You are on page 1of 7

1

A sensitivity –based approach for static and


dynamic load parameter estimation
E. De Tuglie, Member IEEE, G. Patrono and F. Torelli

purposes such as fault diagnosis [15], parameter estimation of


Abstract—In this paper a parameter estimation, based on unknown parameters of static and dynamic load models [3-7,
Sensitivity Theory, has been applied to power system in order to 16-22], and the study of load equivalents for power system
evaluate static and dynamic load equivalents. Such equivalents dynamic analysis [1], [2], [4], [6-8], [23-26].
are represented by a single induction motor that equivalence all
Modern developments to define equivalent load models are
dynamic load and a voltage dependent static load. The developed
procedure can be implemented in on-line allowing a self-adaptive based on Artificial Neural Networks (ANNs) [9-13] or genetic
identification. Test results demonstrates the goodness and the algorithms [17-18].
practical applicability of the proposed procedure. Accurateness and simplicity characterize load models to be
adopted in power system analysis. These characteristics often
Index Terms—Parameter estimation, Sensitivity Theory, contradict each other and a compromise is necessary. In order
induction motor, static load, composite load, equivalent load to overcome these difficulties some authors have adopted
model.
simplified models to describe power system dynamic
phenomena and simplified parameter estimation procedures as
well.
I. INTRODUCTION
A constrained nonlinear recursive filter based on the data

T he accurate knowledge of static and dynamic load


parameters is one of the crucial factors in power system
transient simulations. Planning and operation studies
obtained from field tests has been adopted to estimate
parameters of an aggregate of loads [1].
Parameters of a composite load constituted by a voltage-
heavily depend on the correctness of such simulation results,
dependent static model and a first order dynamic model can be
which, in turn, are greatly dependent on the accuracy of
estimated by adopting the least square criterion [2-3]. In these
system model parameters. In many cases, dynamic parameters
papers, the dependence form frequency as well as higher order
are unavailable from equipment manufacturers or they may be
inaccurate. In order to guarantee reliable simulations of a models have been neglected since they can be considered
power system, efficient parameter estimator codes need to be unnecessary in voltage stability analyses. Derived
implemented in modern control centers. In stability analysis, computational burdensome is alleviated by aggregating
for instance, a representation of effective power demand at heterogeneous loads and adopting low order models or, at
high voltage buses is needed. This may include the aggregate most, static load models only [23-24].
effect of numerous load devices such as lighting, heating and Simplified methods have been proposed to evaluate
motors, plus some levels of transformers tap-changing and parameters of aggregate load models by fitting their outputs to
other control devices. Building up the aggregate effect by available voltage and frequency measurements. Alternative
combining device characteristics may not be possible. The procedures adopt measurements on generators instead of loads
above arguments are related to the following research [4].
question: what are appropriate mathematical structures to Sequential instead of typical simultaneous parameter
represent high voltage effective load and how can the estimation techniques contribute to minimize computational
aggregate load models be determined in practice? It is referred effort [5]. A sequential method has been applied to a
to these questions as model justification, structure, composite dynamic-static load model expressed in continuous
determination and identification respectively. Basic concepts form by minimizing a continuous error function [6].
of structural, input and numerical identifiability are An alternative approach to define equivalent load models
investigated in [7] in order to classify and theoretically argue can be represented by predictive procedures [8]. In [25] a
on existing parameter identification criteria.
discrete time prediction error model has been adopted in order
Parameter identification has long been adopted for various
to define equivalent load models in the frequency domain.
These models are particularly suitable for transient stability
E. De Tuglie is with the Dipartimento di Ingegneria dell’Ambiente e per lo analysis.
Sviluppo Sostenibile of Politecnico di Bari, 70125 Bari, Italy (e-mail:
detuglie@poliba.it) This paper deals with an equivalent load modeling for
G. Patrono and F. Torelli are with the Dipartimento di Elettrotecnica ed dynamic stability studies. For this purpose voltage
Elettronica of Politecnico di Bari, 70125 Bari, Italy (e-mail: measurements at load nodes act as input for the adaptive
g.patrono@poliba.it, torelli@poliba.it)
model developed according to the sensitivity theory [14]. The

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
2

required system excitation can be represented by typical where Q is a symmetric positive definite matrix weighting
power system perturbations deriving from normal system individual components ei = ŷi − yi of the output error, and V
events as well as contingencies. is a positive definite scalar function.
Parameters to be identified refer to a load model composed
by its heterogeneous static and dynamic aggregates. y
u System model to be
Parameters of third order dynamic equivalent have been
identified
derived fixing well defined percentages of total power demand
for both static and dynamic aggregates. In order to reduce the ŷ -
computational effort an optimal choice of parameter groups to Adaptive model
+
be identified has been suggested. e
α̂

II. A BASIC OVERVIEW OF THE APPLIED SENSITIVITY THEORY α̂ Sensitivity


model
The aim of this section is to give theoretical concepts of the
parameter estimation technique based on sensitivity theory, α̂&
that will be applied to a power system in the next sections. For
this purpose, let the dynamic and static load to be identified be ∫
given by the general state equation: α̂

x& = f(x,α, u)
(1) Fig. 1. Scheme of parameter identification.
y = g(α , x)
From (3) it is possible to determine the time derivative of
where α represents the vector of unknown parameters to be the function V:
identified. It is assumed that α is time-invariant or slowly
varying with time. ∂V
Following the methodology outlined in [14], the adaptive V& = e& = e T Qe& (4)
∂e
model will have the same structure of the real system whose
parameters need to be identified. Denoting the parameter Substituting the time derivative of the output error with the
vector, the state, and output vector of the adaptive model by following:
α̂ , x̂ and ŷ , respectively, equations describing the adaptive
model will be: ∂e &
e& = αˆ (5)
∂αˆ
xˆ& = ˆf(xˆ , αˆ , u)
(2) Eqn. (4) can be written as:
ˆy = gˆ (xˆ , αˆ )

∂e
Comparing outputs of both systems (1) and (2), the output V& = e T Q αˆ& (6)
∂αˆ
error e = ˆy - y is calculated. It is assumed that e = e( t , αˆ ) is a
unique function of α̂ and that e = 0 as αˆ = α . The goal is to Suppose that α̂ changes accordingly to the gradient of V:
change α̂ until the error e is zero.
The basic idea of the adopted methodology is reported in T T
⎛ ∂V ⎞ ⎛ ∂e ⎞ T
Fig. 1. The suggested scheme adopts a self-adaptive procedure αˆ& = − k ⎜ ⎟ = −k ⎜ ⎟ Q e (7)
including a sensitivity model. Such procedure allows real time ⎝ ∂αˆ ⎠ ⎝ ∂αˆ ⎠
identification of unknown parameters based on input-output
measurements. where the gradient is a row vector and k is a positive constant.
In order to identify parameters, the methodology requires Substituting eqn. (7) in (6) the following quadratic form
that the power system is excited by perturbations deriving can be derived:
from normal system events as well as contingencies.
The procedure will be based on the Lyapunov function T
∂e ⎛ ∂e ⎞ T
method. For this purpose, we adopt the following Lyapunov V& = − ke T Q ⎜ ⎟ Q e (8)
∂αˆ ⎝ ∂αˆ ⎠
function:
Eqn. (8) represents the time derivative of the adopted
1 T
V (e) = e Qe (3) Lyapunov function. In particular, this is a quadratic form
2 negative semidefinite. This structural characteristic guarantees

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
3

that the error function e( t , αˆ ) approaches the origin in terms of voltage u = V h = [ v Dh vQh ] T at the h-th node.
asymptotically.
The Jacobian matrix ∂e/∂αˆ appearing in (7) can be written Static load. The adopted output for this generic h-th load
as:
component will be represented by vector ˆy Sh = [ P̂Sh Q̂Sh ] T .
∂e ∂ˆy ∂gˆ ∂xˆ ∂gˆ In particular we assume a voltage dependent model as follows:
= = + (9)
∂αˆ ∂αˆ ∂xˆ ∂αˆ ∂αˆ k
P̂Sh = P0hSVh ph
(12)
The term ∂xˆ /∂αˆ represents the sensitivity matrix λ of the Q̂Sh = Q0hSVh qh
k

model to be adapted. The value of the parameter α̂ estimated


at any time is the so-called “nominal value”.
The sensitivity matrix λ is the solution of the state
2
where V h = v Dh + vQh = uT u
2
( ) 1
2
is the voltage magnitude
sensitivity equation:
at the h-th bus and the vector α Sh = [ k ph k qh ] T is the
∂f ∂f parameter vector to be identified.
λ& = λ+
∂xˆ ∂αˆ (10) The adaptive model can be expressed in vector form as
λ(0) = 0 follows:

can be generated by the sensitivity model, as shown in Fig. 1. ⎡ k̂ ph ⎤

The procedure starts with arbitrarily chosen initial ⎢ P h


ˆy Sh = ⎢ 0 S u ( )
T
u 2 ⎥
(13)
k̂qh ⎥
conditions, synthetically represented by αˆ (0) = αˆ 0 .
With the aid of the sensitivity equation (10) the following
( )
⎢ h T 2 ⎥
⎣Q0 S u u ⎦
expression is formed:
The output error is defined by the following relationship:
T
⎛ ∂gˆ ∂gˆ ⎞ T
αˆ& = − k ⎜ λ + ⎟ Q e (11) e Sh = ˆy Sh − y Sh (14)
⎝ ∂xˆ ∂αˆ ⎠

This matrix Q weights the individual components of the The goal is to change k̂ ph and k̂ qh until the error e Sh is
output error e in a suitable manner and k is a gain factor. zero or reaches a minimum value. The method, described in
Commonly, k as well as Q are chosen empirically. In the previous section and applied to a nonlinear static system is
particular, if k is chosen too large, the procedure can diverge based on the evaluation of the Jacobian matrix (9) as follows:
whereas if it is chosen too small the identification will take too
much time. ⎡ k̂ ph ⎤
( ) ( )
1
∂e Sh ⎢ P0hS uT u 2 ln uT u 2 0 ⎥
=⎢ ⎥ (15)
∂αˆ Sh ⎢ k̂qh
( ) ( )
1

III. PARAMETER ESTIMATION APPLIED TO STATIC AND ⎣ 0 Q0hS uT u 2 T
ln u u 2 ⎦
DYNAMIC LOAD MODELS
The adaptive model of the generic load, connected to the Expression (15) represents the sensitivity matrix of the
system network, through the h-th node, is characterized by model to be adapted. The adaptation procedure starts with
two components: a static component and a dynamic arbitrarily chosen initial conditions k̂ ph ( 0 ) and k̂ qh ( 0 ) .
component.
Suppose the static and dynamic load demand be previously
assigned as a percentage of the total load demand, whose Dynamic load. The dynamic load has been supposed to be
an asynchronous motor modeled by a third order model. In
active and reactive power will be denoted by P̂ h and Q̂ h , as
this representation transformer voltages, stator resistances and
follows: saturation are neglected. The differential equations describing
the induction motor connected to the h-th load node of the
P̂Sh = ρ Sh P̂ h ( Q̂Sh = ρ Sh Q̂ h ) for the static load; network can be synthesized, in the (D, Q) frame as follows
P̂Dh = ρ Dh P̂ h ( Q̂Dh = ρ Dh Q̂ h ) for the dynamic load; [27]:
h
ê&aD = −â h êaD
h
+ b̂ h vDh + σˆ hϖêaQ
h
(16)
where ρ Sh + ρ Dh = 1 . h
ê&aQ = − â h êaQ
h
+ b̂ h vQh − σˆ hϖêaD
h
(17)
According to the parameter estimation procedure suggested
in the previous Section, assume the input vector be expressed σˆ& h = − ĥ h êaD
h h
v D + ĥ h êaQ
h h
(
v D + d̂ h 1 − σˆ h )
k̂h
(18)

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
4

h
where êaD h
and êaQ are the two D,Q components of the Also for this case, the aim is to change the vector α̂ Dh until
internal voltage, σ h is the slip, and â h , b̂ h , ĉ h , d̂ h , ĥ h , k̂ h the error e Dh vanishes.
represent parameters to be identified. These parameters have The method, described in the previous section and applied
the following expressions: to a nonlinear dynamic system is based on the evaluation
following Jacobian matrix:
βϖ βϖ
ah = bh = (1 − γ ) c h = 1 dh =
K
γ γ xa γ 2H ∂e Dh ∂ˆy h ∂xˆ h ∂ˆy Dh
= Dh + (25)
h 1 h
∂αˆ D ∂xˆ ∂αˆ Dh ∂αˆ Dh
h =
xa γ 2 H
∂e Dh
with: In order to calculate the matrix it is necessary to
∂α̂ Dh
Rr xm 2
βh = γ h = 1− ∂xˆ h
xr xa x r evaluate the sensitivity matrix . For this purpose, the
∂αˆ Dh
where Rr is the rotor resistance, xr is the rotor reactance, xa following sensitivity equation needs to be solved:
is the stator reactance xm magnetizing reactance, H is the
∂xˆ& h ∂ˆf h ∂xˆ h ∂ˆf h
inertia constant, K and k h are constant characterizing the load = h + (26)
∂αˆ Dh ∂xˆ ∂αˆ Dh ∂αˆ Dh
mechanical torque.
In this formulation parameters â h , b̂ h , ĉ h , d̂ h , ĥ h have not
physical meaningful since they represent only an aggregation IV. NUMERICAL EXAMPLE
of physical parameters. This hypothesis has been adopted to
reduce the unknown parameters and, consequently, to The proposed approach is applied to the multimachine
decrease the computational effort. At the end of the system shown in Fig.1. The load at bus #5 is composed by a
identification process, such parameters will be turned into parallel connection of a static load and two induction motors.
their effective parameters of the adaptive model by simple The absorbed power of the static load represents the 33%
algebraic equations. of the total active power sunk at bus 5.
The output vector of the adaptive model for this load The static load has been described through a nonlinear
component is represented by: voltage-dependent model. In particular this static load has the
following parameter values:
ˆy Dh = [ P̂Dh Q̂Dh Î Dh h
îaD h T
îaQ ] (19) kp=2.0 kq=1.3
where:
The two induction motors are identical and are defined
2 2 through the following physical as well as aggregate
Î ah = îaD
h h
+ îaQ (20) parameters:
h
îaD (
= −ĉ h êaD
h h h h
vQ − êaQ vD ) (21)
Rr=0.07; xr=xa=3.47; xm=3.40; H=0.334; K= 0.98; kh=0.2;
−ĉ h ⎛⎜ v Dh êaD − vQh ⎞⎟
h h 2 2
îaQ = + vQh êaQ
h
− v Dh (22)
⎝ ⎠ The load at busbar under study is expressed through an
adaptive model composed by a static nonlinear (voltage-
Eqns. (31)-(34) can be synthetically rewritten in terms of dependent) component and a dynamic asynchronous load with
state variables by adopting the state vector well defined percentages with regard to the total load demand.
h h h h T
xˆ = [ eaD eaQ σ ] as follows: A three phase fault at bus 2 cleared in 0.20 s with a
subsequent reclosure operation after 1,3 s has been applied to
excite the power system.
xˆ& h = ˆf h (xˆ h , αˆ Dh , u h ) The adopted outputs for identifying the static part of the
(23)
ˆy Dh = gˆ h (xˆ h , αˆ Dh ) load consist in the active and reactive powers PS5 and QS5 .

The output error of the generic h-th asynchronous machine


will be:

e Dh = ˆy Dh − y Dh (24)

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
5

1.35

G2
G1 G3 1.3

1 1.25

1.2

kq
1.15

3
1.1
2

1.05
L1 L2

1
0 1 2 3 4 5 6 7
Time [s]

4 Fig.4. Identification of the parameter “kq”.

L3 L4
Figures from 5 to 10 show the identification process of
parameters related to asynchronous motors whose final
5 estimated values are reported in Table I.
L5
15
L5
M1 M2
10

5
Fig.2 The test system.
Paramater "a"

0
In our simulations we assumed k̂ p ( 0 ) = 1 and k̂ q ( 0 ) = 1
as initial values. As can be seen from figures 3 and 4, the -5

methodology exactly matches the two supposed known


parameters, obtaining the estimated values of, respectively, -10

k̂ p = 2 and k̂ q = 1.3 .
-15
As can be noted, the reclosure event does not change 0 1 2 3 4
Time [s]
5 6 7 8

anything in terms of static parameters since they are estimated


Fig 5. Parameter “a”.
during the phase fault.

15
2.4

2.2 10

2 5
Parameter "b"

1.8
0
kp

1.6

-5

1.4

-10
1.2

-15
1 0 1 2 3 4 5 6 7 8
0 1 2 3 4 5 6 7 Time [s]
Time [s]

Fig 6. Parameter “b”.


Fig.3. Identification of the parameter “kp”.

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
6

16 4

15 3.5

14 3

13 2.5

Parameter "Km"
Parameter "c"

12 2

11 1.5

10 1

9 0.5

8 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
Time [s] Time [s]

Fig 7. Parameter “c”. Fig 10. Parameter “kh”.

4
10

3
9

8
Current I Error [p.u.]

1
Parameter "d"

7
0

6
-1

5 -2

-3
4 0 1 2 3 4 5 6 7 8
0 1 2 3 4 5 6 7 8 Time [s]
Time [s]

Fig 8. Parameter “d”. Fig 11. Error function of current magnitude at busbar #B5.

TABLE I: ESTIMATED VALUES OF DYNAMIC LOAD PARAMETERS.


11.6 Parameter Estimated value
a 1.42
11.4
b 1.36
11.2 c 9.28
d 7.3
h 10.65
Parameter "h"

11

km 3.04
10.8

10.6
From these estimated parameters, by adopting relationships
between these parameters and the physical parameters of the
10.4 equivalent induction motor, the Tab. II follows:
TABLE II.
10.2
0 1 2 3 4 5 6 7 8
PHYSICAL PARAMETERS OF THE EQUIVALENT INDUCTION MOTOR
Time [s]
Parameter Estimated value
Fig 9. Parameter “h”. Rr 0.15
xr=xa 2.57
Fig. 11 shows the behavior of the current error. As can be xm 2.52
noted, after the first phase characterized by high errors the H 0.44
methodology rapidly identify the parameters. K 6.35
km 3.04

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.
7

In order to test the goodness of the equivalent load another [5] Ju P., Handshin E., Wei Z.N., Schlucking U., “Sequential parameters
estimation of a simplified induction motor load model”, IEEE Trans. on
test was performed considering the loss of load at bus #2. For Power Systems, Vol.11, No.1, February 1996.
purpose of comparisons, following a loss of load at bus #3, we [6] Ju P., Handshin E., Karlsson D., “Nonlinear dynamic load modelling:
tested outputs in terms frequency behavior at generating unit model and parameter estimation”, IEEE Trans. on Power Systems,
Vol.11, No.4, November 1996.
G2 of the real system and the system with the identified
[7] Ju P., Handschin E., “Identifiability of load models”, IEE Proc.-Gen.
equivalent load. As can be noted both systems experience the Transm. Distr., Vol.144, No.1, January 1997.
same frequency behavior. [8] Knyazkin V., Canizares C.A., Soder L.H., “On the parameter estimation
and modeling of aggregate power system loads”, IEEE Trans. on Power
1.025 System, Vol.19, No.2, May 2004.
[9] M. Bostanci, J. Koplowitz, C.W. Taylor, “ Identification of power
1.02
system load dynamics using artificial neural networks”, IEEE
Transactions on Power Systems, Vol. 12, No. 4, Novembre 1997, pp.
1.015
1356-1361.
1.01 [10] T. Hiyama, M. Tokieda, “Artificial neural network based dynamic load
modeling”, IEEE Transactions on Power Systems, Vol. 9, No. 3,
Frequency [p.u.]

1.005 Novembre 1998, pp. 876-880.


[11] A.M. Stankovic, A.T.Saric, M.Milosevic, “Identification of
1
nonparametric dynamic power system equivalents with artificial neural
networks”, IEEE Trans. On Power Systems, Vol.18, No.4, November
0.995
2003.
0.99 [12] B.Y. Ku, R.J. Thomas, C.Y. Chiou, C.J. Lin, “Power system dynamic
load modeling using artificial neural networks”, IEEE Trans. On Power
0.985 Systems, Vol.9, No.4, November 1994.
[13] A.P. Alves da Silva, C. Ferreira, A.C. Zambroni de Souza, G. Lambert-
0.98
0 1 2 3 4 6 7 8 Torres, “A new costructive ANN and its application to electric load
Time [s]
representation”, IEEE Trans. On Power Systems, Vol.12, No.4,
November 1997.
Fig 12. Frequency at generating bus 2. Comparison between the case of
[14] P. M. Frank, “Introduction to Systems Sensitivity Theory”, Accademic
real load (continuous line) and equivalent load (dotted line).
Press, 1989.
[15] R.J. Patton, P.M.Frank, R.N. Clark, “Issues of fault diagnosis for
dynamic systems”, 1987.
V. CONCLUSIONS [16] M.R. Akbarzadeh, G. Faezian, H. Tabatabaei, Y.N. Sargolzaei, “ A new
variable structure control methodology for electrical/mechanical
In this paper a parameter estimation procedure, based on parameter estimation of induction motor”, Proc. of American Control
Sensitivity Theory, has been applied to power systems. Conference, Denver, Colorado, June 4-6, 2003.
[17] X. Zhou, H. Cheng, P.Ju, “ The third order induction motor parameter
The procedure has been applied to identify an equivalent estimation using an adaptive genetic algorithm”, Proc. of the 4-th World
load, composed by its static as well as dynamic components, Congress on Intelligent Control and Automation, June 10-14, 2002,
able to approximate the dynamic behavior of the real load as Shangai, China.
[18] F. Alonge, F. D’Ippolito, G. Ferrante, F.M. Raimondi, “Parameter
better as possible. identification of induction motor model using genetic algorithm”, IEE
Test results demonstrated that the proposed procedure Proc.- Control Theory and Appl., Vol.145, No.6, November 1998.
rapidly identify the unknown parameters giving rise to output [19] R. Corteletti, P.R. Barros, A.M.N. Lima, “Parameters estimation of
induction motor using subspace methlods”, IEEE 2003.
differences between the real system and the equivalent system [20] K.Wang, J. Chiasson, M. Bodson, L.M. Tolbert, “A nonlinear least
vanishes. The identification period strictly depend on the square approach for identification of the induction motor parameters”,
perturbation nature. 43rd IEEE Conference on Decision and Control, Dec. 14-17 2004,
Atlantis, Bahamas.
[21] J.A. de Kock, FS var der Merwe, HJ Vermeulen, “Induction motor
parameter estimation through an output error technique”, IEEE Trans.
VI. REFERENCES On Energy Conversion, Vol.9, No.1, March 1994.
[22] H. Chai, P.P. Acarnley, “Induction motor parameter estimation algorithm
[1] Ma Da-Qiang, Ju Ping, “A novel approach to dynamic load modeling”,
using spectral analysis”, IEE Proc.-B, Vol.139, No.3, May 1992.
IEEE Trans. on Power System, Vol. 4, No.2, May 1989.
[23] P. Ju, L.Q. Ni, F. Wu, “Dynamic equivalents of power systems with
[2] Karlsson D., Hill. J.D., “Modeling and identification of nonlinar
online measurements. Part. 1: theory.
dynamic loads in power systems”, IEEE Trans. on Power Systems,
[24] P. Ju, L.Q. Ni, F. Wu, “Dynamic equivalents of power systems with
Vo.19,No.1, February 1994.
online measurements. Part. 2: applications.
[3] Begovic M.M., Mills R.Q., “Load identification and voltage stability
[25] J.C. Wang, H.D. Chiang, C.L. Chang, A.H. Liu, “Development of a
monitoring”, IEEE Trans. on Power Systems, Vol.10, No.1, February
frequency -dependent composite load model using the measurement
1995.
approach”, IEEE Trans on Power Systems, Vol.9, No.3, August 1994.
[4] O’SullivanJW., O’Malley M.J., “Identification and validation of
[26] I.A. Hiskens, “Nonlinear dynamic model evaluation from disturbance
dynamic global load model parameters for use in power system
measurements”, IEEE Trans. on Power Systems, Vol.16, No.4,
frequency simulations”, IEEE Trans. on Power Systems, Vol.11, No.2,
November 2001.
May 1997.
[27] M. Brucoli, F. Torelli, M. Trovato, “State space representation of
interconnected power systems for dynamic interaction studies”, Electric
Power System Research, 5 (1982), pp. 315- 330.

Authorized licensed use limited to: Reva Institute of Tehnology and Management. Downloaded on October 6, 2008 at 7:3 from IEEE Xplore. Restrictions apply.

You might also like