Power Quality Distrubance
Power Quality Distrubance
Power Quality Distrubance
f
f m
t
f f
+ + 4 2
0ar#onics
( ) [ ]
n
n
n
t
f
h
+
4 sin
he Pure sine wa'e is with #agnitude IaJ and ti#e
.eriodItJ in #illisecond, and IthJ re.resents the .hase
angle shift. &oltage sag is a #o#entary 'oltage dro. in
#agnitude ranging fro# B.A.u to B.2.u. It has a ty.ical
duration fro# half a cycle to 2#in. ,n the other hand,
'oltage swell is a #o#entary 'oltage rise ty.ically
caused by unbalance fault fro# the ad!acent .hase. he
swell #agnitude is .ositi'e ranging fro# 2.2.u to
2.@.u. I(nJ is the #agnitude of the signal. :licker is a
.ower /uality disturbance where the 'oltage #agnitude
fluctuates o'er ti#e. It is ty.ically caused by electric
arc furnace and wind far#. he sinusoidal function
with fm flicker #agnitude and ff flicker fre/uency is
#odulated with the .ower syste# sinusoidal wa'efor#.
0ar#onic is a .eriodic .ower /uality disturbance that
is caused by nonlinear loads at the distribution syste#.
0ar#onic li#its are defined in IEEE standard 92A3
2AA4 and it is #easured as total har#onic distortion.
B. Wavelet Transform
he wa'elet transfor# is useful in analy6ing the
transient .heno#ena associated with trans#ission line
faults and Kor switching o.erations. he wa'elet
transfor# can be acco#.lished in three different ways
in signal analysis as CC, C( and DC. Co#.ared
with continuous wa'elet transfor#, the discrete wa'elet
transfor# is sufficient to deco#.ose and reconstruct
#ost .ower /uality .roble#s. It .ro'ides enough
infor#ation and offers high reduction in the
co#.utation ti#e
1) Discrete Wavelet Transform he discrete
wa'elet transfor# is one of the three for#s of wa'elet
transfor#. It #o'es a ti#e do#ain discriti6ed signal
into its corres.onding wa'elet do#ain. his is done
through a .rocess called Lsub3band codificationM,
which is done through digital filter techni/ues.
Dasically, the DC e'aluation has two stages. he first
consists on the wa'elet coefficients deter#ination. he
first stage gets the coefficients fro# this signal. :ro#
these coefficients, the second stage is achie'ed with the
calculation of both the a..ro=i#ated and the detailed
'ersion of the original signal, in different le'els of
resolutions, in the ti#e do#ain. *t the end of the first
le'el of signal deco#.osition, the resulting 'ectors will
be the le'el 2 wa'elet coefficients of a..ro=i#ation
and of detail. In fact, for the first le'el, these wa'elet
coefficients are called c*2 ;n< and cD2 ;n<, as stated
bellow in ;2< and ;4< res.ecti'elyF
( ) ( ) ( )
+ =
!
n ! hd n v n cA 4 . 2
N
;2<
( ) ( ) ( )
+ =
!
n ! gd n v n cD 4 . 2
N
;4<
Ne=t, in the sa#e way, the calculation of the
a..ro=i#ated ;c*4;n<< and the detailed ;cD4;n<<
'ersion associated to the le'el 4 is based on the le'el 2
wa'elet coefficient of a..ro=i#ation ;c*2;n<<. he
.rocess goes on, always ado.ting the Ln32M wa'elet
coefficient of a..ro=i#ation to calculate the LnM
a..ro=i#ated and detailed wa'elet coefficients. 0ere
we are using first le'el of deco#.osition.
22
Ca'efor# *c/uisition
:eature E=traction
Classifier using Neural Network
Classified ,ut.ut
Ca'elet ransfor#
Proceedings of ational Conference on Advanced Computing and CommunicationCACC!!, April"!#, $%!!
:ig. 4 shows .ower disturbance wa'efor# and
its detail. :ig. 4;a< shows the coefficients of detail le'el
D2 of a .ure sinusoidal signal. :ig. 4;b< re'eals the
coefficients of detail le'el 2 of a sinusoidal signal
ha'ing inter#ediate sag. (i#ilarly :ig. 4;c< and :ig.
4;d< shows the coefficients of detail le'el 2 of the
signals with inter#ediate swell and flicker. :ig. 4;e<
shows the coefficients of detail le'el 2 of the signals
with inter#ediate har#onics
.)
*a+
*b+
*c+
*d+
*e+
:igure 4 "agnitude of detail coefficients of DC ;a<
(inusoidal signal ;b< (inusoidal signal with
inter#ediate sag ;c< (inusoidal signal with inter#ediate
swell ;d< (inusoidal signal with inter#ediate flicker ;e<
(inusoidal signal with inter#ediate har#onics
*ll these signals were analy6ed using Ca'elet
ransfor#. 0ere we ha'e chosen Daubechies E at le'el
2 for the analysis .ur.ose. he gra.h .ertaining to
detail coefficients at le'el 2 i.e., D2 .ro'ides the
necessary infor#ation to .recisely locate the
occurrence of the disturbance. he sudden increase in
the 'alue of the coefficient in detail 2 indicates the
occurrence of a disturbance e'ent. E=tracting #ore
features fro# the deco#.osition wa'e can hel. in the
classification .rocess
". #eature E$traction
Now the signals with abo'e .ower /uality
disturbances are analy6ed by wa'elet transfor# and the
following fi'e distinct features inherent to different
ty.es of .ower /uality e'ents ha'e been e=tracted. he
Energy characteristics of each signal is re.resented in
able II
*DGE II. :I&E DI::E+EN )IND ,: ENE+GO
"E*(-+E(
Nor# 2
=
=
n
!
! % " e
2
4
2
2
4
Nor# 4
=
=
n
!
!
"
%
e
2
4 4
2
Entro.y
( )
=
=
n
!
! ! c " e
%
2
4 4
4 7
log
2
(tandard
de'iation
( )
4
2
4 E
2
=
=
n
!
!
"
%
e
*'erage
residual
( )
=
=
n
!
! " e
2
9
Ck is the Discrete wa'elet ransfor# ;DC<
for the sa#.les contained in the nth data
window
P is the #ean obtained fro# the DC of each
disturbance
ei is the different inherent features of DC
III. CG*((I:IC*I,N -(ING NE-+*G
NEC,+)
Neural networks ha'e been used in a board range of
a..lications 12?5. he basic architecture consists of
three ty.es of neuron layersF in.ut, hidden, and out.ut
layers. In feed3forward networks, the signal flow is
fro# in.ut to out.ut units, strictly in a feed3forward
direction. In the signal classification a..lication, the
training e=a#.les are classified according to their
energy signature 'alue. he energy signatures are
selected using the o.ti#i6ation techni/ue of Genetic
*lgorith#. he Network is trained and the Inco#ing
signal is 'erified with the trained neurons and the
classified out.ut is taken.
I&. +E(-G *ND DI(C-((I,N
A. Data &eneration
he si#ulation data was generated in "*G*D
based on the #odel in .a.er 12>5. :i'e classes of
different PQ disturbances, na#ed undisturbed sinusoid
;nor#al<, swell, sag, flicker, har#onics, were
considered. 2B2 cases of .ower /uality disturbance were
generated for training and another 49 cases of each class
were generated for testing. Doth the training and testing
signals are sa#.led at 4B .ointsKcycle and the nor#al
fre/uency is 9B 06. en .ower fre/uency cycles which
contain the disturbance are used for a total of 4BB .oints
B. 'imilarit( Measurement
he trained data are stored in the database. he
synthesis training wa'efor# is co#.ared with the
energy signatures of reference wa'efor#. -sing
#ini#u# distance #ethod the .ower disturbance is
classified. able III shows the sa#.le 'alues of energy
signature.
*DGE III. ENE+GO &*G-E ,: 0E DI(-+D*NCE(
24
Adhiparasakthi Engineering College, Melmaruvathur
Norm
,
Norm
-
'ntropy (tandard
$eviation
.verage
#esidual
(ine B.B2EE @.?A7> 3B.B>7E B.22A@ 37.994?
(ag B.B49A B.B@79 3B.22EB B.2>B? 3B.7?A7
(well B.B4>A B.2B22 3B.2294 B.2>7? 3B.7?A7
!licker B.74>E B.B>2> 34.>E@E B.9>AA 3B.7?A7
"armo
nics
B.7?A7 4.E>>7 39E.9?4 3B.2B99 >.2272
Energy signature #ethods considered in the .re'ious
sections are tested on the testing set of different .ower
/uality disturbance in this section. o test the effect of
detection infor#ation on classification of disturbance,
two different #ethods of feature e=traction and
detection based on :ourier transfor# are studied. *
total of 249cases are used for testing, so each e'ent has
49 cases. he features of 'oltage signal like standard
de'iation, entro.y, and co#bination of nor#s, a'erage
residual of the signal are calculated se.arately for each
ty.e of disturbances and finally the .erfor#ance of
different ty.es of feature e=traction is co#.ared here.
Classification results can be described in ter#s of a
confusion #atri=. he table I& shows the classification
result of .ro.osed neural network #odel.
*DGE I&. CG*((I:IC*I,N *CC-+*CO "*+I% ,:
NE-+*G NEC,+)
(ine (ag (well !licker "armonics
(ine 49 B B B B
(ag B 44 4 B 2
(well B B 4E 2 B
!licker B B 4 44 2
"armonic
s
B B B 4 47
his shows that the .ro.osed #ethod has the ability to
classify the disturbance #ore accurately. *ccording to
the results of able I&, it is shown that .ower /uality
disturbances of sine is classified .erfectly, but loses
one in swell and 4 in har#onics and 7 in sag and
flicker. . able & shows the .ercentage of recognition
rate for 9 .ower Quality disturbances, res.ecti'ely.
able & shows the total correct classification rates are
about A4.@Q. his shows that .ro.er selection of
energy signature will increase the accuracy with less
consu#.tion ti#e in neural network. :or that we use
genetic algorith# for selecting the energy signature
*DGE &. PE+CEN*GE ,: +EC,GNII,N +*E
%ower
&uality
'vent
#ecognition #ate
!!T ($ 'ntropy
and
(tandard
$eviation
/inimum
distance
method
%roposed
/ethod
(ine 2BB 2BB 2BB 2BB 2BBQ
(ag >B E@ >E ?> @@Q
(well ?> EE ?4 @@ A>Q
:licker EB >B >B ?4 @@Q
0ar#onics 4B ?4 @@ A4 A4Q
otal 9A.4 >E.@ ?>.@R @9.> A4.@Q
&. C,NCG-(I,N
In this .a.er, a new #ethod for the detection and
classification of Power Quality Disturbances in .ower
syste#s using wa'elet based Neural Network is
.resented. he results show that the .ro.osed #ethod
has the ability of detection and classification of different
.ower disturbance ty.es efficiently. his work leads us
to belie'e that the .ro.er wa'elet based energy
signature increases the accuracy of classification with
less co#.utation ti#e.
+E:E+ENCE(
125 (uriya )aewarsa, )itti *ttakit#ongcol, Cichai )rongkitsiri,
LCa'elet3Dased Intelligent (yste# for +ecognition of Power
Quality Disturbance (ignalsM, (.ringer Gink, GNC( 7A?4, ...
27?@327@9, 4BB>
145 Chei3"in Gin Chien30sien Cu Chia30ung Gin :u3(heng
Cheng, LDetection and Classification of "ulti.le Power3
Quality Disturbances Cith Ca'elet "ulticlass (&"M, IEEE
ransactions on Power Deli'ery, &ol. 47, No. E, ... 49?9 8
49@4, ,ct. 4BB@
175 (udi.ta Nath, *rinda# Dey and *bhi!it Chakrabarti,
LDetection of Power Quality Disturbances using Ca'elet
ransfor#M, World Academ( of 'cience) Engineering and
Technolog(, 'ol. EA, ... @>A3@?7, 4BBA
1E5 &aldo#iro &ega, Nelson )agan, Gabriel ,rdone6, Cesar
Duarte, L*uto#atic .ower /uality disturbance classification
using wa'elet, (u..ort 'ector #achine and artificial neural
network0 *+th ,nternational "onference on Electricit(
Distribution, Praga, @322 June 4BBA,Pa.er 2B4B
195 D. ). Panigrahi, P. ). 0ota, and (. Dash , LPower /uality
analysis using .hase correlated Ca'elet transfor#M, ,ranian
-ournal of electrical and com.uter engineering, 'ol. 7, no. 4,
su##er3fall 4BBE
1>5 O. Gu and E.(ty'aktakis, LDridge the ga.F(ignal .rocessing for
.ower /uality a..lications,M Electric Power '(stem /esearch,
'ol.>>, no.2, ..3@73A>, July 4BB7.
1?5 G.. 0eydt and *.C. Galli, Lransient .ower /uality .roble#s
analy6ed using wa'eletsM, ,EEE Trans. Power Deliver(, 'ol.
24, no. 4, ... AB@3A29, *.r. 2AA?.
1@5 Pedro ". +a#os, o#as +adil, :ernando ". Janeiro, *. Cru6
(erra2, LDe'elo.#ent and I#.le#entation of *lgorith#s for
Power Quality "easure#entsM, un.ublished.
1A5 ".)arthikeyan and &."alathi, LCa'elet 3 (u..ort &ector
"achine *..roach for classification of Power Quality
DisturbancesM, ,nternational 0ournal of /ecent Trends in
Engineering ,&ol 2, No. 7, "ay 4BBA
12B5 *le=andre *. Carniato, +uben D. Godoy, Joao ,nofre P. Pinto,
L)al#an :ilter and Ca'elets ransfor# Dased hree3Phase
Power Quality Disturbances Detection, Classification and
Diagnosis ool I#.le#entation 3 0ardware and (oftwareM,
,nternational "onference on /enewable Energies and Power
1ualit( ,&alencia ;(.ain<, 29th to 2?th *.ril, 4BBA.
1225 .O. Ji, S. Gu and Q.0. Cu, LDetection of .ower disturbances
using #or.hological gradient wa'eletM, 'cience Direct, in .ress
1245 ". (usha#a, G. ulasi +a# Das, *. Jaya la=#i, LDetection,
classification localisation, and controlling of 'oltage swells
using iu./c through wa'elet based neural networksM, 0ournal
of Theoretical and A..lied ,nformation Technolog(, ... 2@3
4@,4BB9
1275 Doris Di6!ak, Peter Planinsic, LClassification of Power
Disturbances using :u66y GogicM, 23E4EE3B2423>KB>KT4B.BB
U4BB> IEEE
12E5 (. "ishra, C.N. Dhende and D.). Panigrahi, LDetection and
classification of .ower /uality disturbances using ( transfor#
27
Proceedings of ational Conference on Advanced Computing and CommunicationCACC!!, April"!#, $%!!
and .robabilistic neural networkM, ,EEE Trans. Power Deliver(,
'ol. 47, no. 2, ... 4@B34@?, January 4BB@
1295 Jaeho Choi, 0ongkyun )i#, Jin#ok Gee and Gyo3Du# Chung,
L Power disturbance classifier using wa'elet3based neural
networkM, 0ournal of .ower Electronics, 'ol >, no E, ..37B?3
72E,4BB>.
12>5 +odney 0 G. tan, +a#achandara#urthy & ), LNu#erical
#odel fra#ework of .ower /uality e'entsM, Euro.ean -ournal
of scientific research, I((N 2E9B342>, E7;4B2B<, no.2, 7B3E?
2E