SYNCHRONOUS MACHINE PARAMETER IDENTIFICATION

BY A MULTI-TIME SCALE TECHNIQUE.

0. Touhami - H. Guesbaoui - C. Iung

Centre de Recherche en Automatique de Nancy
(CRAN: CNRS URA 821)
ENSEM-INPL 2, avenue de laforst de Haye
545 16 Vandoeuvre-les-Nancy Cedex. FRANCE

-
Abstract In this paper we propose a new method for the It is then possible to find some 'parasitic' parameters (noted q), el,
synchronous machine parameter identification. We use ...) which can be neglected to reduce the system (1).
the multi-time E-ale decomposition to reduce the
operational reactances. The identification method is If, for example, we know by experience that this system has two
based on the frequency responses. The synchronous kind of different dynamics of dimension n and m. respectively noted
machine is a t standstill and supplied by PRBS voltages.
The output data are the currents in the windings. All the x and z ( x*=(x z)T and n+m = p), the choice of the first parameter
Park's model parameters of a n alternator of
13.5kVA/19.5A/50Hz are estimated by an ARX model a c a n be the ratio of thederivatives: 1 I / I l'z I I .
structure.
These variables can be expanded as [ 161:

I. INTRODUCTION
The synchronous machine parameter identification can be gathered where x and i are slow variables and T is a fast variable with the
into three groups of methods. The first ones are based on the theory s a t c h e d time: '0 = ("%)/Eo.
of electromagnetic field [ 11. For an operating point, the magnetic
state of the machine can be computed by a finite-element method There is no explicit expressions of the reduced subsystems obtained
from which the parameters are derived. But such a technique of for non linear systems. In the linear case, from the singularly
calculation is heavy and expensive. perturbed form:
The second kind of methods is based on the two-axis theory. The
time constants of the machine are given from a set of recorded
curves obtained during decay tests. The classical sudden three-phase
short-circuit [2-31 gives no information about the q-axis parameters.
With some other tests made on the field windings [4-61 all the time
constants of the machine can be approximated. These methods are
essentially graphical ones and then inaccurate. For a high number of the m dimension and respectively n dimension reduced fast and slow
damper windings [7-81 they cannot be used successfully. subsystems are:
Nevertheless, the parameter identification can be improved using the
least-square approximate method (91.
Indirects methods using parametric optimization have also been
($= A4.Z + B2.G (4)
used. Among them the adjustment of the armature resistance is the
most widely encountered [lo]. The difficulty of these methods is the
choice of the initial guess. If this choice is not well made a bad = A0.x + B0.u (5)
precision can be obtained -1 -1
A class of stochastic identification schemes based upon frequency with: A0 = A1 - A2.Aq .A3 Bo = B 1 - A2.A4 .B2
responses [ 11-141 gives more precise results. The system is excited
by digital perturbations, as PRBS (Pseudo-Random Binary
Sequences). Such a method had been used for turbogenerators [ 151 and the control vector is written: U = + E('@.
in an on-line identification. The m dimension slow subsystem is an algebrical equation:
In this paper, we present an off-line identification technique with a
machine at standstill. There is then no magnetic coupling between -z = - A &
1 .A3. x - A4-1 .B2.u- (6)
the d and q-axis. A multi-time scale model of the synchronous
machine is used to obtain reduced operational reactances. It yields If now the subsystem (5) has two distinct dynamics xl and x2 with
to separate the model for the various time constants. respective dimensions n l and n2 (x=(xl x2)T and nl+n2 =m), we
can get:

II MODELLING (7)

A. Multi-Time Scale Technique: and write the equation (5):

Consider a p order system written:

x* = h( x*,u) x*(to) = x*o (1)

with a vector control U, UE Rq, where the state vector x* can be

divided into several dynamics.
0-78034634-1/92$03.00 OEEE 178
to obtain now the fast and slow subsystems: Then the expression (2) can be written:

= AOj.x'2 + B2.U
0
(9) Yr 5 'kr

1 0 =
= AO.x i
0-
+ BO.u ( 10) and the equation (4) gives the subtransient behaviour [18]
(EO=l/oo).
With the following expansion of the currents:
with:
0 0
A o = A1-A2.Aq
0 0 - 1 0
.A3 - Bo=
0 0 0 0 - 1 0
B1-A2.A4 .B2 .
id,q = + G,q(zo)
The synchronous machine model has the multi-time scale property.
Then the previous two level simplification can be applied. the equation (6) can be written under the form:

B.Application to the synchronous machine model:

The synchronous machine is supposed to be modelled with one The field flux Yf is slower than the damper winding fluxes YD and
damper winding (rD. XD) for the d-axis and two ones (rQl, X Q ~ ) ,
Y ;then the expression (7) becomes:
(rQ2, X Q ~ ) for the q-axis (2.2 model). The sixth-order elecmcal Q
model ,with reduced variables. is: - -
yf = Ff yD,Q = +D,Q + yD,Q(z 1) (14)

where: Tl=(t -tO)Wd.

- -
And, with vf = vf , the equation (9) gives the transient behaviour
and (10) the steady-state one. It is not necessary to explain here these
where Yr, V r , yd,q, Vd,q are respectively the rotor and the stator expressions.
vector fluxes and voltages The fluxes \y and the currents $ q , can be expanded in such
d*q
vdyf vD,Q)T y D . Q = ( ~ D ' Y Q , ~ Qyd,q'(yd
~ ) ~ Ylq)T transient and steady-state behaviours as:

V r =(Vf 0 0 O)T Vd,q=(Vd Vq)T

YD,Qare the damper fluxes.

The matrix A given by:

A = R.X- 1
-
(the machine is at standstill, the frequency is zero) and: With the s Laplace operator, it is possible to explain these fluxes as:
r
z = WO.' WO= 1oo.x:

The 6x6 matrices of the resistances R and reactances X are:

Rr O
R=( 0 RE) with the reduced operational reactances (211:

with the 4x4 and 2x2 matrices:

Rr = diag(rf '
D rQ1 rQ2) Rs = diag(ra r,)

and:
xmd 1
G ( s )= --
rf ' I + sT'dO

'd.q=(
d'
0 x, ) The reduced field equation is:
- -
- -
-
~

Vf = (rf +S.Xf).i
f + S.Xmd. i d
The system ( 1 1) has the two time scale property: the stator fluxes
are faster than the rotor ones Yr [ 171. The equations (13) are then written:

179
and:
In the same way, for the transient operation, we have :

Thanks to the multi-time scale technique the various time constants that yields to:
can be expressed in term of the physical panuneters of the machine.
The detailled analyticsl expressions had been previously published
[ 19-2 11.

C- Ti.ansferfunctions.
-1 h3+A4 )
X'd = - o O r a ( - + -
All the parameters are determined from the analysis of input-output cL2 113x4
signals. The experimental procedure is :
- the direct and quadrature-axis operational reactances Xd(s), x'd(s) The parameters of subtransient operation can be computed from :
and Xq(s), Xq(s) are given from equations (18a) and (18b) with a
short-circuit on the field winding (vf=O);Vd is a PRBS voltage,
- the field winding parameters can be obtained from equation (17)
with a d-axis armature short-circuit (vd4); vf is a PRBS voltage,
- the direct-axis magnetizing reactance Xmd is then computed from
(16a) and (16c) : . q-axis operational reactances.

The parameters are computed from the quadrature admittances.

(19) Two phases are series-connected, then the stator resistance and the
reactance must be divided by 2.

111. EXPERIMENTAL PROCEDURE.

These definitions (22-231 are the principal ones which power
system analysts have found convenient when describing the
response of synchronous machines. They generally express the A. Power circuit design.
relationship between one machine variable and another.
These methods entail the measurement of operational inductances These operations are applied to the synchronous machine supplied
and srator/rotor transfer functions of a generator over a range of by a power circuit, fig 1 :
discrete frequencies, with the generator rotor at standstill. When
fnx&ency response concepts are envisaged, information is presented
over appropriate range of discrete frequencies. These operational
quantities yield information which enables a chosen equivalent
circuit to be derived.

D- Determination of the synchronous machine parameters.

. d-axis operational reactances.

The identification procedure gives an ARX model which will be
S-Pi
converted in the continuous form . This model will be
(s-hi)(s-hj)
compared to the expressions given in (18) and then the classical
reactances of the synchronous machine will be obtained.
In steady-state operation the equality :

Fig. 1. Power circuit supplying the synchronous machine windings.

180
The Mosfet transistor IRF530 is controlled by PRES through a data (q- is the unit delay operator).
acquisition system with a 70 kHz frequency sampling. A The vector parameter is in this case
microcomputer collects the data currents and voltages. With such
PRBS supply voltages, the process input signals have persistent
excitation properties. Then the estimated parameters of the model e = [a1 a2 ..... ana bo .......bnb]T (26)
converges towards their nomina1 values.
where the coefficients ai (i = 1,..., na) and b. (i = 0..... nb) are to
J
be adjusted. The equation error model (241 has a very important
property that makes it a prime choice in many applications: the
predictor defines a linear ?egression.

IV.MODEL VALIDATION.

v)
In practice, the choice of the model order is as important as the
Pp structure to be taken in system identification. We have validated the
2 2.2 different models successively by the analysis of the output signals
simulated by Promatlab tools, and using the Akaike's Final
2.1 . Prediction Error formulated as :

2, -
FPE = l+PM . 1
1-P/N
0 20 40 60 80 100 120 140 160 180 ZOO
time (ms)
where J is the criterion function to be minimised and P is the
dimension of the parameter vector.
The criterion described by Akaike reflects the prediction-error
variance that will be obtained, on the average, when the model is
Fig.2 Pseudo Random Binary Sequence voltages. applied as a predictor to other data sets than those used for the
identification.
For a model structure that is parametrized i n terms of physical
B. Data acquisition and aploiiation. parameters, an important validation in to confront the estimated
values and their estimated variances. This is reasonable from our
knowledge of the process.
To estimate the parameters in the d and q-axis of the synchronous An identification gives, on one hand, a validation of the
machine, we use a sampling period of T=lms and an observation mathematical model and, on the other hand, an experimental model
time of 1 second and then N=1000data points are collected. from which we will be able to estimate synchronous machine
The input-output signals are filtered to reduce the perturbation parameters.
effects. We use second order Tchebyscheff band-pass filters giving A commonly applied procedure that can be considered a test of the
informations to the estimator only for an interesting band of model validity for simulation is to simulate the system with the
frequency which can be the transient or steady-state behaviour. actual input and compare measured output with the simulated model
These filters have the straigther cutting of all the numerical filters, output.
but their transient behaviour fluctuate. In our test, the model is then evaluated by comparison between the
real and simulated output data, either visually in a plot or by some
formal distance measure. A data set different from those used for
C'. The model structure.
the identification of 8 was used for this comparison (cross-
The ARX (Auto Regressive external input) idendticarion process is validation).
schematized in fig.3 : The same data gave the comparison between the measured output
and the model output, for a d-axis model of a synchronous machine,
is given in figure 4 on steady-state operation and figure 5 on
transient operation. In this test, the Akaike's coefficients are very
small (510-4 1.
The field winding model is represented by figure 6. The real output
data coincides with the simulated output data. The Akaike's
coefficient values are neglectable.
The figures 7 and 8 give the results for the q-axis model respectively
ut0 w(t) YO) for the steady-state and the transient behaviour. The Akaike's
coeficient values are of order IOe6 .

Fig.3 The A R X model structure.

V. TRANSFORMATION OF DISCRETE ADMITTANCES IN
which has a single input and single output; ~ ( t is) the PRES voltage CONTINUOUS ADM ITTANCES.
supply, y(t) is the current and e(t), a white noise.
The (na , nb) order process is defined as follow : To extract the physical parameters of the machine from the discrete
type model, we transform it into a continuous one. The method used
is the following one. At first, we consider a particular input Hij and
A ( q ' l ) = 1 + a 1 q - l +........... + a n a q - n a Yi ( 2 )
a temporal test to estimate the coefficients of -
ui ( z ) l with :
B(q-') = b o + b1q-l + ....,.. + b n b q s n b

181
 alzn-l+...... +an-lz+an
Hij(z) =
z"+b 12"' + ....... +bn z = exp(sT) (28)
01

The solution for this problem is a bilinear transformation, which is 0 05

difficult to be computed. Then, we have used a mixed method [25-
261 to get H(s) from Hfz) . 0
Let a state discrete modelkilization in the form :
s
h

9 -00s

-0 I

4 15

-0.2 1 I
where x is a State vel x,F is a Companion matrix, observable by 20 40 60 80 100 im
construction:
time (ms)
Fig.4 D-axis. Real output and simulated model output
in steady operation.

0 15, 1

01
and G,Care the following matrices :
005
G = (0.....01)T and C = (an ....... (29~)
0

The discrete system is equivalent to the continuous system written : 3

.z -005

x = A.x + B.u 41
(30)
y = c.x I

when A and B are given by : "._0 m 40 60 80 100 im

time (ms)
F = exp(AT)
Fig.5 D-axis. Real output and simulated model output
in aansient operation.

0.15
I
The matrices A,B and C are a minimal realization. We have used an II
algorithm (271 to compute the transfer functions by :

k
n(s-Pj, $ 0
j= I 5
G i j ( ~=
) K I kc I (32)
-0 os

.O.f

This algorithm eliminates the parts of no-controllability and 4.d 20

1
inobservability of each triplet (A,B,C) and then the system (30) is 40 60 80 100 I20
observable and controllable. time (ms)
Fig.6 D-axis. Real output and simulated model output
in field winding.
182

~
 The parameters r a , XQ are very small to be identified.
01

0 05
VI. CONCLUSION.

Our identification process consists in finding a reasonable model

sa n structure based on an analytical input-output representation.
A multi-time scale model of the synchronous machine had been used
n os to obtain reduced operational reactances.
This technique allows us to connect the identified parameters to the
01 physical ones. We have then used a separated identification for the
different time constants.
We have organized the identification as follow :
-01 s t
20 40 60 80 100
I
120
- data handling, filtering and plotting,
- p h e u i c identification methods by an ARX model structure.
time (ms) - presenting model properties, comparison of the simulated curves
Fig.7 Q-axis. Real output and simulated model output to the real ones,
- validation procedures.
in steady operation. This method of tests had been used sucessfully on a synchronous
n 15, machine at standstill and gave all the Parks model parameters. Then
1 pseudorandom binary sequences can be used for the supply of
i 1 I indusaial power circuits.
We notice that with a supplementary degree of amplification
between the power and the control circuit, it can be possible to
realize a parameter identification of large power synchronous
0051
machine as turbogenerators.

List of Symbols.

derivative and transpose of x

stator and field resistances
direct-axis damper resistance
quadrature-axis damper resistances
I
-0.151
20 40 60 80 100 im direct-axis transient open circuit and short-circuit
time (ms) time constant
direct-axis subtransient open circuit and short-
Fig.8 Q-axis. Real output and simulated model output circuit time constant
quadrature-axis transient open circuit and short-
in transient operation.
circuit time constant
. Results.
quadrature-axis subtransient open circuit and short-
circuit time constant
direct and quadrature-axis synchronous reactance
The identified parameters are :
field leakage reactance
direct-axis transient and subtransient reactances
quadrature-axis transient and su btransient reactances
direct-axis damper reactance
quadrature-axis damper leakage reactances
direct and quadrature-axis magnetizing reactances

REFERENCES.

[ l ] S.H. MINNICH. M.V.K. CHAR1 and J.F. BERKERY.

Operational inductances of turbine generators by the finite-element
method, IEEE trans on P A S , vol.PAS-102, n o I, pp 20-27, 1983.
[2] ANSI/IEEE standard: Test procedures for synchronous
machines. IEEE Guide,115-1983.
[ 31 GSHASCKSHAFT. New approach to the determination of
synchronous machine pardmeters from tests. Proc IEE. vol-121,
n ' l l . pp 1385-1392, 1974.
[4] H. KAMINOSONO and K. UYEDA. New measurement of
synchronous machine quantities. IEEE Trans on PAS, vo1.PAS-87,
n 7 1 , pp 1908-1912 , 1968.
183
151 F. DE MELLO and L. HANNETY. Determination of [Z]B.C. MOORE.Principal component analysis in linear systems:
synchronous machine electrical characteristics by tests. IEEE Trans controllability observability and model reduction. IEEE Trans on
on PAS. vol.Pa~-102, n"l2, pp 3810-3815, 1983. AC. volAC-26, n o l , pp 17-31, 1985.
161 M.E COULTES and W. WATSON. Synchronous machine [26] S.P. BINGULAC. On the calculation of the tranfer function
models by standstill frequency response tests. IEEE Trans. on PAS, mamx. IEEE Trans on AC, vo1.AC-22, pp 134-135, 1985.
vol.PAS-100, n"4, pp 1480-1489, 1981. [27] A.VARGA. Numerical stable algorithm for transfer function
[7] Y.TAKEDA and B. ADKINS. Determination of synchronous- mamx evaluation. Int. J. of Conrrol , vo1.33, n06 , pp 1123-1133,
machine parameters allowing for unequal mutual inductances. Proc 1981.
IEE, vol.I21, n"12, pp 1501-1504, 1974.
[8] F.L ALVARADO and C. CANIZkRES. Synchronous machine
parameters from sudden-short circuit tests by back-solving. IEEE
Trans on EC, vol.EC-4(2), pp 224-236, 1989.
[9] M.RAFIAN and M.A LAUGHTON. Determination of
synchronous-machine phase coordinate parameters. Proc.lEE,
~01.123,n"8, pp 818-824, 1976.
[IO] ZHOU DONGSLNG; XI. WONGFU and C. XIANFENG.
The direct fitting method for measuring the parameters of
synchronous machine using direct current decay test. E.M.P.S.,
pp.239-248, 1989.
[ 111 S.D. UMANS; J.A. MALLICK and G.L. WILSON.Modelling
of solid rotor turbogenerators equivalent circuit from standstill
frequency measurements. IEEE Trans on PAS, vol.PAS-97, pp
269-291 , 1978.
[121 1. KAMWA; P. VIAROUGE and E.J. DICKINSON.
Experimental modeling of a synchronous machine-chopper system
using standstill normal operating records. IMACS- TCl'90, Nancy,
pp 247-253, 1990.
[ 131 PL. DANDENO and A.T. PORAY . Development of detailled
turbogenerator equivalent circuit from standstill frequency response
measurement. IEEE Trans on PAS, vol.PAS-100, n04, pp 1646-
1654, 1981.
[ 141 E.EITELBERG and R.G. HARTLEY. Estimating
synchronous machine electrical parameters from frequency response
tests. IEEE Trans EC, vol.EC-2, n o l , pp 132-138, 1987.
[I51 D.J.G MORELL and BW HOGG. Identification and
validation of turbogenerator models. Auromarica , vol-26, n "1, pp
135- 136, 1990.
1161 V.R.SASKENA ; J. OREILLY and P.V. KOKOTOVIC.
Singular perturbations and time scale methods in control theory.
Survey: 1976-1983.Automarica , vol-20, n"3, pp 273-293 , 1984.
[17] P.C.KRAUSE and al. The theory of neglecting stator
transients. IEEE Trans on PAS. vol.PAS-98. pp 141- 148, 1979.
(181 P. AHMED-ZAID ; P.W. SAUER ; M.A. PA1 and M.K.
SARIOGLU. Reduced order modeling of synchronous machine
using perturbations. IEEE Trans on C S , vol.CA.7-29, noli, pp
782-786, 1982.
[ 191 H. GUESBAOUI and C. IUNG. Simplification of the model
of a current-fed controlled synchronous machine by the multi-time
scale method. Modelling and Simulation of Systems- JC BUZZER
AG, Scientific Publishing Co. 0 IMACS, pp 329-331, 1989.
[ZO]H. GUESBAOUI, 0 .TOUHAMI, C. IUNG. Multi-time scale
decomposition of the synchronous machine model. 131h IMACS
World Congress, Dublin, pp 1566-1568. July 1991.
[21] H. GUESBAOUI. Contribution a la mise sous forme standard
des systemes a multiechelles de temps. Application a la modelisation
des machines a courant alternatif. These de doctorat es-sciences,
INPL-Nancy, 1992.
[22] IEEE TASK FORCE ON DEFINITION (PL.DANDEN0,
chairman). Supplementary definitions and associated test methods
for obtaining parameters for synchronous machine stability study
for simulations. IEEE Trans on PAS, vol.PAS-99. pp 1625- 1633 ,
1980.
[23] IEEE TASK FORCE ON DEFINITION (PL.DANDEN0,
chaiman). Current usage & suggested practices in power system
stability simulations for synchronous machines. lEEE Truns on EC,
vol.EC-l(l),pp 77-93, 1986.
[24] L. LJUNG. System identification. Theory for the user.
Prentice-Hall, Inc., Englewood Cliffs, N.J . , 1987.

184

--I--