1

On-Line Parameter Estimation of Saturated

Synchronous Machines
Gustavo Valverde, Student Member, IEEE, Elias Kyriakides, Senior Member, IEEE, Gerald T. Heydt,
Life Fellow, IEEE and Vladimir Terzija, Senior Member, IEEE

windings is later omitted for steady state conditions. The

Abstract—This paper presents a novel method based on the second part presents the steps of the Unscented Kalman Filter
Unscented Kalman Filter (UKF) to estimate the saturated for non-linear estimation problems. Finally, the proposed
parameters of synchronous machines. The proposed methodology method is tested using noisy measurements captured from a
uses voltage and current measurements taken from the stator
saturated machine modeled in SIMULINK.
and the field windings to update the parameters of the
synchronous machine during steady state operation. The method
also utilizes measurements of the power angle in order to relate II. MODEL OF THE SYNCHRONOUS MACHINE
the measurements (in abc components) with the voltage equations The classical model of the synchronous machine consists of
in the qd0 reference frame. The proposed methodology has been three stator windings, one field winding, one damper winding
tested and validated for a saturated synchronous machine.
in the d axis and two damper windings in the q axis. Fig. 1
presents the equivalent circuit model of the synchronous
Index Terms—Unscented Kalman Filter, parameter
estimation, saturation, synchronous machines. machine operating as a generator. The mutual inductances are
not shown in Fig. 1 for simplicity.
I. INTRODUCTION
rFD

E STIMATION of synchronous machine parameters is an

open field of research looking for the actual parameters of
machine models under various operating conditions. Accurate
v FD i FD L FD
ia

rD rs
models are needed to obtain more realistic simulations for Laa va
vD = 0 iD L bb rs ib
planning and operation studies. LD

Previous work has pointed out the importance of rQ1

Lcc
identifying the machine parameters during saturated rn vb
vQ1 = 0 i Q1 L Q1 vn rs
conditions [1], [2]. In fact, efforts have been concentrated on Ln ic
extracting the parameters from operating data [1]-[5]. The rQ2
vc
main advantage of using on-line estimation techniques is that
i Q2 L Q2
generating units are not disconnected from operation and the vQ2 = 0
in
extraction of information is harmless for the machine. Fig. 1. Park’s model of a synchronous machine.
The purpose of the work described in this paper is to
calculate the parameters of a saturated synchronous machine
using on-line measurements from the stator and rotor, through A. Voltage Equations
a robust non-linear approach. The voltage equations of the circuits shown in Fig. 1 are
Once the equivalent saturated reactances of the machine are [7]:
determined, it is easy to approximate the operation of the v abc = − R s i abc + pψ abc + v n (1)
machine under saturated conditions without using the v qdr = R r i qdr + pψ qdr (2)
manufacturer’s open circuit test or other approximations as T T
where vabc = [va vb vc] and vqdr = [vQ1 vQ2 vFD vD] are the
presented in [6]. machine voltages presented in Fig. 1 and vn is the neutral
The first part of this paper introduces the classical model of point to ground voltage. Vectors ψabc = [ψa ψb ψc]T and ψqdr =
the synchronous machine. The representation of the damper [ψQ1 ψQ2 ψFD ψD]T are the fluxes in the stator and rotor
windings, respectively. In addition, Rs=diag [rs rs rs] and Rr =
diag [rQ1 rQ2 rFD rD] and p is the derivative operator.
This work was partially supported by the EPSRC, the Cyprus Research
The fluxes can be represented in terms of currents as:
Promotion Foundation, and the ESF-COST Action IntelliCIS.
G. Valverde (e-mail: [email protected]) and V. Terzija (e-mail: ⎡ ψ abc ⎤ ⎡ L s L sr ⎤ ⎡ − i abc ⎤
[email protected]) are with The University of Manchester, United Kingdom. ⎢ ⎥=⎢ T ⎥⎢ ⎥ (3)
E. Kyriakides (e-mail: [email protected]) is with the University of Cyprus, ⎣ ψ qdr ⎦ ⎣ L sr L r ⎦ ⎣ i qdr ⎦
Nicosia, Cyprus.
G. T. Heydt (e-mail: [email protected]) is with Arizona State University, where iabc = [ia ib ic]T is the vector of stator currents and vector
Tempe, AZ, USA. iqdr = [iQ1 iQ2 iFD iD]T contains the rotor winding currents.

978-1-4577-1002-5/11/$26.00 ©2011 IEEE

 2

Matrices Ls, Lsr and Lr contain the (self and mutual) stator, In the per unit system, the inductance and reactance are the
stator-rotor, and rotor winding inductances. The elements in same if the base speed is the rated speed ωB. By using (7)-(9)
matrices Ls and Lsr change with respect to the position of the and solving for the fluxes per second [9]:
rotor whereas all elements in Lr are time invariant. ⎛ r ⎞
ψ& q = ωB ⎜ vq − ωψ d + s (ψ mq −ψ q ) ⎟ (10)
B. Park’s Transformation ⎝ xls ⎠
By using a transformation of variables it is possible to ⎛ rs ⎞
obtain a set of equations where all inductances are constant ψ& d = ωB ⎜ vd + ωψ q + (ψ md −ψ d ) ⎟ (11)
⎝ xls ⎠
irrespective of the position of the rotor.
This change of variables consists of changing the abc stator ⎛ rs + 3rn ⎞
reference frame into a qd0 rotor reference frame. The new ψ& 0 = ωB ⎜ v0 − ψ0 ⎟ (12)
⎝ xls + 3xn ⎠
variables can be determined by,
i qd 0 s = Pi abc , v qd 0 s = Pv abc , and ψ qd 0 s = Pψ abc ⎛ rFD ⎞
(4) ψ& FD = ωB ⎜ vFD + (ψ md −ψ FD ) ⎟
(13)
where, vqd0s = [vq vd v0]T , iqd0s = [iq id i0]T and ψqd0s = [ψq ψd ⎝ xlFD ⎠
ψ0]T. The subscript s indicates that these variables are referred where ω is the steady state operating speed (1 p.u.) of the
from the stator. Matrix P is defined by the Park’s machine rotor. The magnetizing fluxes are:
Transformation [8]: ⎛ψ ⎞
ψ mq = X aq ⎜ q ⎟ (14)
⎡ cos θ cos(θ − 2π ) cos(θ + 2π ) ⎤ ⎝ xls ⎠
⎢ 3 3 ⎥
P = 2 ⎢ sin θ sin(θ − 2π ) sin(θ + 2π ) ⎥ (5) ⎛ ψ d ψ FD ⎞
3⎢ 3 3 ⎥ ψ md = X ad ⎜ + ⎟ (15)
⎢ 1 1 1 ⎥ ⎝ xls xlFD ⎠
⎢⎣ 2 2 2 ⎥⎦ −1
⎛ 1 1 ⎞
where the angle θ is given by, X aq = ⎜ + ⎟ (16)
⎜ xmq xls ⎟
θ = ωs t + δ − π 2 (6) ⎝ ⎠
−1
⎛ 1 1 1 ⎞
In (6), ωs is the synchronous electrical speed in rad/s and δ is X ad = ⎜ + + ⎟ (17)
the power angle in electrical radians. x
⎝ md xls xlFD ⎠

The new voltage equations in the rotor reference frame are with all variables in p.u. except ωB.
v qd 0 s = − R sn i qd 0 s + ω ψ dqs + pψ qd 0 s (7) In steady state conditions, the damper windings carry a
current equal to zero and do not contribute to the magnetic
v qdr = R r i qdr + pψ qdr (8) field in the q and d axes. In fact, the fluxes in damper
where Rsn = diag [rs rs (rs+3rn)], ψdqs = [ψd -ψq 0]T and ω is the windings for steady state conditions become ψQ1=ψQ2=ψmq
operating electrical speed, all in p.u. After the Park’s and ψD=ψmd.
Transformation, the rotor equations presented in (2) remain Therefore, the equations and parameters corresponding to
the same, as presented in (8). damper windings have been neglected to reduce the size and
The inductances in the new reference frame are all time complexity of the estimation process without introducing any
invariant, irrespective of the rotor position. A proper selection error.
of the rotor base quantities (voltage and power) allows The currents can be easily expressed in terms of fluxes [7]:
representing the relation flux-current in terms of the 1
magnetizing inductances Lmd and Lmq and the leakage iq = (ψ mq −ψ q ) (18)
xls
inductances for each winding Lls, LlQ1, LlQ2, LlFD and LlD. This
relation can also be written in terms of reactances with the 1
id = (ψ md −ψ d ) (19)
resulting fluxes per second: xls
Ld = Lls + Lmd Lq = Lls + Lmq 1
LFD = LlFD + Lmd LQ1 = LlQ1 + Lmq i0 = − ψ0 (20)
xls + 3xn
LD = LlD + Lmd LQ 2 = LlQ 2 + Lmq
1
L0 = Lls + 3Ln iFD = (ψ FD −ψ md ) (21)
xlFD
⎡ψ q ⎤ ⎡ − Lq 0 0 Lmq Lmq 0 0 ⎤ ⎡ iq ⎤
⎢ψ ⎥ ⎢ 0 −L 0 0 0 Lmd Lmd ⎥ ⎢ id ⎥ (9)
⎢ d ⎥ ⎢ d
⎥⎢ ⎥ C. State Space Equations
⎢ψ0 ⎥ ⎢ 0 0 − L0 0 0 0 0 ⎥ ⎢ i0 ⎥
⎢ψ Q1 ⎥ = ωB ⎢ − Lmq 0 0 LQ1 Lmq 0 0 ⎥ ⎢ iQ1 ⎥ The set of equations (10)-(21) can be represented in the
⎢ψ ⎥ ⎢− L 0 0 Lmq LQ 2 0 0 ⎥ ⎢iQ 2 ⎥ compact matrix form:
⎢ Q2 ⎥ ⎢ mq ⎥⎢ ⎥ x& = f (x, u) (22)
⎢ψ FD ⎥ ⎢ 0 − Lmd 0 0 0 LFD Lmd ⎥ ⎢iFD ⎥
⎢⎣ ψ D ⎥⎦ ⎢⎣ 0 − Lmd 0 0 0 Lmd LD ⎥⎦ ⎢⎣ iD ⎥⎦ y = h ( x) (23)
 3

where x is the set of state variables, u is the input variables X k -1 = [x k -1 L x k -1 ] + c [0 Pk -1 − Pk -1 ]

and y is the algebraic (measurable) variables:
T
where Xk-1 is a n × (2n+1) matrix containing the sigma points
x = ⎡⎣ψ 0 ,ψ q ,ψ d ,ψ FD , δ ⎤⎦ (24) calculated from xk-1 and Pk-1 and c is a n+λ vector.
u = ⎡⎣v0 , vq , vd , vFD ⎤⎦ (25)
Step 2: Kalman Filter State Prediction
y = ⎡⎣i0 , iq , id , iFD ⎤⎦ (26)
In this step, the sets of sigma points calculated in Step 1, are
Since the operating speed of the machine is known for evaluated, one by one, through the prediction function defined
steady state conditions, it is not included in the set of states. as,
However, the power angle is needed as a state variable to ˆ i = f ( Xi , u ) i = 0,..., 2n
X k −1 (34)
relate the abc and qd0 components. k k -1

Due to the discrete nature of measurements, (22) and (23) ˆ

The resulting X is a n × (2n+1) matrix containing the
k
are approximated by the discrete form: propagated sigma points. Next, the predicted state mean
x k = f (x k -1 , u k -1 ) (27) vector x −k and the predicted covariance matrix Pk− are
y k = h( x k ) (28) computed:
In this work, the continuous model was approximated using 2n

the 4th order Runge Kutta discretization method. x −k = ∑ Wim X

ˆi
k (35)
i =0
2n
III. THE UNSCENTED KALMAN FILTER Pk− = ∑ Wic ⎡⎣( X
ˆ i − x − )( X
k k
ˆ i − x − )T ⎤ + Q
k k ⎦ k −1 (36)
i =0
The UKF is an efficient discrete-time recursive filter able to
The weights in (35)-(36) can be calculated using,
solve estimation problems represented in the following form
[10], [11]: λ λ
W0m = , W0c = + (1 − α 2 + β ) (37)
x k = f (x k -1 , u k -1 ) + q k -1 (29) n+λ (n + λ )
y k = h(x k ) + rk (30) 1
Wi m = (38)
where x is the state vector and y is the measurement vector, 2(n + λ )
whereas qk-1 and rk are the system and measurement Gaussian 1
noises respectively, with zero mean and uncorrelated Wi c = (39)
2(n + λ )
covariance matrices Q and R, respectively. Functions f and h
are nonlinear equations representing the system and where the variable β takes a value of two, used typically for a
measurement models in terms of the state variables and the Gaussian distribution [11].
system inputs u.
Step 3: Kalman Filter State Correction
Step 1: Sigma Points Calculation For the Kalman Filter state correction, the sigma points
corresponding to the predicted state mean vector and the
For an initial nx1 state vector xk-1, and the corresponding
covariance matrix are:
n × n covariance matrix Pk-1, a set of 2n+1 vectors is obtained.
These vectors are called sigma points and they are X −k = [ x −k L x k− ] + c [0 Pk− − Pk− ] (40)
deterministically chosen. They capture the mean and These new points are evaluated, one by one, in the non-linear
covariance of the original distribution of xk-1 exactly. update function h defined in (30), as follows:
X0k -1 = x k −1 (31) Yki − = h( Xik− ) i = 0,..., 2n (41)
Xik -1 = x k −1 + ( (n + λ )Pk -1 ),
i i = 1,..., n. (32)
The mean of the propagated points is calculated as,
X nk +-1i = x k −1 −( (n + λ )Pk -1 ),
i i = 1,..., n. (33) 2n
μ k = ∑ Wim Yk−
i
(42)
where ( ( n + λ )Pk -1 ) i is the ith column of the matrix i =0

The measurement covariance matrix and the cross-

( n + λ ) Pk -1 and parameter λ is defined as λ=α2(n+υ)-n.
covariance of the state and measurement are obtained as,
Parameter υ can be used to reduce the higher order errors of 2n
S k = ∑ Wic ⎡ (Yk− − μ k )(Yk− − μ k )T ⎤ + R k
i i

the mean and the covariance approximations; it can be υ=3-n ⎣ ⎦ (43)

i =0
or zero [12]. 2n
The square root matrix of Pk-1 can be calculated from the Ck = ∑ Wic ⎡ ( X −k − x −k )(Yk− − μ k )T ⎤
i i

⎣ ⎦ (44)
Cholesky factorization of Pk-1. For the purpose of the i =0
estimation initialization (i.e. when k=0), the initial state vector The filter gain Kk, the state xk and the covariance Pk are
and the initial covariance matrix have to be defined in advance computed as,
according to a priori knowledge of the system. K k = Ck S −k 1 (45)
Equations (31)-(33) can be expressed in the equivalent
compact form: x k = x −k + K k [y k − μ k ] (46)
 4

Pk = Pk− − K k S k K Tk (47) diagonal elements of matrix P0 were set to 10-3.

where yk is the set of measurements at time k. The elements in Q were selected from multiple simulation
trials. The selection depends on uncertainty of inputs,
IV. PARAMETER ESTIMATION sampling frequency and accuracy of the model. In this work,
the diagonal elements of matrix Q are kept constant at 10-5
The state vector is extended with the set of parameters that
during the whole estimation process.
are to be updated:
T
x ss = ⎡⎣ψ 0 ,ψ q ,ψ d ,ψ fd , δ , p ⎤⎦ (48)
1.2
The subscript ss indicates steady state conditions. The set of
parameters p vary during the estimation process to minimize 1

Armature Voltage, p.u.

the difference between the model and the measurement at k.
The transition of suspicious parameters from k-1 to k in (27) is 0.8

represented as:
0.6
p k = p k -1 (49)
In addition, the set of inputs and measurements needed to 0.4
perform the estimation process is:
u ss = [ va , vb , vc , vFD ]
T 0.2
(50)
y ss = [ia , ib , ic , iFD , δ m ]
T
(51) 0
0 500 1000 1500 2000 2500 3000
Field Current, A
where δm is the measured power angle which can be obtained
Fig. 2. Saturation curve of the synchronous machine.
by installing a sensor in the rotor of the machine to calculate
the angular displacement of the rotor with respect to the
terminal voltage in phase a [13], [14]. A. Results
Figs. 3 and 4 present the estimated saturated values for xmq
V. SIMULATION TEST and xmd. The initial state guess x0 was calculated from the
A gas turbine synchronous machine with rated capacity of unsaturated values listed in Table I. The noise in the estimated
483 MVA and 22 kV has been simulated in SIMULINK under parameters appears due to the noisy nature of inputs and
saturated conditions. The machine delivers 428.4 MW and measurements. Smaller values of elements in Q can be used to
54.5 MVAr at 21.795 kV line to line voltage. The parameters reduce the noise of the estimated parameters for each time
of the machine from manufacturer’s data are presented in sampling.
Table I [15]. 1.62

TABLE I 1.6
SYNCHRONOUS MACHINE PARAMETERS
1.58
Value Description
Parameter
(p.u.) 1.56
xmq , p.u.

rs 0.0046 Stator winding resistance for all three phases

1.54
xls 0.1600 Stator winding leakage reactance
1.52
xmq 1.5600 Mutual magnetizing reactance in q axis
xmd 1.6400 Mutual magnetizing reactance in d axis 1.5
rFD 0.000972 Field winding resistance
1.48
xlFD 0.11791 Field winding leakage reactance
1.46

The manufacturer’s open circuit test presented in Fig. 2 has 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
time, s
been used to simulate the saturation of the machine in
Fig. 3. Estimation of magnetizing reactance in q axis, xmq.
SIMULINK. The dashed line represents the voltage in the air
gap. The noisy inputs and measurements in (50)-(51) were
collected with a sampling frequency of 10 kHz, for a period of From simulation results it was found that estimation of xmq
0.32 s. directly depends on the measurement of the power angle, δm.
The diagonal elements of matrix R are kept constant at 10-3 In the absence of δm, it is not possible to guarantee accurate
during the whole estimation process. These values were estimations for xmq. On the other hand, the estimation of xmd is
selected according to the level of uncertainty of the very sensitive to the field current measurement and much less
measurements. The selected suspicious parameters to be dependent to δm.
updated are: Fig. 5 presents the estimated field winding resistance for
p = ⎡⎣ xmq , xmd , rFD ⎤⎦ (52) every time sample. The negative values in the first 50 ms are
caused due to the initial errors in xmd.
The initial guess of states x0 needed to start the UKF was
calculated using the available manufacturer’s parameters. The
 5

The estimation of rFD is 0.000954, less than 2% difference δ = ∠ Eq (54)

with respect to the value listed in Table 1. This estimation is The Park’s Transformation is then used to calculate the
very accurate taking into consideration the presence of noisy currents in the d and q axes. The voltage applied to the field
measurements, which are in the order of 10-4 (about the scale winding referred to the stator is:
of the magnitude of rFD).
Exfd = Eq + ( X dsat − X qsat ) ⋅ id (55)
1.6
Finally, the voltage and currents in the rotor are determined
1.58 by:
1.56 Exfd
iFD = sat (56)
xmd
1.54
xmd , p.u.

vFD = rFD ⋅ iFD (57)

1.52
Table II presents a comparison of the calculated rotor
1.5 variables from (53)-(57) and the measured ones for
comparison purposes. The calculated variables are obtained
1.48 using the estimated parameters from the UKF. The
1.46
compassion shows that the estimated saturated parameters are
able to represent the saturated conditions of the machine at the
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
given operating point.
time, s
Fig. 4. Estimation of magnetizing reactance in d axis, xmd.
TABLE II
COMPARISON OF MEASURED AND CALCULATED VARIABLES
-4
x 10 Variable Measured Calculated % Difference
25
δ [rad] 0.8939 0.8967 0.31
iFD [A] 2545 2550 0.19
20
vFD [V] 295.7 292.4 1.11

15
VI. CONCLUSIONS
rFD , p.u.

10 This paper presented the application of the UKF for on-line

parameter estimation of synchronous machines during steady
5
state conditions. The non-linearities of the estimation problem
were effectively approximated with the sigma point
0
calculation.
-5
It was found that the estimation of xmq is very sensitive to
the power angle measurement whereas the estimation of xmd
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
depends more on the measurement of the field current instead.
time, s
The estimation of the field winding resistance is very
Fig. 5. Estimation of field winding resistance, rFD.
accurate as long as measurements of iFD and vFD are available.
However, the estimation of the stator resistance is very poor
The stator resistance was not included in the set of
during steady state conditions. Therefore, it is preferable to
suspicious parameters as it is not possible to perform a good
estimate the stator resistance from dynamic conditions where
estimation of this parameter during steady state conditions.
more information of the machine parameters could be
extracted.
B. Validation of results The resulting parameters were validated by re-calculating
It is not possible to know the exact equivalent saturated the rotor variables and compared to the average value of the
parameters of xmd and xmq. However, it is possible to validate corresponding measurements. The comparison confirmed the
the estimations by re-calculating the rotor variables δ, iFD and accuracy of the proposed parameter estimator.
vFD by having knowledge of voltages and currents in the stator
and the saturated machine parameters. VII. ACKNOWLEDGEMENTS
The calculation of the rotor variables starts from calculating The authors would like to thank Arizona Public Service
the excitation voltage: Co. for providing the synchronous machine’s manufacturer
Eq = V + (rs + jX qsat ) ⋅ I (53) data.
where V and I are the rms voltage and current in the
VIII. REFERENCES
machine terminals and X qsat represents the saturated
[1] E. Kyriakides, G. T. Heydt, and V. Vittal, “On-line parameter estimation
reactance in q-axis. The angular difference between V and of round rotor synchronous generators including magnetic saturation,”
IEEE Transactions on Energy Conversion, vol. 20, no. 3, pp. 529-537,
Eq is the power angle. If the machine terminal voltage is used Sep. 2005.
as the reference, the power angle is:
 6

[2] H. B. Karayaka, A Keyhani, B. L. Agrawal, D. A. Selin, and G. T. Heydt Gerald T. Heydt (S’62, M’64, SM’80, F’91, LF’08) is from Las Vegas, NV.
“Identification of armature, field, and saturated parameters of a large He holds a BEEE degree from the Cooper Union in New York and MSEE and
steam turbine-generator from operating data” IEEE Transactions on PhD. degrees from Purdue University. He spent approximately 25 years as a
Energy Conversion, vol. 15, no. 2 , Jun. 2000. faculty member at Purdue, and in 1994, he took the position of site director of
[3] J. Melgoza, G. T. Heydt, A. Keyhani, B. Agrawal and D. Selin, the NSF and industrially supported Power Systems Research Center at ASU,
“Synchronous machine parameter estimation using Hartley series”, IEEE where he is now a Regents’ Professor. He is a member of the National
Transactions on Energy Conversion, vol. 16, no. 1, pp. 49–54, Mar. Academy of Engineering and a Fellow of the IEEE. In 1995 he was named
2001. Power Engineering Educator of the Year by the IEEE Power Engineering
[4] H. B. Karayaka, A. Keyhani, G. T. Heydt, B. L. Agrawal, and D. A. Society.
Selin, “Synchronous generator model identification and parameter
estimation from operating data”, IEEE Transactions on Energy Vladimir Terzija (M’95, SM’2000) is the EPSRC Chair Professor in Power
Conversion, vol. 18, no. 1 , Mar. 2003. System Engineering in the School of Electrical and Electronic Engineering,
[5] E. Kyriakides, G. T. Heydt, and V. Vittal, “On-line estimation of The University of Manchester, where he has been since 2006. From 1997 to
synchronous generator parameters using a damper current observer and a 1999, he was an Associate Professor at the University of Belgrade. In 1999, he
graphic user interface,” IEEE Transactions on Energy Conversion, vol. was awarded a prestigious Humboldt Research Fellowship. From 2000 to
19, no. 3, pp. 499-507, Sep. 2004. 2006, he was with ABB AG, Germany, working as an expert for switchgear
[6] F. P. de Mello and L. N. Hannett “Representation of saturation in and distribution automation. His main research interests are application of
synchronous machines,” IEEE Transactions on Power Systems, vol. 1, intelligent methods to power system monitoring, control, and protection,
no. 4, pp. 8-14, Nov. 1986. switchgear and fast transient processes, as well as DSP applications in power
[7] P. Krause, O. Wasynczuk, and S. Sudhoff, Analysis of Electric systems.
Machinery and Drive Systems, IEEE Press - Series on Power
Engineering, 2nd Edition, 2002.
[8] R. H. Park, “Two-reaction theory of synchronous machines –
Generalized methods of analysis – Part I,” AIEE Transactions, vol. 48,
pp. 716-727, July 1929.
[9] C. H. Thomas, “Discussion of analogue computer representations of
generators in voltage-regulation studies,” Transactions of AIEE, vol. 75,
pp. 1182-1184, Dec. 1956.
[10] S. Julier, J. Uhlmann, and H. F. Durrant-Whyte, “A new method for the
nonlinear transformation of means and covariances in filters and
estimators,” IEEE Transactions on Automation & Control, vol. 45, pp.
477-482, 2000.
[11] D. Simon, Optimal State Estimation: Kalman, H infinity, and Nonlinear
Approaches. New Jersey: John Wiley & Sons, Inc., 2006.
[12] S. Sarkka, "On unscented Kalman filtering for state estimation of
continuous-time nonlinear systems," IEEE Transactions on Automation
& Control, vol. 52, pp. 1631-1641, 2007.
[13] M. Despalatovic, M. Jadric, and B. Terzic. “Real-time power angle
determination of salient-pole synchronous machine based on air gap
measurements,” Electric Power Systems Research, vol. 78, no.11, pp.
1873-1880, Apr. 2008.
[14] E. Barrera-Cardiel and N. Pastor-Gomez, “Microcontroller-based power-
angle instrument for a power systems laboratory,” IEEE PES Summer
Meeting, Edmonton, Canada, pp. 1008–1012, 1999.
[15] E. Kyriakides, “Innovative Concepts for On-line Synchronous Generator
Parameter Estimation” PhD Thesis, Arizona State University, Tempe,
AZ, Dec, 2003.

IX. BIOGRAPHIES
Gustavo Valverde (S’08) obtained the B.Sc. degree in Electrical Engineering
from the University of Costa Rica in 2005 and the M.Sc degree in Electrical
Power Systems from the University of Manchester, in 2008. Currently he is a
PhD student at the University of Manchester working on power system state
estimation. His research interests include monitoring of power systems using
synchronized measurements, state estimation, parameter estimation, dynamics
and voltage stability assessment.

Elias Kyriakides (S’00, M’04, SM’09) received the B.Sc. degree from the
Illinois Institute of Technology in Chicago, Illinois in 2000, and the M.Sc. and
Ph.D. degrees from Arizona State University in Tempe, Arizona in 2001 and
2003 respectively, all in Electrical Engineering. He is currently an Assistant
Professor in the Department of Electrical and Computer Engineering at the
University of Cyprus, and a founding member of the KIOS Research Center
for Intelligent Systems and Networks. He is the Action Chair of the ESF-
COST Action IC0806 “Intelligent Monitoring, Control, and Security of
Critical Infrastructure Systems” (IntelliCIS) (2009-2013). His research
interests include synchronized measurements in power systems, security and
reliability of the power system network, optimization of power system
operation techniques, modelling of electric machines, and renewable energy
sources.