own techniques.

Index TermsFault Location, Phasor Matching, Genetic
Algorithm, Simulation, Digital Fault Recorders.
I. INTRODUCTION
HERE are many fault location methods for a specific
transmission line applications, such as one-end[1], two-
end[2], three-end[3], utilizing the voltages and currents
measured at the line end(s) to calculate the impedance. Another
method, based on a stand-alone recording device to capture the
high frequency transient signal generated by faults [4], is
utilizing traveling wave-based algorithm to locate the fault.
However, it is difficult to locate a fault in a transmission network
when data obtained from only a limited number of recording
devices is available. This paper proposes an approach using
waveform matching to locate fault even when only limited data
is available.
The proposed approach utilizes the data obtained from the
recording devices installed in the power system. Here, the
recording devices may include digital fault recorders (DFRs),
digital relay or other intelligent electronic device. The recorded
data may include analog quantities (voltages and currents), as
well as digital quantities (breaker and relay operation status).
When a fault occurs in the system, some devices are triggered
and corresponding records are sent to the central office. The
data may be limited but it can still be used to locate the fault
based on the proposed approach.

This work was supported by PSerc Consortium, an NSF I/UCRC, and in
part by Texas A&M University.
Mladen Kezunovic is with Dept. of Electrical Engineering, Texas
A&M University, College Station, TX 77843 (e-mail:
kezunov@ee.tamu.edu)
Shanshan Luo works presently as a research associate with Dept. of
Electrical Engineering, Texas A&M University, College Station, TX
77843. (e-mail: shanshan@ee.tamu.edu)
Donald R. Sevcik is with Reliant Energy HL&P, P.O. Box 1700,
Houston, TX 77251-1700, (e-mail: don-sevcik@reliantenergy.com)
In our work, we define the limited data as sparse data.
Sparse data may result for two reasons: 1. DFR or digital relays
may not be installed at every substation or bus for monitoring
purpose. 2. Every DFR may not always be triggered under fault
condition. Whatever the case, for most fault cases, only limited
measurements may be available. Under this circumstance, the
mentioned methods to locate a fault in a transmission network
cannot be utilized. This paper aims at proposing a more flexible
fault location method, which matches the recorded during-fault
waveform with the simulated during-fault waveform.
First, the waveform matching idea is explained. Next, the
algorithm implementation is discussed. Test results and
conclusions are given at the end.
II. W
AVEFORM MATCHING

In the waveform matching approach, a power system model
is used to carry out simulation study. A fault needs to be posed
to obtain the during-fault simulated waveforms. Then the
matching is made between the simulated waveforms and the
recorded waveforms to determine the fault location based on the
degree of matching. Theoretically, the simulated fault waveform
will match completely with the recorded fault waveform if the
assumed fault location and fault resistance correspond to the
real fault condition.
The process to determine the fault location is iterative
because several lines in the system and variety of possible fault
resistances should be searched to obtain the optimal matching.
In the practical operation, most probable fault locations are
searched firstly by selecting a certain fault resistance. Changing
the fault resistance according to a specific increment, fault
locations are searched thoroughly. The process will proceed till
the selected sections in power system and possible fault
resistance range are exhausted.
After the search is completed, the fault location is determined
based on the optimal matching scheme. There are two possible
schemes - the phasor matching and transient matching. In the
phasor matching, short circuit studies are carried out to
obtain phasors under fault condition. In the transient
matching, transient simulations are carried out to obtain
transient waveform. We utilize the phasors for matching. The
matching degree can be represented by a value obtained from
the following criterion.
(
)
=
= + =
v
i
N
k
N
k
kr
ks
ki
kr
ks
kv
f
c
I
I
r
V
V
r
R
x
f
1
1
,
(1)
where,
Mladen Kezunovic, Fellow, IEEE, Shanshan Luo, Donald R. Sevcik, member, IEEE

A Novel Method for Transmission Network Fault Location
Using Genetic Algorithms and Sparse Field Recordings
T
2
( )
f
c
R
x
f
,
-the cost function using phasors for matching, it is a
non-negative number
f
R
x,
-the fault location and fault resistance.
ki
kv
r
r ,
-weights for the errors of the voltages and currents
respectively
kr
ks
V
V ,
-the during-fault voltage phasors obtained from the
short-circuit studies and from recorded waveforms respectively
kr
ks
I
I ,
-the during-fault current phasors obtained from the
short-circuit studies and from recorded waveforms respectively
i
v
N
N ,
-the numb er of the selected voltage and current
phasors respectively
k - the index of the voltage or current phasors
The cost function will be zero when the phasors obtained
from the simulation studies exactly match those obtained from
the recorded waveforms. The best fault location estimation will
be achieved when the cost function is at a minimum. Therefore,
the problem of fault location estimation is actually the
optimization problem.
Since there may be several local minimum and maximum
points, it is difficult to use the gradient-based method to find
the global minimum. The GA based optimization approach is a
good choice to search for the global optimal solution. We have
to convert the minimization problem to maximization problem in
order to utilize the GA. That requires us to convert the cost
function to a fitness function of GA. The simplest conversion is
to multiply the cost function by a minus one. We have to add a
constant to make the corresponding fitness function positive.
The fitness function is as follows:
(
)
(
)
f
c
f
R
x
f
C
R
x
f
,
,
max =
(2)
where,
(
)
f
R
x
f ,
is fitness function
max
C
is the maximum fitness value in the current population.
In Equation (2), fault location and resistance are selected as
two variables. They are represented as binary strings in GA.
Three GA operators are generally used: selection, crossover and
mutation. The selection operator mimics the process of natural
selection of strings to create a new generation, where the fittest
members reproduce most often. The crossover, applied with a
probability, acts on a pair of selected members providing the
exchange of binary string. The mutation, applied with a
probability, randomly affects the single bit in a member. The GA
search process is as follows: at the beginning, the initial
population is generated randomly. Then posing the fault
according to the initial population, the short circuit study is
carried out to obtain the simulated during-fault phasors and
further calculate the fitness value for each individual. The next
generation is produced by applying the three steps as described
above. The process is repeated until the best match is found.
III. T
HE IMPLEMENTATION OF THE FAULT LOCATION SYSTEM

A. Overall Architecture
The overall architecture of the fault location solution is
shown in Fig.1. Two commercial software packages, represented
as dotted line in Fig.1, are utilized. One is DFR Assistant [5],
utilized to analyze the fault waveform, relay breaker and
communication channel data based on an expert system. It also
converts the DFR raw data into COMTRADE format [6]. DFR
Assistant can generate an analysis report including the fault
type and possible faulted line. Another is PSS/E (PTI Power
System Simulator) [7]. It can calculate the power flow and carry
out the short circuit study.
The input modules represented by broken lines include DFR
data files, the interpretation files, fault information entered by
user, and power system model. The first three items are
necessary for each monitored substation. The last one is used
as the input to PSS/E.
The main modules of the software are discussed next.




































Fig. 1 Architecture of the fault location software
B. Data Requirement
The data requirement includes: static system model of power
system, fault data, substation interpretation data and fault
DFR Assistant
DFR raw data
DFR data in
COMTRADE format
Interpretation
file
Recorded phasors
and breaker status
Static system
model
PSS/E load flow
program
Pre-fault
simulated phasors
Synchronizing
DFR data
Algorithm
selection
Special cases
apply?
Tuning system
model
Generating
candidates
Fault
information
entered by users
Initializing two
variables randomly
PSS/E short
circuit study
During-fault
simulated phasors
Matching phasors
using (2)
Convergence
criterion met?
Selection, crossover
and mutation
Apply two-end,
three-end or one-
end algorithm
Fault location
report
stop
stop
Y
N

Y

N
3
information entered by the user.
The static system model refers to the saved case of PSS/E. It
should contain the power flow raw data, sequence impedance
data and system top