1

A Wavelet-Based Algorithm for Disturbances

Detection Using Oscillographic Data
F. B. Costa, K. M. Silva, K. M. C. Dantas, B. A. Souza and N. S. D. Brito

Abstract This paper presents a discrete wavelet transform approach to disturbance detection by the analysis of electromagnetic
transients in transmission lines using oscillographic data. The
detection is carried out by the analysis of the detail coefficients
energy of the phase currents. The performance of the method
was evaluated for actual oscillographic data and excellent results
were obtained.
Index Terms Disturbance detection, transmission lines,
wavelet transform.

I. I NTRODUCTION
Electromagnetic transients in power systems are characterized by high frequency components during a short period
of time. These signals superpose both voltage and current
waveforms in a variety of disturbances such as faults, capacitor bank switching transients and lighting strikes. As a
consequence, excessive currents or over voltages may appear
in power system, reducing its reliability.
The wavelet transform (WT) has been widely used to
analyze non-stationary phenomena, such as electromagnetic
transients, due to its ability to analyze a localized area of a
signal, revealing discontinuities and related frequency spectrum [1], [2].
The aim of this paper is to introduce a method for disturbance detection at wavelet domain by the analysis of
electromagnetic transients in transmission lines, using actual
oscillographic data. The detection is carried out by the analysis
of the wavelet coefficients energy related to phase currents.
Disturbances produced by faults, capacitor switching and
transmission line energization and deenergization are taken
into account. The performance of the method was evaluated for
actual oscillographic data and excellent results were obtained.
Software based on the proposed method has been used by
Hydro Electric Company of Sao Francisco (CHESF), an utility
company of Brazil, with a satisfactory performance.

The discrete wavelet transform (DWT) is normally implemented by Mallats algorithm [4]. Its formulation is related
to filter bank theory, since it uses the high-pass h(k) and
the low-pass g(k) filters to divide the frequency-band of the
input signal into high- and low-frequency components. This
operation may be repeated recursively, feeding the downsampled low-pass filter output into another identical filter pair,
decomposing the signal into approximation c(k) and detail
d(k) coefficients for various scales of resolution. In this way,
the DWT may be computed through a filter bank framework:
in each scale, h(k) and g(k) filter the input signal of this
scale, yielding new approximation and detail coefficients,
respectively. So, this process divides the frequency spectrum
of the original signal into octave bands [3]. This filter bank
framework is depicted in Fig. 1. The down-pointing arrows
denote a decimation by two and the boxes denote a convolution
by h(k) or g(k).

Fig. 1.

DWT filter bank framework.

The coefficients of the filters pair are associated with the

selected mother wavelet. Daubechies wavelet family is the
most used for analysis of power system transients [5]. In
this paper, the Daubechies 4 (db4) wavelet was used as the
mother wavelet, due to its good time resolution that provides
an accurate detection of the fast transients [6].
III. E NERGY OF THE D ETAIL C OEFFICIENTS

II. WT AND F ILTER BANKS

The WT is a powerful time-frequency method to analyze a
signal within different frequency ranges, by means of dilating
and translating of a single function named mother wavelet [3].
This work was supported by Brazilian National Research Council (CNPq)
and CHESF.
The authors are with Department of Electrical Engineering at Federal
University of Campina Grande, 882 Aprgio Veloso Av, Bodocongo, Campina
Grande - PB, CEP:58.109-970, Brazil. E-mail: {flabc, kms, karcius, benemar,
nubia}@ee.ufcg.edu.br.
Presented at the International Conference on Power Systems Transients
(IPST07) in Lyon, France on June 4-7, 2007.

In order to compute the wavelet coefficients energy, named

detail-spectrum-energy (DSE), at scale j, a moving data window goes through the detail coefficients shifting one coefficient at a time, viz
k+Nw /2j

Ew (k) =

d2j (n),

(1)

n=k

where Nw is the window length (number of samples contained

in one cycle of the fundamental frequency of the original
signal), k = {1, 2, . . . , (Ns Nw )/2j } and Ns is the total
number of samples of the original signal.

Fig. 2 depicts the windowing process to obtain the DSE, at

scale 1, for a faulted current (Fig. 2(a)). The detail coefficients
(Fig. 2(b)) are obtained after applying one stage of the DWT
in the original signal and the DSE (Fig. 2(c)) is obtained by
(1). According to windowing process, the sample related to
the beginning of the disturbance is anticipated from the actual
sample on original signal by Nw /2j samples.
IV. D ETECTION OF P OWER S YSTEM T RANSIENTS
The most common sources of transients in power systems
are atmospheric phenomena and switching of power system
apparatus. In this way, transient phenomena are present in
faults, voltage sags, capacitor switching transients and transmission line energization and deenergization.
The transient phenomena analyzed in this paper and their
respective detail coefficients analysis are discussed in the
following.
A. Capacitor Switching
Fig. 3 depicts a typical actual transient current waveform
related to a capacitor switching, that was obtained from a
transmission line. This kind of transient occurs for a very
short duration, usually less than one cycle, and the electrical
circuit is quickly restored to original operation [7]. The detail
coefficients of the current depicted in Fig. 3 are shown in Fig.
4 for different levels of resolution.

According to multiresolution decomposition theory, at scales

with high resolution, the wavelet is most localized in time and
oscillates rapidly within a very short period of time. In these
scales, the wavelet is must suitable to detect fast and short
changes on the signal (Fig. 4(a)). Whereas, at scales with low
resolution, the dilated wavelet version becomes less localized
in time and less oscillatory. In these scales, the wavelet is more
suitable to detect low frequency components (Fig. 4(d)).
The higher detail coefficients are related to transient phenomena (Fig. 4). At first scale, these coefficients can be used
for disturbance detection and for beginning and end time
identification [8].

B. Transmission Lines Energization and Deenergization

Transmission lines energization is a typical switching operation accomplished by circuit-breakers and may cause several
transients in power systems.
Fig. 5 depicts a typical actual current waveform due to a
transmission line energization. Theirs related detail coefficients
are depicted in Fig. 6. According to these figures, the major
value of the detail coefficients indicates the circuit-breaker
operation and it can be used for disturbance detection.

Fig. 3.

(a)
window 1

Detail coefficients

0.6

Normalized current waveform of a capacitor energizing transient.

window

window 300

(Ns Nw)
2

(a)

window 301
0.3
0
-0.3

Ns
2

-0.6

100

200

300

400

500

600

700

800

(b)

900

Detail coefficient index

(b)
Detail-spectrum-energy

10
10
10
10
10

0
-1

e1(300)

-2

e1
-3

Ns Nw
2

10

-4 e1(1)

Ns
2

-5

(c)

e1(301)

100

200

300

400

500

600

700

800

900

Detail coefficient index

(c)
Fig. 2.

Windowing: (a) original signal; (b) detail coefficients; (c) DSE

(d)
Fig. 4.

Detail coefficients: (a) scale 1; (b) scale 2; (c) scale 3; (d) scale 4.

Fig. 5. Normalized current waveform of a transmission line energization

transient.

Fig. 7. Normalized current waveform of a transmission line de-energization

transient.

Fig. 6.

Fig. 8.

Detail coefficients at scale 1.

In the same way, transmission line deenergization transients

are switching operation accomplished by circuit-breakers, that
interrupt the power flow through the transmission line.
Fig. 7 depicts a typical actual current waveform due to a
transmission line deenergization. Theirs related detail coefficients are depicted in Fig. 8. It can be observed that the major
value of the detail coefficients indicates the disturbance and
circuit-breaker operation.

Detail coefficients at scale 1.

In this paper, voltage sags related to faults on a parallel

transmission line are analyzed.
Fig. 11 depicts an actual voltage sag. The detail coefficients
for voltage and current are shown in Fig 12.
A common event associated with transmission line voltage
sag is the transient phenomena at currents and voltages waveforms. Therefore, the first scale may be suitable to detect these
fast electromagnetic transients.

C. Faults
The current and voltage signals obtained from a transmission line when a fault occurs have high frequency components,
that are quickly dumped [9]. These components are present
around the beginning and end time of the fault and they are
appropriately detected at first wavelet scale.
Fig. 9 depicts the voltage and current at the faulted phase,
for an actual single-phase-to-ground fault. The related detail
coefficients, at first scale, are shown in Fig. 10.
D. Voltage Sags
Voltage sags are related to power quality problems. They
are characterized by rms voltage reduction (between 0.1 and
0.9 pu) from steady-state normal system operation, during a
period of time from 0.5 to 30 cycles [10].
A fault on a transmission line generates sags and swells at
voltage signals. Sags occur on the faulted phases during a fault
until the protection system acts to clear the fault. On the other
hand, swells are not as common as sags, but they may occur
at phases that are not involved on the fault, specially during a
single-phase-to-ground fault [10].
A very common occurrence of voltage sag is related to faults
on other transmission lines from the same power system. In
these case, the transmission lines will experiment a sag until
the protection system acts to clear the fault.
Voltage sags can also be related to switching of large inductive loads due to inrush currents, but this kind of occurrence
is less common than those aforementioned.

(a)

(b)
Fig. 9.

Normalized waveform of an actual fault: (a) voltage; (b) current.

(a)

(b)
Fig. 10.

Detail coefficients at scale 1: (a) voltage; (b) current.

Current (pu)

4
1.5
0
-1.5

384

768

1152

1536

1920

2304

2688

3072

Number of the sample

(b)
Fig. 11.
current.

Normalized waveforms of an actual voltage sag: (a) voltage; (b)

Fig. 13.
Normalized current waveform of an actual oscillographic data
without disturbance.

Detail-spectrumenergy

(a)

10
10
10
10

-3
-4

E2

-5

E1

-6

192

384

576

768

960

1152

1344

1536

Energy index

Fig. 14.

DSE of the current, at scale 1, without disturbance.

(a)

(a)
(b)
Fig. 12.

Detail coefficients at scale 1 for phase A: (a) voltage; (b) current.

V. P ROPOSED M ETHOD
It is proposed a method for detection of disturbances in
transmission lines by the analysis of electromagnetic transients
using actual oscillographic data. The detection is carried out
by the analysis of the DSE of the phase currents. In other to
develop the method, exhaustive studies related to DSE were
accomplished and it was observed that:
1) The first wavelet scale is most appropriated for disturbance detection.
2) DSE of the currents provides a pattern that may be used
for transients detection.
3) For oscillographic data without disturbance (e.g. Fig. 13)
the DSE may range between thresholds E1 and E2 (Fig.
14), which are discussed in Section VI.
4) For oscillographic data with disturbance, the following
statements can be derived from the DSE illustrated
at Fig. 15, which is related to the current waveforms
depicted in Section IV.
a) The energy values related to steady-state normal
system operation, usually, range between thresholds E1 and E2 .
b) In records due to transmission line energizations,
deenergizations and faults (e.g. Figs. 15(b), 15(c)
and 15(d)), the energy may reach values below E1 .
c) There is a sharp variation corresponding to the
transients beginning time.
d) The energy values are greater than E2 during the
transient phenomena.

(b)

(c)

(d)

(e)
Fig. 15. DSE of the current, at scale 1, related to: (a) capacitor switching;
(b) transmission line energization; (c) transmission line deenergization; (d)
fault; (e) voltage sag.

Detail-spectrum-energy

Fig. 16 is used to introduce the method. It depicts the DSE

curve related to a capacitor switching and the parameters used
to disturbances detection, which are described in the following:
1) E1 and E2 : used to delimit the region in which the
energy values are related to steady-state normal system
operation.
2) Emax : maximum energy value during disturbance time.
3) E: energy variation in five consecutive energy points.
4) Epos : mean energy at one cycle of the fundamental
frequency, computed after disturbance.
5) k1 : sample related to the beginning time of the disturbance.
6) k2 : sample related to the end time of the disturbance.
7) k = k2 k1 : delimit the time of the disturbance.
10
10
10
10
10
10

Steady-state normal
system operatiom

-1
-2

Steady-state normal
system operatiom

Emax

-3

VI. P ROPOSED M ETHOD I MPLEMENTATION

The method was based on the actual oscillographic records
presented in Tab. I, which ware obtained from transmission
lines with different voltage rates (138, 220 and 500 kV) and
different sampling rates (1.2, 2.4 and 15.36 kHz).
TABLE I
T YPE AND AMOUNT OF AVAILABLE ACTUAL OSCILLOGRAPHIC RECORDS
FOR DETECTION RULES ELABORATION .

Epos

iii) if k > kmin , so the disturbance was detected (analysis ended).

iv) Otherwise, get k = k2 and go to step 5a.
c) Otherwise, the oscillographic record do not correspond to a disturbance (analysis ended).
Emin is the minimum energy variation to disturbance
detection and kmin is the number of samples in 2 cycles.

Oscillographic record

Amount of available
records

Without disturbance
transients due to capacitor switching
energization of transmission line

40
33
25

deenergization of transmission line

fault
sag

20
33
50

E2

-4
-5

k/2
k1/2

-6

192

384

576

E1

k2/2
768

960

1152

1344

1536

Energy index

Fig. 16.

Parameters at DSE curve for disturbance detection.

The major energy variation of DSE (Emax ) is related

to the beginning time of the disturbance. This is the main
characteristic used by the proposed method for detection. For
each sample k, E is computed considering five consecutive
points of the DSE curve:

Fig. 17 depicts the histograms with minimum and maximum

energy values, related to available oscillographic records without disturbance (Tab. I).

E = max{E1 (k), E1 (k + 1), . . . , E1 (k + 4)} E1 (k), (2)

where: E1 (k) is the energy value at sample k; max{.} returns
the maximum energy value among related samples.

(a)

A. Summary of the Proposed Method

The following algorithm is carried out for disturbance
detection:
1) Normalize the phase currents (IA , IB , IC ) of the oscillographic record.
2) Apply the DWT for each phase current to get the detail
coefficients at scale 1.
3) Calculate the DSE of the phase currents (EA , EB and
EC ), according to (1).
4) If Emax < E2 , so the oscillographic record do not
correspond to a disturbance (analysis ended).
5) Otherwise, get k = 1.
a) From the k, identify some E(k) > Emin at
the energies EA (k), EB (k) and EC (k).
b) If E(k) > Emin was obtained.
i) Get k1 = k.
ii) From the k1 , identify the set of energy samples
with main value less than E2 , getting Epos , k2
and k = k2 k1 .

(b)
Fig. 17. Energy histogram for the oscillographic records without disturbance:
(a) minimum values; (b) maximum values.

According to Fig. 17, the smaller energy value is 1.105

(Fig. 17(a)) and the bigger energy value is 3.104 (Fig. 17(b)).
Therefore, E1 = 1.105 and E2 = 3.104 .
Fig. 18 depicts the histograms with Emax of the transients
due to capacitor switching, fault and sag (Tab. I).
For each oscillographic record, it was observed that Emax
corresponds to beginning of the disturbance. According to Fig.
18, for transients due to capacitor switching, Emax ranges
from 1.103 to 1.101 ; for faults, it is bigger than 8.102 and
for sags, Emax ranges from 4.104 to 6.
Similar energy histograms for transmission line energization and deenergization was available, and it was stabilized
Emin = 1.104 .

Some problems was observed in few oscillographic data, in

which the transient signals are very dumped. In these cases,
the DSE curve can not present a sharp energy variation needed
to disturbance detection.
A software based on the proposed method was installed at
oscillographic network, of an utility company of Brazil and
excellent results have been obtained.

(a)

R EFERENCES

(b)

(c)
Fig. 18. Histogram of the maximum energy variation for the oscillographic
records with: (a) transients due to capacitor switching; (b) fault; (c) sag.

VII. P ROPOSED M ETHOD P ERFORMANCE E VALUATION

The proposed method was evaluated for 772 actual oscillographic records (Tab. II) from different transmission lines of
CHESF, with different voltage rates (138, 220 and 500 kV)
and different sampling rates (1.2, 2.4 and 15.36 kHz). It was
obtained a success rate of 97.40%.
TABLE II
P ERFORMANCE OF THE METHOD .

Real

Diagnostic
Desired

Amount of
records

Right
diagnostic

Without disturbance
Energization
Deenergization

Without disturbance
With transients
With transients

233
85
60

233
85
59

Fault
Sag
Transients due to
capacitor switching

With transients
With transients
With transients

63
284
47

63
270
42

772

752

C ONCLUSION
A novel method for transmission lines disturbance detection
by the analysis of electromagnetic transients using oscillographic data was presented. The method was evaluated with
actual data and the obtained results reveal that it is independent
of the voltage rate of the transmission lines.
As a byproduct, the method provides the beginning and
end time of the disturbances, but due to energies windowing
process, the obtained beginning time is backward of the actual
value for about one cycle of fundamental frequency.

[1] S. Santoso, E. J. Powers, W. M. Grady, and P. Hofmann, Power quality

assessment via wavelet transform analysis, IEEE Trans. On Power
Delivery, vol. 11, no. 2, Apr 1996.
[2] F. Mo and W. Kinsner, Wavelet modelling of transients in power
systems, IEEE Conference on Communications, Power and Computing,
pp. 132137, May 1997.
[3] I. Daubechies, Ten Lectures on Wavelets. Philadelphia, USA: CBMSNSF Regional Conference Series, SIAM, 1992.
[4] S. G. Mallat, A theory for multresolution signal decomposition: The
wavelet representation, IEEE Transaction on Pattern Anal. Machine
Intell., vol. 11, pp. 674693, July 1989.
[5] C. H. Kim and R. Aggarwal, Wavelet transform in power system: Part
2 examples of aplication to actual power system transients, Power Eng.
Journal, pp. 193202, August 2001.
[6] K. M. Silva, B. A. Souza, and N. S. D. Brito, Fault detection and
classification in transmission lines based on wavelet transform and ann,
IEEE Transaction on Power Delivery, vol. 21, no. 4, pp. 20582063, Oct.
2006.
[7] C. Sankaran, Power Quality, C. PRESS, Ed., 2002, vol. 1.
[8] S. R. S. Baki, M. Z. Abdullah, and A. F. Abidin, Combination
wavelets and artificial intelligent for classification and detection transient
overvoltage, 2002 Student Conference on Research and Development
Proceedings, 2002.
[9] M. Solanki, Y. H. Song, S. Potts, and A. Perks, Transient protection
of transmission line using wavelet transform, Proc. IEE 7th Int.
Conf. Developments in Power System Protection, no. 479, pp. 299302,
Amsterdam, Netherlands, Apr. 2001.
[10] R. C. Dugan, M. F. McGranaghan, S. Santoso, and H. H. Beaty,
Electrical Power Systems Quality, 2nd ed., McGraw-Hill, Ed. New
York: McGraw-Hill, 2003.
F. B. Costa (M05) was born in Salvador, Brasil, 1978. He received his
BS and MS in Electrical Engineering from Federal University of Campina
Grande (UFCG), Brazil, in 2005 and 2006, respectively. He is currently a PhD
candidate at the same university. His research interest are electromagnetic
transients, power quality and fault diagnosis with wavelet transforms and
artificial neural network.
K. M. Silva (M06) was born in Joao Pessoa, Brazil, 1980. He received his BS
and MS in Electrical Engineering from Federal University of Campina Grande,
Brazil, in 2004 and 2005, respectively. He is currently a PhD candidate at the
same university. His research interest are electromagnetic transients, power
quality, fault diagnosis and power system protection.
K. M. C. Dantas (M04) was born at Campina Grande, Brazil, 1982. He
received his BS in Electrical Engineering from Federal University of Campina
Grande (UFCG), Brazil, in 2005. He is currently working toward in Master
degree at the UFCG. He is interested in electromagnetic transients, artificial
intelligence techniques applied to power systems and substations automation.
B. A. Souza (M02 - SM05) received his PhD in Electrical Engineering
from Federal University of Paraba, Brazil, in 1995. He works currently as a
professor at the Department of Electrical Engineering of Federal University
of Campina Grande, Brazil. His main research activities are on optimization
methods applied to power systems, electromagnetic transients, power quality
and fault diagnostic.
N. S. D. Brito (M05) was born in Antenor Navarro, Brazil, 1965. She
received her BS and PhD in Electrical Engineering from Federal University
of Paraba, Brazil, in 1988 and 2001 respectively. She received her MS from
Campinas State University, Brazil. She works currently as a professor at
the Department of Electrical Engineering of Federal University of Campina
Grande, Brazil. Her main research activities are on power quality, especially on
applications that involve fault detection and classification in electric systems.