Design of a digital protection scheme for power

transformers using optimal state observers

Y.V.V.S. Murty
W.J. Smolinski
S. Sivakumar

Indexing terms: Transformers. System protection, Digital control, Fault detection

inherent in the provision of an adequate ground-fault

~

Abstract: The paper presents the design of a sensitivity in differential protection often require that
digital protection scheme for three-phase power ground-fault protection should be added to each winding
transformers. The design is based on an optimal of the transformer. Overcurrent relays are used to protect
state observer for estimating the fundamental and the transformer against damage due to external faults.
harmonic components of the signals. The scheme Since the relay must not operate under emergency
provides primary protection for internal faults, loading conditions, it requires a high current setting.
ground faults and backup protection for external Also, the time setting may have to be high in order to
faults. Internal fault protection is provided by co-ordinate with other overcurrent relays in the system.
means of a percentage differential characteristic, The overcurrent relays provide negligible protection for
which also includes even- and fifth-harmonic faults inside the transformer bank and they are usually
restraints to prevent misoperation during inrush employed only as backup protection for external faults.
and overexcitation conditions, respectively. Pro- Several digital relay algorithms have been proposed by
tection for external faults is provided by means of various authors for calculation of the fundamental and
a three-phase digital overcurrent relay. The pro- harmonic components of current signals [ 2 , 31. These
tection scheme is implemented on a TMS320 algorithms require about one cycle of past data to be
digital signal processor and tested in the labor- stored in memory and, since the weighting coefficients for
atory using a three-phase transformer. Relay oper- each sample are different, they also require a large
ation times are in the range of $ to 2 of a 60 Hz number of coefficients to be stored in memory. The
cycle for internal and ground faults. The percent- response time of these algorithms for internal faults is
age differential relay provides complete restraint around one cycle.
during inrush, overexcitation and external fault Recently the authors [4, 51 have applied a Kalman
conditions. filtering technique for the estimation of harmonic com-
ponents. The algorithm is recursive and does not require
past values of data. The response time of the algorithm
1 Introduction
for internal faults is around half a cycle. The gains of the
Kalman filter are time-varying and a large number of
The advent of low cost digital processors has made it gains must be stored for implementation.
possible to design and implement digital protection In this paper an algorithm, based on an optimal state
schemes for power system applications. observer, for estimation of the fundamental and harmo-
The protection of three-phase power transformers nic components of the signal is described. The aspects of
poses a challenging problem for the relay designer. The speed of response and noise rejection are considered in
design should consider various operating conditions, viz. designing the observer. The algorithm does not require
magnetising inrush during the energisation of the trans- any past data and thus reduces the data memory require-
former, overexcitation resulting from the dynamic over- ment. In contrast to the Kalman filter algorithm, only a
voltage condition, on-load tap changing, current few gain constants must be stored. The response time of
transformer (CT) saturation and ratio errors. The protec- the algorithm for internal faults is in the range of $ to 2
tion scheme should provide primary protection for inter- of a cycle.
nal and ground faults and backup protection for external The following Sections of the paper describe the design
faults [l]. of the observer, percentage differential protection, ground
Protection for internal faults is usually provided by fault protection, external fault protection and implemen-
means of a differential relay. To alleviate the problems of tation of the protection scheme on TMS320 digital signal
transformer on-load tap changing and CT saturation, the processor.
differential relay is often provided with a percentage
(biased) differential characteristic. Harmonic restraint is 2 Harmonic estimation using an optimal state
used to guard against misoperation during magnetising observer
inrush and overexcitation conditions. The difficulties
In earlier papers [4,51 the authors have applied a time-
Paper 6000C (P7, Pll), first received 6th October and in revised form varying Kalman filter approach to estimate harmonic
22nd December 1987 signal components for transformer differential protection,
The authors are with the Department of Electrical Engineering, Uni- assuming that the signals are generated by a deterministic
versity of New Brunswick, Frederickton, N.B. E3B 5A3, Canada system corrupted by noise. In this approach a relay
224 I E E P R O C E E D I N G S , Vol. 135, Pt. C. N o . 3, M A Y 1988
response time of about half a cycle has been realised. (F - KH) within the unit circle. The locations of the
However, since the gains are time-varying in this eigenvalues determine, among other things, the transient
approach, a number of gains have to be precomputed response of the observer. For the purpose of transformer
and stored for this implementation. A variation of this protection, the speed of response and tracking ability are
technique is the use of state-space observers with con- of particular importance.
stant gains. The estimation of spectral components using A simple choice of the gain matrix K is to assign all
state-space observers is of considerable interest in digital the eigenvalues at the origin, which yields a dead-beat
signal processing [6, 73 and the use of observers for state- observer. With this design the state estimates i kconverge
estimation is well known in control [SI. In the present to the actual states x k in 2n + 1 steps, giving the fastest
paper this technique is used for the estimation of the har- response. However, the practical use of this design is
monic components of the differential- and through- limited because it is highly sensitive to unmodelled har-
currents of a power transformer. monics and noise. In general, a fast convergence rate
Consider the following deterministic state-variable implies a poor noise rejection property; conversely, a
equation for a periodic signal having harmonic com- good noise rejection property implies slow response.
ponents up to nth order with samples z k , at time tk, Thus, the design must be based on some trade-off
+
(2n 1 ) samples per period 16, 71. between these two conflicting objectives. The linear quad-
ratic optimal observer design provides a suitable frame-
xk+l = Fxk
work for this effect.
Z k = HXk The optimal observer design is aimed at minimising a
where (2n + 1)-dimension state vector x k is as follows: quadratic cost functional of the form
xk(2i- 1) = real component of the ith harmonic
(4)
phasor
xk(2i)= imaginary component of the ith harmonic where Q and R are symmetric weighting matrices. Q is
phasor positive semidefinite and R is positive definite. The super-
script = denotes the transpose of a matrix.
xk(2n + 1 ) = DC component The observer gain matrix K which minimises J is given
where the ith element of x k is represented by xk(i). by
K = FPHT(R + HPHT) ’ (5)
where P is a stable steady-state solution of the Riccatti
F= equation
Pk+’ = FPkFT+ Q - FPkHT(R+ H P k H T ) - ’ H P k F T(6)
.
The superscript -’ denotes the inverse of a matrix.
where $ = oT,w is the fundamental supply frequency in
Standard algorithms are available for solving the Riccatti
radiens/sec and T is the sampling interval in seconds. equation (e.g. the McFarlane-Potter-Fath method [SI).

f(i+)
cos (i$)
[sin ( i $ )
-sin (i$)
cos (i$) 1
i = 1, 2, ..., n

and H is a 1 x (2n + 1) matrix which gives the connec-

The weighting matrices Q and R play an important
role in shaping the observer response. Normally, Q and R
are chosen as diagonal matrices. In general terms, the
‘higher’ the Q (with reference to R), the smaller the error
tion between the measurement ( t k ) and the state vector in state estimation and faster the response; on the other
(xk). The sampled value of the signal is considered to be hand, the ‘higher’ the R, the slower the response and the
the sum of the real components of the harmonic phasors better the noise rejection properties. Thus with a judi-
and the DC component. Therefore, the matrix H is given cious choice of Q and R one can make trade-offs between
by the speed of response and the noise rejection properties.
H = [ l , O , l , O,..., l,O,l]. The Q and R matrices are similar to the covariance
matrices of the system and measurement noise, respec-
The harmonic components hi(RMS) are given by
tively [7], and the resulting observer is equivalent to a
hi2 = ( x t ( 2 i - 1) + x,2(2i))/2, i = I, . . ., n steady-state Kalman filter. However, the noise com-
hO = x,(2n + 1). ponents are not modelled in the present design and hence
Q and R are treated as weighting matrices.
The sampling frequency should be at least twice the
highest harmonic frequency in the signal. If the signal
contains an exponentially decaying DC component with 3 Design of a protection scheme for power
known decay time constant l/L, the last diagonal element transformers
of the F matrix (which represents the DC component) can
be changed to e-’=. Transformer current signals are assumed to contain har-
The problem of estimating the present state of the monics up to fifth, and higher-order harmonics beyond
signal model (eqn. 1) from output measurements Z k fifth are assumed to be negligible. An eleven-state optimal
involves the design of standard state observers [SI. The observer is designed to estimate the components up to
observer state can be represented by fifth harmonic. A sampling frequency of 16 samples per
cycle is used. In designing the observer, the weighting
+ 1 = Fik + K(zk - H i , )
i-, (3) factors in Q should be chosen taking the ‘relative impor-
where f k denotes the estimate of the state vector xk and K tance’ of the individual components for the objective of
is the observer gain matrix. The primary objective in differential protection into account. After studying
choosing K is to obtain a stable observer, which is various choices of Q and R matrices, the following case is
achieved by assigning the eigenvalues of the matrix considered.
I E E PROCEEDINGS, Vol. 135, Bt. C , N o . 3, M A Y 1988 225
 Q = diag. [lo, 10, 5, 5, 1, 1, 0.1, 0.1, 2, 2, 21, R = 100 A combined fifth-harmonic signal (I5), which is used for
restraint during overexcitation conditions, is formed by
where ‘diag’ denotes the diagonal elements of a square combining fifth-harmonic components of differential cur-
matrix, and the resulting gain matrix is rents as follows:
K = C0.248 0.0513 0.173 0.046 0.0674 0.0434
0.00916 0.0236 - 0.000415 0.1 13 0.0802]’.
The frequency response of the observer for the above case
is shown in Fig. 1. It can be seen from this Figure that
the lower the weighting in the Q matrix for a given har-
monic, the faster the fall-off of the side lobes in the fre-
quency response of the corresponding harmonic filter. It
can also be seen that the observer structure allows the
flexibility of shaping the frequency response in a desired
manner by appropriately choosing the weighting matrices
Q and R.
The protection scheme tested in the laboratory is
shown in Fig. 2. As noted earlier, the protection scheme
consists of percentage differential protection, ground fault normalised frequency
protection and external fault protection, each of which is a
now described in that order.

3.’I Percentage differential protection

The percentage differential characteristic (PDC) imple-
mented in the protection scheme is shown in Fig. 3. The
PDC is based on the fundamental components of the
through and differential currents.
The differential currents are
i d t ) = i l a ( t ) - Ci2a(t) - izc(t)I
idb(t)

idc(t)
= ilb(t) - [iZb(t) - i20(t)1 ’ N2/N1
= i ~ c ( t )- Ciz,(t) - i2dt)l . Nz/”i
NZ/N,
} (7)

normalised frequency
and the through currenls are b

itow

itb(t)

itc(t)
= (il,(t)
=
= (ilc(t)
+ Ci2.W
+ [i2b(t)
+ [i2c(t)
- izc(t)l
- i20(t)1
- i2b(t)l
‘

’
Nz/N,)/2
NZ/Nl)/2

NZ/Nl)/2
} (8)

where NJN, is the turns ratio of the transformer.

The PDC has a variable threshold (C, and Cb); the
maximum threshold Cb will be active during the over-
excitation condition. This threshold corresponds to a
setting of 1.3 times the rated voltage. The slope (C,) of
the PDC must be adjusted to make the relay insensitive
to CT inaccuracies anti off-nominal tap positions of the
transformer.
Six 11-state observers (eqn. 3), three for differential norma I(sed frequency
currents and three for through currents, are used to c
process the data. The initial estimate of the state vector
10-
io is taken as zero. The fundamental, second, fourth and
fifth harmonic components of the differential currents
0 8-
and the fundamental components of the through currents h

are computed using eqn. 2. It has been suggested that a -u 0 6 -

L

secure harmonic restraint function can be obtained by U

3 -
combining the harmonics of all the three phases [SI. A
combined fundamental ( I , ) is formed by combining the
fundamental components of the differential currents over
all the three phases as follows: 0 2-

I: = C h:.
a. b . c
(9) 8
norrnalised frequency
A combined even-harmonic signal ( I 2 & which is used for
d
restraint during inrush conditions, is formed by combin-
ing second- and fourth-harmonic components of differen- Fig. 1 Frequency response of the observer
tial currents as follows: a Fundamental component
b Second harmonic component
I:, = 2 ( h i + hi).
a. b. c
c Fourth harmonic component
d Fifth harmonic component

226 I E E P R O C E E D I N G S , Vol. 135, Pt. C, No. 3, M A Y I988

The percentage differential characteristic should be based the maximum value of the worst case inrush current to
on the fundamental components of through and differen- make the relay insensitive during magnetising inrush.
tial currents. The fundamental component ( h , ) of the dif-
3.2 Ground fault protection
7 5 kVA , 2 L O V / 2 4 O V , deltaistar transformer The sensitivity of the overall differential protection is
circuit breaker somewhat limited for ground faults, owing to the magni-
tude of ground fault impedance. Sensitive ground fault

even harmontc

restraint

Fig. 2 Digital protection scheme for power transformers

I I
I no no
I I

repeat this block

for a l l three phases

If,P u percintage
Fig. 3 Percentaye differential characteristic differential
Characteristic
I , = differential current
I , = through current
signal

ferential current is denoted as I , and the through current

as I , for each phase. The overall relay logic for the per-
centage differential relay is given in Fig. 4. The following
relay settings are chosen for the sample design: Fig. 4 Flowchart ofpercentage differential relay logic
C , , Cb and C , are as shown in Fig. 3
C , = 0.0625 P.u.’, Co = 0.25 P.u.’ C, is the threshold setting for the even-harmonic restraint
C , is the threshold setting for the fifth harmonic restraint
C, = 0.0625 (6.25% slope: 1; against 1: CD,, ~, are delay counters
~

or 25% slope: I , against I T ) .

C , = 0.0625 P.u.’, C , = 0.0078 P.u.’, T D D = 3. protection is usually obtained by using separate primary
and secondary ground fault protection. The delta-side
These settings are only an example, and other values can ground fault current is
be used instead depending on the application.
Typical current signals of the transformer, recorded in ilgf(t) = ila(t) + ilb(t) + i l c ( t ) (12)
the laboratory are given in Fig. 5. The performance of the and the star-side ground fault current is
observer algorithm (OBA), along with the discrete
Fourier transform algorithm (FA) [lo] under various i2gf(t) = + i 2 b ( t ) + i2t(t) + i 2 g ( f ) ’ (13)
conditions, is given in Figs. 6 and 7. From these Figures Two 11-state observers are used to estimate the fun-
it can be seen that the relay is restrained during magne- damental components of the delta and star side ground
tising inrush and overexcitation conditions and the relay fault currents (Ilgfand I z g f ) . If these currents are above
operates in less than 2 of a cycle for internal faults. These the threshold value (Cgt= 0.15 P.u.) for a specified dura-
Figures also show the performance of the relay for inter- tion, a ground fault is assumed and the relay gives a trip
nal and external faults with partial saturation of CTs, signal. Fig. 8 shows the ground relay response for a high
and that the OBA estimates the harmonic components impedance ground fault on the secondary.
faster than the FA, resulting in faster relay operation
time. It is observed that the estimates obtained from both 3.3 External fault protection
the algorithms during steady state converge to the same A fault on a power system network external to the trans-
value in all the cases. former can result in a considerable damage to the trans-
The relay also has an instantaneous trip feature for former if the fault is not cleared promptly. Overcurrent
heavy internal faults. If the magnitude of any of the dif- relays (OCR) may be used to clear the transformer from
ferential current exceeds a predefined threshold and stays external faults before the transformer is damaged. The
for two consecutive samples then a trip signal is given. pickup setting of the OCR should be high enough to
The threshold value is set at a value much higher than permit overloading of the transformer. A setting of 200-
I E E PROCEEDINGS, Vol. I35 Pt. C , No. 3, M A Y 19R8 227
 c

Ida Ida Ido

'dbr
IdaL

'db 'tb

r t

' d c b dc 'dc 'tc

L,
,,.
,,,
a b C d
Fig. 5 Transformer currenr signal under various conditions
a Inrush
x: 1.96 midivision
y : 0.243 p.u./division
h Internal fault (short across phase c winding on secondary)
x : 1.96 ms/division
y : 0.608 p.u./division
c Overexcitation
x : 1.96 ms/division
y : 0.05 p.u./division
d External fault (short across phase b to ground on secondary)
x : 1.96 ms/division
y : 0.334 p.u./division

-
300% of the self-cooled current rating of the transformer L 8 12 16
is common. The time setting is usually chosen for co- tirne,rns
ordination with relays on adjacent equipment and the a
upper limit for the time setting is based on the trans-
former heat-curve [111.
The fundamental components of the input currents of
the transformer are obtained from the through and differ-
ential current components. Three overcurrent relays, one
for each phase, are used at the input side of the trans-
former.
The general form of OCR characteristic can be given
as follows [123 :
To = K'/(In - 1) (14)
time,ms
where To is the tripping time and I is the current magni-
tude in multiples of relay setting. Various characteristic b
curves could be obtained by using different values of K'
and n. As an example, K' = 80 and n = 2 (extremely
inverse) are used to obtain the characteristic curve shown
in Fig. 9.
The digital implementation of the OCR is as follows:
let 1, be the current magnitude (in multiples of the relay
setting) at kth sample. Consider the following integration
with the initial sum U , set to zero
Uk+l = uk + 1; - 1 (15)
and if the relay is operated when the running sum reaches
a threshold value ( K t J then the relay operating time for a O,; 4 8 12 16
constant magnitude of current I is time,rns

To = K,,T/(I* 1) c
- (16)
Fig. 6 Even harmonic restraint performance of the relay
where T is the sampling interval.
FA
The overcurrent relay must be reset whenever the ~ OBA
current magnitude falls below its pickup value. A variety (I Inrush h Internal fault c External fault

228 I E E PROCEEDINGS, Vol. 135, Pi.C , N o . 3, M A Y 1988

 ~

The total execution time for the relay software is

around 670 p s which includes data acquisition time of
180ps. The total program occupied about 640 words of
program memory and 143 words of data memory.

006-
w- \.
-
m:w 0.0.4 - \.
\.
.'
002- i.
\., t r i p p i n g zone-
\ '.. '\
,
0- ' --?.-,

time,ms
a

011 I
1 10 100
c u r r e n t , multiples of relay setting
Fig. 9 Overcurrent relay characteristic curve

.- I

OO 4 a 12 16
time, m s
b
Fig. 7 Fijih harmonic restrairit performance ofthe relay
FA
__ OBA
a Internal fault with partial CT satuiation
h Overexcitation

inputs
Fig. 10
- 'la Ilb 'IC I2a I2b '2c I29
Hardware architecture ofthe proposed d~qitalrelay

0 a 12 16 20 Analogue prefiltering of the current signals is not used

time,ms since the high-frequency components (above fifth
Fig. 8 Performance of the ground relay during a high impedance harmonic) are found to be very small. If the current
groundfault on the secondary side signals contain higher levels of these high-frequency har-
FA
monics, then either these harmonics should be included
~~

~ OBA
in the signal model or analogue prefiltering should be
used to remove these components. This, of course, would
pickup and time setting used for the relay are arbitrary add additional time delay to the relay operating time.
and they can be set to any values, depending on the
application. 4 Conclusion

3.4 Hardware details of the implemented relay The design of a digital protection scheme based on an
The proposed relay hardware is shown in Fig. 10. It con- optimal state observer is described. The protection
sists of a TMS32010 processor [13], and a data acquisi- scheme includes a percentage differential protection,
tion system which includes analogue scaling circuits primary and secondary ground fault protection and over-
(LM741), sample and hold circuits (LF398), analogue current protection for external faults. The percentage dif-
multiplexer (MPC801), 12-bit A/D convertor (AD ferential protection has an even harmonic restraint for
ADC80), and a sampling clock generator. The on-chip magnetising inrush current and a fifth-harmonic restraint
memory (program and data) available on the TMS320 for overexcitation conditions.
(TMS320M10) is sufficient to store the relay software and The protection scheme is implemented on a TMS320
there is no need for additional memory interfacing. The digital signal processor and tested in real time. The tests
relay software is developed on a TMS320 evaluation also include partial CT saturation during internal and
module [141 and tested in real time. external faults. Heavy faults with deep AC CT saturation
I E E PROCEEDINGS, Vol. 135, i't. C,N o . 3, M A Y 1988 229
could not be tested in the laboratory due to limitations in 4 MURTY, Y.V.V.S., and SMOLINSKI, W.J. : ‘Design and implemen-
the laboratory setup. The results indicate that relay oper- tation of a digital differential relay for a 3-phase power transformer
ation times for internal faults are in the range of $ to 2 of based on Kalman filtering theory’, IEEE PES Winter Meeting,
Paper 87 WM113-4, New Orleans, Feb. 1-6, 1987
a cycle. The described percentage differential relay pro- 5 MURTY, Y.V.V.S., and SMOLINSKI, W.J.: ‘A Kalman filter based
vides complete restraint during inrush, overexcitation digital percentage differential and ground fault relay for a 3-phase
and external fault conditions. power transformer’, Accepted for presentation at the IEEE PES
The protection scheme is very economical since, with Winter Meeting, New York, Jan. 31 to Feb. 5, 1988.
6 HOSTETTER, G.H.: ‘Recursive discrete Fourier transformation’,
minimal hardware, complete electrical protection of a I E E E Trans., 1980, ASP-28, pp. 184-190
multiple-winding transformer could be achieved. 7 BITMEAD, R.R., TSOI, A.C., and PARKER, P.J.: ‘A Kalman filter-
ing approach to short-time Fourier analysis’, I E E E Trans., 1986,
ASSP-34, pp. 1493-1501
8 KAILATH, T.: ‘Linear systems’ (Prentice-Hall, New Jersey, 1980)
5 Acknowledgment 9 EINVALL, C.H., and LINDERS, J.R. : ‘A three-phase differential
relay for transformer protection’, I E E E Trans., 1975, PAS-94,
The authors are grateful for many useful discussions with pp. 1971-1980
Dr. K. Natarajan, Department of Electrical Engineering, 10 THORP, J.S., and PHADKE, A.G. : ‘A microprocessor-based three-
phase transformer differential relay’, I E E E Trans., 1982, PAS-101,
University of British Columbia. The authors also thank pp. 4 2 W 3 2 .
him for allowing his control system design package [lS] 11 ‘Guide for protective relay applications to power transformers
used in some parts of this work. C37.91’, ANSIiIEEE Standard, 1985
12 REDFERN, M.A.: ‘The application of microprocessors to power
system protection relays’, Presented at the Pennsylvania Electric
Association Fall Meeting, Sept. 18-19, 1984, Hershey, Pennsylvania
6 References 13 ‘TMS32010 user’s guide’, Digital Signal Processing Products, Texas
Instruments, 1983
1 WARRINGTON, A.R. van C.: ‘Protective relays: their theory and 14 ‘TMS32010 evaluation module user’s guide’, Digital Signal Pro-
practice’ (Chipman and Hall, 1971) cessing Products, Texas Instruments, 1984
2 ‘Computer Relaying’, IEEE Tutorial Course, 79 EH0148-7-PWR 15 NATARAJAN, K.: ‘POLYFREQ - A computational package for
3 RAHMAN, M.A., and JEYASURYA, B.: ‘A state-of-the-art review frequency domain design of M I M O control systems in APLT,
of transformer protection algorithms’, IEEE PES Winter Meeting, System Dynamics and Control Group, University of New Bruns-
Paper 87 W M 114-2, New Orleans, Feb. 1-6, 1987 wick, 1983

230 I E E P R O C E E D I N G S , Vol. 135, P t . C , N o . 3, M A Y 1988