See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/260670463

Calculation of Inrush Currents – Benchmarking of Transformer Models

Conference Paper · January 2011

CITATIONS READS
12 2,139

4 authors, including:

Nicola Chiesa Hans Kristian Høidalen

Equinor Norwegian University of Science and Technology
37 PUBLICATIONS   679 CITATIONS    119 PUBLICATIONS   1,945 CITATIONS   

SEE PROFILE SEE PROFILE

Manuel Martinez
Électricité de France (EDF)
28 PUBLICATIONS   248 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Zero sequence impedance modeling of transformers View project

ProDig - Power system protection and control in digital substations View project

All content following this page was uploaded by Hans Kristian Høidalen on 24 March 2015.

The user has requested enhancement of the downloaded file.

 Calculation of Inrush Currents –
Benchmarking of Transformer Models
N. Chiesa, H. K. Høidalen, M. Lambert, M. Martı́nez Duró

Abstract–The paper investigates the behavior of four trans- w3 w4

former models in electromagnetic transients programs (EMTP)
when calculating inrush currents due to shell-type transformers w2 yoke w2/w1=0.5082

mid. limb

out.limb
energization. These models are single-phase Saturable Trans- l2
former (STC), BCTRAN, Hybrid Transformer (XFMR) and
w3/w1=0.5082
l3 l4 w4/w1=0.5082
UMEC. The measurements of a real transformer energization l1
performed by EDF in France are used to evaluate the accuracy w1 l2/l 1=1
leg l3/l 1=0.6939
of the simulation results provided by each model. For the first
periods of the inrush, the results show that only a topologically l4/l 1=0.6939
correct transformer model as the hybrid transformer model is
able to reproduce the current amplitude in the three phases and
the overall wave shape. BCTRAN, STC and UMEC may predict
the first inrush current peak, but the amplitudes in the other
two phases are far from the measurements. The current decay Fig. 1. Shell-form core and relative core dimensions.
in the 2 seconds after the energization is poorly predicted by
all of the models underlining a poor losses representation in the
transformer model and in the rest of the network. II. TRANSFORMER MODELS
Keywords: Power transformers, inrush currents, measure- The test object is a 96 MVA four-winding auxiliary trans-
ments, simulations, ATP, EMTP-RV, PSCAD/EMTDC. former with a shell-type core. The transformer is energized
from its primary 400 kV Y-coupled winding. The secondary
I. I NTRODUCTION side consists of three 6.8 kV delta-coupled windings.
Available data for the modeling of the transformer are the
T RANSIENT energization studies of transformers are im-
portant for power system reliability consideration and
particularly for the design of power system restoration strate-
standard test report shown in Table I, relative core dimensions
and the air-core inductance. The relative core dimensions are
gies after blackout. During power system restoration, the shown in Fig. 1. The air-core inductance provided by the
supplying network exhibits high harmonic impedance; there- manufacturer is 1.4 H referred to the HV side. The tap-changer
fore, the inrush currents may generate dangerous temporary position is set to tap 1, corresponding to 410 kV. Since there is
overvoltages that could damage the transformers or other no tap-changer option in the selected models, the tap-changer
equipment [1]. voltage is specified as nominal voltage.
However, the standard available models in EMT-type pro- The three secondary windings are merged into one equiva-
grams could in many cases have insufficient accuracy and lent winding in order to be able to implement the transformer
capabilities. Uncertainties are in estimating the residual fluxes, in models that are limited to two- or three-windings. This
the behavior in extreme saturation, the dependence of core is plausible since the three windings are identical and are
structure, and the influence of multi-windings designs [2]. considered as connected in parallel.
In this paper, the simulation results obtained for various The target transformer is represented with the Saturable
standard available transformer models in ATP, EMTP-RV and Transformer Component (STC), BCTRAN, Hybrid Trans-
PSCAD/EMTDC are compared to the measured inrush current
of a real energization performed at EDF in France. TABLE I
AUXILIARY T RANSFORMER ’ S T EST R EPORT

This work received partial financial support from the KMB project “Electric Main data [kV] [MVA] [A] Coupling
power systems for subsea processing and transportation of oil and gas” HV 400±2.5% 96 138.6 YN
financed by the Norwegian Research Council Petromaks programme and LV 6.8 96 8151 d11
industry partners (www.subseapowersupply.com).
N. Chiesa is with SINTEF Energy Research, Trondheim, Norway (e-mail of Open-circuit E0 [%] [MVA] I0 [%] P0 [kW]
corresponding author: nicola.chiesa@sintef.no). LV 90 96 0.171 75.25
H. K. Høidalen is with the Norwegian University of Science and Technology, 100 96 0.421 101.5
N-7491 Trondheim, Norway (hans.hoidalen@elkraft.ntnu.no). 105 96 0.634 117.0
M. Lambert is with Ecole Polytechnique de Montréal, Canada (math- 110 96 0.971 137.8
ieu.lambert@polymtl.ca). 115 96 1.601 165.6
M. Martı́nez Duró is with Electricité de France, R&D Division, Clamart, Short-circuit [kV] [MVA] Ek [%] Pk [kW]
France (manuel.martinez@edf.fr). HV/LV 410/6.8* 96 5.59 332.95
Paper submitted to the International Conference on Power Systems * The test with all three LV windings paralleled and shorted is
Transients (IPST2011) in Delft, the Netherlands June 14-17, 2011 used.
former (XFMR), and UMEC models. Among these, only the A’ A B C
XFMR model can handle a shell-type core topology directly. B’
For comparison, a triplex core equivalent is considered for both Limb-AB Limb-BC
C’
XFMR and UMEC cases. Single-phase transformer models
cannot account for differences in direct- and zero-sequence
behavior of three-phase single-core transformers. However, a
triplex representation may be enough if a delta winding is

Yoke

Yoke

Yoke
Leg

Leg

Leg
Out

Out
present and rules the zero-sequence behavior of the trans-
former [3]. Furthermore, for shell-type and four- or five-limb
cores, the core provides an unwounded return path for the
phase flux. Therefore, differences between direct- and zero-
sequence behaviors are less significant than for three-legged
cores.
Fig. 2. Shell-form core model in the Hybrid Transformer.

A. Saturable transformer model

The Saturable Transformer Component (STC) [2], [4] is it offers better control over the magnetization characteristic
a two- and three-winding single phase transformer model. It through external nonlinear inductances.
is the nonlinear version of the classic Steinmetz model [5]. In this paper, the core representation is connected to the LV
It represents the short-circuit impedances between windings, winding terminal in a delta configuration.
the load and magnetization losses, and the nonlinear inductive
magnetization. When it is used to model three-phase trans- C. Hybrid transformer model
formers, the inter-phase magnetic coupling is not represented The Hybrid Transformer model [9]–[11] is an engineering
in this model. STC is based on an equivalent star circuit where transformer model based on limited input data, available in
the core representation can be connected either to the star ATPDraw. The duality-derived electrical circuit of a shell-
point or to the terminal of the winding closest to the core. form core is shown in Fig. 2. In Test Report mode, the
This model may suffer from instability problem when three model requires data like the one given in Table I and the
windings are modeled and the core is connected to the star magnetization curve is fitted to a modified Frolich equation
point [4], [6]. As proven in [7], this model is not valid for more [11] internally.
than three windings per phase. The STC model is implemented The relative core dimensions must be entered according to
in both ATP and EMTP-RV. This model is available in PSCAD Fig. 1, except that the relative areas must be multiplied by a
as transformer model based on “classical modeling approach factor of 2. Since relative widths in Fig. 1 are equal for the
(not ideal)”. outer legs, yokes and middle limbs, the middle phase’s polarity
In this paper, the core representation is connected at the has to be reversed to get 1 p.u. flux in all core parts, as shown
star point, in-between the separated leakage reactances and in Fig. 2 (shell-form type B in ATPDraw version >5.7).
winding resistances (the short-circuit impedance). While in In the present case, the air-core inductance is specified
ATP and EMTP-RV the nonlinear inductances are modeled and the final slope of the magnetization characteristic is
with piecewise-linear characteristic, in the PSCAD model approximated by [12]:
they are modeled as a two-slope characteristic. The winding
resistance and leakage inductance are equally split on a p.u. L∞ ≈ Lair−core − (1 + k)LHL (1)
base between primary and secondary winding (0.5 splitting
where LHL is the short circuit inductance from the test report
factor) [4].
and k is a calibration factor representing the leakage between
the inner winding and the core assumed to be 0.5 in the Hybrid
B. BCTRAN model Model. This gives L∞ = 0.8295 mH referred to the LV side.
The BCTRAN model [8] is a n-phase transformer model
where inter-winding coupling can be taken into account. The D. UMEC model
model is linear and assumes phase symmetry. It consists The UMEC transformer model [13], [14] is based on the
of a coupled RL or RL−1 matrix representing short-circuit concept of a unified magnetic equivalent circuit. A normalized
impedances between windings, load losses at rated frequency core is used in order to remove the requirement of design data;
and optionally linear inductive magnetization. Nonlinear mag- only relative dimensions are required.
netization and core loss components may be added externally, The magnetic network is derived from the transformer
usually at the terminals of the winding nearest to the core. core topology. Three-limb, five-limb, and three-phase bank
BCTRAN is implemented in EMTP-RV, ATP. In PSCAD is (triplex) transformer core constructions are possible config-
available as transformer model based on “classical modeling urations in the model. The core saturation characteristic is
approach (ideal)”. In PSCAD, the classical transformer model specified directly as a rms I-V curve. The magnetic network
allows only for a two-slope saturation curve. The EMTP-RV representing the core and leakage inductances is described
and ATP implementation of BCTRAN is investigated here as with a matrix formulation using a permeance matrix. Both the
magnetic coupling between windings of different phases and 1.3
the coupling between windings of the same phase are taken
1.2
into account.
Winding and core losses are not included in the magnetic 1.1

Flux linkage [pu]

circuit and are represented by an equivalent linear resistance
1
at the winding terminals. Load losses are equally divided on a
p.u. base and represented by linear series resistances connected 0.9

at one terminal of each winding. Core losses are assumed XFMR shell (leg)
0.8 XFMR shell (yoke)
linear and equally divided on a p.u. base between primary XFMR shell (m&o limb)
and secondary windings. The use of a magnetic network does 0.7 XFMR triplex
BCTRAN & STC
not allow a simple representation of topological core losses, 0.6
0 1 2 3 4 5 6 7
therefore they are represented by linear shunt resistances Current peak [%]
connected at the terminals of each winding.
The UMEC model is implemented in PSCAD. Fig. 3. Saturation curves.

E. Saturation curves and open-circuit model responses 10

2

XFMR shell
One of the most critical steps in the creation of a transformer XFMR triplex
model for inrush current calculation is the construction of the BCTRAN
STC
saturation curves accounting for the nonlinear behavior of the 10
1
UMEC (orig)

Current [%]
core. These are calculated to match the open-circuit test data UMEC (cond)
Test Report
of Table I.
A routine for the conversion from rms to peak value [15] 0
10
is used for the generation of the saturation curve required by
BCTRAN and STC. An additional point is defined beyond the
last calculated value to set the saturated inductance according
−1
to the air-core inductance: 10
0.8 0.9 1 1.1 1.2 1.3
Voltage [pu]
Lsat = Lair−core − LHL (2)
(a) No-load current.
and resulting in the curve shown in Fig. 3. For the STC model, 220
the saturated inductance is calculated by removing only a XFMR shell
200
fraction of the total leakage inductance LHL , corresponding XFMR triplex
BCTRAN
to the fraction of leakage placed on the primary side: 180 STC
UMEC (orig)
Losses [kW]

160
Lsat = Lair−core − ξ LHL (3) UMEC (cond)
Test Report
140
where ξ is the fraction of total leakage placed on the primary 120
side.
100
In XFMR, the conversion from open-circuit test data to
saturation curve is performed by a sophisticated algorithm 80

based on a curve fitting approach. The resulting saturation 60

0.8 0.9 1 1.1 1.2 1.3
curves are shown in Fig. 3 for the triplex and shell topologies. Voltage [pu]
For the shell-core transformer case, four saturation curves are (b) No-load losses.
calculated by the XFMR model based on the relative area and
length of each limb. One may see in Fig. 3 that for the same Fig. 4. Open circuit test of the transformer models compared to input test
report values.
flux-linkage value the curve fitting approach used in XFRM
leads to lower magnetization currents than those obtained with
curves where Lsat is located right after the last magnetization
point. For all models, the core losses are modeled with linear resis-
Similarly to XFMR, the UMEC model has also the ability tor to match the losses at rated excitation. Nonlinear resistors
to create the saturation curves from open-circuit data. In Fig. 4 could be used to match more accurately the nonlinear loss
the curves labeled “UMEC (orig)” are with direct input of test behavior; however, their transient performances is somewhat
report data according to Table I. As also pointed out in [10] uncertain [16].
the input has to be conditioned to match the no-load test data. Open-circuit tests of the transformer models are performed
In Fig. 4 the curves labeled “UMEC (cond)” are obtained by to ensure the accurate match with the input data of Table I.
specifying the saturation
p curve labeled “BCTRAN √ & STC” With the exception of the original UMEC model, all models
in Fig. 3, scaled by 3/2 for the current and ω/ 2 for the can accurately match the input no-load current values as shown
flux-linkages. in Fig. 4(a). The area beyond the last known point shows that
some uncertainties and discrepancies can be observed with the 10
different models. Fig. 4(b) shows the simulated no-load losses
versus the test report input values. All models can match the
5
losses at rated excitation and again some differences can be

Voltage LV [kV]
noticed at higher excitation.
0
III. TEST CASE
A. Network topology −5

The field test network used to benchmark the transformer

models is represented in Fig. 5. The unloaded auxiliary trans- −10
former at the target plant is energized from a supplying 2.6 2.65 2.7 2.75
Time [s]
network consisting in a 315 km 400 kV overhead line fed by
(a) Voltages at the LV-side of the transformer.
a source plant. The source plant’s step-up transformer consists
of three single-phase units rated 550 MVA each. The source 30
plant’s auxiliary transformer is identical to the one at the
target plant and is loaded. The overhead line is double-circuit 20

Flux−linkage LV [Wb−t]
(albeit only one is energized) and untransposed. Two iron-core
10
compensation reactors of 100 MVAr each are present near the
target plant. 0
The network supplying the energized transformer is mod-
eled according to the findings in [17]. The generator is −10
modeled by an ideal voltage source behind an impedance
−20
accounting for the subtransient reactance and the armature
resistance. The step-up transformer is modeled with single- −30
phase transformer models with nonlinear core. The overhead 2.6 2.65 2.7 2.75
Time [s]
line is modeled by cascaded PI cells calculated at the power
frequency (50 Hz) from the conductors’ geometrical and phys- (b) Flux-linkages (referred to the LV-side) and final residual fluxes.
ical characteristics. The compensation reactances are modeled Fig. 6. Measurements of de-energization transient.
as constant inductances (they are linear in the voltage range
of the measurements) with parallel resistances accounting for
losses. the energized transformer and switching times of the circuit-
breaker. The switching times were obtained by identifying
B. Measurements sudden changes of the voltages at the terminals of the trans-
former. A delay in the poles of the breaker is observed from
The measurements include LV-side voltages at the de-
the measurements where phase B and C close 11.6 ms after
energization of the transformer and HV-side voltages and cur-
phase A.
rents at the energization, as well as induced LV-side voltages.
Fig. 6 highlights a rather long ring-down transient lasting
The flux-linkages are calculated as the time integral of the
a few periods, supposedly due to large transformer capaci-
LV-side voltages. The measured waveforms and the calculated
tances [18], resulting in relatively low final residual fluxes.
flux-linkages are shown in Figs. 6-7. Throughout the paper, the
The residual fluxes estimated from the flux-linkages after the
colors blue, green, and red are used to identify the waveforms
ringdown transient are 6.3%, 5.5%, and -10.2% of the rated
for phase A, B and C, respectively.
flux, respectively for the three phases. Due to their low values,
Inrush currents and subsequent harmonic overvoltages are
they have been neglected in the simulations.
highly dependent on two initial conditions: residual fluxes of

IV. R ESULTS
A. Inrush current waveforms
Figs. 8-13 show the first four periods of the inrush current
waveforms calculated with the different models. Each curve is
compared with the measured inrush current.
In general, all the models based on a triplex or single
phase representation give similar results in terms of current
amplitude and shape of the waveforms as shown in Fig. 8,
Fig. 9, Fig. 11 and Fig. 13. STC, BCTRAN, and UMEC (with
conditioned data) give very similar result as they are based
on the same saturation curve (Fig. 3). In XFMR with triplex
Fig. 5. Field test case. core, the current is slightly lower due to a higher saturation
 100 150

100
50
50
Currents HV [A]

Current HV [A]
0
0

−50
−50

−100
−100
−150

−150 −200
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0.46 0.48 0.5 0.52 0.54 0.56 0.58
Time [s] Time [s]

(a) Inrush current for the target transformer.

Fig. 8. Inrush current with STC. Color lines: simulation. Black lines:
Measurements.
100
150

50 100
Currents HV [A]

50
0

Current HV [A]
0
−50
−50

−100 −100

−150
−150
0.46 0.48 0.5 0.52 0.54 0.56 0.58
Time [s] −200
0.46 0.48 0.5 0.52 0.54 0.56 0.58
(b) Close-up of measured inrush current. Time [s]

10
Fig. 9. Inrush current with BCTRAN. Color lines: simulation. Black lines:
Measurements.
100
5
Line Voltage LV [kV]

50

0
Current HV [A]

−5 −50

−10 −100
0.46 0.48 0.5 0.52 0.54 0.56 0.58
Time [s]
−150
(c) Induced voltages on the transformer LV-side. 0.46 0.48 0.5 0.52 0.54 0.56 0.58
Time [s]
Fig. 7. Measurements of the energization transient.
Fig. 10. Inrush current with XFMR (shell core). Color lines: simulation.
Black lines: Measurements.
curve. The inrush currents calculated with these models are
relatively close to the first peak of phase C. However, they fail
voltages are less sensitive than currents and similar waveforms
to reproduce the other phases’ first peak and the overall shape
are obtained with all the analyzed models.
of the waveforms. The simulation results obtained with the
XFMR model using a topologically correct shell-core model
can match both the inrush current and waveform shape in the B. Current Decay
first periods of the inrush transient, as shown in Fig. 10. The absolute value of the inrush current envelope for the
Fig. 12 shows that the UMEC model does not succeed to first two seconds after the energization of the transformer is
accurately represent inrush transient when no-load data are not shown in Fig. 15 for the considered models, as well as for the
conditioned. measurements. All the models can poorly predict the current
Fig. 14 shows the induced voltage on the LV-side and com- decay. Curves are all parallel reflecting that the models have
pares the simulation result obtained with XFMR (shell) with similar losses’ representation (linear winding and core losses).
the measurements. Since there is no significant overvoltage, The only exception is BCTRAN as it has a much larger decay
 100 10

50
5

Line Voltage LV [kV]

Current HV [A]

0
0
−50

−5
−100

−150 −10
0.46 0.48 0.5 0.52 0.54 0.56 0.58 0.46 0.48 0.5 0.52 0.54 0.56 0.58
Time [s] Time [s]

Fig. 11. Inrush current with XFMR (triplex). Color lines: simulation. Black Fig. 14. Induced LV-side voltages with XFMR (shell core). Color lines:
lines: Measurements. simulation. Black lines: Measurements.
100

effect of the switching instant, the inrush current patterns for

50
the case of zero residual flux are presented in Fig. 16 and
Fig. 17. These patterns are obtained by varying the switching
Current HV [A]

0
in instant with the introduction of a delay between 0 and
20 ms. Then the maximum inrush current is recorded and
−50
plotted as a function of the switching in delay. The time t=0
is equivalent to the test case switching instant.
−100
Fig. 16 shows the inrush current pattern when a delay of
11.6 ms is maintained between the first pole to close (phase
−150
0.46 0.48 0.5 0.52 0.54 0.56 0.58 A) and the other two poles. From this figure, it is evident that
Time [s]
the analyzed test case (in t=0) is close to the maximum inrush
Fig. 12. Inrush current with UMEC (original data). Color lines: simulation. current for phase C, but is less than half the most severe inrush
Black lines: Measurements. current that can be experienced by this transformer.
The effect of a different delay between the circuit breaker
150
poles is also shown in Fig. 16. The thinner lines show the
100 inrush pattern for poles delay of 10.6 and 12.6 ms. The inrush
current first peak is highly sensitive to a variation of this
Current HV [A]

50 parameter.
0
Fig. 17 compares the inrush pattern obtained with XFMR
(shell core) and BCTRAN for simultaneous three-pole closing.
−50 The two models predict a similar maximum inrush current for
−100
the transformer of about 220 A (1.15 p.u.). While BCTRAN
predicts an equal pattern for the three phases, XFMR with its
−150 topologically-correct shell-core representation can calculate a
0.46 0.48 0.5 0.52 0.54 0.56 0.58 lower current for the middle phase. While the maximum inrush
Time [s]
current estimated by the two models is fairly similar, at specific
Fig. 13. Inrush current with UMEC (conditioned data). Color lines: switching instants the difference can be up to 200%.
simulation. Black lines: Measurements.
V. D ISCUSSION
than the other models. In BCTRAN, the core is connected The transformer models analyzed in this paper can estimate
to the LV winding. Therefore, current flows in both HV and the first peak of the inrush current with good accuracy. The
LV winding resistances. This gives higher losses and faster main reasons are the knowledge of the air-core inductance
current decay. Hence, it is incorrect as only the current’s zero- (provided by the manufacturer) and the availability of an
sequence component of the current should circulate in the extensive no-load test report (with maximum induction level
LV winding; therefore, the contribution to the inrush current of 115%).
damping is minimal. The knowledge of the air-core inductance allows to treat
the final slope of the saturation curve equally in all models.
At 115% excitation, the current is only few per-cent of the
C. Inrush patterns rated current. However, during inrush current transients the
The simulations presented until now are representative of current peaks are in excess of the rated current. The unknown
a single switching sequence. In order to better investigate the area beyond the last known point in the saturation curve
 80 300
Measurement
XFMR Shell
70 200

Maximum inrush current [A]

XFMR Triplex
BCTRAN
60
Current HV [A]

STC 100
UMEC (cond)
50 0

40 −100

30 −200

20 −300
0.5 1 1.5 2 2.5 0 5 10 15 20
Time [s] Switching in instant [ms]
(a) Phase A
Fig. 16. Inrush current pattern with XFMR shell core. Continuous line:
160 11.6 ms delay. Dotted line: 10.6 ms delay. Dashed lines: 12.6 ms delay.

140
300
120
Current HV [A]

200

Maximum inrush current [A]

100

80 100

60 0

40
−100
20
0.5 1 1.5 2 2.5 −200
Time [s]

(b) Phase B −300

0 5 10 15 20
Switching in instant [ms]
180

160 Fig. 17. Inrush current pattern. Continuous lines: XFMR shell core. Dotted
lines: BCTRAN.
140
Current HV [A]

120

100 (STC), at the LV-terminals (BCTRAN) or at internal points

80
(UMEC and XFMR) makes a difference in both the maximum
inrush and the decay. The shell-form core’s geometry is not
60
too far away from that of a triplex core. However, the use of a
40 topologically-correct shell-form core representation influences
20 the waveform for the currents by considering asymmetries that
0.5 1 1.5 2 2.5
Time [s] exist between phases. This can be an important factor to take
into consideration when overvoltages have to be analyzed, as
(c) Phase C
they depend both on the current peak and the waveform.
Fig. 15. Inrush current decay, 2 s. The sensitivity of the simulation results to the switching
instant is an important issue. Both switching instant and delay
between each pole have to be accurately considered as they
is therefore critical for the accuracy of the inrush current greatly influence the inrush current magnitude.
estimation. As outlined in [12], [19], an accurate representation Residual flux is quite low as seen from the measurements
of the saturation and its final slope is of uttermost importance and has not been taken into account in the analysis to avoid
for the simulation of transformer energization. including an additional complicating factor. Initialization of
The main difference between the analyzed models is due to nonlinear inductors, especially when included in topologically-
the approach used for the estimation of the saturation curve. correct core models, is still an issue and needs to be further
In the STC, BCTRAN and UMEC models, the slope after investigated.
the last test report value (115%) is constant and equal to Lsat , In order to improve the estimation of the inrush current
while in the XFMR model, the slope after this point decreases decay, the loss modeling has to be greatly enhanced. This may
smoothly to reach asymptotically the value of Linf (Frolich be achieved with the use of frequency dependent winding re-
equation fitting). sistances and the representation of nonlinear and topologically
In addition, the modeling approach of the core has shown to correct core losses. As shown in [19], the frequency dependent
be important. The connection point of the core at the star-point winding losses are important for transients between 0.1 Hz
and 3 kHz, while an accurate representation of hysteresis and [9] B. A. Mork, F. Gonzalez, D. Ishchenko, D. L. Stuehm, and J. Mitra,
iron losses may be unnecessary. Hence, a frequency-dependent “Hybrid transformer model for transient simulation: Part I: development
and parameters,” IEEE Trans. Power Del., vol. 22, no. 1, pp. 248–255,
winding-resistance model is required to represent the currents Jan. 2007.
decay more accurately. The loss modeling in the other network [10] H. K. Høidalen, B. A. Mork, F. Gonzalez, D. Ishchenko, and N. Chiesa,
components is also important. “Implementation and verification of the hybrid transformer model in
ATPDraw,” Electr. Power Syst. Res., vol. 79, no. 3, pp. 454 – 459, Mar.
2009, special Issue: Papers from the 7th International Conference on
VI. C ONCLUSION Power Systems Transients (IPST).
[11] H. K. Høidalen, N. Chiesa, A. Avendaño, and B. A. Mork, “De-
Several transformer models have been evaluated in their velopments in the hybrid transformer model – Core modeling and
capability to accurately predict inrush currents due to the optimization,” in IPST’2011, 2011.
[12] N. Chiesa, B. A. Mork, and H. K. Høidalen, “Transformer model for
energization of three-phase shell-core transformers. These inrush current calculations: Simulations, measurements and sensitivity
models are readily available in several simulation softwares. analysis,” IEEE Trans. Power Del., vol. 25, no. 4, pp. 2599–2608, Oct.
The evaluation consists in comparing the field measurements 2010.
[13] W. Enright, O. B. Nayak, G. D. Irwin, and J. Arrillaga, “An electro-
for a real transformer energization case performed at EDF and magnetic transients model of multi-limb transformers using normalized
the simulation results provided by each model. core concept,” in IPST’97 - International Conference on Power System
The simulation results show that a topologically correct Transients, Seattle, Washington, Jun. 1997, pp. 93–98.
[14] W. Enright, O. Nayak, and N. Watson, “Three-phase five-limb unified
core model is required if a higher accuracy is desired when magnetic equivalent circuit transformer models for PSCAD V3,” in
simulating highly nonlinear and unbalanced electromagnetic IPST’99 - International Conference on Power System Transients, Bu-
transients. Good accuracy in the estimation of the inrush dapest, Hungary, Jul. 1999, pp. 462–467.
[15] N. Chiesa and H. K. Høidalen, “Analytical algorithm for the calculation
current’s first peak can also be achieved with equivalent of magnetization and loss curves of delta connected transformers,” IEEE
models if an accurate representation of the saturation and its Trans. Power Del., 2010, accepted for publication, TPWRD-00589-2009.
final slope is employed. [16] N. Chiesa and H. K. Høidalen, “Hysteretic iron-core inductor for
transformer inrush current modeling in EMTP,” in PSCC 2008 - 16th
In general, the investigated models manage to represent the Power Systems Computation Conference, Glasgow, Scotland, Jul. 2008.
inrush currents very well. The main reason is that no-load test [17] M. Martı́nez Duró, “Damping modeling in transformer energization
report up to 115% excitation and air-core inductance were studies for system restoration,” in PowerTech 2009, Bucarest 2009.
[18] N. Chiesa, A. Avendaño, H. K. Høidalen, B. A. Mork, D. Ishchenko, and
available. Hence, it is important for transformer owners to A. P. Kunze, “On the ringdown transient of transformers,” in IPST’07
request such parameters when purchasing their units. - International Conference on Power System Transients, no. IPST-229,
On the other hand, none of the examined models is able to Lyon, France, Jun. 2007.
[19] “Guidelines for representation of network elements when calculating
match the current decay of the measurements. This issue would transients,” CIGRE WG 33.02, 1990.
require further investigation. Particularly, the transformer core
and load losses modeling would need to be improved in
order to take into account nonlinear behavior and frequency
dependency. The loss modeling of other network components
is also important in the determination of the inrush current
decay.

R EFERENCES
[1] D. Povh and W. Schultz, “Analysis of overvoltages caused by trans-
former magnetizing inrush current,” IEEE Trans. Power App. Syst., no. 4, Nicola Chiesa was born in Italy in 1980. He received the MSc degree in
pp. 1355–1365, Jul. 1978. electrical Engineering from Politecnico di Milano in 2005, and the PhD degree
[2] J. A. Martinez and B. A. Mork, “Transformer modeling for low- and from the Norwegian University of Science and Technology in 2010. He is now
mid-frequency transients - a review,” IEEE Trans. Power Del., vol. 20, a research scientist at SINTEF Energy Research.
no. 2 II, pp. 1625–1632, 2005.
[3] M. R. Iravani, A. K. S. Chaudhary, W. J. Giesbrecht, I. E. Hassan,
A. J. F. Keri, K. C. Lee, J. A. Martinez, A. S. Morched, B. A. Mork,
M. Parniani, A. Sharshar, D. Shirmohammadi, R. A. Walling, and D. A.
Woodford, “Modeling and analysis guidelines for slow transients. III. Hans Kr. Høidalen was born in Norway in 1967. He received his MSc and
The study of ferroresonance,” IEEE Trans. Power Del., vol. 15, no. 1, PhD from the Norwegian University of Science and Technology in 1990 and
pp. 255–265, 2000. 1998 respectively. He is now a professor at the same institution with a special
[4] H. W. Dommel and et.al., Electromagnetic Transients Program Refer- interest of electrical stress calculations and modeling.
ence Manual (EMTP Theory Book). Portland, OR: Prepared for BPA,
Aug. 1986.
[5] C. P. Steinmetz and E. J. Berg, Theory and calculation of alternating
current phenomena. New York: Electrical World and Engineer, inc.,
1897. Mathieu Lambert was born in Canada in 1984. He received his MSc from
[6] T. Henriksen, “How to avoid unstable time domain responses caused the Ecole Polytechnique de Montréal in 2009. He is now a PhD candidate at
by transformer models,” IEEE Trans. Power Del., vol. 17, no. 2, pp. the same university.
516–522, Apr. 2002.
[7] L. F. Blume, Transformer engineering : a treatise on the theory,
operation, and application of transformers, 2nd ed. New York, N.Y.:
John Wiley and Sons, Inc., 1951.
[8] V. Brandwajn, H. W. Dommel, and I. I. Dommel, “Matrix representation Manuel Martı́nez Duró was born in Spain in 1975. He received the electrical
of three-phase n-winding transformers for steady-state and transient engineering degree from the Universitat Politècnica de Catalunya, Barcelona,
studies,” IEEE Trans. Power App. Syst., vol. PAS-101, no. 6, pp. 1369 Spain, in 1999. He joined Electricité de France, R&D Division, France, in
–1378, June 1982. 2001.

View publication stats