You are on page 1of 7

# IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 22, NO.

3, JULY 2007

1799

## Electromagnetic Transient Simulations

Task Force on Frequency Domain Methods for Transient Studies
Working Group on Modeling and Analysis of System Transients Using Digital Simulation
General Systems Subcommittee, IEEE Power Engineering Society
Taku Noda, Member, IEEE, and Abner Ramirez, Member, IEEE

## AbstractThis is an introductory paper on -transform-based

methods for electromagnetic transient simulations of power systems. Since the theory of the -transform was originally developed
for the analysis of time series data defined at equidistant time
steps, simulation models developed using -transforms can readily
be used in electromagnetic transient simulations based on the
same time step. First, this paper briefly introduces the basics of the
-transform, and then applications to electromagnetic transient
simulations of power systems are reviewed.

II.

-TRANSFORM

Definition
In Fig. 1, the solid line shows a continuous-time waveform
, and it is sampled at equidistant time steps as marked by
the circles. The time scale is normalized by the sampling interval
. We first define a train of impulses (Diracs delta functions)
as

## Index TermsElectromagnetic transient analysis, equivalent

circuits, power systems, signal processing, switching transients,
time series, transforms.

(1)
Using

I. INTRODUCTION
-TRANSFORM has both frequency-domain and time-domain properties. It has a close relationship with the Laplace
transform. In fact, the -transform of a waveform can be obtained by simple replacement of a variable in its Lapace transform expression. Thus, a -transformed function can be considered as a frequency-domain expression. However, at the same
operator implies a one-sample delay in the time
time, the
domain.
The theory of the -transform was originally developed for
the analysis of time-series data defined at equidistant time steps.
A pioneering work in the development of the -transform theory
for sampled-data control systems can be found in [1]. Reference
[2] is a good introductory text book focusing on discrete-time
signal processing. For electromagnetic transient (EMT) analysis
of a power system, Humpage et al. [3][6], Hauer [7], Angelidis
and Semlyen [8], and the first author of this paper [9], [10] applied -transforms for the modeling of transmission lines. Girgis
et al. [11][13] and Wang and Watson [14] used -transforms
for obtaining a reduced-order equivalent of a power network.
First, this paper briefly introduces the basics of the -transform, and then the above mentioned applications to EMT simulations of power systems are reviewed.
Manuscript received May 2, 2006. Paper no. TPWRD-00236-2006.
Task Force Members: Luis Naredo (Chairperson), Abner Ramirez, Akihiro
Ametani, Alberto Gutirrez, Andrea Mansoldo, Ani Gole, Antonio Lima, Atef
Morched, Bjrn Gustavsen, David Wilcox, Felipe Uribe, Fernando Moreira,
Francisco de Len, Juan Martnez, Leonardo Guardado, Marisol Dvila, Michel
Ritual, Naoto Nagaoka, Neville Watson, Pablo Gmez, Pablo Moreno, Reza
Iravani, Sandoval Carneiro, Taku Noda, Venkata Dinavahi, Victor Ortiz, Washington Neves.
Digital Object Identifier 10.1109/TPWRD.2006.886793

## , the sampled signal

can be expressed by
(2)

## and, thus, its Laplace transform is

(3)
If we use the notation

for

and define
(4)

## equation (3) can be brought into

(5)
This is the definition of the -transform of
and denoted
. As shown in the above, the
hereafter by
-transform of a waveform can be obtained simply by replacing
the complex frequency by using (4) in its Laplace transform
expression.
A. Mapping Between the

and

Planes

## Equation (4) maps the complex plane into the complex

plane as shown in Fig. 2. An interesting property of this mapping
is that the left-hand side of the plane is mapped inside the unit
circle on the plane. The stability of a discrete-time system
can be assessed by checking whether all poles of the system
are inside the unit circle or not. A discrete-time system having

1800

## in the do2) Shift Theorem: Multiplying the operator

main corresponds to shifting the time-domain waveform in the
for
positive time direction by one time step. Thus,
is equivalent to delaying the waveform
by samples.
3) Initial Value Theorem: The initial value of a waveform
can be obtained by its -domain function with
(7)
4) Final Value Theorem: The final value of a waveform can
be obtained by the following formula:
(8)
Fig. 1. Continuous-time waveform is plotted by the solid line, and it is sampled
at equidistant time steps as marked by the circles. The time scale is normalized
by the sampling interval.

## 5) Periodic Function: Let us assume that

be a single(othershot waveform having values only for
with
wise zero). The -transform of the periodic function
for each period is given by
(9)

C. Inverse -Transform
Here, three methods to obtain the time-domain waveform
from its -domain expression
are introduced.
with
1) Power Series: If the power series expansion of
is obtained
respect to

## Fig. 2. Laplace s plane (a) and the z plane (b).

all poles inside the unit circle on the plane is stable, since the
corresponding continuous-time system having all poles on the
left-hand side of the plane is stable. The perimeter of the unit
of the plane ( is
circle corresponds to the imaginary axis
the angular frequency).
Another interesting property is the fact that the mapping is not
one by one. Equation (4) indicates that if the angular frequency
is increased along the imaginary axis of the plane, the corresponding locus on the plane turns around on the perimeter
of the unit circle any number of times. The sampling theorem
tells that frequency components higher than the critical angular
cannot be preserved by sampled data.
frequency
of the upper
This is equivalent to that the end point
perimeter of the unit circle corresponds to
;
if the angular frequency is increased beyond this point, the locus
overlaps on the lower perimeter that is the frequency response
for negative frequencies. Aliasing is also closely related to
this topic.
B. Properties of -Transform

(10)
then
.
can be ex2) PFE (Partial Fraction Expansion): If
panded into partial fractions of the form (here, we describe a
contains first-order
method applicable to the case where
fractions only)
(11)
then, using the -transform
by

is given

(12)
3) By Definition: It can be proved that the inverse -transis analytically obtained by
form of
(13)

## 1) Linearity: -transforms are linear, and thus the following

relation is satisfied:
(6)

.
where is a closed loop enclosing the convergent area of
In most cases, the integration above can be evaluated by applying Cauchys residue theorem.

1801

(19)

## mulThis indicates that the output can be calculated by

additions, which are very efficient if the
tiplications and
model order is small. Equation (19) is closely related to the
concept of recursive convolution which was originally developed for transmission-line modeling [15]. The same arithmetic
operation as (19) is called an IIR (Infinite Impulse Response)
filter in the field of digital filter design and also an Auto-Regressive Moving-Average (ARMA) model in the field of time-series
analysis.

## The following approximate relation between and is often

useful for replacing with in a frequency-domain expression:

B. Norton Equivalent

Fig. 3. Norton and Thevenin equivalent circuits derived by the rational function
of z
in (17).

(14)

## is considered to be an admittance, (19) can be

If
rewritten in the form
(20)

(15)

where

(21)

## This can be interpreted into the Norton equivalent circuit shown

in Fig. 3(a), which is compatible with the nodal conductance
formulation used in EMTP-type programs.

## III. BASIC CONCEPTS

A. Rational Function and Recursive Convolution
The inputoutput relation of a linear time-invariant system
can be described by

C. Thevenin Equivalent
If

## is an impedance, (19) can be rewritten in the form

(16)
is the transfer function of the system and
and
are the input and the output, respectively. If
and
are, respectively, considered to be the voltage and the current of a network component,
is the admittance of the comand
are the current and the voltage,
ponent. If
is the impedance. In most -transform applications to EMT simis replaced (approximated)
ulations, the transfer function
by a rational function of
in the form
where

(17)
Some methods identify the coefficients
,
in the frequency domain by fitting the frequency response
of (17) with given
, and others in the time domain by
matching the time response of (17) with recorded or calculated
data. Let us assume that (17) has been identified by some
identification method. Substituting (17) into (16) gives

(18)

(22)
where
(23)
and can be interpreted into the Thevenin equivalent circuit
shown in Fig. 3(b).
D. Representation of Wave Propagation Function
When the propagation function
of a transmission line
of length with its frequency dependence is modeled in an
into a
EMTP-type program, it is common to decompose
corresponding to the traveling time and
pure time delay
representing the wave deformation
a transfer function
(24)
is replaced by (17), the wave
If the deformation function
deformation in the time domain can efficiently be calculated by
term does not
(19). Since the time delay is associated, the

1802

## have to be represented separately as in the equivalent circuits

shown in Fig. 3.
IV. APPLICATIONS TO EMT SIMULATIONS
A. Modal-Domain Transmission Line Modeling
The dynamics of an -phase transmission line is described
by coupled partial differential equations (PDEs). The modal
theory proposed in [16] can be used to transform the coupled
PDEs into decoupled PDEs which can be solved independently. Since each PDE describes a natural mode of propagation, the decoupled domain is called the modal domain. Voltages
and currents
in the phase coordinates are interfaced
and
in the modal domain by the linear
with those
transformation
(25)
where
and
are called the voltage and current transformation matrices. When the frequency dependence due to the
skin effects of the conductors and the ground is neglected, Bergerons solution for each modal PDE can be expressed in the following form in the time domain [17]:

(26)
where and
are the current and the voltage at the sending
end, and and
are those at the receiving end. The characteristic admittance is denoted by , and the current sources
and
are calculated by

(27)
Next, we consider the case where the frequency dependence is
taken into account, and thus, the characteristic admittance and
the propagation function are frequency dependent. If the frequency-dependent characteristic admittance is denoted by
in the frequency domain and the propagation function
is
and a wave deformadecomposed into a pure time delay
as described in Section III-D, (26) and (27)
tion function
are formally rewritten using Laplace transforms in the form

## have already been calculated in (28), the convolutions due only

are calculated. As explained in Section III-D, the conto
and
are
volutions are carried out by (19), and the outputs
added to the current sources of the Norton equivalents at both
ends. Finally, for each mode, we obtain the -domain equivalent
and are approxcircuit shown in Fig. 4(a). Assuming that
imated to be real and constant, the phase- and modal-domain
quantities are interfaced using (25) as shown in Fig. 4(b). In this
way, a modal-domain transmission-line model is implemented
in an EMTP-type program.
The remaining task is to identify rational functions of
for
and
. Humpage et al. [3][6] first fit given frewith the rational function of
quency response of
(30)
using a two-stage least-squares procedure. They use
(31)

(28)
(29)
denotes an inverse Laplace transform. The inverse
where
Laplace transforms in (28) and (29) can also be denoted by
can be denoted by
convolutions. For instance,
, where
is the impulse response of the characteristic admittance. Let us assume that
and
are,
in (17). Then,
respectively, replaced by rational functions of
for the
according to Section III-B, the terms
for the receiving end in (28)
sending end and
can be realized respectively by Norton equivalents shown in
and
Fig. 3(a). Regarding (29), since the terms

## for the characteristic impedance (instead of the characteristic

(32)
Equations (30) and (31) are then transformed into the -domain
.
using BLT in (14) to obtain rational functions of
Hauer [7] uses
(33)
and
in the frequency domain via a nonto fit both
linear optimization technique, and then BLT is used to obtain
rational functions of
.

1803

## Fig. 6. Example of a phase-domain deformation function with time-domain

discontinuity.
Fig. 5. z -domain equivalent of a phase-domain transmission-line model (the
three-phase case is shown here but this equivalent is applicable to general
n-phase case).

## B. Phase-Domain Transmission-Line Modeling

In a modal-domain model, the modal transformation matrices
are assumed to be real and constant. This assumption is appropriate for most overhead lines. However, the transformation
matrices of some vertically-arranged overhead lines and underground cables show significant frequency dependence. Phasedomain transmission-line models [8][10] have been proposed
to avoid errors due to the use of real constant transformation matrices for such transmission lines by directly formulating equations in the phase domain rather than in the modal domain. In
(28) and (29), all quantities have been scalar for the modal-doas vectors
main formulation. If we consider , , , and
with length of the currents and voltages at the sending and
and
as by
receiving ends, respectively, and
matrices, (28) and (29) become the equations for a phase-domain formulation and this leads to the equivalent circuit shown
in Fig. 5.
and
with the
Angelidis and Semlyen [8] replace
matrix rational function of
(34)
which directly realizes a matrix-vector convolution, where
and
are the coefficient matrices of size by . The elements
of the coefficient matrices are identified by a frequency-domain
term of
is used for the
least-squares procedure. The
in the equivalent circuit shown in Fig. 5.
conductance matrix
On the other hand, the first author of this paper [9], [10] reand
by a rational function
places each element of
in (17), and the terms of
are used to form the
of
by conductance matrix
in Fig. 5. The coefficients of the
are identified by a frequency-domain
rational functions of
least-squares procedure.
It is interesting to note that the model proposed in [9] explicitly represents modal traveling-time differences by utilizing the
operator corresponds to one sample delay in
fact that the
the time domain. The phase-domain propagation function of an
-phase transmission line consists of modal components with
different traveling velocities. This means that the phase-domain
deformation function still exhibits discontinuities in the time domain due to the modal traveling-time differences after removing

Fig. 7. Linear interpolations are applied both to the input and the output of a z
function to interface with an external network with a different time step.

the traveling time of the fastest mode as in (24). Such a discontinuous response can accurately be fitted by

(35)
where is the number of modal components with distinct traveling times and the modal traveling-time differences are repreterms. The numbers of numerator and desented by the
nominator coefficients are the same, and thus the sum of
for
is equal to
. For instance, a
three-phase overhead line has the three modes; an earth-return
mode and two aerial modes. The earth-return mode is slower
than the aerial modes, and the velocities of the aerial modes are
usually almost the same as shown in Fig. 6. Thus, this can be
(the two aerial modes are represented
fitted by (35) with
by one numerator term).
Another interesting attempt proposed in [10] is to apply interpolations both to the input and the output of the calculation
in (19) as shown in Fig. 7, in order to interface with an external
network with a different time step. This practically overcomes
one of the most disadvantageous points of -transform methods;
an identified model is tied to a specific time step.
C. Reduced-Order Network Equivalents
A reduced-order network equivalent is basically used to represent a large subnetwork to reduce computation requirements,
especially in the following situations.
When a large number of simulation cases have to be
performed with different parameter values, a sub-network
which is not affected by the parameter values, is replaced
by a reduced-order equivalent to reduce computation time.

1804

## Fig. 8. Network for the application example.

When the equivalent circuit of a sub-network (or a network component such as a transformer for high-frequency
studies) is not known but its frequency or time response is
known, and if the sub-network can be replaced by one of
the equivalent circuits shown in Fig. 3, then the simulation
can be carried out.
The model proposed in [11][13] uses the Thevenin equivalent
shown in Fig. 3(b) and the coefficients of the rational function
in (17) are identified by a least-squares procedure in the time
domain. On the other hand, the model in [14] uses the Norton
equivalent shown in Fig. 3(a) and the coefficients of the rational
function are identified by a least-squares procedure formulated
in the frequency domain.
V. APPLICATION EXAMPLE
The network for this example is shown in Fig. 8. The voltage
source corresponds to
with
,
and
and its internal resistance and inductance are
respectively. In this network, there are three 10-km,
15-m height, single-phase transmission lines (TLs) represented
by their frequency domain admittance matrix:
(36)
where
propagation constant, and the length.
The per-unit-length series impedance and shunt admittance
of the TLs are calculated considering the skin effects of both
the conductors and the ground soil. At the receiving end of each
and
line there is an - (in parallel) load with
. The frequency response of the driving-point admittance
seen from Node 1 is calculated, and based on that response a
reduced-order network equivalent in the rational function form
in (17) is identified in the following way.
The basic idea of the identification is to determine the coef,
of the rational function model
ficients
using a least-squares procedure formulated in the frequency domain [9]. However, the model order has also to be identified
at the same time. Thus, is increased from 1 to 10, and for each
order the least-squares identification process mentioned above

## Fig. 9. Frequency response of the driving-point admittance seen from Node 1

of the network shown in Fig. 8.

## Fig. 10. Current entering Node 1 of the network shown in Fig. 8.

is repeated. The model order that gives the smallest rms error
is finally chosen. In those cases yielding unstable poles their
reciprocals were used. In this way, a network equivalent with
has been identified. Since the system shown in Fig. 8 includes distributed-parameter lines that theoretically possess infinite number of poles, the order of the identified model can be
regarded as a greatly reduced one. Fig. 9 compares the frequency
response of the identified model with that of the given system,
where good agreement is observed in the frequency range from
1 Hz to 1 kHz.
is connected
Assuming zero initial conditions, the source
, yielding the current delivered by the
to the network at
source as shown in Fig. 10. The identified -domain model in
the form of (17) can readily be used as a time-domain ARMA
model in (19), and the waveform shown by the solid line was calculated using the ARMA model. For comparison, the same transient was calculated by a frequency domain technique, i.e., the
Numerical Laplace Transform (NLT) [18], and superimposed
was used for the
using a dashed line in Fig. 10. Since
error tolerance in the NLT calculation, the dashed line can be
considered as the rigorous solution. The result by the identified
reduced-order equivalent closely reproduces the rigorous solution.
VI. CONCLUSION
This paper has introduced the basics of the -transform and
applications to EMT simulations have been reviewed. The application of the -transform to EMT analysis is proposed as a

## complimentary tool for the existent time and frequency domain

methods. However, more research is needed in the case of representing nonlinear elements and switched devices in the -domain. The authors hope that this material will help the progress
of research and development in this field.

REFERENCES
[1] J. R. Ragazzini and L. A. Zadeh, Analysis of sampled data systems,
Trans. AIEE, vol. 71, pt. II, pp. 225234, 1952.
[2] A. V. Oppenheim, R. W. Schafer, and J. R. Buck, Discrete-Time Signal
Processing, 2nd ed. Englewood Cliffs, NJ: Prentice Hall, 1999.
[3] W. D. Humpage, K. P. Wong, T. T. Nguyen, and D. Sutanto, z-transform electromagnetic transient analysis in power systems, Proc. Inst.
Elect. Eng. C, vol. 127, no. 6, pp. 370378, Nov. 1980.
[4] W. D. Humpage, K. P. Wong, and T. T. Nguyen, Development of
z-transform electromagnetic transient analysis methods for multimode
power networks, Proc. Inst. Elect. Eng. C, vol. 127, no. 6, pp. 379385,
Nov. 1980.
[5] , Time convolution and z-transform methods of electromagnetic
transient analysis in power systems, Proc. Inst. Elect. Eng. C, vol. 127,
no. 6, pp. 386394, Nov. 1980.
[6] , z-transform electromagnetic transient analysis of crossbonded
cable transmission systems, Proc. Inst. Elect. Eng. C, vol. 128, no. 2,
pp. 5562, Mar. 1981.
[7] J. F. Hauer, State-space modeling of transmission line dynamics via
nonlinear optimization, IEEE Trans. Power App. Syst., vol. PAS-100,
no. 12, pp. 49184925, Dec. 1981.
[8] G. Angelidis and A. Semlyen, Direct phase-domain calculation of
transmission line transients using two-sided recursions, IEEE Trans.
Power Del., vol. 10, no. 2, pp. 941949, Apr. 1995.
[9] T. Noda, N. Nagaoka, and A. Ametani, Phase domain modeling of frequency-dependent transmission lines by means of an ARMA model,
IEEE Trans. Power Del., vol. 11, no. 1, pp. 401411, Jan. 1996.
[10] , Further improvements to a phase-domain ARMA line model
in terms of convolution, steady-state initialization, and stability, IEEE
Trans. Power Del., vol. 12, no. 3, pp. 13271334, Jul. 1997.
[11] A. A. Girgis, J. Qiu, and R. B. McManis, A time-domain approach for
distribution and transmission network modeling, IEEE Trans. Power
Del., vol. 5, no. 1, pp. 365371, Jan. 1990.

1805

## [12] A. A. Girgis, W. H. Quaintance, III, J. Qiu, and E. B. Makram, A

time-domain three-phase power system impedance modeling approach
for harmonic filter analysis, IEEE Trans. Power Del., vol. 8, no. 2, pp.
504510, Apr. 1993.
[13] J. Qiu, H. Chen, and A. A. Girgis, Dynamic modeling and parameter
estimation of a radial and loop type distribution system network, IEEE
Trans. Power Del., vol. 8, no. 2, pp. 483490, May 1993.
[14] Y. P. Wang and N. R. Watson, Z-domain frequency-dependent
AC-system equivalent for electromagnetic transient simulation, Proc.
Inst. Elect. Eng., Gen. Transm. Distrib., vol. 150, no. 2, pp. 141146,
Mar. 2003.
[15] A. Semlyen and A. Dabuleau, Fast and accurate switching transient
calculations on transmission lines with ground return using recursive
convolutions, IEEE Trans. Power App. Syst., vol. PAS-94, no. 2, pp.
561571, Mar./Apr. 1975.
[16] L. M. Wedepohl, Application of matrix methods to the solution of
travelling-wave phenomena in polyphase systems, Proc. Inst. Elect.
Eng., vol. 110, no. 12, pp. 22002212, Dec. 1963.
[17] H. W. Dommel, Digital computer solution of electromagnetic transients in single- and multiphase networks, IEEE Trans. Power App.
Syst., vol. PAS-88, no. 4, pp. 388399, Apr. 1969.
[18] L. M. Wedepohl, Power system transients: Errors incurred in the
numerical inversion of the laplace transform, in Proc. 26th Midwest
Symp. Circuits and Systems, Aug. 1983, pp. 174178.
Taku Noda (M97) received the B.Sc., M.Sc., and Ph.D. degrees in engineering
from Doshisha University, Kyoto, Japan, in 1992, 1994, and 1997, respectively.
From 2001 to 2002, he was a Visiting Scientist at the University of Toronto,
Toronto, ON, Canada. He is also an Adjunct Professor at Doshisha University.
His research interest includes electromagnetic transient analysis of power systems and analysis of surges using electromagnetic-field computations.

Abner Ramirez (M96) received the B.Sc. degree from the University of Guanajuato, Mexico, in 1996, the M.Sc. degree from the University of Guadalajara,
Guadalajara, Mexico, in 1998, and the Ph.D. degree from the Center for Research and Advanced Studies of Mexico (CINVESTAV) Campus Guadalajara,
in 2001.
He was a Postdoctoral Fellow in the Department of Electrical and Computer
Engineering, University of Toronto, Toronto, ON, Canada, from 2001 to 2005.
Currently, he is a Professor at CINVESTAVGuadalajara. His interests are electromagnetic transient analysis in power systems and numerical analysis of electromagnetic fields.