Você está na página 1de 6

Fault Location Algorithms for

Super High Voltage Power Transmission Lines


Antans Sauhats, Member, IEEE, and Marija Danilova, Student member, IEEE

I. INTRODUCTION

RANSMISSION lines of any voltage level are subject to


faults. To expedite repairs and restoration of power it is
important to know where the fault is located. Several fault
location schemes for transmission lines that utilise digital
techniques and microprocessor-based systems have been
developed during the last few decades [1]-[3]; some of them
use the information available at both ends of the line to obtain
the accurate fault location, while others utilise data at only one
terminal of the line [1],[2]. The single ended approach to
transmission line fault location is important, as it is less
expensive than the double-ended approach (no communication
link required between the ends of the transmission line) and
more reliable (the ability to operate requires only that the local
end equipment is in operation). The one-terminal data
algorithms determine the impedance and, as a result, distance
to the fault. However, several factors affect the accuracy of the
distance to the fault calculation. One of the main factors
results from the combined effect of the load, fault resistance
and equivalent impedances of the power systems connected to

A. Sauhats is with Faculty of Power and Electrical Engineering, Riga


Technical University, 1 Kronvalda str., Riga, LV-1010, Latvia (phone:
+371 7089 930, fax: +371 7089 931, e-mail: sauhatas@eef.rtu.lv)
M. Danilova is with Joint Stock Company Latvenergo, 12 Pulkveza
Brieza
str.,
Riga,
LV-1230,
Latvia,
(phone: +371 7328 364,
fax: +371 7328 362, e-mail: marija.danilova@energo.lv)

5%

Error (percent of the line length)

Abstract-- This paper presents a new approach for calculation


of fault location on high voltage transmission lines. The proposed
fault location algorithm utilizes statistical information about the
equivalent impedances of the system at the unmonitored end of
the transmission line. Knowledge about the distribution laws of
these values or their numerical characteristics, which are more
accessible for practical use, results in more accurate fault
location for lines with grounded neutral, especially in case of
distant short circuit through the large transient resistance. The
proposed algorithm is based on modelling of the faulted line and
the method of Monte-Carlo. The utilized line model with
distributed parameters is described by two-port network theory
equations. The algorithm automatically eliminates the effect of
line shunt capacitance and calculates not only the expected value
of the distance to the fault, but also another important additional
characteristic for the fault location, namely, the length of the line
segment, where short circuit may have occurred. Furthermore,
the possibility to reduce the costs on the inspection of
transmission line faulted segment and fault point location is
described in the paper.
Index Terms-- Fault Location, Power Transmission Lines,
High voltage, Statistics

the ends of transmission line [3]-[5]. The value of the fault


resistance may be high, especially for ground faults and,
accordingly, the accuracy of fault location may be insufficient.
The operation experience of one terminal data based fault
locators shows that in the majority of cases (80%) the
accuracy of fault location may be considered as satisfactory.
The fault location error may comprise 1-2 % of the monitored
line length. In relatively infrequent cases (10%) the error
may reach 5% and even more [5]-[7]. The possibility of
appearance of considerable errors weakens the confidence in
the results of the measurements and causes the necessity to
search fault location on the long line segments. In real practice
of power systems operation the length of the segment
subjected to the inspection is defined assuming the presence of
maximal error, even in those cases when it is not likely that
such error will appear.
In the previous works [5],[8] we have described the
possibility to calculate not only the expected value of the
distance to the fault, but also the distribution density function
of the distance. Such calculation can be based on common
utilization of the measured values of currents and voltages,
information about statistical characteristics of the equivalent
impedances of power system feeding the opposite end of the
transmission line and the Monte-Carlo method. The algorithm
[5],[8] is based on line model with concentrated parameters
and, therefore is applicable to relatively short transmission
lines. Starting with 150-200 km length of the line, the errors
become significant due to neglected influence of the line shunt
capacitance. Fig. 1 shows the fault distance estimation error
arising from ignored charging effect.

500 kV

4%

3%

330 kV

2%
750 kV
1%

0%
0

100

200
300
400
Line length (km)
Fig. 1. Fault location error as a function of 330 kV, 500 kV and 750 kV
transmission line length

Such inaccuracies are common, even in favourable conditions


for single ended approaches, i.e. when the resistance of the
fault is equal to zero. This paper describes the extension of the
approach [5],[8] for fault location on long transmission lines.
II. THE THEORETICAL BACKGROUND OF THE METHOD
Let us consider a faulted transmission line with distributed
parameters connecting two power systems with known
equivalent impedances (Fig. 2.). In this paper, we will only
refer to the most common failure case: the single line-toground short circuit. Nevertheless, it can be noted that the
described approach can be easily extended for application to
other fault types conditions.
I'1I
I''1I
I''2II
E1

Z1I
U'1I

A1I B1I

RF

A2I B2I

E2

I''2II
I2 U''0I

C2ID2I

A2II B2II

A0II B0II

A0I B0I
I0

U''0I

Z0II

C 0II D0II

Fig. 2. Diagram of the single phase faulted transmission line

The two-port network theory [9] assumes two equations of


the following form (1) to express voltage and current at one
pair of terminals in terms of quantities at the other pair.

Ii''I = CiIU i'I + AiI Ii'I

(1)

where U i , Ii is respectively voltage and current of


sequence i in the fault point.
By defining the propagation constant per unit length for
symmetrical transmission line with distributed parameters:

( R + j L )( G + j C )

(2)

R + j L
G + j C

the AiI , BiI , CiI , DiI parameters can be evaluated as [9]:


Ai = Di = ch ( LF )

where LF - is the distance to the fault.


Taking into consideration that at the fault point there is an
equality of powers [9]:

U1 I1* = (U 2 I 2* + U 0 I 0* ),

(7)

where I * is conjugate value of current I .


Proceeding with rearrangements, it is easy to obtain:

2
2

Im Si = Im U i"I I i* = 0 .

i =0
i =0

(8)

From the above, expressing the fault point currents Ii


thought the currents in the monitored end of the line, the
following can be obtained:

) = 0 ,
*

(9)

where Zi inII , Zi inI are the equivalent impedances of the


four-pole and the system behind as respects to the fault point.
To determine a solution of unknown distance to the fault
LF to the non-linear equation (9), one of the known numerical
methods can be employed (for instance regula falsi [10]). In
addition, it is necessary to keep in mind that there possibly
exist two roots of the equation (9) [5],[8]. In case when both
roots of the equation (9) correspond to the length of the
monitored line, it is difficult to choose the proper root. This
problem should be solved by determination of the
corresponding values of the transient resistance RF (10). If the
distance LF is known, the resistance is defined as one third of
voltage U i and current I i quotient:

) C U k' iI+ A I '

RF = 13 DiIU i'I BiI Ii'I

iI iI

(10)

iI iI

The proper root of the equation (9) corresponds only to the


positive value of the resistance RF (considering the physical

and characteristic impedance of the line:

Zc =

(6)

where ki*I is the conjugate value of the current distribution


coefficient:
BiII + ZiII AiII
Zi inII
DiII + Z iII CiII
kiI =
=
,
Zi inI + Zi inII BiI + Z iI DiI BiII + Z iII AiII
+
AiI + ZiI CiI DiII + ZiII CiII

RF

U i''I = DiIU i'I BiI Ii'I

1
sh ( LF )
Zc

Ci =

Z2II

I''0II

RF

(5)


CiIU i'I + AiI Ii'I
2
Im DiIU i'I BiI I i'I
ki*I
i = 0

C 2II D2II
I''0I

C0ID0I

Z1II

C 1IID1II
I''2I

I'0I
Z0I
U'0I

U''1I

C1ID1I

I'2I
Z2I
U'2I

A1II B1II
I1

Bi = Z c sh ( LF )

(3)

(4)

sense of the problem). When both RF values are positive, the


proper solution for the distance LF corresponds either to the
least value of RF , or the value LF closer to the remote end of
the line should be chosen [5],[8].

III. THE MONTE-CARLO BASED FAULT LOCATION ALGORITHM


Assuming that the equivalent impedances ZiII are random
values, the unknown distance LF can also be determined as a
random value and fault location results can be expressed, for
example, by the set of the following numerical characteristics:
expected value of the distance to the fault point E [ LF ]

standard deviation

s [ LF ] , which is the value,

describing accuracy of the fault point location for the


certain conditions of emergency situation.
For calculation of E [ LF ] and s [ LF ] values in common
case, it is necessary to apply Monte-Carlo method.
The algorithm for estimation of these numerical
characteristics is demonstrated in Fig. 3.
The Monte-Carlo method should be employed to obtain the
distribution density function f ( LF ) of the estimated distance
to the fault LF . The density function forms information basis
for decision-making on recommended margins of the segment
to inspect (more details below). Considering that obtained by
Monte-Carlo method density function f ( LF ) is located

within the certain line segment LF [ LF min ; LF max ] , it can

be asserted that, as well as s [ LF ] , the segment length is an


appropriate characteristic of the performance of the fault
location algorithm.
Empirical or
uniform
distribution law

N=N+1
Random value
generator
Z1I,Z2I,Z0I,Z1II,Z2II,Z0II
Model of the line
(model Fig. 2.)

measured
values U, I

Calculation of the
current distribution
coefficients

Let us incorporate both modelling of the processes in the


faulted transmission lines and the algorithm for fault distance
estimation to describe the performance and main properties of
the discussed algorithm. It is also emphasized that MonteCarlo method was utilized twice:
to model the processes in the faulted line for calculation
of the measured values of the controlled voltage and
currents;
to estimate the distance to the fault according to the
algorithm depicted in Fig. 3. Since the reviewed algorithm
defines the evaluation results of distance as random
values, the accuracy will be characterized by marginal
estimations: the minimal LF min and maximal LF max
values of the determined distances to the fault.
Fig. 4 illustrates the modelling procedure.
For algorithm testing purposes, the PSS/E program
package [11] was used. Two main arguments have contributed
to the decision: PSS/E has powerful ready-to-use fault
simulation tools, and allows to automate the calculations by
running user macro-programs in built-in Iplan programming
language. However, a number of difficulties had to be
overcome during the system modelling in PSS/E: the standard
PSS/E routine, which invokes in-line fault simulation and
parameter calculation, virtually divides the line into two
models [9] and utilizes the line model with concentrated
parameters. That model could provide non-satisfactory
precision of modelling for algorithm testing purposes.
It was easy to
conquer the mentioned barriers
implementing additional fictitious bus Fault, and
recalculating and substituting the lines parameters from the
Iplan program in accordance with distributed parameter line
model and respectively to the fault location on the line [11].
In this paper, the number of trials N (Fig. 3) is equal to
100. It was experimentally determined as able to provide an
appropriate accuracy. The algorithm properties were tested by
system simulation (Fig. 4) under 50 random operating
conditions from strong to weak ones.

...

PSS/E Iplan program


Random value generator
Z1I,Z2I, Z0I, Z1II,Z2II,Z0II
Number of trials M = 50

Calculation of the
fault power S
Im [Si] = 0

NO

Search procedure of
LF

IV. SIMULATION PROCEDURE

NO
PSS/E Iplan program
Calculation of line model
parameters
(model Fig. 2.)

YES
N>Number of
trials
YES
Calculation of the
expected value and
deviation
Estimation of the line
segment limits to
recommend for
inspection

PSS/E Iplan program


Estimation of the limits of line
segment recommended for
inspection
Number of trials N = 100
(the algorithm Fig. 3.)

Fig. 4. The structure of simulation procedure

Fig. 3. The algorithm for estimation of distance to the fault

PSS/E main program


standard routines

PSS/E is the commercial tradename for PTIs transmission planning and


operation software

V. SIMULATION RESULTS

6%

500
750

3 ACO 500
4 ACO 600

0.022
0.014

0.940 0.0122
0.895 0.0130

0.170
0.162

2.37
2.23

0.0065
0.0075

TABLE II
POSITIVE AND ZERO SEQUENCE IMPEDANCES OF THE MONITORED AND
REMOTE SYSTEM
Strong state
Weak State
Sequence
System
R,
X,
R,
X,
Ohm
Ohm
Ohm
Ohm
Monitored
0.157
57.33
0.02
64.41
Positive,Negative
Remote
0.189
39.35
0.436
62.24
Monitored
0.013
66.60
0.012
71.97
Zero
Remote
0.0049
28.14
0.011
33.49

The results of the tests of the algorithm properties for the


500 kV power transmission line are shown in Fig. 5. In this
case fault location on the line is constant and equals to 90% of
the line length, fault resistance equals to 100 Ohm, the remote
bus load is 200 MW at load factor 0.9.
The varying parameter is the length of the line. It could be
concluded that despite sufficient increase of the length of the
line, utilization of the proposed algorithm provides possibility
to determine the mathematical expectation with close to zero
value of the average error. The deviations of the maximal and
minimal determined distance to the fault from the real fault
point are shown relatively to the line length. Depicted
deviations simultaneously provide information on the length
of line segment to inspect as suggested by the algorithm,
which is equal to sum of the deviations absolute values.
It could be noticed that the deviations are relatively smaller
on longer lines. This trend can be explained by decrease of
variant impedance share in the system and therefore reduction
of the influence of the remote system impedance uncertainty.
For comparison, the results obtained by previously
described algorithm [8] are given. The algorithm is based on
line model with concentrated parameter and neglects the
influence of the charging. As it can be concluded from Fig. 5,
the error will rapidly increase for line lengths exceeding
200 km, while in cases of the short lines with the length less
than 100 km, the difference between the results of the methods
is insignificant.

Error (percent of the line length)

4%
3%
2%

Proposed
algorithm:
error of
Mathematical
expectation,
Maximal and
Minimal
determined
distance

1%
0%
-1%
-2%
-3%
50

100 150 200 250 300 350


Length of the line (km)

400

Fig. 5. Algorithm precision depending on length of typical 500 kV


transmission line.
5%
Method [8]: error of
Mathematical expectation,
Maximal and
Minimal
determined
distance

4%
Error (percent of the line length)

TABLE I
POSITIVE AND ZERO SEQUENCE RESISTANCE, REACTANCE AND
CONDUCTANCE OF THE TESTED LINES
Voltage Conductor
Rpos.
Lpos.
Cpos.
Rzero
Lzero
Czero
mark
kV
Ohm/km mH/km F/km Ohm/km mH/km F/km

Method [8]: error of


Mathematical expectation,
Maximal and
Minimal
determined
distance

5%

3%
2%

Proposed
algorithm:
error of
Mathematical
expectation,
Maximal and
Minimal
determined
distance

1%
0%
-1%
-2%
-3%
50

100

150 200 250 300


Length of the line (km)

350

400

Fig. 6. Algorithm precision depending on length of typical 750 kV


transmission line.

7%
Method [8]: error of
Mathematical
expectation,
Maximal and
Minimal
determined
distance

6%

Error (percent of the line length)

The described algorithm was tested and verified using


faulted lines models and digital records of the contingency
processes on 500 and 750 kV transmission lines. The tested
lines were up to 400 km long.
The parameters of the typical transmission line used for
testing procedures are given in table I. During the tests the
impedances of the monitored and remote (unmonitored)
systems were chosen randomly in the impedance interval
corresponding to the strong and weak system operating
conditions as specified in table II. The data is derived from the
real power system measurements.

5%
4%
3%
2%

Proposed
algorithm:
error of
Mathematical
expectation,
Maximal and
Minimal
determined
distance

1%
0%
-1%
-2%
-3%
0

100

200

300

400

Distance to the fault (km)


Fig. 7. Algorithm precision depending on fault location on the typical 500 kV
transmission line.

VI. DECISION-MAKING ON LINE SEGMENT TO INSPECT IN


FAULT LOCATION TASK
Let us suppose that the estimated distances to the fault LF
are normally distributed. E X is the mathematical expectation
of the distribution and X is the standard deviation. To find
an optimal, in sense of costs, length of the line segment to
inspect, let us additionally assume that:
q is known (cost of fault location per unit length).

if the fault cannot be located on the recommended


segment, then the whole line route with the length L is
subjected to the inspection.
Considering these assumptions the mathematical
expectation E [ q ] of the costs of fault location can be
evaluated as follows:

q
+ (1- p) L q
(11)
100
is the probability of fault location in the
E [ q ] = p l L

where

recommended for inspection segment, l is the length of the


segment in percents of the line length.
Assuming the segment length l is chosen as proportional
to the standard deviation X , the centre of the segment is
coincident with E X :

l = s X [1]

(12)

where s - is the constant of proportionality.


Taking into consideration that:

l
l
p = u
u

2X
2X

(13)

where u ( X ) is the function of normal distribution [10] it is


possible to obtain:

E [ q ] = Lq 0.02s X u

( 2s ) + 1 2u ( 2s )) .

(14)

Thus, the average costs are dependent on both s and X .


1.0

3.0%

1.5%

0.0%
400

100
200

50
30
10
Fault resistance (Ohm)

40
0
Distance to the fault (km)

Fig. 8. Dispersion as a function of fault resistance and fault location on


500 kV transmission line.

Mathematical expectation of costs, E[q]/Lq

Fault distance dispersion (percent of the line length)

The results of the similar investigation for 750 kV voltage


line model are presented in Fig. 6. The load at the remote end
was modelled equal to 800 MW. The characteristics behaviour
is similar to the described above. However, it might be noticed
the algorithm would determine smaller dispersion of the
possible fault point location and therefore would suggest for
inspection shorter line segment. The error of the method [8] is
smaller as well, despites increasing charging in 750 kV line
configuration. This phenomenon originates from angle of the
line impedance, which in case of 750 kV lines is convenient
for fault location task.
Fig. 7 represents the results of the algorithms properties
analysis depending on the location of the faulted point on the
line. The system configuration was identical to the described
above. Fault resistance was modelled equal to 100 Ohm. The
average error of the mathematical expectation is weakly
swinging around zero. The inspection zone in this case is
becoming wider moving away from the monitored line end
and reaches 3.2 percents of the line length at the very end of
the line. The advantages of the new method are obvious: the
previously developed method [8] would not find any solution
for determination of the distance to the fault on approximately
last 5 percent of the line. Furthermore, starting at the average
with 180 km distance, the line segment recommended for the
inspection will not contain the factual fault point.
Fig. 8 shows the results of the algorithm application for
determination of the fault distances dispersion on 500 kV
power transmission line depending on the fault locations and
fault resistance values. The faults with resistances over
100 ohm are very unlikely to happen and therefore have not
been considered. The chart surface is rather smooth, which
approves that the trial number choice is appropriate. For each
of the fault resistance values, at the end of the line, it can be
noticed rapid length increase of the recommended for the
inspection segment. The increase of the fault distances
dispersion is more even towards fault resistance axis.

x = 10 %
x = 5 %

0.8

x = 2 %
0.6

x = 1 %

0.4

0.2

0.0
0

2
4
6
8
Proportionality coefficient, s

10

Fig. 9. Mathematical expectation of costs as a function of proportionality


constant s at various standard deviation values

The graph of the function (14) is shown in Fig. 9. It is


obvious that given the X value, one can determine s and
consequently l corresponding to the minimum of average
costs of the fault location. For illustration, let us assume that
resulting value of distance estimation is X = 2% , then as it
follows from Fig. 9 the minimal value of the mathematical
expectation of the costs corresponds to factor s 4 value,
therefore the recommended for inspection segment length is
4X .
Alternative fault location strategy can be implemented. For
instance, it can consist of the following three steps: first
relatively short segment is subjected for the inspection, then
wider searching zone is defined, at last, in case of
unsuccessful efforts, the whole line should be inspected. Still,
the described cost minimization approach can be employed to
define the length of the segment to inspect.
VII. RESULTS OF TESTS AND IMPLEMENTATION
The algorithm precision was verified using the real test
records shown in Fig. 10. An artificial short circuit was
applied to 396.73 km long 750 kV line with the following
parameters:
Z1 = 0.017 + j 0.289, Z 0 = 0.18 + j 0.618 Ohm/km ,

B1 = 3.92 106 , B0 = 2.66 106 Siemens/km


During the test, the line was operated at 500 kV. The
remote end impedance was 1000 Ohm, purely inductive. The
fault current was conducted thought the tree at 390 km
distance.

The simulation results show that the accuracy of the fault


location was sufficient under the tested situations that had
included various fault resistances, fault locations, prefault
loading conditions and impedances of the connected systems.
The algorithm efficiency is confirmed by processing of
measurements data obtained during forced short circuit state
on 750 kV line.
It has been shown that it is possible to minimize the fault
location costs utilizing the distribution density function of
fault distance estimation. The fault location strategy depends
on combination of several accuracy-affecting factors.
IX. REFERENCES
[1]

T.Takagi, Y.Yamakoshi, M.Yamuaura, R.Kondow, T.Matsushima,


Development of a New Type of Fault Locator Using One Terminal
Voltage and Current Data, IEEE Trans., vol. PAS-101, No 8, Aug.
1982, pp. 2892-2898.
[2] L.Eriksson, M.Saha, S.D.Rockfeller, An Accurate Fault Location with
Compensation for Apparent Reactance in the Fault Resistance Resulting
from Remote-end in feed, IEEE Trans. on PAS, PAS-104, No 2, 1985,
pp. 424-436.
[3] D.Novosel, D.G.Hart, M.M.Saha, S.Gress, Optimal fault location for
transmission system, ABB Review 8/1994, pp. 20-27.
[4] R.K.Aggarwal, D.V.Coury, A.T.Johns, A.Kalam, A Practical Approach
to Accurate Fault Location on Extra High Voltage Teed Feeders, IEEE
Trans. on Power Delivery, vol. 8, No. 3, Jul. 1993.
[5] A.Sauhats, A.Jonins, M.Danilova, Statistical Adaptive Algorithms for
Fault Location on Power Transmission Lines based on Method of
Monte-Carlo, in Proc. 7th International Conference on Probabilistic
Methods Applied to Power Systems, September 22-26, 2002, Naples,
Italy, pp.485-490.
[6] I.V.Jakimec, A.V.Narovljanski, I.A.Ivanov, Determination of Fault
Location on Transmission Line based on Power Flows Measurement,
Electrichestvo, No.5, 1999, pp. 5-9 (in Russian).
[7] A.K.Belotelov, A.S.Sauhatas, I.A.Ivanov, D.R.Ljubarskij, Functional
Algorithms and Operation Experience of Microprocessor-based devices
for Fault Location on Transmission Lines, Elektricheskie Stancii,
No.12, 1997, pp.7-12 (in Russian).
[8] A.Sauhats, A.Jonins, V.Chuvychin, M.Danilova, Fault location
algorithms for power transmission lines based on Monte-Carlo method,
in Proc. 2001 IEEE Porto Power Tech Conf.
[9] G.I.Atabekov, Distant approach in long power transmissions protection,
Akademija Nauk Armjanskoj SSR, 1953 (in Russian).
[10] G.Korn, T.Korn, Mathematical handbook, McGraw-Hill Book
Company, 1968.
[11] PSS/E-27 Online Documentation, Power Technologies, a division of
S&W Consultants Inc., December 2000.

X. BIOGRAPHIES

Fig. 10. Records of the contingency processes during forced fault on 750 kV
transmission line.

In these conditions, the average distance to the fault was


determined with error 0.38 percent of the line length or
1.5 km.
The algorithm is integrated in program package for digital
records processing, displaying and analysis.
VIII. CONCLUSIONS
The algorithm for fault location and estimation of the
segment length to inspect has been presented. The algorithm
takes into account the probabilistic nature of the equivalent
impedance of the power systems feeding the unmonitored end
of the long transmission line.

Antans Sauhats received Dipl.Eng., Cand.Techn.Sc.


and Dr.hab.sc.eng. degree from the Riga Technical
University (former Riga Polytechnical Institute) in
1970, 1976 and 1991 respectively. Since 1991 he is
Professor at Electric Power Systems. Since 1996 he
is the Director of the Power Engineering Institute of
the Riga Technical University.

Marija Danilova graduated from the Riga Technical


University, Latvia in 2001. She is currently a Ph.D.
student at the faculty of Electrical and Power
Engineering and planning engineer at the national
power company Latvenergo.