A Black-Box Approach to Interfacing White-Box

Transformer Models With Electromagnetic Transients

Programs
Bjørn Gustavsen, FIEEE Alvaro Portillo, SMIEEE
SINTEF Energy Research WEG Transformers
Trondheim, Norway Blumenau, Brazil
[email protected] [email protected]

Abstract—The standard lightning impulse voltage wave used in subnetworks by reducing the full nodal admittance matrix to a
transformer factory testing does not properly cover all the set of of external terminals and by calculating the voltage
overvoltage stresses that a transformer can experience while in transfer functions from external to internal terminals. The
service. To properly asses the internal overvoltages that can
occur in practice, we propose to interface the manufacturer's
admittance matrix and the voltage transfer functions are fitted
white-box transformer model with EMTP-type simulation tools with rational functions to obtain an EMTP-compatible model.
using a black-box modeling approach. From the white-box A similar approach was adopted in [10] for transformer
model we compute in the frequency domain data for the black- modeling starting from a transformer white-box model, but
box model which are fitted with rational functions, including with internal voltages calculated via the numerical Laplace
voltage transfer functions from external terminals to internal Transform.
nodes. The resulting black-box model is easily interfaced with
EMTP-type circuit solvers via a Norton equivalent and recursive
In the present work, we apply the original procedure in [8].
convolution. Application to the so-called fictitious transformer We start from a white-box model description in terms of
used by CIGRE A2/C4.39 demonstrates that ground fault RLCG matrices for establishing the complete terminal
initiation on a connected cable can generate internal admittance matrix at a set of discrete frequencies and
overvoltages that exceed those of the lightning impulse test. introduce the required connections between winding sections
via nodal analysis. This large admittance matrix is reduced
Index Terms-- Transformer, white box model, black box model,
EMTP, simulation. with respect to the transformer's external terminals and fitted
with a rational function approximant which satisfies the
physical conditions of stable poles and passivity. This rational
I. INTRODUCTION function-based model is interfaced with an EMTP-type solver
Transformer white-box models [1-4] are used by the using a Norton equivalent and convolutions. In addition to
manufacturers to verify the winding's ability to withstand the this terminal equivalent we calculate the voltage transfer
standard test voltages used in the factory tests, i.e. the function from the transformer external terminals to selected
standard lightning impulse voltage and sometimes also points along the winding which are subjected to modeling via
chopped waves. These white-box models are proprietary in rational functions. That way, also the internal overvoltages
the sense that they are normally not disclosed to the customer can be explored via the EMTP-type circuit simulator.
or third parties. CIGRE WG A2/C4.39 recommends however
that the manufacturer should upon request provide the II. WHITE BOX MODELING
customer with a terminal equivalent (black-box model) to We consider one common type of white-box models where
allow the inclusion of the transformer in studies of high- the windings are subdivided into parts, the so-called lumped-
frequency transient interaction between the transformer and
parameter type model [1-4]. Each part is represented by a
the power system [5]. Traditionally, the black-box models series RL branch in parallel with a capacitance (and
were obtained via frequency sweep measurements on the
conductance) which represents the winding series
transformer terminals and subsequent modeling via fitting
capacitance. These basic circuits are connected in cascade to
with rational functions [6],[7]. Often, however, such represent the complete winding structures, and mutual
measurements are not available and so the utilization of the
inductive coupling is introduced between all RL circuits. This
white-box model becomes an attractive option. defines a circuit with N nodes and M (inductive) branches.
Reference [8] shows a procedure for modeling of linear Finally, capacitances and conductances are introduced

This work was supported the consortium participants of the SINTEF-led

project "EM Transients", and by WEG Transformers.

978-1-4799-6415-4/14/$31.00 ©2014 IEEE

between nodes and ground to finalize the model, giving a j with a resistor r, one adds the conductance g to elements
model in terms of matrices C, G ∈  N × N and R, L ∈  M × M . (i,i) and (j,j) of Y, and the conductance −g to elements (i,j)
This modeling approach can take into account frequency and (j,i) of Y, where g=1/r. To ground a node i with a
dependency in the matrices if they are known. resistor r, one adds the conductance g to elements (i,i) of Y.
From Kirchoff's current law we get C. Extraction of Black-Box Admittance Data
iC + i L + iG =
iS (1) The next step is to calculate the admittance Yext defining
the relation between voltage vext and current iext at the external
where iS is the specified (total) current injection from external transformer terminals,
ground into the nodes. The capacitive and conductive iC and
iG are related to the node voltages by (2) and (3). The i ext ( s ) = Yext ( s ) v ext ( s ) (11)
inductive branch currents i are given by (4) which are related
To achieve this, we invert Y to get Z,
to the total inductive current flowing into node by (5).
Z( s ) = Y −1 ( s ) (12)
i C = Cv (2)
Next, we delete all rows and columns of Z except for those
i G = Gv (3)
corresponding to the transformer external terminals. The
retained (small) matrix is denoted Zext. Finally, we invert Zext
e Li + Ri
= (4)
to give the terminal admittance matrix,
i L = TT i (5) −1
(13)
Yext ( s ) = Z ext (s)
The incidence matrix T in (5) is dependent on the internal D. Voltage Transfer to Internal Nodes
connections. It relates the inductive branch currents i with the
total iL of the inductive branch currents flowing out of the The voltage at internal nodes is easily calculated via the
branch reference nodes. It also relates the node voltages v full Z in (12) when the voltage vext at the external nodes is
with the inductive branch voltages e (6), known. Assume that the nodes have an ordering such that the
external nodes come first,
e = Tv (6)
 v ext   Z A Z B  i ext 
 v  = Z (14)
Combining eq. (1)-(6) we get (7a), (7b),
 int   C Z D   i int 
Cv = −Gv − TT i + i S (7a)
Combining (14) and (11) we get the voltage transfer given by
 Tv − Ri
Li
= (7b) (15) and (16).
v int ( s ) = H transfer ( s ) v ext ( s ) (15)
III. COMPUTATION OF BLACK-BOX DATA
H transfer ( s ) = ZC Yext (16)
A. Admittance Matrix
In the frequency domain, we get for (7), IV. BLACK-BOX MODELING
(G + sC) v + T i =
iST
(8a) E. Approximation With Rational Functions
( R + sL ) i =
Tv (8b) In order to include the information of Yext and Htransfer in a
transient simulation, we calculate rational models on standard
Solving (8b) for i and inserting in (8a) gives the total state-space form,
admittance matrix Y
) CY ( sI − AY ) −1 BY + DY + sEY
Yext ( s= (17)
Y( s ) =(G + sC) + TT (R + sL) −1 T (9)
) C H ( sI − A H ) −1 B H + D H + sE H
H transfer ( s= (18)
which relates node voltages and current injection
The actual computation is performed using Vector Fitting
i S ( s) = Y( s) v( s) (10) [6] which gives a model with guaranteed stable poles. In the
case of Yext we fit all columns with a common pole set and
subject the model to passivity enforcement [7] to guarantee a
B. Internal Connections and Groundings
stable time domain simulation. In the case of Htransfer, we
In order to complete the model, we must introduce the calculate a separate model for each matrix column (19) so as
connections and groundings in the transformer. This is to obtain a more compact model with a given model order.
achieved via nodal analysis by adding short-circuits No passivity requirements exist for the model of Htransfer since
represented by small resistances. To connect two nodes i and it does not interact with the system.
 H transfer = [h1 h 2  h n1 ] , Section III-C. In addition we calculate the voltage transfer to
(19) internal nodes 24, 58, 59, 72, 73, 82, 83 using the approach in
h j (s) =
d j + C j ( sI − A j )b j
Section III-D.

F. Model Interface With Circuit Simulator I. Extraction of Black-Box Model

The model (17) is conveniently interfaced with EMTP-type The 2×2 Yext is fitted with a rational model (17) using
circuit simulators using a Norton equivalent with a fixed vector fitting (VF) with N=80 pole-residue terms [6]. In this
conductance matrix G and controlled current sources ihis case, the model is passive and requires no passivity
[8],[9], see Fig. 1. In each time step, the history current enforcement. Fig. 3 shows the elements of Yext as well as the
source is updated based on past history values of the terminal rational approximation. It is observed that a highly accurate
voltage, vext. The internal voltages are computed via the fitting has been achieved. The elements of H are also fitted
convolution between the terminal voltage vext and the impulse with a rational function using VF with N=80 poles (19)
response Htransfer. giving a highly accurate result, see Figs. 5 and 6.

iext(t)
G H*(:)
vext(t) vint(t)
vext ihis(t)

Fig. 1 Transformer with external ports and computation of internal voltages.

V. EXAMPLE: MODELING OF THE FICTITIOUS

TRANSFORMER

G. White-Box Model
We consider the fictitious transformer model described in
[5], see Fig. 4. The two windings are subdivided to give N=93
nodes and M=92 branches. The winding details are shown in
Fig. 3. We wish to create a model with respect to the four
external nodes indicated in Fig. 4.

Fig. 2 Fictitious transformer with external connection points.

Matrices R, L, G and C are computed for this winding,

with R, L evaluated at 1650 kHz. Fig. 3 Winding structure.

H. Calculation of Black-Box Data

From matrices R, L, G and C we calculate the 93×93
terminal admittance Y(s) using (9) over a band of discrete
frequencies between 50 Hz and 1 MHz. We use a
combination of 1001 logarithmically spaced samples and
2001 linearly spaced samples. The grounding of nodes 1 and
48 and the short circuit between nodes 57 and 60, 70 and 71,
and between 81 and 84 is achieved by adding to Y(s) the
stamp of large conductances g=106 as described in III-B.
The terminal admittance matrix is reduced to a 2×2 matrix
Yext with respect to nodes 47 and 70 using the approach in Fig. 4 Elements of Yext and its rational approximation.
 coincides with one of the transformer internal resonances.
To study this situation, we connect a 1000 m cable to the
transformer HV terminals as shown in Fig. 8. The cable has a
800 mm2 Cu conductor, paper-oil insulation and a lead
sheath. The cable is modeled using a frequency-dependent
traveling wave type model. The LV terminal is assumed to
be loaded with a 100 nF capacitor. We apply a unit step
voltage to the far cable end and calculate the overvoltages in
PSCAD.
Fig. 9 shows the voltage at the two cable ends as well as
the transferred overvoltage on the transformer LV side. It is
Fig. 5 Voltage transfer h47 from external node 47 to internal nodes 24, 58, observed that the transformer affects the impinging
59, 72, 73, 82, 83, and its rational approximation. overvoltage by its input impedance loading effect.
47

0.1 µF H.V Close @ t=0 1 p.u.

70
L.V
1000 m cable

1 48

Fig. 8 Cable ground fault initiation

Fig. 6 Voltage transfer h70 from external node 70 to internal nodes 24, 47,
58, 59, 72, 73, 82, 83, and its rational approximation.

VI. TIME DOMAIN SIMULATION RESULTS

J. Interface With Circuit Solver

The rational model (17) is interfaced with PSCAD v4.2
using the procedure described in [8],[9]. A terminal
equivalent (FDTE in [8]) is used for interfacing Yext with the
Fig. 9 Voltage response on cable and transformer external terminals
PSCAD main routine. The overvoltages on the transformer's
external terminals are used as input when calculating the Fig. 10 shows the resulting overvoltages at the internal
internal overvoltages using the transfer function component nodes 24, 58, 59, 72, 73, 82, 83. Clearly, a resonant condition
(FDTF in [8]). Both blocks FDTE and FDTF are based on occurs whereby internal voltages at the LV winding midpoint
recursive convolution via trapezoidal integration. reaches 4 p.u. of the cable voltage at the instant of the ground
fault initiation.

Fig. 7 User-defined components for model interface in PSCAD.

K. Cable Energization Study Fig. 10 Internal transformer voltages at nodes 24, 58, 59, 72, 73 , 82, 83.
One situation which can lead to high internal transformer
L. Overvoltages During Factory Test
overvoltages is ground fault initiation on a feeding cable. The
ground fault leads to an oscillating waveform on the The above result can be compared with the overvoltages
transformer terminal which can lead to excessive voltages during the transformer factory test. The transformer is
inside the transformer if the dominating frequency component intended for three-phase operation with 230 kV primary
voltage and insulation design for 1050 kV BIL. The peak internal overvoltages may result incorrect, mainly because of
value of the operating voltage (1 p.u.) is insufficient representation of the damping. One should
230 ⋅ 2 / 3 = 188 kV so that the peak value of the impulse therefore be careful when interpreting the results when
applied in general simulations.
voltage test is 1050/188=5.59 p.u. During such test, the other
terminals are grounded as shown in Fig. 11. VIII. CONCLUSIONS
Fig. 12 shows the resulting overvoltages. It is observed that We have proposed a procedure for interfacing transformer
the voltage at note #27 in the 66 kV LV winding is limited to manufacturer's white box models with EMTP-type circuit
below 1.5 p.u., which is much lower than the 4 p.u. in Fig. 10. simulators.
In fact, 4 p.u. corresponds to a voltage of 752 kV which is 1. The input to the model is white-box description in terms
much higher than the 325 kV (1.73 p.u.) test voltage that of RLCG matrices. The matrices can be frequency-
would be applied in a lightning voltage application to the LV dependent.
winding. Fig. 13 shows that the resulting waveforms when 2. The RLCG matrices are converted into an admittance
applying 325 kV to the LV terminal (#47) with the HV matrix description with respect to the external terminals,
terminal (#70) grounded. taking into account the internal connections of the
47 transformer. The admittance matrix is fitted with rational
functions, allowing interface with circuit simulators via a
5.59 p.u. Norton equivalent.
1.2/50 µs
70 H.V 3. Voltage transfer functions from the external terminals to
L.V internal nodes are calculated and fitted with rational
functions, again defining a model which can be included
in time domain simulations.
4. Time domain simulations showed that oscillating
1 48 overvoltages on the transformer terminals can lead to
Fig. 11 Factory test with standard lightning impulse voltage excessive internal overvoltages via voltage
magnification.
IX. REFERENCES
[1] J.A. Martinez-Velasco, Power system transients parameter
determination, CRC Press, 2010.
[2] W.J. McNutt, T.J. Blalock, R.A. Hinton, “Response of transformer
windings to system transient voltages”, IEEE Trans. Power Apparatus
and Systems, Vol. PAS-93, Issue 2, 1974, pp 457-467.
[3] Degeneff, R.C. "A general method for determining resonances in
transformer windings", IEEE Trans. Power Apparatus and Systems,
vol. PAS-96, no. 2, pp. 423–430., March/April 1977.
[4] A. Greenwood, Electrical transients in power systems, 2nd edn., Wiley,
New York, 1991.
[5] Technical Brochure CIGRE A2/C4.39 (To be published).
Fig. 12 Internal voltages with lighting impulse voltage to node #70 [6] B. Gustavsen and A. Semlyen, “Rational approximation of frequency
domain responses by vector fitting”, IEEE Trans. Power Delivery, vol.
14, no. 3, pp. 1052-1061, July 1999.
[7] B. Gustavsen and A. Semlyen, “Enforcing passivity for admittance
matrices approximated by rational functions”, IEEE Trans. Power
Systems, vol. 16, pp. 97-104, Feb. 2001.
[8] B. Gustavsen and O. Mo, “Interfacing convolution based linear models
to an electromagnetic transients program”, Proc. Int. Conf. on Power
Systems Transients, June 4-7, 2007, Lyon, France, 6 p.
[9] B. Gustavsen and H.M.J. De Silva, “Inclusion of rational models in and
electromagnetic transients program – Y-parameters, Z-parameters, S-
parameters, transfer functions", IEEE Trans. Power Delivery, vol. 28,
no. 2, pp. 1164-1174, April 2013. B. Gustavsen and A. Portillo,
"Interfacing white-box transformer models with electromagnetic
transients programs via state equations and diagonalization", IEEE
Trans. Power Delivery, vol. 28, no. 2, pp. 1164-1174, April 2013.
Fig. 13 Internal voltages with lighting impulse voltage to node #47 [10] C. Álvarez-Mariño, X.M. Lopez-Fernandez, A.J.M. Jacomo Ramos,
R.A.F. Castro Lopes, J.M. Duarte Couto, (2012) "Time domain severity
VII. DISCUSSION factor (TDSF): Induced transient voltage between transformer and
vacuum circuit breakers", COMPEL: The International Journal for
As discussed in [5], the white-box models by Computation and Mathematics in Electrical and Electronic
manufacturers are mainly intended for assessing the internal Engineering, vol. 31, no. 2, pp.670 – 681.
voltage distribution when the winding is subjected to the
standard lightning wave, i.e. a unipolar impulse voltage.
When applied with oscillating voltage applications, the