Professional Documents
Culture Documents
Component Analysis
E. Vzquez, Member, IEEE, I. I. Mijares, Student, O. L. Chacn, Member, IEEE, and A. Conde, Member, IEEE
I. INTRODUCTION
When transformer internal faults occur, immediate
disconnection of the faulted transformer is necessary to avoid
extensive damage and/or preserve power system stability and
power quality [1]. The percentage differential principle is
widely used to detect faults in transformers [2], providing
excellent results toward improving the security of differential
protection for external faults with current transformers (CT)
saturation.
A misoperation of differential relays is possible due to
inrush currents, which result from transients in transformer
magnetic flux. Some solutions to this problem include to
introduce an intentional time delay in the differential relay, to
desensitize the relay for a given time, or to supervise the
differential relay [3],[4]. In the latter option, the harmonic
content of the differential current provides information that
helps differentiate between inrush conditions and actual faults;
these methods are known as harmonic restraint or blocking
methods.
These methods have problems in cases with low harmonic
content in the differential current [5]. Transformer
overexcitation is another possible cause of differential relay
misoperation. Protective relay
manufacturers have
implemented an additional fifth-harmonic restraint to prevent
misoperations on overexcitation [6], and several methods
based on waveform recognition to distinguish faults from
inrush [7],[8],[9]. However, these waveform recognition
techniques do not identify transformer overexcitation
conditions.
This paper describes a new algorithm for transformer
differential protection. The proposed algorithm explores the
possibility to characterizing the differential current waveform
behavior to discriminate internal faults from inrush and
E. Vzquez (evm@ieee.org), I. I. Mijares, O. L. Chacn
(ochacon@mail.uanl.mx) and A. Conde (con_de@yahoo.com) are with
Universidad Autnoma de Nuevo Len, Monterrey, NL, Mxico
I O = I1 + I 2
(1)
I R = k I1 I 2
CT1
CT2
IO
I1
I2
IR
IO > k I R + I P
(2)
Sw j = j w j
(3)
xn = W T ( y n y )
(4)
iDIF ( a b ) I AB I ab
I DIF = iDIF ( b c ) = I BC I bc
iDIF ( c a ) I CA I ca
(5)
where IAB, IBC, ICA, Iab, Ibc and Ica are delta signals [14] of the
secondary CT currents in both transformer ends, expressed in
per unit of the CT rated current. The delta signals were
selected to improve the algorithms ability to discriminate fast
changes in the differential current waveform that are typical of
capacitive effects in transmission lines.
The delta signals have been used in transmission line
protection using high-frequency components of traveling wave
phenomena [10]. In transformer protection, the delta signals
reduce the load effect because they are almost zero in stable
state. When a fault occurs, by virtue of the superposition
theorem, the fault injected components can be acquired by
subtraction of the steady-state components from the post fault
signals (incremental signals) using a delta filter [14]; the fault
injected components can be expressed as:
i (t ) = i (t ) i (t nT )
(6)
(7)
iDIF ( a b )1
iDIF ( a b ) 64
iDIF ( b c )1
M
I T =
iDIF ( b c ) 64
i
DIF ( c a )1
M
DIF ( c a ) 64
(8)
X
+ 1 I f Zb
R
yn = [I T I T ] / I CT
(9)
(10)
Y = y (k ,1)
M
y (n,1)
y (1,2)
M
L
M
y (k ,2)
M
L
M
y (n,2)
y (1, p )
M
y (k , p)
M
y (n, p)
(11)
zone
zone
zone
(12)
(13)
(14)
(192 x1)
(15)
where y is the row vector with the mean value for each 192
dimensions of the original data, and PC is the representation
of yn vector in the 2D-subspace of the first two principal
components.
Therefore, the differential protection principle is reduced a
pattern recognition process in a 2D subspace, and each point
in this space represents a waveform of the differential current
for a specific disturbance.
V. ALGORITHM
0.05 <
PC1
PC1
< 0.05
Inrush
(16a )
< 0.20 Overexcitation (16b)
0.20 <
PC1
Fault
(16c)
VI. RESULTS
The algorithm was evaluated in distinct scenarios simulated
in the power system test (see Fig. 4). Every scenario was
formed as a combination of situations, including changes in
transformer loading, in the magnetizing curve shape, in the
power system time constant (source impedance) and in the
differential CT ratios. A total of 3170 simulation cases were
run to show the feasibility to implementing this algorithm for
transformer differential protection. These cases are described
in Table 2; every scenario was simulated for all-352
conditions describe in Table 1.
1
2
3
4
5
6
7
8
9
10
TABLE 2.
SCENARIOS USED TO TEST THE ALGORITHM.
Scenarios
Number of
cases
Increase reactive load 10%
352
Reduce reactive load 10%
352
Move knee in magnetizing curve +10%
352
Move knee in magnetizing curve -10%
352
Modify generator impedance from resistive to RL
352
Increase generator inductance 10%
352
Reduce generator inductance 10%
352
Increase CT ratio in 10%
352
Reduce CT ratio in 10%
352
Combination of previous scenarios
2
Total
3170
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
Fig. 10. 352 simulation cases for scenario 9.
VII. CONCLUSIONS
The percentage differential principle is widely used to
detect faults in transformers, providing excellent results in
improved security of the differential protection for external
faults with CT saturation. However, it has problems
discriminating fault currents from the false differential
currents caused by magnetizing inrush and transformer
overexcitation.
The methods used to solve these problems use harmonic
content or waveform recognition, but cases with low harmonic
content in the differential current are yet unsolved.
This paper describes a new algorithm for transformer
differential protection; the proposed algorithm explores the
possibility to characterize the waveform behavior of the
differential current to discriminate internal faults from inrush
or overexcitation events. The PCA method extracts the
features from the differential current in order to implement a
pattern recognition process in a 2D space called feature space.
The representation of the original current signals in the feature
space shows a linearly separable structure, not requiring a
classification technique to solve it.
The algorithm was evaluated with 3170 simulation cases,
which include changes in transformer loading, in the
magnetizing curve shape, in the source impedance and in the
differential CT ratios. In all cases, the algorithm successfully
discriminates internal faults from inrush or overexcitation
events.
VIII. REFERENCES
[1]
[2]
[3]
[4]
[5]
[14]
[15]
[16]
IX. BIOGRAPHIES
Ernesto Vzquez received his B.Sc. in Electronic and
Communication Engineering in 1988, and his M.Sc. and
Ph.D. in Electrical Engineering from the Universidad
Autnoma de Nuevo Len, Mxico, in 1991 and 1994
respectively. Since 1996 has worked as Research Professor
in Electrical Engineering for the UANL. He is IEEE
member. In 2000 he did a research stay in University of
Manitoba, Canada, where he was working in traveling wave
protection algorithms. His research areas are power electrical systems
protection and the artificial intelligence application in power systems.
Ivn I. Mijares received his B.Sc in Electrical Engineering and his M.Sc in
Electrical Engineering from the Universidad Autnoma de Nuevo Len,
Mxico, in 2002 and 2005 respectively. His research area is power system
protection.
Oscar Chacn received his B.Sc. in Chemical Engineering
from the Universidad Autnoma de Nuevo Len, Mxico, in
1968. In 1974 he received his M.Sc. from the University of
Houston and his Ph.D. in 1987 from the University of Texas
in Austin. From 1968, he is full time professor in the
Universidad Autnoma de Nuevo Len, Mxico. He is IEEE
member. His research areas are optimization and artificial
intelligence techniques applied in power systems.
Arturo Conde received the B.Sc. degree in mechanical and
electrical engineering from the Universidad Veracruzana,
Mxico, in 1993, and his M.Sc. and Ph.D. in Electrical
Engineering from the Universidad Autnoma de Nuevo
Len, Mxico, in 1996 and 2002 respectively. From 2003,
he is full time professor in the Universidad Autnoma de
Nuevo Len. His research areas are power system protection and power
quality.