See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3275227

A Novel Digital Directional Transformer Protection Technique Based on

Wavelet Packet

Article in IEEE Transactions on Power Delivery · August 2005

DOI: 10.1109/TPWRD.2005.848646 · Source: IEEE Xplore

CITATIONS READS
80 437

1 author:

M.M. Eissa
Helwan University
112 PUBLICATIONS 2,121 CITATIONS

SEE PROFILE

All content following this page was uploaded by M.M. Eissa on 14 June 2015.

The user has requested enhancement of the downloaded file.

1830 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

A Novel Digital Directional Transformer Protection

Technique Based on Wavelet Packet
M. M. Eissa, Senior Member, IEEE

Abstract—This paper presents a novel digital technique for The effect of current transformer saturation in one or all of
transformer protection. The technique is based on deriving a di- three current transformers of a three-phase feeder has been dis-
rectional quantity proportional to the fault current signal and the cussed in [18]–[20].
prefault voltage signal. Standard fast wavelet transform (FWT)
schemes may not be as effective for data that has chiefly oscillatory A wavelet-based signal processing technique is an effective
features. An effective solution to discrimination involves examining tool for power system transient analysis and feature extraction.
the signal in both the time and frequency domains simultaneously. A few papers [21]–[23] are used to distinguish between the in-
The wavelet packet transform is an extension of the FWT that ternal faults and inrush currents.
allows for finer characterization of signal content for both time The paper proposed a new wavelet packet method to distin-
and frequency together. A 11/132-kV transformer connected to a
132-kV power system was simulated using Alternative Transient guish between the internal and external faults to the transformer
Program/Electromagnetic Transient Program (ATP/EMTP). protection zone. The technique uses the fault current and pre-
Results indicate that the proposed technique is stable, reliable, fault voltage signal as a directional signal. The proposed tech-
and fast during the discrimination between internal and external nique is also used to distinguish between the magnetizing in-
faults, magnetizing inrush currents, and internal faults, ratio-mis- rush and internal faults in power transformers. The proposed
match, and saturation of current transformers (CTs).
technique has stability during CT saturation and ratio-mismatch
Index Terms—Current-transformer (CT) saturation, inrush cur- conditions.
rent, ratio-mismatch, transformer faults, wavelet packet transform The suggested directional signal combined from the voltage
(WPT).
and current measurements has distinguishing characteristics
that, in theory, enable it to be identified. The suggested signal
I. INTRODUCTION can be mainly oscillatory in character, impulsive, or has a
transient nature. Previous algorithms based on the frequency
D IFFERENTIAL protection techniques and recent ap-
proaches have been proposed for transformer protection.
The differential protection concepts are based on the assump-
characterization of the data using a fast Fourier transform (FFT)
will not perform well for impulsive or transient data. On the
tion that during an internal fault, the fundamental component of other hand, standard fast wavelet transform (FWT) schemes
the differential current becomes higher than the no-load current. may not be as effective for data that has chiefly oscillatory
In this area, several protection schemes have been developed features. The wavelet packet transform (WPT) is an extension
[1]–[8] to solve various difficulties (such as magnetizing current of the FWT that allows for finer characterization of signal con-
inrush, fixed current transformer ratios on variable ratio power tent for both time and frequency together. The wavelet packets
transformers due to tapping, and the unavoidably different are a more sensitive and computationally more flexible way
characteristic of CTs on the high- and low-voltage sides of the to do signal discrimination than FFT- or standard FWT-based
transformer). methods.
Some other techniques have been proposed to distinguish
between the magnetizing inrush currents and internal faults II. WAVELET PACKETS AND THEIR IMPLEMENTATION
[9]–[13]. A. Discrete Wavelet Transform
A technique for residual flux measurements and controlled
The discrete wavelet transform (DWT) is briefly discussed
closing to eliminate transformer inrush transients is proposed in
that is needed for definitions of wavelet packets. The wavelet
[14].
scaling function and mother wavelet are functions that satisfy
Some techniques have been developed that use the voltage
the following two-scale equations ( is an integer):
and current data [15]–[18]. References [15]–[17] used the dif-
ferential current and the flux to discriminate between faults and
(1)
magnetizing inrush cases. Reference [18] utilizes the concept of
symmetrical components.
(2)

Manuscript received May 11, 2004; revised July 15, 2004. Paper no. TPWRD- for some set of coefficients and for
00220-2004. which . That is, is the reverse of , in-
The author is with the Electrical Engineering Department, Faculty of Engi-
neering, Helwan University, Cairo 11421, Egypt. terleaved with sign changes. Each wavelet scaling function and
Digital Object Identifier 10.1109/TPWRD.2005.848646 mother wavelet function is represented by a set of coefficients
0885-8977/$20.00 © 2005 IEEE
EISSA: A NOVEL DIGITAL DIRECTIONAL TRANSFORMER PROTECTION TECHNIQUE BASED ON WAVELET PACKET 1831

Fig. 2. (a) WP algorithm with five levels of decomposition. (b) The

corresponding plot in the time-frequency plane of the transform coefficients.

The discrete wavelet transform, by way of its bandpass

H G
Fig. 1. (a) FWT for five levels of decomposition with the basis represented by filtering and downsampling of the original data set, algorithm
FIR filters and ; (b) the corresponding plot in the time frequency plane of
the transform coefficients.
illustrated in Fig. 1(a), has the time frequency illustrated in
Fig. 1(b).

and . The coefficients and are low- B. Wavelet Packet Transform

pass and highpass FIR filter coefficients, respectively. They are
used to construct the set of wavelet basis functions and The standard basis functions mentioned earlier are special
defined below; these are scaled and translated copies of cases of wavelet packets. References [24], [25] provide a
the scaling function and the mother wavelet more detailed exposition. Consider Figs. (1a) and (2a), the
recursive pyramidal FIR filter bank implementation of the
(3) FWT described earlier can be extended by varying the tree
structure; that is, the sub–band information represented by
(4) the detail coefficients can be decomposed even further
[Fig. 2(a)]. The advantage of this further series of operations
If and are integers, then the basis functions are scaled by is that the time frequency plane, Fig. 2(b), is partitioned more
a factor of , and translated time units. If the time resolution precisely; in fact, the filtering sequence can be adjusted even
of a discretely sampled function is , then or further in order to represent the signal energy more efficiently
is the same as or . onto a minimal number of nonzero coefficients, which has
Let be a discretely sample current function of length , obvious advantages in signal compression. If we have a signal
. The decomposition of in of length , then for levels of decomposition, we have
the wavelet basis is done by recursive filtering with and (at sets of sub-band coefficients of length . Each set is a
wavelet function db1, current signal with 50 Hz and a sampling downsampled, bandpass-filtered copy of the as with regular
frequency 20 kHz) (Fig. 1). Since has a length of , then wavelets, but with more resolution in time.
there is a maximum of levels of decomposition, or a max- Wavelet packets are then generalizations of wavelets. Let
imum of applications of and . The wavelet transform to , where is being defined earlier as the
coefficients/decomposition of is given by depth of the filtering tree or number of levels of decomposition.
Let us define the even and odd wavelet packets and
as with wavelets in (1) and (2) by
(5)
(6)
where is the number of levels of decomposition,
or depth of the tree representing the decomposition algorithm and
presented in Fig. 1. Each set of wavelet transform coefficients
and , for , represents a bandpass filtered (7)
and downsampled version of the original signal .
1832 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

Fig. 3. Voltage and current signals for an internal fault.

Fig. 4. Decomposition results of the combined signal.
Using this notation it is clear that and
TABLE I
. Since wavelet packets are generalizations of wavelets, WINDOWS AND THEIR FREQUENCY BANDS
then a series of basis functions can be generated by
scaled and translated versions of as in (3) and (4)

(8)
In wavelet packet analysis, a signal is represented as a
sum of orthogonal wavelet packet functions at dif-
ferent scales, oscillations, and locations
(9)

where is the wavelet packet coefficient.

III. SIGNAL ANALYSIS AND FEATURE EXTRACTION

A. Feature Extraction
In monitoring of different conditions of power transformers,
the current and voltage signals contain complicated information.
To ensure the correct discrimination for different fault condi-
tions, it is important to extract the features of the current and Fig. 5. Simulated power transformer system.
voltage signals.
Fig. 3 shows typical changes of the current and voltage sig- bands are indicated in Table I. These decomposed results of the
nals during an internal fault measured at the low-voltage side of signal not only keep the same features, but also provide more
a power transformer. Wavelet packet transform is represented information, such as the time-domain constituent parts of the
as a compressed signal method. Therefore, it is ideal to use the signal at the frequency band.
wavelet packets as the extracted features [26]. According to the
above, each wavelet packet transform represents certain infor- B. Detection of Internal and External Fault
mation on the signal analysis and feature extraction in a specific Fig. 5 shows a power transformer protected by two direc-
time-frequency window. Fig. 4 shows the decomposition results tional relays. These relays are looking into the transformer. A
of the proposed signal combined from the voltage and current trip output is produced when all of the directional functions op-
waveforms at 50 Hz and a sampling rate 20 kHz. By choosing a erate which will only be the case for a transformer fault.
set of mother and father wavelet functions with a known oscil- The technique is based on deriving a directional function pro-
lation index, one can decompose the combined signal to create portional to the current and voltage signals after fault inception.
a wavelet packet table. The wavelet packet table is created using The prefault voltage is used as a directional reference. The above
the db1 wavelet function. As shown in the figure, each resolu- process can be described in terms of the combined signals ,
tion level is subdivided into blocks which contain the wavelet , , , , and , computed at the low- and
packet coefficients. high-voltage sides of the transformer due to the product of the
Level 0 reproduces the input, and resolution levels 1 to 4 are fault current and prefault voltage. The wavelet packet transform,
plotted in descending order. The frequency windows and their which consists of the main part of the relay, is then applied to
EISSA: A NOVEL DIGITAL DIRECTIONAL TRANSFORMER PROTECTION TECHNIQUE BASED ON WAVELET PACKET 1833

current during transformer energization. In this section, a new

approach is proposed to discriminate between magnetizing
inrush and an internal fault. As shown in Fig. 4, for each level,
the leftmost block corresponds to an oscillation index
and contains the low-frequency, approximated coefficients. The
rightmost block has oscillation and contains the high-frequency
coefficient created from the highpass mother wavelet. The
decomposed components can be considered as the features
of the original signal. So the wavelet packet transform is
used to capture important features of the decomposed signal
that are sensitive to the changes of the inrush signal and CT
saturation. These feature are described based on the sum of
the different wavelet coefficients from window 1 to window
7 as . This value represents the level of the tran-
sient inrush signal. For process of discrimination, this value is
compared with the wavelet coefficient in window 0 . To
achieve a reliable discrimination inrush signal, is computed
as

(11)

where is used to distinguish between transformer inrush cur-

rents or power system fault currents. For a power system fault
Fig. 6. Overall arrangement of the technique at one side. current, exceeds a positive level value.

the signals to extract the desired constituent parts of the input IV. SIMULATION RESULTS
signals at each frequency band. Fig. 6 shows the basic arrange- The ATP/EMTP program [27] is used to generate data to test
ment of this technique. For simplicity of explanation, only one the performance of the proposed technique. The transformer
signal is used in the analysis. The results of feature extrac- model considered is a typical 11/132 kV Y/Y. The neutral point
tion from wavelet coefficients corresponding to the input signal of the Y connection is grounded. The source parameters and line
are computed at a certain frequency band (level-3 and db1 length are shown in Fig. 5. The CT saturation model given in
wavelet function). The directional signal that discriminate [28] is used. For verifying the criterion developed for checking
between internal or external faults is calculated from the coeffi- switch-on faults, the value of the level was set at 3.75 (extensive
cient as studies establish that this level is suitable for the discrimination).
To ensure an algorithm reliability for discrimination between in-
(10) ternal and external faults, the THR value is given as 10% of the
prefault conditions.
where “ ” is the most recent samples and “ ” is the number of
A. Impact of Ratio-Mismatch
samples in the window. is the directional signal that has rich
information about the directionality and during CT saturation. The errors of transformation of the CTs may differ from each
This value can be positive or negative. The final performance of other; thus leading to significant differential current when there
the technique is identified as follows: If the directional signal is normal load flow, or an external fault. In any case, even with
goes lower than negative threshold value THR, the technique some adjustments, there remains some residual ratio-mismatch,
will identify that the fault is external. On the other hand, if the which leads to a small differential current during normal condi-
directional function goes higher than some positive threshold tions. This is one of the shortcomings in the current differential
value, a forward fault is identified. relays. The proposed approach is based on the stand-alone de-
The overall directional transformer protection introduces in- cision from the two directional signals located at each side of
dependent decisions at each voltage side. A trip output is pro- the power transformer. This leads to the fact that the CT error
duced when the directional signals at both voltage sides operate and ratio-mismatch problems do not affect on the fault decision
which will only be the case for an internal fault. The final trip made by the relay.
logic combines the decisions from all of the directional trans-
former protection relays using AND gate logic. B. Test Results
The performance of the proposed technique was evaluated for
C. Detection of Magnetizing Inrush different types of internal and external faults.
The major drawback of power transformer protection is the 1) Internal Fault: A three-phase-to-ground fault on the
possibility for false tripping caused by the magnetizing inrush high-voltage side of the power transformer (location F1
1834 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

Fig. 7. Current waveforms at the low-voltage side and the relay responses
(3LG internal fault at F1).

Fig. 9. Current waveforms at the low-voltage side and the relay responses
(3LG external fault at F2).

Fig. 8. Current waveforms at the high-voltage side and the relay responses
(3LG internal fault at F1).
Fig. 10. Current waveforms at the high-voltage side and the relay responses
(3LG external fault at F2).
in Fig. 5) was simulated. The performance of the algo-
rithm is shown in Fig. 7. Fig. 7(a) shows the waveforms
of the fault currents (ia, ib, and ic) at the low-voltage side. different polarities. This led to the decision that the fault
Fig. 7(b) shows the corresponding directional signals is outside the protection zone of the transformer
, , and for phases a, b, and c, respec- 3) Magnetizing Inrush: The transformer shown in Fig. 5
tively. Fig. 8(b) shows the relay performances at the is energized from the 11-kV side. The inrush signal
high-voltage side. As shown in the 11- and 132-kV sides, is computed. Fig. 11(a) shows the inrush currents
the directional signals are positive and cross the threshold measured at the low-voltage side. Fig. 11(b) shows the
boundary. At that time, the trip logic confirmed that the corresponding three inrush signals , , and for
fault is that in the protection zone of the transformer. phases a, b, and c, respectively. As shown in the figure,
2) External Fault: For the power system shown in Fig. 5, the signals do not reach the threshold level. This has led
consider a three phase-to-ground fault outside the trans- to the transformer not experiencing a fault. The case is
former zone . The results are shown in Figs. 9 and correctly classified as a magnetizing inrush current. The
10. Figs. (9a) and (10a) show the current waveforms relay will be inhibited from issuing a trip signal.
measured at the low- and high-voltage sides, respectively. 4) Switching a Fault: Fig. 12(a) shows the responses of the
Figs. (9b) and (10b) show the corresponding directional inrush to an internal fault. Fig. 12(b) shows the responses
signals. As shown in figures, the directional signals have of the inrush detection circuit given in Fig. 6. As shown
EISSA: A NOVEL DIGITAL DIRECTIONAL TRANSFORMER PROTECTION TECHNIQUE BASED ON WAVELET PACKET 1835

Fig. 11. Magnetizing inrush current and the corresponding inrush signals (no
Fig. 13. Inrush current and the corresponding inrush signals due to a 3LG
fault).
internal fault at F1 and CT saturation.

has the features that it is direct relation to the frequency

changes not to the current value.

V. CONCLUSION
The wavelet packets are more sensitive and a computation-
ally more flexible way to conduct signal discrimination than
wavelet transform-based methods. The principle and applica-
tion of the wavelet packet transform-based transformer protec-
tion is analyzed and discussed in this paper. The discrimination
between internal and external faults is presented. The technique
also includes a new approach for discrimination between mag-
netizing inrush and internal faults. An accurate inrush signal
based on the constituent parts of the decomposed signal at dif-
ferent frequency bands is provided. Several features were de-
rived from wavelet packet transform, and the optimal features
sensitive to the fault currents, inrush currents, and CT satura-
tion. The technique is stable during the CT saturation and is not
affected by CT error and ratio-mismatch. The feature extraction
Fig. 12. Inrush current and the corresponding inrush signals due to 3LG
internal fault at F1. with wavelet packet transform can be implemented in real time
since the wavelet packet transform requires only a small amount
of computation.
in the figure, the inrush signals , , and asso-
ciated with the internal fault are crossing a positive level.
REFERENCES
This led to the final decision that the fault is inside the
protection zone. [1] M. S. Sachdev and D. V. Shah, “Transformer differential and restricted
earth fault protection using a digital processor,” Trans. Eng. Operating
5) CT Saturation: The secondary current can be quite dis- Div. Can. Elect. Assoc., vol. 20, pp. 1–11, 1981.
torted relative to the primary current. If conditions are se- [2] J. S. Thorp and A. G. Phadke, “A microprocessor based three-phase
vere enough, it is possible that the distortion may be even transformer differential relay,” IEEE Trans. Power App. Syst., vol. PAS-
101, pp. 426–432, Feb. 1982.
worse and the saturation can start to occur even sooner. [3] P. Liu, D. Chen, Y. Guo, O. P. Malik, and G. S. Hope, “Improved opera-
The severe saturation can cause problems in the trans- tion of differential protection of power transformers for internal faults,”
former differential relays. Fig. 13(a) shows the responses IEEE Trans. Power Del., vol. 7, no. 4, pp. 1912–1919, Oct. 1992.
[4] B. Kasztenny, E. Rosolowszi, B. Hillstrom, and M. M. Saha, “A self
obtained due to inrush waveforms with 3LG internal fault organizing fuzzy logic based protective relay-an application to power
at F1 and CTs (located at the low-voltage side) are satura- transformer protection,” IEEE Trans. Power Del., vol. 12, no. 3, pp.
tion. As shown in figure, the variations in inrush wave- 1119–1127, Jul. 1997.
[5] A. A. Girgis, D. G. Hart, and W. Bin Chang, “An adaptive scheme for
forms due to CT saturation do not have any effect on digital protection of power transformers,” IEEE Trans. Power Del., vol.
the relay responses [Fig. 13(b)], while the suggested 7, no. 2, pp. 546–553, Apr. 1992.
1836 IEEE TRANSACTIONS ON POWER DELIVERY, VOL. 20, NO. 3, JULY 2005

[6] A. Wiszniewski and B. Kasztenny, “A multi-criteria differential trans- [19] G. Hosemann and H. M. Steigerwald, “Model saturation detector for
former relay based on fuzzy logic,” IEEE Trans. Power Del., vol. 10, digital differential protection,” IEEE Trans. Power Del., vol. 8, no. 3,
no. 4, pp. 1786–1792, Oct. 1995. pp. 933–940, Jul. 1993.
[7] Z. Bo, G. Weller, and T. Lomas, “A new technique for transformer pro- [20] M. Mikru, W. Winkler, and B. Witek, “Performance of differential for
tection based on transient detection,” IEEE Trans. Power Del., vol. 15, three-winding power transformers during transient CT’s saturation,” in
no. 3, pp. 870–875, Jul. 2000. Proc. 4th Int. Conf. Developments Power System Protection, 1989, Inst.
[8] X. Li, X. Yin, D. Chen, and Z. Zhang, “The multi-differential relaying Elect. Eng. Pub. 302, pp. 40–44.
for transformer protection,” in Proc. IEEE/Power Engineering Soc. [21] M. G. Morante and D. W. Vicoletti, “A wavelet-based differential
Transmission Distribution Conf. Exhibition Asia Pacific, vol. 3, 2002, transformer protection,” IEEE Trans. Power Del., vol. 8, no. 4, pp.
pp. 1723–1726. 1351–1358, Oct. 1993.
[9] J. Shaohua and L. Wanshun, “A novel scheme to discriminate inrush [22] A. S. Omar, “A wavelet-based technique for discrimination between
current and fault current based on integrating the waveform,” Proc. Conf. faults and magnetizing inrush currents in transformers,” IEEE Trans.
Software Engineering Education, vol. 19, no. 8, pp. 35–38, 1999. Power Del., vol. 18, no. 1, pp. 170–176, Jan. 2003.
[10] T. S. Sidhu and M. S. Sachdev, “On-line identification of magnetizing in- [23] Y. Sheng and M. Steven, “Decision trees and wavelet analysis for power
rush and internal faults in three-phase transformers,” IEEE Trans. Power transformer protection,” IEEE Trans. Power Del., vol. 17, no. 2, pp.
Del., vol. 7, no. 4, pp. 1885–1890, Oct. 1992. 429–433, Apr. 2002.
[11] Y. Kukiaki, “Power differential method for discrimination between dif- [24] R. R. Coifman and Y. Meyer, Wavelets and Their Applications: Jones
ferential transformer protection,” IEEE Trans. Power Del., vol. 12, no. and Bartlett, 1992.
2, pp. 1109–1115, Jul. 1997. [25] M. V. Wickerhauser, Adapted Wavelet Analysis from Theory to Soft-
[12] L. G. Perez, A. J. Flechsig, J. L. Meador, and Z. Obradovicc, “Training ware. Wellesley, MA: A. K. Peters, 1994.
an artificial neural network to discriminate between magnetizing inrush [26] M. V. Wickerhauser, Lectures on Wavelet Packet Algorithm. St. Louis,
and internal faults,” IEEE Trans. Power Del., vol. 9, no. 1, pp. 434–441, MO: Dept. Math., Washington Univ., 1991.
Jan. 1994. [27] ATP/EMTP Can. EMTP Users Group, Version PC Salford 486 version
[13] A. Kulidjian, B. Kasztenny, and B. Campbell, “New magnetizing inrush 19, Jan. 1998.
restraining algorithm for power transformer protection,” in Inst. Elect. [28] M. M. Eissa, “A novel digital directional technique for bus-bars protec-
Eng. Developments in Power System Protection, 2001, Conference Pub- tion,” IEEE Trans. Power Del., vol. 19, no. 4, pp. 1636–1641, Oct. 2004.
lication no. 479, pp. 181–184.
[14] J. H. Brunke and K. J. Frohlich, “Elimination of transformer inrush cur-
rents by controlled switching-part I: theoretical considerations,” IEEE
Trans. Power Del., vol. 16, no. 2, pp. 276–280, Apr. 2001.
[15] A. G. Phadke and J. S. Thorp, “A new computer-based flux restrained
current-differential relay for power transformer protection,” IEEE Trans.
Power App. Syst., vol. 102, no. 11, pp. 3624–3629, Nov. 1983. M. M. Eissa (M’96–SM’01) was born in Helwan, Cairo, Egypt, on May 17,
[16] K. Inagaki, M. Higaki, Y. Matsui, M. Suzuki, K. Yoshida, and T. Maeda, 1963. He received the B.Sc. and M.Sc. degrees in electrical engineering from
“Digital protection method for power transformer based on an equivalent Helwan University, Cairo, Egypt, in 1986 and 1992, respectively, and the Ph.D.
circuit composed of inverse inductance,” IEEE Trans. Power Del., vol. degree from the Research Institute for Measurements and Computing Tech-
3, no. 4, pp. 1501–1510, Oct. 1988. niques (Hungarian Academy of Science), Budapest, Hungary, in 1997.
[17] M. S. Sachdev, T. S. Sidhu, and H. C. Wood, “A digital relaying algo- Currently, he is an Associate Professor at Helwan University. In 1999, he was
rithm for detecting transformer winding faults,” IEEE Trans. Power Del., invited to be a Visiting Research Fellow at the University of Calgary, Calgary,
vol. 4, no. 3, pp. 1638–1648, Jul. 1989. AB, Canada. His research interests include digital relaying, application of neural
[18] T. S. Sidhu, H. S. Gill, and M. S. Sachdev, “A power transformer pro- networks to power systems, dispatcher training simulators, and local area and
tection technique with stability during current transformer saturation and wavelet applications in power systems.
ratio-mismatch conditions,” IEEE Trans. Power Del., vol. 4, no. 3, pp. Dr. Eissa received the Egyptian State Encouragement Prize in advanced sci-
798–804, Jul. 1989. ence in 2002.

View publication stats