- i i;~~Ert
Fig. 1. Single-phase transmission line connecting 12
nodes k and m. E
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
For the frequency-domain equivalent of (4), the variables bk,
etc. are simply replaced by Bk(j), etc. Applying this t
al(t)+
transformation to the well-known line equations in the t) +
frequency domain,8 _ + a2(t
^ k 2 trs2 Z
Vk(jc) = cosh
rsinh T m- cos
(y cd) Vm(jc) -
Zosinh (yd)
' Im(j")
m'J'-'b ZI
Earth~~~~~a2m
Ik(j -) =
(csinh(yd) Vm(j) - cosh ( d) Im(j _
(5)
R + jcL Fig. 4. Shock-excitation of line yields ai(t), a2(t).
Zo= y = '(R+jcoL)(G`+jcoC)
G + jcoC
the terminal voltages to ground provide a physical measurement
results in of the weighting functions:
Bk(ic) = A2(ic)Fk(jco) + Al(jo)Fm(jO) al(t) = 2vm(t)
Bm(jco) = A1(j)Fk(jco) + A2(Ico)Fm(i) )
(6a)
a2(t) = 2vk(t)
~~~~~~~~~~~~~(for
t >0O) (10)
1- 1 1 1 1
a z t in microseconds t in microseconds
_ s ~~~~~~~~
Fig. 5. Sample weighting functions ai(t) and (b)
(a)
a2(t), for
Fig. 3. Equivalent circuit representing (4) and (7a). zero-sequence- mode of 10-mile, 500-ky 'line.
1402
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
Close Switch at t = O The small total area under a2 is produced largely by
cancellation of the positive and negative areas (both of which
=3] are about 0.06 in magnitude).
Vk
r_*'
w
J 2(y)d t
I conductance G [mho/m] is easy. Fig. 6 still applies,
though now Fig. 7 has the lumped resistance R-d replaced
0 Vm= 1 2
J
Z1
- O|(y)dy y 1
[a by a distributed-parameter resistive network. Solution can be
o obtained using (5) to represent the line, provided one therein
zi > t l l
1V _- |
sets X = 0. The result is to replace the right-hand sides of
(12) by those of (6b).
k-11- ++X
I A ~~~% AA
lv - . R-d +ttk ++ to 0
T vk vm t Z1 Vk
Earth
Fig. 7. Circuit representing (11) as t-o. Fig. 8. Infinitely-long transmission line.
1403
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
~6000
4000 20
C" 6IK \~ ~ ~ ~ ~ ~ ~ ~ ~ ~a(u)
_T
___ __ __\__ __ _l
0 20 40 60 80 loo . _.U__u
Time t in microseconds U0 uo+t&t uo+2A1
Fig. 9. Impulse-response a2(t) for infinite line Fig. 11. Illustration of problem with fast transient
having parameters of Fig. 2. in a(u).
1404
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
using this estimate. Now instead of extrapolating using just a point is clearly simplest, it is not generally the most accurate
history points, it is believed more accurate and consistent to or economical treatment. An improvement is provided by the
interpolate using the currently-unknown point f(t) together assumption of an exponential tail,
with one or more history points. Such is the procedure for
implicit integration. Using just linear interpolation on a(t) = P-eqt (for t >Tmax) (29)
f(t) and f(t-At),
f(t-u) f(t) +
frt)
f(t- At),)
and
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
Point 1: The Modified Fourier Transformation (MFT), 3
recently used by other authors, was rejected on theoretical X - 3 5 -- -
grounds. With R( ) and L(co) only defined at discrete -
points (typically 4 or 5 per decade) as input, and using linear 0.0 - -------
1
interpolation on these to produce the complete curve, it can - - time t
be shown that the MFT generally produces a result which is --- -- -
not independent of convergence parameter "a". Some authors > - 12.5 -
have further simply replaced R(c) +j&.L(co) by -
R(c) + (a +j&,)L(&j)---clearly erroneous, in that the variation - ----- ------_
of R and L has been overlooked during the variable
substitution. Short of evaluation of Carson's infinite series6 Fig. 14. Experimentally-measured voltage at receiving end
using (a +jco) as a variable rather than ji, the MFT would of line, phase b.
seem to yield only a good approximation .at s best.
rejected for this reason. .
. . It was ~~~~~~~~~~~damping, with the peak overvoltage of 3.3 ddivisions ca
excessive. Using higher-frequency constant parameters (237 Hz
clearly
Point 2: The Fast Fourier Transform (FFT)5 is fast, but for
but (b);
has 1000
other Hz for (c)) does of course improve the damping,
the natura-spacing of points in the time domain sometimes shortcomings. The peak overvoltage remains too
turns out to be too large for the desired application. Other large, (3.15 and 3.20 units, respectively), while the steady-state
not-so-obvious complications could be enumerated as well. amplitude becomes excessive as t - oo*. Fig. 15d shows the
Because al(t) and a2(t) are only calculated once, the corresponding result using frequency dependence, wherein the
peak overvoltage is reduced to 2.77 units, and the general
associated numericalpeffort
FFT was dropped as not being justifiableseignifjustificant
not significant anyway. Thus
associated numerical effort isis not
b under
the
Thus the
the waveshape has been
accurate, having beenimproved.
These solutions (a)4d) are highly
computed using At = 50,sec (typical
for accurate production runs at BPA). Doubling At to
Point 3: Regular inverse Fourier transformation was 100,sec introduces only minor error, as shown in (e) and
selected, with the different frequency contributions for each (f). Curves (d) and (e) have no exponential tails, using the
time point calculated one at a time. This allows the code to point storage of (21) over a time span of 6 T . For (e) this
minimal cor storage,independentorequired
execute in inimal
executen s ag iepen e ofhwsall the
core (6r + (6-1) 90 cellswhich
T ) /At, is small approximately
of memory, compared withequal to
current
frequency increment Acoi may be. Extreme accuracy thus program capacity of. 4000. The exgqnential tails of (f)
costs in execution time only, not in core storage requirements. contained areas A1ail =-051 and A2t =-.045, reducing the
point from 90 to 42 cells. The central-processor
Point 4: Uniform spacing of frequency points for inverse solutionstorage
time for a CDC- 6400 is given in the following
transformation is generally inadequate. The uniform tabulation:
Aco normally picked would be sufficient to track the
variation of integrands Al(jw) and A2(j0o) if R and
L were constant. But near zero frequency, the Case (a) - (c) (d) (e) (f)
L(co) variation is such that drastic changes typically occur
over [0, Aco ].- It was found that the use of
geometrically-increasing step size for maybe 50 steps over the pRu time-mstec,
per time-step
.0056 .0175 .0118 .0087
range [ 0, 5Aco ] resolves the aforementioned difficulty.
Thereafter, uniform spacing Aw is used.
As a simple illustrative application, consider a
single-line-to-ground fault
on an open-ended, 3-phase, 500-kV
f4
overhead line of length 138 miles (see Fig. 13). A field
measurement of this test is shown in Fig. 14----the oscillogram °
for phase -b voltage at the receiving end of the line, vb(t).
The model of Fig. 13 is somewhat oversimplified, but serves to
A;
illustrate the digital simulation. Using a typical value of/
p = 100 Q_-m for ground resistivity, line constants for the c 0
zero and positive sequence at 60 Hz are calculated to be 0
Ro =.3055Q92/mi Lo =5.819 mH/mi CO= .0121,uLF/mi X 0 R E l2L f|/ \
303/1200 3 _i
303/240° ~ 39.8 nH b
m
398 H a i2i2QtU,1 1
303/ 0
x '1 __ ohn
~~~J Lower 0
-
10 1 20 30 40 50 60
Day Monumental -Ti me t in msec
Fig. 15. Digital simulations of vb(t) for problem of
Fig. 13. Configuration of test problem. Fig. 13. Details are given in the text.
1406
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
IX. CONCLUSIONS x). For anything happening at x = 0 (node k), the former appears as a
cause, the latter as an effect; Cj (X) represents the incident wave and
Convolution theory has been shown to provide a simple, C2(w) the reflected wave in k:
efficient, and accurate representation for the
frequency-dependent zero-sequence parameters of transposed Ci = -(Vk - ZO Ik) (a)
transmission lines. The numerical techniques presented generally 2
impose no time-step restrictions not present in the C2 = Vk Cl (b)
constant-parameter problem. As a result, selective use of
frequency-dependence does not add great numerical burden to From Eq. (16), for x = d (with I(d) = - Im)
typical transient studies. The BPA transients program ' now yd
has this representation available as a standard feature. 2 (Vm + Zo Im) = CI eYd = C2m (c)
X. REFERENCES where C2m represents the reflected wave from node m. In (c)
1. H. W. Dommel, "Digital computer solution of 'Y = +Ayo (d)
electromagnetic transients in single- and multi-phase where is the purely imaginary propagation constant of the lossless
networks.", IEEE Trans. Power Apparatus and Systems, line. Equations (a) and (c) can be combined into
o
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
Luke Yu (c/o IECO, San Francisco, Calif.): The authors are to be means that the FFT gives a more accurate solution than that of the CFT
congratulated for their fine work on studying line transients. It has be- by
come important to have frequency dependent parameters included in the Comparison of computation tine
analysis.
By means of modal transformation, the multiphase system is trans- number of samples computation time (sec.)
formed into an n-single conductor system. In fact the eigenvalue matrix
of the modal transformation is obtained from the line impedance and 128 CFT FFT
admittance matrix which are frequency dependent. The authors' com- 256 117 6
ment of this effect on the result of transient study is appreciated. 512 300 11
Regarding the illustrative application as shown in Fig. 13 of the 1024 1200* 24
paper, since the analysis of 3 phase system is based on the modal trans-
formation, then the 2 ohm-resistor simulating the tower footing resist- Computer:
*: theoretica'lICL 1906A
estimation
ance becomes a matrix form with mutual elements through the same
modal transformation in the analysis. I would appreciate the authors' the same computation time. The FFT may give a better solution than
detail description on how to handle it in the study. that of the CFT even with geometrically-increasing step size. The authors
stated that because a 1(r) and a2(T) are only calculated once, the associ-
Manuscript received February 11, 1974. ated numerical effort is not significant anyway. In my program of switch-
ing surge calculation which is one of travelling-wave techniques (4,5),
the total computation time for a 3-phase sequential closing on a similar
system to Fig. 13 with maximum observation time 40 ms and time step
M. Ramamoorty (University of Waterloo, Waterloo, Ontario, Canada): 40 ps was 2.5 sec. when using the CFT for evaluating step responses,
The authors describe a very interesting and novel approach to include the and was reduced to 2.0 sec. by the FFT. (Computer CDC 7600) Thus,
variation of line parameters with frequency, in an electromagnetic tran- the computation tinme of the numerical Fourier inversion for evaluating
sient programme. In reference 1 the discusser has given a modification step responses or for evaluating al(r) and a2(r) is a quite significant
to the Uram and Miller method2 to include the same parameter variation part of the total computation. What is the computation time of a I (r)
with frequency. The following assumptions which are identical to those and a2(T) in comparison with the total computation time of one case
made by the authors are used. shown in Fig. 15?
1. Line ig ideally transposed. Line resistance is lumped and skin The numerical evaluation of convolution is the most time consum-
effect is neglected. ing part in line transient calculations by travelling-wave techniques. This
2. Parameter variation is considered only for the zero sequence computation time can be greatly reduced, if one of two functions to be
mode. convoluted is given by
3. Shunt conductane is neglected.
4. Carson's equations are used for parameter calculation. Al A2 + +. Am
Computations are made in real time so that any switching opera- H( = A + - + Am
tions can be easily incorporated. Non linear elements at the line termi- 0 p + a1 p + a2 p + am
nals can be included if the V-I characteristics are known.
The discusser feels that the method of Ref. I is more simple than
that described by the authors in the paper. In a large system it may be where p is Heaviside operator and h(t) = £-1 H(p).
necessary to accurately represent only one or two lines in which a dis- Then convolution g(t) = f(t) * h(t) is approximated by (6)
turbance like a lightning discharge, a shunt fault or a switching opera-
tion has occured. It may be possible to use Uram and Millers's method m
with the modification suggested by the discusser for the study of wave g(t) = Aof(t) + z Ak - Bk(t): simplified convolution
propagation on these lines and for the rest of the system the method sug- k= 1
gested by Dr. Dommel in Ref. 1 of the paper may be used. The discusser
would like to know the opinion of the authors. where
REFERENCES Bk(t) ~ 2 { f(t) + f(t - A\t) exp (akA\t)f A\t + Bkc(t - A\t) *exp (-akAt)
1. M. Ramamoorty; Approximate method for including ground ef- B(t)
fects in wave propagation on transmission lines. Proc. IEEE Vol. =k0 0
120 1 973, pp. 702-703.
2. Uram R. Miller R. W. Mathematical analysis and solution of trans-.-* -
mission line transients. IEEE Transactions PAS 1964, pp. 1116-1123. For example, if the authors' weighting function a(r) is represented by
Manuscript received February 15, 1974.a()=A 1-ex(ar}+A21ep(arI
Then, Eq. (19) is evaluated by the simplified convolution.
A. Ametani (Doshisha University, Kyoto 602, Japan): Concerniing the Eq. (19)~ Z Ak * Bk(t)
authors' comments on numerical inverse Fourier transformation, k=l
point 1: It is true that a solution given by the modified Fourier trans- Thus, the computation time can be easily reduced to less than one tenth.
form is not independent of constant 'a' to change an integral route. How- The detail is given in reference (6).
ever, it has been found from a long experience of using the MFT that
there is an optimum value of 'a' and this, in most cases, gives a quite RFRNE
good solution (1,2). A minor change of 'a' from the optimum value RFRNE
causes no significant effect on the solution. When dealing with frequency- l11] 5. Day, N. Mullineux and J. Reed, "Developments in obtaining
dependent impedances by the MFT, it should be clear~from the theory transient response using Fourier transforms, Pt. II: Use of modified
of the MFT that R&.o) + jwL(o) is replaced by R(co - ja) + jQX- ja) * Fourier transform", Int. J. Elec. Eng. Educ., vol.4, pp. 3140, 1966.
L(cw - ja). There is no difficulty in the evaluation of the frequency-de- [2] L. M. Wodepohl and S. E. T. Mohamed, "Multiconductor trans-
pendent impsedance using variable (c, -ja). Simply set a complex variable mission lines: Theory of natural modes and Fourier integrals ap-
(w- ja) instead of a real variable X in a computer, and evaluate the plied to transient analysis", Proc. IEE, vol. 116, pp. 1553-1563,
impedance in the same manner as that by w. The MFT gives a better September 1969.
accuracy than that of the conventional Fourier transform (CFT). As an [3] A. Ametani, "The application of the fast Fourier transform to
inherent nature of a numerical frequency-time transformation, the time electrical transient phenomena", Tnt. J. Elec. Eng. Educ., vol. 10,
solution tends to diverge in the later part of the solution,.and therefore pp. 277-287, 1972.
the solution is correct only before the divergence. A solution by the [4] A. Amatani, "Modified travelling-wave technique, to solve electrical
MFT, in general, is effective up to 90% of the total observation time, transients on lumped and distributed constant circuits", Proc. lEE,
while that by the CFT is effective up to 70%o. Thus, the MFT gives more vol. 120, pp. 497-504, April 1973.
effective information than the CFT. The authors' further statement on [5] A. Amatani, "Refraction coefficient method for switching-surge
Point 1 would be appreciated. calculations on untransposed transmission lines", IEEE 1973 PES
On Point 2: The following table shows comparison of computation Summer Meeting, paper C73 444-7.
Dimes by the CFT and the fast Fourier transform (FFT) (3). The table [6] J. Umoto and T. Hara, "A new digital analysis of surge perform-
ances in electric power networks utilizing convolution integral",
Manuscript received February 15, 1974. J. TEE Japan, vol. 91, pp. 907-916, May 1971.
1408
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.
W. Scott Meyer and Hermann W. Dommel: Dr. Ametani raises many distributed nature of line resistance is taken into account in (5) - (7). As
interesting points, worthy of more detailed discussion than can be pro- for the significance of "real time" computation for transients problems,
vided here. Yet concerning Fourier transformation, several observations it is wondered whether "time domain" is not meant instead - the use of
might be made: time as the principle problem variable; instead of frequency or some
1) The authors' Point- 1 comment about MFT was meant to apply other parameter. The authors agree that this is important when modeling
only to the case of R and L defined at discrete points. With 4 points per nonlinear elements (such as transformer hysteresis, surge arresters, ac/dQ
decade and typically 8 decades, this represents a total of 32 points of inverters, etc.).
input data. While such data is sufficiently smooth (see Fig. 2 example) Concerning Dr. Yu's comment about the frequency dependence of
to produce accurate weighting functions by CFT, such has not been the series line impedance and shunt admittance per unit length, the out-
found to be the case with MFT. This in no way contradicts the dis- lined procedure is rigorous provided the line is continuously transposed
cusser's good experience with the MFT, it will be noted. (see point 1, paragraph 1, Sect. VIII). If not, the diagonalizing trans-
2) Tabulated figures such as the discusser's, comparing CFT with formation is frequency dependent, and complications arise. While the
FFT, are common in the FFT literature (see Table I, Ref. 5); and for theory is clear4, practical, efficient computer solution of the general
transformation over a complete cycle of the result, they are valid. But case may require considerable experimentation and adaptation. The
in (7b) the authors were not typically interested in a complete cycle, so numerical solution of Fig. 13 was provided by the BPA Electromagnetic
the CFT was only required to solve a small fraction of the problem in- Transients Program, using the basic procedure of Ref. 1. As phase co-
herent to FFT (and upon which the usual comparisons are based). Using ordinates are used therein for the network formulation, the 2-ohm
a uniform frequency discretization Af = AW/2ir with N points, the FFT resistor stays as a single uncoupled branch, while the transmission line
inherently returns a time function with the same number of points yields the interphase coupling resistances shown in Fig. 16. Note that
spaced from time zero through time Tmax = 1 /Af. For the line of Fig. 5, among all nodes at each end of the line there is complete interconnec-
using N = 2000 and Af = 330 Hz to numerically integrate (7b), one tion, although no coupling from one end of the line to the other end
finds that Tmax = 3000 psec and At = 1.5 psec. Here Tmax is almost exists.
15 times larger than necessary, while the uniform spacing of 1.5 ,see is Prof. Semlyen's most recent research using recursive convolution is
marginal for aI near the peak (see Point 2) and inadequate for a2 near very interesting and relevant. The replacement of point-weighting by
time zero (see Point 4). Also to be considered is the numerical effort recursive updating would indeed make the convolution effort a negligible
required to form the integrand values of (6b) for each frequency - a part of the overall solution process. While the authors used such a
complication generally overlooked in CET vs. FFT comparisons. This procedure in Sect. VII for only the tails of their weighting functions,
represents an equal, added burden for both methods, and is not in- the discusser has now proposed that the entire convolution be handled
significant. by means of such formulas. The idea is exciting, and potentially very
3) Sample computation time for aI and a2 using program default promising, although accuracy in the exponential fitting procedure
tolerances is 51 sec on a CDC-6400 computer (perhaps 5 sec on CDC- might potentially pose a problem. But even if accuracy were less than
7600), based on 2000 frequency-domain points with spacing Awc- 160/r. desired, the availability of a superfast procedure for lines other than the
In theory, this calculation would only be performed once in the life of a principal one (or ones) could be of great utility. One benefit might be to
transmission line. Compared with the 65000 seconds (and 13500 pages increase overall system damping and attenuation with frequency,
of printed output) of current monthly computer usage billed to Transi- probably better approximating the actual system. The authors look
ents Program calculations at BPA, such time is negligible. forward to Prof. Semlyen's work on this subject.
Finally, the authors are indebted to Dr. Ametani for pointing out the
Umoto-Hara reference, now apparently the earliest known application
of closed-form convolution to the problem in question. The reader is JDA MA
referred to Prof. Semlyen's current research using this procedure.
It is difficult to respond to Prof. Ramamoorty, since little quanti-
tative data is available for comparison. Perhaps'a reasonable interpreta-
tion of "simplicity" would involve the effort to program a given algo- _ L '
rithm on a digital computer, while "efficiency"' might be measured by
required core storage and numerical effort for y execution. No data con-
lettr.
cerningCncering he points
either of these ssumtion found intedsueritholbe2i
lised
could be the discusser's Proc. IEE
letter. Concerning the assumptions listed by the discusser, it should be
pointed out that his treatment of line resistance (point 1) would not - -
seem to agree with ours; the skin effect is included in Fig. 2, and the
Fig. 16. Structure of resistive network for Fig. 13 (injected node cur-
Manuscript received April 22, 1974. rents not shown).
1409
Authorized licensed use limited to: UNIVERSIDAD TUCUMAN. Downloaded on June 14,2010 at 20:15:08 UTC from IEEE Xplore. Restrictions apply.