Electrical Power and Energy Systems 69 (2015) 67–75

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Analysing a power transformer’s internal response to system transients

using a hybrid modelling methodology
Steven D. Mitchell a,⇑, Gustavo H.C. Oliveira b
a
School of Electrical Engineering and Computer Science, University of Newcastle, Callaghan, NSW 2308, Australia
b
Electrical Engineering Department, Federal University of Parana, Curitiba, PR 81531-990, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: This article presents a novel approach to analysing a power transformer’s internal response to system
Received 16 September 2013 transients. In this approach a hybrid modelling methodology is adopted which leverages the distinct
Received in revised form 16 December 2014 advantages offered by both Black and Grey Box modelling techniques. The Black Box model of the trans-
Accepted 24 December 2014
former is used within the EMTP system study environment in order to take advantage of its mathematical
flexibility and modelling accuracy. Transients derived from network switching operations within the
study can then be used for injection tests within the Grey Box modelling environment. The Grey Box
Keywords:
model, which is based upon the physical structure of the transformer, will facilitate analysis of the trans-
Power transformer
Black Box
former’s internal voltage response to the external stimulus. A fundamental difference between the
Grey Box approach described in this paper and more traditional approaches is that it does not require prior knowl-
Model edge of the internal geometry of the transformer. All of the modelling parameters are derived from exter-
Resonant overvoltage nal tests, nameplate details and an intrinsic understanding of common transformer design principles.
EMTP This can be a distinct advantage since in most cases a transformer’s design specifications are not readily
available outside of the laboratory due to the manufacturer’s intellectual property restrictions. A study of
a gas insulated substation within a hydroelectric power plant in Brazil is used to demonstrate the pro-
posed technique.
Ó 2015 Elsevier Ltd. All rights reserved.

Introduction Power transformers each have their own characteristic fre-

quency response [8]. To predict how a power transformer will
Electrical power system switching operations can generate a behave under different transient conditions, a modelling approach
broad spectrum of transient frequencies [1]. The transient ampli- may be adopted. In fact, mathematical modelling of dynamic sys-
tude may not be sufficiently high to initiate a reaction by surge tems can generally be divided into two basic approaches, in terms
protection, however the frequency content of the transient may of procedures for selecting the model structure and calculating the
be such that there is a match with the natural frequency modes model parameters [9–11]: White-Box (or physical) modelling and
of equipment connected to the electrical network. A case in point Black-Box modelling. A methodology that is a compromise
is power transformers [2]. When a switching transient frequency between these two approaches is the Grey-Box model. This termi-
component aligns with an internal resonance frequency within a nology is associated with methods and models that can be put on a
power transformer, voltage amplification can occur which can scale ranging from a pure White-Box physical model to a pure
result in a breakdown of the transformer’s insulation system. This Black-Box parameterized model [9–11]. Therefore, this will be
is an area of study with a long history [3], however the area is now the nomenclature for transformer models used here.
receiving increased attention due to an increasing number of trans- A White Box model uses intimate knowledge of the internal
former failures which have been attributed to internal resonance geometry and material properties of the transformer to build a
overvoltage conditions [4,5].Working groups from both IEEE [6] lumped parameter electrical network representation of the
and CIGRE [7] have been established to investigate ways of transformer [12–14]. Another common approach is to build a dis-
mitigating the problem. tributed electrical model which views the windings as multi-
conductor transmission lines (MTL) [15]. Simple White Box models
can be incorporated into an electrical system model within an Elec-
⇑ Corresponding author.
tromagnetic Transients Program (EMTP). However their applica-
E-mail addresses: [email protected] (S.D. Mitchell), gustavo@
eletrica.ufpr.br (G.H.C. Oliveira). tion within EMTP becomes difficult when implementing a more

http://dx.doi.org/10.1016/j.ijepes.2014.12.064
0142-0615/Ó 2015 Elsevier Ltd. All rights reserved.
68 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75

comprehensive model which will be accurate across a broader fre- procedure for transient analysis is described in Section ‘Black Box
quency spectrum. Such a model will need to take into account var- modelling for electrical system transient analysis’. The Grey Box
ious non-linear frequency dependent parameter properties such as modelling procedure and the estimation of an internal winding
the complex permittivity of the transformer’s insulation system, transient response is presented in Section ‘Grey Box modelling to
magnetic skin effects associated with the transformer core, and estimate the internal transient response’. The electrical system
the skin and proximity effects within the transformer’s windings description and study results are presented in Section ‘Example:
[16,8,17,18]. Another disadvantage of the White Box model is that generator transformer within a GIS substation’. Concluding
their construction is directly based upon the transformer’s design remarks are given in Section ‘Conclusion’.
blueprint. Rarely is this information made available due to the
manufacturer’s intellectual property restrictions. This makes the
construction of a true White Box model difficult outside of the lab- Hybrid modelling methodology
oratory or without close collaboration with the manufacturer.
In contrast to the physically representative White Box approach Power transformer frequency response measurements are used
is the application of a Black Box model. The Black Box model is a to build both the Black and Grey Box transformer models used in
purely mathematical representation of the terminal response of the proposed methodology. The Black Box model is incorporated
the transformer. Its parameters are derived using system identifi- into an EMTP simulation of the electrical system under study.
cation methods on experimentally recorded time and/or frequency EMTP analysis will facilitate the determination of worst case sys-
domain data in order to establish the dynamic behaviour of the tem transient behaviour scenarios for the transformer terminals.
system [19–22]. A Black Box model can achieve high levels of accu- These transients are then injected into the Grey Box model in order
racy and can be readily incorporated into an EMTP electrical sys- to estimate the internal response at nominal ‘‘nodal’’ points
tem model. The disadvantage is that this modelling approach throughout the transformer’s windings. This is accomplished by
does not provide any information regarding the internal behaviour taking the Fourier Transform of the terminal transient signal and
of the windings. multiplying it by each of the Grey Box model’s terminal to node
A compromise between the Black and White Box modelling transfer functions. This will provide an estimate for the transient
approaches is the Grey Box. Unlike the Black Box approach, both output spectrum at each of the model’s winding nodes. The appli-
the White and Grey Box models are based on a transformer’s phys- cation of an Inverse Fourier Transform on each nodal spectrum will
ical structure. Like the White Box approach, a comprehensive Grey then determine the transient voltage response at each of the Grey
Box model can incorporate non-linear frequency dependent terms Box model’s winding nodes. A diagrammatic representation of the
which will make it unsuitable for implementation within EMTP. proposed hybrid modelling methodology is given in Fig. 1.
The difference between these two methods is primarily in the The following sections will discuss the implementation of both
determination of the model parameters. For the Grey Box model, the Black and Grey Box modelling approaches.
many of the physical parameters may be unknown and will need
to be estimated. One way of estimating the parameter values is
by fitting the model’s transfer function to external measurements,
such as Frequency Response Analysis (FRA) [18,23,24]. It is critical
Start
however that the estimated parameters are representative of the
transformer. An estimator that is not appropriately constrained
can converge on a parameter set which may satisfy the objective
function but is not physically representative of the transformer Measure Transformer
[23]. Such risks can be minimised by constraining parameter val- Frequency Response
ues using acknowledged transformer design principles supported
by targeted external measurements [25]. The advantage of the
Grey Box model is that the limitations associated with access to
Black Box Model Grey Box Model
the transformer’s construction details can be removed, however Non-linear frequency dependent
Pure mathemacal model based on
this will inevitably require modelling assumptions to be made orthonormal basis funcons model based upon an esmate of the
which can lead to some modelling inaccuracy. physical structure of the transformer
This article proposes a novel hybrid modelling methodology to
facilitate the analysis of system transients within the internal
EMTP System Simulaon Nodal Transfer Funcons
winding structure of a power transformer. The hybrid approach
Facilitates the study of worst case Generate transfer funcons between
leverages the advantages offered by both the Black and Grey Box system transient behaviour scenarios terminals and internal nodes
modelling techniques. The high levels of modelling accuracy
offered by the Black Box approach, together with its compliance
within an EMTP system study environment are leveraged to deter-
Terminal Electrical Transient Nodal Output Spectrum
mine worst case terminal transient conditions for the transformer. Esmated electrical transient at the Determine the output spectrum for
The nominal transient conditions are then injected into a Grey Box transformer terminals each node
model which facilitates the determination of the resulting internal
winding response. Unlike traditional transient study approaches,
this methodology does not require access to the manufacturer’s
Transient Voltage Esmate
design specifications [26,27], but will facilitate a prediction for Calculate the transient voltage
the transient response at nominal positions within the winding observed at each node
structure. The approach is demonstrated using data from a gas
insulated substation (GIS) within a hydroelectric power plant in
Brazil.
The paper is structured in the following manner. Section End
‘Hybrid modelling methodology’ will discuss the proposed hybrid
modelling methodology. A discussion on the Black Box modelling Fig. 1. Hybrid modelling methodology.
 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75 69

Black Box modelling for electrical system transient analysis

X0 H0
1 1
2 Cgx1 2 CHx1
When conducting electromagnetic transient analysis on a
power system, its power transformers can be represented by an
n terminal admittance matrix Y where,
1
2 CgH1
I ¼ YV ð1Þ
or, in terms of its frequency response,
Rx1 RH1
2 3 2 3 2 3
CSx1
I1 ðjwÞ Y 11 ðjwÞ Y 12 ðjwÞ    Y 1n ðjwÞ V 1 ðjwÞ
6 I ðjwÞ 7 6
6 2 7 6  Y 22 ðjwÞ    Y 2n ðjwÞ 7 6 7
7 6 V 2 ðjwÞ 7 Cgx2 Lx1 LH1 CSH1
6 7 6 76 7:
6 .. 7 ¼ 6 .. .. 7 6 .. 7
4 . 5 4 . . 5 4 . 5
In ðjwÞ   Y nn ðjwÞ V n ðjwÞ CgH2
ð2Þ
Rx2 CHx2 RH2
In this equation, Ii ðjwÞ and V l ðjwÞ are the frequency responses of the CSx2
transformer current and voltage at terminals i and l, respectively.
Y il ðjwÞ is the frequency response of the element ði; lÞ. The symbol Cgx3 Lx2 LH2 CSH2
 indicates a symmetric structure.
The calculus of Y il ðjwÞ requires the measurement of Ii ðjwÞ and CgH3
V l ðjwÞ for w across a wide band of frequencies. Given these mea-
surements, a transformer MIMO (multi-input–multi-output) non-
Rx3 CHx3 RH3
parametric model in the frequency domain is obtained. The goal CSx3
is to find a parametric model in the state-space realisation of Y
such as, Cgx4 Lx3 LH3 CSH3

_
xðtÞ ¼ A xðtÞ þ B v ðtÞ
iðtÞ ¼ C xðtÞ þ D v ðtÞ
; ð3Þ
CgH4
CHx4
where i and v are vectors comprised of the terminal time-domain
currents Ii and voltages V l respectively for i; l ¼ 1; . . . ; n. The deter-
mination of a parametric model for matrix YðsÞ, or for each element
of Y il ðsÞ from the frequency response measurements, is known as
frequency-domain system identification [20].
Cgxn
Matrices A; B; C and D of model (3) can be estimated directly
from a subspace system identification method [28]. However, each CgHn
element Y il ðsÞ of Y can also be parametrized as a rational transfer Rxn CHxn RHn
function such as:
Bil ðsÞ Csxn
Y il ðsÞ ¼ ð4Þ
LHn CSHn
;
Ail ðsÞ
Lxn
where Ail ðsÞ and Bil ðsÞ are polynomials in s. This model can be
expanded as a truncated series of basis functions f/ilm ðtÞg1
m¼1 [29],
1 1C
1 2 CHx(n+1)
as follows:
2 gH(n+1)
Bil ðsÞ XN 2 Cgx(n+1) X1 H1
Y il ðsÞ ¼ ¼ cil0 þ cilm Uilm ðsÞ; ð5Þ
Ail ðsÞ m¼1
Fig. 2. Grey Box model.
where Uilm ðsÞ is the Laplace transform of /ilm ðtÞ. Selections for this
basis function can assume different forms, such as a single-pole par-
tial fraction decomposition [30] or the Takenaka–Malmquist func-
where Y il ðjwk Þ is the measured frequency response of an element of
tions [31]. b il ðjw Þ is the estimated model, with
(1) at frequencies fwk gKk¼1 , Y k
parameters hil used to approximate the system dynamics. Nil ðjwk Þ
Model parameter estimation
is a weighting function.
The problem of estimating the parameters of Y b il ðsÞ has been
Once there is a model structure, such as those previously pre-
studied by several authors. In the context of power transformer
sented, a set of parameters for fitting the model with the measured
models and frequency domain data, some examples are
admittance matrix elements is required.
[21,19,32]. A subspace method for system identification has been
The estimation problem of interest can be stated as the problem
applied by [32] for the problem of estimating power transformer
of minimising the following objective function. For the sake of sim-
models from frequency response data. Tests were conducted with
plicity, lets consider the problem of estimating one element of Y,
data sets obtained from two identical 132/66/11 kV 30 MVA units.
that is, Y i;l ðsÞ.
In [19], a system identification method, where the main character-
X
K  2 istic is the use of a basis function (see Eq. (5)) called Frequency
 b il ðjw Þ ;
Jðhil Þ ¼ jNil ðjwk ÞjY il ðjwk Þ  Y k ð6Þ Localizing Basis Function, has been applied to determinate a para-
k¼1
metric model for a power transformer. The data used in this work
70 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75

has been collected from a frequency response analysis test per- (3) Select a new set of poles for Uilm ðsÞ.
formed on the A-phase high voltage winding of an ABB power With the estimated parameters hil ; A b il ðsÞ (Eq. (9)) is defined.
transformer with a rating of 132 kV 60 MVA. The method pre- The zeros of A b il ðsÞ, or roots of N A ðsÞ, are the values for the
il
sented by Oliveira et al. [21] is similar to the one discussed here, poles of Uilm ðsÞ in the next iteration. Unstable poles are
however it focused on using discrete-time models instead of con- always flipped to the stable region by making their real parts
tinuous-time models. Two power transformer data sets were used negative. The procedure is repeated (that is, steps 1, 2 and 3)
to illustrate the model parameter estimation method. The data sets until the poles have converged.
were obtained from a three-phase 500/345/13.8 kV transformer, (4) Final parameter estimation.
and a mono-phase 345/230/13.8 kV transformer. Once the poles have converged (meaning that A b il ðsÞ tends to
An issue related to the approach taken in (6) is the selection of Dil ðsÞ), the objective function (10) tends to (6). The model
the dynamics of the basis functions /ilm ðtÞ (or poles of Uilm ðsÞ). This parameters for the model (4) can be obtained minimising
is frequently called basis function pole selection (refer for instance the following objective function:
[33]). Such problems are non-linear and may converge to local  
XK  XN 
minima.  
Jðhil Þ ¼ jNil ðjwk ÞjY il ðjwk Þ  ðcil0 þ cilm Uilm ðjwk ÞÞ:
An alternative approach is to look for the pole of the basis func- k¼1
 m¼1

tions using the so-called Sanathanan–Koener iterations [34,31]. ð12Þ
The procedure can be summarised as follows.
b il ðsÞ.
Model (3) can be computed based on each estimated Y
(1) Select the model order N and a initial set for the poles basis
functions. Grey Box modelling to estimate the internal transient response
Let us recall model (4), which will be now defined as:
Grey Box transformer model
b
b il ðsÞ ¼ B il ðsÞ ;
Y ð7Þ
b il ðsÞ As in the case of the White Box model, the Grey Box model is a
A
mathematical representation of the complex electromagnetic rela-
with tionships that exist within a transformer. A common approach is
the use of the ladder network model which can originally be traced
B XN
b il ðsÞ ¼ Nil ðsÞ ¼ b þ
B bilm Uilm ðsÞ; ð8Þ
back to Blume and Boyajian’s work of 1919 [3]. Their work has sub-
il0
Dil ðsÞ m¼1
sequently been improved by many other researchers over the fol-
A XN lowing decades including [35–37]. A single phase ladder network
b il ðsÞ ¼ Nil ðsÞ ¼ 1 þ
A ailm Uilm ðsÞ: ð9Þ model based on [18] is given in Fig. 2 where L; R and C represent
Dil ðsÞ m¼1 the frequency dependent inductance, resistance and capacitance
elements respectively. The inductance element takes into account
Note that, once the basis functions dynamics (poles) have
the mutual inductance it shares with the other inductive elements,
been defined, the model parameters becames
it also includes its leakage inductance contribution and losses asso-
hil ¼ fail1 ; . . . ; ailN ; bil0 ; bil1 ; . . . ; bilN g.
ciated with the core. More comprehensive detail on this model and
(2) Run an estimation parameter procedure.
its application can be found in [38]. Given the Grey Box trans-
An objective function Jðhil Þ related with this problem is:
former model structure, the next step is to estimate the model
 2 parameters.
XK NA ðjw Þ N Bil ðjwk Þ
 il k
Jðhil Þ ¼ jNil ðjwk Þj Y ðjwk Þ   ð10Þ
k¼1
Dij ðjwk Þ il Dij ðjwk Þ Terminal transfer functions

A search for optimal parameters hil that minimises the objec- Analysis of the Kirchhoff current and voltage relationships that
tive function (10) is then performed, that is: exist within the transformer model structure will facilitate the
hil ¼ argminJðhil Þ: ð11Þ construction of a state space representation of the model [39].
The state space model will be of the form,
Since Uilm ðsÞ has been previously defined in step (1) and is
_
xðtÞ ¼ AðuðtÞÞxðtÞ þ BðuðtÞÞuðtÞ
assumed known in this step, such a problem is linear in hil
and can be solved using standard least square algorithms. vðtÞ ¼ CðuðtÞÞxðtÞ ð13Þ

u(t) H1 H1 H2 H3 H8 H9 H0

X1 X0
X1 X2 X3 X8 X9

400 400

Fig. 3. Grey Box model terminal connections.

 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75 71

Fig. 4. Section of the GIS substation.

700

600

500

400
kV

300

200

100

0
0 5 10 15 20 25 30 35 40 45
t(us)
Fig. 5. Black Box model vector fittings to the High Voltage Winding End to End
Open Circuit FRA test measurements. Fig. 6. Estimated transient voltage for transformer terminal H1.

where x is the state vector, v is the voltage on the nominated output End to End Open Circuit, and the Capacitive Interwinding FRA tests
terminal, u the voltage on the nominated input terminal, A the state [40]. The model parameters were determined by finding a global
matrix, B the input matrix, and C the output matrix. Note that A, B cost function minima using a constrained nonlinear optimisation
and C are non-linear frequency dependent matrices due to a range algorithm. The cost function used for this case study was,
of phenomena. These include permeability attenuation as a result of  !2  !2
 b H H ðjwÞ 
H  b H X ðjwÞ 
H
magnetic skin effect within the transformer core, winding skin and    
J ¼ log10 1 0
 þ log10 1 1

proximity effects, and the complex permittivity of the dielectric  GH1 H0 ðjwÞ   GH1 X 1 ðjwÞ 
materials [16–18].  !2
 Hb X X ðjwÞ 
A transfer function of the model, H, is derived by taking the  
þ log10 2 1
 ; ð15Þ
Laplace transform of (13),  GX2 X 1 ðjwÞ 
VðjwÞ 1 b is the estimated model transfer func-
where G is the FRA data and H
HðjwÞ ¼ ¼ CðwÞðjwI  AðwÞÞ BðwÞ ð14Þ
UðjwÞ tion from (14).
The transfer function (14) is used to determine the model’s fre- Now that a Grey Box model for the transformer has been devel-
quency response for a given set of parameters. Note that there will oped, estimates for the internal winding response to unique sys-
be a transfer function corresponding to each FRA test. tem transients can now be determined.
The next step is to determine the best fit between each transfer
function and its corresponding FRA data set by finding the model Estimated internal winding response to system transients
parameters that will minimise an appropriate cost function. Three
different FRA tests were conducted. These tests were the High Volt- In order to simplify the transient simulation it is assumed that
age Winding End to End Open Circuit, the Low Voltage Winding H0 is connected to ground and the low voltage terminals X 1 and X 2
72 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75

Fig. 7. Grey Box model transfer function fitted to the High Voltage End to End Open
Circuit FRA test measurement.

Fig. 10. H1 terminal to HV node transfer functions.

[41,42]. Fig. 3 also shows the input transient signal, uðtÞ, as well
as highlighting the location of the model’s HV winding nodes, kH1
through to kH9 , and the model’s LV winding nodes, kX1 through to
Fig. 8. Grey Box model transfer function fitted to the Low Voltage End to End Open kX9 . Note that the winding nodes, k, are representative of their dis-
Circuit FRA test measurement. tributed locations within the transformer’s physical winding struc-
ture. For example, kHðn2Þ represents the electrical midpoint of the HV
winding.
The input transient signal is determined from the EMTP tran-
sient study using the Black Box model as discussed in the previous
section. Since the Grey Box model takes into account the frequency
dependent behaviour of the transformer core and the skin and
proximity effects within the transformer’s windings, it is necessary
for the analysis to be conducted in the frequency domain. Note that
by using frequency domain analysis, the output spectrum for any
node can be derived by multiplying the input signal spectrum by
the corresponding node transfer function.
To determine the input signal spectrum we apply a Fourier
Transform to uðtÞ, that is, UðjwÞ. The next step is to determine
the transfer function between the high voltage input terminal H1
and a node within the winding structure to be analysed (kH1
through to kH9 for the high voltage winding and kX1 through to
kX9 for the low voltage winding). With reference to (14) and
acknowledging that the output matrix will facilitate node selection
for the transfer function, transfer functions can be determined for
Fig. 9. Grey Box model transfer function fitted to the Capacitive Interwinding FRA
each of the internal winding nodes within the Grey Box model.
test measurement. After estimating the transfer function for nominal winding node
b H k ðjxÞ, the spectrum at this point due to the transient input
k, H 1

signal is,
are terminated via 400 X resistors (Fig. 3). The termination resis-
b k ðjxÞ ¼ UðjxÞ H
V b H k ðjxÞ; ð16Þ
tors are representative of the transmission line surge impedance 1
 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75 73

Example: generator transformer within a GIS substation

This example demonstrates the proposed methodology by esti-

mating the distributed internal voltage stress within a power
transformer due to system switching transients generated within
the substation. The substation is part of a hydroelectric power
plant belonging to the Brazilian power grid. A system drawing of
the section of the substation under consideration is given in Fig. 4.
In this example it is considered how system transient voltages
generated by switching on disconnectors or circuit breakers will
propagate through the windings of the transformer T1 which is
connected to generator U3. The transformer under consideration
is a single-phase 525/18 kV 256 MVA generator transformer. In
order to study such a problem and to simulate the above men-
tioned phenomena, the substation section was modelled using
EMTP-RV software and only single-phase models of the elements
were considered. The transmission lines between the components
of the circuit were modelled within EMTP using a distribution
parameter model (modal surge impedance and propagation
velocity).

Black-Box modelling results and system transient response

The Black-Box identification procedure described in Section

‘Black Box modelling for electrical system transient analysis’ is
now applied to fit a state-space model to the High Voltage End
to End Open Circuit FRA test data obtained from the generator
transformer. The Black-Box model structure used here is given by
(5),with i ¼ l ¼ 1, and a set of eight Takenaka–Malmiquist func-
tions to form the basis (N ¼ 8). After only 40 iterations using the
identification procedure described in ‘Model parameter estima-
tion’, the model parameters have converged to achieve an excellent
Fig. 11. H1 terminal to LV node transfer functions. fit between the model frequency response and the measurement
data as shown in Fig. 5.
After incorporating the generator transformer’s Black Box
where V b k ðjxÞ is the resulting output spectrum. Finally an estimate
model into the GIS substation electrical system EMTP simulation,
for the transient voltage in the time domain v^ k ðtÞ, at nominal wind- switching transient studies were then conducted. The scenario
ing node k can be determined by taking the Inverse Fourier Trans- considered in this paper was based on the closure of circuit breaker
form of Vb k ðjxÞ,
05U03 whilst circuit breaker 45U34 was open. This is equivalent to
energising the GIS point PM1 from BUS-A, which is already energ-
v^ k ðtÞ ¼ F 1 f Vb k ðjxÞg: ð17Þ
ised from the system, that is, from LT IPU MD2. The estimated

Fig. 12. 525 kV/18 kV 256 MVA generator transformer HV node transient response.
74 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75

Fig. 13. 525 kV/18 kV 256 MVA generator transformer LV transient response.

transient voltage for the transformer’s high voltage terminal H1 former models to the problem. The Black Box model’s mathemati-
under these conditions is shown in Fig. 6. cal flexibility is leveraged within the EMTP system study to
The next step in the hybrid modelling methodology was to determine terminal transient conditions for the transformer. The
inject the estimated transformer terminal transient into its Grey transient conditions are then injected into the physically represen-
Box model. tative non-linear Grey Box model in order to determine the inter-
nal transient response within the transformer’s windings. The
Grey Box modelling results and internal voltage estimation parameters associated with both models are derived from external
measurements only and are therefore not subject to intellectual
By applying the procedure presented in Section ‘Grey Box mod- property and internal access restrictions as is typically the case.
elling to estimate the internal transient response’, the Grey Box A case study of switching operations within a large GIS substation
model parameters were determined by fitting its corresponding and their influence on the internal structure of a generator trans-
transfer functions to the transformer FRA measurements (HV End former was used to demonstrate the utility of this approach. Since
to End Open Circuit, LV End to End Open Circuit, Capacitive Inter- internal access to the generator transformer is not a viable propo-
winding FRA tests). As shown in Figs. 7–9, a satisfactory fitting sition, future work includes further validation of this paper’s pro-
which includes the dominant resonant activity was achieved for posed methodology utilising a scaled laboratory equivalent.
frequencies up to 500 kHz.
With the Grey Box model parameters derived, transfer func- Acknowledgments
tions between the high voltage input terminal H1 and the various
nodes within the model’s winding structure were then determined The authors would like to acknowledge Ing. Robson A. Oliveira
(Figs. 10 and 11)). It is observed that the estimated frequency and Ing. José G.R. Filho for their valuable input. The second author
responses of the internal nodes of the transformer do not contain would like to acknowledge SETI/Fundação Araucária and CNPq for
significant resonances. This aligns well with the data set used in supporting this research.
the system identification procedure since this data too shows only
low levels of resonant activity (Figs. 7–9). References
Given the nodal transfer functions and the injected transient
signal, from (16) and (17), an estimate for the transient voltage [1] De A, Chakrabarti A, Hazra P. Resonant overvoltages produced in EHV
transformer windings due to power system transients. In: India annual
in the time domain at each nodal location was calculated. Figs. 12 conference, 2004. Proceedings of the IEEE INDICON 2004. First; 2004. p. 74–
and 13 show the respective HV and LV node transient responses. 7. http://dx.doi.org/10.1109/INDICO.2004.1497709.
This example shows the benefits the proposed methodology [2] Gustavsen B. Study of transformer resonant overvoltages caused by cable-
transformer high-frequency interaction. In: Power and energy society general
offers. An estimation of the internal voltage profile of a transformer meeting, 2011 IEEE, vol. 1; 2011. ISSN 1944-9925. http://dx.doi.org/10.1109/
without requiring prior knowledge of its internal design PES.2011.6038890.
specifications. [3] Blume LF, Boyajian A. Abnormal voltages within transformers. Trans Am Inst
Electr Eng 1919;XXXVIII(1):577–620. http://dx.doi.org/10.1109/T-
AIEE.1919.4765613. ISSN 0096-3860.
Conclusion [4] Gustavsen B, Brede A, Tande J. Multivariate analysis of transformer resonant
overvoltages in power stations. IEEE Trans Power Deliv 2011;26(4):2563–72.
http://dx.doi.org/10.1109/TPWRD.2011.2143436. ISSN 0885-8977.
This research proposes a new methodology for the determina- [5] Massaro URR. Electrical transient interaction between transformers and power
tion of the internal voltage profile and, thus, potential resonant system, brazilian experience. In: Proceedings of the international power
overvoltage conditions within a power transformer due to external system transients conference; 2009.
[6] IEEE guide to describe the occurrence and mitigation of switching transients
switching events within the power network. The methodology is induced by transformer, switching device, and system interaction. IEEE
based on the cooperative application of Black and Grey Box trans- unapproved draft std PC57.142/D6; April 2009.
 S.D. Mitchell, G.H.C. Oliveira / Electrical Power and Energy Systems 69 (2015) 67–75 75

[7] Electrical transient interaction between transformers and the power system. Trans Power Deliv 2008;23(2):760–7. http://dx.doi.org/10.1109/
CIGRE working group A2/C4.39. TPWRD.2008.915815. ISSN 0885-8977.
[8] Harlow JH, editor. Electric power transformer engineering. The electric power [25] Mitchell S, Welsh J. Initial parameter estimates and constraints to support gray
engineering series. CRC Press; 2004. box modeling of power transformers. IEEE Trans Power Deliv
[9] Ljung L. System identification, theory for the user. 2nd ed. Prentice Hall; 1999. 2013;28(4):2411–8. http://dx.doi.org/10.1109/TPWRD.2013.2259266. ISSN
[10] van den Bosh PPJ, van der Klauw AC. Modeling, identification and simulation of 0885-8977.
dynamic systems. CRC Press; 1994. [26] Popov M, van der Sluis L, Smeets RPP, Roldan J. Analysis of very fast transients
[11] Nelles O. Nonlinear system identification. Springer; 2001. in layer-type transformer windings. IEEE Trans Power Deliv
[12] Martinez JA, Mork BA. Transformer modeling for low and mid frequency 2007;22(1):238–47. http://dx.doi.org/10.1109/TPWRD.2006.881605. ISSN
transients a review. IEEE Trans Power Deliv 2005;20(2):1625–32. 0885-8977.
[13] Bjerkan E, Hoidalen H, Moreau O. Importance of a proper iron core [27] Lopez-Fernandez X, Alvarez-Marino C, Couto D, Lopes R, Jacomo-Ramos A.
representation in high frequency power transformer models. In: Modeling and insulation design methodology in power transformer under fast
International symposium on high voltage engineering, Beijing, China. transients. In: 2010 XIX international conference on electrical machines
[14] Bjerkan E. High frequency modeling of power transformers – stresses and (ICEM); 2010. p. 1–6. http://dx.doi.org/10.1109/ICELMACH.2010.5608019.
diagnostics. PhD thesis. Faculty of Information Technology, Mathematics and [28] Katayama T. Subspace methods for system identification. Springer; 2005.
Electrical Engineering; 2005. [29] Heuberger PSC, van den Hof PMJ, Wahlberg B. Modelling and identification
[15] Zhang Q, Wang S, Qiu J, Jing X, Gao C, Zhu JG, et al. Application of an improved with rational orthogonal basis functions. Springer; 2005.
multi-conductor transmission line model in power transformer. IEEE Trans [30] Gustavsen B, Semlyen A. Rational approximation of frequency domain
Magn 2013;49(5):2029–32. responses by vector fitting. IEEE Trans Power Deliv 1999;14:1052–61.
[16] Abeywickrama KGNB, Podoltsev AD, Serdyuk YV, Gubanski SM. Influence of [31] Dirk Deschrijver BH, Dhaene T. Orthonormal vector fitting: a robust
core characteristics on inductance calculations for modeling of power macromodeling tool for rational approximation of frequency domain
transformers. In: First international conference on industrial and responses. IEEE Trans Adv Pack 2007;30(2):216–25.
information systems; 2006. p. 24–9. http://dx.doi.org/10.1109/ICIIS.2006. [32] Akcay H, Islam SM, Ninness B. Subspace-based identification of power
365629. transformer models from frequency response data. IEEE Trans Instrum Meas
[17] Mitchell SD, Welsh JS. The influence of complex permeability on the 1999;48:700–7004.
broadband frequency response of a power transformer. IEEE Trans Power [33] Silva TO. Optimal pole conditions for laguerre and two parameter kautz
Deliv 2009(99):1. http://dx.doi.org/10.1109/TPWRD.2009.2036358. ISSN models: a survey of know results. In: Proc of the IFAC symp on system
0885-8977. identification; 2000.
[18] Mitchell S, Welsh J. Modeling power transformers to support the [34] Sanathann CK, Koerner J. Transfer function synthesis as a ratio of two complex
interpretation of frequency-response analysis. IEEE Trans Power Deliv polynomials. IEEE Trans Autom Control 1963;9:56–8.
2011;26(4):2705–17. http://dx.doi.org/10.1109/TPWRD.2011.2164424. ISSN [35] Abetti PA. Transformer models for the determination of transient voltages.
0885-8977. Power Ap Syst Part III. Trans Am Inst Electr Eng 1953;72(2):468–80. http://
[19] Welsh J, Rojas C, Mitchell S. Wideband parametric identification of a power dx.doi.org/10.1109/AIEEPAS.1953.4498656. ISSN 0018-9510.
transformer. In: Universities power engineering conference, 2007. AUPEC [36] Wilcox D, Hurley W, McHale T, Conlon M. Application of modified modal
2007. Australasian; 2007. p. 1–6. http://dx.doi.org/10.1109/AUPEC.2007. theory in the modelling of practical transformers. IEE Proc C Gener Transm
4548053. Distrib 1992;139(6):513–20. ISSN 0143-7046.
[20] Pintelon R, Schoukens J. System identification: a frequency domain approach. [37] de Leon F, Semlyen A. Complete transformer model for electromagnetic
2nd ed. IEEE Press; 2012. transients. IEEE Trans Power Deliv 1994;9(1):231–9. http://dx.doi.org/
[21] Oliveira GHC, Maestrelli R, Rocha ACO. An application of orthonormal basis 10.1109/61.277694. ISSN 0885-8977.
functions in power transformers wide band modeling. In: IEEE international [38] Mitchell SD. Power transformer modelling to support the interpretation of
conference on control and automation, 2009; 2009. p. 831–6. frequency response analysis. PhD thesis. School of Electrical Engineering and
[22] Oliveira GHC, Mitchel SD. Comparison of Black-Box modeling approaches for Computer Science, University of Newcastle, Australia; 2011.
transient analysis: a GIS substation case study. In: Proc of the 2013 [39] Rohrer RA. Circuit theory: an introduction to the state variable
international conference on power system transients. CA: Vancouver; 2013. approach. McGraw-Hill; 1970.
[23] Brozio C, Vermeulen H. Wideband equivalent circuit modelling and parameter [40] Mechanical condition assessment of transformer windings using Frequency
estimation methodology for two-winding transformers. IEE Proc Gener Response Analysis (FRA), ELECTRA-CIGRE WG A2.26 Report 342 237.
Transm Distrib 2003;150(4):487–92. http://dx.doi.org/10.1049/ip- [41] Deshpande MV. Electrical power system design. McGraw-Hill; 1985.
gtd:20030422. ISSN 1350-2360. [42] Singh S. Electric power generation transmission and distribution. 2nd ed. PHI
[24] Ragavan K, Satish L. Construction of physically realizable driving-point Learning Pvt. Ltd; 2008.
function from measured frequency response data on a model winding. IEEE