Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
1, JANUARY 2004
VH,1 H L
Transf.
1 1
VLH,31 = VL,3 / VH,1 2 2
3 3 VL,3
Jumpers
to ground Attenuator
6
5
4
3 Fig. 4. Measuring Y .
2 Current
sensor
uration used for measuring an off-diagonal element . The
1 A
following components were used:
1) network analyzer: HP3577A;
2) attenuator: HP8491A, 30 dB (a special attenuator with
low-pass characteristic was used when connecting the
source to the low-voltage winding);
Fig. 3. Measuring Y . 3) current sensor: Pearson model 2100, 1 V/1 A.
All measurements were done using the following settings of
where and denote the high-voltage and low-voltage wind- the network analyzer: logarithmically distributed samples be-
ings. and are matrices of size 3 by 3. tween 50 Hz and 2 MHz, 100-s sweep time, 10-Hz resolution
A useful test is to compare the calculated and with bandwidth. After each sweep, the measured results were down-
a direct measurement of these quantities. The measurement pro- loaded to a PC for further processing.
cedure is illustrated in Fig. 2.
B. Correction for Measurement Cables
III. MEASUREMENT SETUP The dominant effect of the measurement cables is their capac-
itance to ground because the high impedance of the transformer
A. Instrumentation prevents any significant voltage drop to develop along the ca-
A measurement setup has been designed for the purpose of bles. It follows that each cable results in that the corresponding
measuring the admittance matrix as well as voltage ratios diagonal element is modified by an added contribution
(e.g., and ). The setup consists of a network analyzer, with being the cable capacitance. Thus, the major effect of
a connection board with a built-in current sensor (Fig. 3), and the cables can be removed by correcting the measured using
shielded cables which connect the transformer terminals with the expression
the connection board. The cable shields are grounded at both
ends; at the transformer and at the connection board. All re- (6)
connections needed for measuring different matrix elements are
where is a diagonal matrix with entries being the capacitance
done at the connection board using jumper connections, thus
of the respective cables. The short stub cable “A” in Fig. 3 is
giving reproducible results. The cables between the network an-
taken into account by adding its capacitance to all diagonal ele-
alyzer and the connection board are made as short as possible.
ments of .
The built-in calibration feature of the network analyzer is used
for taking into account the frequency dependency of the current
sensor, voltage probes, and the attenuator. IV. ACCURACY CONSIDERATIONS
Fig. 3 shows the connection board as configured for the mea- It follows from (3) that an open-circuit condition is equiva-
surement of a diagonal element . Fig. 4 shows the config- lent to solving an equation with and being sub-
416 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004
10
the elements are strongly frequency dependent and that some
elements are much smaller than others. The large difference in
element sizes is mainly a result of the high voltage ratio of the 0
10
transformer (25:1).
Off diagonal elements
Fig. 7 compares the measured elements of with the cor- -1
10
responding elements calculated by (5). It is seen that the voltage
transfer from high voltage to low voltage is at 200 kHz, about
ten times higher than the voltage transfer at 50 Hz. The calcu- -2
10
2 3 4 5 6
lated ratio is seen to closely match the measured ratio. 10 10 10 10 10
Frequency [Hz]
A similar observation is made for the elements of when
comparing measured and calculated quantities, see Fig. 8. Two
Fig. 8. Measured and calculated elements of V .
distinct peaks of the transferred voltage can be observed around
15 kHz.
We next look at a situation with open terminals at both the calculated elements for the voltage ratio as calculated by (3). It is
HV and LV side, see Fig. 9. Fig. 10 compares the measured and seen that a large deviation results at frequencies below 10 kHz.
GUSTAVSEN: WIDE BAND MODELING OF POWER TRANSFORMERS 417
3
10
Measured
Calculated
2 1
10 2
3
Voltage ratio
1
10
Fig. 9. Open terminals at both windings.
0
10
5
–1
10
–2
10
2 3 4 5 6
10 10 10 10 10
Frequency [Hz]
Fig. 12. Calculated versus measured voltage ratio after introducing measured
voltage ratios V and V in the admittance matrix Y .
TABLE I
TRANSFORMER DATA AT 50 Hz, REF. HIGH VOLTAGE SIDE
3
This situation can be modeled by making the following substi-
10
tutions:
2 (7)
10
(8)
1
(9)
10 a)
(10)
b)
0
10 The modified gets a voltage ratio from high to low and
2 3 4 5 6
10 10 10 10 10 from low to high, which equals the measured voltage ratios.
Frequency [Hz]
It was also found that this substitution greatly improves the
Fig. 11. Condition number for matrix subject to inversion. (a) Y , (b) Y , accuracy for the terminal condition in Fig. 9 (see Fig. 12). Com-
(c) Y for situation in Fig. 9. parison with Fig. 10 shows that the error has been strongly
reduced.
The accuracy at 50 Hz can be assessed by comparison with
C. Analysis
nominal data. Table I shows results for the positive sequence
The above results regarding accuracy can be understood by system as calculated from , before and after making changes
considering the condition number of the submatrix which is sub- (7)–(10). Both measurements give a short circuit impedance
ject to inversion, as explained in Section IV. Fig. 11 compares which agrees well with the nominal one. However, the open cir-
the condition number for the terminal situation in Figs. 7, 8, cuit impedance using the original is much too low. Normally,
and 10. It can be seen that for the situation in Fig. 10 (trace “c”), the magnetizing current is much smaller than the nominal load
is large, particularly at low frequencies which is where the de- current. 566 gives a current (10.2 A) which is higher than the
viation between the measured and calculated voltage ratio is the nominal current (1.73 A). With the modified , the current is re-
highest. The small for the situation in Figs. 7 and 8 (traces duced to 0.49 A. When measuring the admittance with open ter-
“b,” “a”) explains the good agreement between the measured minals at the LV-side, a value of 22 002 was achieved, which
and calculated voltage ratio. gives a current of 0.27 A.
418 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 19, NO. 1, JANUARY 2004
5
120 10
1m cables Original voltage ratio
5m cables Reduced voltage ratio
5m cables, corrected
100 4
10
c)
Voltage ratio
80
Condition number
3
10
60
c)
2
10
40
1
10
20
a) , b)
0
0 10
4 5 2 3 4 5 6
10 10 10 10 10 10 10
Frequency [Hz] Frequency [Hz]
Fig. 13. Calculated voltage ratio. Mitigating the effect of the measuring cables. Fig. 14. Effect of reduced voltage ratio on condition number for matrix subject
to inversion. (a) Y , (b) Y , and (c) Y for situation in Fig. 9.
1
four approximations were finally combined into a single com- 10
Magnitude [S]
–2
10
(13)
–3
10
–4
(14) 10
–5
10
(15) –6
10
Original
Approximation
7
10
(16) 10
2
10
3
10
4
10
5
10
6
Frequency [Hz]
(17)
Fig. 15. Rational function approximation of Y (s).
150
The procedure for fitting the blocks of with a common set Measured
of poles follows the one outlined in [11]. This procedure utilizes Calculated
Rational approximation
a sparse version of VF which permits individual, frequency de-
pendent weighting of the elements of the considered block of
. The particular weighting adopted was 100
1
2
Voltage ratio
(18)
50
This gives increased weight where elements are small, thus 3
tending to provide a fitting with a high relative accuracy rather
than a high absolute accuracy.
5
The number of iterations needed by VF was strongly reduced
0
by calculating a suitable set of starting poles. This was achieved 10
2
10
3
10
4 5
10
6
10
by first calculating the modes of the considered block of , Frequency [Hz]
stacking the modes into a single vector and subjecting this vector
to fitting by VF. The resulting poles were then taken as starting Fig. 16. Voltage transfer of rational approximation versus measured response.
poles for VF when fitting the block.
Fig. 15 shows the fitted elements of using 80 poles per C. Time Domain Implementation
submatrix. The increased error in the voltage transfer due to the The resulting approximation in the form of matrices —
rational approximation was negligible for the example in Fig. 7 can be included in EMTP-type programs as a conductance
and 8 (open HV side, open LV side). However, for the example matrix in parallel with a vector of past history current sources
in Fig. 9, the error increased noticeably around 3 kHz, as shown (see [11] for an explanation). A user-defined routine was
in Fig. 16. This is probably a result of the high condition number written for this purpose and included in MatTran [14] which is
causing a magnification of the fitting error. an EMTP-type program implemented in Matlab.
200
Measured
Simulated
10xV4
100
V2
Voltage [Volt]
Fig. 17. Excitation on high-voltage side. V3
–100
1500
–200
Measured
Simulated
V1
1000 V1
–300
500
10xV5 , 10xV6 –400
Voltage [Volt]
–1500 0
0.01 0 0.01 0.02 0.03 0.04 0.05
Voltage [Volt]
Time [ms] V3
–100
–300 V1
–400
–500
0 0.1 0.2 0.3 0.4 0.5
Time [ms]
Fig. 19. Excitation on low-voltage side.
Fig. 21. Voltage responses when the cable capacitances are not accounted for.
300
V2 Measured
V1 20xV4 Simulated
200
100
Voltage [Volt]
Fig. 22. Open terminals at both windings. 20xV6
0
500
Measured
400 V2 Simulated
100
300
200 200
10xV4
Voltage [Volt]
100
V3 300
0 10 0 10 20 30 40 50
Time [ms]
–100
V1
–200 Fig. 26. Measured and simulated voltage waveforms.
–300
accurate measurements, fast reconnections, and reproducible [5] Q. Su, R. E. James, and D. Sutanto, “A z-transform model of trans-
results. formers for the study of electromagnetic transients in power systems,”
IEEE Trans Power Syst., vol. 5, pp. 27–33, Feb. 1990.
Application to a distribution transformer has demonstrated [6] A. Morched, L. Marti, and J. Ottevangers, “A high frequency trans-
the accuracy of the procedure by comparison with redundant former model for the EMTP,” IEEE Trans. Power Delivery, vol. 8, pp.
measurements. The resulting model was, however, found to be 1615–1626, July 1993.
[7] T. Hasman, “Reflection and transmission of traveling waves on power
sensitive to errors when terminals were open-circuited at both transformers,” IEEE Trans. Power Delivery, vol. 12, pp. 1684–1689,
windings. It was found that the accuracy could be greatly im- Oct. 1997.
proved for this situation by including measured transferred volt- [8] B. Gustavsen and A. Semlyen, “Application of vector fitting to the state
equation representation of transformers for simulation of electromag-
ages in the model. The sensitivity became reduced at low fre- netic transients,” IEEE Trans. Power Delivery, vol. 13, pp. 834–842,
quencies when reducing the voltage ratio of the transformer. July 1998.
A linear model suitable for EMTP-type programs was ob- [9] Course EE263: Introduction to Linear Dynamical Systems, Lecture
notes from Prof. S.P. Boyd, Stanford University, CA.
tained by subjecting to rational approximation and subse- [10] L. N. Trefethen and D. Bau, Numerical Linear Algebra. Philadelphia,
quent passivity enforcement. Comparison between measured PA: SIAM, ISBN 0–89871–361–7, 1997, p. 95.
and simulated time domain voltages have demonstrated a high [11] B. Gustavsen, “Computer code for rational approximation of frequency-
dependent admittance matrices,” IEEE Trans. Power Delivery, vol. 17,
accuracy of the model both at high frequencies and at 50 Hz. pp. 1093–1098, Oct. 2002.
[12] B. Gustavsen and A. Semlyen, “Rational approximation of frequency
ACKNOWLEDGMENT domain responses by vector fitting,” IEEE Trans. Power Delivery, vol.
14, pp. 1052–1061, July 1999.
The author expresses his thanks to Prof. K. Feser and the [13] , “Enforcing passivity for admittance matrices approximated by ra-
tional functions,” IEEE Trans. Power Syst., vol. 16, pp. 97–104, Feb.
staff at the Institut für Energieübertragung und Hochspannungs- 2001.
technik, University of Stuttgart, Stuttgart, Germany, for pro- [14] J. Mahseredjian and F. Alvarado, “Creating an electromagnetic tran-
viding laboratory facilities and support of this project. Useful sients program in Matlab: MatEMTP,” IEEE Trans. Power Delivery, vol.
12, pp. 380–388, Jan. 1997.
discussions with Dr. T. Henriksen at SINTEF Energy Research [15] B. Gustavsen, “Frequency dependent modeling of power transformers
is appreciated regarding the usage of transferred voltages in with ungrounded windings,” IEEE Trans. Power Delivery, to be pub-
transformer modeling. This work was carried out at the Univer- lished.
sity of Stuttgart.
REFERENCES
[1] P. I. Fergestad and T. Henriksen, “Transient oscillations in multiwinding
transformers,” IEEE Trans. Power App. Syst., vol. PAS-93, pp. 500–509,
Mar./Apr. 1974.
[2] R. C. Degeneff, “A general method for determining resonances in Bjørn Gustavsen (M’94–SM’03) was born in Harstad, Norway, in 1965. He
transformer windings,” IEEE Trans. Power App. Syst., vol. PAS-96, pp. received the M.Sc. and Dr.-Ing. degrees from the Norwegian Institute of Tech-
423–430, Mar./Apr. 1977. nology, Trondheim, in 1989 and 1993, respectively.
[3] P. T. M. Vaessen, “Transformer model for high frequencies,” IEEE Currently, he is with SINTEF Energy Research, Trondheim. His interests in-
Trans. Power Delivery, vol. 3, pp. 1761–1768, Oct. 1988. clude simulation of electromagnetic transients and modeling of frequency-de-
[4] G. B. Gharehpetian, H. Mohseni, and K. Möller, “Hybrid modeling of pendent effects. He spent 1996 as a Visiting Researcher at the University of
imhomogenous transformer windings for very fast transient overvoltage Toronto, Toronto, ON, Canada, and the summer of 1998 at the Manitoba HVDC
studies,” IEEE Trans. Power Delivery, vol. 13, pp. 157–163, Jan. 1998. Research Center, Winnipeg, MB, Canada.