Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
net/publication/3360417
Article in Generation, Transmission and Distribution [see also IEE Proceedings-Generation, Transmission and Distribution], IEE Proceedings C · December 1990
DOI: 10.1049/ip-c.1990.0060 · Source: IEEE Xplore
CITATIONS READS
15 48
3 authors, including:
Some of the authors of this publication are also working on these related projects:
Design of a high voltage pulse transformer. Application to the klystron modulators for the compact linear collider (CLIC) View project
All content following this page was uploaded by Innocent Kamwa on 14 May 2014.
I. Karnwa, PhD
P. Viarouge
J. Dickinson, PhD
and
U
I ... I The voltage values can be derived from eqns. l a and lb:
b
1 dY
Fig. 1 Generalised equivalent circuits of the synchronous machine ud = r a i d +-
wn dt
- O,Y,
(a) d-axis: e, = w , Y q
(b) q-axis: e, = o,Y,
1 dY
vq=raiq+--9+comyld
on dt
mise between older models and present requirements for
increasing accuracy and realism in describing complex 1 dYf
rotor machines [ 2 ] .
vf = rfif +-
on dt
~
with
X = [ xd
anq+l,nd+Z
Ond+2.nq+1
instant
Y = Y' + Y K - 1 (64
Xd= may be applied, knowing that id(O-) = i,(O-) = 0.
Direct identification of the turboalternator consists
essentially of being able to construct a prediction model
(eqn. 6) of the short-circuit of the machine, which is
xqq xqQi x4Q2
... optimal in the sense of minimising the quadratic predic-
... tion error, based on certain equivalent circuits which a
'Qiq XQiQi 'QiQz
x, = xQ2q xQ:Qi xQ2Q2
... priori are unknown. The problem of optimally estimating
the latter (in the short-circuit test sense), from p observa-
... tions of the current vector, can be formulated as follows
LXQ.,q XQn,Qi XQ.,Q2 XQ.,Q., J ~171.
Gd = CXdd xd f X d D l .' ' xdDdl ' q = Cxqq XqQi ' ' ' XqQnql Let
(44 x T = [ r a r f rD, . . . I
rDOd r Q I . IQ", I
O i j is the null matrix with dimensions i x j . These Xf X D , ". xD,, IX Q ~ .. XQ,, I xCI
electromechanical transients with the generator con- e(k) = y(k) - j ( k ) q = 3nd + 2n, +5
nected to an infinite grid through a transmission line, and
the modelling of small perturbations of the state variables w = diag (wd wf w,) (8b)
under the same conditions, are good examples for these k is the discrete time index and j ( k ) is the measure of y at
kind of problems. time k. Consequently, e(k) represents a prediction error
vector. An optimal estimation of the equivalent circuit
3 Direct estimation of the equivalent circuits parameters consists of determining the q-dimensional
vector ?. which is the solution of the following nonlinear
Faced with the difficulties encountered in inferring the optimisation problem:
d-axis circuit from the given operational impedances [ 13,
251, the question arises as to whether it would be possible min f (x)
to determine the equivalent circuits directly from the
experimental data. Electrotechnical tradition generally xmin(i)< x(i) < xm.,(i) i = 1, . . . , q (9)
attempts first to characterise the synchronous machine by where x,, and xminrepresent bounds fixed by experience
its time constants, and only then to deduce the circuits. It and common sense.
would, however, seem just as reasonable to determine the The choice of weighting factors wi can be quite arbi-
circuits first and subsequently calculate the operational trary, but we have found in practice that choosing wd =
impedances. This inverse procedure allows, the various wx = w,/2 gives good results. The detailed numerical sol-
approximations required to construct the algorithms ution of the above nonlinear least-squares optimisation
published in Reference 13 to be avoided. In the present problem (eqn. 9) involves a Newton-Raphson procedure
section we will present a nonlinear programming formu- well documented in the mathematical programming [18,
1EE PROCEEDINGS, Vol. 137, Pt. C , N o . 6 , N O V E M B E R 1990 447
191 and system identification [17, 221 literature. The original machine (in Table 3). Although the solution
present authors have had recourse to a code adapted differs (mainly in the q-axis) from the true solution x*, the
from the NL2SOL program described in Reference 19. short-circuit data associated with 2 and x* are not very
The program, chiefly conceived to determine the vector different.
of parameters providing the best least squares fitting of
Table 1 : Direct estimates of the equivalent circuits: Model
experimental data, has proved particularly efficient over a
2.2
wide range of applications. It seems to be one of the best
programs available for the solution of nonlinear least
squares problems in general, and more specifically, those ra 0.0033000 0.0030000 1.36 0.00
involving, at the same time, mediocre initial conditions r, 0.0008361 0.0009290 9.44 -0.00
and an eventual nonzero sum of residues at the solution rD1 0.0146740 0.013361 6 0.783 -0.1 6
point [1 71. ra, 0.001 6020 0.0015255 2.05 14.30
ro2 0.0092510 0.0085698 0.260 -1.90
x, 0.127260 0.141 388 0.0686 0.01
3.2 Numerical validation xD, 0.897750 0.0812405 0.106 0.01
Using a slight modification of the NL2SOL code [19], xo, 0.731 250 0.7048457 0.01 1 1 13.25
xaz 0.103290 0.0947551 0.1 13 -0.91
the proposed method will now be illustrated by means of x ~ ,0.0000010
~ 0.0000045 651 0 0.00
the simulated identification of two turboalternators with xmd 1.44900 1.608191 7 0.00143 0.112
solid rotors. xmQ 1.77100 1.8959851 0.00164 17.76
(&) f(x,) = 0.349; f(R) = 0.316 x 10-5
Example 1 : nonlinear estimation of the equivalent circuits
of a machine with rotor 2.2 We would like to stress here that the particular solu-
The first machine studied was a steam turbine generator tion R obtained in this example depended on the short-
described by Krause in Reference 20 (page 240). Its circuit data available, and could have been substantially
experimental model provided by this author involves a improved with an increased number of data points and a
rotor (2.2) with two equivalent windings in each axis, i.e. best sampling period.
one damper in the direct axis and two in the quadrature
axis. We have used this a priori model, termed the true Example 2: Nonlinear estimation of the equivalent circuits
equivalent circuits in the sequel (x*), to calculate a of a machine with rotor 3.3
sudden short-circuit solution of the generator. The simu- The preceding example is not really complicated enough
lated oscillographs were covered by p = 1500 points with to illustrate the efficiency of the method. Therefore, we
a sampling rate of T = 0.8 ms and o,(O-) = 0.1 p.u. To analysed another alternator requiring a more complex
validate the applicability of our estimation scheme, an model. This time, however, we chose richer test condi-
attempt is then made to reassemble the equivalent circuit tions, both from the point of view of sampling rate and
which produced the simulated responses; the latter being number of data points. Let us consider the Nanticoke
considered to have been measured in a noise-free machine [14], modelled with nd = 2 and n, = 3 dampers
environment for the purpose of our demonstration. In the in the d and q axes, respectively. This type of equivalent
analysis, we assumed that x , and 8, were known, which circuit realisation represents the level of model soph-
seems reasonably realistic. In addition, we took istication widely accepted for turbogenerator and solid-
w = diag (1, 1, 2) to achieve an adequate weighting of the pole machines [2, 71.
three currents. For this example, the initial estimates xo The well established model 3.3 of the Nanticoke
were obtained by forcing a mismatch of f 10% on all the turboalternator used in the sequel, has three windings in
exact parameters defined by x*. each axis and was derived by Dandeno and Poray [14]
The results of the estimation are summarised in Table from experimental standstill frequency response tests.
1. The last column clearly shows that the final circuit is Once again, we numerically simulated a short-circuit test,
very different from the true one. Nevertheless, with basing our calculations on the parameters given in Refer-
respect to the initial circuit, there is a considerable ence 14 and with the following: T = 0.6 ms, v,(O-) =
improvement; the residual error being reduced by a 0.1 pu and p = 2020. The question is whether it is pos-
factor of 100,OOO to 1 (passing from 0.349 to sible to recover directly the equivalent circuits from this
0.316 x lo-'). In fact, plotting the measured responses 'test' data. The proposed formulation led to a nonlinear
* A
i d , .-,L 2, and the responses i d , i,, i , , calculated from the least squares' problem whose solution is summarised in
solution 2, it is virtually impossible to discern any differ- Table 4. The initial estimate xo was obtained by per-
ence. To quantify this statement, we calculated the ana- turbing the true solution x* by _+ 15%.
lytical values of the short circuit for x = 2 (see Table 2). As the last column indicates, the simulation was gener-
The reader can compare these values with those of the ally most successful. Apart from x p l and x,,, which
i F,U) '
i
d rlf r/0
1 exp (-4.446t)cos (376.89t) 0.4163 -0.1481 -3.781x 1O-j
2 exp (-4.446t)sin (376.89t) -4.032x 1 0-3 -0.0128 -0.3704
3 exp (-32.431f) -0.0949 -0.1379 1.520x 1 0-3
4 exp (-13.616f) -5.250x 7.832x 6.351x
5 exp (-1.101t) -0.2660 0.2859 -2.083x
6 exp (-0.678t) 1.263x -1.540x 1.126x
7 exp (0.000f) -0.0556 4.994x 10-9 -9.259x 10-5
9
IC) i,(t) = x r l lF,(f), I = d , f, 9
,=1
Table 4: Direct estimates of equivalent circuits: Nanticoke tests is a badly conditioned problem. Nevertheless, the
3.3
resulting solution remains the best source of information
xb"' p' Vf(2) Error, % that it is possible to extract from short-circuit testing, if
r, 0.0046000 0.0039995 13.6 0.01 one adopts the least squares based formulation as pro-
rt 0.000799000 0.0009386 75.1 0.14 posed in this paper. Because of this we will call it the
rD, 0.131330 0.1132936 0.741 0.79 suboptimal solution. In fact, one can demonstrate that,
rD, 0.00503200 0.0058192 19.9 1.70 under identical experimental conditions, the estimation of
ro, 0.00618700 0.0055384 4.12 -2.94 the operational impedances leads asymptotically (p + 00)
ro2 0.0918850 0.1082664 1.09 -0.15
ro3 0.0216200 0.0182397 5.33 2.98 to the true solution of the problem (eqn. 9), whereas the
xf 0.0131 750 0.0151 821 0.369 2.05 circuit method leads to biased results.
xD, 3.14180 2.7017588 0.0226 1.1 0 All the above deficiencies are related to the well
xD, 0.00640050 0.0073506 5.42 2.38 known impossibility of recovering a unique solution for
xo, 1.90555 3.5151 30 0.0358 -1 12.15
0.1193854 0.555 -0.07 the equivalent circuits from a single test at constant field
xoI 0.101405
xo3 0.518995 0.4421580 0.0267 2.03 voltage, with the only perturbation coming from the
xkt, -0.599725 -0.5166927 0.181 0.92 stator terminals. This fact can be demonstrated by the
xkt2 1.03213 0.8922227 0.122 0.59 machine equations in operational form [12, 161 as
xmd 1.82920 2.1764536 0.0283 -1.14 follows:
Xm, 2.36555 1.3025091 0.0443 36.68
(Ir) f(x,) f(R) = 0.835x
= 0.276; 1 O-4. Era + Pxd@)lid - O r xq(P)iq = ud - PG@)uf ( 104
x h ) i d + Era + PX&)liq = uq - o r %)vf (lob)
mainly determine the late transient and the steady state with or= o m o nx&), . G(p) and x,(p) are three rational
regime of i,, all the other parameters were estimated with fractions defined by
reasonable accuracy. The reader is referred to Table 5
which illustrates the analytical representation of the pre-
vious short-circuit test. It can be noted that the mode
- 1.114 ( j = 7), which represents the q-axis transient, is
the least vigorous mode, as far as the three currents, i d , if
and i , , are concerned. The fact that the effects of this
mode are too small to be experimentally detectable
renders estimation by short-circuit testing the most diff-
cult. The model is not sufficiently parsimonious, at least
along the q-axis. In fact, it is possible to build a reduced
order model for this machine, with only two dampers in x (1 +pT;)(l +pT,2).-(1 + p T ? + ' )
x,(p) = =9
the q-axis, which is relatively satisfactory if only the accu- Xqd Wn(1 +pT:o)(1 + p T & ) . . . ( 1+ p T : z + ' )
rate prediction of the short-circuit is considered.
(1 1 4
3.3 Discussion of some deficiencies in the They represent a generalisation of the so-called standard
short-circuit testing operational impedances [161. During the short-circuit
The first example of Section 3.2 demonstrates that the operation, the synchronous machine looks like a SIMO
estimation of the equivalent circuits from short-circuit system (single input multiple outputs) with three outputs
Table 5: Analytical expression for the no-load short-circuit: Nanticoke 3.3
i F,(O '
i
d r/t r10
1 exp (-5.058t)cos (376.40f) 0.314 -0.08149 -0.06657
2 exp (-5.058t)sin (376.40t) -3.427x -0.03219 0.3585
3 exp ( - 180.454f) -7.140x -2.754.x 0.06514
4 exp (-113.875t) -3.239x 1 0-3 -0.11 1 7 6.848x
5 exp (-20.275t) -0.1133 0.04857 -1.688x lo-'
6 exp (-11.791t ) -6.858x 1 0-6 1.441x 1 0 - 5 -3.658x 1 0-3
7 exp (-1.114f) 8.930x 1 0 - 6 -6.457x lo-' -5.390x 1 0-4
8 exp (-0.557t) -0.1542 0.1449 7.089x 1 0 - 5
9 exp (0.000t) -0.04303 5.978x 1 O-' -7.718 x 1 0 - 5
9
(&) i,(t) = 1rI, F,(f), I = d , f, 9.
1=1
prove that a short-circuit test can bring out the exact 2 CANAY, I.M.: ‘Physical significance of sub-subtransient quantities
q-axis equivalent circuit. We only assert that, with in dynamic behavior of synchronous machines’, IEE Proc. E, 1988,
models that involve low numbers of q-axis dampers (n,), 136, (6),pp. 334340
this test can be very effective in providing adequate 3 MAUN, J.C.: ‘New procedure for the identification of dynamic
parameters to this usually weakly defined component. parameters of the synchronous machine by the three-phase short-
circuit test and extension to others tests’, ICEM’84 Proceedings, pp.
It seems worthwhile to stress that the methodology we 1.2.1-1.2.8
advocate in this paper can be extended to cope with data 4 SALVATORE, L., and SAVINO, M.: ‘Experimental determination
collected from a loaded machine. Among the suggested of synchronous machines parameters’, IEE Proc. E, 1981, 128, (4),
tests, we can mentioned a short-circuit test on a loaded pp. 212-218
5 CANAY, I.M.: ‘Determination of model parameters of synchronous
machine during which an excitation control system drives machines’, IEE Proc. E, 1983, 130, (2), pp. 86-94
the field voltage to ceiling or simply removes the excita- 6 IEEE Task force on definition (DANDENO, P.L., Chairman): ‘Sup-
tion. However, it is our opinion that this will require the plementary definitions and associated test methods for obtaining
addition of a mechanical equation to the electrical state parameters for synchronous machine stability study for simulations’,
equation provided in this paper (eqns. 5), since the rotor IEEE Trans., 1980, PAS-99, pp. 16251633
7 IEEE Task force on definition (DANDENO, P.L., Chairman):
speed would probably vary more substantially during the ‘Current usage and suggested practices in power system stability
on-line data logging process. An additional item needed simulations for synchronous machines’, IEEE Trans., 1986, EC-1,
in this context is the no load magnetisation curve. This (l), pp. 77-93
curve would be necessary in accounting for saturation at 8 ANSIjlEEE Standard 115-1983: ‘IEEE Guide: Test procedures for
synchronous machines’
each time step of a Runge-Kutta routine, when solving 9 DANDENO, P.L., KUNDUR, P., PORAY, A.T., and ZEINEL-
the model for the predicted currents [lS]. Since the DIN, H.M.: ‘Adaptation and validation of turbogenerator model
equivalent circuits derived by such a procedure are only parameters through on-line frequency response measurements’,
valid at the given operating point with no means to IEEE Trans., 1981, PAS-100, pp. 1656-1665
extrapolate the results to another one, we suspect that IO. SCHULTZ, R.P., JONES, W.D., and EWART, D.N.: ‘Dynamic
models of turbine generators derived from solid rotor equivalent cir-
the cost of this increased complexity is greater than the cuits’, IEEE Trans., 1973, PAS-92, pp. 926933
benefits that result from the use of more realistic data. 1 1 KAMWA, I., VIAROUGE, P., and DICKINSON, J.: ‘Discrete
technique for synchronous machine operational impedances estima-
tion from time-domain standstill tests’, in ROBERT, J., and
5 Conclusions MIDVIDY, W. (Eds.): ‘IMACS Annals on Computing and Applied
Mathematics’, 6 (J. C. Baltzer AG, Scientific Publishing Co., Basel,
We have presented an efficient nonlinear least squares- 1989), pp. 101-104
based procedure for estimating the equivalent circuits of 12 SRIHARAN, S., and HIONG, K.W.: ‘Synchronous machine model-
synchronous machines directly from short-circuit data. ling by standstill frequency
. - resuonse
. tests’, IEEE Trans., 1987,
Since the latter are generalised to include an arbitrary EC-2, i2), pp. 239-245
13 UMANS., S.D... MALLICK. J.A.. and WILSON. G.L.: ‘Modelline. of
number of dampers in each axis, our method seems the solid rotor turbogenerators’ (in two parts), i E E E Trans., 1678,
only feasible approach for analysing short-circuit oscillo- PAS-97, pp. 269-29 1
graphs when the machine is a solid rotor turbogenerator 14 DANDENO, P.L., and PORAY, A.T.: ‘Development of detailed
[2, 251. The proposed one-step method for identifying turbogenerator equivalent circuits from standstill frequency
equivalent circuits arose from the recognition of the need responses measurements’, IEEE Trans., 1981, PAS-100, pp. 1 6 2 5
1633
to remove the various approximations usually employed 15 HARLEY, R.G., LIMEBEER, D.J.N., and CHIRRICOZZI, E.:
to calculate the appropriate parameters from dynamic ‘Comparative study of saturation methods in synchronous machine
constants. models’, IEE Proc. B, 1980, 127, (I), pp. 1-6
Unfortunately, the standard short-circuit test leads to 16 CONCORDIA, C.: ‘Synchronous machines’ (Wiley, New York,
1951)
poor observability of the q-axis eigenstructure, which 17 LJUNG, L.: ‘System identification - theory for the user’ (Prentice-
consequently gives rise to a weakly identifiable circuit Hall, Englewood Cliffs, NJ, 1987)
model, at least for a high numper of q-axis dampers. 18 BETTS, J.T.: ‘Solving the nonlinear least squares problem: applica-
Concerning a more critical issue,! this paper claims that, tion of a general method’, J . of Optimization Theory and App., 1976,
whatever the accuracy of our es imation algorithm, the 18, (4). pp. 469482
19 DENIS, J.E., GAY, D.M., and WELSH, R.E.: ‘An adaptive non-
no load short-circuit test is, by i!iself, insufficiently infor- linear least-squares algorithm’, A C M Trans. on Math. Software,
mative for the development of a machine model applic- 1981,7, (3), pp. 348-368
able to all operating conditions. This results from the fact 20 KRAUSE, P.C.: ‘Analysis of electric machinery’ (McGraw-Hill, New
that it does not involve any perturbation of the field York, 1986)
21 JACK, A.G., MENG, L.G., and BEDFORD, T.J.: ‘Frequency
voltage. In general, there is no guarantee that equivalent domain-based methods to predict turbogenerator transients with
circuits obtained by this approach will accurately model particular emphasis on field current’, IEE Proc. C, 1989, 136, (3), p.
the actual field input admittance, which means that they 2 0 6 2 14