534 IEEE Transactions on Power Delivery, Vol. 3, No.

2, April 1988

A STATE-OF-THE-ART REVIEW OF TRANSFORMER PROTECTION ALGORITHMS

M.A. Rahman, Senior Member, lEEE B. Jeyasurya, Member, IEEE

Faculty of Engineering and Applied Science

Memorial University of Newfoundland
St. John's, Newfoundland, Canada

ABSTRACT Various claims have been made on the speed,

accuracy, computational burden, etc. of these
This paper presents a comparative study of algorithms. It seems quite difficult to grasp the
algorithms for digital differential protection of real significance of these algorithms since they are
power transformers. The mathematical basis for each evaluated using different models. The purpose of this
algorithm is briefly described. The algorithms are paper is to provide a state of the art review of most
compared as to their speed of response, computational of the algorithms for digital protection of power
burden and the capability to distinguish between an transformer. The algorithms are evaluated on a common
inrush and transformer internal fault. This basis. The paper also identi€ies the algorithms which
theoretical comparison of algorithms for transformer may be implemented in a practice1 design.
digital protection is presented in a comprehensive
form for the first time.
PRINCIPLE OF TRANSFORMER PROTECTION

INTRODUCTION T h e differential relaying principle [8] is

commonly used for the protection of power
For the last 1 5 years, there has been transformers. This is based on a comparison of the
considerable interest in the area of digital primary and secondary currents. These currents are in
protection of power apparatus [l]. In the early 1970s a known relationship under healthy conditions. When
dedicated digital relays using minicomputers were the currents deviate from the pre-defined relationship
proposed. Some utilities also tested minicomputer an internal fault is assumed and the transformer is
based experimental on-line systems for digital switched off. This method is quite satisfactory in
protection of transmission lines in their substations most respects but is subject to false tripping on the
[2,3]. The main features which have encouraged many transient magnetizing inrush current which flows when
researchers to investigate the feasibility of the transformer is energized. This current flows in
designing digital relays for power system protection one winding only of the transformer (the winding being
are its economy, reliability, flexibility, improved energized) and therefore appears to the protection as
performance over conventional relays and the an internal fault condition. The most common
possibility of integrating a digital relay into the technique used for preventing false trips during
hierarchial computer system within the substation. energization is the use of a harmonic restraint relay
The introduction of microprocessors has brought about [9]. The inrush current differs from an internal
novel and low-cost possibilities for the development fault current in so far as its waveform comprises a
of protection devices for power systems and power high percentage of harmonics. Of these, the second
apparatus. The capability of the currently available harmonic has particular prominence under all switching
microprocessors is such that all the digital relaying conditions. This fact is used in the design of a
functions performed by the minicomputers of the 1970s harmonic restraint relay. If the second harmonic
can now be done by microprocessor systems. As a content of the differential current exceeds a
result, many investigators published results of work pre-defined percentage of the fundamental, inrush is
on specific hardware and software techniques for assumed and the protection is prevented from
microprocessor based transmission line relays. A few tripping.
prototype distance relays using multiple
microprocessors have also been tested on-line [ 4 , 5 ] . The published literature on the digital
protection of power transformers has received
As the utilities gain experience with the use of attention as far as the development of algorithms for
these prototype digital relays, one can foresee that differential protection is concerned. These
low cost, dedicated microprocessor based relays will algorithms sample the differential current at regular
appear in commercial service during the next few time intervals and determine its fundamental and
years. Considerable attention has not been given to harmonic content. The decision as to trip or restrain
the o n - l i n e implementation of digital power is then based on the relative levels of the
transformer protection. Only during the last few fundamental and second harmonic content. Sykes and
years, researchers have been investigating Morrison [LO] suggested the use of recursive band pass
microprocessor based three phase transformer relays filters to calculate the fundamental and second
[6,7]. The technical literature available on this harmonic content. Malik, Dash and Hope [ll] proposed
subject is mainly on the algorithms for power the application of a cross correlation approach based
transformer protection [10-18]. on F,ourier techniques. Schweitzer et a1 [12] proposed
finite impulse response (FIR) filters to calculate the
magnitude of the fundamental and second harmonic
87 WM 114-2 A paper recommended and approved components. Degens [13] and Rahman et al. [ 1 4 ]
by the IEEE Power System Relaying Committee of the presented algorithms based on the least-squares curve
IEEE Power Engineering Society €or presentation at fitting technique to find the second harmonic and
the IEEE/PES 1987 Winter Meeting, New Orleans, fundamental components of the differential current.
Loujsiana, February 1 - 6, 1987. Manuscript sub- Rahman and Dash [15] have presented an algorithm using
mitted August 18 1986; made available €or printing the rectangular transform technique, which involves a
November 12, 1986. few number of multiplications. In a recent paper [16]
Fakruddin et al. have investigated the application of
Haar functions for differential protection.

0885-8977/88/O4OO-0534$01.00@1988 IEEE
 535
A p p l i c a t i o n of Walsh f u n c t i o n s has a l s o been proposed
by Rahman e t a l . [17,18] f o r d i g i t a l p r o t e c t i o n of From t h e s a m p l e s g( t i ) , r e c t a n g u l a r c o e f f i c i e n t s can
transformers. The following s e c t i o n summarizes the be obtained as:
mathematical f o r m u l a t i o n of t h e commonly u s e d
algorithms f o r transformer d i f f e r e n t i a l p r o t e c t i o n .

ALGORITHMS FOR TRANSFORMER DIFFERENTIAL PROTECTION

I n e q u a t i o n (8), [ S K I i s t h e v e c t o r of F o u r i e r s i n e
F o u r i e r Analysis Approach [11L c o e f f i c i e n t s and [SUI is the v e c t o r of corresponding
rectangular coefficients. The F o u r i e r c o e f f i c i e n t s
Any waveform g ( t ) having a f i n i t e energy i n the can be c a l c u l a t e d from the r e c t a n g u l a r c o e f f i c i e n t s
i n t e r v a l (0,T) can be expressed i n the i n t e r v a l as a
as
F o u r i e r Series:

g ( t ) = Fo + JT F1 sin(-) + @ F2 cos(+) +
T
JT F3 sin(&)
T
+ 4 F4 COS(%)
T
+ ... (1)
For example, t h e matrix [A-'] takes the form:
with c o e f f i c i e n t s
1 0 -1 I3 0 -115

1 0 0 0 -113 . . -
F1 IT g ( t ) s i n ( 9 ) dt 1 0 0 0 . . .
[A-l] =
1 0 0 . . .
0 1 0 . . .

Using t h i s approach, t h e d i f f e r e n t i a l c u r r e n t is L * _ I
analyzed in terms of i t s F o u r i e r s e r i e s and the peak Using t h i s approach f o r transformer d i f f e r e n t i a l
of the fundamental and second harmonic c o n t e n t s a r e
given by:
p r o t e c t i o n , F1 ..
F4 can be obtained as:

AMP1 = (F12 + FZ2)'I2

AMP2 = (F32 + F42)1/2

F1...F4 a r e o b t a i n e d by c o r r e l a t i o n of s a m p l e d
d i f f e r e n t i a l c u r r e n t with s t o r e d samples of r e f e r e n c e
s i n e and cosine waves.
SIR, CIR, ... a r e the rectangular coefficients
Rectangular Transform Approach [ 1 5 1 o b t a i n e d f r o m s a m p l e d d i f f e r e n t i a l c u r r e n t g( t i ) ,
From t h e samples of the d i f f e r e n t i a l c u r r e n t using equation ( 8 ) and ( 9 ) . The fundamental and
g ( t i ) , F o u r i e r s i n e and cosine c o e f f i c i e n t s can a l s o second harmonic content of the d i f f e r e n t i a l c u r r e n t
can now be determined using equation (3).
be defined as:
Algorithm based on Walsh Functions [17,18]

Walsh f u n c t i o n s form a complete s e t of orthogonal

functions. These f u n c t i o n s which a r e denoted by
w a l ( k , t ) bear a number of resemblances t o s i n e and
c o s i n e f u n c t i o n s appearing as the squared up v e r s i o n s
of them. They take only t h e values fl and change s i g n
I f t h e s i n e and c o s i n e terms i n equation ( 4 ) and ( 5 ) only when t is a m u l t i p l e of a power of 112. I n terms
a r e replaced with the equivalent 'rectangular' of t h e s e f u n c t i o n s , t h e Walsh expansion of g ( t ) i n the
f u n c t i o n s [5] i n t e r v a l (0,T) is defined a s

s i n y t ( x ) = sgn ( s i n x) (6 1 g ( t ) = k L4 Wk wal(k, T
")

cos Yt(X) = sgn (cos x) (7 1 where

where
T
sgn(y) = y1 lyl f o r Y f 0
wk = 1 1 g ( t ) wal(k, -r>
T
dt
T O
= 0 fory=O
T h e W a l s h c o e f f i c i e n t s ( W k ) a r e o b t a i n e d by
then corresponding 'rectangular transform'
c o r r e l a t i n g the samples of d i f f e r e n t i a l c u r r e n t g( t i )
c o e f f i c i e n t s w i l l be denoted as SKR and t&.
w i t h r e f e r e n c e Walsh functions. The Walsh method is
convenient t o implement only f o r a number of samples
536

which is an i n t e g r a l power of two. The r e l a t i o n s h i p

b e t w e e n F o u r i e r and Walsh c o e f f i c i e n t s c a n b e
expressed in m a t r i x form as

W = AF (18)

w h e r e any e l e m e n t Akk of m a t r i x A is t h e r e s u l t of
s u b s t i t u t i n g t h e k t h s i n u s o i d t h a t appears in e q u a t i o n
(1) f o r g ( t ) in ( 1 6 ) . T h e F o u r i e r c o e f f i c i e n t s F1,
F 2 , F3 a n d F4 c a n b e d e t e r m i n e d f r o m t h e Walsh
c o e f f i c i e n t s as:

F1 = 0.9 W 1 - 0.373 W5 - -074 Wg (19)

F2 i
i 0.9 w2 + 0.373 w6 - -074 w10 (20)

F3 = 0.9 W3 - 0.373 W11 (21 1 As in t h e previous a l g o r i t h m s ,

t h e fundamental and
second harmonic c o n t e n t can then be determined.
F4 = 0.9 W4 + 0.373 W12 . (22 )
F i n i t e Impulse Response Approach [12]
T h e Walsh c o e f f i c i e n t s W1 ... W12 a r e o b t a i n e d by
In t h i s approach f o u r f i n i t e impulse response
a d d i t i o n and s u b t r a c t i o n of t h e s i g n a l samples only in (FIR) f i l t e r s are used; two each f o r the fundamental
a c c o r d a n c e w i t h t h e r e f e r e n c e s q u a r e waves. Once F1, and second-harmonic. The f i l t e r s cover one c y c l e of
F 2 , F3, F4 a r e a v a i l a b l e , t h e fundamental and second time and have v a l u e s of e i t h e r p l u s o r minus one a t
harmonic c o n t e n t of t h e d i f f e r e n t i a l c u r r e n t can be any i n s t a n t d u r i n g t h a t period. The impulse responses
e a s i l y obtained. are r e s t r i c t e d t o p l u s o r minus one so t h a t f i n d i n g
t h e responses of t h e f i l t e r s t o a r b i t r a r y i n p u t s is
Haar F u n c t i o n s Approach 1161 reduced t o adding t o g e t h e r samples of t h e i n p u t , w i t h
a p p r o p r i a t e sign changes as determined by t h e impulse
The Haar f u n c t i o n set forms a complete set of responses. Using t i m e d i s c r e t e d i f f e r e t i a l c u r r e n t
o r t h o g o n a l r e c t a n g u l a r f u n c t i o n s . Haar f u n c t i o n s have s a m p l e s , i,, s a m p l e d N t i m e s p e r c y c l e , t h e f i l t e r
t h r e e p o s s i b l e states 0, +A and -A where S is a o u t p u t s are c a l c u l a t e d as
f u n c t i o n of 4 2 . They may be expressed as HAR(n,t).

For a time i n t e r v a l 0 < t < 1, t h e f u n c t i o n s are g i v e n

as

HAR ( 0 , t ) = 1 for0 < t (1

HAR ( 1 , t ) = 1 0 6 t (112

-1 for 112 < t 1

0 elsewhere

p = 1, 2, ... n = 0, 1, 2, ... (2P-1) where k s t a n d s f o r t h e k t h sample period. The

fundamental and second harmonic c o n t e n t s are then
In terms of t h e s e f u n c t i o n s , t h e Haar expansion of g i v e n by e q u a t i o n (3).
g ( t ) i n t h e i n t e r v a l (0,T) is d e f i n e d as
Least-Squares Curve F i t t i n g Algorithm [13]

I n this approach, t h e i n r u s h c u r r e n t is assumed

t o contain t h e fundamental frequency component, a
where decaying DC component and u p t o 5 t h harmonic. In a
1 c e r t a i n time t h e i n r u s h c u r r e n t can be w r i t t e n as
CK = 1 g ( t ) . HAR(k,t) d t (24 5
t=O il(t) =s po exp(-At) + 2'
k=l
pk sin(k W 0 t 4- @k> (33)
The Haar c o e f f i c i e n t s (%) are o b t a i n e d by c o r r e l a t i n g
I f t h e time window is small compared w i t h t h e
t h e samples of d i f f e r e n t i a l c u r r e n t g ( t i ) with time c o n s t a n t l/A of t h e decaying DC, t h e DC component
r e f e r e n c e Haar f u n c t i o n s as given by e q u a t i o n (24). c a n be approximated by
T h e F o u r i e r c o e f f i c i e n t s F1 ..
F4 are then o b t a i n e d
as
E q u a t i o n (33) s i m p l i f i e s t o
 531
multiplications to yield the four unknowns of equation
5
il(t) = po - poht + E pk cosOk sinkwot (40). This method i s basically a process of digital
filtering. Every time a new sample is taken, the old
k-1
5 (35) ones are shifted and digital filtering according to
+ E pk sinek coskwot equation (40) is carried out again on the new Set Of
k=l samples.
If N samples of the differential current are FREQUENCY RESPONSE OF ALGORITHMS
considered at time t , t2, ...,
t , the sampling
process results i n the jollowing set 0% equations: All the above mentioned algorithms may be
considered as implementation of non recursive digital
filtering techniques. The objective of each algorithm
is to determine the peak of the fundamental and
harmonic content of the differential current as each
new sample is taken. The digital filter outputs
depend on the present and past input signal samples
only. Only the filter weighing factors are unique to
each algorithm. The frequency response plot of the
Fourier Algorithm, FIR algorithm and Curve fitting
algorithm are shown in Figures 1-3. Frequency

4s the vector of the unknown contain 12 elements, it

is necessary that N > 12 to solve the unknown. If
equation (36) is written as

[A] m X = I (37)

then the least-squares solution of equation (37) is

given by

X = [(At . A)-l . At] . I = B . I (38)

From equation (38) it follows that matrix B can be

calculated by

B = [At . A1-l . At (39) Figure 1 Frequency Response of Fourier

Algorithm
The elements that occur in matrix B can be calculated
off-line from equation ( 3 9 ) . Matrix B that results
contains known elements, which can be used to
calculate the vector of unknown X from the sampled
differential current. To determine the amount of
fundamental and second harmonic present in the
differential current, only the unknown pl cosQl, p1
sine1, p2 cosQ2 and p2 sine2 i n equation (36) are of
importance. This can be found by

Figure 2 Frequency Response of Finite Impulse

The peak of the fundamental and second harmonic can Response Algorithm
then be calculated as
response is a measure of the filtering characteristic
p1 = (p1*cos2~1+ p12sin201)1/2 (41)
of the algorithm. By transforming the algorithm into
the complex-z domain and substituting frequency for
the complex variable z , the frequency response plot is
p2 = + p22sin2Q2)1/2 obtained. This plot is symmetric about the Nyquist
frequency (one-half the sampling frequency) and
It has been found that a sampling frequency of 12 repeats after each integer multiple of the sampling
samples per cycle, data window (N) of 13 samples and frequency. The frequency response plots for
time reference tl = 0 needs the lowest number of Rectangular transform, Walsh functions and Haar
functions are very similar to that of the Fourier
algorithm since all these algorithms basically obtain
F1, P2, F3 and F4 of equation (1) by correlating the
538
between the total magnetising current in two phases.
:: Hence, the inrush phenomena, from a relay point of
view is the same with star-connected current
transformers measuring inrush currents into a delta
winding as it is with delta-connected current
transformers measuring inrush currents into a
star-connected winding.

In this simulation, the values of the inrush

currents in individual phases are taken as independent
inrushes [20]. The inrush current in each phase is
obtained using Specht's [ 2 1 ] formula. A s no mutual
effect is considered between the limbs in the
simulation, it is assumed that three single phase
transformers are electrically connected, but
magnetically isolated to achieve three-phase
representation.
r.- 1

Figure 3 Frequency Response of Curve Fitting

Algorithm
signal samples with reference waveforms. The
frequency response plots show that all these
algorithms are able to effectively extract components
of the fundamental and second harmonic frequencies.
Each algorithm may be considered as four digital
filtering computations, two each for fundamental ad
second harmonic. In the frequency response plots
shown in Figures 1-3, the frequency response of the
two filters for fundamental are combined into a single
response (AUP1) and similarly for the second harmonic
filter (AMPZ). The responses are plotted as a
function of normalized frequency ratio of F/Fao, where
Fa, is the base frequency. The fundamental amplitlude
AMP1 shown in the + marked curves has peak at
fundamental frequency. Similarly, AMP2 shown in the
13 marked curves, has its peak at second harmonic
frequency of 2.0 per unit. The AMP 2 response of
Finite Impulse algorithm has relatively higher minor
peak at 6th harmonic frequency as shown in Fig. 2.

TRANSFORMER MODEL FOR ALGORITHM EVALUATION

In the work on digital differential protection of

transformers proposed in the past, a single phase
transformer and its model were used for testing the
prototype relays. In this paper, a digital computer
model of a three phase transformer is considered for
fat00 Ib.00 2b.00 d.00
T I N E 1.81
rho0 rb. M d. 00

evaluating the algorithms. The amount of inrush

depends on the polarity of the voltage waveform,
switching angle and residual flux associated with the
respective cores. Earlier studies [19] on inrush in
3-phase transformers show that there may be an inrush
current into more than one phase, and these separate
inrush currents may be affected by the electric
connections and/or magnetic coupling between phases.
When a star-delta transformer is energized from the
star side, the inrush current in two phases aid the
third phase and as a result, the current in one phase
is oscillatory. This phenomena is called 'helping
effect'. In differential relaying it is almost
universal practice to connect the current transformers
in delta on the star-side of a star-delta bank. This
connection eliminates zero-sequence current from the
relay, and yields the correct phase angles to go with
star-connected current transformers on the delta side Figure 4 Inrush Current on Star Side of a
of the bank. Thus the current flowing to any one S tat9 elta Transformer
relay is the difference between the magnetizing
currents in two phases. The amount of helping effect Figure 4 shows inrush current of a star-delta
is of no consequence, since it has disappeared in the transformer energized from the star side. It is
delta relay currents. The current transformers on the assumed that the switching is done when the phase A
delta side of a star-delta-connected bank are usually voltage angle A is zero. Phase A is assumed to have a
connected in star when they are used to energize remnant flux equal to 90% of the peak steady-state
differential relaying. Thus the relays receive the flux and phase C is assumed to have a remnant f l u x
phase currents. The relay current is the difference equal to 50% of the peak steady-state flux. The core
 539
is assumed to saturate at a flux level equal to 140%
8 of the peak steady-state flux. From Figure 4 it can
be seen that the phase B current is oscillatory.
Figure 5 shows the corresponding relay currents which
are obtained by connecting the current transformers in
delta on the star side of the transformer.
Figure 6 shows phase A fault current for a
+phase internal fault. Before the fault occurs power
is delivered by the source to a load through the
transformer. The fault-current is seen to have a very
high offset.

SIMULATION RESULTS

A three phase transformer is simulated based on

the previously discussed model [ 2 0 ] using a VAX-11/785
mainframe computer. The simulations provide samples
of currents in each phase when the transformer is
energized or when an internal fault occurs. For
simulation purposes, three single phase transformers
are electrically connected, but magnetically isolated
to form the 3-phase representation. Data from the
simulations are used as input to the different
algorithms to identify the response of the algorithms.
A sampling frequency of 960 Hz has been considered for
all the algorithms except curve fitting algorithm.
For the curve fitting algorithm a sampling frequency
of 720 Hz has been considered in order to simplify the
computations.

00 iaoo reoo d.00

T I M E IDS)
aaoo rb. on 1
moo
It is reported in literature [ 2 2 ] that a secure
restraint function for a three-phase transformer can
be obtained if a single harmonic restraint signal is
? derived by combining the harmonics of the three
I phases. In the simulation results shown here, the
restraining signal is based on the combined second
harmonic content of the relay currents in the three
phases and the operating signal is based on the
combined fundamental component.

The performance of all the algorithms for inrush

is shown in Figures 7(a) to (f). The inrush currents
in all three phases and the corresponding relay
currents are shown earlier in Figures 4 and 5
respectively. A s seen from Figure 7, the restraining
signal is well above the operating signal and hence
there is no possibility of relay maloperation. The
performance of all the algorithms for a 3-phase
internal fault is shown in Figures 8(a) to 8(f).
Phase A current for this simulation is shown earlier
Figure 5 Relay Currents (Delta) Corresponding in Figure 6 . The restraining signal is greater than
to Figure 4 the operating signal for the signal samples
immediately following the fault. This is due to the
fact that the algorithm initializes the data window
with signal samples of 0 magnitude. As the data
window is filled with actual signal samples of the
differential current, the operating signal becomes
greater than the restraining signal and a secure trip
decision is obtained.

COMPARISON OF ALGORITHMS

Before designing a microprocessor based

differential relay for transformer protection, one
should consider the computational requirements
for the different algorithms and evaluate the
capability of the microprocessors available in the
I market. In a transformer digital relay the
d
“
‘0.00 10.00 20.00 Yh.00 4b.00 Sb.00
I
60.00
transformer current in each phase is sampled at
TIME Ins1 regular intervals and stored in the processor memory
through sample-hold circuits and analog to digital
Figure 6 Transformer Internal Fault Current converter. During each sampling interval the
fundamental and second harmonic content of the
(Phase A )
differential currents are evaluated using the
particular algorithm. The relay logic then determines
the operating and restraining signals. If a fault is
540
0 0
0 0

Y
o uv
a
8
*

T ime a f t e r energ I z a t I an [BSI T I m e a f t e r energ I -rat I an (11s)

Figure 7.a Fourier Algorithm Performance for Haar Functions Algorithm Performance
Figure 7.d
Inrush 0 for Inrush
0 0 0

vA
i Um vi

lb.00 2b.00 3b.00 rb.00 I50

Time a f t e r e n e r g l z a t i a n l m r l Time a f t e r e n e r g l z a t l a n (nrl

Figure 7.b Rectangular Transform Algorithm Figure 7.e Finite Impulse Response Algorithm
Performance for Inrush 0 Performance € O K Inrush 0
0 0 0

I
0

lb.00 2b.00 3b.00 tb.00 SO50

T 1 mea f t er c n c r g I z o t I on ( a d
Figure 7.c Walsh Functions Algorithm Figure 7.f Curve Fitting Algorithm Performance
Performance for Inrush for Inrush
 541

8 8

Time a f t e r f a u l t Im.1

Figure 8.a Fourier Algorithm Performance for Figure 8.d Haar Functions Algorithm Performance
Fault for Fault
0 0 0 O

n I
n

..
0
+a 0".
-&
+

3
-- 3

LO 0-
ri
-0

6"
n-
-f ;
-
0
-
CD

'"8 0"'"
7 - i
5
L
Y

.
o u
.f 2
0

20.00 30.00 40.00 5aJo

Im.1
Time a f t e r f a u l t
Figure 8.b Rectangular Transform Algorithm
Figure 8.e Finite Impulse Response Algorithm
Performance for Fault
0 0
Performance for Fault
0 0

Time after fault lmsl

Figure 8.c Walsh Functions Algorithm Figure 8.f Curve Fitting Algorithm Performance
Performance for Fault for Fault
542
confirmed it issues a trip signal and in case of an
inrush tripping is blocked. REFERENCES
A major part of the sampling interval is required
for the determination of the fundamental and second "Computer Relaying Tutorial Text", IEEE Power
harmonic content of the differential currents. The E n g i n e e r i n g S o c i e t y P u b l i c a t i o n No.
computational requirements for the different 79EH0148-7-PWR, 1979.
algorithms studied are shown in Table 1.
G.B. Gilchrest, G.D. Rockefeller, E.A. Udren,
Table I "High-speed Distance Relaying Using a Digital
Computer", IEEE Trans. on Power Apparatus and
Computational Requirements for Transformer Systems, Vol. PAS-91, No. 3, May/June 1972, pp.
Relay Algorithms 1235-1258.

M.M. Chen, W.G. Breingan, "Field Experience with

Digital System for Transmission L i n e
age of Protection", IEEE Trans. on Power Apparatus and
Systems, Vol. PAS-98, No. 5, Sept./Oct. 1979,
(1 Phase) pp. 1796-1805.

P: Bornard, J.C. Bastide, "A Prototype of

Fourier Multiprocessor Based Distance Relay", IEEE
Rectangular Trans. on Power Apparatus and Systems, Vol.
Walsh 50% PAS-101, No. 2 , Feb. 1982, pp. 491-498.

Finite 298 29% A.L. S t . J a q u e s , G. S a n t e r r e , " A

Curve 32% Multiprocessor-based Distance Relay: Design
Features and Test Results", IEEE Trans. on Power
Apparatus and Systems, Vol. PAS-102, No. 12,
The time required for these computations in each Dec. 1983, pp. 3842-3849.
sampling interval is also shown. This is based on an
INTEL 8086 16 bit microprocessor and INTEL 8087 J.S. Thorp, A.G. Phadke, "A Microprocessor based
Numeric Data Processor with a clock frequency of 8 Threephase Transformer Differential Relay",
MHz. During the early stages of research in digital IEEE Trans. on Power Apparatus and Systems, Vol.
relaying the investigators were concerned with PAS-101, No. 2, Feb. 1982, pp. 426-432.
replacing the multiplication and division operations
by arithmetic shift in order to reduce the time A. J . Degens, J . J.M. Langedi jk, "Integral
required for the computation. This is not a major Approach to the Protection of Power Transformers
concern now since the currently available by Means of a Microprocessor", Electrical Power
microprocessors and supporting components can perform & Energy Systems, Vol. 7, No. 1, Jan. 1985, pp.
high speed arithmetic also. 37-47.
As may be seen from this table, all algorithms A.R. van C. Warrington, 'Ptotective Relays;
can complete the computations for one phase within the Their Theory and Practice", Vol. 2, Chapman and
specified sampling interval, but, Curve fitting Hall, London, 1977.
Fourier and Finite Impulse Response algorithms can do
the computations with more than 60% of the time Applied Protective Relaying, Westinghouse
available for data acquisition and relay logic. A Electric Corporation, Newark, N.J., 1976.
major advantage of fast computation is that excess
time may be utilized for additional monitoring and J.A. Sykes and I.F. Morrison, "A Proposed Method
control functions. This is very useful if the of Harmonic Restraint Differential Protection of
microprocessor based transformer relay forms part of Transformers by Digital Computer", IEEE Trans.
an integrated microprocessor-based system for relaying on Power Apparatus and Systems, Vol. PAS-91, No.
and control of substations [23]. The relay designer 3, May/June 1972, 1266-1272.
can decide on the algorithm to be implemented based on
the application and the capability of the hardware O.P. Malik, P.K. Dash, G.S. Hope, "Digital
available. Protection of a Power Transformer", IEEE
Publication 76CH1075-1 PWR, Paper No. A76 191-7,
CONCLUSIONS IEEE PES Winter Meeting, New York, January 1976,
pp. 1-7.
Most of the published algorithms for digital
differential protection of transformers are compared. E.O. Schweitzer, R.R. Larson, A.J. Flechsig Jr.,
All the algorithms are able to distinguish between an "An Efficient Inrush Current-Detection Algorithm
inrush and internal fault. In case of an internal for Digital Computer Relay Protection o f
fault, the algorithms provide a reliable trip decision Transformers", Paper A77 510-1, IEEE Power
within a cycle from the instant of fault occurrence. Engineering Society Summer Meeting, Mexico City,
Based on this study, the Curve fitting, Rectangular 1977.
Transform, Fourier and Finite Impulse Response
algorithms are identified as those which may be easily A.J. Degens, "Algorithm for a Digital
implemented in a microprocessor based transfomer Transformer Differential Protection Based on a
differential relay. These algorithms perform the Least-Squares Curve Fitting", Proceedings IEE
computations well within the sampling interval. The (London), Vol. 128, Part C, No. 3, May 1981, pp.
additional time available in the sampling interval may 155-161.
be utilized for other monitoring and control
functions.
 543
For example, the algorithm should be checked for performance during loads
M.A. Rahman, P.K. Dash, E.R. Downton, "Digital that contain harmonic currents in addition to the fundamental frequency
Protection of Power Transformer Based o n currents.
Weighted Least Square Algorithm", IEEE Trans. on 2) The authors have stated in the section "Simulation Results" that the
Power Apparatus and Systems, Vol. PAS-101, N o . data window is initialized with samples of zero magnitude before subjecting
11, NOV. 1982, pp. 4202-4209. each algorithm to a magnetizing inrush or fault condition. Because of the
initialization of the data windows, the values of restrain and operate signals
M.A. Rahman, P.K. Dash, "Fast Algorithm for should be zero at time zero, i.e., at the inception of a fault. However, this is
Digital Protection of Power Transformers", not so in Fig. 8(a), 8(b), 8(c), 8(d), and 8(f) of the paper. Would the authors
Proceedings IEE(London), Vol. 129, Part C, No. explain the cause of these discrepancies.
2, March 1982, pp. 79-85. 3) It seems to us that the number of some arithmetic operations listed in
Table 1 of the paper are not correct. In our opinion, this number of + / -
D.B. Fakruddin, K. Parthasarathy, L. Jenkins, operations for Fourier, Walsh, and Finite Impulse Response algorithms

, ri
B.W. Hogg, "Application of Haar Functions for should be 54,94,and 62, respectively. With these changes in the numbers
Transmission Line and Transformer Differential of + / - operations, the corresponding numbers in columns 6 and 7 of
Protection", Electrical Power & Energy Systems, Table 1 would also change.
Vol. 6, NO. 3, July 1984, pp. 169-180. 4) The authors state that the frequency response of two filters for
measuring the fundamental frequency and the two filters for measuring the
M.A. Rahman, B. Jeyasurya, A. Gangopadhyay, second-harmonic frequency component have been combined to provide
"Digital Differential Protection of Power single responses shown in Figs. 1-3. We fail to find from the paper how the
Transformers based on Walsh Functions", responses of two filters designed to pass the selected frequency were
Transactions of CEA Engineering h Operating combined to obtain single responses. In our opinion, combining the outputs
Division, Vol. 24, Part 3, Paper 85-SP-149, of two algorithms to determine the fundamental frequency components
1985. provides a band of values. This can be illustrated as follows:
Consider that two algorithms provide orthogonal outputs of A, sin wt and
B. Jeyasurya, M.A. Rahman, "Application of Walsh A, cos wt when the input to the algorithms is A sin (wt + 0) as shown in
Functions for Microprocessor-Based Transformer Fig. D. 1 . Since the filters are nonrecursive, the outputs are orthogonal (they
Protection", IEEE Trans. on Electromagnetic
Compatability, Vol. EMC-27, No. 4, Nov. 1985, A q;(*t+e) As Sfn(wt)
~

pp. 221-225.

W.K. Sonnemann, C.L. Wagner, G.D. Rockefeller,

"Magnetizing Inrush Phenomena in Transformer
Banks", AIEE Transactions, Vol. 77, October
1958, pp. 884-892.
& Sin(wt)

M.A. Rahman, A . Gangopadhyay, "Digital

Simulation of Magnetizing Inrush Currents in Fig. D.1. Output of orthogonal digital filters.
Three -Phase Transformers", IEEE Paper No. 86WM
149-9, PES Winter Meeting, New York, Feb. 2-7, are designed to be) at all frequencies but the peak values of the two outputs
1986. are not equal at all frequencies.
1 The outputs of the orthogonal algorithms can be combined as follows to
T.R. Specht, "Transformer Inrush and Rectifier obtain the peak value of the phasor:
Transient Currents", IEEE Trans. on Power A ,w f ) 2 ] 0 5
A , = [ ( A c sin w ~ ) ~ + (cos
Apparatus and Systems, Vol. PAS-88, No. 4, April
1969, pp. 269-276. = [ A : sin2 w t + A ; cos2

C.H. Einvall, J.R. Linders, "A Three-phase

Differential Relay for Transformer Protection",
IEEE Trans. on Power Apparatus and Systems, Vol.
PAS-94, No. 6, Nov./Dec. 1975, pp. 1971-1980. The outputs are time-invariant for all frequencies for which A, is equal to
A,. At all other frequencies, the outputs are time-variant. If A, is greater
J.S. Deliyannides, E.A. Udren, "Design Criteria than A,, A, would have a maximum value of A, at wl = 0, 2a, . . ., and a
for a n Integrated Microprocessor-Based minimum value of A, at wt = a, 3a, . . .. Similarly, if A, is greater than
Substation Protection and Control System", IEEE A,, A, would have a maximum value of A, and a minimum value of A,.
The estimated peak value of the phasor would therefore be in the range A,
Trans. on Power Apparatus and Systems, Vol.
PAS-101, No. 6, June 1982, pp. 1664-1673. and A, as shown in Fig. D.2. It is interesting to note that the average value

Discussion
0.8
M. S. Sachdev, T. S. Sidhu, and A. Srivastava (Power System Research U
01

Group, University of Saskatchewan, Saskatoon, Sask., Canada S7N OWO): :0.6

The paper compares digital relaying algorithms for their application to m
E

differential protection of transformers. An attempt has been made to 9

compare the speed of response and ability to discriminate between inrush 0.4
and internal fault conditions of the selected algorithms. The discussers offer
the following comments on the paper.
0.2
1) The algorithms have been evaluated for two types of operating
conditions. They are a high-energy magnetizing inrush and a severe three-
phase internal fault. From the results of these tests, the authors have 0
concluded that the selected algorithms are able to distinguish between 0 1 2 3 4 5 6
magnetizing inrush and internal fault conditions. We are of the opinion that Normalised Frequency
the suitability of an algorithm for implementing in a practical design should
be checked by testing the algorithm driving marginal operating conditions. Fig. D.2. Range of estimated values of the phasor.
544

of A: is 0.5(A: + A f ) but the average value of A, is not [+(A: + after simulating a transformer digital relay, implementing the various
A similar phenomenon is observed, when the responses for the second- algorithms on a Personal Computer. The computational requirements are
harmonic filters are combined. based on an INTEL 8086 Microprocessor and an INTEL 8087 Numeric
5) We would also like to point out that the Fourier Algorithm should Data Processor which is required to do multiply, divide, and square root
have been implemented using 12 samples per cycle. This would have operations. The sampling rate is 16 samples per cycle. The programs
required fewer computations in comparison to the author’s approach of written in Assembly language are available from the authors from which
one may verify the total number of required arithmetic operations. The
using 16 samples per cycle. Twelve sample Fourier algorithm requires only number of arithmetic operations is correct. However, it is to be noted that
42 additionslsubtractions, 8 multiplications, and 2 square root operations.
the number of operations can be reduced further. But it requires additional
6) We would also like to point out that on-line application of the least MOVE, STORE, and LOAD instructions which are more time-consuming
square error approach was developed at University of Saskatchewan in the than simple ADD/SUB instructions.
1970’s [I]. The algorithm was used to design an algorithm for distance 4) The frequency response plot for an algorithm is obtained by
protection of transmission lines. Later, the least square error approach was transforming the algorithm into the complex Z-domain and substituting
applied to differential and earth fault protection of transformers [2]. frequency for the complex variable Z(Z = UT, where T i s the sampling
interval). The outputs are time-invariant for a specified sampling interval as
References per [I] of the paper and [A]. The responses shown in Figs. 1-3 of the paper
are obtained by evaluating the square root of the sum of the squares of the
[I] M. S. Sachdev and M. A. Baribeau, “A new algorithm for digital individual responses. The discussers provided a valid argument on two
impedence relays,” IEEE Trans. Power App. Syst., vol. PAS-98, orthogonal algorithms. In the paper it is assumed that A, is equal to A,.
no. 6, pp. 2232-2240, NovJDec. 1979. Thus it is implied that both the filter outputs are time-invariant. For the sake
[2] M. S. Sachdev and D. V. Shah, “Transformer differential and of completeness, the typographical errors of the discussion are corrected as
restricted earth fault protection using a digital processor,” in Trans.
follows
CEA, vol. 20, no. 4, paper 81-SP-155, Mar. 1981.
Manuscript received March 3, 1987. A,=[(A, sin ~ t ) ~ + ( A , c o~s t ) ~ ] ’ ~
= [ A i sin2 wt+Af cos2
M. A. Rahman and B. Jeyasurya: The authors thank the discussers for The output of filter F2 is A, cos w t .
their comments on the paper. In fact, Prof. M. S. Sachdev and his associates 5) The twelve-sample Fourier algorithm will reduce the total computa-
have extended the paper by the discussion contribution. tional requirements as pointed out by the discussers. Some of the algorithms
1) The authors agree with the discussers that performance of the studied specifically require that the number of samples per cycle be an
algorithms should also be checked during marginal operating conditions. integral power of 2. Hence a rate of 16 samples per cycle was also
Since the objective of the paper was to compare the algorithms with respect considered for the Fourier algorithm.
to their capability to distinguish between an inrush and internal fault, 6) The authors took note of the discussers’ research on the application of
saturation of current transformers, loads containing additional harmonic the least square error approach for digital protection as noted in [ 141 of the
currents, etc., have not been simulated. A practical design implementing
any suitable algorithm s h d d incorporate additional logic to ensure that the paper.
relay provides a secure restraining signal during operating conditions which Reference
call for undesirable trip signal ([7] of the paper).
2) The filter coefficients of the algorithms corresponding to the [A] M. Habib and M. A. Marin, “A comparative analysis of digital
performance shown in Figs. 8(a), 8(b), 8(c), 8(d), and 8(f) of the paper relaying algorithms for the differential protection of three phase
have finite values at the beginning of the data window. At t = O+ , the fault transformers,” in IEEE 1987 PICA Conf. Rec. (Montreal, Que.,
current also has a finite value. This is multiplied by the finite filter Canada, 1987), pp. 460466.
coefficient resulting in finite values of restraint and operating signals.
3) The arithmetic operations listed in Table 1 of the paper are arrived at Manuscript received June 30, 1987.