle University of Thessaloniki he is serving as Editor in Chief of the Electric Power Systems Research
(AUTh), Thessaloniki, Greece. journal, Elsevier. Prof. Nucci is doctor honoris causa of the University
In 2011, she joined the Distributed Electrical System laboratory (DESL) at Politehnica of Bucharest and member of the Bologna Science Academy.
the École Polytechnique Fédérale de Lausanne (EPFL), Lausanne,