Académique Documents
Professionnel Documents
Culture Documents
3, JULY 2005
I. INTRODUCTION (1)
applied, but they invariably involve the series impedance and the line, respectively, while , , , and are the line parameters
shunt capacitance of the line. expressed in per-unit length.
Two types of time-domain models have been developed for These parameters are frequency dependent, although can
overhead lines: lumped and distributed parameter models. The be assumed constant, and can usually be neglected. Given the
appropriate selection of a model depends on the line length and frequency dependence of the series parameters, the approach to
the highest frequency to be simulated. the solution of the line equations, even in transient calculations,
Lumped-parameter models represent transmission systems is performed in the frequency domain. The behavior of a mul-
by lumped elements whose values are calculated at a single ticonductor overhead line is described in the frequency domain
frequency. These models are adequate for steady-state calcula- by two matrix equations
tions, although they can also be used for transient simulations
in the neighborhood of the frequency at which parameters were (3)
evaluated.
The most accurate models for transient calculations are those (4)
that take into account the distributed nature of parameters [1],
[2]. Two categories can be distinguished for this type of model: where and are, respectively, the series impedance
constant parameter and frequency-dependent parameter models. and the shunt admittance matrices per-unit length.
Fig. 1 shows the reference frame and the equivalent circuit of The series impedance can be decomposed as follows:
a differential section of a single-conductor overhead line.
(5)
TABLE I
MODELING GUIDELINES FOR OVERHEAD LINES
for cylindrical conductors [4]. Other programs base the calcu- II. DATA INPUT
lations on closed-form approximations [5], [6]. Reference [7]
gives a detailed description of the procedures. Since transient studies evolved after load flow, short circuit,
The shunt admittance can be expressed as follows: and stability studies, existing databases of transmission-line
parameters may consist of only synchronous frequency (50- or
(6) 60-Hz) line impedances. Short-circuit line data are often just
the positive, negative, and zero impedances. Load-flow line
where elements can be usually neglected, except at very low databases might contain only a per-phase positive-sequence
frequencies. representation. In all cases, line data are stored only as
Table I shows a summary of modeling guidelines for overhead impedances.
lines; it is a revised version of those guidelines proposed by LC routine users enter the physical parameters of the line,
CIGRE [2]. select the desired type of line model, and the model is created.
These guidelines are related to phase conductors only. If the Since all models are developed from physical transmission-line
aim of a transient simulation is to determine whether the line parameters, it is highly recommended that a database of physical
will flashover or not, then a representation of line insulations line parameters be created.
must be included. Specific transients are those related to stresses In order to develop line models for transient simulations, the
caused by an external source represented as a current source following input data must be available:
(i.e., a lightning stroke). In such cases, other pieces of the line • (x,y) coordinates of each conductor and shield wire;
(shield wires, towers or poles, footing impedances) must be in- • bundle spacing, orientations;
cluded in the model. • sag of phase conductors and shield wires;
The length of an overhead line that must be included in a • phase and circuit designation of each conductor;
model depends on the type of transient to be duplicated or, more • phase rotation at transposition structures;
specifically, on the range of frequencies involved in the transient • physical dimensions of each conductor;
process. As a rule of thumb, the lower the frequencies, the more • DC resistance of each conductor and shield wire (or resis-
the length of line is to be represented. For low and mid frequency tivity);
transients, the whole line length is included in the model. For • ground resistivity of the ground return path.
fast and very fast transients, a few line spans will usually suffice.
This paper deals with data input that is required for proper Other information, such as segmented grounds, can be
modeling of overhead lines in transient simulations. important.
Users of EMTP-like tools obtain overhead line parameters by Note that all of the above information, except conductor re-
means of a dedicated supporting routine that in this document sistances and ground resistivity, is from geometric line dimen-
will be denoted “Line Constants” (LC) [7]. In addition, several sions; therefore, the availability of these parameters should not
routines are presently implemented in transients programs to be a problem.
derive line models considering different approaches [8]–[10]. LC routine users can request the following models:
The next section describes the most basic input requirements of • lumped-parameter equivalent or nominal pi-circuits at the
LC-type routines. It is followed by a section that investigates the specified frequency;
sensitivity of line parameters ( , , , ) to variations in the • constant distributed-parameter model at the specified fre-
representation of an overhead line and by an example that shows quency;
the influence that some parameters can have on the transient • frequency-dependent distributed parameter model, fitted
response. for a given frequency range.
2040 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
III. EXAMPLE
A. Test Line
An example case is included to illustrate
• proper input of physical parameters;
Fig. 3. Relationship between overhead line parameters and ground resistivity.
• examination of line constants output; (a) Zero-sequence resistance (ohm/mi). (b) Zero-sequence inductance (H/mi).
• benchmarking impedances , ; (c) Positive-sequence resistance (ohm/mi). (d) Positive-sequence inductance
• benchmarking for frequency response; (H/mi).
• application considerations.
Fig. 2 shows the geometry of the 345-kV transmission line B. Sensitivity Analysis of Line Parameters
studied in this example. Conductor data for this line are pre- A parametric study of sequence parameters was performed.
sented in the following table. To obtain the frequency dependence of the resistance and the
MARTINEZ et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS—PART I: OVERHEAD LINES 2041
inductance of conductors, users can assume either a solid con- tion. Skin effect entails that the highest current density is at the
ductor or a hollow conductor and apply the skin effect correc- conductor surface. To include skin effect for hollow conductors
2042 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
V. CONCLUSIONS
When only phase conductors and shield wires are to be in-
cluded in the line model, the line parameters can be calculated
from the line geometry, as well as from physical properties of
phase conductors, shield wires, and ground. A great accuracy is
not usually required when specifying input values if the goal is
to duplicate low-frequency and slow-front transients, but more
care is needed, mainly with the ground resistivity value, if the
goal is to simulate fast transients.
REFERENCES
[1] Modeling and Analysis of Power System Transients Using Digital Pro-
grams, A. Gole, J. A. Martinez, and A. Keri, Eds., IEEE Special Publ.
TP-133-0, IEEE Catalog no. 99TP133-0, 1999.
[2] CIGRE Working Group 33.02, “Guidelines for representation of net-
work elements when calculating transients,” CIGRE Brochure 39, 1990.
[3] J. R. Carson, “Wave propagation in overhead wires with ground return,”
Bell Syst. Tech. J., vol. 5, pp. 539–554, 1926.
[4] S. A. Schelkunoff, “The electromagnetic theory of coaxial transmission
Fig. 10. Zero-sequence energization of a 30-mi untransposed overhead line lines and cylindrical shields,” Bell Syst. Tech. J., vol. 13, pp. 532–579,
(Source = 1 V , 50-s pulse). (a) With ground wires. (b) Without ground wires.
1934.
[5] L. M. Wedepohl and D. J. Wilcox, “Transient analysis of underground
power transmission system; system-model and wave propagation char-
exceeds the so-called corona inception voltage, see, for in- acteristics,” Proc. Inst. Elect. Eng., vol. 120, no. 2, pp. 252–259, Feb.
1973.
stance, [12]. This inception voltage decreases as the travel [6] A. Deri, G. Tevan, A. Semlyen, and A. Castanheira, “The com-
distance increases and depends on the propagation wave steep- plex ground return plane. A simplified model for homogeneous and
ness. Corona increases the conductor capacitance, decreases multi-layer earth return,” IEEE Trans. Power App. Syst., vol. PAS-100,
no. 8, pp. 3686–3693, Aug. 1981.
the surge impedance, increases the coupling factor between [7] H. W. Dommel, Electromagnetic Transients Program Manual (EMTP
conductors, and increases losses. Corona causes additional Theory Book). Portland, OR: Bonneville Power Administration, 1986.
attenuation and distortion, mainly on the wavefront and above [8] J. R. Marti, “Accurate modeling of frequency-dependent transmission
lines in electromagnetic transient simulations,” IEEE Trans. Power App.
the inception voltage, so a noncorona model will provide Syst., vol. PAS-101, no. 1, pp. 147–155, Jan. 1982.
conservative results. Some programs allow users to include [9] T. Noda, N. Nagaoka, and A. Ametani, “Phase domain modeling of fre-
this effect in transient simulations. Several approaches can quency-dependent transmission lines by means of an ARMA model,”
IEEE Trans. Power Del., vol. 11, no. 1, pp. 401–411, Jan. 1996.
be considered. The simplest one includes corona effect from [10] A. Morched, B. Gustavsen, and M. Tartibi, “A universal model for
line geometry, although some models also consider the air accurate calculation of electromagnetic transients on overhead lines
density factor and even an irregularity factor. In fact, corona and underground cables,” IEEE Trans. Power Del., vol. 14, no. 3, pp.
1032–1038, Jul. 1999.
is a very complex phenomenon whose accurate representation [11] IEEE Task Force on Fast Front Transients, “Modeling guidelines for fast
should be based on a distributed hysteresis behavior. Perhaps transients,” IEEE Trans. Power Del., vol. 11, no. 1, pp. 493–506, Jan.
the most important study for which corona can have a strong 1996.
[12] A. R. Hileman, Insulation Coordination for Power Systems. New
influence is the determination of incoming surges in substations York: Marcel Dekker, 1999.
[12]. When an accurate representation of the corona effect is [13] P. S. Maruvada, D. H. Nguyen, and H. Hamadani-Zadeh, “Studies on
possible, then additional input parameters are required for a full modeling corona attenuation of dynamic overvoltages,” IEEE Trans.
Power Del., vol. 4, no. 2, pp. 1441–1449, Apr. 1989.
characterization of the model [13], [14]. [14] S. Carneiro and J. R. Martí, “Evaluation of corona and line models in
The concept of “nonuniform line” has been introduced to electromagnetic transients simulations,” IEEE Trans. Power Del., vol.
deal with line geometries where the longitudinal variation of 6, no. 1, pp. 334–342, Jan. 1991.
[15] A. I. Ramirez, A. Semlyen, and R. Iravani, “Modeling nonuniform trans-
line parameters can be significant. Examples of this type of line mission lines for time domain simulation of electromagnetic transients,”
are lines crossing rivers or entering substations. In such cases, IEEE Trans. Power Del., vol. 18, no. 3, pp. 968–974, Jul. 2003.
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005 2045
Abstract—EMTP-type programs include dedicated support “cable constants” (CC). These programs have some shortcom-
routines (cable constants) for calculating an electric representa- ings in representing certain cable features. The paper provides
tion of cable systems in terms of a series impedance matrix guidelines on how to apply CC routines to the most common
and a shunt admittance matrix , based on cable data defined by
geometry and material properties. and are the basic input of types of high-voltage cable systems. The discussion considers
the various cable models that are used in time-domain transient both cables with extruded solid insulation (XLPE, PE) and ca-
simulations. This paper discusses the modeling of high-voltage bles with oil-impregnated paper.
cables: single-core, three-phase, and pipe-type cables. Material
properties are given for commonly used conductive and insulating
materials, and how to represent semiconductive screens, lossy in- II. INPUT DATA
sulation materials, and magnetic armors is shown. The significance
of the grounding condition of sheaths and armors is discussed. The basic equations used to represent overhead lines and in-
In transient calculations, it is always important to accurately
represent the core conductor, insulation, semiconductive layers, sulated cables have the following form:
and the metallic sheath. Frequency-dependent losses of paper-oil
insulation need to be taken into account for very-high-frequency (1)
transients. The significance of conductors external to the cable
depends on the shielding effect of the cable sheath, which depends
(2)
on the sheath design and the frequency content of the transient.
The conclusions are supported by numerical simulation results. where and are, respectively, the series resistance, se-
Index Terms—Insulated cables, modeling, power system tran-
ries inductance, shunt conductance, and shunt capacitance per-
sients, simulation. unit length of the cable system. These quantities are ma-
trices, being the number of (parallel) conductors of the cable
system. The variable reflects that these quantities are calcu-
I. INTRODUCTION lated as function of frequency.
and are calculated by means of CC routines, using cable
S EVERAL line models have been implemented in com-
monly available EMTP-type programs which can accu-
rately represent the frequency dependence of cable systems
geometry and material properties as input parameters. In gen-
eral, users must specify:
[1]–[3]. All of these models require the same type of input pa- 1) Geometry
rameters, namely the series impedance matrix and the shunt • location of each conductor ( - coordinates);
admittance matrix . Sufficiently accurate input parameters • inner and outer radii of each conductor;
are, in general, more difficult to obtain for cable systems than • burial depth of the cable system.
for overhead lines as the small geometrical distances make 2) Material properties
the cable parameters highly sensitive to errors in the specified • resistivity and relative permeability of all conduc-
geometry. In addition, it is not straightforward to represent tors ( is unity for all nonmagnetic materials);
certain features, such as wire screens, semiconductive screens, • resistivity and relative permeability of the surrounding
armors, and lossy insulation materials. The situation is made medium ;
further complicated by uncertainties in the geometrical data • relative permittivity of each insulating material .
as provided by the manufacturer as they define guaranteed The calculation of and from the geometry and mate-
measures, but not necessarily the actual measures. rial properties follows similar steps for all CC routines. The
Most EMTP-type programs have dedicated support routines reader is referred to [4]–[6] for details. The main challenge is
for calculating cable parameters. These routines have very sim- the impedance calculation which is based on computing sur-
ilar features, so hereinafter they will be given the generic name face impedances and transfer impedances of cylindrical metallic
shields, as well as self and mutual ground impedances. CC rou-
Manuscript received March 1, 2004; revised August 9, 2004. Paper no. tines differ in the actual expressions that are used in the calcula-
TPWRD-00106-2004. tion of these quantities. It is worth noting that these routines take
Task Force Members: J. A. Martinez (Chairman), D. Durbak, B. Gustavsen,
B. Johnson, J. Mahseredjian, B. Mork, R. Walling. the skin effect into account but neglect any proximity effects. A
Digital Object Identifier 10.1109/TPWRD.2005.848774 procedure for including proximity effects is given in [7].
0885-8977/$20.00 © 2005 IEEE
2046 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
TABLE I TABLE II
RESISTIVITY OF CONDUCTIVE MATERIALS RELATIVE PERMITTIVITY OF INSULATION MATERIALS
TABLE III
PARAMETERS OF SEMICONDUCTIVE LAYERS (INDICATIVE VALUES)
TABLE IV
TEST CABLE PARAMETERS
The receiving end voltages are shown in Fig. 8, assuming Fig. 9. Effect of semiconductor thickness on overvoltage.
1-mm semiconductive layers. It can be seen that reducing the
thickness of the lead sheath from 2 to 1 mm leads to a strong simulation is performed with the following representations of
increase of the attenuation, whereas a reduction from 3 to 2 mm the main insulation:
has little effect. This can be understood by considering that
a) Lossless insulation [i.e., dc value in (4)].
the dominant frequency component of the transient is about
b) Lossy insulation by (4).
10 kHz. At this frequency, the penetration depth in lead is
2.4 mm, according to (6). Thus, increasing the thickness of the Fig. 10 shows an expanded view of the initial transient (re-
lead sheath beyond 2.4 mm will not lead to a significant change ceiving) end. It can be seen that the lossy insulation gives a much
in the response. stronger reduction of the peak value for the narrow pulse (2 s)
3) Sensitivity to Semiconductor Thickness: Assuming a than that for the lossless insulation. This reduction is an effect of
0.215-mm Cu sheath, the step response is calculated for dif- both attenuation and frequency-dependent velocity. It is further
ferent thicknesses of the semiconductor layers: 0, 1, 2, 3 mm. seen that the travel time of the lossy insulation is smaller than
The responses in Fig. 9 show that the semiconductors lead to that of the lossless insulation, which is caused by the reduction
a decrease of the propagation speed, as previously explained in in permittivity at high frequencies, according to (4).
Section VIII.
4) Sensitivity to Insulation Losses: In this example, the B. Armored Cable
XLPE main insulation is replaced by paper-oil insulation. It In this example, an armor of 5-mm steel wires and a 5-mm
is further assumed that the cable has a 2-mm lead sheath and outer insulation are incorporated into the cable design. It is fur-
no semiconductive screens. The open-end voltage is calculated ther assumed XLPE main insulation, a 2-mm lead sheath, and
by applying a 2 and a 10- s width square-voltage pulse. The 1-mm semiconductive screens. Only one cable is considered.
2050 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
REFERENCES
[1] L. Marti, “Simulation of transients in underground cables with fre-
quency-dependent modal transformation matrices,” IEEE Trans. Power
Del., vol. 11, no. 3, pp. 1099–1110, Jul. 1988.
[2] T. Noda, N. Nagaoka, and A. Ametani, “Phase domain modeling of
frequency-dependent transmission line models by means of an ARMA
model,” IEEE Trans. Power Del., vol. 11, no. 1, pp. 401–411, Jan. 1996.
[3] A. Morched, B. Gustavsen, and M. Tartibi, “A universal model for
accurate calculation of electromagnetic transients on overhead lines
and underground cables,” IEEE Trans. Power Del., vol. 14, no. 3, pp.
1032–1038, Jul. 1999.
[4] H. W. Dommel, Electromagnetic Transients Program Manual (EMTP
Theory Book). Portand, OR: Bonneville Power Administration, Aug.
1986.
[5] L. M. Wedepohl and D. J. Wilcox, “Transient analysis of underground
power-transmission systems. System-model and wave-propagation
characteristics,” Proc. Inst. Elect. Eng., vol. 120, no. 2, pp. 253–260,
Feb. 1973.
[6] A. Ametani, “A general formulation of impedance and admittance of
cables,” IEEE Trans. Power App. Syst., vol. PAS-99, no. 3, pp. 902–909,
Fig. 11. Effect of armor permeability on overvoltage.
May/Jun. 1980.
[7] Y. Yin and H. W. Dommel, “Calculation of frequency-dependent imped-
The resulting voltage of the open-circuit step response is cal- ances of underground power cables with finite element method,” IEEE
culated for different values of the armor permeability: Trans. Magn., vol. 25, no. 4, pp. 3025–3027, Jul. 1989.
[8] G. Bianchi and G. Luoni, “Induced currents and losses in single-core
, being the cable length of 50 km, see Fig. 11. It is submarine cables,” IEEE Trans. Power App. Syst., vol. PAS-95, no. 1,
seen that increasing the permeability strongly increases the ef- pp. 49–58, Jan./Feb. 1976.
fective attenuation of the voltage. The reason is that a perme- [9] O. Breien and I. Johansen, “Attenuation of traveling waves in single-
phase high-voltage cables,” in Proc. Inst. Elect. Eng., vol. 118, Jun. 1971,
ability increase reduces the penetration depth in the armor, thus pp. 787–793.
increasing the resistance of the inner armor surface impedance. [10] B. Gustavsen, “Panel session on data for modeling system transients: In-
For a 5-km cable length, the significance of the armor was found sulated cables,” in Proc. IEEE Power Engineering Soc. Winter Meeting,
2001.
to be small as the magnetic field would not appreciably pene- [11] K. Steinbrich, “Influence of semiconducting layers on the attenuation
trate the sheath conductor, due to the increased frequency of the behaviour of single-core power cables,” Proc. Inst. Elect. Eng., Gen.,
Transm. Distrib., vol. 152, no. 2, pp. 271–276, Mar. 2005.
transient. [12] A. Ametani, Y. Miyamoto, and N. Nagaoka, “Semiconducting layer
impedance and its effect on cable wave-propagation and transient
X. CONCLUSION characteristics,” IEEE Trans. Power Del., vol. 19, no. 4, pp. 1523–1531,
Oct. 2004.
This paper has considered cable data for simulating transients [13] J. P. Noualy and G. L. Roy, “Wave-propagation modes on high-voltage
on phase conductors of single core cables, three-phase cables, cables,” IEEE Trans. Power App. Syst., vol. PAS-96, no. 1, pp. 158–165,
Jan./Feb. 1977.
and pipe-type cables. The main conclusions can be summarized [14] A. Ametani, “Wave propagation characteristics of cables,” IEEE Trans.
as follows. Power App. Syst., vol. PAS-99, no. 2, pp. 499–505, Mar./Apr. 1980.
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005 2051
I. INTRODUCTION
Manuscript received June 21, 2004; revised August 23, 2004. Paper no. losses , short-circuit current , short-circuit voltage
TPWRD-00289-2004. , short-circuit losses . Saturation curves and capaci-
Task Force Members: J. A. Martinez (Chairman), D. Durbak, B. Gustavsen,
B. Johnson, J. Mahseredjian, B. Mork, R. Walling. tances can also be obtained from tests, but they are not always
Digital Object Identifier 10.1109/TPWRD.2005.848752 available.
0885-8977/$20.00 © 2005 IEEE
2052 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
A. Matrix Representation
The steady-state equations of a single-phase multiwinding
transformer can be expressed by an impedance equation [11]
(1)
The formulation can be extended to three-phase transformers
by replacing any element of by a (3 3) submatrix
(2)
(3a)
(3b)
(4)
(10) (12)
MARTINEZ et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS—PART III: TRANSFORMERS 2055
(15)
the innermost windings, for paths between the legs and the
space between the two windings, for paths between the legs
and outside the windings, and for the leakage path through
the air in parallel with the yokes. Reluctances , also known
as zero-sequence paths, have been halved.
After using duality, reluctances due to paths through iron will
transform into nonlinear inductances, while flux paths in air will
transform into linear inductances. Fig. 11(c) shows the equiva-
lent circuit, where the effect of reluctances and has been
neglected.
• and represent the winding resistances of the high-
and low-voltage windings.
• The linear inductances represent the flux leakages be-
tween both windings.
• Zero-sequence flux paths are represented by the linear in-
ductances ; eddy current losses resulting from zero-se-
quence fluxes are represented by resistances .
• The parallel combination of resistances and saturable
inductances represents the flux paths through legs.
• Each parallel combination of and represents the
yoke section between a pair of phases.
Capacitances between terminals and ground (core and tank),
between windings, and between phases could be added to the
equivalent circuit.
The elements and represent leakage fluxes and asso-
ciated losses that circulate outside the concentric coil pairs on
each phase. Dividing the flux paths into two parts results in a
more symmetric and convenient connection for the core equiv-
alent in the resulting electrical equivalent circuit.
Some discussion can be made about zero-sequence flux paths
at this point. In the case of an unbalanced excitation that in-
cludes zero sequence, the total fluxes linked by the three sets
of coils on the phase legs will not add to zero and this zero-se-
quence flux will circulate through the surrounding oil and air
and through fittings and tank walls. Following the assumptions
made in the development of this equivalent circuit, this flux flow
is distributed across the tops of the coils on the three phase
legs and is in addition to the relatively smaller amount of flux
that would flow there for balanced three-phase excitation. Alter-
nately, the zero-sequence effect might have been concentrated
all at the center phase, or divided into two parts at each of the
other phases. In any case, the total effect is the same. In the
equivalent electrical circuit, it is seen that zero-sequence oper-
ation results in an additional current (it can be seen in terms
of superposition) which circulates through the and ele-
ments. Regardless of how the and effect is distributed,
the total impedance of the zero-sequence path must be the same
as is shown in Table I, where the zero-sequence test is carried
out.
The portion of the equivalent circuit inside the ideal trans-
formers is valid regardless of winding connections; delta or wye
connections are made up at the outside. However, procedures
for parameter determination will depend on transformer con-
Fig. 11. Duality-derived equivalent circuit of a three-legged stacked-core nections since the same zero-sequence tests cannot be applied
transformer. (a) Physical structure. (b) Magnetic circuit. (c) Equivalent circuit. to all transformers.
2058 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
TABLE I
TESTS FOR DETERMINATION OF WINDING PARAMETERS OF WYE-WYE TRANSFORMERS
TABLE II
TESTS FOR DETERMINATION OF WINDING PARAMETERS OF DELTA-WYE TRANSFORMERS
The diagram of the two tests required to obtain the pa- be used if transformer geometry is known. As illus-
rameters of the core is shown in Table III. In both tests, the trated in several works, see, for instance, [15] and [18],
delta-connected HV side is left open. an accurate enough estimation of the different core re-
1) The first one is performed by shorting one outer leg and luctances and, therefore, of the saturable inductances,
exciting the other outer leg; it will be used to determine can be performed by using some simple expressions
leg characteristic (i.e., , ). presented in textbooks [28].
2) The second test is performed by shorting the center 2) Other tests to obtain transformer parameters have been
leg and exciting an outer leg; after obtaining the com- proposed by some authors. Those tests fit the model
bined characteristic of leg and yoke characteristic, the they are using for representing a three-legged trans-
leg characteristic, deduced in the previous test, is sub- former. For instance, Fig. 13 shows the diagram of
tracted. the measurements proposed in [33] for estimation of
The tests are also used to obtain core losses and, there- the nonlinear characteristics of the three legs and the
fore, the full representation of legs and yokes. As for zero-sequence parameters. However, one can also note
single-phase units, some care is needed to account for that both types of measurements are based on open-cir-
capacitances effects during excitation (open-circuit) tests. cuit tests. Therefore, the same concern for delta-con-
Parameters and can be also derived from a zero- nected transformers still remains.
sequence open-circuit test for both transformer connec- 3) Some researchers suggest a core-loss representation
tions. As for the tests shown above, the delta side of a using nonlinear resistors in parallel with each non-
delta-wye connected transformer should also be open. In linear inductance, with the nonlinear resistance values
such a case, the equivalent circuit would be that shown in determined from test data at different excitation levels.
Fig. 12. A modified version of (18) must be used. In fact, However, as discussed earlier for single-phase trans-
only (18b) has to be modified formers, there are significant limitations of nonlinear
resistance representation because the hysteresis losses
(19) depend on the maximum flux level and not the max-
imum voltage level. A fit of nonlinear resistances to
• Discussion loss data, which is accurate for one excitation wave-
1) Tests shown in Table III cannot be always performed, form, may not provide a match for excitation of a dif-
since opening the delta side or even separating the three ferent frequency or waveform.
windings of the wye side is only possible with very 4) Eddy current effects in windings can be incorporated
special transformers. Therefore, an alternative proce- by following the same procedure suggested for single-
dure should be considered. An efficient approach can phase units. The most reliable representation should be
2060 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
TABLE III
TESTS FOR DETERMINATION OF CORE PARAMETERS [19]
Fig. 12. Equivalent circuit of a zero-sequence open-circuit test (delta side must
be open in a delta-wye connection).
Although accurate and detailed models require both test and [16] X. Chen, “A three-phase multi-legged transformer model in ATP using
design data, extreme detail is rarely justified. Simple assump- the directly-formed inverse inductance matrix,” IEEE Trans. Power Del.,
vol. 11, no. 3, pp. 1554–1562, Jul. 1996.
tions can be generally applied, having available data sufficient [17] G. R. Slemon, “Equivalent circuits for transformers and machines in-
for most transient simulations. The effect of capacitances as well cluding nonlinear effects,” Proc. Inst. Elect. Eng., pt. IV, vol. 100, pp.
as the frequency dependence of some parameters should not be 129–143, 1953.
[18] C. M. Arturi, “Transient simulation and analysis of a three-phase five-
neglected, mainly for large size transformers. limb step-up transformer following an out-of-phase synchronization,”
IEEE Trans. Power Del., vol. 6, no. 1, pp. 196–207, Jan. 1991.
[19] D. L. Stuehm, Three Phase Transformer Core Modeling, Bonneville
Power Administration, Feb. 1993.
[20] A. Narang and R. H. Brierley, “Topology based magnetic model for
steady-state and transient studies for three-phase core type trans-
REFERENCES formers,” IEEE Trans. Power Syst., vol. 9, no. 3, pp. 1337–1349, Aug.
1994.
[1] A. Greenwood, Electrical Transients in Power Systems, 2nd ed. New [21] B. A. Mork, “Five-legged wound-core transformer model: derivation,
York: Wiley, 1991. parameters, implementation, and evaluation,” IEEE Trans. Power Del.,
[2] IEEE Standard Test Code for Liquid-Immersed Distribution, Power, and vol. 14, no. 4, pp. 1519–1526, Oct. 1999.
Regulating Transformers, IEEE Std. C57.12.90, 1999. [22] A. K. Sawhney, A Course in Electrical Machine Design. New Delhi,
[3] IEEE Standard Test Code for Dry-Type Distribution and Power Trans- India: Dhanpat Rai & Sons, 1996.
formers, IEEE Std. C57.12.91, 2001. [23] R. A. Walling, K. D. Barker, T. M. Compton, and I. E. Zimmerman,
[4] IEEE Guide for Transformer Loss Measurement, IEEE Std. C57.123, “Ferroresonant overvoltages in grounded padmount transformers with
2002. low-loss silicon-steel cores,” IEEE Trans. Power Del., vol. 8, no. 3, pp.
[5] CIGRE Working Group 33.02, “Guidelines for representation of net- 1647–1660, Jul. 1993.
work elements when calculating transients,” CIGRE Brochure 39, 1990. [24] E. E. Fuchs, D. Yildirim, and W. M. Grady, “Measurement of eddy-
[6] IEEE Task Force on Slow Transients, “Modeling and analysis guidelines current loss coefficient P , derating of single-phase transformers,
for slow transients. Part I: torsional oscillations; transient torques; tur- and comparison with K-factor approach,” IEEE Trans. Power Del., vol.
bine blade vibrations; fast bus transfer,” IEEE Trans. Power Del., vol. 15, no. 1, pp. 148–154, Jan. 2000.
10, no. 4, pp. 1950–1955, Oct. 1995. [25] F. de León and A. Semlyen, “Time domain modeling of eddy current
[7] IEEE Task Force on Slow Transients, “Modeling and analysis guide- effects for transformer transients,” IEEE Trans. Power Del., vol. 8, no.
lines for slow transients. Part II: controller interactions; harmonic inter- 1, pp. 271–280, Jan. 1993.
actions,” IEEE Trans. Power Del., vol. 11, no. 3, pp. 1672–1677, Jul. [26] E. J. Tarasiewicz, A. S. Morched, A. Narang, and E. P. Dick, “Frequency
1996. dependent eddy current models for nonlinear iron cores,” IEEE Trans.
[8] IEEE Task Force on Slow Transients, “Modeling and analysis guidelines Power Syst., vol. 8, no. 2, pp. 588–597, May 1993.
for slow transients—Part III: the study of ferroresonance,” IEEE Trans. [27] A. Gaudreau, P. Picher, L. Bolduc, and A. Coutu, “No-load losses in
Power Del., vol. 15, no. 1, pp. 255–265, Jan. 2000. transformer under overexcitation/inrush-current conditions: tests and a
[9] A. M. Gole, J. A. Martinez-Velasco, and A. J. F. Keri, Eds., Modeling new model,” IEEE Trans. Power Del., vol. 17, no. 4, pp. 1009–1017,
and Analysis of System Transients Using Digital Programs: IEEE PES Oct. 2002.
Special Publication, 1999. [28] V. Del Toro, Electric Machines and Power Systems. Englewood Cliffs,
[10] J. A. Martinez and B. A. Mork, “Transformer modeling for low- and NJ: Prentice-Hall, 1985.
mid-frequency transients. A review,” IEEE Trans. Power Del., vol. 20, [29] IEEE WG C-5 of the Systems Protection Subcommittee, “Mathemat-
no. 2, pp. 1625–1632, Apr. 2005. ical models for current, voltage, and coupling capacitor voltage trans-
[11] V. Brandwajn, H. W. Dommel, and I. I. Dommel, “Matrix representa- formers,” IEEE Trans. Power Del., vol. 15, no. 1, pp. 62–72, Jan. 2000.
tion of three-phase n-winding transformers for steady-state and tran- [30] G. C. Paap, A. A. Alkema, and L. van der Sluis, “Overvoltages in power
sient studies,” IEEE Trans. Power App. Syst., vol. PAS-101, no. 6, pp. transformers caused by no-load switching,” IEEE Trans. Power Del.,
1369–1378, Jun. 1982. vol. 10, no. 1, pp. 301–307, Jan. 1995.
[12] H. W. Dommel, Electromagnetic Transients Program Manual (EMTP [31] P. Sarma-Maruvada and N. Hylten-Cavallius, “Capacitance calculations
Theory Book). Portland, OR: Bonneville Power Administration, 1986. for some basic high voltage electrode configurations,” IEEE Trans.
[13] R. Yacamini and H. Bronzeado, “Transformer inrush calculations using Power App Syst., vol. PAS-94, no. 5, pp. 1708–1713, Sep./Oct. 1975.
a coupled electromagnetic model,” Proc. Inst. Elect. Eng., Sci. Meas. [32] F. de León and A. Semlyen, “Efficient calculation of elementary parame-
Technol., vol. 141, no. 6, pp. 491–498, Nov. 1994. ters of transformers,” IEEE Trans. Power Del., vol. 7, no. 1, pp. 376–383,
[14] J. Arrillaga, W. Enright, N. R. Watson, and A. R. Wood, “Improved simu- Jan. 1992.
lation of HVDC converter transformers in electromagnetic transient pro- [33] E. F. Fuchs and Y. You, “Measurement of -i characteristics of asym-
grams,” Proc. Inst. Elect. Eng., Gen. Transm. Distrib., vol. 144, no. 2, metric three-phase transformers and their applications,” IEEE Trans.
pp. 100–106, Mar. 1997. Power Del., vol. 17, no. 4, pp. 983–990, Oct. 2002.
[15] N. D. Hatziargyriou, J. M. Prousalidis, and B. C. Papadias, “Generalized [34] F. de León and A. Semlyen, “Complete transformer model for electro-
transformer model based on the analysis of its magnetic core circuit,” magnetic transients,” IEEE Trans. Power Del., vol. 9, no. 1, pp. 231–239,
Proc. Inst. Elect. Eng. C, vol. 140, no. 4, pp. 269–278, July 1993. Jan. 1994.
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005 2063
I. INTRODUCTION
TABLE I
MODELING GUIDELINES FOR ROTATING MACHINES [1]
The paper has been organized into two main sections covering
synchronous and induction machines, respectively.
TABLE II the manufacturer (short-circuit test data) are listed below; this
SYNCHRONOUS MACHINE MODELS FOR TRANSIENT STUDIES list corresponds to Model 2.2 (i.e., a model with two circuits on
each rotor axis, as shown in Fig. 2). This model is usually the
most complex model that can be derived from short-circuit test
data. However, the same procedure can be applied to machines
with any number of rotor circuits on each axis, as presented in
[7], if the required information is available.
Short-circuit tests can be only used for the determination of
d-axis values; that is, q-axis characteristic values can be calcu-
lated but cannot be directly derived from short-circuit test data.
The sections that describe the calculation of q-axis quantities
are titled “unconfirmed” in IEC 34-4 [8] and “informative” in
IEEE Standard 115 [2].
Several procedures have been proposed for the determination
of the internal parameters from short-circuit test data. For in-
stance, an informative Appendix is included in IEEE Standard
1110 with relationships between manufacturer data and equiva-
lent circuit parameters. The procedure summarized in this sec-
tion was presented in [9]. Table IV shows the relationships upon
which the procedure is based and the steps to be performed to
TABLE III
SYNCHRONOUS MACHINE PARAMETERS obtain the parameters of each circuit rotor. A procedure appli-
cable to any type of ac machine with any number of circuits on
any rotor axis was presented in [7].
2) Standstill Frequency Response (SSFR) Tests: An accu-
rate identification of synchronous machine parameters can be
performed from low-voltage frequency-response tests at stand-
still. These tests are becoming a widely used alternative to short-
circuit tests due to these advantages.
• They can be performed either in the factory or in site at a
relatively low cost.
• Equivalent circuits of an order higher than Model 2.2 (
Table II) can be derived.
• Identification of field responses is possible. According to
IEEE Standard 115, measurable parameters are:
• The d-axis operational impedance .
• The q-axis operational impedance .
• The standstill armature to field transfer function .
• The standstill armature to field transfer impedance
.
where is the alternative to the third measurement .
In addition, the measurement of field to armature transfer
to be performed, the user’s experience, and the available infor-
impedance is occasionally required.
mation. Depending on the characteristic parameter source, the
Table V shows the test setups for each of the measurable pa-
most complex models very often cannot be used due to a lack
rameters and the main relationships derived from each test.
of data.
The procedure for the identification of d-axis parameters from
As mentioned in the introduction, these representations are
SSFR can be summarized as follows [2].
suitable for simulation of low-frequency transients, such as tran-
sient stability studies, subsynchronous resonance, load rejec- 1) Use the best available estimate for the armature leakage
tion, short circuits, generator tripping, generator synchroniza- inductance ; it could be the value supplied by the man-
tion, and inadverted energization. ufacturer.
The conversion procedures that have been proposed for de- 2) Obtain , which is the low-frequency limit of ,
termination of the parameters for these models are discussed then determine
in the subsequent sections. Two sources of data are considered: (2)
short-circuit and standstill frequency-response (SSFR) tests.
3) Find the field to armature turns ratio using the
B. Determination of Electrical Parameters armature to field transfer impedance
1) Short-Circuit Test: Table III shows the parameters to be
(3)
specified in equivalent circuits of Fig. 3, while those supplied by
2066 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
TABLE IV
RELATIONSHIPS BETWEEN CHARACTERISTIC VALUES AND EQUIVALENT CIRCUIT PARAMETERS
For a discussion on the factors to be used for adjusting this 2) Obtain , which is the low-frequency limit of ,
value, see [2]. then determine
4) Calculate the field resistance referred to the armature
winding (6)
(4)
3) Define an equivalent circuit structure for the q-axis.
4) Use the available parameters and a fitting technique to find
5) Define an equivalent circuit structure for the direct axis. values for the unknown parameters that produce the best
6) Use the available parameters and a fitting technique to find fit for .
values for the unknown parameters that produce the best 5) Adjust to its unsaturated value .
fit for and sG(s). As with the d-axis, these equivalent circuit parameters can be
7) Adjust to its unsaturated value , see normalized to per-unit values.
Section II-B4. The most complex step in the above procedures is the appli-
8) Measure the field winding resistance, convert it to the de- cation of a fitting technique to derive those parameters of the
sired operating temperature, and refer it to the stator equivalent circuits that match the obtained frequency response.
Many techniques have been proposed for this purpose; some of
(5) the techniques presented during the last years were based on
• maximum-likelihood estimation [10];
where is the operating temperature in C, is the mea-
• noniterative parameter identification procedure [11];
sured field resistance in ohms, and is the average field
• network synthesis technique [12];
winding temperature in C during the measurement.
• vector fitting [13].
Note that and substitute the parameters used to find
unknown values during the fitting procedure. Once the param- The experience gained with SSFR testing and analysis of
eters have been determined, they can be normalized to per-unit salient pole machines was presented at a panel session held at
values. the 1997 IEEE PES WM and summarized in [14]; see also [15]
The procedure for identification of q-axis parameters could and [16]. However, SSFR tests have also disadvantages, as dis-
be as follows [2]. cussed below [17], [18].
1) Use the best available estimate for the stator leakage in- • The effect of eddy current losses on the armature resis-
ductance . tance during the SSFR is not accounted for when the
MARTINEZ et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS—PART IV 2067
TABLE V
STANDSTILL FREQUENCY-RESPONSE TESTS
operating reactances are deduced using the following the system as near to zero as possible and the excitation
expression: system on manual control [17], [20]. The generator cir-
(7) cuit breaker is opened, and the transient in the terminal
voltage, field voltage, and current are recorded. To obtain
where is the dc armature resistance; unsaturated and saturated values, the test is carried out
• Test equipment requires very linear, very high power under both under and overexcited conditions.
amplifiers. 3) Time-domain small disturbance test: The linear parame-
• Standstill measurements are made at low currents; how- ters of the machine are identified from lightly loaded, un-
ever, and can vary up to 20% in the range derexcited conditions, while saturation characteristics are
from no load to rated current. identified from a wide range of operating conditions [21].
• Tests are conducted at unsaturated conditions. 4) Time-domain large disturbance test: A large disturbance
• Centrifugal forces on damper windings are not accounted is abruptly introduced in the excitation reference voltage
for, becoming difficult to assess the error introduced by with the machine under normal operating conditions [22].
them. The recorded variables are terminal and field voltages,
• The resistance in the contact points of damper windings armature and field currents, as well as rotor speed.
can be higher at standstill than it is during running. As with any other data conversion procedure, steps involved
3) Online Testing: To avoid some of the drawbacks and lim- when online measurements are provided are similar to offline
itations of offline tests, the so-called online tests have been de- procedures:
veloped. They can be either time-domain or frequency-domain
tests. Some of these tests are described below as follows. • select the order of every rotor circuit;
• assume initial parameter values;
1) Online frequency response test: It is carried out with the
• use an estimation or a fitting technique to adjust their
machine running at rated speed and loaded below full
values.
load. The frequency response is obtained by applying si-
nusoidal signals to the voltage regulator reference and Approaches presented to estimate machine parameters have
measuring the steady-state changes in field voltage and been based, among others, on the Newton–Raphson method
current, the rotor speed, the terminal voltage, and active [19], the weighted least-square method, and the maximum-like-
and reactive power outputs [17], [19]. lihood method [10], [21].
2) Load rejection test: It is performed with the machine 4) Saturation: Saturation effects can significantly affect
running at synchronous speed, with power injected to some transients (e.g., transient stability). In addition, an ac-
2068 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
(8)
(9)
Fig. 4. Saturation characteristic.
• Since an ideal machine is assumed, saturation does not af-
fect the sinusoidal space distribution of the magnetic field
curate representation of saturation is not only important for or the sinusoidal waveform of the induced voltage.
transient simulations; it can be also important for exciter design If the saturation curve is represented as a two-slope piece-
and sizing [6]. wise linear curve, see Fig. 4, the saturation effects can be im-
According to IEEE Standard 1110–1991, the main concerns plemented by using the following coefficient:
for saturation representation can be summarized as follows.
(10)
• Saturation is assumed to affect only direct axis parameters
in salient pole machines, while in solid rotor machines, its
effect can be significant in both direct and quadrature axis where and are the slopes of the unsaturated and satu-
parameters. rated region, respectively.
• Saturation correction can be limited to adjustment of the The coefficient is set to 1.0 in the unsaturated region, but
magnetizing inductances and during the simula- whenever the solution moves to the saturated region, it is used
tion of large disturbances (i.e., disturbances that vary sig- to obtain the saturated values of the mutual inductances
nifycantly from steady-state values).
• During small disturbances, generally related to oscillation
modes in the 0.1-to-10-Hz range, the - path in a solid (11)
iron rotor is different from the path followed during steady
state; therefore, circuit constants are different also; the Much attention has been paid during the last years to the
analysis can be done by replacing the steady-state perme- cross-magnetizing phenomenon: a magnetic coupling exists be-
ability by the corresponding incremental permeability in tween the direct and the quadrature axis when the machine is
every region of the rotor. saturated [24]–[30].
The open-circuit saturation curve can be obtained with the The cross-magnetizing effect was illustrated in [31] using a
machine running offline at synchronous speed. First, set the flux phasor diagram. Fig. 5(a) shows the different saturation
field current to zero. Then give the excitation a small step in- curves. Assume that is a point of the d-axis satura-
crease and measure the steady-state field current and terminal tion curve, is a point of the q-axis saturation curve, and
voltage. Increase the excitation a number of times until 105% is a point of the air-gap saturation curve that is the satura-
of the rated voltage is reached. The set of measured field cur- tion characteristic deduced when the machine is simultaneously
rents and terminal voltages can be plotted to show the saturation excited from both axes. Resolving into its components, it can
characteristic. be found that and are not equal to and [Fig. 5(b)].
The differences are due to the cross-magnetizing effect.
The usual practice in many EMTP-like tools has been to de-
The determination of the saturated reactances including the
termine saturated parameters from the open-circuit saturation
cross-magnetizing phenomenon has been studied by many au-
curve. The representation of saturation effects has been based
thors, see, for instance, [31], [32] and [33]; the procedure de-
on the following simplifying assumptions [4], [23].
scribed below was presented in [34]. This paper includes also
• The flux linkage of both direct and quadrature axis is a method for calculation of the q-axis saturation characteristic
the sum of a mutual flux and leakage flux; saturation from the d-axis characteristic.
affects only mutual fluxes, its effects on leakage fluxes If the total magnetizing current is split into its components
are ignored.
• The rotor is isotropic (i.e., hysteresis effects are
neglected). (12)
MARTINEZ et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS—PART IV 2069
(15)
TABLE VI
INDUCTION MACHINE PARAMETERS
C. Parameter Estimation Procedure This algorithm ignores the core losses and the leakage reac-
As mentioned above, it is common to estimate the electrical tance of the outer cage represented in Fig. 6 by and , re-
parameters from standard specification data [3]. Among the pro- spectively. Therefore, the simplified circuit of an induction ma-
cedures based on standard data, that presented in [37] has be- chine will be that shown in Fig. 7.
come very popular. It was originally developed for implemen- The standard specifications are split into two parts:
tation in the universal machine module of the Electromagnetic • those which define the normal operating conditions at full
Tranisents Program (EMTP) [4]. load: efficiency, power factor, slip;
MARTINEZ et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS—PART IV 2071
IV. CONCLUSION
A significant effort has been made during the last 30 years to
determine rotating machine parameters from test measurements.
Only machine models for low-frequency and switching tran-
sients have been analyzed, in all cases considering a terminal
machine model (i.e., models can be used to simulate the inter-
action between electrical machines and the power system).
Although many reliable conversion procedures aimed at
obtaining electrical machine parameters for the most adequate
model have been developed to date, very few of these proce-
dures are presently implemented in EMTP-like tools.
REFERENCES
Fig. 9. Procedure for induction machine parameter estimation.
[1] CIGRE WG 33.02, Guidelines for representation of network elements
when calculating transients, in CIGRE Brochure 39, 1990.
• those which define the starting performance: starting cur- [2] IEEE Guide: Test Procedures for Synchronous Machines, IEEE Std. 115,
rent and starting torque. 1995.
[3] IEEE Guide: Test Procedure for Polyphase Induction Motors and Gen-
Two additional values are used: the breakdown torque, which erators, IEEE Std. 112, 1996.
is used to determine the saturation threshold current, and the [4] H. W. Dommel, ElectroMagnetic Transients Program. Reference
design ratio, which is defined as Manual (EMTP Theory Book). Portland, OR: Bonneville Power
Administration, 1986.
[5] J. R. Marti and K. W. Louie, “A phase-domain synchronous generator
(16) model including saturation effects,” IEEE Trans. Power Syst., vol. 12,
no. 1, pp. 222–229, Feb. 1997.
The value of this ratio can be approximated to unity for [6] IEEE Guide for Synchronous Generator Modeling Practices in Stability
Studies, IEEE Std. 1110, 1991.
double-cage machines, and between 0.5 and 0.6 for deep-bar [7] I. M. Canay, “Modeling of alternating-current machines having multiple
rotor machines. rotor circuits,” IEEE Trans. Energy Convers., vol. 8, no. 2, pp. 280–296,
The parameter estimation procedure will be then aimed at Jun. 1993.
[8] Rotating Electrical Machines—Part 4: Methods for Determining Syn-
translating standard specification data to the parameters of the chronous Machine Quantities from Tests, IEC 60034-4, 1985.
equivalent circuit shown in Fig. 7 (Table VI). [9] I. M. Canay, “Determination of model parameters of synchronous ma-
chines,” Proc. Inst. Elect. Eng. B, vol. 130, no. 2, pp. 86–94, Mar. 1983.
If the unsaturable and saturable parts of the total leakage in- [10] A. Keyhani and H. Tsai, “Identification of high-order synchronous gen-
ductance are equally divided between the stator and the rotor, erator models from SSFR test data,” in Proc. IEEE/PES Winter Meeting,
the equivalent circuit will be that shown in Fig. 8. Paper no. 94 WM 022-4 EC, New York, Jan. 30–Feb. 3, 1994.
[11] S. Henschel and H. W. Dommel, “Noniterative synchronous machine
Table VII shows the relationships to be used for determining parameter identification from frequency response tests,” IEEE Trans.
induction machine parameters from standard specifications as Power Syst., vol. 14, no. 2, pp. 553–560, May 1999.
defined above. [12] J. Verbeek, R. Pintelon, and P. Guillaume, “Determination of syn-
To determine the parameters shown in Table VII, the fol- chronous machine parameters using network synthesis techniques,”
IEEE Trans. Energy Convers., vol. 14, no. 3, pp. 310–314, Sep. 1999.
lowing facts were taken into account. [13] B. Gustavsen and A. Semlyen, “Rational approximation of frequency
• The efficiency was modified to account for core, friction, domain responses by vector fitting,” IEEE Trans. Power Del., vol. 14,
no. 3, pp. 1052–1061, Jul. 1999.
and windage losses. [14] IEEE PES WG 12, “Experience with standstill frequency response
• Both stator and rotor leakage paths are saturated during (SSFR) testing and analysis of salient pole synchronous machines,”
start up. IEEE Trans. Energy Convers., vol. 14, no. 4, pp. 1209–1217, Dec. 1999.
[15] D. Y. Park et al., “Standstill frequency-response testing and modeling of
• Unsaturable and saturable parts of the leakage reactances salient-pole synchronous machines,” IEEE Trans. Energy Convers., vol.
were divided equally between stator and rotor. 13, no. 3, pp. 230–236, Sep. 1998.
2072 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
[16] R. M. Sanders, “Standstill frequency-response methods and salient pole [30] J. Verbeeck, R. Pintelon, and P. Lataire, “Influence of saturation on esti-
synchronous machines,” IEEE Trans. Energy Convers., vol. 14, no. 4, mated synchronous machine parameters in standstill frequency response
pp. 1033–1037, Dec. 1999. tests,” IEEE Trans. Energy Convers., vol. 15, no. 3, pp. 377–283, Sep.
[17] P. M. Anderson, B. L. Agrawal, and J. E. Van Ness, Subsynchronous 2000.
Resonance in Power Systems. New York: IEEE Press, 1990. [31] A. M. El-Serafi and A. S. Abdallah, “Saturated synchronous reactances
[18] I. M. Canay, “Determination of the model parameters of machines from of synchronous machines,” IEEE Trans. Energy Convers., vol. 7, no. 3,
the reactance operators xd(p), xq(p) (Evaluation of standstill frequency pp. 570–579, Sep. 1992.
response test),” IEEE Trans. Energy Convers., vol. 8, no. 2, pp. 272–279, [32] A. M. El-Serafi and J. Wu, “Determination of the parameters rep-
Jun. 1993. resenting the cross-magnetizing effect in saturated synchronous
[19] P. L. Dandeno, P. Kundur, A. T. Poray, and H. M. Zein El-din, “Adapta- machines,” IEEE Trans. Energy Convers., vol. 8, no. 3, pp. 333–342,
tion and validation of turbogenerator model parameters through on-line Sep. 1992.
frequency response measurements,” IEEE Trans. Power App. Syst., vol. [33] E. Levi and V. A. Levi, “Impact of dynamic cross-saturation on accu-
PAS-100, no. 4, pp. 1656–1664, Apr. 1981. racy of saturated synchronous machine models,” IEEE Trans. Energy
[20] F. P. de Mello, L. N. Hannett, and J. R. Willis, “Determination of Convers., vol. 15, no. 2, pp. 224–230, Jun. 2000.
synchronous machine stator and field leakage inductances standstill [34] N. C. Kar, T. Murata, and J. Tamura, “A new method to evaluate the
frequency response tests,” IEEE Trans. Power Syst., vol. 3, no. 4, pp. q-axis saturation characteristic of cylindrical-rotor synchronous gener-
1625–1632, Nov. 1988. ator,” IEEE Trans. Energy Convers., vol. 15, no. 3, pp. 269–276, Sep.
[21] H. Tsai, A. Keyhani, J. Demcko, and R. G. Farmer, “On-line syn- 2000.
chronous machine parameter estimation from small disturbance [35] S. H. Minnich, “Small signals, large signals, and saturation in generator
operating data,” IEEE Trans. Energy Convers., vol. 10, no. 1, pp. 25–36, modeling,” IEEE Trans. Energy Convers., vol. 1, no. 1, pp. 94–102, Mar.
Mar. 1995. 1986.
[22] R. Wamkeue, I. Kamwa, X. Daï-Do, and A. Keyhani, “Iteratively [36] D. Lindenmeyer, H. W. Dommel, A. Moshref, and P. Kundur, “An in-
reweighted least squares for maximum likelihood identification of duction motor parameter estimation method,” Elect. Power Energy Syst.,
synchronous machine parameters from on-line tests,” IEEE Trans. vol. 23, pp. 251–262, 2001.
Energy Convers., vol. 14, no. 2, pp. 159–166, Jun. 1999. [37] G. J. Rogers and D. Shirmohammadi, “Induction machine modeling for
[23] V. Brandwajn, “Representation of magnetic saturation in the syn- electromagnetic transient program,” IEEE Trans. Energy Convers., vol.
chronous machine model in an electromagnetic transients program,” 2, no. 4, pp. 622–628, Dec. 1987.
IEEE Trans. Power App. Syst., vol. PAS-99, no. 5, pp. 1996–2002, [38] B. K. Johnson and J. R. Willis, “Tailoring induction motor analytical
Sep./Oct. 1980. models to fit known motor performance characteristics and satisfy par-
[24] I. Boldea and S. A. Nasar, “A general equivalent circuit (GEC) of elec- ticular study needs,” IEEE Trans. Power Del., vol. 6, no. 3, pp. 959–965,
tric machines including crosscoupling saturation and frequency effects,” Aug. 1991.
IEEE Trans. Energy Convers., vol. 3, no. 3, pp. 689–695, Sep. 1988. [39] J. R. Willis, G. J. Brock, and J. S. Edmonds, “Derivation of induction
[25] A. M. El-Serafi, A. S. Abdallah, M. E. El-Sherbiny, and E. H. Badawy, motor models from standstill frequency response tests,” IEEE Trans. En-
“Experimental study of the saturation and the cross-magnetizing phe- ergy Convers., vol. 4, no. 4, pp. 608–615, Dec. 1989.
nomenon in saturated synchronous machines,” IEEE Trans. Energy Con- [40] J. A. de Kock, F. S. van der Merwe, and H. J. Vermeulen, “Induction
vers., vol. 3, no. 4, pp. 815–823, Dec. 1988. machine parameter estimation through an output error technique,” IEEE
[26] A. M. El-Serafi and A. S. Abdallah, “Effect of saturation on the steady Trans. Energy Convers., vol. 9, no. 1, pp. 69–76, Mar. 1994.
state stability of a synchronous machine connected to an infinite bus [41] S. I. Moon and A. Keyhani, “Estimation of induction machine parame-
system,” IEEE Trans. Energy Convers., vol. 6, no. 3, pp. 514–521, Sep. ters from standstill time-domain data,” IEEE Trans. Ind. Appl., vol. 30,
1991. no. 6, pp. 1609–1615, Nov./Dec. 1994.
[27] S. A. Tahan and I. Kamwa, “A two-factor saturation model for syn- [42] P. Ju, E. Handschin, Z. N. Wei, and U. Schlücking, “Sequential param-
chronous machines with multiple rotor circuits,” IEEE Trans. Energy eter estimation of a simplified induction motor load model,” IEEE Trans.
Convers., vol. 10, no. 4, pp. 609–616, Dec. 1995. Power Syst., vol. 11, no. 1, pp. 319–324, Mar. 1996.
[28] E. Levi, “Modeling of magnetic saturation in smooth air-gap syn- [43] J. Holtz and T. Thimm, “Identification of the machine parameters in a
chronous machines,” IEEE Trans. Energy Convers., vol. 12, no. 2, pp. vector-controlled induction motor drive,” IEEE Trans. Ind. Appl., vol.
151–156, Jun. 1997. 27, no. 6, pp. 1111–1118, Nov./Dec. 1991.
[29] , “Saturation modeling in d-q axis models of salient pole syn- [44] J. Stephan, M. Bodson, and J. Chiasson, “Real-time estimation of the
chronous machines,” IEEE Trans. Energy Convers., vol. 14, no. 1, pp. parameters and fluxes of induction motors,” IEEE Trans. Ind. Appl., vol.
44–50, Mar. 1999. 30, no. 3, pp. 746–759, May/Jun. 1994.
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005 2073
I. INTRODUCTION
with series or parallel spark gaps can also be represented with
T HE functions of a surge arrester are to do nothing, which
is to conduct little or no current for normal operating volt-
ages, or conduct current during overvoltages, without causing
the model described below where the device has two curves:
1) prior to sparkover, and 2) after sparkover.
a fault. Thus, the surge arrester must have an extremely high MO varistor materials have a temperature dependence that is
resistance during normal system operation and a relatively low evident only at low current densities. Temperature dependence
resistance during transient overvoltages. In other words, it must does not need to be represented in simulations for typical over-
have a nonlinear voltage versus current (V-I) characteristic. voltage studies where the arrester currents exceed 10 A. The
Early overvoltage protective devices used spark gaps con- temperature dependence factors into the selection of arrester rat-
nected in series with discs made with a nonlinear silicon-car- ings for steady-state and temporary overvoltages. The tempera-
bide (SiC) material. The spark gaps provide the high impedance ture-dependent V-I characteristic is important only for the eval-
during normal conditions while the SiC discs impede the flow of uation of energy absorbed by the surge arrester, and should not
current following sparkover. The V-I characteristic of SiC-type influence the insulation protective margins.
surge arresters is a combination of both the SiC disc and the gap
The V-I characteristic depends upon the waveshape of the
behavior.
arrester current, with faster rise times resulting in higher peak
The metal-oxide (MO) varistor material used in modern high-
voltages. Table I shows modeling guidelines derived from
voltage surge arresters has a highly nonlinear voltage versus
CIGRE WG 33–02 [1]. The commonly used frequency-inde-
current characteristic as shown in Fig. 1. This characteristic ob-
pendent surge arrester model is appropriate for simulations
viates the need for series spark gaps. Therefore, the electrical
characteristics are determined solely by the properties of the containing low and most switching frequencies (Groups I and
MO blocks. A higher voltage rating is achieved by adding disks II). However, a frequency-dependent model should be used
in series. Higher energy ratings are achieved by using larger di- when very high frequencies are simulated (Groups III and IV);
ameter discs or parallel columns of discs. such a model has to incorporate the inherent inductance of the
Several manufacturers of medium-voltage applications also surge arrester. A lumped inductance of about 1 H per meter
market MO surge arresters with spark gaps. MO surge arresters for the ground leads should also be included in models for high
frequencies.
The subsequent sections present the rationale for models rep-
resenting MO surge arresters in low- and high-frequency tran-
Manuscript received March 1, 2004. Paper no. TPWRD-00107-2004. sients simulations. Model characteristics, parameter determina-
Task Force Members: J. A. Martinez (Chairman), D. W. Durbak, B. Gus-
tavsen, B. Johnson, J. Mahseredjian, B. Mork, R. Walling. tion, if needed, and an illustrative example are presented for
Digital Object Identifier 10.1109/TPWRD.2005.848771 each model.
0885-8977/$20.00 © 2005 IEEE
2074 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
TABLE I
GUIDELINES TO REPRESENT METAL-OXIDE SURGE ARRESTERS
(1)
Fig. 5. MOV surge arrester model for fast front surges [6].
B. Other Models
Interested readers should also consult other models presented
in the literature, see [4]–[13]. All models incorporate an in-
Note that values presented in Fig. 6 and Table II have been ductor to represent a frequency-dependent behavior.
scaled from those presented in [6], by using the factor 1.6 pro- • One of the first models was presented in [4]; the equivalent
posed in the original reference. circuit consisted of a series combination of a nonlinear
Formulas to calculate parameters of the circuit shown in resistor and a nonlinear inductor (Fig. 7). An alternative
Fig. 5 were initially suggested in [6]. They are based on the approach for parameter determination was later presented
estimated height of the arrester, the number of columns of MO in [12].
disks, and the curves shown in Fig. 6. • The model initially presented in [9] was later simplified
The information required to determine the parameters of the and adopted by a CIGRE Working Group [10]. The equiv-
fast front model is as follows: alent circuit was reduced to a series combination of a linear
• height of the arrester (in meters); resistor, a nonlinear resistor, and a linear inductor.
• number of parallel columns of MO disks; • A simplified version of Durbak’s model was proposed in
• discharge voltage for a 10 kA, 8/20 s current (in [13] (Fig. 8). According to the authors of this model, the
kilovolts); capacitance can be eliminated since its effect is negli-
• switching surge discharge voltage for an associated gible, and the two resistances and can be replaced
switching surge current (in kilovolts). by a single resistance of about 1 M , placed between
Linear parameters are derived from the following equations: model terminals.
A procedure to determine parameters was presented for all of
(2) the above models.
Fig. 9. Fast front arrester model with IEEE example parameters [11].
Fig. 11. Test results with a concave lightning surge current waveshape.
(a) 10-kA, 8/20-s time-to-crest current surge. (b) 10-kA, 2/20-s time-to-crest
current surge.
REFERENCES
[1] CIGRE Working Group 33.02, “Guidelines for representation of net-
work elements when calculating transients,” CIGRE Brochure 39, 1990.
[2] H. W. Dommel, Electromagnetic Transients Program Manual (EMTP
Theory Book). Portland, OR: Bonneville Power Administration, 1986.
[3] E. C. Sakshaug, “Influence of rate-of-rise on distribution arrester pro-
tective characteristics,” IEEE Trans. Power App. Syst., vol. PAS-98, no.
2, pp. 519–526, Mar./Apr. 1979.
[4] S. Tominaga et al., “Protective performance of metal oxide surge arrester
based on the dynamic v-i characteristics,” IEEE Trans. Power App. Syst.,
vol. PAS-98, no. 6, pp. 1860–1871, Nov./Dec. 1979.
[5] M. V. Lat, “Analytical method for performance prediction of metal oxide
surge arresters,” IEEE Trans. Power App. Syst., vol. PAS-104, no. 10, pp.
2665–2674, Oct. 1985.
[6] D. W. Durbak, “Zinc-oxide arrester model for fast surges,” EMTP
Newslett., vol. 5, no. 1, Jan. 1985.
[7] C. Dang, T. M. Parnell, and P. J. Price, “The response of metal oxide
surge arresters to steep fronted current impulses,” IEEE Trans. Power
Del., vol. PWRD-1, no. 1, pp. 157–163, Jan. 1986.
[8] D. W. Durbak, “The choice of EMTP surge arrester models,” EMTP
Newslett., vol. 7, no. 3, Sep. 1987.
[9] W. Schmidt et al., “Behavior of MO-surge-arrester blocks to fast tran-
Fig. 12. Influence of the current slope. (a) Current surges. (b) Arrester sients,” IEEE Trans. Power Del., vol. 4, no. 1, pp. 292–300, Jan. 1989.
voltages. [10] A. R. Hileman, J. Roguin, and K. H. Weck, “Metal oxide surge arresters
in AC systems. Part IV: Protection performance of metal oxide surge
arresters,” Electra, no. 133, pp. 132–144, Dec. 1990.
V. ELECTROTHERMAL MODELS [11] IEEE Working Group on Surge Arrester Modeling, “Modeling of
metal oxide surge arresters,” IEEE Trans. Power Del., vol. 7, no. 1, pp.
All previous models were mainly developed to determine 302–309, Jan. 1992.
residual voltages between arrester terminals. The energy ab- [12] I. Kim et al., “Study of ZnO arrester model for steep front wave,” IEEE
sorption capability is another important issue when selecting Trans. Power Del., vol. 11, no. 2, pp. 834–841, Apr. 1996.
[13] P. Pinceti and M. Giannettoni, “A simplified model for zinc oxide surge
an arrester. The thermal behavior can therefore be of concern. arresters,” IEEE Trans. Power Del., vol. 14, no. 2, pp. 393–398, Apr.
Several models based on an analog electrical equivalent circuit 1999.
have been proposed to include the thermal part of the arrester, [14] IEEE Guide for the Application of Metal-Oxide Surge Arresters for Al-
ternating-Current Systems, IEEE Std. C62.22, 1997.
see, for instance, [5] and [18]. Parameter determination for [15] H. J. Li, S. Birlasekaran, and S. S. Choi, “A parameter identification
these models is based on experimental measurements. technique for metal-oxide surge arrester models,” IEEE Trans. Power
Del., vol. 17, no. 3, pp. 736–741, Jul. 2002.
[16] F. Heidler, J. M. Cvetic, and B. V. Stanic, “Calculation of lightning cur-
VI. CONCLUSION rent parameters,” IEEE Trans. Power Del., vol. 14, no. 2, pp. 399–404,
Apr. 1999.
An adequate model of an MO surge arrester for low-fre- [17] IEEE TF on Parameters of Lightning Strokes, “Parameters of lightning
quency and slow front overvoltages can be described by its strokes: A review,” IEEE Trans. Power Del., vol. 20, no. 1, pp. 346–358,
nonlinear V-I characteristic. The implementation of such a Jan. 2005.
[18] A. Petit, X. D. Do, and G. St-Jean, “An experimental method to de-
model in a transients program is straightforward from the termine the electro-thermal model parameters of metal oxide arresters,”
manufacturer’s ratings and characteristic curves. IEEE Trans. Power Del., vol. 6, no. 2, pp. 715–721, Apr. 1991.
IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005 2079
Abstract—A detailed dynamic arc model may be used to roughly transient recovery voltage (TRV) withstand capability for
evaluate the interrupting capacity of a breaker and its influence the breaker.
on the deformation of the interrupted current. In some cases 2) A more elaborate model considers the arc as a time-
the correct computation of the actual arcing time is of crucial
importance for assessing the first current-zero crossing of the varying resistance or conductance. The time variation
breaker. Arc models in conjunction with surrounding network is determined ahead of time based on the breaker char-
details are also used to understand complex arc instability prob- acteristic and perhaps upon the knowledge of the initial
lems. This paper recalls gas circuit breaker models and presents interrupting current. This model can represent the effect
data requirement for such models. The document includes some of the arc on the system, but requires advanced knowl-
illustrative examples and typical data.
edge of the effect of the system on the arc. Arc parameters
Index Terms—Circuit breakers (CB), modeling, power system are not always easy to obtain and the model still requires
transients, simulation. the use of precomputed TRV curves to determine the
adequacy of the breaker.
I. INTRODUCTION 3) The most advanced model represents the breaker as a dy-
namically varying resistance or conductance, whose value
TABLE I
MODELING GUIDELINES FOR CBs [3]
scribing arc behavior using parameters with different physical cooling power. An extra parameter is needed to specify the
interpretation. number of breaks per pole.
Basic descriptions of arc behavior were first described by Parameters for this model have been derived from experi-
Cassie and Mayr. mental tests and this model has been successfully used and val-
• The Cassie model is given by the following equation: idated in [5], [10], [12], and [13].
This model can be used to represent thermal failure near cur-
rent interruption and conductivity in the post-arc region. Re-
(1) sistance instability near current interruption can cause current
chopping. Although parameters for this model are best derived
It assumes an arc channel with constant temperature, cur- from short-line fault (SLF) breaking tests, it is, however, feasible
rent density, and electric field strength. Changes of the arc to provide some typical data. It has been observed through prac-
conductance result from changes of arc cross section; en- tical cases that the applicability range of (4) is actually longer
ergy removal is obtained by convection. than the near current-zero region.
• The Mayr model is given by the following equation: Although in some cases it is feasible to simulate arc model
equations using control diagram blocks and sufficiently small
time steps, the best solution is achieved using a hard-coded arc
(2)
model with sophisticated iterative techniques, as explained in
[5]. The highly nonlinear model of (4) still requires a small
This model assumes that changes of arc temperature are integration time step to correctly account for its time constants
dominant, and size and profile of the arc column are con- and achieve nonlinear solution method convergence. Such
stant. Thermal conduction is the main mechanism of en- a small time step (0.1 s typically) requires a prohibitive
ergy removal. computer time for statistical studies. Another difficulty is the
where is the arc conductance, is the arc voltage, is the availability of parameters from the breaker manufacturer. This
arc current, is the arc time constant, is the steady-state is why a simplified arc model must be used in statistical studies
power loss, and is the constant part of the arc voltage. The [10], [17]. Such a model combines a current-arc voltage char-
is in the region of 1 s and the is between 0.1 and acteristic per break and a time-dependent function
0.5 s . These parameters are not strictly constant for an to account for blast pressure, arc length, and arcing time.
actual arc, but observations indicate that during the brief time Contrary to the Avdonin equation, this simplified model cannot
around current-zero, these parameters vary sufficiently slowly account for thermal reignition, but it can correctly compute
to assume them to be constant. A combination of both models the breaker’s arcing time with a larger integration time step
gives the Cassie-Mayr model (typically 70 s). The functions and must be
provided by the manufacturer.
Other arc models are presented in the Appendix.
(3)
V. CONCLUSION
The resistance of a CB dynamic arc during the thermal period
can pass from zero to infinity in a few microseconds. Several
models have been developed and implemented for duplicating
the performance of a CB near current zero.
Modeling of such nonlinear behavior must be based on pa-
rameters determined from accurate measurements. Validation
must be also carried out with accurate models of the test cir-
cuits. In addition to the arc model, it is essential to correctly
model the network surrounding the breaker. Sensitivity analysis
approaches must be used to observe complex arc-network inter-
Fig. 6. Kilometric fault strength of the modeled arc.
action problems.
Some default values of arc model parameters are provided in
the literature. However, users must be warned about the validity
of these values; they should be considered as an initial estima-
tion or just a reference.
APPENDIX
OTHER ARC MODELS
1) The Urbanek model can represent arc interruption and
both thermal and dielectric failure [14]. Both current
chopping and reignition are also represented. It is charac-
terized by the following equation for the arc conductance:
(10)
[2] L. van der Sluis, W. R. Rutgers, and C. G. A. Koreman, “A physical arc [11] A. V. Avdonin et al., Some problems of EHV and UHV air-blast circuit
model for the simulation of current zero behavior of high-voltage circuit breakers, in CIGRE 1980 Session. Paper 13-04.
breakers,” IEEE Trans. Power Del., vol. 7, no. 2, pp. 1016–1022, Apr. [12] G. St-Jean, M. Landry, M. Leclerc, and A. Chenier, “A new concept in
1992. post-arc analysis to power circuit breakers,” IEEE Trans. Power Del.,
[3] CIGRE Working Group 13.01, “Applications of black box modeling to vol. 3, no. 3, pp. 1036–1044, Jul. 1988.
circuit breakers,” Electra, no. 149, pp. 40–71, Aug. 1993. [13] G. St-Jean and R. F. Wang, “Equivalence between direct and synthetic
[4] J. A. Martinez-Velasco, “Circuit breaker representation for TRV calcu- short-circuit interruption tests on HV circuit breakers,” IEEE Trans.
lations,” in EEUG Meeting, Hannover, Germany, Nov. 13–15, 1995. Power App. Syst., vol. PAS-102, no. 7, pp. 2216–2223, Jul. 1983.
[5] J. Mahseredjian, M. Landry, and B. Khodabakhchian, “The new EMTP [14] L. Blahous, “The problem of defining a test circuit for circuit breakers
breaker arc model,” in Proc. IPST, Seattle, WA, Jun. 22–26, 1997, pp. in terms of prospective voltage,” Proc. Inst. Elect. Eng., vol. 126, no. 12,
245–249. pp. 1291–1294, Dec. 1979.
[6] J. Kosmac and P. Zunko, “A statistical vacuum circuit breaker model for [15] , “Derivation of circuit breaker parameters by means of Gaussian
simulation of transients overvoltages,” IEEE Trans. Power Del., vol. 10, approximation,” IEEE Trans. Power App. Syst., vol. PAS-101, no. 12,
no. 1, pp. 294–300, Jan. 1995. pp. 4611–4616, Dec. 1982.
[7] M. T. Glinkowski, M. R. Gutierrez, and D. Braun, “Voltage escalation [16] E. Thuries, P. Van Doan, J. Dayet, and B. Joyeux-Bouillon, “Synthetic
and reignition behavior of vacuum circuit breakers during load shed- testing method for generator circuit breakers,” IEEE Trans. Power Del.,
ding,” IEEE Trans. Power Del., vol. 12, no. 1, pp. 219–226, Jan. 1997. vol. PWRD-1, no. 1, pp. 179–184, Jan. 1986.
[8] D. A. Woodford and L. M. Wedepohl, “Impact of circuit breaker [17] B. Kulicke and H. H. Schramm, “Clearance of short-circuits with de-
pre-strike on transmission line energization transients,” in Proc. IPST, layed current zeros in the Itaipu 500 kV substation,” IEEE Trans. Power
Seattle, WA, Jun. 22–26, 1997, pp. 250–253. App. Syst., vol. 99, no. 4, pp. 1406–1414, Jul./Aug. 1980.
[9] CIGRE WG 33.02, Guidelines for Representation of network elements [18] B. Khodabakhchian, J. Mahseredjian, M.-R. Sehati, and M. Mir-Hos-
when calculating transients, in CIGRE Brochure 39, 1990. seini, “Potential risk of failures in switching EHV shunt reactors in some
[10] Q. Bui-Van, B. Khodabakhchian, M. Landry, J. Mahseredjian, and J. one-and-a-half breaker scheme substations,” in Proc. IPST, New Or-
Mainville, “Performance of series-compensated line circuit breakers leans, LA, Sep. 28–Oct. 2, 2003.
under delayed current-zero,” IEEE Trans. Power Del., vol. 12, no. 1,
pp. 227–233, Jan. 1997.
2086 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
Abstract—Power-electronics converters are presently used at user to view it as a “black box.” In some transient simulations,
all voltage levels. The representation of power converters and detailed modeling of the power semiconductor devices is re-
their control units in transient simulations may be made using quired to accurately assess switching transients, device stresses,
different modeling levels, from a detailed representation of each
semiconductor device to an average representation of the con- and losses. However, in many cases, simpler, “ideal” switch
verter without explicitly modeling semiconductor devices at all. models can be used for modeling the power semiconductor
When semiconductors are included, data specification can be a devices [1], [4].
critical step. Gathering data required to model power converters The objective of this paper is to present the data requirements
in transient simulations can be a difficult task. This paper deals for modeling power-electronic devices within power converters
with data requirements for modeling power semiconductors when
using an EMTP-like tool. The document includes a summary of for transient simulation studies. Different levels of models for
guidelines for modeling power converters, proposes a procedure power converters and their power-electronic devices will be pre-
for creating approximate semiconductor models and includes a sented along with guidelines for choosing the appropriate level
detailed study of two common semiconductor devices. A discus- of modeling detail. As part of this discussion, detailed device
sion on the application of very detailed models using specialized models will be presented. This will be followed by a discus-
software packages is also included.
sion on where to get appropriate device data for different levels
Index Terms—Modeling, power system transients, semiconduc- of modeling and how to convert these data for use in the device
tors, simulation.
models. This paper targets applications of power semiconductor
devices commonly used in medium-to-high-power applications:
I. INTRODUCTION power diodes, insulated-gate bipolar transistors (IGBTs), thyris-
tors (SCR), and gate turn-off thyristors (GTOs). Low-power,
P OWER converters are used in many applications in power
systems, both in the power delivery system and as part
of end-use applications [1], [2]. Power delivery applications
single-phase loads are beyond the scope of this paper.
TABLE I
SEMICONDUCTOR CHARACTERISTICS
priate. In other cases, the snubber circuit can be ignored, al- IV. SWITCH MODELING
though a numerical snubber may be needed as part of the switch
model with some transients programs. An example is any of the A. Ideal Device Models
standard library diode models in SPICE-based simulators; to en- The converter terminal characteristics are often sufficient for
able the dc analysis requires a numerical snubber. many simulations involving power converters. In such cases,
Similarly, the degree of detail in the power-electronic device it is often appropriate to model series and parallel-connected
model depends on the relationship between the time periods power-electronic devices as one or two equivalent devices. For
of the frequencies of interest and the switch transition times example an HVdc converter could have 60 thyristors connected
of the power semiconductor devices. The latter are usually in series in each switch location, but the converter could be rep-
much shorter than the intervals between converter switching resented with a single thyristor modeled in each switch loca-
operations. There will be transients associated with these tion. It is sometimes sufficient to represent a converter made
turn-on/turn-off transitions. If these transients (or transients up of many converter modules as a simpler converter. For ex-
with similar frequencies) are of interest, then more detailed ample, a 48-pulse VSC could be represented with a simpler,
device models will be required. When the slowest transition lower pulse-order model if the response is sufficient for the
time of the power semiconductor device (which is often at studies to be performed. If the converter is connected to a system
turn-off) approaches the period between switch operations, where the time scales of the dynamic response of interest are
more detailed device turn-on and turn-off models are required. very long compared to the device turn-on and turn-off times,
ideal switch models can be used [1]. In this case, the power-elec-
Use of these models also requires more detail in modeling the
tronic device is assumed to open or close in one time step, as the
parasitic inductances and capacitances in the converter.
simulation progresses (essentially instantaneously as far as the
Simpler or aggregate device models can be used in many
external system is concerned).
cases when performing most power systems transients simula-
The behavior of an ideal switch device models can be sum-
tions [1]. In these cases, the converter can be reduced to a sim- marized as follows (Table I and [1]):
pler equivalent as discussed in Section IV-a.
• when the device is off, it behaves as an open circuit;
• when the device is on, it behaves as a short circuit;
• the device turns on at the next time step after a firing com-
III. POWER SEMICONDUCTOR DEVICES
mand;
The following power semiconductor devices are most • the device turns off at the next time step after a firing com-
common in power system simulations: mand, or for diodes and thyristors at the next time step
after the next current zero crossing;
• power diodes; • switch transition time is equal to one simulation time step.
• thyristors;
These models can be applied when:
• GTO thyristors;
• frequencies of interest are much slower than switch
• IGBTs.
turn-on and turn-off times;
The following semiconductor devices may see increased use
• series/parallel combinations of devices into one equivalent
in the near future: switch are acceptable;
• gate commutated thyristors (GCT/IGCTs); • converter losses, device voltage stresses, and device cur-
• MOS-controlled thyristors (MCTs); rent stresses are not important.
• MOS turn-off thyristors (MTOs);
• static induction transistor/thyristor (SIT/SiTh). B. Detailed Device Models
Each type of device listed above has specific turn-on and
More detailed device models are required in other circum-
turn-off characteristics that are visible in the voltage and current stances, usually in cases where the transient response of the con-
characteristics. See Table I for typical characteristics of some verter and the immediate converter subsystems are of interest.
commonly used devices. Each will have conduction losses while Examples of situations where more detailed switch models are
the device is turned on. In some devices, the conduction losses needed include:
are best modeled as a steady-state voltage drop; in others, it
• studies for switching and conduction loss prediction;
appears more as a somewhat nonlinear, temperature-dependent • simulations to evaluate voltage and current stresses on the
resistance. power-electronic devices;
Models representing these characteristics can be added to • simulations of converters with high switching frequencies
EMTP-like programs. However, some degree of approximation and slow devices;
will be necessary in most cases to work within the constraints • electromagnetic-interference (EMI) studies;
of these programs. In some cases, resistances, inductances, and • thermal analysis;
dependent voltage sources are sufficient to represent device be- • design of device protection.
havior. In other cases, more detailed equations describing de- In these cases, a reasonably accurate switch model is critical
vice behavior are used to create device models to include in the to the performance of the study. Note that the cases mentioned
circuit. above are usually of more interest to the converter designer.
JOHNSON et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS. PART VII: SEMICONDUCTORS 2089
Fig. 6. IGBT circuit model using voltage and current sources [19].
C. IGBT Model and fall is approximated as being linear with time. Occasionally,
two time segments are employed [1], but data sheets ordinarily
The IGBT is a hybrid device, behaving at its input like a do not give enough information for such a model. Such data may
MOSFET and at its output, like a BJT. Hence, the models tend be gathered empirically if needed [1].
to be the same as common models for those two devices as il- Because the IGBT exhibits MOSFET behavior on its gate
lustrated in Fig. 3. The IGBT turns on by first turning on the side, gate capacitance may be important in finding storage be-
MOSFET section of the device, which draws a base current haviors. A simple series RC model formed by the gate-emitter
through the IGBT section. (The parasitic BJT shown in Fig. 3 capacitance of the IGBT and the series resistance of the
is normally not conducting and will turn on only at very high drive circuit may be appropriate. is the input capacitance
values of current through the main IGBT.) To turn the IGBT on some data sheets. Some gate drive models also use the
off, a similar procedure to turn-on occurs. The MOSFET sec- Miller capacitance ( on the data sheet). Fig. 7 illustrates
tion is gated off, which extinguishes the base current in the BJT these elements. The notation is similar to MOSFETs, and the
section. An ideal device model may be appropriate in situations circuit is the same.
described in Section IV of this paper.
Model calculations are a sequence of piecewise linear circuit
Piecewise linear RC models are common. Appropriate
calculations. As conditions are met for transition from one cir-
switching models of this nature are found in [1]. A fairly simple
cuit approximation to the next, the appropriate changes in the
example of an IGBT model in forward conduction, illustrating
linear circuit models must be made. Otherwise, calculations are
voltage- and current-source models, is shown in Fig. 6. A
governed by linear circuit theory. Under blocking conditions, for
circuit model of the input of an IGBT under turn-on conditions
example, the small leakage current allows an IGBT to be mod-
is shown in Fig. 7. There are four sequential modes of operation
eled as an open circuit. During turn-on, current rise is modeled
during turn-on; conduction progresses from one mode to the
as a ramp current source. Rise transients may be influenced by
next depending on the state of charge of the two capacitors.
antiparallel diodes, particularly during diode reverse recovery
The model and the degree of complexity chosen depends upon
and also in the case in which the diode is significantly slower
what behaviors the user wants to simulate.
than the IGBT. Transitions between piecewise linear modes can
As was the case for the diode, it is also true that most of the
be governed by a number of conditions or inputs, for example
necessary data is available from data sheets. Forward voltage
drops are usually given both as nominal values and as a graph • passive switching (as shown in the diode model of Fig. 5);
against current with temperature contours. Because device pro- • active switching (as shown in the case of the parasitic BJT
tection methods often use forward voltage drop as an indicator in Fig. 3);
of overcurrent, it is important to capture a good representation • conditions on circuit elements (such as the charge on the
of the voltage versus current behavior, particularly for slight to capacitors of Fig. 7);
moderate overcurrents. Rise and fall times of the currents are • recognition of crossing a current or voltage level (such as
prominently listed on IGBT data sheets. These are more signif- what terminates most turn-on or turn-off piecewise linear
icant than voltage rise and fall times because current behavior modes).
is normally so much slower as to render voltage rise and fall As was the case for the diode, average conduction losses in
times nearly instantaneous. For most IGBT models, current rise the IGBT are the product of the switching frequency (or the
JOHNSON et al.: PARAMETER DETERMINATION FOR MODELING SYSTEM TRANSIENTS. PART VII: SEMICONDUCTORS 2093
switching period T, which is the inverse of the switching fre- expressed as a set of nonlinear first-order differential equations.
quency), the forward voltage drop , the forward current MATLAB/SIMULINK works best on simulations of system dy-
, and the duty cycle namics and control algorithms. These simulators work on other
types of problems and often do quite well, but their power is
(3) usually best manifested within the class of problems for which
they are designed.
The time the product of the duty ratio and the switching These models provide remarkably accurate predictions of cir-
period . cuit behavior. Loss estimates and voltage and current transients
Switching losses are the product of the IGBT output voltage obtained from simulations based on these models are usually
and the IGBT collector current during switching quite good and match up with experimental results closely.
The most significant tradeoff involved in using these detailed
(4) models, as one might expect, is the simulation time required.
It does not require many devices in a circuit to lengthen the
where and are the rise time and the fall time of the current simulation time excessively. It is often necessary (and worth
through the device during turn-on and turn-off, respectively. the effort) to simplify the rest of the circuit as much as possible
If a constant bus voltage and a linear rise and fall of current to reap the benefits of the detailed models and sophisticated
are assumed, then the average switching losses are the product simulation techniques available in these software packages.
of the bus voltage, the load current, and half the sum of the rise
and fall times. Instantaneous losses are, as always, the product VIII. CONCLUSION
of the device voltage and current at hand.
Thyristors and GTOs tend to exhibit behaviors of both passive Several different levels of detail are available when mod-
and active switches. The concepts in building models for them eling power-electronic converters. In some cases, average, non-
appear as a combination of the models for the diode and IGBT. switching models will suffice. However, switch-based models
In the interest of brevity, the details are presented in Chapters are required in many cases. The ideal switch models available in
23 and 24 of [1]. most EMTP-like programs are sufficient for most power system
simulations requiring switching models.
In cases where a more detailed switch model is required,
VII. MORE DETAILED MODELS the user can employ an approximate model or a highly de-
If piecewise linear models are inadequate for finding switch tailed equation-based model. Several options for implementing
behavior and its effect upon system behavior, more detailed approximate models were described. Approximate models
models exist in specialized software packages, such as SPICE are often easier to implement in fixed time-step simulation
and Saber. Both of these programs can utilize variable time step packages.
numerical integration models that are better suited to the accu- These approximations may also be sufficient since the appro-
rate modeling of turn-on and turn-off behavior of power semi- priate data on the devices, passive circuit elements, and controls
conductor devices than a fixed time step method. SPICE also for a specific converter may be difficult to obtain. When the spe-
serves as the engine for several circuit capture software pack- cific device type and model numbers are known, parameter data
ages, such as Electronics Workbench. Most device manufac- for the devices are available through the manufacturers specifi-
turers will send a SPICE or Saber model upon request; some cation sheets. If the specific device model numbers are not avail-
post the most commonly requested models to their company able, model data can be approximated by assuming: 1) the kind
websites. These models tend to be “plug and play.” The inter- of device, 2) voltage, current, speed class of the device, and then
face is usually quite simple, often merely being an icon in a using data for similar devices. Very detailed device models are
schematic capture. The specifics of how these models were de- available for Saber and SPICE, but these models will not port
veloped sometimes appears in the literature, such as when stan- to EMTP-like programs easily and will require very small inte-
dards agencies develop the models [13]–[15]. Unfortunately, gration time steps.
this information is more often proprietary when developed by However, approximate device models are often adequate
private individuals or companies. Nonetheless, a little under- for most applications where EMTP-like programs will be
standing of the fundamental purposes and methods of the sim- used. Keys for the program user are 1) know the application
ulation software at hand lends a great deal of flexibility to the and define the level of detail well and 2) if a detailed power
simulation. For example, understanding a SPICE netlist yields semiconductor device model is used, is the rest of the model
an ability to modify the model under changing conditions (e.g., accurate enough now?
temperature, voltage,and current levels, etc.). An understanding
of the underlying simulation algorithms helps one to identify REFERENCES
artifacts and numerical errors. Knowing the history of simula-
[1] N. Mohan, T. M. Undeland, and W. P. Robbins, Power Electronics: Con-
tion software packages can help one to select the appropriate verters, Applications, and Design, 2nd ed. New York: Wiley, 1995.
tools. For example, simulators employing SPICE engines work [2] P. T. Krein, Elements of Power Electronics. London, U.K.: Oxford
best on electronic circuits because its developers intentionally Univ. Press, 1998.
[3] N. G. Hingorani and L. Gyugyi, Understanding FACTS: Concepts and
employed algorithms and formulations that were suited to the Technology of Flexible AC Transmission Systems. New York: IEEE
typical behaviors of such circuits. ACSL works best on systems Press, 2000.
2094 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005
[4] IEEE TF on Power Electronics (L. Tang, Chairman), “Guidelines for [12] R. Kraus and H. J. Mattausch, “Status and trends of power semiconduc-
modeling power electronics in electric power engineering applications,” tors device models for circuit simulation,” IEEE Trans. Power Electron.,
IEEE Trans. Power Del., vol. 12, no. 1, pp. 505–514, Jan. 1997. vol. 13, no. 3, pp. 452–465, May 1998.
[5] N. Mohan, W. P. Robbins, T. M. Undeland, R. Nilsen, and O. Mo, “Simu- [13] D. W. Berning and A. R. Heffner Jr., “IGBT model validation,” IEEE
lation of power electronics and motion control systems—An overview,” Ind. Appl. Mag., vol. 4, no. 6, pp. 23–34, Nov./Dec. 1998.
Proc. IEEE, vol. 82, no. 8, pp. 1287–1302, Aug. 1994. [14] A. R. Heffner Jr. and D. M. Diebolt, “An experimentally verified IGBT
[6] D. W. Novotny and T. A. Lipo, Vector Control and Dynamics of AC model implemented in the saber circuit simulator,” IEEE Trans. Power
Drives. London, U.K.: Oxford Univ. Press, 1996. Electron., vol. 9, no. 5, pp. 532–542, Sep. 1994.
[7] D. W. Novotny and T. King, “Equivalent circuit representation of current [15] A. R. Heffner Jr., “Modeling buffer layer IGBT’s for circuit simulation,”
driven synchronous machines,” IEEE Trans. Power App. Syst., vol. PAS- IEEE Trans. Power Electron., vol. 10, no. 2, pp. 111–123, Mar. 1995.
100, no. 6, pp. 2920–2926, Jan. 1981. [16] M. Rashid, Microelectronic Circuits: Analysis and Design. Boston,
[8] R. J. Kerkman, D. Leggate, and G. L. Skibinski, “Interaction of drive MA: PWS-Kent, 1999.
modulation and cable parameters on AC motor transients,” IEEE Trans. [17] M. H. Rashid, SPICE for Circuits and Electronics Using PSPICE, 2nd
Ind. Appl., vol. 33, no. 3, pp. 722–731, May/Jun. 1997. ed. Englewood Cliffs, NJ: Prentice-Hall, 1995.
[9] M. Melfi, A. M. J. Sung, S. Bell, and G. L. Skibinski, “Effect of surge [18] A. Hefner, “An investigation of the drive circuit requirements for the
voltage risetime on the insulation of low-voltage machines fed by power insulated gate bipolar transistor (IGBT),” IEEE Trans. Power
PWM converters,” IEEE Trans. Ind. Appl., vol. 34, no. 4, pp. 766–775, Electron., vol. 6, no. 2, pp. 208–219, Apr. 1991.
Jul./Aug. 1998. [19] K. Sheng, S. J. Finney, and B. W. Williams, “New analytical IGBT
[10] J. A. Martinez-Velasco, R. Abdo, and G. Capolino, “Advanced represen- model with improved electrical characteristics,” IEEE Trans. Power
tation of power semiconductors using the EMTP,” in Proc. IPST, Lisbon, Electron., vol. 14, no. 1, pp. 98–107, Jan. 1999.
Portugal, Sep. 3–7, 1995, pp. 505–510.
[11] C. Wong, “EMTP modeling of IGBT dynamic performance for power
dissipation estimation,” IEEE Trans. Ind. Appl., vol. 33, no. 1, pp. 64–71,
Jan./Feb. 1997.