IET Generation, Transmission & Distribution

Research Article

ISSN 1751-8687
Transmission line fault detection and Received on 15th August 2014
Accepted on 14th December 2014
localisation methodology using PMU doi: 10.1049/iet-gtd.2014.0788
www.ietdl.org
measurements
Pathirikkat Gopakumar 1 , Maddikara Jaya Bharata Reddy 1, Dusmanta Kumar Mohanta 2
1
Department of EEE, NIT, Tiruchirapallli, Tamil Nadu, 620015, India
2
Department of EEE, MVGR College of Engineering Vizianagaram, Andhra Pradesh, 535005, India
E-mail: [email protected]

Abstract: Rapid advancements pertaining to measurements and computational technology have brought a paradigm shift
for operational architecture of power grids across the globe. Self-healing, a vital operational feature of emerging power
grids, necessitates real-time identification and localisation of transmission line faults for the entire power network. This
study proposes a novel support vector machine-based fault localisation methodology to precisely identify and localise
all types of transmission line faults occurring at any location in the power grid based on phasor measurement unit
(PMU) measurements. Detection of fault is achieved through PMU measurements only from a single generator bus for
the entire grid. Bus associated with fault, faulty branch and location of fault in faulty branch are calculated using fast
Fourier transform analysis of variations pertaining to equivalent voltage phasor angle (EVPA) and equivalent current
phasor angle (ECPA). The proposed methodology has been validated through extensive case studies for Western
System Coordinating Council (WSCC)-9 and IEEE-14 bus systems. The main contribution of the proposed methodology
is that the fault location information can significantly contribute to system protection center for restoration of the line
within shortest time span and initiate appropriate wide area control actions to maintain stability.

Nomenclature using wide area synchrophasor measurements have been proposed

in [4]. The algorithm presented in [4] is insensitive to power
ev equivalent voltage phasor angle (EVPA) system parameters and deliver acceptable accuracy in fault location
ei equivalent current phasor angle (ECPA) estimation. However, it seems to be complex and very challenging
i ith frequency coefcient in fast Fourier transform (FFT) task to have discrimination between measurements for the fault
spectra of EVPA variation and no-fault conditions. As a remedial measure, an adaptive PMU
i ith frequency coefcient in FFT spectra of EVPA deviation measurement-based fault location algorithm has been presented in
of a bus from a generator bus [5]. Measurements from both ends of a transmission line have
i ith frequency coefcient in FFT spectra of EVPA deviation been utilised for estimating fault location. Similar approach for
between a parent bus and connected bus multi-section compound transmission lines has been presented in
[6]. However, impacts of FR and FIA on fault location accuracy
have not been studied in these papers. Fault location estimation
1 Introduction method for double circuit transmission lines using PMU
measurements has been presented in [7]. The proposed method
Identifying the location of transmission line fault in a power grid is can accurately estimate the fault distance in the case of ground
of great importance to facilitate self-healing of the system and for faults such as line-to-ground (LG) fault or line-to-LG (LLG)
maintaining reliability of power supply [13]. Rapid advancements faults. However, accuracy of fault location is poor for line faults
pertaining to measurements and computational technologies have such as line-to-line (LL) faults and LL-to-line (LLL) faults.
paved the way for deployment of global positioning system (GPS) Majority of the research works reported in this literature, including
based synchronised phasor measurement units (PMUs) for the [17], are limited to specic network congurations such as
power grids sprawling over large geographical area. Precise transmission lines fed at one end or from both the ends and for
real-time dynamics of gigantic power grid can be achieved using multi-terminal transmission lines. For achieving complete fault
PMUs having a common time frame. Consequently, development monitoring, these methods [17] necessitate acquisition of fault
of robust and precise fault location techniques based on wide area data from all transmission lines in the grid for fault classication.
PMU measurements is one of important research interest. Moreover, the information regarding fault classication and
Numerous research works in the area of PMU-based fault location localisation need to be sent to system protection centre (SPC)
estimation have been reported in the literature. Fault location through proper communication channels. This results in heavy
estimation using synchronised voltage and current phasor computational burden. PMU-based fault location estimation
measurements from PMUs at both ends of a transmission line has technique for estimating the fault location for any branch of a
been presented in [1]. Transmission line impedance-based fault network has been presented in [8]. The proposed technique
location estimation methodology for multi-terminal transmission depends on the sequence model of the network, and hence the
lines using PMU measurements has been presented in [2]. accuracy of the fault location estimation is sensitive to network
However, inuence of fault inception angle (FIA) has not been modelling errors.
studied in these papers. Estimation of fault distance during This paper presents a transmission line fault detection and
dynamic conditions of power system has been reported in [3]. The localisation methodology for power grids, to accurately identify
method has been shown to have immunity against fault resistance and estimate the location of the fault occurring anywhere in the
(FR) and FIA. Travelling wave based fault location algorithms network using wide area PMU measurements. Real-time detection

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
& The Institution of Engineering and Technology 2015 1033
Fig. 1. Functional block diagram of the proposed methodology

of fault location is achieved using fast Fourier transform (FFT) the proposed fault localisation methodology in smart grids can
analysis of variations of equivalent voltage phasor angle (EVPA) assist SPC to isolate the fault at the earliest and initiate wide area
at any one of the generator buses. EVPA is estimated through control actions to heal the disturbances [13, 14].
Parks transformation of PMU measurements at the generator bus.
During normal operating conditions, EVPA has a constant value
and undergoes through variations during transmission line faults. 2 Formulation of the proposed methodology
Frequency domain analysis of variations pertaining to EVPA over
a time span equal to inverse of nominal power grid frequency is The proposed methodology for fault detection and localisation in
the vital constituent of the proposed methodology. Once any power grids has been corroborated in this section. The
transmission line fault is detected, the proposed methodology can
methodology encompasses various operational stages as illustrated
precisely pinpoint the bus associated with the faulty branch using in the form of a functional block diagram given in Fig. 1. In this
FFT analysis of deviations in EVPA of all buses, except for that
paper, the power grid is assumed to be completely observable
related to generator bus. This identied bus is termed as parent through PMUs installed at all buses or in optimal locations [15]. It
bus in this paper. Next step for the fault localisation is to is also assumed that all high-speed communication and
discriminate the faulty branch, as a parent bus can have multiple
computation facilities are available without any constraints so that
branches connected to it. This has been achieved using FFT latency issues can be neglected.
analysis of EVPA deviations between parent bus and all buses
As given in Fig. 1, PMU measurements across the power grid are
connected to the parent bus during the fault. Once faulty branch is
acquired and recorded at SPC through high-speed communication
detected, variations related to equivalent current phasor angle channel. On the basis of this data, real-time fault detection,
(ECPA) for that faulty branch over a time span of inverse of
detection of bus involved in the faulty branch, faulty branch
nominal power grid frequency is analysed using FFT. Fault discrimination and localisation of fault are carried out as illustrated
distance in the faulty branch from the parent bus is estimated with
in following sections.
the aid of multiclass support vector machine (SVM) using FFT
coefcients extracted from ECPA variations.
The main contribution of this paper is that the proposed 2.1 Real-time fault detection
methodology can precisely pinpoint the bus associated with the
faulty branch in a large power network. The fault location Real-time fault detection is accomplished by means of FFT analysis
necessitates discrimination of the faulty branch as many branches of EVPA at any one of the generator buses. EVPA at a given
may be connected to the identied bus. This is an advantage over generator bus can be calculated from Parks transformation of
conventional approaches. The conventional approaches only PMU measurements of voltage phasors at that bus, as given in (1)
demarcate regions within the grid where fault might have
occurred. The fault location information obtained from the
proposed methodology can signicantly contribute to SPC to
restore the faulty line at the earliest. Another application of this
type of central fault localisation algorithm bears signicance in the
upcoming power parlance such as smart grids equipped with
inverter driven distributed generation (DGs) units and microgrids
(MGs). DGs and MGs are connected to main grid although
inverters and the overall system inertia of these smart grids
become much lower than that of conventional power grids [911].
Hence, disturbances owing to occurrence and clearance of
transmission line faults can initiate power oscillations and affect
the total stability of the system [912]. Therefore application of Fig. 2. Single line diagram of WSCC-9 bus system

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
1034 & The Institution of Engineering and Technology 2015
Table 1 FFT coefficients of EVPA variations during various transmission line fault conditions for WSCC-9 bus system
Fault Branch (bus Fault conditions (fault location distance from 0 Hz 50 Hz 100 Hz 150 Hz 200 Hz 250 Hz 300 Hz
type bus) the first bus, FR, FIA) (0f ) (1f ) (2f ) (3f ) (4f ) (5f ) (6f )

LG fault 4-5 10 km, 0 , 0 0.606 0.033 0.087 0.002 0.0002 0 0

6-4 70 km, 70 , 0 0.515 0.031 0.084 0.002 0.0002 0 0
8-9 10 km, 0 , 60 0.506 0.027 0.077 0.001 0.0002 0 0
7-8 70 km, 0 , 120 0.415 0.024 0.074 0.001 0.0002 0 0
LL fault 4-5 10 km, 0 , 0 0.866 0.169 0.213 0.077 0.0027 0 0
6-4 70 km, 70 , 0 0.861 0.161 0.209 0.071 0.0021 0 0
8-9 10 km, 0 , 60 0.854 0.159 0.203 0.067 0.0018 0 0
7-8 70 km, 0 , 120 0.851 0.154 0.199 0.061 0.0014 0 0
LLG 4-5 10 km, 0 , 0 0.883 0.164 0.182 0.067 0.0018 0 0
fault 6-4 70 km, 70 , 0 0.878 0.159 0.179 0.061 0.0016 0 0
8-9 10 km, 0 , 60 0.873 0.154 0.172 0.055 0.0015 0 0
7-8 70 km, 0 , 120 0.868 0.149 0.167 0.047 0.0011 0 0
LLL 4-5 10 km, 0 , 0 1.182 0.163 0.031 0.014 0.0011 0 0
fault 6-4 70 km, 70 , 0 1.179 0.16 0.028 0.011 0.001 0 0
8-9 10 km, 0 , 60 1.172 0.153 0.021 0.009 0.0009 0 0
7-8 70 km, 0 , 120 1.167 0.150 0.018 0.007 0.0007 0 0
no fault condition 0.498 0 0 0 0 0 0
load change condition 0.597 0 0 0 0 0 0

and (2) [16] (see (1) and (2)) FFT coefcients of 0, 50, 100, 150, 200, 250, 300, 350, 400 and
450 Hz are represented by 0, 50, 100, 150, 200, 250, 300, 350,
where VRM, VYM and VBM stand for peak values of voltage phasors 400 and 450, respectively. It is obvious from Table 1 that during
in R, Y and B phases. Vdse and Vqse refer to direct and quadrature axes transmission line faults of any kind, frequency coefcients 50,
components of three-phase voltage phasors. Frequencies and s 100, 150 and 200 exist which are zero during normal operating
correspond to the frequency in radians of three-phase voltages and conditions and load change conditions. Hence, 50, 100, 150 and
rotating reference frame, respectively. The EVPA (ev) can be 200 are utilised for real-time fault detection. Values of these
calculated from direct and quadrature axes voltages as (3) [16] coefcients and 0 can be estimated mathematically as given in
(4)(8), respectively


Vqse
 N
Fev = tan1 (3) 1 1
2pn
Vdse D50 = F (n) e N (4)
N n=0 ev
During normal operating conditions, EVPA has a constant value and
 N1
when transmission line fault occurs, it undergoes variation. Variation 1  4pn
D100 = F (n) e N (5)
in EVPA is because of the impact of harmonic voltages created in the N n=0 ev
system as a consequence of ow of fault generated harmonic currents

 N1
through transmission line impedances. Harmonic currents are 1  6pn
generated in the system because of waveform distortion happening D150 = F (n) e N (6)
N n=0 ev
during occurrence and isolation of transmission line faults.
Frequency components present in the EVPA variation are
 N1
estimated using FFT analysis over a time span of inverse of the 1  8pn
D200 = F (n) e N (7)
nominal power grid frequency (1/f ). Since ECPA has constant N n=0 ev
value during normal conditions, it will be reected as a constant

 N
1
value at 0 Hz in FFT spectra. During fault conditions, multiple 1
frequencies exist because of EVPA variations. Hence, frequency D0 = F (n) (8)
N n=0 ev
spectra of EVPA at a generator bus are a direct measure for
detection of transmission line fault that occurred at some location
within the power grid. For better understanding, a case study is where N is the number of samples of EVPA in the time span
presented with WSCC-9 bus system, as shown in Fig. 2 [17]. considered.
All the three generators in the system are assumed to be of same
rating with 50 Hz as operating frequency. Transmission lines are 2.2 Identication of the parent bus (bus connected to the
modelled with positive and negative sequence values of resistance, faulty branch)
inductance and capacitance as 1.17 , 47.55 mH and 0.62 F per
phase per 50 km, respectively. The zero sequence values of the Identication of the parent bus is achieved through frequency
same are 19.42 , 162.54 mH and 0.422 F per phase per 50 km, domain analysis of deviation of EVPA of each bus in power grid
respectively [12]. FFT spectral coefcients of EVPA during from that of a generator bus. Any generator bus can be taken as
various fault conditions such as LG, LL, LLG and LLL faults are reference bus. For calculating EVPA deviations of all the buses, it
given in Table 1. For a comparative study, the same coefcients is not necessary that all the buses to be installed with PMUs.
for no-fault and load change conditions are also included in However, the system must be completely observable with PMUs,
Table 1. PMU measurements of voltage phasors at bus-1 are taken so that voltage phasors at buses not installed with PMUs can be
for study. estimated from a neighbouring PMU installed buses. In the FFT






2   2p 2p 2p 2p
Vdse = VRM sin(vt ). sin vs t + VYM sin vt . sin vs t + VBM sin vt + . sin vs t + (1)
3 3 3 3 3






2   2p 2p 2p 2p
Vqse = V sin(vt ). cos vs t + VYM sin vt . cos vs t + VBM sin vt + . cos vs t + (2)
3 RM 3 3 3 3

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
& The Institution of Engineering and Technology 2015 1035
Table 2 Fault conditions considered in WSCC-9 bus system Table 3 FFT coefficients of EVPA deviations of each bus from that of G1
in WSCC-9 bus system
Fault type Branch Parent bus Distance, km FR, FIA,
Fault details Reference Bus FFT coefficients of EVPA deviation of
LG 9-6 9 10 0 0 bus each bus from that of G1
LL 4-5 4 50 10 20
LLG 8-7 8 70 50 45 0 50 100 150 200
LLL 6-4 6 85 100 60
LG fault in bus-1 1 0 0 0 0 0
branch (9-6), 2 0.089 0.002 0.01 0.001 0.001
10 km from 3 1.592 0.007 0.022 0.003 0.006
spectrum of EVPA deviation, it can be observed that the parent bus bus-9, FR = 0 4 0.877 0.015 0.113 0.011 0.011
, FIA = 0 5 1.514 0.017 0.124 0.012 0.014
will have highest coefcient for 100 Hz component (100) value. 6 2.373 0.066 0.399 0.034 0.05
This is illustrated here with a case study conducted in WSCC-9 7 1.446 0.017 0.127 0.012 0.014
bus system with fault conditions as given in Table 2. 8 2.399 0.052 0.322 0.029 0.038
Frequency coefcients corresponding to 0, 50, 100, 150 and 9 2.822 0.113 0.628 0.051 0.098
LL fault in bus-1 1 0 0 0 0 0
200 Hz in the FFT spectra of EVPA deviation of each bus in the branch (4-5), 2 0.156 0.062 0.637 0.056 0.162
grid from generator bus-1 are represented by 0, 50, 100, 150 50 km from 3 1.688 0.07 0.776 0.068 0.175
and 200. FFT is computed over a time span (1/f ). FFT spectral bus-4, FR = 4 0.731 0.187 2.252 0.171 1.536
coefcients for all the fault conditions considered in Table 2 are 10 , FIA = 5 1.468 0.042 0.448 0.051 0.166
20 6 2.235 0.088 0.868 0.089 0.388
given in Table 3. It is evident from Table 3 that for the parent bus, 7 1.384 0.022 0.224 0.024 0.079
100 has greater numerical value than other coefcients (50, 150 8 2.224 0.033 0.341 0.037 0.114
and 200) for all unbalanced faults. This implies that for all 9 2.424 0.033 0.37 0.04 0.125
unbalanced faults, 100 has more inuence to the fault than that of LLG fault in bus-1 1 0 0 0 0 0
branch (8-7), 2 0.072 0.014 0.189 0.022 0.011
50, 150 and 200. The parent bus number and corresponding 70 km from 3 1.616 0.007 0.08 0.01 0.006
value of 100 are highlighted in Table 3 for better clarity. Hence, bus-8, FR = 4 0.761 0.027 0.312 0.037 0.02
100 of EVPA deviation from generator bus can be utilised for 50 , FIA = 5 1.428 0.096 0.989 0.124 0.177
parent bus identication. 45 6 2.059 0.05 0.516 0.062 0.051
7 1.402 0.129 1.291 0.166 0.297
Since EVPA of generator G1 has been taken as reference for 8 2.465 0.218 1.949 0.262 0.676
parent bus identication, EVPA deviation of bus-1 and 9 2.398 0.076 0.706 0.085 0.096
corresponding FFT coefcients reects as zeros. Moreover, 0 LLL fault in bus-1 1 0 0 0 0 0
cannot be opted for faulty bus identication as it has inuence branch (6-4) 2 0.143 0.004 0.004 0.002 0.001
85 km from 3 1.713 0.013 0.01 0.003 0.006
with load changing conditions (results given in Table 1 validate bus-6, FR = 4 0.554 0.027 0.017 0.003 0.009
this statement). Hence, utilising this coefcient may result in 100 , FIA = 5 1.275 0.022 0.008 0.003 0.009
wrong conclusions regarding faulty bus. 60 6 9.997 1.182 0.094 0.188 0.185
7 1.217 0.021 0.007 0.003 0.009
8 1.966 0.026 0.01 0.004 0.013
9 2.232 0.013 0.004 0.003 0.012
2.3 Faulty branch discrimination (in the parent bus)

A parent bus may have multiple number of branches connected to it.

Exact faulty branch connected to the parent bus can be discriminated given (9) and (10) [16] (see (9) and (10))
using FFT analysis of deviation of EVPA of all connected buses from
that of parent bus. Similar to above cases, this EVPA deviations are where IRM, IYM and IBM stand for peak values of current phasors in
also analysed over a time period of (1/f ). Frequency coefcients R, Y and B phases, respectively. Idse e
and Iqs refers to direct and
corresponding to 0, 50, 100, 150 and 200 Hz in the FFT spectra of quadrature axes components of three-phase voltage phasors. The
EVPA deviation of a connected bus from parent bus are angle stands for the power factor angle.
represented by 0, 50, 100, 150 and 200. It has been observed The ECPA (ei) can be calculated from direct and quadrature axes
from the FFT spectrum that 100 will have least value in EVPA currents as (11) [16]
deviation spectrum that corresponds to the connected bus which
share the faulty branch. This is because EVPA of parent bus and
the connected bus that share faulty branch will have similar EVPA
e
Iqs
variation. Hence, when both get subtracted (for nding the Fei = tan1 e (11)
deviation in EVPA), 100 results in low value. The same case study Ids
conducted in the preceding section has been extended here for
validating the proposed faulty branch discrimination. The results of
the case study are tabulated in Table 4, which validate that the It has been observed from extensive case studies that magnitude of
the frequency components present in the ECPA variation during
connected bus sharing faulty branch with parent bus has least value
of 100. The branch number of faulty branch and corresponding transmission line fault depend on the location of the fault
occurred. Similar to EVPA discussed in the preceding section,
value of 100 are highlighted in Table 4 for better clarity.
ECPA also has a constant value during no-fault conditions and
undergoes variations during transmission line fault conditions.
2.4 Fault location estimation (from parent bus) Frequency coefcients of 0, 50, 100, 150 and 200 Hz are found to
be existing in FFT spectrum during fault conditions, in similar
A novel fault location algorithm using FFT analysis of ECPA is way that of EVPA. The frequency coefcients of ECPA for the
presented here. ECPA is estimated in similar fashion of EVPA but above-mentioned frequencies are represented by 0, 50, 100,
with three-phase line currents. Park transformations of currents are 150 and 200, respectively. The same can be estimated as given






2     2p 2p 2p 2p
e
Ids = IRM sin vt f . sin vs t + IYM sin vt f . sin vs t + IBM sin vt + f . sin vs t + (9)
3 3 3 3 3






2     2p 2p 2p 2p
e
Iqs = I sin vt f . cos vs t + IYM sin vt f . cos vs t + IBM sin vt + f . cos vs t + (10)
3 RM 3 3 3 3

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
1036 & The Institution of Engineering and Technology 2015
Table 4 FFT coefficients of EVPA deviations between parent bus and connected buses in WSCC-9 bus system
Fault details (type, branch, distance, FR, FIA) Parent bus Connected bus Branch FFT coefficients of EVPA deviation of connected
bus from parent bus

0 50 100 150 200

LG, (9-6), 10 km, 0 , 0 9 3 9-3 1.08 0.07 0.42 0.03 0.05

6 9-6 0.31 0.01 0.03 0.01 0
8 9-8 0.33 0.03 0.2 0.01 0.03
LL, (4-5), 50 km, 10 , 20 4 1 4-1 0.56 0.1 1.09 0.09 0.43
6 4-6 0.88 0.05 0.57 0.03 0.28
5 4-5 0.74 0.01 0.22 0.01 0.14
LLG, (8-7), 70 km, 50 , 45 8 7 8-7 1.06 0.09 0.69 0.10 0.39
9 8-9 0.06 0.14 1.24 0.17 0.58
LLL, (6-4), 85 km, 100 , 60 6 9 6-9 1.88 0.05 0.04 0.01 0
4 6-4 0.06 0.03 0.01 0 0

in (12)(16)

 N
1
1
C0 = F (n) (12)
N n=0 ei

 N
1
1
C50 = F (n) e(2pn/N) (13)
N n=0 ei

 N
1
1
C100 = F (n) e(4pn/N) (14)
N n=0 ei

Fig. 3. In WSCC-9 bus system, variation of FFT coefcients of ECPA

 N
1
1
during no-fault condition C150 = F (n) e(6pn/N) (15)
N n=0 ei

Fig. 4. In WSCC-9 bus system, variation of FFT coefcients of ECPA with fault distance during
a LG fault
b LL fault
c LLG fault
d LLL fault

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
& The Institution of Engineering and Technology 2015 1037
 Fig. 6. Block diagram of a single SVM module utilised in the proposed fault
localising methodology

Fig. 5. Functional block diagram of the fault localising methodology

 N1
1 
C200 = F (n) e(8pn/N ) (16)
N n=0 ei

Variations of the coefcients given in (12)(16) with fault location

are illustrated in Figs. 3 and 4 for the same case studies considered
in the preceding section for WSCC-9 bus system. Fig. 3 shows the
value of coefcients during no-fault condition, which indicates that
during this condition 0 is constant and all other components
remain at zero. Variations in the coefcients given in (12)(16)
during LG, LL, LLG and LLL fault conditions are illustrated in
Fig. 4. It is obvious from Fig. 4 that the FFT coefcients of ECPA
variations vary in magnitude with location of the fault occurred.
Since the FFT coefcient data are clustered, machine intelligence Fig. 7. Single line model of IEEE-14 bus system
techniques are essential for extracting fault distance from the same.
SVM has been opted in this paper for the real-time estimation of
fault location from FFT coefcients. Development of SVM SVM classier. Excellent generalisation properties and robustness
classiers for fault location estimation is illustrated in the in dealing with multi-dimensional data make multiclass SVM
following sections. classier to be an ideal tool for fault localising in power grids [18,
19]. In classication problems, SVM seeks a linear decision
2.4.1 SVM classier for fault localisation: Fault location is surface that separates classes of data such that maximum distance
estimated from the FFT coefcients with the help of multiclass can be achieved between data lay in border line. These border

Fig. 8. In IEEE-14 bus system, EVPA deviation of buses from that of G1 (bus-1) during
a No-fault condition
b During LG fault condition
c During LL fault condition

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
1038 & The Institution of Engineering and Technology 2015
Table 5 FFT coefficients of EVPA deviations during various transmission line fault conditions in IEEE-14 bus system
Fault details Reference bus Bus FFT coefficients of EVPA deviation of each bus
from that of G1

0 50 100 150 200

no fault bus-1 1 0 0 0 0 0
2 0.029 0 0 0 0
3 0.066 0 0 0 0
4 0.087 0 0 0 0
5 0.079 0 0 0 0
6 0.145 0 0 0.000 0
7 0.119 0 0 0.000 0
8 0.138 0 0 0.000 0
9 0.134 0 0 0.000 0
10 0.156 0 0 0.000 0
11 0.16 0 0 0.000 0
12 0.162 0 0 0.000 0
13 0.161 0 0 0.000 0
14 0.157 0 0 0.000 0
LG fault at branch (3-2), 25 km from bus-3, FR = 0 , FIA = 0 bus-1 1 0 0 0 0.000 0
2 0.029 0.003 0.001 0.002 0
3 0.073 0.003 0.015 0.001 0.002
4 0.089 0.001 0.005 0.001 0.001
5 0.079 0.002 0.003 0.001 0
6 0.145 0.001 0.004 0.001 0.001
7 0.121 0.001 0.005 0.001 0.001
8 0.14 0.001 0.005 0.001 0.001
9 0.135 0.002 0.005 0.001 0.001
10 0.157 0.001 0.005 0.001 0.001
11 0.161 0.001 0.004 0.001 0.001
12 0.163 0.002 0.004 0.001 0.001
13 0.161 0.002 0.004 0.001 0.001
14 0.158 0.001 0.004 0.001 0.001
LL fault at branch (11-6), 60 km from bus-11, FR = 25 , FIA = 30 bus-1 1 0 0 0.00 0 0
2 0.029 0.003 0.002 0.002 0
3 0.061 0.002 0.004 0.001 0.001
4 0.078 0.002 0.008 0.001 0.001
5 0.068 0.001 0.011 0.001 0.002
6 0.135 0.001 0.034 0.001 0.011
7 0.108 0.001 0.012 0.001 0.002
8 0.127 0.001 0.012 0.001 0.002
9 0.122 0.001 0.017 0.001 0.004
10 0.147 0.001 0.029 0.001 0.008
11 0.158 0.001 0.047 0 0.02
12 0.153 0.001 0.031 0.001 0.01
13 0.151 0.002 0.029 0.001 0.008
14 0.145 0.001 0.022 0.001 0.006
LLG fault at branch (5-4), 70 km from bus-5, FR = 60 , FIA = 50 bus-1 1 0 0 0 0 0
2 0.029 0.003 0.001 0.001 0
3 0.061 0.003 0.009 0.001 0.001
4 0.083 0.003 0.021 0.001 0.004
5 0.073 0.002 0.026 0.001 0.006
6 0.141 0.002 0.025 0.001 0.006
7 0.115 0.003 0.022 0.001 0.005
8 0.134 0.001 0.022 0.001 0.005
9 0.13 0.002 0.023 0.001 0.005
10 0.152 0.001 0.024 0.001 0.006
11 0.156 0.001 0.024 0.001 0.006
12 0.158 0.001 0.025 0.001 0.007
13 0.157 0.001 0.025 0.001 0.006
14 0.152 0.001 0.024 0.001 0.006
LLL fault at branch (12-13) 80 km from bus-12, FR = 100 , FIA = 75 bus-1 1 0 0 0 0 0
2 0.029 0.001 0.001 0.001 0
3 0.055 0.002 0.001 0.001 0
4 0.07 0.003 0.001 0.001 0
5 0.051 0.001 0.001 0.001 0.001
6 0.123 0.001 0.002 0.001 0.001
7 0.101 0.002 0.001 0.001 0.001
8 0.119 0.001 0.001 0.001 0.001
9 0.114 0.001 0.001 0.001 0.001
10 0.135 0.002 0.001 0.001 0.001
11 0.139 0.001 0.001 0.001 0.001
12 0.14 0.001 0.004 0.001 0.001
13 0.139 0.001 0.002 0.001 0.001
14 0.136 0.001 0.001 0.001 0.001

lines at different types of data are called support vectors. If no such set that lie at the border of a particular class distributions. This
linear decision surface exists to separate the data, SVM maps the data facilitates training cases included in the other classes to be
into a higher-dimensional space known as the feature space where effectively discarded. Therefore SVM approach to classication
the separating linear decision surface exists and can be determined. necessitates the training samples that lie on the class boundaries
The function used for mapping the data to higher-dimensional alone for class discrimination. This helps SVM to generalise the
space is called kernel function. SVM explores optimal separating data more accurately on unseen cases, and hence the training error
hyperplane between data classes by focusing on the training data comes much smaller than other popular articial intelligence

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
& The Institution of Engineering and Technology 2015 1039
Table 6 FFT coefficients of EVPA deviations between parent bus and connected buses in IEEE-14 bus system
Fault details (type, branch, distance, FR, FIA) Parent bus Connected bus 0 50 100 150 200

LG, (3-2), 25 km, 0 , 0 3 2 0.019 0 0.005 0 0.001

4 0.04 0 0.008 0 0.001
LL, (11-6), 60 km, 25 , 30 11 10 0.011 0 0.018 0 0.011
6 0.022 0 0.013 0 0.009
LLG, (5-4), 70 km, 60 , 50 5 6 0.067 0 0.017 0 0.008
4 0.01 0 0.005 0 0.002
1 0.073 0 0.026 0 0.006
LLL, (12-13), 80 km, 100 , 75 12 6 0.011 0 0.002 0 0
13 0.006 0 0.001 0 0

approaches such as neural networks [18]. Multiclass classication branch number. Branch selector at input side directs the input data
problem can be accomplished in either one-against-one strategy or (0, 50, 100, 150 and 200) to that SVM module which is
one-against-all strategy or one-step strategy. This paper adopts trained for the detected faulty branch. Branch selector at output
one-step strategy for fault localisation, since the classication of side directs the SVM module output (fault location distance) to
data from a cluster of data occurs in a single step. In general, output port. Faulty branch number input facilitates precise
one-stage strategy usually necessitates fewer support vectors in selection of SVM module in synchronous with that at the input
comparison with the other multiclass classication strategies based side to obtain output.
on binary SVMs. However, in contrast to other strategies, one-step The FFT coefcients 0, 50, 100, 150 and 200 form the input
strategy necessitates a series of analysis to be conducted that set for SVM classier. Set of fault location distances corresponding
directly depends on the number of classes. Detailed discussion of to each entry on the input data set form the output data set. Block
SVM classication is available in the literatures [18, 19]. In this diagram of single SVM module deployed in the proposed fault
paper, library for SVM toolbox in MATLAB/SIMULINK localising methodology is shown in Fig. 6 for better clarity.
environment has been utilised for training and testing of SVM Sufcient number of samples in the training data set with higher
classier. Radial basis function kernel has been adopted as it is diversity has been maintained by incorporating samples pertaining
excellent for non-linear input mapping [20]. to various FRs and FIAs for achieving enhanced boundary to
The functional block diagram of the proposed fault location exactly determine fault distance. Application of articial
methodology using multiclass SVM classiers is shown in Fig. 5. intelligence methods including SVM suffers an inherent
It consists of m number of SVM modules trained for each branch shortcoming that they cannot automatically adapt to new network
of the network, where m is the number of branches in the topology. Manual retraining is necessary once the system changes
network (b1, b2, , bm). Branch selector is present at two sides of from its actual topology. However, this scenario is not considered
the SVM modules which operate synchronously with the faulty in this paper.

Fig. 9. In IEEE-14 bus system, variation in 0, 50, 100, 150 and 200 with fault distance during
a LG fault condition
b LL fault condition
c LLG fault condition
d LLL fault condition

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
1040 & The Institution of Engineering and Technology 2015
3 Results and discussion of 100 for all four fault cases are highlighted in Table 5 for better
clarity. After identication of the parent bus, EVPA deviations of
The proposed transmission line fault detection and localising all connected buses from the parent bus are estimated and FFT has
methodology for power grids has been validated through case been performed on this deviation. The FFT coefcients extracted
studied conducted on IEEE-14 bus system, as shown in Fig. 7 from these EVPA deviations between buses for all four fault cases
[16]. All buses in the system are assumed to be installed with are tabulated in Table 6. It is clear from Table 6 that FFT
PMUs. The results of the case study are discussed in this section. coefcient 100 of EVPA deviation between parent bus and that bus
The system has been modelled in MATLAB/SIMULINK platform connected through faulty branch will have least value compared
similar to that of WSCC-9 bus system presented in Section 2.1. with that of other connected branches. The branch number of faulty
EVPA deviation of all buses except generator buses from that of branch and corresponding value of 100 are highlighted in Table 6
G1 (bus-1) during various operating conditions are illustrated in for better clarity. . Location of the fault has been identied through
Fig. 8. Fig. 8a displays the EVPA deviations during no-fault FFT analysis of ECPA variation in the faulty branch current. FFT
conditions, which illustrate that EVPA deviations are of constant coefcients 0, 50, 100, 150 and 200 are utilised for
value during this operating conditions. The FFT spectral identication of fault location. Variations of these FFT coefcients
coefcients of the same are tabulated in Table 5. The constant with fault distance for all the fault conditions assumed in Table 5
value of EVPA deviations reect as a non-zero value for 0 Hz are shown in Fig. 9. Multiclass SVM has been employed for
component and zero value of all other frequency components in estimating fault location from these FFT coefcients.
FFT spectra. EVPA deviations during LG fault occurred in branch
(3-2) connecting bus-3 and bus-2 at 25 km from bus-3 with FR and
FIA of 0 and 0, respectively, is shown in Fig. 8b. Fault is 3.1 Comparative study with other machine intelligence
occurred at 0.04 s, and hence up to this time EVPA deviation has a techniques
constant value and afterwards undergoes variations as a result of
fault occurred. The FFT spectral coefcients 0, 50, 100, 150 and Efciency of SVM classiers for the proposed fault localising
200 are tabulated in Table 5. EVPA variations during LL fault methodology has been compared with other popular machine
occurred in branch (11-6) at 60 km from bus-11 with FR of 25 intelligence methods such as articial neural network (ANN) and
and FIA of 30 is shown in Fig. 8c. Similar to LG fault, up to 0.04 adaptive neuro-fuzzy inference system (ANFIS). IEEE-14 bus
s EVPA is of constant value and then experiences variation. system has been opted as test case for validating the superiority of
Similar behaviour can be observed for LLG and LLL faults. The SVM classiers.
fault conditions assumed and frequency coefcients are given in ANN-based classier is developed with one input layer, two
Table 5. In contradiction to no-fault case, frequency coefcients hidden layers and one output layer trained using feed-forward back
other than 0 exist in fault cases. This condition is utilised for propagation method [21]. Input layer contains ve neurons, hidden
real-time detection of transmission line faults. It is obvious from layers consist of 20 neurons and the output layer has one neuron.
Table 5 that the bus to which faulty branch is connected has the When training the network with feed-forward back propagation
highest 100 value. The parent bus number and corresponding value method, the error was observed to be reducing after 2000 epochs.

Fig. 10. For IEEE-14 bus system, %error in fault location estimation during
a LG fault in branch 3-2 at 25 km away from bus-3 with FIA = 0 and FR = 0
b During LL fault in branch 11-6 at 60 km away from bus-11 with FIA = 30 and FR = 25
c LLG fault in branch 5-4 at 70 km away from bus-5 with FIA = 50 and FR = 60
d LLL fault in branch 12-13 at 80 km away from bus-5 with FIA = 75 and FR = 100

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
& The Institution of Engineering and Technology 2015 1041
An ANFIS-based classier is implemented with hybrid optimisation 3 He, Z.Y., Mai, R.K., He, W., Qian, Q.Q.: Phasor-measurement-unit-based
method for training fuzzy inference system [22]. The ANFIS transmission line fault location estimator under dynamic conditions, IET Gener.
Transm. Distrib., 2011, 5, (11), pp. 11831191
classier consists of 650 membership functions. 4 Korkali, M., Lev-Ari, H., Abur, A.: Traveling-wave-based fault-location technique
SVM, ANN and ANFIS are trained for all four types of for transmission grids via wide-area synchronized voltage measurements, IEEE
conditions, as considered in preceding section, with the same Trans. Power Syst., 2012, 27, (2), pp. 10031011
inputoutput data set of 800 samples, which consists of 200 5 Joe-Air, J., Jun-Zhe, Y., Ying-Hong, L., Chih-Wen, L., Jih-Chen, M.: An adaptive
PMU based fault detection/location technique for transmission lines. I. Theory and
samples for each type of four faults. After successful training of algorithms, IEEE Trans. Power Deliv., 2000, 15, (2), pp. 486493
the networks, the percentage error in the fault location estimation 6 Liu, C., Lin, T., Yu, C., Yang, J.: A fault location technique for two-terminal
for all four cases has been shown in Fig. 10. It is clear from the multisection compound transmission lines using synchronized phasor
gures that percentage error of fault location is almost zero for measurements, IEEE Trans. Smart Grid, 2012, 3, (1), pp. 113121
7 Kang, N., Liao, Y.: Double-circuit transmission-line fault location utilizing
SVM, whereas that of ANN and ANFIS varies with the fault synchronized current phasors, IEEE Trans. Power Deliv., 2013, 28, (2),
distance. This validates the superiority of SVM classiers for the pp. 10401047
proposed fault location methodology over ANN and ANFIS. 8 Jiang, Q., Li, X., Wang, B., Wang, H.: PMU-based fault location using voltage
measurements in large transmission networks, IEEE Trans. Power Deliv., 2012,
27, (3), pp. 16441652
9 Donnelly, M.K., Dagle, J.E., Trudnowski, D.J., Rogers, G.J.: Impacts of the
4 Conclusion distributed utility on transmission system stability, IEEE Trans. Power Syst.,
1996, 11, (2), pp. 741746
10 Slootweg, J.G., Kling, W.L.: Impacts of distributed generation on power system
Smart fault location methodologies for real-time protection of power transient stability, Proc. IEEE Power Eng. Soc. Summer Meet., 2002, 2,
grids are gaining worldwide attention from power engineers. This pp. 862867
paper proposes a novel fault detection and localisation 11 Gopakumar, P., Reddy, M.J.B., Mohanta, D.K.: Letter to the editor: stability
concerns in smart grid with emerging renewable energy technologies, Electr.
methodology to precisely detect and localise all types of Power Compon. Syst., 2014, 42, (34), pp. 418425
transmission line faults occurring at any location in power grid. 12 Ma, J., Zhang, P., Fu, H., Bo, B., Dong, Z.: Application of phasor measurement
Detection of the fault is achieved based on PMU measurements unit on locating disturbance source for low-frequency oscillation, IEEE Trans.
obtained from a single generator bus for the entire grid. Smart Grid, 2010, 1, (3), pp. 340346
13 Seethalekshmi, K., Singh, S.N., Srivastava, S.C.: A synchrophasor assisted
Identication of the bus associated with the fault, discrimination of frequency and voltage stability based load shedding scheme for self-healing of
the faulty branch and location of the fault in that branch are power system, IEEE Trans. Smart Grid, 2011, 2, (2), pp. 221230
achieved through FFT analysis of variations in EVPA and ECPA. 14 Hashiesh, F., Mostafa, H.E., Khatib, A., Helal, I., Mansour, M.M.: An intelligent
Multiclass SVM classier has been deployed for estimating wide area synchrophasor based system for predicting and mitigating transient
instabilities, IEEE Trans. Smart Grid, 2012, 3, (2), pp. 645652
location of the fault. The main contribution of this paper is that 15 Lien, K.P., Liu, C.W., Yu, C.S., Jiang, J.A.: Transmission network fault location
the proposed methodology can precisely localise any transmission observability with minimal PMU placement, IEEE Trans. Power Deliv., 2006, 21,
line fault occurring at any location within power grid based on (3), pp. 11281136
single generator bus PMU measurements. The fault location 16 Kundur, P.: Power system stability and control (Tata McGraw-Hill Education Pvt.
Ltd., New Delhi, 2012)
information achieved from the proposed methodology can 17 Seethalekshmi, K., Singh, S.N., Srivastava, S.C.: A classication approach using
signicantly contribute to SPC to restore the faulty line at the support vector machines to prevent distance relay maloperation under power swing
earliest and initiate necessary wide area control action for healing and voltage instability, IEEE Trans. Power Deliv., 2012, 27, (3), pp. 11241133
smart grid stability issues owing to occurrence and isolation of the 18 Salat, R., Osowski, S.: Accurate fault location in the power transmission line using
support vector machine approach, IEEE Trans. Power Syst., 2004, 19, (2),
fault. The methodology has been validated through extensive case pp. 979986
studies conducted for WSCC-9 and IEEE-14 bus systems. 19 Ravikumar, B., Thukaram, D., Khincha, H.P.: Comparison of multiclass SVM
classication methods to use in a supportive system for distance relay
coordination, IEEE Trans. Power Deliv., 2010, 25, (3), pp. 12961305
20 Chang, C.C., Lin, C.J.: LIBSVM: a library for support vector machines, ACM
5 References Trans. Intell. Syst. Technol., 2 (27), pp. 12727, Available at http://www.csie.
ntu.edu.tw/~cjlin/libsvm
1 Lin, Y.H., Liu, C.W., Yu, C.S.: A new fault locator for three-terminal transmission 21 Gracia, J., Mazon, A.J., Zamora, I.: Best ANN structures for fault location in
lines using two-terminal synchronized voltage and current phasors, IEEE Trans. single-and double-circuit transmission lines, IEEE Trans. Power Deliv., 2005,
Power Deliv., 2002, 17, (2), pp. 452459 20, (4), pp. 23892395
2 Liu, C.W., Lien, K.P., Chen, C.S., Jiang, J.A.: A universal fault location technique 22 Reddy, M.J.B., Mohanta, D.K.: Adaptive-neuro-fuzzy inference system approach
for N-terminal (N 3) transmission Lines, IEEE Trans. Power Deliv., 2008, 23, for transmission line fault classication and location incorporating effects of power
(3), pp. 13661373 swings, IET Gener. Transm. Distrib., 2008, 2, (2), pp. 235244

IET Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 10331042
1042 & The Institution of Engineering and Technology 2015