This article was downloaded by: [University of Sydney]

On: 06 September 2014, At: 19:35

Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number:
1072954 Registered office: Mortimer House, 37-41 Mortimer Street,
London W1T 3JH, UK

Electric Power Components

and Systems
Publication details, including instructions for
authors and subscription information:
http://www.tandfonline.com/loi/uemp20

A Systematic Approach in
Estimating the Generator
Parameters
Azah Mohamed , Aini Hussain , Samila Mat Zali &
Aznan Ariffin
Published online: 30 Nov 2010.

To cite this article: Azah Mohamed , Aini Hussain , Samila Mat Zali &
Aznan Ariffin (2002) A Systematic Approach in Estimating the Generator
Parameters, Electric Power Components and Systems, 30:3, 301-313, DOI:
10.1080/153250002753598492

To link to this article: http://dx.doi.org/10.1080/153250002753598492

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all
the information (the “Content”) contained in the publications on our
platform. However, Taylor & Francis, our agents, and our licensors
make no representations or warranties whatsoever as to the accuracy,
completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of
the authors, and are not the views of or endorsed by Taylor & Francis.
The accuracy of the Content should not be relied upon and should be
independently verified with primary sources of information. Taylor and
Francis shall not be liable for any losses, actions, claims, proceedings,
demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in
relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study
purposes. Any substantial or systematic reproduction, redistribution,
 reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access
and use can be found at http://www.tandfonline.com/page/terms-and-
conditions
Downloaded by [University of Sydney] at 19:35 06 September 2014
 Elec tric P o we r C ompo ne nts and Syste ms, 30:301–313, 2002
C opyright ® c 2002 Taylor & Francis
1532-5008/ 02 $12.00 + .00

A Systematic Approach in Estimating the

Generator Parameters

AZAH MOHAMED
AINI HUSSAIN
Department of Electrical, Electronics and Systems Engineering
Universiti Kebangsaan Malaysia
43600 Bangi
Downloaded by [University of Sydney] at 19:35 06 September 2014

Selangor DE, Malaysia

SAMILA MAT ZALI

AZNAN ARIFFIN
TNB Research Sdn. Bhd.
Kawasan Institusi Bandar Baru Bangi
Selangor DE, Malaysia

One of the growing needs in transmission planning is to provide accurate sim-

ulation models of the generator, excitation system, and governor; however, the
degree of accuracy of the generator model is often questioned because model pa-
rameters currently used may have changed with time from its initial values. The
accuracy of generator parameters is one of the crucial factors in correctly simu-
lating power system dynamic events. This paper describes a systematic approach
for deriving and estimating the generator parameters in which the implemen-
tation of the method is automated by using Matlab. In the proposed method
for estimating the generator parameters, a gradient-based simulation optimiza-
tion technique is employed. The objective is to minimize the error between the
voltage response from the simulation software and the actual voltage response
from staged tests. The generator parameters are estimated by updating the pa-
rameters until the responses of the simulated outputs match the actual recorded
data. EŒectiveness of the simulation optimization technique for estimating the
generator parameters is shown by demonstrating on an example test system.

K eywords parameter estimation, synchronous generator model, optimiza-

tion technique

1. Introduction
Estimation of dynamic parameters of generators has been a challenging problem in
power systems. The accurate determination of generator parameters as operating

Manuscript received in Žnal form on 13 March 2001.

The work reported in this paper is funded by TNB Research under research project
TNB 265/ 99. The authors also wish to express gratitude to Power Technologies, Inc. ( Asia)
for assisting in this project.
Address correspondence to Dr. Azah Mohamed.

301
 302 A. Mohamed et al.

conditions change is important for power system analysis and control system design.
The operational behavior of a generator and the behavior of the power system
are decided by its dynamic parameters. Therefore, the accuracy of power system
simulation results depends greatly on the accuracy of generator parameters. Various
methods of parameter estimation have been adopted in the estimation of generator
parameters, such as the exciter, governor, transient, and subtransient parameters
[1{4].
The correctness of the present generator model parameters is often questioned
because model parameters usually provided by equipment manufacturers from de-
sign data and factory tests are not veriŽed by Želd tests. In addition, parameters
may have changed due to re-tuning, aging, and equipment changes, such as gen-
erator rewinding and revamping of excitation system. The use of more accurate
generator models can result in increases in overall power transfer capability and
Downloaded by [University of Sydney] at 19:35 06 September 2014

associated economic beneŽts. In other words, incorrect generator models can result
in the system being simulated to operate beyond the safe margins.
One of the standard procedures for obtaining synchronous machine parameters
are by means of standard machine tests so that the frequency response tests and
the physical relationships between the measured frequency responses and the ma-
chine parameters can then be established [3,5]. The method for identifying machine
parameters from frequency response tests basically use an iterative technique which
Žrst matches a set of time-constants with the measured response by curve Žtting,
and then solves a set of nonlinear equations for obtaining the machine parameters.
In this way, the operational inductances of the quadrature and direct axes, and also
the armature to Želd transfer function, can be obtained; however, the disadvantage
of this method is that the order and initial estimates of the machine parameters
need to be known in order to get accurate machine parameters.
In [6], the Prony method is used for identifying a transfer function of a lin-
earized synchronous machine model. In this method, the estimated parameters will
only be accurate when the system is close to linear and the noise levels are within
a tolerable range; however, in a generator unit, there are many nonlinear compo-
nents. Furthermore, the capability of the transfer function identiŽcation method to
estimate nonlinearities such as exciter gains and time-constants from the transfer
function is cumbersome and needs symbolic manipulation of dynamic models.
Another well-known method for obtaining the machine parameters is the method
used widely by Power Technologies, Inc. [7,8]. In this method, a series of staged tests
are conducted to obtain the responses of the machine so that the models that are
derived are closer to the response of the generator to power system disturbance. The
staged test provides su cient information to identify the values of the key generator
parameters for the computer simulation model. The method developed a test pro-
cedure to measure voltage and Želd current deviations caused by staged generator
trips starting from speciŽc loaded conditions. The staged tests can be conducted
with ease and without hazard on the generators and can be implemented during
normal operation of the machine. The terminal voltage responses obtained from
staged tests are plotted manually, and then the machine parameters are derived by
hand calculations. This procedure is considered ine cient, and therefore a method
to automate this process for the derivation of generator parameters is proposed by
implementing it in a power system and generic analytical software. In the proposed
method, the data of the generator terminal voltage response is Žrst obtained from
the staged tests. The method also developed a graphical user-friendly interface for
 Estimating the Generator Parameters 303

use in generator parameter derivation in which the data of the terminal voltage
response is inserted into the interface that will compute values of the direct and
quadrature axes’ reactances and time-constants. Initially, the generator parameters
derived are far from the actual values. Therefore, to compute acceptable parameters
for use in a power system simulation software, an estimation procedure based on
the gradient-based optimization technique is used and illustrated in this paper.

2. Generator Mathematical Description

The modeling and analysis of the generator is based on the circuit shown in Figure 1.
The stator circuit consists of three-phase armature windings carrying alternating
currents. The rotor circuits are comprised of Želd and amortisseur windings. For
purposes of analysis, the currents in the armotisseur is assumed to ow in two sets
Downloaded by [University of Sydney] at 19:35 06 September 2014

of closed circuits, one set whose ux is in line with that of the Želd along the direct-
axis and the other set whose ux is at right angles to the Želd axis or along the
quadrature-axis. From the circuits, the generator model which is based on Park’s
direct-axis and quadrature-axis representations can be derived in which the detailed
derivation of the equations can be found in [9].

3. Generator Parameters Derivation

The method for deriving the machine d-axis and q-axis parameters considers two
types of test procedures; that is, the steady state test and the dynamic test [7]. For
the steady state test, two tests are required in which they are the open-circuit test

Figure 1. Mathematical description of generator.

 304 A. Mohamed et al.

and the on-line steady-state measurements. The open-circuit test is carried out to
obtain the generator saturation curve. For the dynamic test, partial load rejection
is considered in order to obtain the terminal voltage response, hence deriving the
generator transient and subtransient reactances and time-constants.
To determine the d-axis parameters, the dynamic test procedure is to have the
machine connected to the system at zero power and the excitation system manually
controlled to establish the under-excited or over-excited initial conditions. The test
involves opening of the circuit breaker that connects the machine to the system
and recording the terminal voltage and the Želd current derivation. The recorded
terminal voltage is plotted against time. The d-axis parameter values are then
derived from the A, B, and C values, which are indicated in Figure 2. The d-axis
time-constants Td0 o and Td00o can also be obtained from the terminal voltage plots as
shown in Figure 2. Given that the value of the interrupted current is i o , which is
Downloaded by [University of Sydney] at 19:35 06 September 2014

known, the d-axis parameters are obtained using

C
Xd = ; (1)
io
B
Xd0 = ; (2)
io
A
Xd00 = : (3)
io
In the derivation of the q-axis parameters, a loading condition is considered
where the armature current is composed only of a quadrature axis component. A
load rejection test with the generator initially loaded under conditions to produce
zero d-axis armature current components would result in a terminal voltage tran-
sient response. Using the response, the quantities of the ux and current, iq o are
computed as follows:
Vt
Áq (t) = sin(¬ ); (4)
!
· µ ¶¸
P ¡ jQ
iq o = Real ¤ [cos(¬ ) ¡ j sin(¬ )] ; (5)
Vo
where ¬ is the power angle, Vt is the terminal voltage, and V o is the terminal voltage
at time t = 0.

Figure 2. Terminal voltage variation af-

ter load rejection of a purely reactive load
( machine initially under-excited) .
 Estimating the Generator Parameters 305

Figure 3. Analysis of ux for quadrature

axis parameters.
Downloaded by [University of Sydney] at 19:35 06 September 2014

The q-axis parameters of Xq , Xq0 , and Xq00 values can then be derived using
equations (6) to (8), obtained from the ux response as shown in Figure 3, as
follows:
C ¤!
Xq = ; (6)
iq o
B¤!
Xq0 = ; (7)
iq o
A¤!
Xq00 = : (8)
iq o

4. Parameter Estimation Procedure

The basic framework for the parameter estimation procedure is shown in Figure 4.
Initially, a load rejection test is performed on the generator and the measured

Figure 4. The framework of the estimation procedure.

 306 A. Mohamed et al.

outputs are recorded. One of the outputs, which is the terminal voltage response, is
then inserted into the developed Matlab derivation program to derive the machine
parameters. Using the derived parameters, a dynamic simulation is carried out using
the PSS/E simulation engine. The simulated voltage response is then compared with
the measured voltage response. If the responses do not match, then a gradient-based
optimization algorithm is used to adjust the machine parameters until the simulated
voltage response matches the measured voltage response.
The objective of our estimation task is to make the simulated model response
match the actual response by iterating on the parameters to be estimated. This
can be achieved by minimizing the error between the estimated model response
and the actual response. For this particular problem, the estimation procedures are
run for three segments of the d-axis and q-axis responses as shown in Figures 5a
and 5b. These responses are divided into three segments based on the fact that the
Downloaded by [University of Sydney] at 19:35 06 September 2014

Figure 5a. Sectionalizing the d-axis response.

Figure 5b. Sectionalizing the q-axis response.

 Estimating the Generator Parameters 307

typical values of transient time-constants (Td0 o , Tq0 o ) and subtransient time constants
(Td00o , Tq00o ) lie in the range of 3.0{10.0 secs and 0.01{0.09 secs, respectively [9]. The
choice of segmentation also depends on the terminal voltage and ux responses
obtained from the dynamic tests as shown in Figures 2 and 3, respectively. As such,
speciŽc parameters are determined from only certain segments; that is, Td00o , Tq00o ,
Xd00o , and Xq00o are from segment 1, and Td0 o , Tq0 o , Xd0 o , and Xq0 o are from segment 2;
Xd , Xq are from segment 3. The proposed segmentation procedure is to avoid the
local minimum problem and also to speed up the generator parameter estimation
process. Therefore, there are three diŒerent error functions that are minimized for
the three segments:

Min E(» ) to tal = E1 (» ) + E2 (» ) + E3 (» ); (9)

½2 ¯ 1 , ¯ 2 , ¯ 3
Downloaded by [University of Sydney] at 19:35 06 September 2014

where
v
u N
uX
E1 (» ) = t (Vterm ( ac t ) (tsegme nt 1 ) ¡ Vte rm ( sim ) (tsegme nt 1 ; ­ 1 )) 2 ;
n =1

v
u N
uX
E2 (» ) = t (Vterm ( ac t ) (tsegme nt 2 ) ¡ Vte rm ( sim ) (tsegme nt 2 ; ­ 2 )) 2 ;
n =1

v
u N
uX
E3 (» ) = t (Vterm ( ac t ) (tsegme nt 3 ) ¡ Vte rm ( sim ) (tsegme nt 3 ; ­ 3 )) 2 ;
n =1

­ 1 = Td00o ; Xd00 ; Tq0 o ; Xq00 ;

­ 2 = Td0 o ; Xd0 ; Tq0 o ; Xq0 ;

­ 3 = Xd ; X q ;

E(» ) total : Total error function of parameter » ,

Vte rm ( ac t ) (t): Actual terminal voltage at time t,
Vte rm ( sim ) (t; ­ ): Simulated terminal voltage at time t with parameter » ,
N : Total number of sampling points,
­ 1 , ­ 2 , ­ 3 : Parameter spaces for the range of the estimated parameters.

For minimizing each of the errors E1 (» ), E2 (» ), and E3 (» ), the gradient-based

optimization technique is used in the minimization process. In general, the algo-
rithm is described as follows:

i) Start from an initial guess for parameters to be estimated, say, » 0i for n = 0.

ii) The gradient at nth iteration is computed by using the Žnite-diŒerence
approximation:

@E(» ni ) E(» ni + ¢» n)

i )
g(» n
² n
= i
8» n
i
2­: (10)
@» i ¢» ni
 308 A. Mohamed et al.

iii) The correction of pni is determined by

E(» ni )
¢» n
i = ¡ 8» n
i
2­: (11)
g(» ni )

iv) Update each parameter by

n+ 1 n n n
» i
= » i
¡ ¯ ¢» i
8» i
2­: (12)

The factor ¯ in (12) is a damping ratio used to limit the correction in each step. In
our program, ¢» ni cannot surpass 10% of » ni because it is intended to improve the
convergence of the proposed method.
Downloaded by [University of Sydney] at 19:35 06 September 2014

5. Test Results and Discussion

5.1. Test System Description

The test system under study is shown in Figure 6. It is a four-bus system with
one inŽnite bus. It has two loads, a transformer, and three generators. Synchronous
machine 3 is the machine chosen for the load rejection test.

5.2. Estimation of Direct A xis Parameters

The study test system is used as a test case for studying the eŒectiveness of the
estimation method. Simulations of the test case with some chosen d-axis parameters
are used as the source of actual machine parameters for estimation purpose, and
therefore the simulation outputs are considered as the actual voltage response.
In the process of estimating the unknown d-axis machine parameters, the actual
voltage response is compared with the simulated voltage response. The diŒerence
between these responses are minimized by iterating from the least error to the
largest error out of the three errors obtained from the three segments of the voltage
response. In this case, the error E3 (» ) gives the least error, and it is minimized Žrst
and then followed by the errors E2 (» ) and E1 (» ).

Figure 6. Study test system.

 Estimating the Generator Parameters 309

Table 1
Estimation results of the d-axis parameters

D-axis parameters

Td0 o Td00o Xd Xd0 Xd00

Actual value 6.5 0.06 1.8 0.6 0.3

Initial derived value 4.98 0.39 1.75 0.79 0.30
Final estimated value 6.48 0.049 1.80 0.58 0.30
Deviation 0.02 0.011 0.00 0.02 0.00
Iterations 13 35 3 8 0
Downloaded by [University of Sydney] at 19:35 06 September 2014

Using the estimation method, the unknown d-axis parameters are computed
and the results are tabulated as shown in Table 1. From the table, the actual values
of the d-axis parameters are the chosen parameters of the test case generator. The
initial derived values of the parameters are obtained from the derivation program,
and they are considered as the initial values in the estimation procedure. The Žnal
estimated values of the parameters are the values obtained after several iterations of
the estimation process. Comparing the actual and the Žnal estimated values of the
d-axis machine parameters, it can be observed that the deviations are small, which
is less than 2%. This proves that the proposed estimation method can minimize
the deviations between the two sets of values and accurately estimate the d-axis
machine parameters.
Figure 7a shows a comparison between the actual voltage response and the
initial voltage response, whereas Figure 7b shows a comparison between the initial
voltage response and the estimated voltage response. It can be observed from the
Žgures that the estimated voltage response is almost identical to the actual voltage
response, using the Žnal parameter values tabulated above.

5.3. Estimation of Quadrature A xis Parameters

The results in estimating the q-axis parameters are as shown in Table 2. From the
table, it can be seen that the deviations between the actual and the Žnal estimated
values of the q-axis machine parameters are small; that is, less than 0.04. A compar-
ison between the actual voltage response and the initial voltage response is shown
in Figure 8a, whereas a comparison between the initial voltage response and the
estimated voltage response is shown in Figure 8b. From the Žgures, the estimated
voltage response is almost similar to the actual voltage response. Hence, the pro-
posed estimation method can accurately estimate the q-axis machine parameters.
The results obtained from the proposed estimation method show that the gen-
erator parameters can be estimated well from the actual terminal voltage and ux
responses. The estimated generator parameters are considered accurate because the
simulated voltage and ux responses are almost identical to the actual responses.
The proposed segmentation method can be considered as an eŒective method in
estimating the generator parameters because by segmenting the gradient-based op-
timization problem into three subproblems, faster convergence can be achieved and
the global optimal point can be easily obtained.
 310 A. Mohamed et al.
Downloaded by [University of Sydney] at 19:35 06 September 2014

Figure 7a. Actual and initial voltage response for d-axis.

Figure 7b. Initial and Žnal estimated voltage response for d-axis.

Table 2
Estimation results of the q-axis parameters

Q-axis parameters

Tq0 o Tq00o Xq Xq0 Xq00

Actual value 1.0 0.1 0.7 0.35 0.2

Initial derived value 1.73 0.05 0.95 0.44 0.19
Final estimated value 1.07 0.063 0.701 0.346 0.198
Deviation 0.07 0.037 0.001 0.004 0.002
Iterations 11 11 11 5 1
 Estimating the Generator Parameters 311
Downloaded by [University of Sydney] at 19:35 06 September 2014

Figure 8a. Actual and initial voltage response for q-axis.

Figure 8b. Initial and Žnal estimated voltage response for q-axis.

6. Matlab-Based Generator Parameters Derivation

The graphical user interface developed for automating the generator parameter
derivation procedure is shown in Figures 9a and 9b. Figure 9a shows the plots of the
terminal voltage response and indicates the signiŽcant points, such as the Td0 o and
Td00o . The interface also computes the values of the d-axis reactances (Xd ; Xd0 ; Xd00 )
and the time constants Td0 o and Td00o ; however, Figure 9b shows the plots of the ux
response and the interface computes the values of the q-axis reactances (Xq ; Xq0 ; Xq00 )
and time constants Tq0 o and Tq00o .

7. Conclusions
A systematic approach in deriving and estimating the generator parameters has
been presented. The proposed method for estimating the generator parameters is
based on the the gradient-based optimization technique. From the results, it can
be seen that the method is capable of accurately estimating the d-axis and q-
axis parameters. The method will be a useful tool for estimating the generator
 312 A. Mohamed et al.
Downloaded by [University of Sydney] at 19:35 06 September 2014

Figure 9a. Graphical user interface for d-axis parameters derivation.

Figure 9b. Graphical user interface for q-axis parameters derivation.

 Estimating the Generator Parameters 313

parameters from the actual measured voltage and ux responses. Graphical user-
friendly interfaces for deriving the generator parameters have also been developed.
With these interfaces, the terminal voltage response after a load rejection test and
the ux response are su cient to derive the parameters of a generator.

References
[1] N. Jaleeli, M. S. Bourawi, and J. H. Fish III, 1986, “ A Quasilinearization Based Algo-
rithm for the IdentiŽcation of Transient and Subtransient Parameters of Synchronous
Machines,” IEEE Trans. on Power System, Vol. PWRS-1, pp. 46–52.
[2] L. X. Le and W. J. Wilson, 1988, “ Synchronous Machine Parameter IdentiŽcation: A
Time Domain Approach,” IEEE Trans. on Energy Conversion, Vol. EC-3, pp. 241–248.
[3] S. Henschel and H. W . Dommel, 1999, “ Noniterative Synchronous Machine Parameter
IdentiŽcation from Frequency Response Tests,” IEEE Trans. on Power System, Vol. 14,
Downloaded by [University of Sydney] at 19:35 06 September 2014

No. 2, pp. 553–560.

[4] M. Shen, V. Venkatasubramaniam, N. Abi-Samra, and D. Sobajic, “ A New Framework
for Estimation of Generator Dynamic Parameters,” IEEE PE-097-PR S ( 08-99) .
[5] IEEE Standard 115A-1987, 1995, “ IEEE Standard Procedures for Obtaining Syn-
chronous Machine Parameters by Standstill Frequency Response Testing,” IEEE Inc.
[6] J. R. Smith, J. F. Hauer, and D. T. Trudnowski, 1993, “ Transfer Function IdentiŽcation
in Power System Application,” IEEE Trans. on Power System, Vol. 8, No. 3, pp. 1282–
1288.
[7] F. P de Mello and J. R. Ribeiro, 1977, “ Derivation of Synchronous Machine Param-
eters from Test,” IEEE Trans. on Power Apparatus and System, Vol. PAS-96, No. 4,
pp. 1211–1218.
[8] L. N. Hannet and J. W . Feltes, 1993, “ Derivation of Generator, Excitation System and
Turbine Governor Parameters from Tests,” Colloquium On Power System Dynamic
Performance, Florianopolis, Brazil, September 22–23.
[9] P. Kundur, 1994, Power System Stability And Control, McGraw Hill, Inc.