Smart Discrete Fourier Transform Algorithm’s Comparison with Available

Filtering Algorithms for Distance Relay for Transmission Line Protection

*P. Suresh Babu **Dr. S. V. Jayaram Kumar

*Associate Professor in EEE Dept., St. Ann‟s College of Engineering & Technology CHIRALA-523187 , INDIA E-mail: sureshperli@[Link]
**Professor in EEE Dept., JNTU College of Engineering, Kukatpally, Hyderabad-500082, INDIA. E-mail: svjkumar101@[Link]
.
Abstract: Of all available relays, distance relay attracted a fundamental frequency between the relay location and the
more attention for transmission line protection. Whenever fault point thus determining if a fault is internal or external to
fault occurs in power system the actuating quantities contains a protection zone. Voltage and current signal are used for this
harmonics, decaying dc components and noise. Faults have to purpose and they generally contain the fundamental
be cleared rapidly in order to prevent damage to the operating frequency component in addition to harmonics and the dc
devices and personnel. So many filtering algorithms are offset. With digital technology being ever increasingly
available for the protection of transmission line but all are adopted in power substation more particularly in the
having their own merits and demerits. To obtain exact protection field, distance relays have experienced some
fundamental frequency components for relaying purpose, improvement mainly related to efficient filtering method.
Smart Discrete Fourier Transform (SDFT) algorithm is During last two decades remarkable work has been
proposed. On 17-bus power system, proposed algorithm is demonstrated in the area of distance protection. Many
tested for speed, accuracy, frequency response, filtering algorithms have been proposed numerical relaying
computational burden and capability to distinguish among actions. Every algorithm has its own merits and demerits
different types of faults. The results are compared with
available filtering algorithms. The power system testing is The aim of most of these filtering algorithms is to
done with MATLAB and PSCAD/EMTDC environment. The extract the fundamental frequency component from the
test results shows that proposed SDFT algorithm is better that complex post fault voltage and current signals containing a
existing relaying algorithms. transient dc offset component and harmonic frequency
components in addition to the power frequency component.
Key words: Filtering Algorithms, Faults, Distance relay, The exponentially decaying dc offset present in the relaying
Decaying Dc offset, Harmonics, Power system Protection. signal gives rise to large errors in the pharos estimates unless
the offset terms are removed prior to the execution of the
I. INTRODUCTION algorithms.

According to the historical records, many large scale This paper presents a novel Smart Discrete Fourier
system-wide blackouts involve relay misoperations. Transform (SDFT) algorithm with capability to estimate
Evaluation and improvement of existing relay algorithms and exact fundamental frequency components during faults, that
settings as well as investigation of new techniques for operating signals contain fundamental frequency components
relaying are very important for understanding and mitigating in combination with decaying dc components and harmonics.
relay misoperations. The most common approach used by Proposed SDFT algorithm performance is compared in terms
many researchers for studying relay algorithm performance is of distance relay with existing filtering algorithms regarding
using a simple two-machine system and limited fault their speed, accuracy, computational burden and frequency
scenarios. An algorithm for comprehensive study of different response.
relaying actions and fault analysis under variety of system-
wide disturbances is needed. A 17-bus power system is taken to test the effectiveness
of proposed SDFT algorithm. Different types of faults like
Numeric transmission line distance protection systems single line to ground, double line to ground and three lines to
have been widely applied in recent years primarily because of ground faults are simulated at different lengths of selected
their monitoring and communications capabilities rather than transmission line. The distance relay characteristics are
for improved performance of the protection functions. generated using MATLAB and PSCAD/EMTDC output files
Typical tripping times for digital distance relays range from containing the values of the apparent impedances seen by
one to 3 cycles, with state of art filtering algorithm it can distance relay placed at bus-3, whose fundamental frequency
offer trip times of one-quarter to one cycle. Recent components extraction is based on proposed SDFT algorithm
developments in adaptive algorithms and the use of higher for faults at selected locations on transmission line L5.
sampling rates combine to provide secure high speed
protection not available with previous implementations. The II. PROPOSED SDFT FILTERING ALGORITHM
fast and accurate determination of fault location on electric
power transmission line is utilized as an aid in the fault The voltage and current signals may contain serious
analysis and power restoration. At the same time, the fault harmonics and decaying dc components during fault interval.
destination adversely impacts service reliability, operation The decaying DC seriously decreases the precision and
cost and the quality of power delivery. convergence speed of fundamental frequency signal from
DFT. In order to overcome the above problems, the proposed
Distance relaying techniques have attracted digital multifunction relay with SDFT algorithm can estimate
considerable attention for the protection of transmission line. the DC offset frequency and phasor from a faulted input
The principle of these techniques measures the impedance at
operating signals. Since there are several components in a N
i sin 

 e 
N 1 ----8
fault current signal, the algorithm first takes DC offset into j
 2 e j ( N 1) 2
consideration and uses smoothing windows to eliminate other i 0

sin
components in a fault signal. 2

Consider any fault signal f(t) with fundamental We can rearrange the Equn. (7) as
frequency components and decaying DC offset components
 f  r
can be expressed as f j 2 1  sin N1 j ( N 1)1
t fˆr  e  50  N e
f (t )  F sin(t  1 )  F sin(2 )e ----1 N sin 1
Where  f  r
f *  j 2  1 50  N sin N 2 j ( N 1) 2
F is the amplitude of the faulted signal  e e
N sin  2
1 is the phase angle of the faulted signal 

r
2 F sin(2 ) 1
 2 is the fault angle of the signal e 50  ----9
  2
e 50 N
N   j
 1   is the time constant of the signal e 50 N N
1
Where
f
Suppose f (t ) is sampled with a rate of sampling (50*N) Hz f and  (2  )
to produce the sample set { f (k )} 1  2   60
50N N
 k   k 
f (k )  F sin    1   F sin(2 ) exp     ----2
 50 N   50 N By rearranging Equn. (9) we can get

The signal f(t) is conventionally represented by phasor f sin N1 j 50 N ( f ( 2 r  N 1)100r )
fˆr
( 50 f )
 e
complex number f N sin 1
f  Fe j1  F cos 1  jF sin 1 ----3 
f * sin N 2 j 50 N ( f ( 2 r  N 1)100( r  N 1))
 e
N sin  2
Then f(t) can be expressed as 
 r
2 F sin(2 ) e 50
1 
----10
fe j1  f *e  j1  k    2
e 50 N

f (t )   F sin(2 ) exp     ----4 N   j

2  50 N  e 50 N N
1
Let assign

Fundamental frequency components of Discrete Fourier f sin N1 j 50 N ( f ( 2r  N 1)100r ) =Ar ----11
e
transform of {f(k)} is calculated from the following equation N sin 1
N 1
2  j 2k 
fˆr 
N
 f ( k  r )e N ----5

f * sin N 2 j 50 N ( f ( 2 r  N 1)100( r  N 1)) = B
e r ----12
N sin  2
k 0


Taking frequency deviation   2 (50  f ) into 2 F sin( 2 ) e

50
1 
r
e 50 N = Cr ----13
consideration N 

 j
2
e 50 N N
1
N 1 (k r ) k
f j 2 ( 50  f )  j 2
fˆr 
N
e
k 0
50 N
e N
Equn. (10) can be re written as
(k r )
f * N 1 j 2 ( 50 f ) 50 N  j 2 N
k
 e
N k 0
e fˆr  Ar  Br  Cr ----14

(k r )
2 F sin(2 ) N 1 
k
 j 2 So far the development of the algorithm of SDFT is the

N

k 0
e 50 N
e N ----6
same as the traditional DFT method. So the SDFT can keep
all advantages of DFT such as recursive and half-cycle
We can rearrange the Equn (6) as computing manner. But in the DFT, it doesn't take DC offset
into consideration and it assumes that the frequency deviation
 f  r  f  r
f j 2 1  N 1 j 2   is small enough to be ignored. It always considers fˆr  Ar , so
fˆr  e  50  N  e  50  N
N k 0 traditional DFT based methods incur error in estimating
 f  r  2 f  r
f *  j 2 1  N 1  j 2  
 e  50  N  e  50  N
frequency and phasor when frequency deviates from nominal
N k 0 frequency (50 Hz) or DC offset is present. If we want to
2 F sin(2 )  50 N N 1  50 N
r k k
 j 2 obtain exact solution, we must take B, and C, into

N
e e
k 0
e N ----7
consideration. Then we define

j( ( 2 f 100 ))
Above Equn. (7) can be solved by the following identity ae 50 N
----15
  a 2 fˆr 1  a( fˆr  fˆr 2 )  fˆr 1
be 50 N ----16 Cr  ----35
From Equn (10) following relations are obtained (a  b)(ab  1)
 2
Ar 1  Ar .a ----17   j 
1  Cr N e 1 
50 N N
2  sin  ----36
Br 1  Br .a 1
----18  
 2 X
e 50  1 


Cr 1  Cr .b ----19 
Furthermore, we take noise into consideration and use
Then
smoothing windows to filter noise. Consider a sampled set
fˆr 1  Ar 1  Br 1  Cr 1  Ar .a  Br .a 1  Cr .b ----20 {f(k) } to be a filtered set { z(k)} by a smoothing window
ˆf  A  B  C  A .a  B .a 1  C .b ----21
r 2 r 2 r 2 r 2 r 1 r 1 r 1 {SW( m) with window size „m‟.
s1 , s2 , s3 ......sm
m
Equn.(14) is multiplied both sides with „b‟ and subtract from z (k )   s i f (k  i  1) ----37
Equn (20) gives i 1
Moreover, the DFT of { z(k)} is given by
yˆ r  fˆr 1  fˆr b  Ar (a  b)  Br (a 1  b) ----22
2 N 1 j
2k

yˆ r 1  fˆr  2  fˆr 1b  Ar 1 (a  b)  Br 1 (a 1  b) ----23 zˆr 

N
 z ( k  r )e
k 0
N

yˆ r 2  fˆr 3  fˆr 2b  Ar 2 (a  b)  Br 2 (a 1  b) ----24 N 1

 m
 j
2k
= 2   s i f ( k  r  i  1) e
N

We can rearrange Eq.(22), (23) and (24) as

N k 0 i 1 
2k
m 2 N 1 j 
yˆ r 1a  yˆ r  Ar (a  1)( a  b)
2
----25
=  si   f (k  r  i  1)e N

i 1 N k 0 
yˆ r  2 a  yˆ r 1  Ar 1 (a 2  1)( a  b) ----26 m
Equn (25)/ equn (26) gives = s i fˆr i 1 ----38
A yˆ a  yˆ r 1 yˆ r 3 a  yˆ r  2 i 1
a  r 1  r  2  ----27 From the definition of Equn.(14), we can obtain:
Ar yˆ r 1a  yˆ r yˆ r  2 a  yˆ r 1
Put Equn.(23) & Equn.(24) in Equn.(27) zˆr  Ar ( s1  s2 a  ......  sm a m 1 )

[ fˆr 2 ( fˆr  fˆr 2 )  fˆr 1 ( fˆr 1  fˆr 3 )]b2 +  Br ( s1  s2 a 1  ......  sm a  ( m 1) )

[ fˆr3 ( fˆr  fˆr2 )  fˆr1 ( fˆr2  fˆr4 )]b +  Cr ( s1  s2b  ......  smb ( m 1) ) ----39

[ fˆ ( fˆ  fˆ )  fˆ ( fˆ  fˆ )] =0
r 3 r 1 r 3 r 2 r 2 r 4 ----28
The relations of Equn. (17), Equn. (18) and Equn.(19) are
Solve Equn.(28) to obtain 'b'. From the definition of 'b' in still kept in Equn.(39). Therefore, the same steps from
Equn (16) we can obtain the exact solution of the time Equn.(20) to Equn.(33) can be used in Equn.(39). Hence we
constant. can estimate time constant and frequency without modifying
1 ----29
 equations, but we have to do some change in Equn.(32) and
50 N log b
Equn.(35) when we estimate phasor and fault angle.
Equn.(27) can be rearranged as
yˆ r 1a 2  ( yˆ r  yˆ r  2 )a  yˆ r 1  0 ----30 yˆ r 1a  yˆ r
Ar  ----40
Solve Equn.(30) to obtain 'a'. From the definition of 'a' in (a 2  1)( a  b)( s1  s2 a  ......  sm a m 1 )
Equn. (15) we can get the exact solution of the frequency.
Where
50 N
f  50  f  cos1{Re(a)} ----31 yˆ r  zˆr 1  zˆr b
2
From Equn.(29) and Equn.(3I), it is observed that SDFT can yˆ r 1  zˆr  2  zˆr 1b
provide exact time constant and frequency a 2 fˆr 1  a( fˆr  fˆr  2 )  fˆr 1
Cr  ----41
using fˆr , fˆr 1 , fˆr  2 , fˆr 3 and fˆr  4 in the absence of noise. (ab  1)(a  b)(s1  s 2 b  ...... s m b m1 )
Moreover, we can estimate phasor and fault angle after
getting exact time constant and frequency by the following The phasor obtained from Equn. (40) and fault angle
equations: obtained from Equn.(41) will allay the phase shift and
yˆ a  yˆ r
Ar  2 r 1 ----32 amplitude decay caused by smoothing windows.
(a  1)( a  b)
 N sin 1  ----33 III. FILTERING ALGORITHMS AVAILABLE
F  abs Ar 
 sin( N1 ) 
1  angle( Ar e  j ( N 1) ) 1
----34 i) Infinite Impulse Response Algorithm
In traditional method, transfer of analog filter to digital
filter is practiced to get the desired requirements.
Transformation method used is impulse variance that uses s- As the matrix A has known elements, matrix B can be found
plane conversion to z-plane of differential equations to easily. Matrix b can then be used to compute the vector X
difference equations and then to direct synthesis. from the sampled signal. The Fourier sine and cosine
Analog prototype low pass filter is given as components of fundamental frequency can be obtained from
16
1
H N ( s)  2 ----42 C1  P1 cos(1t n )   B (3, n) x(t n ) ----51
s  2s  1 n 1
16
Analog prototype can be converted into digital by bilinear
S1  P1 sin(1t n )   B(4, n) x(t n ) ----52
transformation, it can be expressed as n 1
2 1  Z 1 ----43 iii) Kalman Filtering Algorithm:
s
T 1  Z 1
The Kalman filtering algorithm has been proven to be the
T is sampling time optimal linear estimator even in noisy environment once a
The bilinear transformed equation that is „s‟ replaced by right signal is represented by a state variable equation of the form
hand term of equation results in X K 1  FX K  WK ----53
1

 4 2 2    2  1  Z K  HX K  VK
H N ( s)  (1  2Z  Z ) 2 
1 2
 1   2 Z 

 T T   T   The covariance matrices for Wk and Vk vectors are given as
1 follows.
 4 2 2  2 
 (1  2Z 1  Z 2 ) 2   1Z 
----44 E[WK Wi T ] = Q ik
 T T  
=0 ik
The fundamental component present in the signal x(t) will be
E[V K T
Vi ] = R ik
extracted by designing band pass filter at 50Hz frequency can
be obtained from below expression =0 ik
Y50 ( N )  0.523 X ( N  1)  0.365 X ( N  2)  0.734 X ( N  3)
E[WK Vi T ] = 0 for all k and I ----54

 0.649 X ( N  4)  0.632Y ( N  1)  0.643Y ( N  2)  Where E denotes the expected values.

Having a prior knowledge of the initial estimation error
0.879Y ( N  3)  0.369Y ( N  4) ----45 covariance matrix Po-, the Kalman gains can be computed
recursively as follows.
ii) Least Square Curve Fitting Algorithm: Kk  Rk H T (H Rk H T  R)1
Rk  ( I  K k H ) Pk
Here it is assumed that the inrush current contains decaying Rk 1  FRk F T  Q ----55
DC and no more than five harmonics, then in a certain time
Where
interval. The inrush signal can be represented as Kk is the Kalman gain matrix at time tk.
t
x(t )  P0 e t   Pk sin(k0t   k ) ----46 Pk is the estimation error covariance matrix at tk;
k 1 Pk, the error covariance matrix for the updated estimate at tk
Where x(t) is instantaneous differential signal sampled at a I is the identity matrix.
time t
Having an initial state estimate X0- , the Kalman filter
P0 is decaying Dc component.
equation, which recursively estimate new values of the state
 is inverse time decay time constant of Dc component vector, is as follows.
  
Pk is peak component of the kth harmonic differential signal X K  X K  KK (Z K  HX K )
 
 0 is fundamental frequency X K 1  FX K ----56

 k is Phase angle of kth harmonic Where X K is the estimate of Xk.
5 The discrete time state space representation of periodic signal
x(t )  P0  P0 t   Pk cos k sin(k0t ) having harmonic components on to nth order with samples Zk
k 1
5
at time tk can be given.
  Pk sin  k cos(k0t ) ----47 X K 1  FX K
k 1
Z K  HX K ----57
In matrix form the above equation can be written as
 f (1 )
X   AY  ----48  f (2 )
0

Least square components obtained from F  
 f (3 ) 
Y ( AT A) 1 AT   BX  ----49 
 0

1
It can be shown that  =wT, w is the fundamental supply frequency in rad/s and
B  ( A T 1
A) A T
----50 is the sampling interval, s and
H = [1, 0, 1, 0 .................1, 0,1]
iv) Block Pulse Functions Algorithm: translation respectively, the scale and translation in terms of
A set of block pulse functions on a unit time interval these parameters will be a  a0 m and b  nb0 a0 m
(0,1) with N number of samples per cycles defined as After discretization in terms of the parameter ao, bo, m and n
n (t )  1 for
(n  1) t n ----58 the mother wavelet can be written as,
N N    (t  nb a m ) 
both for n=1,2,3,…..,N g ' (m, n, t )   1 g
  ----63
a m 
0 0
= 0 otherwise  a0  
m
If there is a function f(t), which is integrable in (0,1) can be 
approximated using BPF as,
m / 2 m
g ' (m, n, t )  a0  nb0 )
N
f (t )   ann (t )
g (ta0 ----64
----59
n 1 After discretization, the wavelet domain coefficients are no
where the coefficient an are block pulse function coefficients longer represented by a simple a and b instead they are
determined so that the integral square error is minimized. represented in terms of m and n. The discrete wavelet
1 N coefficient Wgf (m, n) are given by
----60
 
  f (t )  a  (t ) 2 dt
n n


 f (t ) g (ta
m
Wgf (m, n)  a0  nb0 )dt ----65
m/ 2
0 n 1
0
For such a square fit an is given by 
n
N The transformation is over continuous time but the wavelets
an  N 
( n 1)
f (t )dt ----61 are represented in a discrete fashion like the continuous
wavelet transformation these discrete wavelet coefficient
N
represent the correlation between the original signal and
= average value of f (t) in the interval (n-1)N<t<n/N wavelets for different combination of m and n.

v) Wavelet Transforms Algorithm:

The wavelet transform translate the time domain DISTANCE RELAY
function into a representation localized not only in frequency
but also in time. Wavelet theory is the mathematics In general thirteen input signals, namely, three line-to-
associated with building a model for a non-stationary signal, ground voltages, three line-to-line voltages, three line
with a set of components that are small wave called wavelets. currents Three differences in line currents and residual
There are some conditions that must be met for a function to current are required to obtain phasor quantities necessary for
qualify as a wavelet. They must be oscillatory and have an impedance relay. In this work all the 13 signals are
amplitudes that quickly decay to zero. The product of an obtained from simultaneously taken samples of 6 signals,
oscillatory function with a decay function yields the wavelet. namely, three line-to-ground voltages and three line currents
A number of different wavelets are used to approximate any using following relations
given function with each wavelet generated from one original
VL1 L 2  VL1  VL 2 ----66
wavelet called a mother wavelet. The new elements, called
daughter wavelet are nothing but scaled and translated
mother wavelets. Scaling implies that the mother wavelet is iL1 L 2  iL1  iL 2 ----67
either dilated or compressed and translation implies shifting
of the mother wavelet in the time domain. iR  iL1  iL 2  iL 3 ----68

Let a equation for a mother wavelets follows. The ground impedance seen at the relay point is calculated as
t 2 jt
g (t )  e e Z Li 
VL1 ----69
K 1
I L1 
g (a, b, t )   1  g  (t  b) 
1 ----62 IR
3
 a  a
VL1, IL1 are the RMS values of the relay voltage and current
Where IR is the RMS value of the residual current
a is a constant; K is the degree of compensation, being a ratio of zero to
b is the time translation factor positive sequence impedance of the line that remains constant
a is the scaling factor. for all fault locations within the protected line.
The energy of the scaled daughter wavelet is normalized to The phase impedances at the relay point are calculated as
keep the energy same, as the energy in the mother wavelet.
VL1 L 2
Z L1 L 2  ----70
For computer implementation, the discrete wavelet I L1 L 2
transform is used. A discrete wavelet transform results in a The Function logic supports two commonly used operating
finite number of wavelet coefficients depending upon the
Circular and Quadrilateral characteristics of conventional
integer number of the discretization step in scale and
translation denoted by m and n, respectively. So any wavelet impedance relays. The Function Logic implements a three
coefficient can be described by two integer, m and n. If ao stepped distance protection by accepting three such
and bo are the segmentation step sizes for the scale and characteristics, one for each zone.
 TEST SYSTEM AND RESULTS relay at 10% of line L5. The corresponding characteristics are
as shown below.
Proposed filtering algorithm was tested with 17 bus system as
shown in fig.1

Single line-to-ground faults at 10% of Line L5 from Bus 3

Fig.1 Test System Data

Generator G1 is rated at 2GW and generator G2 is rated at

1GW, both are at 15KV, 0.866 power factor lagging with
transient reactances of 0.14p.u. Load-1 rated at
462300 MVA and Load 2 is rated at 577300 MVA.

Transformers Data

Voltage ratio
Transformer MVA rating
KV/KV
T1 2500
15/500
T2 1800
T3 500
500/15
T4 600
T5 600 500/230
T6 1000
T7 1000
500/230 Double line-to-ground faults at 10% of LineL5 from Bus 3
T8 1000
230/500
T9 1000

Each transformer has reactance of 0.1p.u. Each transmission

line has shunt conductance of 1.0e-10 mhos per meter. Load
connected at bus-14 is 200MW at 500KV, bus-15 is 150MW
at 230KV, bus-16 is 150MW at 500KV and bus-17 is
150MW at 500KV.

Zones and Reach adjustment of Distance relay

The adjustment of zone-1 was set to protect up to 80% of the

impedance of the protected line. Zone-2 was set to protect up
to 50% of the shortest impedance of the lines T6 and T13,
emanating from the remote bus. Zone-3 was set to protect up
to 100% of the remote line T6. Distance relay was placed at
bus-3 for the proposed simulation.

Case-1: Simulation of faults at 10% of Line L5 : Three-phase-to-ground fault at 10% of Line L5 from Bus 3

Different types of faults like single line-to-ground, double Case-2: Simulation of faults at 50% of line L5 :
line-to-ground and three-phase-to-ground faults were
simulated at 10% of transmission line L5 from bus 3. Fault Single line-to-ground, double line-to-ground and three-phase-
impedances from relay location are calculated by distance to-ground faults were simulated at 50% of transmission line
L5 from bus 3. Fault impedances from relay location are
calculated by distance relay at 50% of line L5. The relay‟s location and the fault point by using digitized samples
corresponding characteristics are as shown below. of voltage and current signals. Each algorithm has certain
merits and demerits regarding computational simplicity,
speed, accuracy and frequency response. The selection of any
particular algorithm depends on the protection requirement.
All the various schemes of algorithms proposed advocate a
family of wide spectrum of techniques to determine the
impedance of the transmission line under fault condition.
Different fault types were simulated in two different
locations of transmission line L5. The distance relay‟s
correctness in fault identification, the time of the event they
were monitoring and exhibition of required discrimination in
their operation were compared between proposed Smart
Discrete Fourier Transform algorithm (SDFT) and above said
filtering algorithms. The distance relay calculated the
apparent impedance of the transmission line, and the plotted
apparent impedances in R-X plane supported the correct
operation of the relay with the proposed algorithms are
Single line-to-ground faults at 50% of Line L5 from bus-3 compared.

In the Kalman filtering algorithm uses only present sampled

signals and does not require any past data to be stored in the
memory. The major difference between the Kalman filter and
proposed SDFT algorithm is evaluation of filter gains. In
SDFT there is no need of evaluating gains but case of
Kalman filter requires it. The gains of Kalman filter vary
with time as the gains are non-stationary. Being recursive in
nature, even though Kalman filter is computationally more
efficient but it has limited capability for modeling the
decaying dc component. In addition, statistical properties of
the signal to be processed are needed for calculating the
Kalman gains. So it is concluded that proposed SDFT
algorithm is efficient and more acceptable.

One of the most prominent advantages of algorithms using

Blocked Pulse Function Algorithm coefficients is that the
sampling rate can be any positive integer, where as in other
Double line-to-ground faults at 50% of Line L5 from bus-3 method like in Haar and Walsh transform, one is compelled
to use sampling rates that are equal to integral powers of 2. In
other algorithms generally 16 samples/cycle are used to give
the satisfactory results. For distance protection however, the
BPF algorithm uses 12 samples, which requires less
multiplication, so less memory space and time but obtained
samples contains errors as it is not handling decaying dc
component properly. This drawback is completely eliminated
in proposed SDFT algorithm.

Least Square Curve Fitting Algorithm can also handles

decaying dc offset components to filter from operating
signals as it is in proposed SDFT algorithm but it is suffering
from drawback that it can only work for the harmonics of
order 5. Above which this algorithm outcomes suffers from
large errors. But there is no restriction on order of harmonics
in SDFT algorithm.
Three phase-to-ground fault at 50% of Line L5 from bus-3
Being recursive in nature, an Infinite Impulse Response
Algorithm is an accurate and error free algorithm but it needs
CONCLUSIONS to transform differential equations from s-plane to z-plane
frequently and feedback is desired. Probability of errors in
A number of algorithms for numerical distance protection transformation will result with less knowledge in z-
have been compared with proposed novel SDFT algorithm to transforms. But this feedback and interrelating output with
calculate the apparent impedance of the line between the input is not required in SDFT algorithm.
 [9].J. Z Yang and C. W. Liu, “A Smart Method Makes DFT More
Non-stationary signals, where the interest is only for what Precise for Power System Frequency Estimation:‟ IEEUPES Winter
spectral components exist in the signal but not interested Meetzng, Feb. 1999, pp. 909-913.
where these occur, can be easily handled with wavelet
[10].P Pillay and A Bhattacharjee. „Application of Wavelets to
transforms. It provides the time frequency representation. Model Short Term Power System Disturbances.‟ IEEE Transactions
Although the discretized Continuous Wavelet on Power Systems, vol 11, no 4, November 1996, p 2031.
Transform(CWT) enables the computation of required CWT
by the computer, but it is not a true discrete transform. As a [11]. A A Girgis and D G Hart. „Implementation of Kalman and
matter of fact, the wavelet series is simply a sampled version Adaptive Kalman Filtering Algorithm for Digital Distance
of the CWT and the information it provides is highly Protection on a Vector Signal Processor.‟ IEEE Transactions on
redundant for the signal concerned. This redundancy, on the Power Delivery, vol 4, no 1, January 1989, p 141.
other hand, requires a significant large amount of
computation time and resources. Time of fault clearance [12].J L Pinto de Sa. „A New Kalman Filter Approach to Digital
Relaying.‟ IEEE Transactions on Power Delivery, vol 7, no 3, July
based on fault amplitude computation is more important than 1992, p 1652.
the instant where the fault has occurred. Computation time of
Wavelet is more than proposed SDFT. More than that, in [13]. Norman T. Stringer, „The Effect of DC Offset on Current
some limit, SDFT can handle non-stationary signals as Operated relays,” IEEE Trans. on Industry Applications, Vol. 34,
wavelets. No. 1. JanuarylFebruaty 1998, pp. 30-34.

The discrete Fourier transform algorithm was used to extract [14]. A H Osman and O P Malik. „Transmission Line Distance
real and imaginary components that involves number of Protection based on Wavelet Transform.‟ IEEE Transactions on
multiplication operations, in addition to summation Power Delivery, vol 19, no 2, April 2004, p 515.
/subtraction operations, it is less time consuming algorithm. [15.]Lai, T. M., L. A. Snider, E. Lo, 2006. “Wavelet Transform
Based on the experimental results, proposed SDFT algorithm Based Relay Algorithm for the Detection of Stochastic High
is found to be the best algorithms in terms of accuracy and Impedance Faults”. Electric Power Systems Research, vol. 76, no. 8,
computing time when compared some of existing filtering pp. 626-633.
algorithm. [16] Sheng, Y. and S. M. Rovnyak, 2004. “Decision Tree-Based
Methodology for High Impedance Fault Detection”, IEEE
REFERENCES Transactions on Power Delivery, vol. 19, no. 2, pp 533-536.

[17] Huang, S. and C. Hsieh, 1999. “High-impedance Fault

[1]. M S Sachdev and M Nagpal. „A Recursive Least Error Squares
Detection Utilizing A Morlet Wavelet Transform Approach,” IEEE
Algorithm for Power System Relaying and Measurement
Transactions on Power Delivery, vol. 14, no. 4, pp. 1401-1410.
Application.‟ IEEE Transactions on Power Delivery, vol 6, no 3,
1991, p 1008.
[18] Kim, C. H., R. Aggarwal, 2001. “Wavelet Transforms in Power
Systems,” Power Engineering Journal vol. 15, no. 4, pp. 193-202.
[2].F Liang and B Jeyasurya. „Transmission Line Distance
Protection Using Wavelet Transform Algorithm.‟ IEEE
Transactions on Power Delivery, vol 19, no 2, April 2004, p 545. [19].Akorede, M. F., H. Hizam, 2009. “Wavelet Transforms:
Practical Applications in Power Systems”,Journal of Electrical &
[3].C Fernandez and F Luis Pagola. „Total Least Squares and Technology, vol. 4, no. 2, pp 168-174.
Discrete Time Models in H V Distance Protection.‟ IEEE
Transactions on Power Delivery, vol 14, no 1, 1999, p 74. [20] D.V., Coury, H.G.F., Brito: “Digital filters applied to
computerrelaying”, Proc. IEEE International Conference on Power
[4].A S Alfuhaid and M A EI-Sayed. „A Recursive Least Squares System Technology POWERCOM ’98, New York, NY, USA, pp.
Digital Distance Relaying Algorithm.‟ IEEE Transactions on Power 1062-66, 1998.
Delivery, vol 14, no 4, October 1999, p 1257.

[5].D C Robertson, O I Camps, J S Mayer and W B Gish. „Wavelets

and Electromagnetic Power System Transients.‟ IEEE Transactions
on Power Delivery, vol 11, no 2, April 1996, p 1050.

[6].A Girgis and R G Brown. „Application of Kalman Filtering in

Computer Relaying.‟ IEEE Transactions on PAS, vol 100, no 7,
1981, p 3367.

[7].J Hector, F Altuve, V Ismael Diaz and M Ernesto Vazquez.

„Fourier and Walsh Digital Filtering Algorithms for Distance
Protection.‟ IEEE Transactions on Power Systems, vol 11, no 1,
February 1996, p 457.

[8]Dario D‟Amore and Alessandro Ferrero. „A Simplified

Algorithm for Digital Distance Protection Based on Fourier
Techniques.‟ IEEE Transactions on Power Delivery, vol 4, no 1,
January 1989, p 157.