Ieee Transactions Apparatus and Systems, Vol. PAS-97, No. 5, 1978 1607

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

IEEE Transactions on Power Apparatus and Systems, Vol. PAS-97, No.

5, Sept/Oct 1978 1607

A CORRELATION METHOD FOR TRANSMISSION LINE PROTECTION


M. Vitins, Member, IEEE
Brown Boveri Research Center
CH-5401 Baden, Switzerland
ABSTRACT 13 Most methods for determining the fault
location on a power transmission line known
A method of transmission line protec- today are based on the fundamental sinusoidal
tion based on a distributed parameter model voltage and current signal components. Conse-
of the line is outlined. It is shown that quently, the undesired transient signal com-
the fault distance appears as a time delay ponents must be suppressed prior to evalu-
between two quantities associated with the ating the voltage and current waveforms. In
travelling waves at the measurement site.The 4E conventional electromechanical devices, for
time delay can be extracted from the funda- example, the DC-transient can be compensated
mental voltage and current waveforms in the to a certain extent by the use of an induc-
presence of superimposed line transients by K tive lumped parameter line replica while the
means of a correlation technique. The pro- > high frequency oscillations are normally of
posed method has been tested by numerical '< little concern due to the inertia of mechani-
simulations and by means of a real-time cal parts. In solid state relays the problem
analog-type computer on a model transmission N of high frequency transients can be solved by
line and the results were found to be satis- providing adequate but time consuming filter-
factory. The feasability of a digital comput- N ing. Since the late 1960's, when computer
er implementation is briefly discussed. relaying has come to the attention of protec-
! tion engineers, a number of computer-oriented
INTRODUCTION real-time distance protection algorithms have
been developed. These methods are usually
Modern high voltage networks require based on inductive lumped parameter models of
faults to be removed rapidly and selective- the transmission line and therefore require
ly. One major problem encountered with in the elimination of the high frequency tran-
reducing the response time or in increasing >K sients, a task which - despite the use of
the accuracy of relays for power transmis- r digital filtering and data fitting tech-
sion line protection arises from the fact niques - causes an annoying time delay.
that immediately after the occurence of the I

fault transient travelling wave phenomena In order to develop methods of line pro-
distort the steady state voltage and current tection which cope with high frequency tran-
waveforms. w4sients the use of improved mathematical mod-
+>els of the power transmission line has been
Two basic types of transient phenomena x proposed, in which the transients are incor-
can be distinguished. One type of transient porated into the line model rather than being
is due to the discharge of the magnetic ener- ignored. Kohlasl has described a method in
gy which is stored in the source and line *. which the instantaneous profile of the line
inductance and is characterized by an expo- voltage is computed from the measured values
nentially decaying current offset, known as cof the voltage and current waveforms by means
the DC-transient. The other type of transient \1"of the wave or telegraph equation. The point
is due to the sudden discharge of the elec- > along the line for which this voltage profile
trical energy stored in the line capacitance vanishes or for which its magnitude reaches a
and is characterized by damped high frequency minimum is considered to be the fault loca-
oscillations in the voltage and current wave- >Y tion. Another approach proposed by Gel'fand
forms. This type of transient dominates when and Zisman2 is based on the representation of
the fault occurs at the moment of a voltage Ian overhead line by a multi-section lumped
extremum while the DC-transient dominates for &parameter model consisting of up to two sec-
faults initiated at the moment of a zero tions, in which line inductance, capacitance,
crossing of the voltage.With increasing volt- resistance and conductance are included. In
age levels and line lengths the above men- % this method the fault distance is given by
tioned high frequency oscillations, with fre- the roots of a polynomial, the order of which
quencies ranging from ca. 150 to 1000 Hz, ranges from one to four. In both methods the
become more and more important. computational effort is rather large.
The present paper outlines a novel ap-
proach for the development of methods of line
% protection based on the use of the wave equa-
\_ tion. By adopting a fundamental point of view
F 77 733-9. A paper reocomended and approved by the main features and limitations of the
the IEEE Power System Relaying Camittee of the rtn methods will be discussed.
Pcwer Engineering Societv for presentation at the EEE
PES Sumir Meeting, Mexico City, Mex., July 17-22, THEORY
1977. Manuscript submitted Januaxy 24, 1977; made
available for printing April 21, 1977. The transient phenomena initiated by the

0018-9510/78/0900-1607$00.75 01978 IEEE


1608
occurence of a fault on a power transmission known as the reflection coefficient, due to
line can be described by a set of hyperbolic the presence of a resistance Rf at the fault
partial differential equations, known as the Rw -R f
telegraph equation3. In the present study it
will be assumed that the distributed induct- rR+R (4)
w f
ance and capacitance of the line are fre-
quency-independent and that the losses of the For solid faults, Rf = 0, we have r =
line are concentrated at both ends of the 1.*) This equation is valid for all times t,
line. Under these conditions the telegraph t > 0, if the fault is assumed to occur at
equation simplifies to a form known as the the time t = -ax.
wave equation, which can readily be solved.
The general solution of this equation is a The basic relation (3) just mentioned,
linear superposition of two waves travelling together with the equations (2) defining the
along the line in opposite directions3. wave signals, constitute the line equations
on which protection schemes can be based.
Consider a relaying location at the This basic relation states that in case of a
boundary of a protected zone of the trans- fault the two wave signals a(t) and b(t) are
mission line, at which the voltage and cur- exactly congruent except for a constant time
rent waveforms are measured by means of ideal shift, 2ax, proportional to the fault dis-
voltage and current sensors. Let a(t) be the tance and a constant scaling factor, r,
instantaneous value of the forward travelling related to the fault resistance.
wave at the moment t as it leaves the measu-
ring location heading into the protected zone It should be pointed out that the line
of the transmission line. Similarily let b(t) equations (2), (3) describe the sinusoidal
represent the instantaneous value of the as well as the transient time behaviour of
backward travelling wave at the moment t as the measured waveforms. In this respect the
it arrives at the measuring point coming from present fault relation differs fundamentally
the protected zone. Let us call the two sig- from those given by conventionally used
nals a(t) and b(t) wave signals for short. lumped parameter models or from fault crite-
According to the theory of the wave equation ria based on the evaluation of fault gener-
the instantaneous value of the voltage u(t) ated surges4'5. The actual time behaviour of
and current i(t) at time t at the relaying the signals is given by the resultant effect
point can be split up into wave signals: of the above line relation together with the
boundary condition at or behind the measuring
u(t) = 21[a(t) - b(t)] point. The period of the dominant transient
oscillation is dependent of the source imped-
(1) ance and is given by twice to four times the
i(t) = 2R [a(t) + b(t)], travel time between the source and the fault.6
w
It may be mentioned that relation (3) is
where Rw is the characteristic resistance of valid for resistances Rf of any magnitude,
the lossless line. A proof is offered in Ap- hence ohmic loads at the far end of the line
pendix A. lead to an equation of the same type. In this
case the fault resistance and fault distance
The-wave signals a(t) and b(t) can be in equation (3) are to be replaced by the
obtained from the physically measured quanti- load resistance and line length respectively.
ties u(t) and i(t) by straightforward addi- When the load has the magnitude R the re-
tion or subtraction of the above two equa- flection coefficient vanishes and thus the
tions: wave signal b(t) vanishes as well. As the
a(t) u(t) load resistance increases the reflection co-
= + Rw i(t), efficient changes sign and becomes negative.
(2) These considerations can be employed in order
b(t) = -u(t) + Rw i(t). to distinguish fault from load conditions.

Now, a sudden occurence of a fault with- Basic Principle of Measurement


in the protected zone at a distance x from
the measuring site imposes a new boundary In principle a wide variety of methods
condition at x, such that the forward travel- new to the field of protection can be con-
ling wave is reflected back towards the meas- ceived for determining the time delay asso-
urement point before it reaches the far end ciated with the travel time given in the bas-
of the line. To be more precise, consider the ic fault equation of the previous section.
forward travelling wave a(t) at the measuring However, great care must be exercised in de-
point at time t. This instantaneous value veloping non-conventional real-time methods
will propagate along the line, be reflected in order to avoid sophisticated methods,
at the fault and return to the measuring
site, after the elapse of twice the travel
time, 2ax, where a represents the reciprocal *) In the literature3 the reflection coeffi-
of the propagation speed. Thus cient r and the signal b(t)is usually de-
fined with the opposite sign.This conven-
b(t+2ax) = r a(t), (3) tional notation yields the more awkward
value r = -1 for ideal faults and it is
when r represents an attenuation factor, therefore not adopted in this paper.
1609
which may not meet the requirements with re- nals by applying Fourier-series analysis,
gard to the allowable computation time or compute the phase angle between these two
which may be sensitive to other sources of phasors, obtain the fault distance and check
noise not included in the line model. the result with respect to the boundaries of
the protection zone. When a finite window
As in most protection schemes known to- length is used for the extraction of the fun-
day the present approach will be based on the damental frequency components, non-harmonic
evaluation of the network frequency voltage transients present in the time signals will
and current waveforms at the measuring lo- inevitably cause errors in the computed esti-
cation. In Figure 1 the complex voltage and mates of the phasors. Hence, errors in the
current phasors U and I, respectively, and computation of the phase angle and fault dis-
the phasors A,B of the linear combinations, tance will be produced. The accuracy of the
associated with the fundamental frequency w, computed fault distance is thus directly re-
are depicted for a typical fault situation: lated to the accuracy of the phasor esti-
mates. This difficulty also applies to any
impedance type method.
A An alternative approach is based on the
fact that a phase angle measurement is a spe-
cial technique for determining a more physi-
cal quantity, namely a time delay. A shift of
UjX.I the time function b(t) by a time interval of
the amount 2axR, where xe is for the moment
an arbitrary reference distance, will cause a
rotation of the associated phasor in the fre-
quency domain from B(xR=0) to B(xR) as indi-
0 R-I cated in Figure 1. The compensated angle be-
tween the phasor A and the rotated phasor
B(xR) is given by 2wa(x-xR) and its sign
therefore indicates whether the fault loca-
tion x is smaller or greater than the refer-
ence distance xR. By checking angles of this
type twice, once for each zone boundary, the
zone of the transmission line at hand can be
B = B(O) protected. Such an algorithm will be outlined
in the next sections.
Fig. 1. Phasor Representation
Vectors of Correlation
In the frequency domain a time shift is
represented by a rotation of the associated A basic element of the present method
phasor. The fundamental relation (3) there- is the use of integrals which represent a
fore states that the phasor B must lag the measure of the correlation between the wave
phasor A by a phase angle of the amount 2wax signals and a pair of weighting functions,say
while its amplitude is slightly smaller than w1(t) and w2(t), within a time interval T
that of A. By way of example, the magnitude
of the phase angle 2wax is equal to 120 for T T
a fault distance of 100 km, assuming a funda-
mental frequency of 50 Hz and a velocity of
A,=fw1(t)a(t)dt,
0
Bl(xR)=fwl(t)b(t+2axR)dt,
0
propagation at the speed of light. (5)
T T
For small phase angles ranging up to a- A2=fw2(t)a(t)dt,
0
B2(xR)=fw2(t)b(t+2axR)dt.
0
bout 300, that is to say for fault distances
up to about 250 km, the phase angle and the
fault distance is essentially proportional The time shift 2axR represents twice the
to the reactance X, as can be seen in travel time between the relaying point and a
Figure 1. This is also true when current reference distance xR and it is designed to
weighting factors differing from the charac- compensate the actual time delay 2ax between
teristic resistance Rw are employed. In the the two wave signals a(t) and b(t) in the
present study the main emphasis will beplaced case of a fault at distance x. For the sake
on the protection of long lines for which of simplicity the integrals are initiated at
the linear approximation is not necessarily time t = 0.
valid and, more important, for which the high
frequency fault initiated transients become a In on-line applications it may be more
serious obstacle while attempting to deter- suitable to shift the interval of integration
mine the complex phasors from the measured and the weighting functions rather than the
waveforms. signal waveform b(t). Thus, by replacing t by
t-2axR in equation (5), we obtain equivalent
The primary aim of a line protection expressions
method is to protect a given zone of the
transmission line. One rather straightforward T+-r T+T
approach tosolve this problem would be to ex- B1= w1(t-T)b(t)dt, B2= f W2 (t-T)b(t)dt, (6)
tract the phasors A and B from the time sig- T T
1610
where T = 2axR. In order to compute these
correlation integrals a total measuring in-
[A1
Bl(xR)
terval Tm = T+' must be available. This time B(xR) = (9)
Tm represents the temporal window length in
which the measured waveforms are correlated [A2I [B2 (xR)I
with the weighting functions.
Correlation integrals of the type A1, A2 which in the absence of transients will be
appear either directly or disguised as a bas- related to the complex phasors A and B(xR)
ic element in most distance-type protection associated with the sinusoidal waveforms.
algorithms which have been developed in the When transients in the measured waveforms are
past 10 years. In impedance type protection present these vectors will contain additional
schemes, for example, the weighting functions vector components, thus distorting the basic
must be mutually orthogonal and basically of properties of complex phasors. The motion of
a sinusoidal shape in order to obtain the the vector B(x for a varying time delay
real and imaginary parts of the phasors and 2ax associated with the reference distance
of the resulting impedance.7 In methods based xR is no longer necessarily circular as it
on the differential equation of the lumped was in Figure 1. A more typical motion of
parameter line, the weighting functions are B(xR) is depicted in Figure 2.
usually chosen to be two time shifted con-
stant functions, each being of a finite time
interval.8'9 In other studies various other
weighting functions, such as square wave, saw
tooth, constant, Walsh and piecewise constant
weighting functions, and data fitting tech-
niques have been suggested and tested. Such
A
techniques can also be applied in the present
approach.
Correlation integrals can in principle
be implemented by analog, hybrid or digital
techniques. In the references cited above
digital realizations are considered, in which
s 4 ~B NOR
say N samples taken each At seconds are sum-
med, whereby the k-th sample is multiplied by
a weighting factor ck, hence
N
A1= I
k=l
ck a(k At). (7)
This sum can be represented in the gen-
B1 = B(O)
eral notation(5) chosen above by considering
- in a formal manner - a train of weighted Fig. 2. Vectors of Correlation
dirac functions,
In order to apply the rotation principle
N stated in the previous section two basic con-
w1(t) = E ck 6(t-kAt). (8) ditions must clearly be valid:
k=l
When applying weighting functions of (i) Fault Criterion. The correlation vectors
this type care must be taken to eliminate A and B(xR) must coincide, except for a
high frequency signal components by analog positive multiplicative scalar factor,
filters prior to sampling, as imposed by the when the reference distance xR is equal
sampling theorem. to the fault distance x.

One particularity of the correlation in- (ii) Basic Condition. The tip of the vector
tegrals in (5), which is not encountered with B(xR) must cross the vector A once and
in present day schemes but which is typical only once as the time delay 2axR in
for any time delay correlation method, is b(t+2axR) is augmented. Thus the angle
that the integration intervals and thus also fR must increase monotonically as the
the sampling intervals of the wave signals reference time delay increases and ap-
a(t) and b(t) are not in phase as in existing proaches the actual time delay of the
protection schemes but rather possess a wave signals.*)
mutual reference time delay of the amount T =
2axR that is to say of roughly 0.66 ms per The validity of the fault criterion will
100 km reference distance (if 1/a is equal to in the next section be shown to be a straight-
the speed of light). If the wave signals are forward consequence of the line model (3).
formed prior to sampling, this time delay can
be incorporated into the sampling scheme. *) The monotonic growth of the angle fR is
not required when the fault and reference
Let us now consider the correlation vec- distances lie far apart.
tors
1611
Following this proof, the basic condition of the previously mentioned basic condition
will be discussed. At this stage of the anal- associated with the rotation property of cor-
ysis note that the vector of correlation is relation vectors is offered. It will be as-
required to rotate neither with an angle pro- sumed that the correlation integrals consist
portional to the time delay nor with a con- of a first part due to the' fundamental wave-
stant amplitude, as is the case with phasors. forms and a second part due to the transient
phenomena on the transmission line. All other
Fault Criterion signal noise contained in the measured wave-
forms will be neglected or be considered to
By multiplying the line relation (3) have been successfully suppressed in the
with the weighting function wl(t) and inte- process of the filtering and integration.
grating both sides of the equation over the
integration interval T we obtain High Frequency Transients: Let us as-
sume, for the sake of simplicity, that a time
T T shift of the steady state waveform in the
signal b(t) produces a rotation of the asso-
f wl(t) b(t+2ax) dt = r f w1(t) a(t) dt. (10)
ciated correlation vector B at the rate w
0 0
with a radius Bw about the origin, as is the
Due to the previously introduced definitions case for phasors (see Appendix B). Likewise
(5) we obtain, setting xR = x, B (x) = r Al. let us assume that an arbitrary transient
Similarly we obtain B2(x) = r *2 and hence high frequency oscillation of the frequency
Q also gives rise to a circular rotation with
B(x) = r A. (11) the amplitude BQ at the rate Q. The superpo-
sition of these two circular motions yields
Thus the two correlation vectors B(xR) an epitrochoidal curve, as depicted in Figure
and A are indeed parallel when the reference 3 for various amplitude ratios BQ/Bw.
distance xR coincides with the fault distance
x. Note that this criterion of parallelism is
entirely accurate, despite the fact that the
vectors of correlation A and B (XR=X) need
bear only a very weak resemblance to the com-
B,//Bw
plex phasors corresponding to the fundamen- 0.30
tal waveforms. As a matter of fact the fault tB2 JB2 6 -wla
criterion (11) is valid for any arbitrary
correlation integrals of the general form
(5).
A further increase of the reference dis-
tance xR will cause the B vector to rotate
further on. The location of a fault with res-
pect to the two boundaries of a protection B
an
zone, for example xR = 0 and xR = z, can be
determined by specifying the position of the w
vector A with respect to the two vectors B(O) 0 B1
and B(z) associated with the boundaries of
the protection zone. Thus only the three cor-
relation vectors A, B(O), B(z) need to be
computed to solve the primary protection
problem.
Analytically, the relative positions of Fig. 3. Motion of Correlation Vector in the
these vectors can be described by their vec- Presence of High Frequency Transient.
tor product B(xR) x A. The component vertical
to the plane in which A and B lie is given by Since the angle 4R of the resultant vec-
tor B(xR)must grow monotonically as the
V(xR) = A2Bl(xR) AlB2(xR) (12) reference distance xR increases, the curve
traced by the vector must be void of loops,
and its sign is equal to the sign of hence,
sin(4-4R) and hence also of x-xR. Expressions
of this type can be computed for each zone B9
(13)
boundary thus yielding, for example V(O) and B o

V(z). If V(O) is positive and V(z) is nega-


tive, i.e. if As the frequency of the transient oscil-
lations decreases to lower values, as is the
V(O) O, case for increasing fault distances, the con-
> V(z) < 0
dition on the permissable amplitude B i re-
the fault must lie within the protected zone maining after the filtering becomes less and
of the transmission line. less stringent.

Effect of Line Transients DC-Transient: Consider now a slowly de-


caying DC-transient which is superimposed on-
In this section an approximative analysis to the fundamental current waveform. This
1612
transient will cause a slowly decreasing off- The angle ¢-+R between the vectors A and
set of the sinusoidal signal present in b(t) B(xR) in Figure 2 can be computed explicitly
(and a(t)), and thus offsets in the two cor- from the vector and scalar products,
relation integrals Bl(xR) and B2(xR). The DC-
transient therefore has the effect of dis- V(xR) = IAIIB(xR)Isin(- R),
placing the circular motion of the vector
B(x ) for increasing xR as shown in Figure (16)
4 Kor the case of a;constant DC-transient. W(xR) =IAI B(xR) ICOS(4- R)I
where JAt, IB(xR)I denote the lengths of the
B2 vectors A and B(xR). Hence
V (xR)
¢-¢R R~~~~
= arctan W(x R ) (17)
If the vectors of correlation are of the pha-
sor-type it follows from Figure 1 and 2 that
this angle is related to the fault distance x
by

¢-¢R = 2wac(x-xR). (18)


Another useful quantity is the ratio of
the vector lengths IB(xR)I/IAlwhich, accord-
ing to equation (11), is exactly equal to
magnitude of the reflection coefficient r,
when xR = x and which may be used as a rough
measure of this quantity when xR 0 x.
The fault distance may be of use for de-
termining possibly defective elements of the
line and does not necessarily have to be com-
puted in an on-line fashion. One possibility
of accurately computing the fault distance is
to extend the window length after having de-
tected a fault until the line is interrupted
Fig. 4. Motion of Correlation Vector in the by the circuit breakers in order to obtain
Presence of DC-Transient. accurate phasor estimates and to apply formu-
lae (17) and (18). If the signal waveforms
Again the angle 1R of the resulting vec- are stored over a given time interval an al-
tor B(xR) must increase with increasing xR. ternative way to compute the fault distance
Clearly, therefore, the amplitude of the DC- is to iterate over the reference distance un-
component BDC which remains unfiltered must til the vector product vanishes, thus elimi-
be less than the sinusoidal component Bw nating the need of accurate phasor estimates.

BDC < B . (14) RESULTS


The results (13) and (14) served to il- Results from Theoretical Data
lustrate basic limitations of the method and
they do not necessarily represent worst Theoretical waveforms were calculated on
cases. A more detailed analysis however does the basis of the wave equation with frequency
yield similar results. independent distributed parameters by means
of a digital computer. Out of a large number
Auxiliary Formulae of studies one typical simulation is included
here for illustration. The simulated line had
The vector product is a measure of the the frequency independent resistance r' =
degree of parallelism and hence it does not 0.013 Q/km, inductance Q' = 0.903 mH/km and
completely describe the relative positions of capacitance c' = 13.4 nF/km.The line consist-
the two vectors A and B(x ). Other useful ed of two parts, one of length 200 km and
quantities can be obtained gy applying ele- the other of length 100 km, and it was fed in
mentary vector theory. One such quantity is from both sides with a 50 Hz frequency source
the scalar product, of equal magnitude and a mutual phase shift
of 100 through source impedances of equal
W(xR) = A1 Bl(xR) + A2 B2(xR), (15) magnitude 52 R. At the junction an ohmic load
of 260 R was assumed. The fault had a resist-
which represents the projection of A onto B ance of 13 R and was initiated in the first
(or vice versa). If this product is positive, zone at a distance of 180 km along the line.
the angle between the two vectors is less The voltage and current waveforms were com-
than 900, otherwise the angle is larger than puted at interval of 34.7 ps.
900.
The position of the fault was computed
1613
with respect to the measurement point,
0 km, and with respect to the far end of the
= xR distance, or zone boundary, can very
rately be computed.
accu-

line, xR = z = 200 km, on the basis of the


mentioned samples by applying formulae (17), Another important fact brought out by
(18). A typical result of these distanice com- Figure 5 is that the directional discrimina-
putations is depicted in Figure 5 as ai func- tion of the fault location with respect to
tion of the length of the measuring wi.ndow T the zone boundaries xR = 0 and xR = 200 km is
m correct for window lengths significantly less
than is necessary to estimate the fault dis-
P. U tance itself. In this example a window length
of Tm - 5 ms from the moment of the occurence
of the fault onwards suffices to reliably lo-
VOLTAGE / cate the fault with respect to the zone boun-
/ |,\, CUCURRENT / daries. This decision can be obtained on the
basis of vector products of the type (12)
X . A . . . . i .
0- 2b MS rather than on the complete expression for
the fault distance (17), hence the need for
-1 JiVs. T I ME
TIME
estimating the fault distance itself for sol-
ving the main transmission line protection
_ \,

\ _ _J~ problem can be avoided.


Basically similar results can be ob-
tained by employing fewer samples and larger
sampling intervals, say 0.5 ms.
Results from Experimental Data
In the next stage of the work the pro-
posed method of fault localization was imple-
mented on a real-time basis and tested in an
on-line fashion on a model transmission line.
The aim of this study was to evaluate errors
arising due to the implementation and to in
detail investigate the influence of various
network parameters such as for example the
fault distance, the fault resistance, the
fault inception angle, the source impedance,
the distributed line losses and the load con-
ditions at the far end of the line.
The method was verified on a special
purpose analog device, according to the flow
chart in Figure 6. In order to avoid explicit
multiplications and to reduce computational
errors the weighting functions wl, w2 were
chosen to be simple constant functions over a
time interval Tw with a mutual time delay 2A.
In order to check the auxiliary conditions,
improved phasor estimates, A = al + ja2, were
Fig. 5. Input Waveforms and Calculated Fault obtained by linear superpositions of the cor-
Distance. relation integrals A1, A2, which corresponds
to a resultant weighting of the square wave
starting immediately at the moment when the type.
first fault reflected wave reached the relay-
ing point. The weighting functions were in
this example of the sinusoidal type wl(t) =
sin(wt), w2(t) = cos(wt), and the integration
time follows from T = T -2axR
It should be pointed out that these
weighting functions are not orthogonal to
each other with respect to integration inter-
vals T shorter than the fundamental period
20 ms and that more orthogonal-type weighting
functions will yield improved distance esti-
mates. Nevertheless Figure 5 does illustrate
a very important feature of the method, name-
ly that the accuracy of the computed fault
distance for a given window length Tm improv-
es with decreasing distance between the
fault and reference distance. Thus fault dis-
tances in the neighbourhood of the reference Fig. 6. Block Diagram
1614
lines based on the use of the wave equation
as the line model has been discussed. The
5ms
hi
main aspect brought up in the paper is that
the fault location problem is considered as a
problem of determining a time delay. This ap-
a(t) proach opens the way for the development of a
wide scope of protection algorithms differing
from the schemes known today.
~-b(t)
The application of a very general corre-
lation technique for determining the time de-
lay is elaborated. Simulation studies and re-
sults obtained under laboratory conditions
indicate that the method works well. Further
development will be necessary to more care-
fully evaluate the method as compared with
other conventional and non-conventional
methods.
ACKNOWLEDGEMENT

Tm
V\"\()
\
' V(O)
The author wishes to acknowledge the
very valuable contribution by Mr. O.Lanz in
developing the analog device used in this
study.
Fig. 7. Typical Input and Output Signals. APPENDIX A
One typical result is shown in Figure 7. LINE EQUATIONS
The directional discrimination functions
V(0), V(z) are seen to operate correctly for Let y be the distance coordinate of a
a fault at 168 km near the boundary z = 196 transmission line, where y = 0 corresponds to
km. The device was initiated artificially and the measuring site, and let t be the physical
the parameters were chosen as follows: Tw = time. A lossless transmission line can be
2.5 ms, 2A = 2.5 ms, T = Tw + 2A = 5 ms,hence mathematically described by a set of linear
the total measuring time was Tm = T + 2az = partial differential equations with constant
6.16 ms. The line model consisted of the men- coefficients, known as the wave equation,
tioned section of 196 km length and a further for the instantaneous line voltage u(y,t) and
line section of 98 km length, whereby an ohm- current i(y,t)
ic load of 386 Q was placed at the junction.
The total line was represented by 21 v-type u + 2.'i = 0, i + c'u = 0 (19)
equivalent circuits associated with the (50
Hz) parameters 2' = 1.25 mH/km, c' = 9.3 where the subscripts y, t denote partial dif-
nF/km and r' = 0.02 Q/km. A 50 Hz voltage of ferentiations with respect to y and t and
equal magnitude and phase was fed into the where 2' and c' represent the distributed
two outer ends of the line through the source line inductance and capacitance. The solution
impedances 42 Q and 34 Q, respectively. of this equation is of the general form
Under more hostile transient conditions, u(y,t) = l[a(t-ay) - b(t+ay)],
for example when the X/R-ratio of the network (20)
is very high or when the source impedance is
very large, additional high and low pas fil- i(y,t) = 2 [a(t-ay) + b(t+ay)],
tering becomes necessary, hence longer re- w
sponse times are to be expected. Such fil- where R is the characteristic resistance and
tering will be necessary from an other point a is thwe reciprocal of the velocity of propa-
as well, namely to suppress the transients gation
in present day sensing devices.
R =
w -i,,
c a = rRT. (21)
In the course of the theoretical and ex-
perimental studies various additional probl- The function a(t-ay) represents a for-
ems were solved, the detailed discussion of ward travelling wave, while b(t-ay) repre-
which lies beyond the scope of this paper. sents a backward travelling wave. Setting y =
For example, it was demonstrated that the use 0 in this solution, we obtain the expression
of a memory voltage will aid the directional adopted in the body of the paper.
discrimination of faults close to the relay-
ing site and that the method can be extended Assume now that there is a fault at the
to include mutual couplings in three phase distance x from the relaying location charac-
systems. terized by the relation
CONCLUSIONS u(x,t) = Rf i(x,t), (22)

A method for the protection of overhead where Rf denotes a fault resistance. Substi-
1615
tuting the solution (20) into the fault con- proportional to the frequency. The maximum
dition, we obtain and minimum lengths of the vector B is given
by VT cos(wA)B, and 2' sin(wA)B, respective-
b(t+ax) = r a(t-ax), (23) ly. If sin(wA) = cos(wA) = l/ , i.e. if A is
equal to 2.5 ms and the frequency is 50 HZ,
where the constant r is given by the motion becomes circular. In this special
case the vector of correlation represents a
l-Rf/R
f/ w
R -R f
w complex phasor in the sense of Fourier-series
r =
r
+f/Rw Rw +R f analysis.
A time shift in equation (23) of the amount Needless to say, the above discussion is
ax finally leads to the line equation (3) in valid for any arbitrary frequency component
the main text which is the starting point of of the signal at hand, in particular for fre-
the discussion in the body of the paper. quencies present in the line transients.
APPENDIX B REFERENCES

CORRELATION VECTORS ASSOCIATED WITH SINUSOI- (1) J. Kohlas, 'Estimation of fault locations
DAL WAVEFORMS on power lines.' Proceedings of the 3rd
IFAC Symposium, the Hague/Delft, the
In the following discussion we restrict Netherlands, 12-15 June 1973, p. 393-402.
ourselves to weighting functions wl(t) and
w2(t) which for the sake of simplicity are (2) Ya. S. Gel'fand and L. S. Zisman, 'Algo-
identical to each other up to a time shift of rithms for distance relay protection from
the amount 2A. Thus we assume that inter-phase faults on transmission
lines.' Elektrichestvo, No. 9, pp. 18-24,
w2(t) = w1(t+2A). (25) 1975, english translation in Electric
Technology R 1975 (3), pp. 91-105.
Consequently the corresponding correlation
integrals B2(xR) and Bl(xR) will also related (3) L. V. Bewley, Traveling waves on trans-
to each other. The substitution of (25) into mission systems. John Wiley & Sons, Inc.,
the definitions (5) indeed yields New York, 1951.

B2(xR) =
B1(x R A/), (26) (4) R. F. Stevens and T. W. Stringfield,
transmission line fault locator using
'A
provided both weighting functions are - with- fault-generated surges.' AIEE Transac-
out loss of generality - assumed to vanish tions, Part II, Vol. 67, 1948, p. 1168.
outside of the integration window T.
(5) A. D. Odgen, L. N. Walker, G. E. Ott, and
The contribution of a sinusoidal wave- J. R. Tudor, 'Implementation of high fre-
form component in the signal to be correlated quency transient fault detector'. Present-
to the correlation integral will again be si- ed at IEEE Winter Power Meeting, Jan.
nusoidal of the general form 1970, 70 CP 140.

B1(xR) = B sin(2waxR+y), (27) (6) U. Ratti and T. R. Tudor, 'Power system


fault transient response', L'Energia
where w is the considered frequency and where Elettrica, No. 9, 1972, pp. 590-600.
B, and y are further parameters which depend
on the weighting function and the amplitude (7) G. R. Slemon, S. D. T. Robertson and M.
and phase of the sinusoidal waveform. Ramamoorty, 'High speed protection of
power system based on improved power sys-
The insertion of (27) into (26) yields tem models.'CICRE, paper No.31-09. 1968.

B2(xR) = B sin(2wxxR-2wA+y). (28) (8) A. D. McInnes and I. F. Morrison, 'Real


time calculation of resistance and react-
ance for transmission line protection by
The general form of the motion of the tip of digital computer.' IEA March 1971, p.16.
the vector B(xR), with components B1(x ) and
B2(xR), as a function of xR is an elHipse. (9) R. Poncelet, 'The use of digital comput-
The angular rate of this motion with respect ers for network protection.' CIGRE paper
to xR is given by 2wa and is thus a constant No. 32-08, 1972.
1616
Discussion We feel that his method could be best implemented using an all
digital approach with such devices as CCD (Charge Coupled Device) to
G. S. Hope and 0. P. Malik (The University of Calgary, Calgary, Alberta, mechanize his correlation function. It is also possible to find the time
Canada): We would like to congratulate the author on a fine paper delay by direct cross-correlation between the wave signals.
which highlights some of the basic problems of impedance relaying of We would also like the author's comments on the extension of his
transmission lines. method to cover other types of fault, particularly when the zero-
We were particularly pleased with the author's approach in Figs. 2 sequence current is not negligible and on the influence of the CT and
to 4 to the description of some of the basic points. Although an alter- CVT transient effects.
nate approach to illustrate the basic problems was tried earlier [A,B], Once again, we congratulate the author for a most interesting
we feel that the authors approach will give added strength to those paper.
arguments.
The values of a(t) and b (t) are obtained respectively by adding and
subtracting the actual voltage and current (eqn. 2). Now examining M. Vitins: The author wishes to thank the discussers for their valuable
Fig. 5, one sees that the voltage is approaching zero. In fact, if the vol- comments and their pertinent questions.
tage were zero, the wave signals will coincide perfectly. Thus the lack The references mentioned by Messers. Hope and Malik strengthen
of coincidence gives a measure of the voltage. As the length of the trans- some of the basic points brought forth in the paper and the author
mission line increases, the difference between the two wave signals be- appreciates the opportunity to refer to them. As pointed out by the
comes more noticeable and more measurable. This phenomena has been discussers it is in general more difficult to protect short lines than long
observed by the discussors also in their studies. In fact the more remote ones by means of distance relaying devices due to the unfavorable ratio
the fault on a high voltage transmission line, the faster it can be de- of transient to fundamental signal in the voltage waveform for close-in
tected and the more accurately it can be located. This condition exists faults. In order to achieve an accurate directional discrimination of the
precisely when speed and accuracy requirements are not the most criti- fault with respect to the measuring point for such faults the use of a
cal. However, the basic property of correlation is illustrated, i.e. when memory voltage will be necessary. The replacement of the line voltage
what one wishes to identify has significant parameters correlation is by a memory voltage will keep the two resulting auxiliary signals a(t)
excellent. On the other hand, when the inductance and capacitance in and b (t) from coming to a perfect coincidence for a close-in fault.
the system are small, problem becomes more difficult. The directional discrimination with respect to the measuring point de-
For a close-in fault, voltage, u(t), is almost zero and b(t) = a(t) pends on whether the signal b (t) is lagging or leading the signal a(t), and
with Rf = 0. From eqn. (12), V(z) is also equal to zero. This thus makes the memory voltage will thus guarantee a significant discrimination.
close-in faults very difficult to detect. Such important cases are pre- The memory voltage may not be used for the directional discrimi-
sumably described by the author as hostile in connection with his de- nation of faults with respect to the distant zone boundary. For close-in
scription of Fig. 7 in the paper. faults we do indeed have b(t) = a(t), as the discussers pointed out, how-
Experimental results in the paper show that, for the particular ever it is not true that the directional discrimination function V(z) with
fault described, a fault detection time of 6.16 ms on a 50Hz signal is respect to the boundary zone z is equal to zero. Due to the introduction
practical. Discussors have had similar results for remote faults at of a reference time delay 2az in computing the B-vector, the two vec-
150 kmc. However, longer time is required to accurately detect faults tors of correlation A and B (z) will deviate from each other and thus
that are close to the relaying point. The problem becomes even more lead to a non-vanishing vector product V (z). The larger the protected
complicated when one looks at the correlation vectors given by the zone, the easier it is to detect close-in faults. The minimum time to ac-
author in Figs. 2 to 4. Fundamentally, one is required to increase the curately relay short lines is in general limited by the discrimination of
length of Be,, as the transient frequency increases. One can avoid this faults with respect to the zone boundary and is related to the ratio
problem by using analog filters with very low cut-off frequencies. This, B& /Bw, for faults near the zone boundary. This ratio depends essen-
however, leads to the annoying time delay as described by the author. tially on the length of the line, the infeed conditions and the location
Choose your technique - be it a R-L technique, correlation or reflected of the line in the whole network.
waves. For close-in faults, B2 /Bl, ratio must be small and this can only The final point raised by Messers. Hope and Malik can be discussed
be done by increasing the delay in the analog filters or by increasing the by referring to Appendix B of the paper where the special case of two
window size of the data analyzed. The BQ /B0, ratio must be maintained identical but time shifted weighting functions are considered. It is
and, in the opinion of the discussors, is the limiting factor on the speed shown that a sinusoidal component of frequency Q in the signal b(t)
of relaying. For close-in faults, it is in the neighborhood of 8-9ms. will in general lead to an ellipsoidal motion of the corresponding com-
The discussors wonder if the author would care to speculate on ponent of the vector of correlation BQ2, regardless of the length of inte-
the minimum time for sufficient information to relay accurately for gration T and of the time shift A. For the sake of discussion, circular
close-in faults. motions are considered in Fig. 3. According to Equation (27) and (28)
We would like to know the basis under which the author assumes the motion is circular if the period of the frequency at hand is an odd
that BQ is rotating in the same plane as B0, in light of the fact that he integer fraction of eight times the shift A ( sin (12A) = Icos(12A) I). For
has taken an integration interval, T, shorter than one fundamental fre- example, if A = 2.5 ms, we obtain a circular motion for signal compo-
quency period, 20ms. nents of the frequencies 50, 150, 250, 350 Hz and so on. The rotation
is anticlockwise of every second frequency, that is to say for 50 and
REFERENCES 250 Hz. The resulting motion of the vector of correlation is given by
the linear superposition of these individual rotations.
[A] G. S. Hope and V. S. Umamaheswaran, "Sampling for computer Messers. Nguyen and Robichaud have suggested an interesting
protection of transmission lines", IEEE Trans. on Power Apparatus formula for the distance to fault via the impedance which is essentially
and Systems, Vol. PAS-93, 1974, pp. 1522-1534. based on the computation of the complex input reflection coefficient
[B] G. S. Hope and O. P. Malik, "Sampling rates for computer trans- (Rw - Z) / (Rw + Z) or (RwI - U) / (RwI + U). After introducing the com-
mission line protection", IEEE Power Engg. Society, 1975 Summer plex notation RwI - U = B1 + jB2 and RwI + U = Al + jA2, the required
Meeting, Paper # A 75 544-7. angle k' may be obtained from
[C] G. S. Hope, 0. P. Malik and M. E. Rasmy, "Digital transmission
line protection in real-time", Proc. IEE, Vol. 123(12), 1976, pp.
1349-1354. & = arctan (AlB2 A2B- -
Manuscript received July 28, 1977.
Clearly, this formula requires the determination of the orthogonal
fundamental frequency components of voltage and current and hence
the accuracy of the fault discrimination will be directly related to the
C. T. Nguyen and Y. Robichaud (Hydro Quebec Institute of Research, amount of non-rejected low- and high-frequency fault generated tran-
Varennes, Quebec, Canada): The author is to be commended for a well sients. This equation is a special case, xR = 0, of the more general
written paper on a new technique for high voltage line protection based formula (17), (15), (12) for the fault distance with respect to a prede-
on the wave equation. termined reference point xR. By introducing the time delay 2ctxR as
We would like the author to comment on the relative merits of his described in the paper travelling wave effects are accounted for. A
method over the impedance calculation method since both have similar higher degree of discrimination of faults with respect to the reference
features on the computation level. One can also estimate the distance to distance xR becomes possible via formula (12) despite the use of non-
the fault via the impedance by the following equation: orthogonal components and relaxed filtering conditions.
exp (- 2yx) = (Rw - Z)/(Rw + Z) = IZ'I exp (itt') The author agrees with Messers. Nguyen and Robichaud that a
time delay can be detected by direct cross-correlation of the wave sig-
x =-0'/2a. nals. One should however bear in mind that the magnitudes of the two

Manuscript received August 12, 1977. Manuscript received November 7, 1977.


1617
wave signals will in general differ from each other. An excessive amount near the zone boundary an error in the sensor system will automatically
of cross multiplications may lead either to a slow or somewhat costly result in an error of the estimated fault location with respect to the
solution, at least for the time being. The suggestion of the discussers to zone boundary as in any distance protective scheme. For close-in faults,
employ charge coupled devices may become attractive provided such however, experiments have indicated that larger transient CVT errors
elements fullfill a number of requirements such as sufficient reliability and a limited amount of CT saturation may be tolerated provided a
and dynamic range. memory voltage is employed.
The method described in the paper applies straightforwardly to
interphase faults. In the case of ground faults, for which the zero
sequence current is not negligible, the zero sequence travelling waves
should ideally be taken into account. In order cover all types of fault Michael Vitins studied at the Swiss Federal In-
with a minimum computational effort it is recommended to take full stitute of Technology in Zurich from 1965 to
advantage of the symmetry or near-symmetry of three phase transmis- 1973, where he received the Dipl. Ing. and Dr.
sion lines. math. degrees and two Silver Medals.
In order to equal or to improve on speed and reliability of present Since 1973 he has been at the Brown Boveri
distance relaying systems new schemes must cope with the same prob- Research Center and has been working in the
lems arising from the transient response of sensing devices. For faults field of protective relaying.

You might also like