Received: 24 January 2018 Revised: 22 June 2018 Accepted: 12 July 2018

DOI: 10.1002/etep.2700

RESEARCH ARTICLE

Detection of three‐phase fault during power swing using

zero frequency filtering

Prabhu M. S.1 | Paresh Kumar Nayak1 | Gayadhar Pradhan2

1
Department of Electrical Engineering,
Summary
Indian Institute of Technology (Indian
School of Mines), Dhanbad, India Distance relay is susceptible to maloperation during power swing. Power swing
2
Department of Electronics and blocking function is integrated in distance relays to ensure security during
Communication Engineering, National power swing. However, if a fault occurs during power swing, it must be
Institute of Technology, Patna, India
detected and cleared by the relay as quickly as possible. Since power swings
Correspondence and three‐phase faults are both balanced phenomena, discriminating them is
Prabhu M. S., Senior Research Fellow,
very challenging. This paper proposes a method to detect three‐phase fault dur-
Electrical Engineering Department,
Indian Institute of Technology (Indian ing power swing by computing the Hilbert envelope of the current waveform
School of Mines), Dhanbad 826004, India. followed by zero‐frequency filtering. The average rate of change in amplitude
Email: [email protected]
of the zero‐frequency filtered current is used as the fault detector index. The
discrimination capability between three‐phase fault and power swing of the
extracted feature is the main contribution of this paper. Results of various pos-
sible test cases simulated through EMTDC/PSCAD show that the proposed
method is fast and reliable in detecting three‐phase faults during power swings.

KEYWORDS
distance protection, power swing, power swing blocking, three‐phase fault detection, zero frequency
filtering

1 | INTRODUCTION

Power system disturbances such as faults, switching on/off of large loads, disconnections of generators, line switching
etc., cause oscillations in rotor angles among generators. This can result severe swings in the power‐flow.1 During power
swing, the impedance measured by a distance relay may enter into its operating zones and can cause the relay to trip
lines unnecessarily. Investigations on past blackouts worldwide (including the 2012 blackout in India) depict that

Abbreviations: ANFIS, adaptive neural fuzzy inference system; DFT, discrete Fourier transform; DTFT, discrete time Fourier transform; FFT, fast
Fourier transform; HE, Hilbert envelop; IDFT, inverse discrete Fourier transform; OST, out‐of‐step tripping; PSB, power swing blocking; SVM,
support vector machine; TMI, transient monitor index; WT, wavelet transform; ZFF, zero‐frequency filter; ZFFO, zero‐frequency filter output
Symbols: i(n), current signal measured at nth instant; ia(n), analytic signal derived from the measured current signal at nth instant; ih(n), Hilbert
transform of measured current signal at nth instant; *, signal convolution; h(n), Hilbert transform operator at nth instant; H(ω), Fourier transform
of Hilbert operator; Ih(ω), Fourier transform of current signal; hi(n), magnitude of the analytic signal measured at nth instant; Φ(n), phase of the
analytic signal measured at nth instant; x(n), first‐difference of Hilbert envelope of current signal measured at nth instant; y(n), digital resonators
output at nth instant; N, number of samples per cycle; byðnÞ, digital resonators output after trend removal at nth instant; ak, coefficients of digital
resonators; yðnÞ, local mean of digital resonators output at nth instant; ΔZFFOph , average rate of change of zero‐frequency filtering output; ΔZFFO,
fault detector index; ΔZFFOth , threshold for fault detector index; ΔR, rate of change of apparent resistance; Isup, superimposed components of
current; |ψ(I2)|, magnitude of Teager‐Kaiser energy operator of negative‐sequence current

Int Trans Electr Energ Syst. 2019;29:e2700. wileyonlinelibrary.com/journal/etep © 2018 John Wiley & Sons, Ltd. 1 of 15
https://doi.org/10.1002/etep.2700
2 of 15 M. S. ET AL.

distance relay maloperation during power swing is one of the major contributing reasons for such blackouts in the
power systems.2-5
Power swing blocking (PSB) function is integrated in modern distance relays to avoid unintended tripping during
power swing. However, if a fault occurs during PSB activation must be detected and the relay should be unblocked
to trip the line. Power swing is a balanced phenomenon. Discrimination of unsymmetrical faults from power swing is
easy due to the presence of negative and/or zero‐sequence components of the current and voltage in such faults.
However, distinguishing three‐phase faults from power swing is very challenging as both these phenomena are of
balanced nature.
Quantities such as voltage, current, apparent impedance, active, and reactive powers at the relay location vary as a
function of machine rotor angle during power swing, but the variations in those quantities are slow. On the other hand,
the inception of a fault causes abrupt changes in those quantities. Such discriminations are applied in the PSB algo-
rithms as the criteria to block or unblock the distance relays during power swing.1 The rate of change of the apparent
impedance, the blinder, the rate of change of resistance, and the rate of change of the swing centre voltage are generally
used as features in the conventional PSB schemes for differentiating faults from power swing. The relative advantages
and disadvantages of such schemes are well documented in IEEE and Benmouyal et al.1,6 A fault detector using
superimposed components of current is proposed in Apostolov et al and Khoradshadi7,8 to differentiate three‐phase
faults from power swing. Although the method is simple to implement, it has limitation to detect the three‐phase faults
producing small change in current magnitude during power swing.
In recent years, many alternative solutions have been proposed to detect three‐phase faults during power swing. In
Lin et al,9 a cross‐blocking scheme on the basis of the derivative of the three‐phase active and reactive power is proposed
to detect three‐phase faults during power swing. However, the swing cases demonstrated are mainly of unstable types.
For such unstable swings, the system would have collapsed. A differential power scheme based on the autoregression
technique is proposed in Rao and Pradhan10 to detect three‐phase faults during power swing. The performance of the
presented scheme depends on an appropriate selection of the autoregression parameters for which an extensive simu-
lation study is required. Another scheme based on the fast Fourier transform (FFT) coefficients of the fundamental com-
ponents of the instantaneous three‐phase active power is proposed in Mahamedi and Zhu.11 The method requires both
the voltage and the current data for its implementation. The scheme performs well in discriminating slow swings from
three‐phase faults. However, discriminating fast swings from three‐phase fault still remains a challenge for the
approach. In addition to that, even the non‐fault transients may be misinterpreted as the fault by the method. In Nayak
et al,12 a technique based on cumulative sum of the change in the negative‐sequence current magnitude is presented to
detect faults during power swings in a series‐compensated line. However, determining the optimal threshold settings for
discriminating faults from power swing is very challenging. Further, another approach based on the extraction of the
transient monitor components of the dynamic current signal is proposed in Khodaparast and Khederzadeh.13 As the
approach is based on the capturing of transient components of current signal, transients caused due to non‐fault events
such as load switching may end up being misinterpreted as fault. Recently, a fault detector based on phase‐space
method is proposed in Dubey et al14 for detecting the three‐phase faults during power swing. However, the performance
of the scheme is mainly demonstrated for three‐phase faults occurring in the zone‐1 region of the distance relay during
the unstable power swing only.
Several signal‐processing techniques like the wavelet transform15-17 as well as the supervised learning methods such
as the adaptive neurofuzzy inference system,18 the support vector machines,19 decision‐tree based scheme,20 and the
combination of S‐transform and the probabilistic neural network21 have been applied to the current signals to extract
the relevant feature for discriminating three‐phase faults from power swing. Recently, an integrated approach based
on Teager‐Kaiser energy of instantaneous zero sequence voltage and negative‐sequence current is also proposed in
Kumar and Jena22 for detecting fault during power swing. The fault detection times for the reported schemes range from
5 to 30 ms for a 50‐Hz power system.
However, fault detectors based on the wavelet transform technique require a high sampling rate to effectively cap-
ture the different fault situations. On the other hand, the supervised learning methods require a large number of train-
ing patterns to be generated for different power system operating conditions. Furthermore, such approaches might
require re‐training in case the configuration of the power system changes significantly.
This paper proposes a method to detect three‐phase fault during power swing based on the detection of frequency
components of the measured current signals. The proposed approach confines the analysis around the zero‐frequency
component of the current waveform. Since a single frequency component is used for detecting the fault, the proposed
method does not require a high sampling rate unlike the aforementioned signal processing techniques. Moreover, there
M. S. ET AL. 3 of 15

is no need to acquire a large number of examples to train the statistical models. The proposed method is tested for var-
ious possible three‐phase fault cases during power swing simulated through EMTDC/PSCAD both in the WSCC three‐
machine, nine‐bus system and IEEE 39‐bus New England system. The performance of the fault detector is found to be
fast and reliable. The rest of the paper is organized as follows. The detailed description of the proposed method is pro-
vided in Section 2. The experimental results for the simulation studies are presented in Section 3. The comparative
assessment of the proposed method with four existing techniques is provided in Section 4. Finally, the paper is con-
cluded in Section 5.

2 | PROPOSED METHOD

The work presented in this paper is based on the premise that the occurrence of a three‐phase fault in a transmission
line results in an abrupt change in the frequency components of the current waveform. The different frequency compo-
nents present in the current waveform at the inception of a three‐phase fault are mainly fundamental, harmonics, and
decaying dc components.23 The frequency of the decaying dc component is close to zero Hz. On the other hand, such
low‐frequency components are absent during power swing.11 Consequently, continuous analysis and tracking of those
extremely low‐frequency components of the current waveform can be used to detect the occurrence of a three‐phase
fault during power swing. In order to extract those low‐frequency components, first the Hilbert envelope (HE)24,25 of
the current waveform is computed in the present work. This helps to enhance the energy due to the low‐frequency
components in comparison to the high‐frequency ones. Next, the HE of current waveform is passed through a
zero‐frequency filter (ZFF).26 In general, the signal energy after zero‐frequency filtering will be nearly zero during
power swing. On the other hand, the zero‐frequency filtered signal will be of significantly high energy at the inception
of a three‐phase fault. The rate of change of the zero‐frequency filtered signal over one cycle is then employed to dis-
tinguish a three‐phase fault from power swings. In the following, the detailed descriptions for the two steps involved
in the proposed method are presented.

2.1 | Hilbert envelope of current waveform

The HE of any analytic signal gives the magnitude of the signal.24 An analytic signal is a complex‐valued function that
has no negative frequency components. The HE of a signal can be computed as follows.24,25 Let ia(n) be the analytic
signal derived from the given current waveform i(n) as

ia ðnÞ ¼ iðnÞ þ jih ðnÞ (1)

where, ih(n) is the Hilbert transform of i(n). The Hilbert transform of i(n) is given as

ihðnÞ ¼ iðnÞ*hðnÞ (2)

where, * denotes convolution and

8 9
< 2sin2 ðπn=2Þ =
; n≠0
hðnÞ ¼ πn (3)
: ;
0; n¼0

Alternatively, the Hilbert transform can be computed through the use of discrete time Fourier transform (DTFT) and its
inverse. If the Fourier transform pairs are denoted as

iðnÞ↔I ðωÞ (4)

hðnÞ↔H ðωÞ (5)

then the Hilbert transform of i(n) is given by

ih ðnÞ ¼ IDTFT fI h ðωÞg ¼ IDTFT fI ðωÞH ðωÞg (6)

4 of 15 M. S. ET AL.

where,

þ jI ðωÞ; −π ≤ ω ≤ 0
I h ðω Þ ¼ (7)
− jI ðωÞ; 0≤ω≤π

As already mentioned, the HE envelope of any signal is defined as the magnitude of the corresponding analytic signal.
Hence, the HE of i(n), ie, hi(n) is given as
hi ðnÞ ¼ jia ðnÞj (8)

or
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
hi ðnÞ ¼ i2 ðnÞ þ i2h ðnÞ (9)

At the same time, the phase Φ(n) of the analytic signal ia(n) is given as

ih ðnÞ
ΦðnÞ ¼ tan−1 (10)
iðnÞ

The earlier reported works have shown that the HE of hi(n) captures the relatively low‐frequency components while
the high‐frequency components are captured by the phase of the analytic signal Φ(n).25 Consequently, the HE of the
current waveform is exploited in this work to enhance the energy of the low‐frequency components.

2.2 | Zero frequency filtering of Hilbert envelope

After boosting the low‐frequency components, the next step in the proposed approach is to filter the frequency compo-
nents around zero Hz. This is achieved with the use of the zero‐frequency filtering approach reported in literature. The
ZFF is a resonator with a narrow bandwidth centered around zero‐frequency.26,27 The deviation in ZFF filtered signal
due to the fault is computed by subtracting a local mean at each time instant. The local mean is computed for quarter
cycle. The steps involved in the processing of a signal to obtain the zero‐frequency filtered signal are summarized as
follows28,29:
• Determine the first‐difference x(n) of the HE of the current signal ia(n) as

x ðnÞ ¼ ia ðnÞ þ ia ðn−1Þ (11)

• Next, compute the output of a cascade of two ideal digital resonators at zero Hz as
4
yðnÞ ¼ − ∑ ak yðn−kÞ þ xðnÞ (12)
k¼1

where the values a1 = 4, a2 = −6, a3 = 4, a4 = −1 are determined by cascading two ideal resonators at zero Hz.26,27
• Remove the trend, ie,
byðnÞ ¼ yðnÞ þ yðnÞ (13)
where,
1 N
y ðnÞ ¼ ∑ y ðn Þ (14)
ð2N þ 1Þ n¼‐N

and 2N + 1corresponds to the samples in one quarter period of the signal period.
• The trend removed signal byðnÞ is referred to as the ZFF output signal (ZFFO)

As stated earlier, the inception of a three‐phase fault results in the occurrence of low‐frequency components that
are near and around zero Hz. In order to substantiate this fact, spectral analysis of the current signals for only one
cycle during power swing and three‐phase fault during power swing is performed using discrete Fourier transform
(DFT), and the same are shown in Figure 1A. The solid red line in the plot shows the magnitude spectrum during
M. S. ET AL. 5 of 15

Fault during power swing Only power swing

30

20

Magnitude
(A)
10

0
0 100 200 300 400 500
Frequency (Hz)
60

Magnitude 40
(B)
20

0
0 100 200 300 400 500
Frequency (Hz)
40

30
Magnitude

20 (C)

10

0
0 100 200 300 400 500
Frequency (Hz)

FIGURE 1 A, Magnitude spectra for one cycle of the current signal; B, magnitude spectra after computing the Hilbert envelope of the
considered current signal; and C, magnitude spectra after zero‐frequency filtering of the Hilbert envelope

pure power swing. On the other hand, the spectrum for fault during power swing is shown by solid blue line. It is
clearly visible that during power swing, frequency components near and around zero Hz are almost absent while
their presence can be easily noted for the three‐phase fault during power swing. On computing the HE of the current,
those low‐frequency components are enhanced while the high‐frequency components get suppressed significantly.
The same can be visualized from the magnitude spectra shown in Figure 1B. Zero frequency filtering further removes
the effect of higher‐frequency components, and this is very much evident from the spectra shown in Figure 1C. From
Figure 1B,C, it can be noted that low‐frequency components are more pronounced after the inception of a three‐
phase fault during power swing in comparison to pure power swing. Thus, one can easily discriminate a three‐phase
fault from the power swing through the proposed analysis. From Figure 1C, it is also observed that the rate of change
of ZFFO is more pronounced at the inception of a three‐phase fault during power swing compared with only power
swing. The use of this attribute for detecting three‐phase fault during power swing is described in the following
subsection.

2.3 | Decision criterion

The current waveforms for a three‐phase fault during the power swing and their corresponding HEs and the zero‐fre-
quency filtered outputs are shown in Figure 2A,B,C, respectively. The details of the data used to generate these plots
are given later in Section 3. The HE is known to capture the slow variations of the current waveform as discussed ear-
lier. This is very much evident from Figure 2B. Next, zero‐frequency filtering is performed on the HE in order to empha-
size the extremely low‐frequency components. As already mentioned earlier, the occurrence of fault results in abrupt
change in the frequency components of the current waveform. Some of those frequency components are extremely
low while such low frequencies are not present in non‐fault cases as well as during the power swing. This attribute is
clearly visible from the peaks in the zero‐frequency filtered output (ZFFO) currents shown in Figure 2C. In order to fur-
ther enhance the large valued peaks and to suppress the other ripples having very small amplitudes, average rate of
change of ZFFO for any particular phase is computed over one cycle of the signal as
6 of 15 M. S. ET AL.

FIGURE 2 Sub‐figures A, shows three‐phase currents for a three‐phase fault during power swing, while their corresponding Hilbert
envelopes are shown in B. The resulting output currents after zero‐frequency filtering (ZFFO) are depicted in C. The average rate of
change of ZFFO (ΔZFFOph) for the three‐phases are shown in D

1 N
ΔZFFOph ¼ ∑ jZFFOðnÞ−ZFFOðn−1Þ (15)
N n¼‐N

where, N is the number of samples in one cycle and ZFFO is the zero‐frequency filtered output current for that phase. In
this study, N = 20. It is expected that the ΔZFFOph will be faster at the inception of a fault while the rate of change will be
nearly zero elsewhere. It is evident from Figure 2D that ΔZFFOph assumes a large value only at the inception of fault
while it is insignificantly small and constant at all other points. The most important attribute of the proposed feature
(ΔZFFOph) is that it is independent of the swing cycle silp frequency. On the other hand, the HEs as well as the zero‐fre-
quency filtered outputs do exhibit some variations with respect to the slip frequency. Consequently, we have employed
the ΔZFFOph curve for discriminating three‐phase faults from power swings in this paper.
From Figure 2D, it is observed that the ΔZFFOph obtained for the three‐phase currents are of different magnitudes.
This is due to the fact that there is a 120° phase shift between the different phases in a three‐phase system. To obtain
a purposeful index for faster detection of three‐phase faults during power swing, the sum of the values of ΔZFFOph
obtained from the three‐phase currents is used in this paper. The fault detector index is defined as

ΔZFFO ¼ ΔZFFOa þ ΔZFFOb þ ΔZFFOc (16)

where, ΔZFFOa, ΔZFFOb, and ΔZFFOc are the average rate of change of ZFFO for phase‐a, phase‐b, and phase‐c, respec-
tively. Consequently, the proposed method detects a three‐phase fault during the power swing if,

ΔZFFO > ΔZFFOth (17)

where, ΔZFFOth is a suitable threshold selected for reliably differentiating three‐phase faults from power swings. In this
work, the value of ΔZFFOth is set at 0.5. Such a value of threshold was chosen after a systematic study of a series of fault
cases simulated both on the WSCC three‐machine, nine‐bus system and IEEE 39‐bus New England system. The set
value of ΔZFFOth is immune to fault location, fault inception angle, swing cycle slip frequency, and sudden change of
M. S. ET AL. 7 of 15

large block of loads during power swing. Of course, the setting for ΔZFFOth is not a fixed one for all types of power sys-
tems. When the system configurations are changed significantly, the threshold value needs to be modified accordingly
based on transient study.
As already mentioned earlier, detecting unsymmetrical faults during power swing is easy due the presence of neg-
ative and/or zero‐sequence current in such faults. On the other hand, it is difficult to discriminate three‐phase faults
from power swing since both happen to be balanced in nature. Based on the fault inception angle, the fault location,
and the fault resistance, the low‐frequency components may or may not be present in the case of an unsymmetrical
fault. Consequently, the proposed method will not be suitable for detecting all varieties of unsymmetrical faults during
power swing. Therefore, the proposed method should not be implemented stand alone. Rather, the proposed method
should be used along with the existing sequence component‐based simple fault detection schemes which effectively
detect unsymmetrical faults but fail to detect three‐phase faults during power swing. The same is shown as block dia-
gram in Figure 3.

3 | S I M U L A TI O N R E S U L T S A N D D I S C U S S IO N S

3.1 | Performance evaluation on WSCC three‐machine, nine‐bus system

The proposed method is tested by generating data through EMTDC/PSCAD on WSCC three‐machine, nine‐bus system
as shown in Figure 4A.30 In the presented work, the relay R1 at bus‐7 is considered for performance evaluation of the
proposed method. The zone‐1 of relay R1 is set at 90% of line 7‐8, and zone‐2 reach covers up to 30% of line 8‐9. The
reach of zone‐3 is set 40% more of line 8‐9 such that it can provide backup protection to full length of line 8‐9. The gen-
eral practice for applying PSB function in distance protection scheme is to block all of its zones during power swings
and allow tripping if the swing is unstable by using out‐of‐step tripping (OST) functions available with distance relays
at pre‐selected locations of the power system.1 During PSB activation, three‐phase faults are created at different fault
inception times/angles, slip frequencies, and locations of zone‐1, zone‐2, and zone‐3 regions of the distance relay R1.
The results for the fault cases in zone‐1 and zone‐2 regions of the relay R1 are provided in the following subsections.
The results for the zone‐3 fault cases in the zone‐3 region are provided in Section 4 of the paper.

3.1.1 | Far end three‐phase fault (zone‐2 region) during slow power swing
The relay R1 observes slow swing (slip frequency = 2 Hz approx.) due to the removal of line 7‐5 following the clearance
of a three‐phase fault on it (Figure 4A). During power swing, the locus of the apparent impedance seen at the relay R1 is

FIGURE 3 Block diagram representing the complete process of detecting three‐phase as well as unsymmetrical faults during power swing
8 of 15 M. S. ET AL.

(A) (B)

(C) (D)
FIGURE 4 Performance evaluation on WSCC 3‐machine, nine‐bus system for three‐phase fault during slow and fast power swing. A,
Single line diagram of the WSCC three‐machine, nine‐bus system; B, apparent impedance seen at relay R1 for far end three‐phase fault
during slow power swing (time interval between adjacent points is 0.01 s); C, performance for far end three‐phase fault on line 7‐8 during
slow power swing (swing frequency = 2 Hz approx.); and D, performance for close‐in three‐phase fault on line 7‐8 during fast power swing
(swing frequency = 7 Hz approx.)

plotted in the R‐X diagram shown in Figure 4B. From the R‐X diagram, it is observed that the apparent impedance has
entered both into zone‐2 and zone‐3 characteristics of the relay R1 during power swing and stays there more than their
respective operating times. This can result in maloperation of the relay R1. Thus, to avoid unintended tripping, the dis-
tance relay needs to be blocked for such swing condition.
To test the performance of proposed scheme for detecting three‐phase fault during slow power swing, a three‐phase
fault is created at 10.4 seconds at 95% of line 7‐8 from bus‐7 (in the zone‐2 region of the relay R1). The observations from
this study are provided in Figure 2, while detailed explanations for the same have already been provided earlier. The
performance of the proposed method is shown in Figure 4C. From the shown plot, it is clearly visible that the fault
detector index ΔZFFO is below the preset threshold during power swing and significantly high at the inception of the
three‐phase fault. This helps to identify the three‐phase fault by the proposed method within 5 ms after its inception
even if the change in current magnitude from swing to fault is not significant.

3.1.2 | Close‐in three‐phase fault (zone‐1 region) during fast swing

In the case of close‐in three‐phase faults, the voltage at the relay bus collapses. Due to the subsidence transients in the
coupling capacitor voltage transformer, the fault detectors based on voltage phasors, for example, the rate of change of
impedance or the rate of change of swing‐center voltage will be affected. The performance of the proposed scheme is not
affected by the close‐in faults since only the current signal is made use of. Therefore, the occurrence of close‐in faults is
M. S. ET AL. 9 of 15

not an issue for the proposed scheme. To test the performance of proposed scheme in such a case, a three‐phase fault is
created at 5.53 seconds at a distance of 15 km from relay R1 on line 7‐8 (Figure 4A) during fast power swing (slip fre-
quency = 5 Hz approx.) as discussed earlier. The corresponding results are provided in Figure 4D. From the figure, it is
very much evident that the fault detector index ΔZFFO remains well below the preset threshold during power swing. As
soon as a three‐phase fault has initiated, the value of ΔZFFO increases rapidly to a high value and exceeds the threshold
within 3 ms after the initiation. Thus, the performance of the proposed scheme is not affected in the case of close‐in
three‐phase fault during power swing.

3.1.3 | On detection time for different three‐phase faults during power swing
To see the overall performance of the proposed method, different simulations are carried out by creating three‐phase
faults at different locations of the line 7‐8 (Figure 4A) by varying fault inception time/angle and swing cycle slip fre-
quency. Negative to positive zero‐crossing of phase‐a voltage is taken as reference for fault inception angle calculation.
The response time of the proposed fault detector for different fault cases are provided in Table 1. From the table, it is
clearly observed that for all the fault cases the response time is below a quarter cycle. This shows the strength of the
proposed method.

3.2 | Performance evaluation on IEEE 39‐bus New England System

The performance of the proposed method is also evaluated by generating data on the IEEE 39‐bus test system
(Figure 5A).30 In this system the relay R2 at bus‐29 protecting line 29‐28 is considered for study. The removal of line
29‐26 following a three‐phase fault on it introduces swing condition for relay R2. Different fault cases are simulated
on line 29‐28 during power swing to test the algorithm. Due to page limitation, results of only two important test cases
are discussed below.

3.2.1 | Performance during slow swing

In this case, first a three‐phase fault is created at 0.7 seconds on line 29‐26 at a distance of 100 km from bus‐29, and the
line is removed at 1.0 seconds for clearing the fault. The removal of line 29‐26 introduces swing condition for the relay
R2. During power swing, a three‐phase fault is created at the middle of the line 29‐28 at 3.6 seconds (slip‐fre-
quency = 1 Hz approx.). The results for the test case are provided in Figure 5B. From the result, it is clearly observed
that the fault detector index ΔZFFO is well below the set threshold during power swing, and as soon as a three‐phase
fault has occurred at 3.6 seconds the index exceeds the preset threshold within 5 ms and able to detect the fault. This

TABLE 1 Results on fault detection time

Fault Inception Angle (Deg.)/ Slip Frequency, Hz, Fault Detection

Fault Location, % Fault Inception Time, s (Approx.) Time, ms

10% (30 km) 36°/ (3.93 s) 1.0 3

90°/ (8.635 s) 4.0 3
144°/ (9.988 s) 7.0 3
30% (90 km) 54°/ (3.931 s) 1.0 3
108°/ (8.636 s) 4.0 3
162°/ (9.989 s) 7.0 3
50% (150 km) 36°/ (3.93 s) 1.0 3
90°/ (8.675 s) 4.0 3
144°/ (9.938 s) 7.0 3
70% (210 km) 54°/ (3.931 s) 1.0 4
108°/ (8.676 s) 4.0 4
162°/ (9.939 s) 7.0 4
90% (270 km) 36°/ (3.85 s) 1.0 4
90°/ (8.66 s) 4.0 4
144°/ (9.858 s) 7.0 4
10 of 15 M. S. ET AL.

(A)

(B) (C)

FIGURE 5 Performance evaluation on IEEE 39‐bus New England System for three‐phase fault during slow and fast power swing. A, Single
line diagram of the IEEE 39‐bus New England System; B, performance for three‐phase fault on line 29‐28 during slow power swing (swing
frequency = 1 Hz approx.); and C, performance for three‐phase fault on line 29‐28 during fast power swing (slip frequency = 5 Hz approx.)

shows that for the 39‐bus test system there is no need of changing the threshold setting for discriminating three‐phase
faults from power swing.

3.2.2 | Performance during fast swing

As time progresses, the slow swing discussed in the above case becomes fast swing. At 8.04 seconds, a three‐phase fault
is created at 100 km from bus‐29 on line 29‐28 with a swing cycle slip frequency close to 7 Hz. The results for the test
case are provided in Figure 5C. The observation on the result clearly shows that despite the increase of swing cycle
slip frequency, the fault detector index remains below the preset threshold during power swing. With the
inception of a three‐phase fault at 8.04 seconds, the index ΔZFFO exceeds the preset threshold within 5 ms and can
be able to detect the fault. This shows that the proposed method is able to detect the three‐phase fault even occurring
during fast swing.

4 | CO MP A R ATI V E AS S E S S ME N T

The advantages and disadvantages of different existing methods for detecting the three‐phase faults during power swing
have already been discussed in Section 1. Among the discussed existing approaches, the four important methods, (1) the
superimposed components of current (Isup),7 (2) the method based on FFT coefficients,11 (3) the method based on tran-
sient monitor index (TMI),13 and (4) the method based on magnitude of Teager‐Kaiser energy operator of negative‐
M. S. ET AL. 11 of 15

sequence current (ψ(I2))22 are considered for comparison with the proposed method. The results for the simulation
studies performed for the three typical test cases on WSCC three‐machine, nine‐bus system (Figure 4A) are provided
in the following subsections.

4.1 | Three‐phase fault in the zone‐3 region of distance relay during slow power swing
A three‐phase fault is created at 200 km (50% of line 8‐9) from bus‐8 (Figure 4A) at 9.76 seconds during slow power
swing (slip frequency = 2 Hz approx.). The locus of the apparent impedance seen at relay R1 is plotted in R‐X diagram
as shown in Figure 6A. The comparative assessment for this fault case is provided in Figure 6B. From Figure 6B(i), it is
clearly observed that the inception time and location of the fault result in a very small change in the magnitude of the
current from power swing to fault. As a result, the performance of the techniques based on the superimposed compo-
nents of current (Isup) (Figure 6B(ii)) and TMI of phase current (Figure 6B(iv)) will definitely find difficulty or will
exhibit a delayed response in discriminating such faults from power swing. From Figure 6B(iii), it is observed that
the scheme based on the FFT coefficients has ability to discriminate the three‐phase fault from slow power swing.
However, the amplitudes of all the three aforementioned methods vary with the slip frequency of the swing cycle.

(A)

(B)

FIGURE 6 Comparative assessment for the case of three‐phase fault in zone‐3 region of the relay R1 during slow power swing on WSCC
three‐machine, nine‐bus system. A, Apparent impedance seen at relay R1 (time interval between the adjacent point is 0.01 s) and B,
comparative results of the proposed method with Isup, the method based on FFT coefficients, TMI, and |ψ(I2)|
12 of 15 M. S. ET AL.

Consequently, selecting a suitable threshold for discriminating three‐phase faults from power swing by these methods
will become a difficult task.
On the other hand, the observation on Figure 6B(v) clearly shows that the method based on the magnitude of
Teager‐Kaiser energy operator of negative‐sequence current (ψ(I2)) has limitation to detect such types of three‐phase
faults that occur in zone‐3 region of the relay during slow power swing. However, from Figure 6B(vi), it is clearly
observed that the fault detection index ΔZFFO of the proposed method is independent of swing cycle slip frequency
and exceeds the threshold within 5 ms after the inception of the fault even if the change in current magnitude from
swing to fault is not significant. Thus, the proposed method outperforms the existing approaches.

4.2 | Three‐phase fault in the zone‐3 region of distance relay during fast power swing
The pre‐fault condition of the nine‐bus power system (Figure 4A) is set in such a manner that the removal of line 7‐5
following a three‐phase fault on it introduces fast swing (slip frequency above 3 Hz)15 for the relay R1 at bus‐7. During
fast power swing, the locus of the apparent impedance seen by relay R1 is plotted in the R‐X plane as shown in Figure 7
A. From the figure, it is clearly evident that the seen impedance has crossed the line impedance through zone‐2 and

(A)

(B)

FIGURE 7 Comparative assessment for the case of three‐phase fault in zone‐3 region of the relay R1 during fast power swing on WSCC
three‐machine, nine‐bus system. A, Apparent impedance seen at relay R1 (time interval between the adjacent point is 0.01 s) and B,
comparative results of the proposed method with Isup, the method based on FFT coefficients, TMI, and |ψ(I2)|
M. S. ET AL. 13 of 15

zone‐3 characteristics and has completed two fast swing cycles. Further, the time of stay in the zone‐2 region is around
0.5 seconds which is beyond the delay time of the zone‐2 element. Therefore, the relay would definitely maloperate.
Thus, to avoid unintended tripping, PSB function of the relay is activated.
For the comparative assessment, a three‐phase fault is created at 5.22 seconds at 300 km (75% of line 8‐9) from bus‐8.
The results for this simulation study are provided in Figure 7B. In this case also, as the fault is created far away from the
relay location, the change of current magnitude from swing to fault is insignificant which is clearly seen in the current
waveforms (Figure 7B(i)). From Figure 7B(ii), it is noticeable that the value of Isup depends on swing cycle slip frequency
and as the change in current magnitude from power swing to fault is not significant resulting a low amplitude of Isup.
Consequently, selecting a suitable threshold for discriminating fast power swing from three‐phase fault by using Isup will
definitely become a difficult task. From Figure 7B(iii), it can be noted that the magnitude of the fundamental compo-
nent of the FFT coefficients is significant during fast swing. At the same time, the FFT coefficients at the inception
of the three‐phase fault during fast swing are of comparable magnitude. This clearly shows that the use of FFT coeffi-
cients is not a good indicator for discriminating three‐phase fault from fast power swing.
From Figure 7B(iv), it is observed that using TMI of phase current three‐phase fault during fast swing can be
detected. However, as TMI varies with swing cycle slip frequency setting of proper threshold for discriminating some
typical three‐phase fault cases from fast swing may be difficult by this method. Similar to the previous case, in this case
also the method based on the magnitude of Teager‐Kaiser energy operator of negative‐sequence current (|ψ(I2)|) fails to
detect the three‐phase fault during fast swing (Figure 7B(v)). On the other hand, in the proposed method, the value of
the fault detector index ΔZFFO is independent of swing cycle slip frequency even during fast power swing and lies below
preset threshold (Figure 7B(vi)). As soon as the fault is initiated, ΔZFFO exceeds the preset threshold and is able to detect
the fault at a response time of 5 ms. This is the merit of the proposed scheme over the existing ones for detecting three‐
phase faults during fast swing.

4.3 | Performance during change in power flow

A change in the power flow in a line can affect the voltage and the current waveforms at the relay location. Such a
change should not be misinterpreted by the relay as a fault if it happens during power swing. A change in power flow
in the line during slow power swing is simulated by suddenly withdrawing 50% of the load that is connected to bus‐8 at

FIGURE 8 Comparative assessment of

the proposed method with Isup, the
method based on FFT coefficients, TMI,
and |ψ(I2)| for the case of sudden load
change during power swing on WSCC
three‐machine, nine‐bus system
14 of 15 M. S. ET AL.

9.8 seconds (Figure 4A). The comparative assessment for this non‐fault event is shown in Figure 8. The removal of the
large load at bus‐8 causes a sudden change in the current magnitude at bus‐7 which is observed in Figure 8(i). This sud-
den change in current magnitude results a change in Isup (Figure 8(ii)) as well as in TMI (Figure 8(iv)). This may lead to
misinterpretation of such event as a fault case. The observation on Figure 8(iii) clearly shows that the effect of load
switching on the amplitude of the FFT coefficient is negligible. However, it is observed that the amplitude of the
FFT coefficient increases afterwards. This increase may be misinterpreted as a fault.
The fault detection index based on the magnitude of Teager‐Kaiser energy operator of negative‐sequence current
(|ψ(I2)|) is not affected on load switching which is clearly evident from Figure 8(v). Similar to |ψ(I2)|‐based method,
the proposed fault detection index ΔZFFO falls well below the threshold (Figure 8(vi)). Therefore, such cases are treated
as a non‐fault event by the proposed method. Consequently, the relay will remain silent in such situations. This implies
that the proposed scheme has immunity to non‐fault transients.

5 | CONCLUSION

This paper presents a simple technique based on zero‐frequency filtering of three‐phase currents to detect three‐phase
faults during power swings. The performance of the proposed method is evaluated through simulation studies for
numerous test cases by varying the fault location, fault inception angle, and slip frequency of the power swing. The
results clearly show that using the proposed method, a reliable and fast (within a quarter cycle for a 50‐Hz power sys-
tem) detection of three‐phase fault during power swing can be accomplished. The comparative results also demonstrate
the merits of proposed method over the existing methods.

ORCID
Prabhu M. S. http://orcid.org/0000-0003-3419-2138
Paresh Kumar Nayak http://orcid.org/0000-0001-6853-3266

R EF E RE N C E S
1. IEEE Power system relaying committee of the IEEE power engineering society, power swing and out‐of‐step considerations on transmis-
sion line. Technical Report. PSRC WG D6 2005.
2. NERC. Disturbance Reports North American Electric Reliability Council. Technical Report: NJ:1996 2001.
3. Final report on the August 14, 2003 blackout in the United States and Canada: causes and recommendations. Technical Report U.S.
Canada Power System Outage Task Force. 2004.
4. Report of the enquiry committee on grid disturbance in northern region on 30th July 2012 and in northern, eastern and north‐eastern
region on 31st July 2012. Technical Report. Power Grid, New Delhi, India 2012.
5. Novosel D, Bartok G, Henneberg G, Mysore P, Tziouvaras D, Ward S. IEEE PSRC report on performance of relaying during wide‐area
stressed conditions. IEEE Trans Power Delivery. 2010;1(25):3‐16. https://doi.org/10.1109/TPWRD.2009.2035202
6. Benmouyal G, Hou D, Tziouvaras D. Zero‐setting power‐swing blocking protection. The 31st Annual Western Protective Relay Confer-
ence. 2004: 19‐21.
7. Apostolov AP, Tholomier D, Richards SH. Superimposed components based subcycle protection of transmission lines. IEEE Power Syst
Confe Exposition. 2004;1:592‐597. https://doi.org/10.1109/PSCE.2004.1397508
8. Khoradshadi ZH. Evaluation performance comparison of power swing detection algorithms. Proc IEEE Power Eng Soc General Meeting.
2005;2:1842‐1848. https://doi.org/10.1109/PES.2005.1489280
9. Lin X, Gao Y, Liu P. A novel scheme to identify symmetrical faults occurring during power swings. IEEE Trans. Power Delivery.
2008;1(23):73‐78. https://doi.org/10.1109/TPWRD.2007.911154
10. Rao JG, Pradhan AK. Differential power‐based symmetrical fault detection during power swing. IEEE Trans Power Delivery.
2012;3(27):1557‐1564. https://doi.org/10.1109/TPWRD.2012.2196527
11. Mahamedi B, Zhu JG. A novel approach to detect symmetrical faults occurring during power swings by using frequency components of
instantaneous three‐phase active power. IEEE Trans Power Delivery. 2012;3(27):1368‐1376. https://doi.org/10.1109/TPWRD.2012.2200265
12. Nayak PK, Pradhan AK, Bajpai P. A fault detection technique for the series compensated line during power swing. IEEE Trans Power
Delivery. 2013;2(28):714‐722. https://doi.org/10.1109/TPWRD.2012.2231886
M. S. ET AL. 15 of 15

13. Khodaparast J, Khederzadeh M. Three‐phase fault detection during power swing by transient monitor. IEEE Trans Power Delivery.
2015;5(30):2558‐2565. https://doi.org/10.1109/TPWRS.2014.2365511
14. Dubey R, Samantaray S, Panigrahi BK, Venkoparao VG. Phase‐space‐based symmetrical fault detection during power swing. IET Gener
Transm Distrib. 2016;10(8):1947‐1956. https://doi.org/10.1049/iet‐gtd.2015.1245
15. Brahma SM. Distance relay with out‐of‐step blocking function using wavelet transform. IEEE Trans Power Delivery. 2007;3(22):1360‐1366.
https://doi.org/10.1109/TPWRD.2006.886773
16. Pang C, Kezunovic M. Fast distance relay scheme for detecting symmetrical fault during power swing. IEEE Trans Power Delivery.
2010;4(25):2205‐2212. https://doi.org/10.1109/TPWRD.2010.2050341
17. Dubey R, Samantaray S. Wavelet singular entropy based symmetrical fault‐detection and out‐of‐step protection during power swing. IET
Gener. Transm. Distrib. 2013;10(7):1123‐1134. https://doi.org/10.1049/iet‐gtd.2012.0528
18. Zadeh HK, Li Z. A novel power swing blocking scheme using adaptive neuro‐fuzzy inference system. Electr Pow Syst Res.
2008;7(78):1138‐1146. https://doi.org/10.1016/j.epsr.2007.09.007
19. Seethalekshmi K, Singh SN, Srivastava SC. Svm based power swing identification scheme for distance relays. IEEE PES General Meeting.
2010;1‐8. https://doi.org/10.1109/PES.2010.5588164
20. Swetapadma A, Yadav A. Data‐mining‐based fault during power swing identification in power transmission system. IET Sci Meas
Technol. 2016;2(10):130‐139. https://doi.org/10.1049/iet‐smt.2015.0169
21. Moravej Z, Ashkezari JD, Pazoki M. An effective combined method for symmetrical faults identification during power swing. Int J Electr
Power Energy Syst. 2015;64:24‐34. https://doi.org/10.1016/j.ijepes.2014.07.039
22. Kumar J, Jena P. Solution to fault detection during power swing using teager‐kaiser energy operator. Arab J Sci Eng.
2017;42(12):5003‐5013. https://doi.org/10.1007/s13369‐017‐2538‐7
23. Sachdev M, Nagpal M. A recursive least error squares algorithm for power system relaying and measurement applications. IEEE Trans
Power Delivery. 1991;3(6):1008‐1015. https://doi.org/10.1109/61.85841
24. Ananthapadmanabha TV, Yegnanarayana B. Epoch extraction from linear prediction residual for identification of closed glottis interval.
IEEE Trans Acoust Speech Signal Process. 1979;27(4):309‐319. https://doi.org/10.1109/TASSP.1979.1163267
25. Cohen L. Time‐frequency Analysis: Theory and Applications. Upper Saddle River. NJ. USA: Prentice‐Hall; 1995.
26. Murthy KSR, Yegnanarayana B. Epoch extraction from speech signals. IEEE Trans Audio Speech Lang Process. 2008;16(8):1602‐1613.
https://doi.org/10.1109/TASL.2008.2004526
27. Murty KSR, Yegnanarayana B, Joseph MA. Characterization of glottal activity from speech signals. IEEE Signal Process Lett.
2009;6(16):469‐472. https://doi.org/10.1109/LSP.2009.2016829
28. Prasanna SRM, Pradhan G. Significance of vowel‐like regions for speaker verification under degraded condition. IEEE Trans Audio Speech
Lang Process. 2011;8(19):2552‐2565. https://doi.org/10.1109/TASL.2011.2155061
29. Pradhan G, Prasanna SRM. Speaker verification by vowel and nonvowel like segmentation. IEEE Trans Audio Speech Lang Process.
2013;21(4):854‐867. https://doi.org/10.1109/TASL.2013.2238529
30. Nayak PK, Pradhan AK, Bajpai P. Secured zone 3 protection during stressed condition. IEEE Trans Power Delivery. 2015;1(30):89‐96.
https://doi.org/10.1109/TPWRD.2014.2348992

How to cite this article: M. S. P, Nayak PK, Pradhan G. Detection of three‐phase fault during power swing
using zero frequency filtering. Int Trans Electr Energ Syst. 2019;29:e2700. https://doi.org/10.1002/etep.2700