Implementation and verification of the Hybrid

Transformer model in ATPDraw

Hans K. Høidalen, Bruce A. Mork, Francisco Gonzalez, Dmitry Ishchenko, Nicola Chiesa

Abstract-- The paper documents a new transformer model in CHA-Xa/2

ATPDraw called XFMR. This model handles 3-phase 6
transformers with two or three windings. Autotransformers and CH-GND/2 L3 CX-GND/2
NH:NX NX:NX a
all Wye and Delta couplings are supported. The model includes A L4
CHA-HB/2 RH( f ) Zl RX( f )
an inverse inductance matrix for the leakage description,
’ a’
optional frequency dependent winding resistance, capacitive A’ C
Zy
coupling, and a topologically correct core model (3- and 5-legged) CH-GND/2 HA-HB/2
NH:NX NX:NX b
with individual saturation and losses in legs and yokes. Three
B L3
different sources of data are supported; typical values, standard CHB-HC/2 RH( f ) Zl RX( f )
test reports, and design information. The hybrid model XFMR is ’ b’
B’ Zy
compared to the UMEC model in PSCAD showing good CH-GND/2 CHB-HC/2
agreement at rated, stationary operation, but considerable NH:NX NX:NX c
differences in transient situations. Both models need development C RH( f ) RX( f )
Zl
to reproduce all switching transient behaviors properly. L4 ’ c’
CH-GND/2 C’ L3 CX-GND/2
Keywords: Transformer modeling, saturation, leakage
inductance, winding resistance, capacitance, topologically correct Fig. 1. Electric model of the Hybrid Transformer [4], 2-windings (H and X),
core. 3-phases, 3-legged core.

I. INTRODUCTION Modeling of the transformer can be based on three different

T HE transformer is an essential component in power

systems but the standard models used to represent it in
transient analysis are rather poor. Several approaches are
data sources; typical values, test report, and design
information. The three sources can be selected independently
for resistance, inductance, capacitance, and core. Test report
documented in the literature [1] and this paper focuses on input is based on standard open and short circuits tests, with
models with topologically correct cores. A model called the capacitance measurements as an additional option. This is the
Hybrid Transformer model [2]-[4] is recently implemented in normal choice of data source for existing transformers. Design
ATPDraw and is there called XFMR. The characteristics and data requires the geometry and material parameters of the
implementation of this model is outlined in this paper and its windings and the core. Such data are rarely available so this
response is compared to the UMEC model [5], [6] found in option is more for research purposes. The Typical value option
PSCAD. UMEC is chosen since it is based on similar core uses available text book tabulated values of leakage
geometry data as XFMR. The purpose is not to show which impedance, copper and core losses, and magnetizing current to
model is the best but rather to point out the need for improved estimate a transformer model. This is suitable when the
transformer models for switching transient studies. transformer is not purchased yet, or data is unavailable in an
initial study. However, such model must be used with caution.
II. MODELING A. Leakage inductance
The modeling of the transformer is based on the magnetic The leakage inductance is modeled with an inverse
circuit transformed to its electric dual [2], [3]. The leakage and inductance matrix (A-matrix). The matrix has dimension
main fluxes are then separated into a core model for the main (nw+1)⋅np where nw is the number of physical windings, the
flux and an inverse inductance matrix for the leakage flux. The core is connected to the nw +1 winding, and np is the number
copper losses and coil capacitances are added at the terminals of phases [2], [3], [7]. The coupling (auto, Y, D) and phase
of the transformer. The resulting electrical circuit is shown in shift is produced directly in the A-matrix. All possible phase
Fig. 1. Only standard EMTP elements are used. shifts are supported. The A-matrix has the following structure
for a three-winding, three-phase transformer:
This work has received financial support from Bonneville Power
P S T C
⎡ 11 a14 ⎤
Administration, the Spanish Secretary of State of Education and Universities,
A B C
⎡ Aw 0 0⎤ ⎢a a22 a23 a24 ⎥⎥
the Fulbright, the Norwegian Research Council, and SINTEF Energy
a a a

A = ⎢⎢ 0 Aw 0 ⎥⎥ where Aw = ⎢ 21
Research. 12 13
H. K. Høidalen and N. Chiesa are with the Norwegian University of Science (1)
⎢ a31 a32 a33 a34 ⎥
⎢⎣ 0 0 Aw ⎥⎦
and Technology, Trondheim-Norway, B. A. Mork, F. Gonzalez, and D.
⎢ ⎥
Ishchenko are with Michigan Technological University, Houghton-USA,
⎣ a41 a42 a43 a44 ⎦
(e-mail of corresponding authors:
[email protected], [email protected]).
where ABC are the three phases and PSTC stands for primary,
Presented at the International Conference on Power Systems
Transients (IPST’07) in Lyon, France on June 4-7, 2007 secondary, tertiary, and the core (nw+1) winding.
 The A-matrix is assumed to have no mutual coupling between the windings in the case of a two-winding
between the phases. The entire zero-sequence effect is transformer. In the case of a 3-winding transformer the
modeled in the attached core. The Aw-matrix is established traditional star-equivalent approach is used.
according to the EMTP Theory Book [8] chapt. 6.4, and In the case of an auto-transformer the short circuit
section 5.2.4 p. 31 in [2]. resistances are recalculated according to the power balance
1) Typical values used in [7]. The approach used for reactances (from the
Theory Book [8]) did not work out for the resistances.
The leakage reactance is established from [9] using the
3) Design data
The user can specify the winding conductivity σ, the
lowest value in the typical range. In the case of a three-
winding transformer the leakage reactance (in pu) between the
inner and outer winding is approximated as the sum of the equivalent cross section A of each turn, the average length l of
other two. In this case it is assumed that the medium voltage each turn, number of turns of the inner winding N. The DC
winding is the middle one. resistance is normalized to the power frequency. If the
2) Test report resistance is assumed to be frequency dependent the conductor
The leakage reactance is calculated from the standard test area must be specified in height and width (which determines
report short circuit data (positive sequence). the stray losses).

X [ pu ] = Z [%]2 − ( P[kW ] /(10* S [ MVA])) 2 /100

4) Frequency dependency
(2)
The frequency dependent resistance is calculated between
In the case of an autotransformer the reactances are scaled 0.1 to 10 kHz. The typical values and test report resistances
are assumed to follow R (ω) = R0 ⋅ ω / ω0 where R0 is the
according to the Theory Book [8] chapt. 6.7.

resistance at the angular power frequency ω0. This expression

3) Design data
The leakage reactances are calculated according to classical
results in considerably lower values than suggested in Fig. 26
MMF distribution theory as shown in [2], [3]. Both cylindrical
in [2]. This needs to be further investigated. The design data
and pancake windings are supported.
The calculated R(ω) and ω value pairs are fitted to the
resistances are assumed to follow eq. (37) in [2].
4) Handling of the core winding
The artificial core winding is related to the leakage channel function (two-cell Foster equivalent [11])
R ⋅ ω2 ⋅ L12 R2 ⋅ ω2 ⋅ L22
R (ω) = R0 + 21 +
between the inner physical winding and the core. A parameter

R1 + ω2 ⋅ L12 R22 + ω2 ⋅ L22

K=a1/a2 is defined in [2], [7] where a1 is the width of the inner (5)
leakage channel and a2 is the width of the leakage channel
between the inner and the outer/middle winding. A fixed value with the resistances R1 and R2, and inductances L1 and L2 as
K=0.5 is used in ATPDraw. If the pu leakage reactances XML, unknowns. The fitting routine is based on the Lavenberg-
XMH, and XHL (L=inner, M=middle, H=outer) for a three Marquardt method and a simplified version of mrqmin from
winding transformer are given then the leakage reactances to ref. [12]. A negative inductance L0=-L1-L2 is added in series to
the winding resistance to compensate for the inductance of the
X LC ≈ K ⋅ X ML , X MC ≈ X LC + X ML = ( K + 1) ⋅ X ML , and
the core winding are assumed to be [2], [7]
Foster cells. A constraint is put on the total inductance |L0| <Lw
X HC ≈ X MC + X HM = ( K + 1) ⋅ X ML + X HM
where Lw is the inverse of the diagonal Aw-matrix element, [2]
(3) section 5.4.2. The constraint is handled simply by setting

B. Winding resistance L1=L2=0.5⋅Lw when the constraint is violated and then

continue to obtain new optimized values for R1 and R2.
The winding resistances are added externally at the
terminal of the transformer (A-matrix). Optionally, the C. Capacitance
resistances can be frequency dependent. The C-matrix is split in two halves and connected to each
1) Typical values end of the physical windings. The capacitance matrix C is
based on the following two matrices:
⎡C11 C12 C13 ⎤ ⎡C AA C AB C AC ⎤
The typical winding resistances (at power frequency) are in

C w = ⎢C21 C22 C23 ⎥ and C p = ⎢⎢C BA C BB C BC ⎥⎥ (6)

⎢ ⎥
principle based on [10]. A function (4) is established that takes
in the parameter u [kV] and s [MVA] and returns the
resistance in %. Data for a 290 MVA/ 420 kV transformer ⎢⎣C31 C32 C33 ⎥⎦ ⎢⎣CCA CCB CCC ⎥⎦
(Tab. I) were used to extend the data given in [10]:

⎛u ⎞
Rw = 0.7537 ⋅ ⎜ ⎟ ⋅s
The Cw matrix contains the capacitances between windings
− 0.2759
0.0859

⎝ 15 ⎠
[%] (4) 1-3 equal in all phases. The capacitance matrix Cw is built up
like a nodal admittance matrix. The Cp matrix contains
2) Test report capacitances that are specific to phase A, B, or C. These are
typically connected to the outer windings. The total C-matrix
The test report data are given at power frequency. The per
is then built on these two symmetrical matrices dependent on
unit short circuit resistances are calculated from short circuit
the type of winding (pancake/cylindrical). The concept ‘outer
power losses in the test report (positive sequence). The
winding’ will be different for pancake and cylindrical
winding resistance (in pu) is assumed to be equally shared
windings.
 1) Typical values characteristic and the losses are typically higher. The material
A capacitive coupling factor Kc can be specified by the user list is only used for design data and typical values.
with a default value of 0.3. The concept of transient recovery
voltage (TRV) is used to calculate the effective capacitance
when the inductance is known [13]. IEEE standard C37, Fig.
B2 [14] is used to obtain the typical frequency of the TRV for

3⋅ I
a known voltage level and fault current.
Ceff (U , S , X pu , f ) = ⋅
2π U ⋅ ( fTRV (U , I ) )2
f
[μF] (7)

with U in [kV], S in [MVA] and I =

S
3 ⋅U ⋅ X pu
[kA]

In the case of typical values, the Cp matrix (between Fig. 2. 5-legged stacked core model. The αβγ terminals are the nw+1 winding.
phases) is always set to zero for lack of any better choice. For

Cw [1, 2] = CPS = K c ⋅ Ceff (U S , S , X PS , pu , f )

a two-winding transformer the Cw matrix is calculated as The B/H relationship is assumed to follow the Frolich
equation B =
H
a +b⋅ H
Cw [1,1] = CPP = Ceff (U P , S , X PS , pu , f ) − Cw [1, 2]
(10)

Cw [ 2, 2] = CSS = (1/ K c − 1) ⋅ Cw [1, 2]

( )
(8)
The specific loss is assumed to follow
⎛ f ⎞
p [W / kg ] = ⎜ ⎟ ⋅ d ⋅ B 2 + e ⋅ B 10
1.5

⎝ 50 ⎠
For a three winding transformer the typical capacitance is (11)
more complicated with several coupling factors involved.

Cw [1,3] = CPT = 0
Here a simple approach is used: where f is the power frequency.
The specific loss is traditionally (for instance Westinghouse

Cw [ 2,3] = CST = Ceff (U S , S , X ST , pu , f ) − Cw [ 2, 2]

T&D reference book, 1964) assumed to be
(9) p = K ⋅ ( f ⋅ t ⋅ B ) 2 + K ⋅ f ⋅ B x with x said to be 3 for
Cw [3,3] = CTT = Ceff (UT , S , X ST , pu , f ) − Cw [ 2,3]
e max e max
modern materials in the year of 1964. In the above expression
This approach could be further discussed and improved. t is the thickness of the laminates. The traditional expression
was tested on modern material data with little success.
2) Test report
Fig. 3 shows the fit of the specific losses and DC-
In the test report the capacitances between each winding magnetization curve of ARMCO M4 steel. The Frolich fitting
and ground and between all windings is assumed to be directly is not very good, and in Fig. 3b fitting around the knee point
specified while the Cp matrix is set to zero. All values must be (nominal flux) was preferred at the sacrifice of high field
specified per phase. fitting (B=1.9 T). Similar fitting is performed for the other
3) Design data core materials. A modification of the Frolich equation which
The calculation of design data capacitances are based on improves the agreement is introduced in [15]. This method is
[2] chapt. 5.3, p. 33-42. The user has to specify the winding implemented in ATPDraw but is not described here since the
geometry as well as the various equivalent permittivities of numerical implementation of this method still needs
insulation system. Standard formulas for calculating the improvement.
capacitance between cylinders and for cylinders over plans are
4
used with end effect and tank effect adjustments.
D. Core
The core model is connected to the artificial core winding ARMCO M4 60 Hz
3 Fit
terminal on the A-matrix. Only stacked cores with three and
five legs are supported at this point. Basically the inductive p [W/kg] = (f/50)1.5∗(0.339∗B2+0.00125∗B10)
and resistive core parts are treated independently. The core
p [W/kg]

losses are assumed to be linear and the nonlinear inductances 2 50 Hz

are modeled by the Frolich equation (10). Each part of the
core is modeled with its own core loss resistance and
nonlinear inductance using information about their relative
cross section and length to scale the values. Fig. 2 shows the 1

core model for a 5-legged transformer.

It is assumed that the magnetic material is characterized by
four parameters a, b, d and e. A list of typical steel materials is
0
developed based on fitting the manufacturer’s data from state
0 0.4 0.8 1.2 1.6 2
of the art catalogues. Older steel materials will have a different Fig. 3 a) Core loss curves
 10000 when the basic insulation level (BIL) is known and
−0.2134
⎛ u ⎞ ⎛ s ⎞
I m = 0.855 ⋅ ⎜ ⎟ ⋅⎜ ⎟
0.2283

⎝ 150 ⎠ ⎝ 20 ⎠
[%] (15)
ARMCO M4
1000 Fit Frolich when BIL must be estimated. BIL is in [kV], u is the rated
H = 5.284*B/(1-0.542*B) voltage in [kV] and s is the rated power in [MVA].
For a typical core model the user has to specify the
H [A/m]

maximum B-field (normally 1.5-1.7 Tesla) and the maximum

100
core loss density. First a core material has to be guessed and
this gives the a and b values in the Frolich equation (and
possibly also the c and d values that would replace p).
The following relations are then assumed:
2 ⋅ U rms 2 ⋅ U rms
10

λ max = = Bmax ⋅ A ⋅ N ⇒ A ⋅ N =
ω ω⋅ Bmax
(16)

a ⋅ Bmax
H max = ≈ 2 ⋅ irms ⋅
N
1 − b ⋅ Bmax
1
l
a ⋅ Bmax
0.4 0.8 1.2 1.6 2
(17)
⇒ =
B [T]
N
(1 − b ⋅ Bmax ) ⋅ 2 ⋅ irms
Fig. 3 b) DC-magnetization curve
l
Inductance modeling: which somewhat doubtfully assumes a sinusoidal magnetizing
The basic Frolich equation in (10) is reformulated as a current.
This gives the parameter of the fluxlinkage-current
linkage λ = B ⋅ A ⋅ N and the current i = H ⋅ l / N where N is
current flux-linkage characteristic by introducing the flux
characteristic:
a' = a ⋅ ≈ ω ⋅ (1 − b ⋅ B max ) ⋅ rms and
the number of turns of the inner winding, A is the cross l i
section, and l is the length of the involved core section. This A⋅ N 2 u
ω ⋅ B max
rms

b' = b ⋅ ≈ b⋅
gives
i ⋅ A⋅ N 2 / l i ⋅ Ar
1
λ= = A⋅ N 2 ⋅ u rms
(18)
a + b⋅ | i | ⋅N / l a '⋅ lr + b ' ⋅ | i |
(12)

where the constants a ' = a ⋅ lL /( N 2 ⋅ AL ) and b ' = b /( N ⋅ AL ) ,

We see that the expressions for a’ and b’ are independent
of the magnetic material property a. The typical value of b
based on the absolute length (lL) and cross section area (AL) of seems to be fairly constant for standard core materials and a
the core leg, are determined in an optimization process. value of 0.5 is assumed in ATPDraw.

(1 − b ⋅ B max ) ⋅ 2 ⋅ u rms ⋅ i rms

The final characteristics are determined by inserting the The core loss is estimated to
Ploss = p ⋅ ρ ⋅ A ⋅ l = p ⋅ ρ ⋅
relative dimensions for the corresponding section, Ar and lr.
ω ⋅ a ⋅ B max
(19)
The nonlinear inductances are implemented as optional type 2

where p [W/kg] and ρ [kg/m3] are given and the volume A⋅l is
93 or type 98 inductances in ATP.
estimated from (16) and (17).
Core loss modeling
2) Test report
The core loss is split in parts associated with individual
The user specifies the excitation voltage in [%], the current
core sections. If we assume that the core loss is proportional to
in [%] and the core loss in [kW]. The core loss is used directly
the volume of the core we can set the outer leg (Ro) and yoke
as explained above to obtain the core resistances. At the
Ro = Rl /( Arl ⋅ lrl ) and R y = Rl /( Ary ⋅ lry ) where lr and Ar are
resistances (Ry) proportional to the leg resistance (Rl):
moment the core resistances are assumed to be linear and the
core loss value at 100 % excitation is used.
the relative dimensions. Based on the resistive part of Fig. 2 The inductive magnetizing current for each point is
this gives the total core loss as inverse proportional to the leg calculated as
⎛ P[kW ] ⎞
resistance Rl. The constant K3/5 is dependent on the core
I rms = I 0 [%]2 − ⎜ ⎟ [%]
2

⎝ 10 ⋅ S[ MVA] ⎠
geometry. (20)
Ploss = ⋅ K3/ 5
2
V
(13) This results in a sequence of excitation points (Urms and
Rl
Irms). The magnetic circuit in Fig. 2 assuming sinusoidal fluxes
1) Typical values is solved and the rms values of the line currents are calculated
The estimation of the magnetizing current (Im) is based on and compared to measured ones. Optimized values of a’ and
[10]. Some fitting of the data is performed which results in b’ in (12) are found by a constrained Golden Search [12]
−0.2154
⎛ BIL ⎞ ⎛ s ⎞
I m = 0.73 ⋅ ⎜ ⎟ ⋅⎜ ⎟
0.2933 optimization method implemented in ATPDraw [15]. If a
⎝ 350 ⎠ ⎝ 20 ⎠
[%] (14) single point is specified the core model is linear.
 3) Design data value even if the test report shows nonlinearity. The UMEC
For design data the user specifies the core material directly model did not support the phase shift (150 deg.) specified for
with its B-H relationship (a and b values in (10)). The absolute the tested transformer and this has been handled by using a
core dimensions and the number of inner-winding turns N are lead-coupling and renaming the current abc to cab.
known, so the inductances can be found directly from (12). Figure 4 shows the open circuit test report values and the
Based on manufacturer data the core losses can be established corresponding rms quantities calculated in PSCAD and ATP.
2 ⋅U rms
from (11) with B =
The voltage is applied on the LS side and the source has zero

ω⋅ A⋅ N
and known values of the core impedance. The average current values were established by
RMS tools available in the two programs. At values below
weight (volume and density) the core loss can be estimated. rated voltage the two models correspond fairly well to each
other and to the test report, while at higher excitations there is
III. VERIFICATION a substantial difference. The XFMR model uses a Frolich
This section shows a comparison between the new XFMR equation that goes into complete saturation (the current is
model and the UMEC model. Version 4.1 of PSCAD and leakage reactance limited), while the UMEC model uses a
version 5.0 of ATPDraw were used for the modeling. The test linear extrapolation.
object used in this paper is a 290 MVA generator step-up
transformer with a 5-legged stacked core. Some limited core 1.2

Excitation voltage E0 [pu]

design information was available. The yoke and outer leg area
1.1
is 0.54 times the leg area. The yoke and outer leg lengths are
approximately 1.5 and 2.5 times the leg length. Table I shows 1.0
the test report data for the transformer.
0.9
TABLE I Test report
GENERATOR STEP-UP TRANSFORMER TEST REPORT 0.8 XFMR
UMEC
Main data [kV] [MVA] [A] Coupling
0.7
HS 432 290 388 YNj
0 1 2 3 4 5
LS 16 290 10465 d5 Average excitation current I0 [%]
Open-circuit E0 [kV,(%)] [MVA] I0 [%] P0 [kW]
Fig. 4. Test report values compared to calculated quantities for the XFMR and
LS 12 (75) 290 0.05 83.1
UMEC models.
14 (87.5) 290 0.11 118.8
15 (93.75) 290 0.17 143.6
16 (100) 290 0.31 178.6 60
17 (106.25) 290 0.67 226.5
Short-circuit [kV] [MVA] ek, er [%] Pk [kW] b c a
40
HS/LS 432/16 290 14.6, 0.24 704.4
Line current [A]

20

CALCULATED λ/i –CHARACTERISTICS FOR UMEC INPUT

TABLE II

iΔ ⋅ 3 [%]
0
λ [pu] iZ [%] iL [%]
-20
0.75 0.0707 0.0580 0.0707
0.875 0.2224 0.2120 0.2386
0.9375 -40
0.3590 0.3486 0.3995
1.0 0.7170 0.7085 0.8268
1.0625 1.6509 1.6446 1.9582 -60
0.000 0.005 0.010 0.015 0.020
time [s]
For the UMEC model the test report I0-E0 open circuit data
are converted into a peak current and fluxlinkage Fig. 5. Measured and calculated magnetization current. Solid lines: XFMR,
characteristic externally since this was not a feature of this dashed lines: UMEC, dotted lines: Measurements.
model. In this process the loss component was subtracted and
the harmonics in the line current of the delta-winding handled For the test object some waveform records of the excitation
with 3pl-harmonic elimination [16], [17] using the phase current were available. Figure 5 shows the comparison
current. Table II shows the calculated values (iZ directly from between the XFMR, the UMEC model and measured
I0 in Tab. I, iL after subtraction of the losses, and iΔ is the magnetization current at rated voltage. Generally, the XFMR
phase current in the delta with 3-pl harmonic elimination). In current is a bit higher than the UMEC current peaks. Phase b
order to meet the UMEC model requirements the peak-current is better reproduced by the UMEC model. We also see that the
was divided by 2 and scaled back to line currents with 3 . measured current is not symmetrical for some unknown
Without the proper handling of the delta winding the UMEC reason. The model assumption of equal core sections is not
model produced too small magnetization current. For both necessarily true, as well as the exact core length ratios were
models the core loss is assumed to be constant at the rated not known. We also see some strange behavior around zero
crossing of the current which can be due to a non-sinusoidal V. REFERENCES
excitation voltage. [1] J. A. Martinez and B. A. Mork, “Transformer modeling for low- and
Figure 6 and 7 show calculated XFMR and UMEC mid-frequency transients – a review”, IEEE Trans. Power Delivery, vol.
20, no. 2, pp. 1625-1632, 2005.
response of circuit breaker (CB) operation on the LS side (the [2] B. A. Mork, F. Gonzalez, and D. Ishchenko: “Parameter estimation and
HS is open). The applied voltage was also in this case ideal advancements in transformer models for EMTP simulations. Task MTU-
with rated value (16 kV). Figure 6 shows the calculated 7: Model performance and sencitivity analysis”, Bonneville Power
fluxlinkage for the XFMR and UMEC models. We see a small Administration, Portland, OR, 2004.
[3] B.A. Mork, F. Gonzalez, D. Ishchenko, D. L. Stuehm, J. Mitra.” Hybrid
difference during the ring-down transient period, but Transformer Model for Transient Simulation-Part I: Development and
otherwise the fluxes are almost identical. Fig. 7 shows the Parameters”, IEEE Trans. Power Delivery, Vol. 22, pp. 248-255, 2007.
calculated inrush currents where the XFMR model gives [4] B.A. Mork, F. Gonzalez, D. Ishchenko, D. L. Stuehm, J. Mitra, “Hybrid
Transformer Model for Transient Simulation-Part II: Laboratory
around 20 times higher values than the UMEC model. This Measurements and Benchmarking”, IEEE Trans. Power Delivery, Vol.
can be explained from Fig. 4 and the large differences at 22, pp. 256-262, Jan. 2007
extreme saturation. The reality is probably somewhere in [5] W. Enright, O. Navak, G. Irwin, and J. Arrillaga, “An electromagnetic
transients model of multi-limb transformers using normalized core
between the XFMR and UMEC models, but the XFMR model concept”, in Proc. 1997 Int. conf. Power System Transients, pp. 93-98.
seems to have a more realistic shape of the saturation curve. [6] W. Enright, O. Navak, and N. Watson, “Three-phase five-limb unified
magnetic equivalent circuit transformer models for pscad v3”, in Proc.
1999 Int. conf. Power System Transients, pp. 462-467.
2 [7] B. A. Mork, F. Gonzalez, D. Ichshenko, “Leakage inductance model for
CB-OUT

Autotransformer transient simulation”, in Proc. Int. Conf. on Power

CB-IN

XFMR
UMEC System Transients, paper 248, 2005.
1 [8] H. W. Dommel and et.al., Electromagnetic Transients Program
Fluxlinkage [pu]

Reference Manual (EMTP Theory Book), Portland, OR: Prepared for

BPA, Aug. 1986.
0 [9] J. J. Graininger and W. D. Stevenson: Power System Analysis, McGraw-
Hill 1994.
[10] A. Greenwood: Electrical Transients in Power Systems, Wiley, 1991.
[11] F. deLeon, A. Semlyen, “Time domain modeling of eddy current effects
-1
for transformer transients”, IEEE Trans. Power Delivery, vol. 8, no. 1,
pp. 271-280, 1993.
time [sec] [12] W. H. Press, S. A. Teukolsky, W. T. Vetterling, B. P. Flannery:
-2
Numerical recipes in Fortran, 2nd ed. Cambridge University Press, 1992.
0.01 0.02 0.03 0.04 0.05 0.06 [13] IEEE Working Group 15.08.09, Editors: A. M. Gole, J. Martinez-
Fig. 6. Calculated fluxlinkages for the XFMR and UMEC models. For the Velasco, A. J. F Keri, Modeling and analysis of power system transients
XFMR model the flux is found by an integration of the LS line voltages. using digital programs, IEEE 99TP133-0, pp. 4.12-4.13, 1998.
[14] IEEE Guide for Transient Recovery Voltage for AC High-Voltage
Circuit Breakers Rated on a Symmetrical Current Basis, ANSI/IEEE
40 Standard C37.011-1994.
XFMR [15] N. Chiesa, “Power Transformer Modelling: Advanced Core Model”, M.
30 UMEC*20 SC. Thesis, Politecnico di Milano, Italy, 2005.
Inrush current [kA]

20 [16] W. L. A. Neves and H. W. Dommel, “Saturation curves of delta-

connected transformers from measurements”, IEEE Trans. Power
10 Delivery, vol. 10, no. 3, pp. 1432-1437, 1995.

characteristics for Δ-coupled transformer winding”, presented at the

[17] N. Chiesa, H. K. Høidalen, “On the calculation of fluxlinkage/current-
0
European EMTP User Group meeting in Warzaw-Poland, 2005.
-10

-20 VI. BIOGAPHIES

time [sec.]
-30 Hans Kr. Høidalen was born in Norway in 1967. He received his MSc and
0.04 0.05 0.06 0.07 0.08 PhD from the Norwegian University of Science and Technology in 1990 and
Fig. 7. Calculated inrush line currents for the XFMR and UMEC models. 1998 respectively. He is now a professor at the same institution with a special
interest of electrical stress calculations and modeling.
Bruce A. Mork was born in ND-USA in 1957. He received the BSME,
IV. CONCLUSION MSEE, and PhD from North Dakota State University in 1979, 1981 and 1992
respectively. He joined the faculty of Michigan Technological University in
The XFMR and UMEC models give fairly the same result 1992, where he is now Associate Professor of Electrical Engineering, and
at steady state and rated voltage. The most important Director of the Power & Energy Research Center.
difference is seen in Fig. 4. The XFMR core model uses a Francisco Gonzales was born in Spain. He received the M.S. and PhD from
Frolich equation that goes into complete saturation, while the Universitat Politècnica de Catalunya in 1996 and 2001 respectively. He
worked as Postdoc at Michigan Technological University from 2002 to 2005.
UMEC model uses a linear extrapolation. This results in very He is now with El Sewedy Cables in Egypt.
different over-excitation responses and inrush currents. The Dmitry Ishchenko was born in Krasnodar, Russia. He received his M.S. and
UMEC model requires substantial manual conditioning of the PhD degrees in Electrical Engineering from Kuban State Technological
University, Russia in 1997 and 2002 respectively. In 2003 he joined Michigan
open circuit test report data as shown in Tab. II. Both models Techological University as a postdoctoral researcher.
need to be developed in order to reproduce all kinds of Nicola Chiesa was born in Italy in 1980. He received the M.Sc. degree in
switching transients. The representation of extreme saturation, electrical Engineering from Politecnico di Milano in 2005. In Sept. 2005 he
joined the Norwegian University of Science and Technology as a PhD
zero-sequence, nonlinear core losses, hysteresis, and candidate.
remanence should be added.