Professional Documents
Culture Documents
3,July2013
WaveletFFNNBasedFaultLocation
EstimationofaTransmissionLine
MajidJamil*1,Md.AbulKalam2,A.Q.Ansari3,M.Rizwan4
1,2,3DepartmentofElectricalEngineering,JamiaMilliaIslamia,NewDelhi,India,
DepartmentofElectricalEngineering,DelhiTechnologicalUniversity,Delhi,India
*1majidjamil@hotmail.com;2kalam.a@rediffmail.com;3aqansari@ieee.org;4rizwaniit@yahoo.co.in
Abstract
In order to maintain the quality power supply and reliable
operation of power delivery service to the utilities with
minimum interruption, it is extremely important that the
transmissionlinefaultsneedtobedetectedandlocatedina
reliableandaccuratemanner.Tomeettheserequirementan
approach based on discrete wavelet transform using neural
network (NN) is proposed to estimate the fault location of
the three phase transmission line and the results have been
compared with the conventional method (measuring the
sequencecomponentsofvoltageandcurrent).Theproposed
approach is based on the measurements of the three phase
faultcurrentsignalsatoneterminalofthetransmissionline
(at relaying point), created for different fault operating
conditionsusingMATLABsoftware.Toaccomplishthetask,
features of these measured signals are extracted in terms of
summation of ninth level detail coefficients using multi
resolution analysis (MRA) based on discrete wavelet
transform,selectingdaubechies4(db4)asmotherwaveletat
100kHzsamplingfrequency.Thesefeaturesarethenusedas
an input to train and test the feed forward neural network
(FFNN)toprovidefaultlocationfromtherelayingpointon
the transmission line. The NN estimates the location of the
fault with a significant reduction in percentage error
compared to the conventional method of fault location for
wide variations of operating condition. Hence, the use of
multiresolution analysis (MRA) based on discrete wavelet
transform in combination with NN is quite promising for
faultlocationestimationofthetransmissionline.
Keywords
Transmission Line; Fault Current Signals; Wavelet Transform;
Fault Location; Feed Forward Neural network (FFNN); Ninth
LevelDetailCoefficients
Introduction
In power transmission line protection, faulty phase
identification and location of fault are the two most
important tasks which need to be addressed in a
reliable and accurate manner to ensure quality
performanceofthepowersystem.Sincedifferenttypes
of transient phenomena occurs on the transmission
lineduetowhichthefaultedtransmissionlineexhibits
both high frequency oscillations, localized impulses
superimposed on the power frequency components
and its harmonics. Therefore, from these transient
phenomena, faults on transmission lines may be
detected, classified, located accurately, and cleared as
fast as possible. Numbers of conventional and
intelligenttechniquesareavailableintheliteraturefor
thefaultclassificationandlocationofthetransmission
systems. However, fault classification and location
basedonconventionalmethodshavesomelimitations
in achieving the desired speed, selectivity and
accuracy as they do not have the ability to adapt the
dynamically to the variation of operating conditions
likesystemloadinglevel,faultinceptioninstanceand
fault resistance, etc. Therefore to overcome these
problems several intelligent techniques based
approaches have been implemented such as Fuzzy
logic based by [Biswarup et al.(2005), Mahanty et al.
(2007)], Artificial neural network (ANN) based by
Samantrayetal.(2007),Fuzzyneuralnetworkbasedby
Huisheng Wang et al.(1998) , Wavelet based by
[Chunju Fan et al. (2006), Osman H. et al. (2004), El
safty S. et al. (2009)], and combined WaveletANN
techniquesbasedby[SamiEkicietal.(2008),Chiradeja
P.etal.(2009),NanZhangetal.(2007),AbdollahiA.et
al. (2010), Meisam et al. (2011)]. Among these
approaches the Neuralnetwork based approaches
havebeenquitesuccessfulinidentifyingthetypeand
locationoffaultbutitrequiresaconsiderableamount
oftrainingeffortforgoodperformance,especiallyfora
wide variation of operating conditions as mentioned
above. In addition to a variation in operating
conditions, the accuracy of the fault classification and
locationalsodependsontheamplitudeofthedcoffset
and harmonics content in the transient signal of the
faultedtransmissionline.Therefore,fordetectionand
location of fault, it is pertinent to use waveform
analysis techniques like wavelet transform which has
been found to be an effective tool in monitoring and
77
ElectricalEngineeringResearchVol.1Iss.3,July2013
analyzingpowersystemdisturbancesincludingpower
quality assessment and system protection against
faultsasdiscussedbySamantarayS.R.etal.(2006).As
a matter of fact, most of the transient signals based
relay principle adopts the wavelet transform
technologytoconstructthealgorithms.
In this paper combination of discrete wavelet
transform (DWT) and neural network (NN) to locate
faultsonthetransmissionlineusingsingleendedpost
fault(sampledfaultcurrentsignals)currentsignalsfor
digitalrelayingpurposesisproposed,consideringthe
same network as modeled by Sukumar M.Brahama
(2007).The analysis produces very accurate fault
location results and it is observed that the mean
relative error using waveletFFNN based model is
around 3.2 % whereas the mean relative based on
conventional method (fault location measuring the
sequencevoltageandcurrent)isfoundaround9.36%
includingallcases of phaseandground faults. Hence
the proposed work can be considered as an
improvement over the conventional method of fault
locationonthetransmissionline.
This paper is organized as follows: First section
describes the introduction, the second section
represents the conventional method of fault location,
the third section describe the fault location
methodologyusingwavelettransformincombination
with neural network, the section fourth and fifth
describethesimulationstudies,resultsanddiscussion
respectivelyfollowedbyconclusionandreferences.
Computation of Fault Location using
Sequence Component of Voltage and
Current
l x
ComputedfaultlocationsforLG(linetoground),LLG
(Double line to ground), LL (line to line) and LLL
(symmetrical fault) for different fault operating
conditionsaredepictedinTABLE1.
Fault Location Methodology
The proposed fault location methodology scheme is
illustrated in FIG. 2. At first the three phase fault
current signals are obtained, considering the 100 kHz
sampling rate at 50 Hz base frequency, created for
different fault operating condition using MATLAB.
Then Multiresolution analysis (MRA) based on
discrete wavelet transform is used to obtain the ninth
leveldetailedcoefficientoffaultcurrentsignals(ia,ib
andic),selecting daubechies4 (db4) as mother wavelet
as it has good performance results for power system
analysis, discussed by Osman A.H. etal.(2004)]. Since
in the proposed fault location scheme selected
sampling frequency is 100 kHz, therefore ninth level
detailed coefficients (cD9) include seconed and third
harmonics which are predominant in case of
transmissionlinefaults.
Transientdisturbance
Recordingofthethreephasefaultcurrent
signals(ia,ibandic)atsendingendofthe
transmissionline
Waveletdecompositionofthefaultcurrent
signals
l
E s
ER
Fault
Featuresextractedintermsof
summationofninthleveldetails
coefficients
SendingEnd
Relay
Bus1
ReceivingEnd
Bus3
Bus2
InputtotheFFNN
FIG.1POWERSYSTEMMODEL
78
Faultlocation
FIG.2FAULTLOCATIONSCHEME
ElectricalEngineeringResearchVol.1Iss.3,July2013
TABLE1COMPUTEDFAULTLOCATIONRESULTSUSINGSEQUENCECOMPONENTSOFVOLTAGEANDCURRENT
Case
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
Types
of
fault
AG
BG
CG
ABG
BCG
ACG
AB
BC
AC
ABC
Actual
distance
tofault
from
relaying
point
(km)
Fault
resistance
(ohm)
225
245
105
240
5
200
15
35
15
130
285
295
65
230
255
280
125
250
280
295
1000
1000
50
100
40
40
50
50
1000
200
50
50
10
10
1
1
10
10
10
10
Fault
inception
Angle
Receiving
endphase
voltage
angle()
(degree)
(degree)
180
180
36
90
54
54
54
54
54
54
36
36
54
54
54
36
90
90
90
90
15
15
15
90
15
15
30
30
15
90
15
15
15
15
15
15
15
90
90
90
Sending
endzero
sequence
component
voltage
(atBus1)
Zero
sequence
component
voltageat
faultpoint
(atBus3)
Sending
endzero
sequnence
current
component
Magnid.
(p.u)
2.47E03
2.11E03
8.08E02
2.00E02
1.72E01
5.54E02
1.33E01
1.14E01
1.22E02
2.15E02
2.35E02
2.06E02
4.96E06
3.27E06
4.03E06
8.72E07
3.43E06
3.67E06
4.98E16
2.47E15
Magnid.
(p.u)
1.30E02
1.13E02
2.86E01
1.31E01
1.92E01
3.18E01
1.80E01
2.09E01
1.67E02
8.62E02
1.81E01
1.64E01
1.27E05
1.27E05
1.49E05
7.53E06
1.47E05
9.28E06
6.30E15
5.65E15
Magnid.
(p.u)
1.25E01
1.03E01
4.43E+00
1.04E+00
9.30E+00
2.98E+00
7.36E+00
6.21E+00
6.73E01
1.08E+00
1.29E+00
1.12E+00
2.79E04
1.14E04
9.78E05
6.49E05
2.10E04
6.00E05
4.59E14
3.02E14
Simulation Studies
To validate the proposed fault location estimation
approach, it is extensively tested on a 500 kV, 50 Hz
transmissionlineof300kmlengthconnectedbetween
twosourcesasshowninFIG.1andtheparametersof
thesystemareasfollows:
Sendingandreceivingendsourcevoltage
Es=500kV;ER=475 kV,where isreceivingend
sourcephaseangle;
Sendingandreceivingendpositiveandzerosequence
sourceimpedance:
Zs1=17.177+37.9404j;
Zs0=2.5904+12.2773j;
ZR1=15.31+38.2704j;
ZR0=0.7229+12.6073j;
Zero
sequence
impedance
between
Bus1to
Bus3(p.u)
Comput
edfault
location
distance
8.44E02
8.91E02
4.63E02
1.07E01
2.22E03
8.83E02
6.51E03
1.54E02
6.62E03
5.99E02
1.22E01
1.28E01
2.77E02
8.27E02
1.11E01
1.03E01
5.34E02
9.35E02
1.27E01
1.06E01
210.93
222.84
115.77
266.48
5.55
220.76
16.27
38.5
16.56
149.75
304.43
319.02
69.19
206.72
276.48
256.39
133.39
233.72
316.54
263.75
%
Error
(x)
(km)
6.25
9.04
10.26
11.03
11.09
10.38
8.49
9.99
10.38
15.19
6.82
8.14
6.44
10.12
8.42
8.43
6.71
6.51
13.05
10.59
Z0=0.37432+0.9429j/km;
C1=1.2097E8F/km;C0=7.4982E9F/km.
Insimulationstudies,aseriesoftestcasesarecreated
fordifferentvaluesoffaultresistancesvaryingfrom10
to1000forgroundfaultsandforphasefaults1and10,
variation in distance in steps of every 5 km is
considered from 5 km to 295 km. For simulation, a
permanent fault is created at faultinception anglesof
36, 54, 90 and 180 degree, which includes the zero
crossing of fault signals as well, considering 0.02 sec
simulationtimeatthesamplingfrequencyof100kHz.
For each fault operating condition, fault current
signals at the relaying point are measured in order to
be used for feature extraction using multiresolution
analysisbasedonDWTbyequations(1),(2)and(3).
Transmissionlineparameters:
Z1=0.15483+0.3050j/km;
79
ElectricalEngineeringResearchVol.1Iss.3,July2013
Theerroriscalculatedbyusingthefollowingformula.
The comparisons of fault location results between
conventionalmethodandbyapplyingneuralnetwork
in combination with wavelet transform for different
fault operating conditions have been illustrated in
TABLE2.
% Relative Error =
? 00
TABLE2COMPARISIONOFFAULTLOCATIONRESULTSBETWEENCONVENTIONALMETHODANDAPPLYINGNNIN
COMBINATIONWITHWAVELETTRANSFORM
Case
1
2
3
4
5
Types
of
fault
AG
BG
CG
6
7
8
9
10
11
12
13
14
15
16
17
ABG
BCG
ACG
AB
BC
AC
18
19
20
80
ABC
Actual
distance
tofault
from
relaying
point
(km)
Fault
resistance
(inohm)
Fault
inception
Angle
(degree)
Receiving
endphase
voltage
angle()
(degree)
Computed
fault
location
distance(x)
(km)using
sequence
components
ofvotage
andcurrent
%
Error
Estimatedfaultlocationusingfeaappliedasan
inputtotheFFNNturesvector
sa
Sb
Sc
Estimated
distance
(km)
%
Error
225
1000
180
15
210.93
6.25
0.0597
0.1322
0.1625
219.72
2.35
245
1000
180
15
222.84
9.04
0.0587
0.1319
0.1620
239.80
2.12
105
50
36
15
115.77
10.26
0.5731
1.9092
0.5847
107.25
2.14
240
100
90
90
266.48
11.03
0.2936
0.2898
0.3735
235.03
2.07
40
54
15
5.55
11.09
0.0049
0.0047
0.0342
4.89
2.20
200
40
54
15
220.76
10.38
0.0054
0.0031
0.0125
194.00
3.00
15
50
54
30
16.27
8.49
0.0279
0.0235
0.0088
15.51
3.40
35
50
54
30
38.5
9.99
0.0244
0.0217
0.0087
34.33
1.91
15
1000
54
15
16.56
10.38
0.0065
0.0065
0.0073
14.35
4.33
130
200
54
90
149.75
15.19
0.0206
0.017
0.0176
126.42
2.75
285
50
36
15
304.43
6.82
1.2801
0.5627
1.2876
281.57
1.20
295
50
36
15
319.02
8.14
1.2199
0.5729
1.2550
289.87
1.74
65
10
54
15
69.19
6.44
2.9418
0.5175
2.5280
64.51
0.75
230
10
54
15
206.72
10.12
1.5985
0.5394
1.4920
234.92
2.14
255
54
15
276.48
8.42
0.0067
0.0210
0.0166
250.43
1.79
280
36
15
256.39
8.43
0.5898
1.9720
1.5209
285.38
1.92
125
10
90
15
133.39
6.71
0.3981
0.0777
0.4707
127.48
1.98
250
10
90
90
233.72
6.51
0.3570
0.3460
0.2274
244.95
2.02
280
10
90
90
316.54
13.05
0.2454
0.3431
0.3114
283.12
1.11
295
10
90
90
263.75
10.59
0.2315
0.3292
0.2979
290.00
1.69
ElectricalEngineeringResearchVol.1Iss.3,July2013
Conclusion
Expert
Systems
with
Applications,Vol.34,pp.29372944,2008.
El Safty, S. and ElZonkoly, A., Applying wavelet entropy
principle in fault classification.Electrical Power and
EnergySystems,Vol.31,pp.604607,2009.
Fan,Chunju,Li,K.K.,Chan,W.L.andYu,Weiyong,Study
of protection scheme for transmission line based on
wavelet transient energy. Electrical Power and Energy
Systems,Vol.28,pp.459470,2006.
Grainger, J., Stevenson, W. Jr., Power system analysis
(McGrawHill,1994).
Ha, Hengxu, Zhang, Baohue and lv, Zhilai, A Novel
principleofsingleendedfaultlocationtechniqueforEHV
transmissionlines.IEEETransactionsonPowerdelivery,
Vol.18,No.4,pp.11471151,October2003.
Javad, Sadeh and Hamid, Afradi, A new and accurate
aided
neural
network
decomposition.
REFERENCES
wavelet
packet
based
transmissionlinefaultanalysismethod.ElectricalPower
andEnergySystems,Vol.31,pp.5969,2009.
Biswarup,DasandVittal,ReddyJ,Fuzzylogicbasedfault
classificationschemefordigitaldistanceprotection.IEEE
Systems,Vol.29,pp.669678,2007.
616,April2005.
faultclassificationapproachusingcurrentsamplesonly.
2007.
2003.
distributionsystemsusingwaveletsandneuralnetwork,
discrete
wavelet
transform
and
artificial
neural
2011.
Osman, A.H., Malik, O.P.,Transmission line distance
protection
based
on
wavelet
transform.
IEEE
TransactionsonPowerDelivery,Vol.19,No.2,pp.
andentropybasedfeatureextractionforlocatingfaulton
81
ElectricalEngineeringResearchVol.1Iss.3,July2013
TransactionsonPowerDelivery,Vol.19,No.1,pp.4955,
January2004.
No.2,pp.859869,April2007.
line
using
synchronized
phasor
82