Earth Fault Distance Computation Methods

De
par
tme
nto
fEl
ect
ric
alE
ngi
nee
ringa
ndAut
oma
ti
on
T his disse rtat io n fo c use so n singl eph aset o

Eart
hFaul tD ist
anc e
Mo
al
t
eart hfaul ts in me dium vo ltage( MV)
h
-
D
d Rafi Adz
Computation Me tho ds
D1
dist ribut io n fe e de rs.Singl eph aset oe art h
89
faul tis th emo stc o mmo n faul tt ypein po w er
/
20
dist ribut io n ne tw o rks.I nF inland and in
1
4
Basedo
nT ransients in
mo sto fc o unt rie s in Euro pet heMV
man
ne tw o rks areo pe rat edw ithne ut raliso l
ated
or re so nante art h e d syst e ms.T h ew ay the
PowerDistribution Syste
ms
Eart
ne ut ralis c o nne c te d is impo rtantsinc eit
w illaffe c tth et ypeo f pro t e ction syst em
hFaul
ne ede d and po w e r syst e mc o mpo ne nt
tD
se l
e c t
io n.I n iso lat e d ne ut ralne tw orks e arth
ist
faul tc urre nt s arere lative l y smal lc o mpare d
anc
w ithe arth e d syst e ms.T h einit ialt ransie nts
eCo
ofe art hfaul t s areimpo rt antfo r faul t Mo
hdRa
fiAd
zma
n
mput
locat io ne spe c ially in une art h e d and
co mpe nsat ed ne ut ralne tw o rk.T h e
atio
elec tro magne tict ransie ntpro videt h e
n Me
avail abl einfo rmat io n fo re st imat ing faul t
t
locat io n.T his disse rtat io n pro po se s fault
ho
ds Base
locat io n algo rit h ms base do ne art hfaul t
transie ntsignal s me asure d atse co ndary
sideo f MV and L V dist ribut io nt ransfo rme rs
donT
in orde rt ol oc at ea singl el inet oe art hfaul t
ransie
in une arth e d and c o mpe nsat ed ne ut ralMV
dist ribut io n ne tw o rks.
nts in P
owe
rDist
ribut
ion Syst
ems
I
S BN978-
95 2
-60-
5973-0(pri
nted) BUSI
N E
S S+
9HSTFMG*afjhda+
I
S BN978-
95 2
-60-
5974-7(pdf) ECONOM Y
I
S SN-
L1799-4934
I
S SN1799-
4934(pri
nted) ART+
I
S SN1799-
4942(pdf) DESI
G N+
ARCH I
TECT
U RE
A
Aa
lt
oU ni
versi
ty
a
Sc
hoolofElect
ric
alE ngi
ne e
ring SCI
ENCE+
lt
oU
Depa
rtmentofElect
rica
lE ngi
neeri
nga
ndAut
oma
ti
on T
EC HNOL
O GY
www.
aalt
o.f
i
ni
CROSSOV
ER
ve
DOCTORAL
r
DOCTORAL DI
S SE
R T
ATI
O NS
sit
DI
S SE
R T
ATI
O NS
y
Aal
toUni
vers
it
ypubl
icat
ionseri
es
DOC T
O RALDISSE
R TATIONS1 89
/20
14
EarthF aultD ist

anc eComput ation
Me t
h ods Base donT ransients in
Pow erD ist
ribution Syst
e ms
Mo
hdRa
fiAd
zma
n
Ad octoraldi
ss ert
ati
onc o
mp let
e dforthed eg
reeo fDoct
oro f
Science( Tec
h no l
ogy)t
ob ed ef
e nded,wit
ht hepermi
ssi
o no ft
he
Aalt
oU nivers
ity,S choo
lofE l
ectric
alE ngi
neeri
ng,ata p
ub l
ic
examinationheldattheLectureh al
lS3o ft
h eschoolo
n8D ecember
2014at1 2no on.
Aa
lt
oU ni
versity
Sc
hoo lofE l
ec t
ric
alE ngineer
ing
Depa
rtme nto fElect
ricalEnginee
ringandAut
oma
ti
on
P
ow erS y
s t
ema ndH i
g hV o
lt
a geEngine
eri
ng
Super
vis
ingp ro
fessor
Pro
fes
sorM at
tiLeh
tonen
Prel
iminarye x
ami ne rs
Pro
fessorM ansour H.A bd
el-
R ah
man
Uni
versit
yofM anso ura
Egypt
Ass.P rof.Dr.I
ng .Pe
ter Brac
iní
k
UniversityofŽil
ina
Sl
ov ak i
a
Opp one nts

Ass.P rof.Dr.I
ng .Pe
ter Brac
iní
k
UniversityofŽil
ina
Sl
ov ak i
a
Dr.SeppoHänni
nen
VTTT e
chni
calResearc
hCe
ntreo
fFi
nland
F
inland
Aal
toUni
vers
it
ypubl
icat
ionseri
es
DOC T
O RALDISSE
R TATIONS1 89
/20
14
©Mo
hdRaf
iBi
nAd
zman
I
S BN9 78-952-6
0-5973-0( pri
nted)
I
S BN9 78-952-6
0-5974 -
7( pdf)
I
S SN-L1 7
9 9-
4934
I
S SN1 799-4934(pri
nted)
I
S SN1 799-4942(pdf)
ht
tp:
//urn.
fi
/UR N:
ISBN :
97 8-
9 52-
60-59
74-
7
Uni
graf
iaOy
Hel
sink
i201
4
F
inl
and
A
bst
rac
t
A
alt
oUni
ver
sit
y,P
.O.Bo
x11
000
,FI
-00
076A
alow
t ww.
aal
to.
fi
Autho r
Mo hd Rafi Adzman
Nameo fthed oc t
o raldi
s sert
a t
ion
EarthF aultDistanc eCo mput atio
n Metho ds Base
donTransie
nts in Po
w e
rDist
ribut
ion
Syst
e ms
Publishe rScho olo f El
ectric
alEngineering
UnitD epartme nto f Elect
ricalEnginee
ring
Se
riesAal
toUnive
rsit
y publication se
riesDOCTORALD I
SSERT ATI
ONS 1
89/
201
4
F
iel
do fr
esea
rc hPo wer Syst
e m and HighVol
tageEngine
ering
Ma
nus
cri
pts
ubmi
ttd25 August20
e 14 Da
teo
fth
ede
fe e8 D
nc ec
ember 2014
P
ermi
ssi
ont
opub
li
shg
rant
ed(
dat
e)4 No
vembe
r 20
14 La
nguageEnglish
Mo
nog
rap
h Ar
ti
cled
iss
ert
ati
on(
summa
ry+ o
rig
ina
lar
ti
cle
s)
A b s tra c t
T h emo stc o mmo n faul tt ypein MV dist ribut io n ne tw ork is singl el inet oe art hfaul t.T h e
init ialt ransie nt so fe art hfaul ts areimpo rtante spe cial ly fo r une art h e d and c ompe nsat ed
ne ut ralne tw o rks.T h ee art hfaul tt ransie ntsignal sc onsisto f many diffe re ntfre que nc y
co mpo ne nt s, w h ic hre sul tfro mc h arging and disc h arging o ft h ene tw o rk c apac it anc es.T h e
transie ntc o mpo ne nt s pro videval uabl einfo rmat io n fo r faul tlo cat io n purpo se s.T h ec h arging
co mpo ne nth as h igh e r ampl itudeand l o w er fre que nc yt h an t hedisc h argec ompo ne ntand
h enc eis mo resuit abl et obeuse d fo r faul tl o c at io n purpo se s.
Int his t h e sis, w edisc uss al go rit h ms t ol o cat ean e art hfaul tin une art he do rac o mpe nsat e d
ne ut ralMV ne t w o rks using t h einfo rmat iono ft h eme asure dt ransie ntsignal .T h ene tw o rks
co nside re d areassume dt oberadial lyo pe rat ed and t h ey aremo de led using El e ct ro magne tic
T ransie ntP ro gram- Al t
e rnat iveT ransie ntP ro gram ( EMT P - AT P) .F ivet ype so f faul tl o c atio n
algo rith ms h avebe en de ve lo pe dw h ic harec al le d ge ne ralmo de l(GM)al go rith m, e xac tmo de l
(EM)al go rit hm, c o nt inuo us w ave le tt ransfo rm ( CWT )base d me t ho d, mul tiplere gre ssio n
anal ysis ( MRA)base d me th o d and art ific ialne uralne tw o rk ( ANN) .GM al go rith m is de ve lo pe d
base do n a simpl ifie d mo de lo f symme t ric alc o mpo ne nt sw h ileEM al go rith m is de ve l
o pe dw ith
exac t" pi" -mo de lo f symme t ric alc o mpo ne nt s.Bo thal go rit h ms ut ilizet h efre que nc yo fc h arging
transie ntt oe stimat et h efaul tdist anc e .CWTbase d al go rit h m re quire s bo thvo ltageand
curre nto ft ransie ntsignal st oe stimat et h efaul tpat hinduc tanc e.MRA and NN al go rit h ms
w e rede ve l o pe d using t h et ransie ntsignalme asure d fro mt h ese co ndary sideo ft h eMV/ L V
dist ribut io nt ransfo rme r.I n addit io n, an al go rit h mt ofind a c o rre c tpat ht ow ards t hepo sitio n
of faul tin ne tw o rk w h ichh as many branc h e s is pre se nt e d.
T h ere sul t s fro m int ensivesimul at io ns and e xpe rime nt s in ac tualdist ribut io n ne tw o rks are
alsopre se nt e d in t his t he sis.T h ere sul ts areanal yz ed using signalpro c essing t e c hnique s.T h e
algo rith ms appl yc o nt inuo us w ave le tt ransfo rm ( CWT )tol o cat et h edo minantc h arget ransie nt
fre que nc y and e xt rac tt h espe cificc o effic ie ntc o rre spo nding t ot h ec harget ransie ntfre que nc y.
Int h is th e sis, t hepro pe rtie so fH il be rtt ransfo rmat io n( H T )areuse dt oe stimat et h edamping
atte nuat io no ft het ransie ntsignal .
F inal lyt h epe rfo rmanc eo ft h epro po se d faul tl o cat io n al go rit hms is e val uate d and t h ere sul t s
arec ompare d.Base do nt h esimul at io n re sul t s, itis fo und t h att h epro po se d algo rit hms w o rk at
a re aso nabl el e ve lo f ac curac y.T here sul t s fro m re ale xpe rime ntdat a sh o wt hatbo thCWTand
GM al go rit h ms h avea c omparabl ere sul t .
KeyworsEart
d hfault, faul
tl ocation, dist
ribut io
n network, co ntinuous w avel
ettransfo
rm,
chargetransie nt, neuralne tw ork, multipl
eregre ssion analysis, ATP-
EMT P
I
S BN(pr
int
ed )978- 952- 60 -5973-0 ISBN( pdf)978- 952-60 -5974-7
I
S SNL1
- 799-4934 I
S S N( pr i
nted )1799-4934 I
S S N( pd)1
f 799-4942
L
ocat
iono fp
ubl
ish
erHe
lsinki Loca
ti
o nofp r
int
ingH el
sinki Year2014
Pa
ges234 n h
ur t
tp:/
/urn.
fi/U RN:
ISBN:978-
952-60
-5974-
7
Preface
First of all, thanks and praise is due to almighty Allah, The Most Gra-
cious, The Most Merciful, who guide me in every step I take.
This work would have been impossible without the help and guidance
of several people, whose contribution I would like to acknowledge. First
of all, I would like to express my deepest gratitude to my supervisor Pro-
fessor Matti Lehtonen, for accepting and giving me this wonderful re-
search project. His supervision and his guidance both helped me to chan-
nel and specify the discussed ideas and at the same time provided much
appreciated freedom and support to explore new ways and concepts. I
am also thankful to the pre-examiners Professor Dr. Mansour H. Abdel-
Rahman and Assoc. Professor Dr. Ing. Peter Braciník for reading the
thesis manuscript and I appreciate their valuable comments.
I owe special thanks to Dr. David Topolanek for a useful discussions

and providing me a data used in this research and Mr. Tatu Nieminen
for front cover picture of my dissertation. Many thanks go to a research
colleague at room I344, Muzamir, M. Shafiq, Zoko, Michael, Ilhan, Ali,
Rizk, Gang Liu and to all the wonderful people at Department of Electri-
cal Engineering who have offered me a pleasant and encouraging working
environment.
I would like to express my appreciation for the financial support I re-

ceived from Universiti Malaysia Perlis (UniMAP), Ministry of Education
Malaysia (MOE), as well as research support from Aalto University, Grad-
uate School of Electrical Energy and Engineering, Ella and Georg Ehrn-
rooth Foundation and FORTUM Foundation.
I am personally want to thank all of my Malaysian friends and their
5
Preface
families for their outstanding support, encouragement and concerns during

my stay in Finland.
Last but not least, I want to express my heartfelt thanks to my beloved

family, wife Normiza Mohd Noor and children Nurin Yasmin, Mohammad
Haziq Fiqri and Mohammad Wafiq Ziqri for their love, patience and end-
less support. My sincere appreciations to my parents, Sumyati Sauji and
Adzman Ibrahim, parents-in-law, Endon Ibrahim and Mohd Noor Che
Chik, and my families in Malaysia for their love and prayers.
Vantaa, November 10, 2014,
Mohd Rafi Adzman
6
Contribution of the author
The author has developed the methodology presented and used in the
analyses and made all the analyses himself. The results and conclusions
are the sole work of the author.
7
Contribution of the author
8
Contents
Preface 5
Contribution of the author 7
Contents 9
List of Tables 13
List of Figures 17
List of Symbols 25
List of Abbreviations 31
1. Introduction 33
1.1 Background and Problem Statement . . . . . . . . . . . . . . 33
1.2 Research Objective . . . . . . . . . . . . . . . . . . . . . . . . 35
1.3 Research Contribution . . . . . . . . . . . . . . . . . . . . . . 36
1.4 Organization of Thesis . . . . . . . . . . . . . . . . . . . . . . 37
2. Transients of Single Phase to Earth Fault 39

2.1 The Discharge Transient . . . . . . . . . . . . . . . . . . . . . 39
2.2 The Charge Transient . . . . . . . . . . . . . . . . . . . . . . 41
2.3 Transient of Suppression Coil Current . . . . . . . . . . . . . 44
3. Fault Location Methods Based on Earth Fault Transient Sig-

nals 47
3.1 Method based on Frequency of Charging Transient . . . . . 47
3.2 Wavelet Method . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.3 Differential Equation Method . . . . . . . . . . . . . . . . . . 50
3.4 Least Square Fitting Method . . . . . . . . . . . . . . . . . . 50
3.5 Fourier Transform Method . . . . . . . . . . . . . . . . . . . . 51
9
Contents
3.6 Traveling Wave method . . . . . . . . . . . . . . . . . . . . . . 52

3.7 Artificial Intelligence Method . . . . . . . . . . . . . . . . . . 53
4. Determination of Transient Components with Signal Pro-

cessing Methods 55
4.1 Filtering of 50 Hz Component . . . . . . . . . . . . . . . . . . 55
4.2 Spectrum Analysis using Discrete Fourier Transform . . . . 56
4.3 Time-Frequency Analysis . . . . . . . . . . . . . . . . . . . . 57
4.4 Identification of Charge Transient Component . . . . . . . . 58
4.5 Analytic Representation and Damping Estimation Using Hilbert
Transform (HT) . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5. Fault Distance Estimation Using Transient Signals 65

5.1 General Model (GM) of Symmetrical Components . . . . . . 66
5.1.1 General Model 1 (GM1) . . . . . . . . . . . . . . . . . . 68
5.1.2 General Model 2 (GM2) . . . . . . . . . . . . . . . . . . 71
5.1.3 General Model 3 (GM3) . . . . . . . . . . . . . . . . . . 72
5.1.4 General Model 4 (GM4) . . . . . . . . . . . . . . . . . . 74
5.2 Exact Model (EM) of Symmetrical Components . . . . . . . . 77
5.2.1 Exact Model 1 (EM1) . . . . . . . . . . . . . . . . . . . 78
5.2.2 Exact Model 2 (EM2) . . . . . . . . . . . . . . . . . . . 80
5.2.3 Exact Model 3 (EM3) . . . . . . . . . . . . . . . . . . . 81
5.3 Continuous Wavelet Transform (CWT) . . . . . . . . . . . . . 83
5.4 Transient Signal Recorded from the Secondary side of the
MV\LV Distribution Transformer . . . . . . . . . . . . . . . 85
5.4.1 Pre-processing stage . . . . . . . . . . . . . . . . . . . 87
5.4.2 Multiple Regression Analysis (MRA) . . . . . . . . . . 89
5.4.3 Artificial Neural Network (ANN) . . . . . . . . . . . . 90
5.4.4 Correct path location in branches of feeders . . . . . . 92
6. Performance Evaluation of Proposed Fault Location Algo-

rithms Using EMTP/ATP Simulation Model 95
6.1 Simulated Network Model 1 . . . . . . . . . . . . . . . . . . . 96
6.1.1 Description of the simulation and the network model 96
6.1.2 Result Analysis and Discussion . . . . . . . . . . . . . 97
10
Contents

7. Performance Evaluation of Proposed Fault Location Algo-

rithms Using Transients Recorded in Real Networks 163
7.1 Case study 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.1.1 Description of Real Network Case 1 . . . . . . . . . . 163
7.2 Case study 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.3 Case study 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
8. Conclusions and Future Work 191
References 195
Appendix A. Mathematical Derivations and MATLAB Scripts 209

A.1 Derivation of Equation for General Model 1 Damped Natu-
ral Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . 209
A.2 Derivation of Equation for General Model 2 Undamped Nat-
ural Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . 210
A.3 Derivation of Equation for General Model 3 Damped Natu-
ral Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . . 211
A.4 Derivation of Equation for General Model 4 Undamped Nat-
ural Frequencies . . . . . . . . . . . . . . . . . . . . . . . . . 213
A.5 Example of Matlab Script for MRA model . . . . . . . . . . . 214
Appendix B. ATPDraw OH Line Configurations and Load Vari-

ation 215
B.1 ATPDraw OH Line Geometrical and Electrical Configurations215
B.2 Load Variation for Simulation Model 4 . . . . . . . . . . . . . 216
Appendix C. Tables of Simulation Results 217

C.1 Tabulated Results of Simulated Network Model 1 . . . . . . 217
11
Contents
12
List of Tables
5.1 Description of the composite components. . . . . . . . . . . . 67
6.1 Test results of single line to earth fault for 90◦ fault incep-
tion angle using the WT algorithm. . . . . . . . . . . . . . . . 138
6.2 Test results of single line to earth fault for 45◦ fault incep-
tion angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
6.3 Data generation for MRA and NN formulation. . . . . . . . . 150
6.4 Data generation for testing the MRA and NN algorithm. . . 150
7.1 Case 1 network parameters. . . . . . . . . . . . . . . . . . . . 167

7.2 Calculation results using CWT and GM2 algorithm. . . . . . 167
7.3 Case 2 network parameters. . . . . . . . . . . . . . . . . . . . 175
7.4 Calculation results using CWT and GM2 algorithm. . . . . . 182
7.5 Description of actual single line to earth fault experiment. . 185
7.6 Test result of single line to earth fault using recorded signal. 189
B1 Simulation model 4 data. . . . . . . . . . . . . . . . . . . . . . 216
C1 Comparison of simulation test result to the theory of earth

fault transient using GM1 equations. . . . . . . . . . . . . . . 217
C2 Comparison of the distance estimates obtained using GM
algorithms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
C3 Comparison of simulation test result to the theory of earth
fault transient with the effect of fault resistance using GM1
equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
C4 Comparison of the distance estimates obtained of the GM
algorithm methods with effect of fault resistance. . . . . . . 220
C5 Comparison of simulation test result to the theory of ground
fault transient with effect of fault inception angle using GM1
equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
13
List of Tables

algorithm methods with the effect of fault inception angle. . 221
C7 Comparison of simulation test result to the theory of ground
fault transient with the effect of composite loads using GM3
equations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
algorithm methods with effect of composite loads. . . . . . . 223
C9 Comparison of the distance estimates obtained between EX1
and GM2 algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 224
C10 Comparison of the distance estimates obtained between EM1
and GM2 algorithm with the effect of fault resistance. . . . . 225
and GM2 algorithm with the effect of fault inception angle. . 225
and GM2 algorithm with the effect of resistive loads. . . . . 226
and GM2 algorithm with the effect of inductive loads. . . . . 227
and GM2 algorithm with the effect of loads of 0.9 power fac-
tor. The loads are located at the end of each feeders. . . . . . 228
and GM2 algorithm with the effect of loads of 0.5 power fac-
tor. The loads are located at the end of each feeders. . . . . . 228
C16 Comparison of the distance estimates obtained from multi-
ple regression (MRA) and neural network (NN) algorithm
with the effect of fault resistance. Fault inception angle is
90◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
C17 Comparison of the distance estimates obtained from multi-
ple regression (MRA) and neural network (NN) algorithm
with the effect of fault resistance. Fault inception angle is
45◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
C18 Comparison of the distance estimates obtained from GM al-
gorithms with the effect of fault resistance. Signal mea-
sured from MV/LV side and fault inception angle is 90◦ . . . . 231
C19 Comparison of the distance estimates obtained from GM al-
14
List of Tables
C20 Comparison of the distance estimates obtained from EM al-

C21 Comparison of the distance estimates obtained from EM al-
15
List of Tables
16
List of Figures
2.1 An example of the earth fault transient signal in MV over-

head line network. . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Discharge of faulty lines in isolated neutral network. . . . . 40
2.3 Charge of healthy lines in isolated neutral network. . . . . . 42
2.4 Equivalent circuit for charge transient calculation. . . . . . 42
2.5 An example of composite earth fault transient signal and its
CWT analsyis. . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.6 Charge of healthy lines in a suppression coil network. . . . . 45
2.7 Equivalent circuit model of transient of the suppression coil 46
3.1 Propagation of traveling wave in power network due to earth

fault. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.1 Example of the ouput of pre-processing of transient signal

using CWT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.2 Example of the output from pre-processing of current tran-
sient signals using CWT . . . . . . . . . . . . . . . . . . . . . 62
4.3 Example of instantaneous current amplitude, envelope and
the linear decaying of the extracted CWT coefficient of charge
component. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.1 The general model for earth fault transients. . . . . . . . . . 67

5.2 The basic idea of fault location algorithm using general model
of symmetrical components. . . . . . . . . . . . . . . . . . . . 68
5.3 The equivalent circuit for earth fault transient of general
model 1 (GM1) algorithm. . . . . . . . . . . . . . . . . . . . . 69
17
List of Figures

5.7 Flowchart for fault location algorithm using general model
of symmetrical components. . . . . . . . . . . . . . . . . . . . 75
5.8 Exact π-model equivalent circuit with damping included. . . 77
5.9 The equivalent circuit for earth fault transient of exact model
1 (EM1) algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 79
2 (EM2) algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 80
3 (EM3) algorithm. . . . . . . . . . . . . . . . . . . . . . . . . 81
5.12 Flowchart for fault location algorithm using exact π-model
of symmetrical component. . . . . . . . . . . . . . . . . . . . . 82
5.13 Flowchart for fault location algorithm based on CWT. . . . . 84
5.14 Flowchart for MRA and NN fault location algorithms based
on recorded transient signal of single MV/LV distribution
transformer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.15 An example of simulated transient signal recorded at MV
and MV/LV distribution transformer. . . . . . . . . . . . . . . 87
5.16 Characteristics of extracted CWT coefficient of transient sig-
nal recorded from MV/LV distribution transformer. . . . . . 88
5.17 Architecture of neural network for estimate the fault distance. 91
5.18 Flowchart for correct path location algorithm based on recorded
transients of MV/LV distribution transformers. . . . . . . . . 92
6.1 A schematic diagram and ATPdraw simulated network model

for earth fault simulation analysis in 20 kV unearthed medium
votage network. . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.2 Simulated transient voltage signal in the faulty phase when
a single phase to earth fault happens at a feeder in an ex-
ample of a network with overhead line. . . . . . . . . . . . . 98
6.3 Simulated transient current signal in the faulty phase when
a single phase to earth fault happens at a feeder in an ex-
ample of a network with overhead line. . . . . . . . . . . . . 99
6.4 Example of the output from pre-processing of voltage tran-
sient signals using CWT and Hilbert transformation. . . . . 101
18
List of Figures

sient signals using CWT and Hilbert transformation. . . . . 102
sient signals using CWT with the effect of fault resistance
(Rf =25 Ω, Rf =50 Ω and Rf =100 Ω). . . . . . . . . . . . . . . . 103
sient signals using CWT with the effect of fault resistance
(Rf =200 Ω, Rf =300 Ω and Rf =400 Ω). . . . . . . . . . . . . . 104
6.8 Example of scale-dependent energy percentage distribution
with Rf =25 Ω, Rf =50 Ω and Rf =100 Ω. . . . . . . . . . . . . 106
6.9 Example of scale-dependent energy percentage distribution
with Rf =200 Ω, Rf =300 Ω and Rf =400 Ω. . . . . . . . . . . . 107
6.10 Comparison of simulation test result to the theory of earth
fault transient. . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.11 Comparison of the distance estimates obtained using gen-
eral model (GM) algorithms for the simulated single line to
earth faults with varies fault distance. . . . . . . . . . . . . . 109
6.12 Results of calculated and estimated damped (fdc , fdm ) and
undamped (fnc , fnm ) frequencies with the effect of fault re-
sistance (Rf ). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
6.13 Comparison of results of calculated (fdc ) and estimated (fdm )
damped frequencies with the effect of fault resistance (Rf ). . 111
6.14 Comparison of results of calculated (fnc ) and estimated (fnm )
undamped frequencies with the effect of fault resistance (Rf ).112
6.15 Result of calculated (ζc ) and estimated (ζm ) damping factors
with the effect of fault resistance (Rf ). . . . . . . . . . . . . . 113
6.16 Comparison of results of calculated (ζc ) and estimated (ζm )
damping factors with the effect of fault resistance (Rf ). . . . 114
6.17 Distance estimates with the effect of fault resistance . . . . 115
6.18 Comparison of simulated transient signal and its scalogram
with the effect of inception angle. The fault distance is 4 km
and without fault resistance. . . . . . . . . . . . . . . . . . . 117
6.19 Distance estimates with fault inception angle . . . . . . . . . 117
6.20 Comparison of calculated damped frequencies with the ef-
fect of load and fault resistance . . . . . . . . . . . . . . . . . 118
6.21 Comparison of estimated damped frequencies with the ef-
19
List of Figures
6.22 Comparison of calculated undamped frequencies with the

effect of load and fault resistance . . . . . . . . . . . . . . . . 121
6.23 Comparison of estimated undamped frequencies with the ef-
6.24 Calculated and estimated damping factor (ζ) with the effect
of load and fault resistance. . . . . . . . . . . . . . . . . . . . 123
6.25 Distance estimates with the effect of 5 MVA load and fault
resistances for general model networks. . . . . . . . . . . . . 124
6.26 Distance estimates with the effect of 12 MVA load and fault
resistances for general model networks. . . . . . . . . . . . . 125
6.27 A schematic network diagram and its simulation model for
a single phase to earth fault in a distribution network. . . . 127
6.28 The effect of fault distance, fault resistance and fault incep-
tion angle on fault distance calulation . . . . . . . . . . . . . 129
6.29 Fault location performance with the effect of resistive load
located at the end of each feeder . . . . . . . . . . . . . . . . . 131
6.30 Fault location performance with the effect of inductive load
located at the end of each feeder . . . . . . . . . . . . . . . . . 132
6.31 Fault location performance with the effect of load with 0.9
PF and 0.5 PF located at the end of each feeder . . . . . . . . 134
6.32 A schematic diagram of 20 kV, 227.6 km, unearthed neutral
MV network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.33 ATPdraw circuit for the simulated network model 3. . . . . . 136
6.34 CWT FFT amplitude of MV/LV substation for fault test num-
ber 1 with fault inception angle of 90◦ and 45◦ . The arrow
indicates the substation with the highest amplitude. . . . . . 140
ber 7, 2, 8, 10 and 4 with fault inception angle of 90◦ . The
arrow indicates the substation with the highest amplitude. . 141
20
List of Figures
6.39 Test number 2 (11.70 km) correct path identification. . . . . 145

6.44 Schematic diagram of the simulated 20 kV ungrounded neu-
tral MV network. . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.45 ATPdraw circuit for the simulated network model 4. . . . . . 149
6.46 A comparison of estimated FFT amplitudes of extracted CWT
coefficient at 4 different MV/LV distribution transformers
(DT) with inception angle of 90 ◦ . . . . . . . . . . . . . . . . . 152
(DT) with inception angle of 60◦ . . . . . . . . . . . . . . . . . 153
(DT) with inception angle of 30◦ . . . . . . . . . . . . . . . . . 154
6.49 Comparison of the distance estimate errors obtained from
multiple regression (MRA) and neural network (NN) algo-
rithm with the effect of fault resistance (Rf ). Fault incep-
tion angle is 90◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
multiple regression (MRA) and neural network (NN) algo-
rithm with the effect of fault resistance (Rf ). Fault incep-
tion angle is 45◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
general model (GM1,GM2,GM4) algorithms with the effect
of fault resistance (Rf ). Signal measured from MV/LV sub-
station and fault inception angle is 90◦ . . . . . . . . . . . . . 158
general model (GM1,GM2,GM4) algorithms with the effect
of fault resistance (Rf ). Signal measured from MV/LV sub-
station and fault inception angle is 45◦ . . . . . . . . . . . . . 159
exact model (EM1-EM3) algorithms with the effect of fault
resistance (Rf ). Signal measured from MV/LV substation
and fault inception angle is 90◦ . . . . . . . . . . . . . . . . . . 160
21
List of Figures

exact model (EM1-EM3) algorithms with the effect of fault
resistance (Rf ). Signal measured from MV/LV substation
and fault inception angle is 45◦ . . . . . . . . . . . . . . . . . . 161
7.1 Compensated neutral network test diagram for test 1-2 . . . 164
7.2 Compensated neutral network test diagram for test 3-4 . . . 165
7.3 Isolated neutral network test diagram for test 5-8 . . . . . . 166
7.4 Recorded signals of single line to earth fault at 0.46 km. . . 168
7.5 Recorded signals of single line to earth at 10.40 km. . . . . . 169
7.6 Recorded signals of single line to earth fault at 14.20 km.
Test number 5 and 6. . . . . . . . . . . . . . . . . . . . . . . . 170
Test number 7 and 8. . . . . . . . . . . . . . . . . . . . . . . . 171
7.8 Example of output from pre-processing of test number 1
(0.76 km) transient signal using CWT analysis. . . . . . . . . 172
7.9 Example of output from fault distances algorithm based on
CWT for test number 1. . . . . . . . . . . . . . . . . . . . . . . 173
7.10 Results of fault distance estimation (lf ) using CWT algorithm.173
7.11 Lounais-Suomen Sahko OY network test diagram. . . . . . . 175
Test number 1 and 2. . . . . . . . . . . . . . . . . . . . . . . . 176
Test number 3 and 4. . . . . . . . . . . . . . . . . . . . . . . . 177
7.14 Recorded signals of single line to earth fault at 25.40 km
(test number 5) and 36.0 km (test number 6). . . . . . . . . . 178
7.15 Recorded signals of single line to earth fault at 36.0 km. . . 179
7.16 Example of CWT scalogram and energy percentage distri-
bution of test number 4 (25.4 km) transient signal. . . . . . . 180
7.17 Example of CWT extracted coefficients, FFT frequency spec-
trum of ectracted coefficent and, voltage and current linear
decaying of extracted CWT coefficient of test number 4 (25.4
km) transient signal. . . . . . . . . . . . . . . . . . . . . . . . 181
7.18 Fault distances calculated using CWT algorithm. . . . . . . . 182
7.19 Schematic diagram of Czech distribution network. . . . . . . 183
7.20 Photo taken during the experiment. . . . . . . . . . . . . . . 184
7.21 Recorded signals of experiments 1. . . . . . . . . . . . . . . . 186
22
List of Figures

7.26 Fault distances calculated using CWT algorithm. . . . . . . . 188
A1 The equivalent circuit for ground fault transient model 1 . . 209

B1 Simulation network model 3 ATPDraw OH configurations . 215

B2 Simulation network model 4 ATPDraw OH configurations . 215
23
List of Figures
24
List of Symbols
a Positive scale
am
0 Scaling (dilation) parameters
A Ampere
FFT amplitude of extracted CWT coefficient
Amplitude
b Time shifting factor
Time position
C Capacitance
Coefficient
Phase to phase capacitance
C(a, b) Wavelet coefficient
Ce Phase to earth capacitance
Ceq Equivalent capacitance
Cj,0 Zero sequence capacitance of sound line
Cj,p Positive sequence capacitance of sound line
Cl,n Capacitance per unit length of negative sequence system
Cl,0 Capacitance per unit length of zero sequence system
C l,0′ Zero sequence capacitance of faulty line located behind
of the fault point (F)
Cl,p Capacitance per unit length of positive sequence system
Cl,p′ Positive sequence capacitance of faulty line located behind
Cp Positive sequence capacitances of the network
C0 Zero sequence capacitances of the network
CV (a, b) Wavelet coefficient of voltage transient signals
CI (a, b) Wavelet coefficient of current transient signals
CE Phase to earth capacitance
ESC Sum of percentage energy coefficient associated to same scale
25
List of Symbols
f Frequency
fc Charge transient frequency(Hz)
Calculated transient frequency(HZ)
Natural frequency(Hz)
fd Discharge transient frequency (Hz)
fdc Calculated damped transient frequency(Hz)
fdm Estimated damped transient frequency (HZ)
fm Estimated transient frequency(Hz)
fnc Calculated undamped transient frequency(Hz)
fnm Estimated undamped transient frequency (Hz)
fs Sampling frequency(Hz)
f (t) Original signal
F Fault point
Fc Pseudo-frequency
F0 Center frequency
g Output of filter
i Phase current
îc Initial amplitude of charge transient
îe Uncompensated steady state earth fault current
I Current amplitude coefficient
Ic Charging current
Iω Complex wavelet coefficient of current
k Integer variable parameters
Sample number
Node
kam
0 Translation (time shift) parameters
l Fault distance
lcl Length of connected line
lf Fault distance
L Inductance
Total loads
Line
Lc Inductance of suppression coil
Ldt Inductance of distribution transformer
Leq Equivalent inductance
Lj,0 Zero sequence inductance of sound line
Lj,p Positive sequence inductance of sound line
Ll Fault path inductance
26
List of Symbols
Ll,n Inductance per unit length of negative sequence system

Ll,0 Inductance per unit length of zero sequence system
L l,0′ Zero sequence inductance of faulty line located behind
Ll,p Inductance per unit length of positive sequence system
Ll,p′ Positive sequence inductance of faulty line located behind
Lp Sum of the inductances of coupling transformer, suppression coil,
faulty line and fault
LL Inductance of the line
Inductance of composite loads
LT Substation transformer phase inductance
n Integer variable parameters
N Samples
r Neural network correlation coefficient
R Resistance
Rdt Resistance of distribution transformer
Rf Fault resistance
Rl,n Resistance per unit length of negative sequence system
Rl,0 Resistance per unit length of zero sequence system
Rl,p Resistance per unit length of positive sequence system
Rp Sum of the resistances of coupling transformer, suppression coil,
faulty line and fault
RL Resistance of composite loads
RT Resistance of substation transform
s Complex frequency
S Scale of extracted CWT coefficient
Root of the transient characteristic equation
Complex frequency
SC Percentage of the energy for each CWT coefficient
Scalogram
t Time
td Time difference between two consecutive peaks
T Period of fundamental frequency
ũ(t) Hilbert transform of signal u(t)
U Voltage amplitude coefficient
Uω Complex wavelet coefficient of voltage
U (t) Analytic signal
27
List of Symbols
Uv Analytic form of the extracted wavelet coefficient of the voltage transient

Ui Analytic form of the extracted wavelet coefficient of the current transient
v Phase voltage
Velocity of the travelling wave
Wave propagation velocity
v̂ Phase voltage peak value
ve Instantaneous phase to earth voltage
x Estimated fault distance
Y Admittance
Y012 (s) Line sequence admittance matrix in the complex frequency domain
Z Impedance
Z(ω) Frequency component of impedance
1
Zkk Positive sequence diagonal element in the k node
0
Zkk Zero sequence diagonal element in the k node
Z012 (s) Line sequence impedance matrix in the complex frequency domain
ZT Substation transformer impedance
αc Damping attenuation corresponding to charge transient
αd Damping attenuation corresponding to discharge transient
αm Estimated damping attenuation
β Slope
Coefficient
δ Damping attenuation
∆ Sampling period
Increment
ζ Damping factor
ζc Damping factor corresponding to charge transient
Calculated damping factor
ζd Damping factor corresponding to discharge transient
ζm Estimated damping factor
λ Local coefficient
φ Phase angle
ϕ Phase
Phase shift
Ψ Mother wavelet
ψ Mother wavelet
ω Frequency (rad/s)
ωc Damped charge transient frequency (rad/s)
ωd Damped discharge transient frequency (rad/s)
28
List of Symbols
ωdm Estimated damped transient frequency (rad/s)

ωf Fundamental angular frequency (rad/s)
ωnc Estimated undamped transient frequency for charge transient(rad/s)
ωnd Estimated undamped transient frequency for discharge transient(rad/s)
ωnm Estimated undamped transient frequency (rad/s)
ω0 Undamped natural frequency (rad/s)
1 Primary
Index of transient root
2 Secondary
Index of transient root
29
List of Symbols
30
List of Abbreviations
AC Alternating current
AM R Automatic meter reading
AN N Artificial neural network
AT P Alternative transient program
CW D Choi-Williams distribution
CW T Continuous wavelet transform
DC Direct current
DF T Discrete Fourier transform
DOM ’Domanin’ distribution substation
DT Distribution transformer
DW T Discrete wavelet transform
DW F Discrete wavelet filter
EM Exact model
EM T P Electromagnetic transient program
FD Frequency-dependent
FL Fault location
FT Fourier transform
GM General model
HT Hilbert transform
HV High voltage
Im Imaginary part of a complex number
IEDS Intelligent electronic devices
ln Natural logarithm
LV Low voltage
M LP Multilayer perceptron
M RA Multiple regression analysis
MV Medium voltage
N OV E ’Nove Mesto’ distribution substation
31
List of Abbreviations
OLE ’Olesna’ distribution substation

P OH ’Pohledec’ distribution substation
RM S Root mean square
ST F T Short time Fourier transform
WV T Wigner-Ville transform
32
1. Introduction
1.1 Background and Problem Statement
This thesis focuses on single phase to earth faults in medium voltage (MV)
distribution feeders. Single phase to earth fault is the most common fault
type in power distribution networks. In Nordic countries, about 50% to
80% of the faults are of this type [1, 2]. In Finland, about 70% of the per-
manent faults that occur in rural MV networks are caused by animals
and weather conditions, like thunder, snow, icing, storm, hard wind and
fallen trees [3]. One of the solution to reduce an earth fault in overhead
line feeders is by replacing the bare conductor to covered conductor type
of overhead line. However, the solution requires a huge amount of invest-
ment. Based on the statistics survey done by [4], a typical feeder construc-
tion in MV distribution network was a mixture of both covered conductors
and bare overhead lines on the same feeders. The share of covered conduc-
tors per feeder varies between 10% to 50% and all the permanent faults
at covered conductor feeders were earth faults [4].
In 2010, a fault statistic studied by utility company in Dresden Germany
shows that 75% of all faults are earth faults [5]. In Finland, over 80%
of the annual outage costs of customers are due to to faults in medium
voltage distribution networks [6]. In rural overhead networks, earth fault
currents are typically 10 - 40 A [7]. The small earth fault current is an
advantage to system operation and is unlikely to cause damage to the
power system components. However, the detection and localization of the
fault is difficult due to the small current that reduces the sensitivity of
the conventional relay which is normally based on currents and voltages
at power frequency.
Basically, electricity supply system differs from one country to another.
33
Introduction
In Finland and in most of countries in Europe the medium voltage (MV)

networks are operated with neutral isolated or resonant earthed systems
(i.e earthed via an arc-suppression coil also known as Peterson coil) [8].
The way the neutral is connected is important since it will affect the type
of protection system needed and power system component selection. In
isolated neutral networks earth fault currents are relatively small com-
pared with earthed systems. With unearthed or resonant earthed system,
the earth fault current is small and usually lower than the normal load
current. When earth fault happens, the outage cost are accumulated dur-
ing the time taken to locate and isolate the fault. In addition, a high fault
resistance might also contribute to the losses. It is not possible to de-
tect the high fault resistance using conventional feeder protection meth-
ods [9]. Therefore, it is very important to developed efficient methods for
locating the faults. In recent years, several earth fault location algorithms
in unearthed and compensated neutral networks has been developed and
proposed [10–15]. However, there is no single method that can be used
for locating a fault in all types of earth fault. Since the topology of distri-
bution network usually has many branches compared to the transmission
line, the task is even more tough since bare fault distance calculation
gives several possible places.
In general, fault location algorithms can be categorized to three types
[16], based on using fault generated signals, utilizing external signal sources
and knowledge-based approaches. The fault generated signals can fur-
ther be divided into three groups which are conventional [17–19], tran-
sients [20–23] and travelling waves [24–26]. Fault location algorithms
based on injecting an external signal e.g. with a pulse generator are still
being used by some of electric utilities nowadays. These algorithms use
the concept of analyzing the relationship between the transmitted and re-
flected pulse of the injected signal [27, 28]. In another development the
injected signal is use of a different frequency than are present in the net-
work [29]. Several papers have proposed the use of knowledge based ap-
proach such as artificial neural networks, fuzzy set theory, expert systems
and genetic algorithms for locating a fault in distribution system [30, 31].
The successfulness of these methods is highly depending on the amount
and quality of the supplied data.
The initial transients of earth faults are important for fault location
especially in unearthed and compensated neutral networks. The elec-
tromagnetic transient during a few milliseconds after the fault inception
34
Introduction
will provide the available information for estimating fault location. Even
though the transient may cause problems to the conventional protective
schemes it may work properly under transient protective scheme. The
transient disturbance recorder can be installed at the monitored network
to support the conventional signal measurement that is only based on the
root mean square (RMS) voltage and current of the power frequency com-
ponent. Nowadays, fault localization and detection based on fault tran-
sient signals in transmission and distribution network has been a subject
of keen interest to utility and researchers [32–35].
With the upgrading of the distribution network towards the concept of
“SmartGrids”, more measuring and recording equipment such as smart
meter, microprocessor digital relay, digital fault recorder and multi utility
controller will be installed in low and medium voltage distribution substa-
tions. In order to adapt and apply the concept of distribution automation
system in the network, those equipments are monitored and time syn-
chronized. Thus, the realization of this kind of system will open more op-
portunity to control, optimize and enhance the reliability of current sys-
tem. This will also affect the development and enhancement of current
protection system and fault location algorithms in power networks.
The term fault location can be represented as the determination of the
faulty feeder, branch or line section. It also can be used as a general
term when we discuss about fault distance computation or estimation.
This dissertation proposes fault location algorithms based on earth fault
transient signals measured at secondary side of MV and LV distribution
transformers in order to locate a single line to earth fault in unearthed
and compensated neutral MV distribution networks.
1.2 Research Objective
In order to reduce the outage costs and to satisfy the quality of supply
requirements, fault location algorithms have to locate fault as soon as
possible and with an efficient way. This can be achieved by utilizing the
information of initial transients of earthed faults. The idea of using tran-
sient signal as a basis of developing fault location algorithm is not a new
thing. However, the formulation used to calculate the exact fault distance
may be different. In addition, the implementation and the execution of
algorithm might also be dissimilar. The main objective of this research
is to developed accurate and efficient fault location algorithms for locat-
35
Introduction
ing a single line to earth fault in unearthed and compensated neutral MV

distribution networks. The developed fault location algorithms mainly
utilize the information extracted from measured earth fault transient sig-
nal in MV substation and also from secondary side of MV/LV transformer.
The work is intended to develop a variety of fault location algorithms.
The developed fault location algorithms can be used as a tools to locate
and estimate the fault distance of an earth fault in a faulty feeder.
1.3 Research Contribution
The contribution of the research can be divided into three main cate-
gories. In the first category, there are fault location algorithms based
on measured transient signals using general model(GM) of symmetrical
components, exact model (EM) of symmetrical components and continu-
ous wavelet transform (CWT) based algorithm. The proposed algorithms
mainly use measured transient signal from MV side of the faulty feeder,
for example from MV main substation. In GM and EM algorithm, 4 gen-
eral models (GM) and 3 exact models (EM) of symmetrical components
has been presented and tested using simulated earth fault. Due to the
limitation of information of line parameters in real networks, only GM2
and CWT algorithm were tested using real recorded data. Based on the
results, GM2 and CWT algorithms managed to produce an acceptable ac-
curacy in estimating the fault distance in a real network.
In the second category, were the algorithms based on earth fault tran-
sient signals measured from the secondary side of the MV/LV distribution
transformer (DT). Fault location algorithm based on multiple regression
analysis (MRA) and artificial neural network (ANN) was developed in this
category. In addition, an algorithm which can be used as indication to se-
lect the correct path to the fault position when earth fault happens in a
network that have several branches was developed.
In the third category, a technique to measure the damping factor of
charge transient signal was proposed, based on continuous wavelet trans-
form (CWT) analysis and Hilbert transformation. CWT analysis is a pow-
erful tool to visualize the non-stationary signal such as earth fault tran-
sient. In this case, the calculated damping factor is used to estimate the
undamped charge transient frequency which in turn will be used to esti-
mate the fault distance in the GM algorithm.
36
Introduction
1.4 Organization of Thesis
The thesis is organized into the following chapters:
Chapter 2 presents general knowledge of earth fault transient signals.

The creation of charge and discharge transient signals when earth fault
is established is also presented. In this chapter, transient of suppression
coil current is also discussed.
Chapter 3 presents some known fault location methods based on earth

fault transient signals. Methods based on frequency of charge transient,
wavelet transform, differential equation, least square fitting, Fourier trans-
form, traveling wave and artificial intelligence are presented.
Chapter 4 presents the determination of transient components with a

signal processing method. In this chapter, filtering of 50 Hz components,
spectrum analysis using discrete Fourier Transform (DFT) and Time fre-
quency analysis are presented. Furthermore, identification of charge tran-
sient component is discussed. Then, analytic representation and damping
estimation using Hilbert transformation (HT) is presented.
Chapter 5 presents the proposed fault distance estimation algorithms

in this thesis. The proposed general model (GM), exact model (EM) and
continuous wavelet transform (CWT) fault distance based algorithm are
presented in this chapter. In addition, the proposed fault location algo-
rithms based on transient signals recorded from secondary side of MV/LV
distribution transformer, which are MRA and NN fault location algorithm
are discussed. Furthermore, a fault path indication algorithm will also
presented.
Chapter 6 presents performance evaluation of the proposed fault loca-

tion algorithms using EMTP/ATP simulation model. In this chapter, the
performance of proposed algorithms is evaluated with four simulated net-
work models.
Chapter 7 presents performance evaluation of proposed fault location

algorithms using transients recorded in real networks. In this chapter,
three real case studies are presented.
37
Introduction
38
2. Transients of Single Phase to Earth
Fault
This chapter studies the transient signals which are generated from sin-
gle line to earth fault. In [36], a technical analysis of transient phenomena
associated with earth fault in a three phase system is given. According
to [36], the transients are caused by redistribution of conductor voltage
throughout the whole system, with the fault location as the origin of the
change. When single line to earth fault occurs, two different processes are
superposed [37–39]. Both charge and discharge transient processes start
at the same time but with different duration. Figure 2.1 shows a typi-
cal single line to earth fault transient in the voltage and current of the
faulted phase. The depicted signals were recorded in a 20 kV overhead
line network.
In this chapter, we provide general knowledge about the transients which
arise during the single phase to earth fault. In the following section, the
process of charge and discharge transient will be explained using lumped
model of two feeders and the earth fault at phase 3 of faulty feeder as
shown in Figure 2.2 and Figure 2.3.
2.1 The Discharge Transient
In Figure 2.2, when earth fault is established at phase 3, the electric

charge stored in the earth capacitance of faulty phase of faulty and healthy
feeder has to be drained off starting at the fault location. The discharge
transient will propagate in both directions to the end of line with the speed
of traveling wave at certain characteristic impedance of the line. Then,
the traveling waves are reflected at the end of the line respectively. The
impedance of the faulty and healthy line, fault impedance at the fault lo-
cation and the grounding resistance are the important parameters which
affect the behavior of the discharge transient [37].
39
Transients of Single Phase to Earth Fault
4
x 10
4
3 Ua
Ub
2
Voltage [V]
Uc
1
0
−1
−2
−3
−4
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
Sample Number [k] x 10
4
(a) Phase voltage
600
Ia
Ib
400 Ic
Current [A]
200
−200
−400
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
(b) Phase current
Figure 2.1. An example of the earth fault transient signal in MV overhead line network.
Healthy feeder (Background network) L1
LOAD B
C L2
HV/MV C L3
ZT1 L1
ZT2 L2
N Ce Ce Ce
ZT3 L3
Faulty Feeder Ic3B L1

LOAD A
C L2
C L3
Rf Ce Ce Ce
If
Ic3A
Ic3B
Figure 2.2. Discharge of faulty lines in isolated neutral network.
According to [38], the frequency of the discharge transient can be calcu-

lated as:
v
fd = (2.1)
4lf
where, v is the velocity of the traveling wave and lf is the fault dis-
tance. According to [16], due to the influence of the connected line at
the healthy feeder which contribute to the discharging of the capacitance,
40
Equation (2.1) has to be modified and the discharge transient can be cal-
culated as:
v
fd = (2.2)
4(lf + lcl )
where, v is the velocity of the traveling wave, lf is the fault distance and
lcl represent the length of the connected line. Generally, the discharge
transient components are short duration high frequency components. The
discharge transient components are highly damped and disappear much
quicker than the charge transient component. Based on the real tests
conducted in [40], the amplitude of discharge component was typically 5
to 10% of the amplitude of charge component and their frequencies var-
ied through the range of 500 to 2500 Hz for distribution network. Based
on the Equation (2.1) and Equation (2.2), the fault distance significantly
affects the frequency of the discharge transient. In HV transmission net-
works, the discharge transient frequency is about 300 to 5000 Hz [16].
2.2 The Charge Transient
Due to the ground fault, the voltage of the sound phases rapidly increases
and subsequently increase the charge stored in their capacitances. In
Figure 2.1, rapid increase of two sound phases right after the inception
of the earth fault can be seen clearly. Figure 2.3 show the illustration of
charging process of healthy phases in isolated neutral network. In the
process, all the capacitive charging currents (Ic ) of the healthy and faulty
feeder have to flow over the fault location back to the transformer in phase
3. Assuming an earth fault is located at high voltage or medium voltage
substation, according to [37, 38], based on the simplified lump circuit as
shown in Figure 2.4, the frequency of charge component in undamped
condition can be calculated as:
Leq = 1.5LT (2.3)
Ceq = 2(C + Ce ) (2.4)

1 1
ωc = = (2.5)
Leq Ceq 3LT (C + Ce )
where, LT is the substation transformer phase inductance, C is the phase
to phase capacitance and Ce is phase to earth capacitance of the network.
Equation (2.5) is used when the fault happens at the busbar or when the
inductance of the transformer is much higher compared to the line in-
41
ductance. The faulty line inductance (Lf ) can be added to transformer

inductance (LT ) in Equation (2.5) for more exact solution.

C
LOAD B
L2
HV/MV C L3
ZT1 L1
ZT2 L2
N Ce Ce Ce
ZT3 L3
Ic1B
LOAD A
C L2
C L3
Rf Ce Ce Ce
If
Ic1A
Ic2A
Ic1B
Ic2B
Figure 2.3. Charge of healthy lines in isolated neutral network.
The initial amplitude of the charge transient in an undamped case can

be calculated as:
ve
îc = = ve ωc Ceq (2.6)
ωc Leq
1.5LT
2(C+Ce)
Figure 2.4. Equivalent circuit for charge transient calculation.
Assuming that the fault happens at the instantaneous phase voltage

maximum, then the transient amplitude is calculated as in Equation (2.7):
Ceq ωc
îc = îe (2.7)
3Ce ωf
where ωf is the fundamental angular frequency, ve is the instantaneous

phase to earth voltage at the fault moment and îe is the uncompensated
steady state earth fault current [40].
42
Figure 2.5 shows an example of spectral information of earth fault tran-

sient voltage and current signals recorded from general model of three
sequence component networks. The general model will be discussed in
detail in Chapter 5. The spectral information is obtained using continu-
ous wavelet transform (CWT) analysis.
4
x 10
1
Voltage[V]
0.5
−0.5
Fault Inception
−1
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
0.1
3250 Discharge transient component
0.09
1625 0.08
Charge transient component
Frequency[Hz]
1083.33 0.07
0.06
812.5
0.05
650 0.04
541.667 0.03
0.02
464.286
0.01
406.25
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time(s)
(a) Composite earth fault transient voltage (above) and its CWT Scalogram(below)
100
Current[A]
50
−50
Fault Inception
−100
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
0.1
Discharge transient component
3250 0.09
1625 0.08
Charge transient component
Frequency [Hz]
0.07
1083.33
0.06
812.5
0.05
650 0.04
0.03
541.667
0.02
464.286
0.01
406.25
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Time(s)
(b) Composite earth fault transient current (above) and its CWT Scalogram (below)
Figure 2.5. An example of composite earth fault transient signal and its CWT analysis.
The red box referred most energy of charge transient component while the
black box is for discharge transient component.
In Figure 2.5, it shows that the charge transient component has lower
frequency and longer duration than the discharged transient component.
43
The time-frequency localization of the earth fault transient signals shows

clearly the difference in terms of frequency and duration between these
two transient components. The scalogram in Figure 2.5 represents the
percentage of energy for each coefficient of the CWT analysis. The color
bar in the figure represents the level of percentage of energy calculated.
From the color bar the white color has the lowest energy while the red
color shows the highest. It is found that the charge transient compo-
nents carry much more energy than the discharge transient components.
The energy is at its highest level at the beginning of the fault inception
and then decreases towards the end of the signal. From the CWT analy-
sis, both analyzed signals gave almost similar information with regard
to the charge and discharge component. CWT analysis of simulated and
recorded real earth fault transient signals will be discussed further in
detail in Chapter 6 and Chapter 7.
Since charge transient components have lower band of transient fre-
quency and longer compared to discharge transient components they are
more suitable for the purpose of fault location and fault detection. Based
on real measurements recorded in distribution and transmission systems,
the typical value of charge transient is in the range of 100 - 800 Hz and
80 to 600 Hz, respectively [7, 16].
2.3 Transient of Suppression Coil Current
In resonant earthed network, also known as compensated neutral net-

work, a Petersen coil is connected between the neutral point of the power
system and the earth. In this case, in addition to the discharge and charge
transients, we also have to consider the current of the suppression coil
circuit as shown in Figure 2.6. According to [7], based on the fault loop
impedance of the suppression coil, coupling transformer winding, faulty
line length and the fault, the equation for the voltage becomes:
di
Lp + Rp i = v̂sin(ωf t + φv ) (2.8)
dt
where i is the current that is composed of transient and permanent cur-

rent, v̂ is the phase voltage peak value, ωf is the fundamental angular
frequency, φv is initial phase angle, and Lp and Rp represents sums of the
inductances and resistances of the suppression coil, coupling transformer
winding, faulty line length and fault.
44
LOAD B
C L2
HV/MV C L3
ZT1 L1
N ZT2 L2 Ce Ce Ce
ZT3 L3
Ic1B
LOAD A
Lc C L2
C L3
Rf If Ce Ce Ce
IL
Ic1A
Ic2A
Ic1B
Ic2B
Figure 2.6. Charge of healthy lines in a suppression coil network.
The current i in Equation (2.8), can be expressed as [7]:
t
v̂ (− )
i= [sin(ωf t + φv − φ) − e τ sin(φv − φ)] (2.9)
Z
where:

Z= (Rp )2 + (ωf Lp )2
ωf Lp
φ = arctan
Rp
Lp
τ=
Rp
In Equation (2.9), the transient signal includes fundamental frequency

component and exponential decaying DC component. The DC component
will be zero if the fault happens at the moment of zero of the permanent
current and in the maximum level at the moment of the permanent cur-
rent maximum. Using Thevenin’s theorem, the transient of suppression
coil is investigated using simplified equivalent circuit as shown in Fig-
ure 2.7 [7].
45
3Rf
E 3LC Ce
Figure 2.7. Equivalent circuit model of transient of the suppression coil [7].
The characteristic equation of the circuit in Figure 2.7 is given in com-

plex frequency domain as follow [7]:
S 2 + Sδ + ω02 (2.10)
The roots of the characteristic Equation (2.10) is given as:

S1,2 = −δ ± δ 2 − ω02 (2.11)
where:
1
δ=
6Rf Ce

ω0 = 3Lc Ce
where Lc is the inductance of the suppression coil, Ce is the total earth

capacitance per phase of the system and Rf is the fault resistance.
46
3. Fault Location Methods Based on
Earth Fault Transient Signals
Transient signals have been used and utilized for many purposed in power
systems. Protection system for example is one of the areas in power sys-
tems where researchers and engineers have studied the characteristics
of the transient signal for the purpose of fault location and detection.
The main idea is to improve the efficiency and reliability that can bring a
higher performance of the system itself. Actually, using earth fault tran-
sients for the purpose of fault location is not a new idea. However, the
idea and the technique proposed by the researchers for their algorithms
might be different by one and another. With new technologies which of-
fer better, faster, cheaper and more robust services than their outdated
technologies, studies on fault location becomes more challenging and in-
teresting. In this chapter, an overview of several existing ideas related
to single line to earth fault location algorithms using earth fault initial
transient signals is presented.
3.1 Method based on Frequency of Charging Transient
Based on studies done by [38], when earth fault happens in a network,

the fault line inductance Lf can be added to the transformer inductance
(LT ) as given in Equation (2.5) for more exact solution. The frequency of
the charging transient can be calculated using following equation:

1 1
fc = (3.1)
2π 3(LT + Lf )(C + Ce )
where LT is inductance of the substation transformer, Lf is inductance of

the line from transformer to the earth fault location, C is phase to phase
capacitance of the the whole network and Ce is phase to earth capacitance
of the whole network. The inductance of the line corresponding to the
47
Fault Location Methods Based on Earth Fault Transient Signals
location of the fault can be calculated by rearranged Equation (3.1) as

follow:
1
Lf = − LT (3.2)
3(2πfc )2 (C + Ce )
The method is simple since only one measurement is needed per substa-
tion transformer. Another advantage is that the fault distance at different
feeders can be estimated only by knowing their line inductance [41].
A research done by [42], studied fault location in an unearthed neutral
network with various types of RLC-circuit with and without damping as
a basis of fault location algorithm. In this work, the algorithm used a
measured damped transient frequency and a frequency shifting factor a
for estimating undamped transient frequency. The proposed algorithm
estimates the fault distance with the information of the damped and un-
damped transient frequency, estimated frequency shifted factor, network
parameters and its configuration. The algorithm required a small number
of iterations for estimating the inductance of the faulty line.
In [7], the author proposed a correlation of transient frequency to the
fault distance as the basis of fault location. The method requires only one
measurement per substation transformer for measuring the transient fre-
quency. Furthermore, a data base on network components, real time data
on network connections and a computer model are needed. In this work,
the single line to earth fault circuit was modelled completely. The accu-
racy of the method was tested with simulated and real data. A similar
work has been done by a previous researcher with a simple RLC-circuit.
However, the simplifications and and approximations reduce the accu-
racy of the methods. A parametric on-line fault location method for dis-
tribution MV networks is proposed by [15]. The method uses the upper
frequency of free oscillations caused by the faulted phase capacitance dis-
charge in the transient process of earth fault as the basis of fault location.
In order to produce a distance to frequency response, five single phase arc-
ing ground faults at difference distances were initiated using horn elec-
trodes. The method used discrete Fourier transform (DFT) to calculate
the frequencies of the generated transient signals. According to [15], the
fault distance to frequency response curve has to be corrected when there
is a significant change of load or network structure.
48
3.2 Wavelet Method
In case of a single phase to earth fault, the fault path inductance of the
faulty line is a composed of a series connection of zero, positive and nega-
tive sequence inductance as follows [40]:
1
Lf = (Ll,p + Ll,n + Ll,0 ) · l (3.3)
3
where, l is the fault distance and Ll,0 , Ll,p , Ll,n are the inductance per unit
length of zero-, positive- and negative-sequence systems.
A fault location algorithm based on wavelet transform was presented
by [43] to estimate the fault path inductance of the faulty line length us-
ing transient signal. According to [43], the fault path inductance can be
estimated with the following equation:

1 Uω (k∆t, f )
Lf = imag (3.4)
ω Iω (k∆t, f )
where Uω and Iω are the complex wavelet coefficients of voltage and cur-
rent and ω is the angular frequency. According to [44], the algorithm first
determines the maximum wavelet coefficient of the current including the
amplitude, frequency and location of the wavelet. Using this frequency
with different time translations, the equivalent fault inductance can be
calculated. The 2 ms inductance interval, corresponding to 10 subes-
timates, is then determined with the smallest standard deviation. The
mean value of the inductance, calculated in this interval, is finally used
to determine the fault distance.
Researcher in [16] deals with an earth fault in subtransmission net-
works. His work related to transient based ground fault location using
charge transient. He proposed of to use discrete wavelet filter (DWF), the
FIR filter with the Gaussian wavelet and amplitude correction in filtering
specific coefficient of transient component. Instead of using DWT in many
frequency scales, DWF only uses a single scale where the transient is of
interest. However, the method still required a pre-processing technique
to locate the specific transient frequency. This is done by using discrete
Fourier Transform (DFT). The equation of DWF is given as follows:
(n − b)/a)2
2 −( )+jω0 (n−b)/a
DW F [n] = √ x [n] · e 2δ 2 (3.5)
a 2πδ n
where, x[n] is a discrete function of the samples, a is scaling factor, b is
49
the translation, δ = 3/2 , ω0 = 5.336 rads−1 and n is an integer vari-

able parameter which represents discrete time. In Equation (3.5), the

scaling factor is calculated from sampling frequency (fs ) and the filtered
frequency of the charge transient (fc ) as a = fs /fc [16].
3.3 Differential Equation Method
Considering an earth fault in electrical power network, the network can

be modelled with a simple first order model of a RL-circuit. The voltage
and current signals of the faulty phase have the following relation [45]:
di
u(t) = Ri(t) + L (3.6)
dt
where u, i, R and L are the voltage, current, resistance and the inductance
of the faulty phase, respectively. According to [45], since the differentia-
tion is sensitive to higher frequency noise, the solution of above equation
is done by using the integration technique. With samples of voltage and
current signal, inductance L is calculated by using three equally spaced
pairs of voltage and current using trapezoidal rule as follows:
∆t ((ik+1 + ik )(uk+2 + uk+1 ) − (ik+2 + ik+1 )(uk+1 + uk ))

L= (3.7)
2 ((ik+1 + ik )(ik+2 − ik+1 ) − (ik+2 + ik+1 )(ik+1 − ik ))
An earth fault transient signal has been applied to differential equation

method by Schegner [46]. The total length of the time window was 16 to
24 samples, with sampling frequency of 10 kHz. Differential-equation al-
gorithms work in theory for all the voltage and current components which
satisfy Equation (3.6). The best result is, however, obtained if all the other
frequencies are first filtered out, except the charge transient [40].
3.4 Least Square Fitting Method
According to [47], considering the first-order line modelled with capaci-

tances neglected, the parameters of voltage and current transient wave-
form can be model directly using least square fitting method in time do-
main as follows:
i(t) = I1 es1 t + I2 es2 t (3.8)
u(t) = U1 es1 t + U2 es2 t (3.9)
50
where, s1,2 = δ ± jωc is the complex frequency of the charging transient.

For oscillating transients, the amplitude coefficients U1 , U2 , and I1 , I2 ,
are also pairs of complex conjugates. The voltages and currents of a line
written in complex frequency domain have the following relation:
Us = RI(s) + sLI(s) (3.10)
By separating the sub-components s1 and s2 , we obtain the following

equation:
U1 = RI1 + s1 LI1 (3.11)
U2 = RI2 + s2 LI2 (3.12)
Then, the fault distance is solved as reactance of the faulty line length:
U 2 I1 − U 1 I2
L= (3.13)
I1 I2 (s2 − s1 )
For proper computation, least square fitting methods require signal fil-
tering and pre processing. Equation (3.13) is the basic equation based on
first order line model. For better result, the model can be replaced with
a higher order model which includes capacitances at the close end of the
line [47].
3.5 Fourier Transform Method
An algorithm based on Fourier transform method was suggested by [48].

The basic idea is to solve the line impedance in the frequency domain.
In case of first order line model, the reactance of the faulty line length
is obtained directly as the imaginary part of the impedance calculated
from the corresponding frequency spectrum components of currents and
voltages of the faulty phase [48]. The distance to earth fault location is
obtained as follows:
U (ω)
Z= (3.14)
I(ω)
3Im(Z(ω))
l= (3.15)
(Ll,0 (ω) + 2Ll,p (ω))ω
where, Z(ω), U (ω) and I(ω) are the frequency (ω) component of impedance,
measured voltage and current, Ll,0 and Ll,p are the inductance per unit
51
length of zero- and positive-sequence per length of the faulty line. As

explained in [48], the fault distance is calculated as a weighted average
of the estimates made for the n dominating frequencies in the spectrum.
Also a higher order model, which allows for the phase to earth capaci-
tances, is presented in [48].
3.6 Traveling Wave method
Researchers have long been studying traveling wave transients for fault
location and fault detection purposes [49–53]. A fundamental concept of
using traveling waves when earth fault happens in the network is de-
scribed in Figure 3.1.
Fault location
Amplitude
T1
T2
Time
Figure 3.1. Propagation of traveling wave in power network due to earth fault.
When earth fault happens in a power network, it will create a transient

signal which propagates along the line as waves [54]. The waves will
travel across the line towards line ends and then a reflection will occur.
The process will continue until it diminished. The fault distance can be
estimated precisely by measuring the time difference between two consec-
utive peaks of waves and the knowledge of the wave velocity as given in
following equation [55]:
νtd
x= (3.16)
2
where, x is estimated fault distance, td = (t2 − t1 ) is the time difference

between two consecutive peaks and ν is wave propagation velocity. The
wave propagation velocity depends on the overhead line parameters and
the traveling waves’ propagation path [55–58].
A research done by [57] shows that there is a correlation between the
typical frequencies of generated earth fault transient and the specific path
52
of the network covered by the traveling waves originated by the fault.

Continuous wavelet transform (CWT) is used to identify the typical fre-
quencies. Fault location methods based on traveling waves have better
performance compared to the conventional and other transient fault lo-
cation methods [16]. However the method usually requires a high sam-
pling frequency in the range of MHz and the actual data need to be time-
synchronized [16, 59–62].
3.7 Artificial Intelligence Method
Over the past few decades, a number of fault location algorithms based
on artificial neural networks (ANN) have been developed, including event
signal characteristics neural network approach [63–65], wavelet analy-
sis and neural network approach [66–68], radial basis function neural
network approach [69–71], support vector machine and neural network
approach [72–74] and fuzzy and neural network approach [75–77].
A transient based ANN algorithm for locating single line to earth fault
has been developed in [78]. In this work, Multilayer Perceptron (MLP)
with one hidden layer was used as ANN-structure and was trained with
Backpropagation method. The ANN-structure has only one hidden layer
which is sufficient for representing any given input-output transformation
[78, 79].
Two algorithms has been developed and presented in [14, 78]. The first
algorithm uses either the phase voltage or the phase voltage and current
samples as input data and the second algorithm uses harmonic compo-
nents of the neutral voltage transient as input data. The transient re-
quired filtering and pre processing technique before it can be used as an
input data. Both algorithms use fault distance as an output value. The
first algorithm was compared with conventional algorithms which are dif-
ferential and wavelet algorithm. According to the author [78], the perfor-
mance of ANN was comparable to the conventional algorithms.
53
54
4. Determination of Transient
Components with Signal Processing
Methods
Earth fault transient signal is non-stationary signal and it consists of

many frequency components. Therefore, determination of interesting part
of transient signal is very important and crucial since the fault location
algorithm is mainly based on the information of the specific transient com-
ponent. The error of the fault location algorithm could be high if the
wrong selection of transient component is used. In a real network, the
transient signal is recorded using a disturbance recording system that is
installed in substation. Nowadays, most of disturbance recording systems
are embedded in intelligent electronic devices (IEDS) that are aligned to
the protection, control, measurement and supervision of utility and in-
dustrial power distribution systems. A proper design of filters such as low
pass and anti aliasing filter that are used in IEDS is important in order
to maintain a correct information gathered from recorded transient signal
to be used in fault location algorithm. The transient frequency of inter-
est should fall inside the interested frequency band region. In order to
determine the information of the recorded transient signal, several signal
processing techniques have to be implemented. This chapter examines
a number of signal processing methods used to allocate and extract the
important part of the transient components.
4.1 Filtering of 50 Hz Component
Basically, the 50 Hz frequency component can be removed using comb

filters [80]. According to [40], the steady state of fundamental component
can be filtered using Equation (4.1) as follows:
g(t) = f (t) − f (t + T ) (4.1)
55
Determination of Transient Components with Signal Processing Methods
where g(t) is the output of the filter, f (t) the original signal and T is the
period of fundamental frequency. In case of the fundamental frequency is
50 Hz, T is 20 ms.
4.2 Spectrum Analysis using Discrete Fourier Transform
Discrete Fourier transform (DFT) is probably one of the best known meth-
ods to transform the signal from time domain into frequency domain. It
has been used for a decade to estimate the amplitude and phase of the fun-
damental frequency and associated harmonics of current and voltage [81].
The Fourier transform (FT) and discrete Fourier transform (DFT) is de-
fined as [82]:
∞
F T (f ) = x(t) · e−j2πf t dt (4.2)
−∞
N −1 2πkn
−j
DF T [k] = x[n] · e N (4.3)
n=0
x[n] = x(nTs ), n = 0, 1, 2, ..., N − 1
where t is time, f is frequency, n and k are integer variable parameters,

and x[n] is a sequence obtained by sampling the continuous time signal
x(t) every Ts seconds for N samples.
The fast Fourier Transform (FFT) is an algorithm used for computing
DFT with less computational effort and in an efficient way [83]. Both
forms have performed successfully for harmonic measurement and analy-
sis in many applications of power systems [84–93]. Even though, FFT
provide an efficient way to compute the transform digitally, the result of
the analysis is time averaged. The transformation contains only the av-
eraged information and does not provide any time-domain information
about the signal. It becomes inadequate when one is interested in the
local frequency contents of a signal or location specific features such as
charge and discharge transient component within the earth fault tran-
sient signal. In other words, it will not detect efficiently signals that have
sharp changes and it has the tendency to obscure transient [81,82,94–96].
Therefore, the analyzed signal of FFT must be stationary. In order to over-
come this problem a time frequency analysis is required.
56
4.3 Time-Frequency Analysis
A time frequency signal analysis is very important for interpretation of

transient signal components in both time and frequency. There are num-
ber of time-frequency methods available for decomposition of transient
signal into time-frequency plane such as short time Fourier Transform
(STFT) [97–100], Wigner-Ville transform (WVT) [101–104], Choi-Williams
distribution (CWD) [101, 105–107] and Wavelet transform (WT) [97, 98,
102, 108–111].
Among all of these methods, wavelet transform has emerged in recent
years as one of the most favoured tool used by researchers and engi-
neers for analyzing transient signals in the areas of protection systems,
fault location, power quality, power system transients, partial discharges,
condition monitoring, load forecasting and power system measurements
[112–120]. Wavelet transform can be divided into two categories which
are discrete wavelet transform (DWT) and continuous wavelet transform
(CWT). A DWT can be calculated as follows [121]:
1 n − kam
DW T (f, m, n) = m f (k) · ψ ∗ ( 0
) (4.4)
a0 k am0
where ψ is the mother wavelet, the asterisk in Equation (4.4) denotes

a complex conjugate, am m
0 , ka0 are the scaling (dilation) and translation
(time shift) parameters respectively, and k and m are integer variables.

In mutiresolution analysis, DWT usually works within a frequency of
band. In DWT, the analyzed signal will be decomposed into a discrete
number of logarithmic frequency bands [82]. Using DWT to locate and
measure the transient component frequency may be prone to errors when
the transient component is located at the boundary of the frequency band
[122]. Hence, the error from that measurement will affect the fault loca-
tion algorithm as well. Compared to DWT, instead of using a decompo-
sition of frequency band to locate the charge frequency components, we
prefer determining specific frequencies that will be used in fault location
algorithm. In next section, an identification of charge transient compo-
nent using CWT will be discussed.
57
4.4 Identification of Charge Transient Component
In this section, the identification of charge transient component is per-

formed using continuous wavelet transformation (CWT). The CWT pos-
sesses the capability of illustrating the frequency contents of the tran-
sient signal with respect to the time function. The main purpose that we
used CWT is first to identify the dominant transient frequency that cor-
responds to the charge transient component generated by the earth fault.
Second, after we identified the dominant frequency of the transient sig-
nal, the wavelet coefficients belonging to that dominant frequency will be
extracted which later will be used as the input of the fault location algo-
rithm.
Let us briefly explain the main features of this transform. According
to [43], the CWT is the inner multiplication of a wavelet family Ψa,b (t)
with the signal s(t) and it could be expressed as:
+∞
C(a, b) = s(t) · Ψa,b (t)dt (4.5)
−∞
In Equation (4.5), Ψa,b (t) are derived from a chosen function Ψ(t) called
"mother wavelet" given as [43]:
1 t−b
Ψa,b (t) = Ψ( )dt (4.6)
|a| a
The mother wavelet Ψ(t) must be short and oscillatory, and it must have
zero average and decay quickly at both ends [43]. In Equation (4.6), a and
b are the parameters for positive scale and time position (time shifting
factor). The Ψ in Equation (4.5) is known as the complex conjugate of Ψ
and the output of CWT would be the wavelet coefficient denoted as C(a, b).
In literature, several mother wavelets have been used to analyze transient
signals [8, 20, 22, 23, 43, 57, 123]. In [16, 43, 122, 124, 125], Morlet wavelet
has been applied to earth fault transient signal for the purpose of fault
detection and fault location. Therefore, on the basis of their investigation
results the Morlet wavelet has been chosen in this study.
The expression of Morlet wavelet is given in Equation (4.7) as follows:
2 /2)
Ψ(t) = e(−t ej2πF0 t (4.7)
In Equation (4.8), F0 is the center frequency of mother wavelet Ψ(t) and

∆ is a sampling period. According to [126], the pseudo-frequency Fc asso-
58
ciating to scale a, can be expressed as follows:
F0
Fc = (4.8)
a∆
Based on the output of the CWT, in Equation (4.9) SC represents the

percentage of the energy for each coefficient which is illustrated as so-
called ”scalogram” and can be computed as [126]:
abs(C(a, b)∗ C(a, b))

SC(a, b) = 100 (4.9)
sum(abs(C(a, b)∗ C(a, b)))
From the obtained scalogram, the sum of percentage energy coefficient

associated to the same scale ESC can be calculated as:
N

ESC (a) = SC(a, b) (4.10)
n=1
Then, the detection of charge transient frequency is performed by inspect-

ing the peak amplitude of the obtained ESC . Figure 4.1 shows an example
of CWT scalogram of measured transient signal and its peak energy per-
centage detection. In Figure 4.1, it shows that the dominant transient
frequency is located at scale number 24. Then, with this information the
specific wavelet coefficient will be extracted and the damped charge tran-
sient frequency (ωdm ) will be estimated with FFT algorithm. Since the
identification of charge transient component using CWT requires a step-
by-step process, which is heavy computational and time consuming, in
practical case the application is much more suitable for offline calculation
and analysis application.
59
50
I1
Current[A] 0
−50
50 100 150 200 250 300
49 0.16
46
43 0.14
40
37 0.12
34
31 X= 41
Scales a
Y= 24 0.1
28
25 Level= 0.16376
22 0.08
19
16 0.06
13
10 0.04
7
4
0.02
1
50 100 150 200 250 300
Sample Number[k]
(a) Simulated transient current (above) and its CWT scalogram (below)
10
I
X: 24 EP
9
Y: 8.182
8
Energy percentage[%]
7
6
5
4
3
2
1
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b) Scale-dependent energy percentage distribution
Figure 4.1. Example of the ouput of pre-processing of transient signal using CWT.
4.5 Analytic Representation and Damping Estimation Using Hilbert

Transform (HT)
In this section we discuss the application of Hilbert transformation to

estimate the damping (αm ) of the transient signal. Assuming that the
transient signal has been identified and the specific transient component
has been extracted, as explained in previous section, then the extracted
wavelet coefficient will be transformed into the analytic signal using the
Hilbert transformation (HT). According to [127], the Hilbert Transform
60
ũ(t) of a signal u(t) can be expressed as shown in Equation (4.11):
ũ(t) = (πt)−1∗ u(t) (4.11)
The significance of the Hilbert transform is that it is used to form so called

”analytic signal” or ”pre-envelope” from a real data sequence. The analytic
signal U (t), the real signal u(t) and its Hilbert Transform ũ(t) defines a
complex function as:
U (t) = u(t) + j ũ(t) (4.12)
Since the analytic signal is complex, it can be expressed in time domain

as [23, 128]:
U (t) = A(t)ejϕ(t) (4.13)
where:

A(t) = u2 (t) + ũ2 (t) (4.14)
ũ(t)
ϕ(t) = tan−1 (4.15)
u(t)
dϕ(t)
ω(t) = (4.16)
dt
where A(t) is instantaneous amplitude, ϕ(t) is instantaneous phase and

ω(t) is instantaneous frequency. In this work, the analytic signal of ex-
tracted wavelet coefficient of voltage and current signal can be repre-
sented in Equation (4.17) and Equation (4.18) respectively as follows:
Uv (t) = uv (t) + j ũv (t) (4.17)
Ui (t) = ui (t) + j ũi (t) (4.18)
where, uv (t) and ui (t) are the extracted wavelet coefficients of voltage and
current signal respectively, while ũv and ũi are their Hilbert transforma-
tions.
61
The main steps needed to obtain an estimation of the attenuation (αm )

through Hilbert transformation can be summarized as follows [129],
(i) Assume that we have obtained the interesting part of the transient
signal which is extracted CWT coefficient of charge transient com-
ponent as shown in Figure 4.2(c)
Current[A]
20
0
−20 Fault inception
50 100 150 200 250 300 350 400 450 500
40
37
34 0.15
31
Scales a
28
25 Charge transient component
22 0.1
19
16
13
10 0.05
7
4
1
50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(a)
10
I
X: 24 EP
9
Y: 8.182
8
7
6
5
4
3
2
1
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b)
200
I
CWT
150
100
Current[A]
50
−50
−100
−150
−200
0 50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(c)
Figure 4.2. Example of the output from pre-processing of current transient signals using
CWT. (a) Simulated transient current (above) and its CWT scalogram (below),
(b) scale-dependent energy percentage distribution and (c) extracted CWT
coefficient
62
(ii) Then, the Hilbert transformation is used to obtain the envelope of

the interesting part as a function of time as shown in Figure 4.3(a)
(iii) The damping attenuation (αm ) is estimated from the plot of instan-
taneous amplitude of the signal (Ax(t)) versus time as given in Fig-
ure 4.3 (a). In Figure 4.3 (b), the attenuation (αm ) is the approxi-
mated slope of the logarithm of the amplitude peaks (InAx(t)). A
straight line can be fitted through a plot of the log of the peaks. Ba-
sically, the damping can be estimated slightly after the time of the
earth fault inception. For example, in Figure 4.3 (b), the damping
factor is estimated from the time of 0.03 seconds to 0.06 seconds. If
the damping is linear, a first order least square polynomial curve fit-
ting method will be sufficient to fit the curve. Otherwise, the damp-
ing can be estimated by calculating the slope of the amplitudes of
two successive peaks and then averaging it.
50
40
30
20
A x[t]
10
0
−10
−20
−30
−40
−50
0.03 0.035 0.04 0.045 0.05 0.055 0.06
Time[s]
(a) Instantaneous current amplitude signal decaying characteristic (red) and its envelope
(blue)
4
lnA x[t]
−1
−2
0.03 0.035 0.04 0.045 0.05 0.055 0.06
Time[s]
(b) Linear decaying
Figure 4.3. Example of instantaneous current amplitude, envelope and the linear decay-
ing of the extracted CWT coefficient of charge component.
63
The undamped charge transient frequency (fnm ) can be estimated using

the following equation:

ωnm = αm2 + ω2
dm (4.19)
ωnm
fnm = (4.20)
2π
where ωdm , fnm and αm are estimated damped charge transient frequency
(rad/s) , estimated undamped charged transient frequency (Hz) and atten-
uation of the extracted wavelet coefficients, respectively.
64
5. Fault Distance Estimation Using
Transient Signals
In this chapter, we discuss and propose fault distance algorithms based

on measured transient signals. In first and second section of this chapter,
algorithms based on the correlation of fault distance and the charge tran-
sient frequency are first presented using general network model (GM) and
then an exact model (EM) of symmetrical components. Both algorithms
require only a single measurement of the transient signal to estimate the
transient frequency. This can be obtained, for example from the main
substation of the network. The third algorithm is based on continuous
wavelet transform (CWT) of both voltage and current, and it is essentially
an impedance measurement algorithm. The function of using CWT to ex-
tract the charge transient signal has been explained in Chapter 4. The
proposed CWT algorithm has some similarity to an algorithm developed
in [43]. Their similarity and differences will be discussed in Section 5.3.
In this work, we also developed two earth fault location algorithms based
on the amplitudes and frequencies of the transient signals recorded from
the secondary side of the distribution transformers. The idea of using
transient signals measured from secondary side of MV/LV for the purpose
of fault location and fault detection is still new for most of the researchers,
in the area of power systems. We have developed two algorithms that use
multiple regression analysis (MRA) and artificial neural network (ANN)
to estimate the correlation of fault distance and the measured quantities
using this kind of transient signal. In addition, with the recorded tran-
sient signal from MV/LV side of secondary transformer, we propose an
algorithm that can be used to identify the correct path of fault location
towards the position of the fault in the network with tree structure. The
performance of proposed fault location algorithms in this chapter will be
discussed in the next chapter.
65
Fault Distance Estimation Using Transient Signals
5.1 General Model (GM) of Symmetrical Components
The main principle of the general model (GM) algorithm is based on esti-
mating the fault distance using its correlation with the single frequency
of the charging process during the earth fault. As explained in Chap-
ter 2, the charging transient is usually of higher amplitude and lower fre-
quency than discharge transient. In the algorithm, in order to estimate
the earth fault distance, we use the estimated earth fault charge transient
frequency and compare it with the calculated earth fault charge transient
frequency. The earth fault charge transient frequency is estimated using
CWT and FFT analysis as explained in previous chapter, while the calcu-
lated charge transient frequency is formulated using developed simplified
lumped parameter model.
The idea of using simplified model for locating an earth fault in power
system have been proposed by several researcher [40,42,129,130]. Accord-
ing to [40], in case of single phase to earth fault, an interconnection of its
phase sequence networks can be simplified into a composite forms of se-
quence networks as shown in Figure 5.1 and its description is given in Ta-
ble 5.1. In case of a compensated neutral network, the compensation coil
can usually be omitted since at the transient frequencies its impedance is
high compared to that of the network earth capacitances.
The general models developed by [40] as given in Figure 5.1 were used
to formulate the equation of charge and discharge transient frequency. In
order to examine the best choice of model to be used for estimating the
earth fault distance, four general models have been examined. The mod-
els are called as general model 1 (GM1), general model 2 (GM2), general
model 3 (GM3) and general model 4 (GM4). The details of the models
is explained in the following section. The advantages of the algorithm is
that only one measurement is needed per primary substation transformer.
The developed algorithm is used in simple iteration to compare the esti-
mated and calculated earth fault charge transient frequencies. Figure 5.2
shows the basic idea of estimating earth fault distance based on general
model of symmetrical component algorithm, the details flowchart of the
algorithms is given in Figure 5.7 at the end of this section. The next is
subsection concerned with a formulated equation that was developed from
the general model of symmetrical components of earth fault transient.
66
(a) (b) (c) (d)
L2
2Ll,p+Ll,0
Llp,n Ll,0
R2
C0
C2
LT Cp
C0 R1
2LT (Cp/2) C1
L1
(e) (f) (g)
L2
LDT Llp,n
R2
RDT R1p,n
C2
LDT L3
RT R1
Cp RDT C1 R3
LL RL
LT L1
LL RL L4 R4
Figure 5.1. The general model for ground fault transients. Positive and negative se-
quence networks (a), zero sequence network for isolated networks (b), the
composite connection for isolated networks without (c) and with damping in-
cluded (d), model for load impedances (e), the equivalent circuit with loads
included for positive and negative sequence (f) and the combinations of all
sequence networks (g) [40].
Table 5.1. Description of the composite

components.
Composite
Remarks
components
C1 0.5Cp
C2 Co
L1 2LT
L2 2Ll,p + Ll,0
L3 2LDT
L4 2LL
R1 2RT
R2 2Rl,p + Rl,0 + 3Rf
R3 2RDT
R4 2RL
Notes:
RT and LT = resistance and inductance of sub-
station transformer, Rl,p , Rl,0 and Ll,p , Ll,0 =
sequence resistances and inductances of faulty
line length, RDT and LDT = resistance and
inductance of distribution transformer, Cp and
C0 = sequences capacitances of the network, RL
and LL = resistance and inductance of compos-
ite loads and Rf = fault resistance.
67
Figure 5.2. The basic idea of fault location algorithm using general model of symmetrical
components.
5.1.1 General Model 1 (GM1)
Let us consider an equivalent circuit model for ground fault transients

as shown in Figure 5.3. The equivalent circuit is a composite form of
sequence networks. In the equivalent circuit, we neglected the influence
of sound lines, distribution transformers, low voltage loads and network,
compensation capacitors and the transmission network. From the circuit,
the characteristic equation of the circuit can be obtained by solving the
equations for voltages around the loops of the circuit. The derivation of
the characteristic equation for circuit in Figure 5.3 is given in Appendix
A Section A.1. A simplified characteristic equation of circuit in Figure 5.3
is given as follows:
S 4 L1 L2 C1 C2 + S 3 (L1 R2 C1 C2 + L2 R1 C1 C2 )+
S 2 (L1 C1 + L2 C2 + L1 C2 + R1 R2 C1 C2 )+
S(R1 C1 + R2 C2 + R1 C2 ) + 1 = 0 (5.1)
68
R2 L2 C2
L1
E i2 i1 C1
R1
Figure 5.3. The equivalent circuit for earth fault transient of general model 1 (GM1)
algorithm. [40].
In Equation (5.1), the estimated fault distance (lf ) is included in parameter

L2 , where L2 = lf (Ll,p +Ll,n +Ll,0 ). By substituting all the required param-
eters as given in Table 5.1 into Equation (5.1) and then solving the roots
of the equation, we obtain two pairs of complex frequencies with negative
real parts as follows:
S1,2 = −αc ± jωc (5.2)
S3,4 = −αd ± jωd (5.3)
where, ωd , ωc , αd and αc are damped natural frequency of the discharge

transient, damped natural frequency of charge transient, damping atten-
uation corresponding to discharge transient and damping attenuation cor-
responding to charge transient respectively.
In the presence of damping the charge and discharge transient signals
will oscillate with damped natural frequencies. Due to the damping, the
damped natural frequency is lower than undamped natural frequency. In
case of decaying oscillation at frequency ωd and ωc , the undamped natural
frequency is related to the damped natural frequency as follows [131]:

ωnd = ωd2 + (αd )2 (5.4)

ωnc = ωc2 + (αc )2 (5.5)
By substituting the above Equations into Equation (5.2) and Equation (5.3),
S1,2 and S3,4 now can be written as:

2 − (α )2
S1,2 = −αc ± j ωnc c (5.6)

2
S3,4 = −αd ± j ωnd − (αd )2 (5.7)
where ωnc and ωnd are undamped natural frequency of charge and dis-
69
charge transient. For damping factor ζ < 1, Equation (5.6) and Equa-
tion (5.7), can now be written as:

2 − (α )2
S1,2 = −ζc ωnc ± j ωnc c (5.8)

2
S3,4 = −ζd ωnd ± j ωnd − (αd )2 (5.9)
where αc = ζc ωnc and αd = ζd ωnd . And, the damping factors for charge
and discharge transient are given as:
1
ζd = (5.10)
ωd
1+
αd
1
ζc = (5.11)
ωc
1+
αc
Subsequently, the undamped natural frequency is related to the damping

factor as follows:
−αd
ωnd = (5.12)
ζd
−αc
ωnc = (5.13)
ζc
Equation (5.12) and Equation (5.13) can further be expressed as:
ωnd
fnd = (5.14)
2π
ωnc
fnc = (5.15)
2π
70
Let us consider the LC circuit as shown in Figure 5.4. The equivalent cir-
cuit is similar as in Figure 5.3 but without the effect of damping. From the
circuit, the characteristic equation can be obtained by solving the equa-
tions for voltages around the loops of the circuit. The derivation of the
characteristic equation is given in Appendix A Section A.2. The charac-
teristic equation of circuit in Figure 5.4 is given as follows:
S 4 L1 C1 L2 C2 + S 2 (L1 C1 + L2 C2 + L1 C2 ) + 1 = 0 (5.16)
L2 C2
E i2 i1 C1
L1
algorithm. [40].
By substituting all the required parameters as given in Table 5.1 into

Equation (5.16) and then solving the roots of the equation, we obtain the
roots as two pairs of complex frequencies as follows:
S1,2 = ±jωnc (5.17)
S3,4 = ±jωnd (5.18)
where ωnc and ωnd are undamped natural frequency of charge and dis-
charge transient. Equation (5.17) and Equation (5.18) can further be ex-
pressed as in Equation (5.15) and Equation (5.14), respectively.
71
In this section, we consider the equivalent circuit with the effect of load
impedances on the earth fault transients as shown in Figure 5.5. The com-
posite load impedance is the combination of impedances of the low voltage
network and impedances of the loads. The combination is connected as a
series connection. In a low voltage network, the most significant compo-
nent is the reactance of the distribution transformer [40]. However, in this
model we consider the resistances of the distribution transformer as well.
We may also consider the impedances of the low voltage lines which can
be connected in series to the impedance of the distribution transformer. In
transient analysis, a single load, with reasonable accuracy, can be mod-
eled as resistance and inductance [40]. The load is modeled as a connec-
tion of these components. In isolated and compensated neutral networks,
there is no path for zero sequence current through the transformer, there-
fore the load impedance only affects the positive and negative sequence
networks. In Figure 5.1 (g), the impedance of the composite load is cou-
pled in parallel to the network positive sequence capacitances.
R2 L2 C2 R3 L3
L1
E i2 i1 C1 i3 L4 i4 R4
R1
algorithm.
From the circuit, the characteristic equation can be obtained by solving

the equations for voltages around the loops of the circuit. The derivation
of the characteristic equation for circuit in Figure 5.5 is given in Appendix
A, Section A.3. A simplified characteristic equation of circuit in Figure 5.5
is given as follows:
72
S 6 (L3 L1 L2 C2 L4 C1 )+
S 5 (C1 L4 L3 L1 R2 C2 + C1 L3 R4 L1 L2 C2 + C1 L4 L3 R1 L2 C2
+R4 L4 L1 C1 L2 C2 + C1 L4 R3 L1 L2 C2 )+
4
S (C1 L4 R3 R1 L2 C2 + L4 L1 L2 C2 + C1 L4 R3 L1 R2 C2
+C1 L4 L3 R1 R2 C2 + R4 L4 R1 C1 L2 C2 + R4 L4 L1 C1 R2 C2
+C1 L3 R4 L1 R2 C2 + L4 L3 L2 C2 + C1 R3 R4 L1 L2 C2
+C1 L3 R4 R1 L2 C2 + L3 L4 L1 C2 + L1 L3 C1 L4 )+
S 3 (R4 L4 R1 C1 R2 C2 + C1 L3 R4 R1 R2 C2 + C1 R3 R4 R1 L2 C2
+C1 L4 R3 R1 R2 C2 + C1 R3 R4 L1 R2 C2 + L4 R3 L2 C2
+L4 R4 L2 C2 + L4 L1 R2 C2 + L4 L3 R2 C2 + R4 L1 L2 C2
+L3 R4 L2 C2 + L4 R1 L2 C2 + L1 L3 C1 R4 + L1 L4 C1 R4
+L1 R3 C1 L4 + L3 R4 L1 C2 + L4 R4 L1 C2 + R3 L4 L1 C2
+R1 L3 C1 L4 + L3 L4 R1 C2 )+
2
S (L4 R4 R2 C2 + L4 R3 R2 C2 + L4 R1 R2 C2
+R4 L1 R2 C2 + R3 R4 L2 C2 + L3 R4 R2 C2
+L1 L4 + R4 R1 L2 C2 + C1 R3 R4 R1 R2 C2 +
L 3 R 4 R 1 C 2 + R3 L 4 R 1 C 2 + R3 R 4 L 1 C 2 + L 1 R 3 C 1 R4
+L4 R4 R1 C2 + R1 L3 C1 R4 + R1 L4 C1 R4 + L3 L4
+R1 R3 C1 L4 ) + S(R3 R4 R2 C2 + R4 R1 R2 C2
+L4 R4 + L3 R4 + R3 L4 + L1 R4 + R3 R4 R1 C2 + R1 R3 C1 R4
+L4 R1 ) + R1 R4 + R3 R4 = 0 (5.19)
By substituting all the required parameters as given in Table 5.1 into

Equation (5.19) and simplifying the equation, the roots of the equation
have the following expression:
S1,2 = −αc ± jωc (5.20)
S3,4 = −αd ± jωd (5.21)
λ1 = −a1 (5.22)
λ2 = −a2 (5.23)
where, ωd , ωc , αd , αc , a1 and a2 are damped natural frequency of the dis-

charge transient, damped natural frequency of charge transient, damping
73
attenuation corresponding to discharge transient, damping attenuation

corresponding to charge transient and two local coefficients, respectively.
The undamped natural frequencies and damping factors of charge and
discharge transient can be added into Equation (5.20) and Equation (5.21)
as we explained in Section 5.1.1.
An equivalent circuit for undamped earth fault transient as shown in Fig-

ure 5.6 is similar as in Figure 5.4 except with additional reactance L5
connected in parallel to C1 . L5 represents the imaginary part of total net-
work load impedances. In real network, the loads are scattered all over
the network. Hence, their reactance can be modeled as lumped and con-
nected parallel to the positive sequence network.
L2 C2
E i2 L1 i1 C1 i3 L5
algorithm.
From the circuit, the characteristic equation can be obtained by solving

the equations for voltages around the loops of the circuit. The derivation
of the characteristic equation for circuit in Figure 5.6 is given in Appendix
A Section A.3. A simplified characteristic equation is given as follows:
S 4 C1 C2 L1 L2 L5 + S 2 (C1 L1 L5 + C2 L1 L2 + C2 L1 L5 + C2 L2 L5 )
+L1 + L5 = 0 (5.24)
By substituting all the required parameters as given in Table 5.1 and in-
cluding L5 into Equation (5.24) and simplifying the equation, the roots of
the equation have the expression similar as in Equation (5.17) and Equa-
tion (5.18).
74
Figure 5.7. Flowchart for fault location algorithm using general model of symmetrical
components.
75
In Figure 5.7, let us assume that an earth fault at one of the feeders has
been detected. First, the transient signal is recorded and sampled. Then,
signal processing is made to the recorded transient signal by using CWT
analysis as described in Chapter 4. From CWT, the wavelet coefficient
belonging to identified charge transient component is extracted. In the
next step, the extracted CWT coefficient is used to estimate the damping
factor (ζm ) and damped transient frequency (fdm ). Both estimated pa-
rameters are used to estimate the undamped transient frequency (fnm ).
Basically, the damped transient frequency (fdm ) can be used by general
model 1 (GM1) and general model 3 (GM3) algorithm provided that the
earth fault that happens has zero fault resistance. However, in a real
case, there is always some damping due to the fault resistance. There-
fore, in order to cancel the effect of fault resistance, we used the estimated
undamped transient frequency (fnm ) to be compared with the calculated
frequency (fnc ).
Assume that, general model 1 (GM1) algorithm is selected to estimate
the fault distance. The algorithm use simple iterative process to estimate
the fault distance. A low value of calculated fault distance (lf ) is used as
an initial condition of the iteration process. From the formulated equation
of general model 1 (GM1), we calculate the undamped natural frequencies
(fnc ) and compare them with the estimated undamped charge transient
frequency (fnm ) that was recorded in the previous step. If the estimated
frequency is lower than the calculated frequency the iteration process will
be repeated by calculating the next calculated frequency (fnc ) with the
increased fault distance by increment value (∆lf ). Finally, the estimated
fault distance (lf ) is successfully achieved if the estimated frequency (fnm )
is same or higher than the calculated frequency (fnc ). In order to reduce
the number of iterations required to reach the final solution, the fault
distance increment value (∆lf ) can be increased to accelerate the iteration
process.
76
5.2 Exact Model (EM) of Symmetrical Components
Early investigation done by [130, 132, 133] shows that the transient fre-
quency can be correlated to the fault distance by using the simple RLC-
circuit. However due to the reduction and simplification of the circuit the
accuracy of the methods is reduced. The accuracy can be improved if the
network can be modeled completely. Similar to the GM algorithm, the
main principle of exact model (EM) algorithm that we propose in this sec-
tion is based on estimating the fault distance using the single frequency of
the charging process during an earth fault. However, in exact model (EM)
algorithm, we take into account the effect of inductances of sound lines.
In addition, instead of lumped the line capacitances at the substation, we
model them as an exact π-model solution. In this solution, the lines are
modeled using exact π-model equivalent circuits as shown in Figure 5.8.
In Figure 5.8, the exact π-model based on sequence component equivalent
circuits represents the positive and zero sequence components of earth
fault transients where the distribution transformers, low voltage loads
and LV network have been neglected. In the model, we assume that the
negative sequence network and positive sequence network is identical.
(a)
Ll,0’ L1,0 Lj,0
F
Cl,0’ Cl,0’ Cl,0 Cl,0 Cj,0 Cj,0

2 2 2 2 2 2
j=1...n
(b)
Ll,p’ Ll,p Lj,p
F
Cl,p’ Cl,p’ Cl,p Cl,p LT Cj,p Cj,p

2 2 2 2 2 2
j=1...n
Figure 5.8. Exact π-model equivalent circuit with damping included. Ll,0 , Ll,p are zero
sequence and positive sequence inductance of faulty line located in front of
the fault point (F ). Ll,0′ , Ll,p′ are zero sequence and positive sequence induc-
tance of faulty line located behind of the fault point (F ). Lj,0 , Lj,p are zero se-
quence and positive sequence inductance of sound line. LT is the inductance
of the transformer. n is the number of lines. Capacitances correspondingly.
In the exact model (EM) algorithm, the undamped complex frequency is

defined from the single phase to earth fault bus impedance matrix of the
77
network in complex frequency domain. For single phase to earth fault, the
diagonal elements of the nodal transform impedance matrix in complex
frequency (s) is given as:
1 0 2 1
Zkk (s) = Zkk (s) + Zkk (s) (5.25)
3 3
0 and Z 1 , are the zero and positive sequence diagonal elements

where Zkk kk
in the k node of the corresponding bus impedance matrix. In Equation (5.25),
positive and negative sequence impedance is assumed identical.
1 2
Zkk (s) = Zkk (s) (5.26)
The fault resistance (Rf ) was not required to be added to the equation.
This is because, in this work we purposely formed the impedance ma-
trix for undamped case of ground fault transient. Then, in order to esti-
mate the fault distance, we substitute the estimated undamped complex
frequency (fnm ) of the charge transient for complex frequency (s) in Equa-
tion (5.25). Then, we solve the roots of a polynomial equation of Zkk (s) = 0.
In the computer model, the lowest positive value of the roots from the
simulation results gives the estimated fault distance. The explanation of
exact the π-model equivalent circuit that was used in exact model (EM)
algorithm is the main subject of the next subsection. The flowchart for
estimating the earth fault distance based on exact model (EM) algorithm
is given at the end of this section.
5.2.1 Exact Model 1 (EM1)
An equivalent circuit of exact π-model for undamped case of earth fault

transient is presented in Figure 5.9. In this model, we have neglected the
effect of low voltage loads and LV network, compensation capacitors and
transmission network. The effect of the neglected components on fault lo-
cation performance is discussed in next chapter. As seen in Figure 5.9, we
purposely modeled the equivalent circuit for undamped case of earth fault
transient. This is because we want to avoid the effect of damping due to
fault resistance in fault location calculation process. The idea is to cal-
culate the undamped charge transient frequency from estimated damped
transient frequency and its damping factor of the earth fault transient
signal. The undamped charge transient frequency is then calculated, by
using the technique explained in Section 4.4 through Section 4.5. With
78
the information of network components and real time data of their con-
nections, the equivalent circuit exact π-model can be used in computer
simulation model using nodal analysis. As an example, Equation (5.27)
shows the sequence admittance matrix in complex frequency (s) of Fig-
ure 5.9 that has k number of buses.
(a)
D*Ll,0’ F k*L1,0 Lj,0
N3 N2 N1 Nk
D*Cl,0’ k*Cl,0
2 2 Cj,0 Cj,0
D*Cl,0’ k*Cl,0 2 2
2 2
j=1...n
(b)
D*Ll,p’ F k*L1,p N1 Lj,p
N3 Nk
N2
D*Cl,p’ k*Cl,p LT
2 2 Cj,p Cj,p
D*Cl,p’ k*Cl,p 2 2
2 2
j=1...n
Figure 5.9. The equivalent circuit for earth fault transient of exact model 1 (EM1) algo-
rithm. Ll,0 , Ll,p are zero sequence and positive sequence inductance of faulty
line located in front of fault point (F ). Ll,0′ , Ll,p′ are zero sequence and pos-
itive sequence inductance of faulty line located behind of the fault point (F ).
Lj,0 , Lj,p are zero sequence and positive sequence inductance of sound line.
LT is the inductance of the transformer, k is a fault distance, T is the total
lenght of faulty feeder, D = T −k is the total length of faulty feeder (T ) minus
fault distance (k), N1 to Nk is the number of the buses and n is the number
of lines. Capacitances correspondingly.
⎡ ⎤
Y11 (s) Y12 (s) Y13 (s) · · · Y1k (s)
⎢ ⎥
⎢Y21 (s) Y22 (s) Y23 (s) · · · Y2k (s)⎥
⎢ ⎥
⎢ ⎥
Y012 (s) = ⎢Y31 (s) Y32 (s) Y33 (s) · · · Y3k (s)⎥ (5.27)
⎢ ⎥
⎢ .. .. .. .. .. ⎥
⎢ ⎥
⎢ . . . . . ⎥
⎣ ⎦
Yk1 (s) Yk2 (s) Yk3 (s) · · · Ykk (s)
By inverting Equation (5.27), the sequence impedance matrix is ex-

pressed as follows:
Z012 (s) = [Y012 (s)]−1 (5.28)
Finally, in computer simulation model, the nodal transform impedance

matrix in complex frequency (s) is given as in Equation (5.25).
79
In Exact Model 2 (EM2) algorithm we model a slightly different construc-

tion of exact π-model of undamped earth fault transient equivalent circuit
compared to the EM1 model. Both models have similar construction ex-
cept that in EM2 model, the line positive and zero sequence inductances
(Ll,0′ , Ll,p′ ) located behind the fault point (F ) of a faulty feeder section
is neglected . In addition, the capacitance in the faulty line feeder are lo-
cated depending on the location of the earth fault. For example, if the fault
happens in the middle of the faulty feeder, capacitance at fault point (F )
and bus N1 is shared for each half of the faulty line feeder capacitances.
However, if the fault happens at the end of the line, the capacitance is
higher at the fault point (F ), and vice verse.
We purposely invent the model as it is in order to observe the effect of
fault location with different exact π-model constructions. Similarly as in
EM1, in order to model it in computer simulation model, the information
of network components and real time data of their connections is required.
From the computer simulation model, we solved the impedance matrix in
complex frequency (s) as given in Equation (5.25).
(a)
F k*L1,0 Lj,0
N2 Nk
N1
k*Cl,0 Cj,0
2 Cj,0
D*Cl,0
2
j=1...n
(b)
F k*L1,p Lj,p
N1 Nk
N2
k*Cl,p LT Cj,p
D*Cl,p 2 Cj,p
2
j=1...n
Figure 5.10. The equivalent circuit for earth fault transient of exact model 2 (EM2) al-
gorithm. Ll,0 , Ll,p are zero sequence and positive sequence inductance of
faulty line located in front of the fault point (F ). Lj,0 , Lj,p are zero sequence
and positive sequence inductance of sound line. LT is the inductance of
the transformer, k is a fault distance, T is the total lenght of faulty feeder,
D = T − k is the total length of faulty feeder (T ) minus fault distance (k), N1
to Nk is the number of the buses and n is the number of lines. Capacitances
correspondingly.
80
We propose the exact model 3 (EM3) algorithm with the use of exact
π-model as shown in Figure 5.11. The model is constructed similar to
exact π-model in EM2 algorithm, except that the sequence capacitance
(Cl,0 ,Cl,p ) of its faulty feeder section is located at bus N1 as depicted in
Figure 5.11. Through the developed computer model, we simulate and
solve the impedance matrix in complex frequency (s) as given in Equa-
tion (5.25).
F (a)
k*L1,0 Lj,0
N1 Nk
T*Cl,0 Cj,0 Cj,0

2 2
j=1...n
F (b)
k*L1,p Lj,p
Nk
N1
LT
T*Cl,p Cj,p Cj,p
2 2
j=1...n
Figure 5.11. The equivalent circuit for earth fault transient of exact model 3 (EM3) al-
gorithm. Ll,0 , Ll,p are zero sequence and positive sequence inductance of
faulty line located in front of the fault point (F ). Lj,0 , Lj,p are zero sequence
and positive sequence inductance of sound line. LT is the inductance of
the transformer, k is a fault distance, T is the total lenght of faulty feeder,
D = T − k is the total length of faulty feeder (T ) minus fault distance (k), N1
to Nk is the number of the buses and n is the number of lines. Capacitances
correspondingly.
In Figure 5.12, the flowchart shows the steps used by exact model algo-
rithm for computing the undamped transient frequency (fnm ) is similar as
explained in previous section. Let us assume that Exact model 1 (EM1)
algorithm is selected to estimate the fault distance. Then, using a esti-
mated damped transient frequency (fdm ), we calculate the inductance per
unit length of line network using Frequency-Dependent(FD) Line Model
or known as JMarti FD model in EMTP environment. In EMTP envi-
ronment, we can calculate the sequence impedance of the line at specific
estimated charge transient frequency using line check calculation tools
provided that we have the details of line geometrical and electrical val-
ues. Next, the estimated undamped frequency (fnm ) and calculated line
81
inductance is applied to computer simulation program that was built in

MATLAB environment to calculate Equation (5.25) and to solve the roots
of the polynomial equation. Finally, the lowest positive value of the roots
gives the estimated fault distance.
Figure 5.12. Flowchart for fault location algorithm using exact π-model of symmetrical
component.
82
5.3 Continuous Wavelet Transform (CWT)
In this section, we propose a fault location algorithm based on continuous

wavelet transform (CWT). The developed CWT algorithm is similar to the
algorithm developed by [43] but slightly different process is used to esti-
mate the fault path inductance. [43], applied discrete wavelet transform
(DWT) to transient signal for fault distance estimation algorithm. The
algorithm first determines the maximum wavelet coefficient of current in-
cluding the amplitude, frequency and location of the wavelet. Later, the
fault path inductance is calculated using the maximum wavelet coeffi-
cients of current and voltage and its frequency. Finally, the fault distance
is calculated from estimated inductance using the same equation as is
expressed in Equation (5.30). A similar method can also be found in [16].
Assuming that the process of identification of charge transient compo-
nent and transforming it into analytical signal (Uv (t), Ui (t)) as described
in Section 4.4 and Section 4.5 is accomplished, the fault distance can be
estimated by first evaluating the inductance as follows:

1 Uv (t)
Lf (t) = Im (5.29)
2πfc Ui (t)
where, Lf (t) is instantaneous value of inductance, fc is a estimated charge

transient frequency of the extracted wavelet coefficient, and Uv (t) and
Ui (t) are the analytic forms of the extracted coefficient of voltage and cur-
rent transient respectively. Then, the fault distance (lf ), can be calculated
by using Equation (5.30) as follows [43]:
Lf
lf = 1 ′ ′ ′ (5.30)
3 (L0 + L1 + L2 )
′ ′ ′
where, L0 , L1 , L2 are the inductances per unit length of the zero, positive
and negative-sequence systems respectively. The final value of the fault
distance (lf ) was selected based on the smallest standard deviation of the
mean value of the fault distance calculated in Equation (5.30) for a set
of subsequent samples. The flow chart representing the CWT algorithm
developed in this work is shown in Figure 5.13.
83
Figure 5.13. Flowchart for fault location algorithm based on CWT.
84
5.4 Transient Signal Recorded from the Secondary side of the

MV\LV Distribution Transformer
The concept of "SmartGrids" which has the characteristics of integrated,

merging monitoring, control, protection, maintenance of all the resources
for more efficient network management system [134], opens a great op-
portunity for installation of more measuring and recording devices in the
low voltage network. A study done by [135], proved that an advanced
AMR meter is able to work as an intelligent monitoring device for locat-
ing low voltage network faults and for monitoring voltage level. The in-
formation gathered from advanced AMR meter is transmitted using com-
munication infrastructure to the control center. In case that a transient
or disturbance recorder is installed at the MV/LV substation, a similar
concept as introduced by [135] could be applied for monitoring of tran-
sient signal. In recent studies, the transient signal of earth fault in MV
network is captured using transient recorder located at low voltage side
of a MV/LV distribution transformer [136, 137]. When a fault happens in
medium voltage (MV) distribution feeder, the transient features can be
distinguished from the secondary side of the MV/LV distribution trans-
former. In this section, we propose an alternative approach for earth
fault location, based on the transient signal measured from the MV/LV
distribution transformer. Based on the typical transformer connections
and earthing practices in Finland, the MV/LV distribution transformer is
modeled as delta connected HV winding, star connected LV winding with
neutral brought out (DyN11) [138].
The algorithms are based on the observation, that when measuring from
primary substation towards the fault, the amplitude of the charge tran-
sient voltage is increased towards the fault location. Behind the fault, the
voltage is equal to the voltage in the fault point. In the parallel branches,
the voltage is the same as in the point of common coupling, as seen from
fault point. Hence, the fault location can be found by analyzing the am-
plitudes of transient voltages in different network locations.
The algorithms utilize the information gathered from the measured tran-
sient voltage signal recorded from the secondary side of the MV/LV distri-
bution transformer and tries to find the correlation between fault distance
and measured quantities using either Multiple Regression Analysis or Ar-
tificial Neural network. Using the similar technique as we describe in
Chapter 4 Section 4.4, we used CWT to identify the dominant transient
85
frequency and extract its coefficients. Besides of the characteristics of the

extracted coefficients, the information of the total loads of the distribution
network is also required in this algorithm. In the next subsection, we ex-
plain the pre-processing stage of the measured transient signal gathered
from the MV/LV distribution transformers. The proposed algorithms are
depicted in Figure 5.14.
Figure 5.14. Flowchart for MRA and NN fault location algorithms based on recorded
transient signal of single MV/LV distribution transformer.
86
5.4.1 Pre-processing stage
When an earth fault happens in MV feeder, the transient signal can also
be detected from the secondary side of the MV/LV distribution trans-
former. The transient component can easily be distinguished from the
fundamental frequency component. Figure 5.15 shows an example of sim-
ulated transient signal recorded at the MV and LV side of the network.
In Figure 5.14, let us assume that an earth fault has been detected in
MV network. From the assumption that an earth fault has been detected,
the transient signal measured from MV/LV distribution transformers is
recorded and sampled. Then, the recorded signal will be converted from
phase to line voltage. Later, the signal is analyzed with CWT and the
CWT extraction features process is applied to the transient signal as has
been explained in Chapter 4, Section 4.4. Figure 5.16 shows an exam-
ple of extracted CWT coefficient of transient signal recorded from MV/LV
distribution transformer.
4
3 x 10
Va
Vb
2 Vc
1
Voltage [V]
−1
−2
−3
Fault Inception
−4
0 100 200 300 400 500 600 700 800 900 1000
Sample Number[k]
(a) Phase voltage
800
Vab
Fault Inception Vbc
600 Vca
400
Voltage [V]
200
−200
−400
−600
0 100 200 300 400 500 600 700 800 900 1000
Sample Number[k]
(b) Line voltage
Figure 5.15. An example of simulated transient signal recorded at (a) the MV and (b) LV
side of the MV\LV distribution transformer
87
It has been reported that the loads have a significant effect to the tran-
sient signal [40,139]. Therefore, due to this reason, the total load (MVA) of
the network prior an earth fault happens in MV network is also recorded.
Based on the extracted CWT coefficient characteristics, the peak value of
its FFT amplitude and its coefficient scale is recorded. Actually, the scale
of CWT coefficient is correlative to its frequency. Therefore, either one
of the parameters can be used. After computing and recording the two
characteristics of extracted CWT coefficients, with these two parameters
including the total load (MVA) of the network prior a to the earth fault
they are used as the inputs database of the MRA and ANN algorithms.
Multiple regression analysis (MRA), artificial neural network (ANN) and
correct path algorithms are the main subject of the following subsection.
500
V
250
Voltage[V]
−250
−500
100 150 200 250 300 350 400 450 500
Sample Number[k]
(a) Extracted CWT coefficient

14
Energy
12 X: 17
Y: 10.69
Energy Percentage[%]
10
0
0 5 10 15 20 25 30 35 40
Scales[a]
(b) Calculated Energy Percentage of CWT coefficients
100
FFT
90
80 X: 996.1
Y: 71.21
70
Amplitude[FFT]
60
50
40
30
20
10
0
0 500 1000 1500 2000 2500 3000
Frequency[Hz]
(c) FFT amplitude of Extracted CWT coefficient
Figure 5.16. Characteristics of extracted CWT coefficient of transient signal recorded

from MV/LV distribution transformer.
88
5.4.2 Multiple Regression Analysis (MRA)
A simple linear regression illustrates the relation between the dependent

variable y and the independent variable x based on the regression equa-
tion below:
yi = β 0 + β 1 x 1 + e i (5.31)
where, i = 1, 2, 3, ...., n, β0 is intercept or constant, β1 is called slope or co-

efficient and ei refers to error term. Using the least squares method, the
best fitting line can be found by minimizing the sum of the squares ver-
tical distance from each data point on the line [140]. In this work we de-
veloped the formula in Equation (5.31) using MATLAB matrix commands
and the example of the MATLAB matrix commad is given in Appendix A
Section A.5. As shown in Figure 5.14, the database developed for MRA
algorithm is based on information gathered from measured transient sig-
nals of single MV/LV distribution substations. Since there are several
MV/LV substations in the MV network, based on our studies using simu-
lation model given in Chapter 6, Section 6.4 of this dissertation, we chose
the farthest MV/LV substation in simulation network model for develop-
ing the database. The reason why we chose that particular substation is
that, when the simulated single line to earth fault happens at the farthest
distance from the main MV substation, the estimated FFT amplitude of
extracted CWT coefficient of the chosen substation is the highest com-
pared to other substations. Also, we found that when the fault resistance
is increased and fault inception angle is decreased, the estimated FFT
amplitude of extracted CWT coefficient of MV/LV substations will subse-
quently be decreased. However, the question that which is the most suit-
able MV/LV substation could be used for the purpose of designing MRA
algorithm based on measured transient signal from MV/LV side of distri-
bution transformer is not being studied in this dissertation but perhaps
in future.
According to the multiple linear regression model the dependent vari-
able is related to two or more independent variables. The general model
for k variables is of the form:
yi = β0 + β1 xi + β2 xi2 + .... + βk xik + ei (5.32)
where, i = 1, 2, 3, ...., n and k = 2, 3, ...., n , β0 is intercept or constant,

βk is called slope or coefficient and ei refer to error term. The multiple
89
regression model developed for fault location in this dissertation is defined

in Equation (5.33) as:
l = β0 + β1 A + β2 S + β3 L (5.33)
where, l is fault distance, A is FFT amplitude of extracted CWT coeffi-

cient, S is scale of extracted CWT coefficient and L is total loads (MVA) of
the network prior to a single line to earth fault.
The developed multiple regression model used for fault location is spe-
cific for the network during the observation. The model needs to be rede-
veloped if there is a change in network configuration. The reason is, those
changes will affect the variables used in the developed multiple regres-
sion model. Therefore it is necessary to redeveloped the regression model
according to the condition of the network.
Using typical 20 kV isolated neutral distribution network simulated
using EMTP-ATP program, the proposed multiple regression model was
tested and the result and the performance will be discussed in detail in
Chapter 6, Section 6.4.
5.4.3 Artificial Neural Network (ANN)
An algorithm based on artificial neural network (ANN) has been dis-

cussed in Chapter 3. In case of fault location in MV network, several
researchers mainly used the data measured from MV substation or faulty
feeder in order to design a ANN algorithm. In this research we propose
a new approach for fault location in unearthed or compensated neutral
distribution system using combination of wavelet analysis and artificial
neural network (ANN) and based on measured transient voltage signals
recorded from secondary side of MV/LV distribution transformers. Accord-
ing to [141], using the relationship of dependent and independent number
of n samples, the formula created using artificial neural network (ANN)
analysis is in the following expression:
y = AN N (x1 , x2 , ....., xn ) (5.34)
where ANN is an implicit non mathematical function that can only can be
designated as ’knowledge base’. Basically, using the same dependent and
independent variables as developed for MRA formula can also be used for
artificial neural network analysis. A proposed designed neural network
90
is constructed with a two layers, feed forward network with tangent sig-
moid hidden neurons and linear output neuron. The developed neural
network training function updates weight and bias values according to
Levenberg-Marquardt optimization algorithm. ANN architecture of the
proposed ANN algorithm is depicted in Figure 5.17.
Output value
Output layer
(1 neuron)
Hidden layer
(20 neurons)
A S L Input Layer
(3 neurons)
Inputs
Figure 5.17. Architecture of neural network for estimate the fault distance.
The designed neural network has 3 inputs variables which contain valu-
able and relevant information for estimating fault distance. Each input
unit of neural network has a symbol associated with it where A is repre-
sented as the peak value of FFT amplitude of extracted CWT coefficient,
S is associated its coefficient scale and L is referred to total loads (MVA)
of the network prior to an earth fault. For fault distance estimation al-
gorithm in this work the ANN consists only one hidden layer. For this
kind of input-output task, one single hidden layer is assumed to be good
enough [79]. The outputs of hidden layer has sigmoid activation function
and are transferred to the output layer which is composed of only one
neuron which gives the estimated fault distance. It is important to train
and test the neural network with a large data set. The training process is
done in several times and each time the performance after each iteration
is saved where the performance is defined as the correlation coefficient
(r). Finally the saved trained network that has the highest value of cor-
relation coefficient (r) is used to for fault location algorithm. Since this
method is based on the ’knowledge base’ of the specific network, therefore
the NN have to retrain if there are any changes in terms of the network
topology.
The proposed algorithm have has tested using simulation model based
on a practical 20 kV unearthed network distribution network as shown
in Chapter 6 Section 6.4. The data set used in training and testing the
91
neural network is given in Chapter 6, Table 6.3 and Table 6.4, respec-
tively. The result and performance of the algorithm will be discussed in
that chapter.
5.4.4 Correct path location in branches of feeders
In networks with several branches, indicating the correct fault path is

a separate issue. In fault location algorithm it is a difficult task if only
one measurement is used to identify the correct branch of a complex tree
structure kind of network. Therefore, we propose an algorithm that use
the transient signal recorded from secondary side of MV/LV substations as
the basis for identifying and selecting a correct path to the fault position.
The work flow of the algorithm is shows in Figure 5.18.
Figure 5.18. Flowchart for correct path location algorithm based on recorded transients
of MV/LV distribution transformers.
The line to line fault voltage transient signal measured from the sec-
92
ondary side of MV/LV substation transformer is processed with CWT sim-

ilarly as in Chapter 4 Section 4.4. Then the peak amplitude of the FFT fre-
quency spectrum of the extracted CWT coefficients for each substation is
identified and recorded. For correct path location, a fault distance should
be taken from the measured point at main HV/MV substation towards
the MV/LV substation which has highest recorded FFT amplitude of the
extracted wavelet coefficient. In other words, the highest recorded FFT
amplitude will indicate which MV/LV substation should we locate in order
to determine the correct fault location if there are several branches. The
proposed algorithm has been tested using a simulation model in Chap-
ter 6, Section 6.3. The result and performance of the algorithm will be
discussed in that chapter.
93
94
6. Performance Evaluation of Proposed
Fault Location Algorithms Using
EMTP/ATP Simulation Model
In this chapter, the proposed algorithms are tested using transient signal
generated by a simulation model. The simulation of the voltage and cur-
rent transient is required and important in order to validate the proposed
fault location algorithms. In this research, we used Alternative Transient
Program (ATP/EMTP) as transient simulator. The program was chosen
because it is one of the most common simulation tools used by power en-
gineers and researcher for simulating transients [142]. The program soft-
ware contains extensive modeling capabilities for transmission lines, ca-
bles, breakers, loads, converters, protection devices, non-linear elements,
electromagnetic coupling, and major power electronics devices and equip-
ment. It also has an enhanced graphical user interface called ATPDraw
as a pre processor, which allows an easy entry of system topology and
data [143] . The program software will in future be referred to only as
EMTP.
In this chapter, four simulation models are presented. The first model
is according to general model of symmetrical component as we have dis-
cussed in Chapter 5, Section 5.1. In simulation model 2, we modeled the
unearthed neutral network with exact π-model as described in Chapter 5,
Section 5.2. Then in the following section, we investigate the earth fault
network using the line constants EMTP program and taking into account
their real geometrical and electrical values. In the last section of this
chapter, we tested the fault location algorithms with the modified system
from [144] and modeled the network using line constant EMTP program.
The earth faults are simulated with different scenarios such as fault dis-
tance, fault resistance, fault inception angle and also the effect of load.
Some results of this studies can be found in [137].
95
Performance Evaluation of Proposed Fault Location Algorithms Using EMTP/ATP Simulation Model
6.1 Simulated Network Model 1
In the beginning of this work, we studied the correlation of fault distance

and the transient frequency using the general model algorithm as ex-
plained in Chapter 5 Section 5.1. The simulated network model was based
on the model as shown in Figure 5.1 (d) and (g). The model neglected the
inductance and resistance of sound lines, distribution transformer and
approximated the network capacitances as lumped components located
at the substations. According to the Thevenin theorem, the ground fault
transient can be produced by injecting a voltage of equal magnitude but
opposite sign compared to the phase voltage at the fault point [40]. The
description of the network model is given in the following section.
6.1.1 Description of the simulation and the network model
A schematic diagram of the simulated 20 kV unearthed medium voltage

network and its EMTP simulation model is shown in Figure 6.1(a) and
(b) [129]. The network is radially operated. It consists of 5 feeders with a
total of 100 km of overhead lines.
In EMTP simulation, a single line to ground fault was simulated in a
20 km long line and fault distances varied from 2 km to 20 km with 2 km
steps. The fault resistances varied from 0 Ω to 100 Ω. The fault distance
was maintained at 2 km, 4 km, 10 km and 16 km when the varied fault
resistance was applied to the model. In addition, different fault voltage
inception angles were applied to investigate the sensitivity of transient
frequency and the performance of the fault location algorithm. Fault dis-
tance was fixed to 4 km and 16 km with zero fault resistance when fault
inception angle variation was applied to the model. A sampling frequency
of 20 kHz was used.
The specification of the various elements in the simulation model are
as follows: LT = 5.6 mH, RT = 0.04 Ω, Ll,p = 1.0 mH/km, Ll,0 = 5.0 mH/km,
Cl,p = 38.25 nF/km, Cl,0 = 6.12 nF/km, Rl,p = 0.60 Ω/km, Rl,0 = 1.30 Ω/km,
LDT = 19.09 mH, RDT = 0.18 Ω, LL-5MVA = 584.20 mH, LL-12MVA = 233.68 mH,
RL-5MVA = 88.89 Ω and RL-12MVA = 35.56 Ω. The details of the various ele-
ment abbreviations can be found in Table 5.1. The simulation results are
discussed in the next section.
96
OH line
OH line
OH line OTHER FEEDERS=80 km

110/20kV .
.
.
.
.
.
.
OH line
OH line faulty feeder = 20 km
(a) 20kV 100km unearthed neutral MV network
(b) ATPdraw simulation model
Figure 6.1. A Schematic(a) and ATPdraw simulated network model(b) for earth fault
simulation analysis in 20 kV unearthed medium votage network. The details
of the abbreviations in ATPdraw simulation model can be found in Table 5.1
6.1.2 Result Analysis and Discussion
The performance accuracy of the GM algorithm has been investigated un-

der varied fault distances, several fault resistances, different fault incep-
tion angles and load conditions. In these studies, the performance accu-
racy used to evaluate GM algorithm is defined by mean absolute error
(MAEGM ) index as expressed in equation below:
n n
1 1
M AEGM = |GMi − li | = |∆GMi | (6.1)
n n
i=1 i=1
where ∆GMi is the absolute error fault distance using GM algorithm,

GMi is estimated fault distance using GM algorithm, li is exact fault dis-
tance and n is numbers of observation of fault distance error. Similarly, to
evaluate the error between estimated and calculated frequency and damp-
ing factor in the simulation result, the mean absolute error (MAE { f,ζ } ) is
calculated with the following equation:
n n
1 1
M AE{f,ζ} = |∆{fm − fc , ζm − ζc }| = |∆{fi , ζi }| (6.2)
n n
i=1 i=1
97
where ∆|{fi , ζi }| is the absolute error of the parameters (f, ζ), fm is es-
timated transient frequency, fc is calculated transient frequency, ζm is
estimated damping factor and ζc is calculated damping factor and n is
numbers of observation of parameter error.
Figures 6.2 and 6.3 shows the example of earth fault transient voltage
and current for earth fault located at 2 km, 10 km and 18 km at faulty line
feeder with overhead line as shown in Figure 6.1. The transient signals
were captured at location "M" as shown in Figure 6.1. The earth fault was
simulated at time t=0 and the signal was captured before the time switch
is "on" at t=0.005 second.
8000
6000 V1
4000
Voltage [V]
2000 X: 100
Y: 0
0
−2000
−4000
−6000
−8000
0 100 200 300 400 500 600 700 800 900 1000
Sample Number [k]
(a) Transient voltage, lf =2 km
8000
V1
6000
4000
Voltage [V]
2000 X: 100
Y: 0
0
−2000
−4000
−6000
−8000
0 100 200 300 400 500 600 700 800 900 1000
Sample Number [k]
(b) Transient voltage, lf =10 km
8000
V1
6000
4000
Voltage [V]
2000 X: 100
Y: 0
0
−2000
−4000
−6000
−8000
0 100 200 300 400 500 600 700 800 900 1000
Sample Number [k]
(c) Transient voltage, lf =18 km
Figure 6.2. Simulated transient voltage signal in the faulty phase when a single phase to
earth fault happens at a feeder in an example network with overhead lines.
The fault distances (lf ) are 2, 10 and 18 km from the substation. The fault
resistance is zero ohm.
98
100
I
80 1
60
40
Current [A]
20 X: 100
Y: −0
0
−20
−40
−60
−80
−100
0 100 200 300 400 500 600 700 800 900 1000
Sample Number [k]
(a) Transient current, lf =2 km
50
I
1
40
30
20
Current [A]
10 X: 100
Y: −0
0
−10
−20
−30
−40
0 100 200 300 400 500 600 700 800 900 1000
Sample Number [k]
(b) Transient current, lf =10 km
40
I1
30
20
Current [A]
10
X: 100
Y: −0
0
−10
−20
−30
0 100 200 300 400 500 600 700 800 900 1000
Sample Number[k]
(c) Transient current, lf =18 km
Figure 6.3. Simulated transient current signal in the faulty phase when a single phase to
earth fault happens at a feeder in an example network with overhead lines.
The fault distances (lf ) are 2, 10 and 18 km from the substation. The fault
resistance is zero ohm.
The earth fault is simulated at the instantaneous voltage maximum and

the transient signal is measured at the substation. Figures 6.2 and 6.3
shows that when lower fault distance, the peak amplitude of the tran-
sient signal (voltage and current) is higher. Correspondingly, when the
fault distance is increased the peak amplitude of the transient signals is
decreased.
As an example, a signal pre-processing output for a single phase to earth
fault with fault distance 10 km from substation is presented. The fault is
simulated with zero fault resistance and fault inception angle 90◦ (in-
99
stantaneous voltage maximum). Figure 6.4 shows an example of the out-

put from pre-processing by CWT, Hilbert transformation (HT) and fast
Fourier transform (FFT) as described in Chapter 4.
In Figures 6.4 and 6.5, simulated voltage and current transient signals
from 10 km earth fault simulated case was analyzed and sampled starting
from the beginning of the fault inception. In Figure 6.4(a), a high density
of energy spectrum corresponding to the charge and discharge component
can be seen clearly from the instant of fault occurrence. Both charge and
discharge components can be seen from CWT scalogram of voltage tran-
sient. However, in case of CWT scalogram of current transient signal as
shown in Figure 6.5(a), only charge component is distinguished. Both sig-
nals have higher density of energy level (red color bar) appeared at the
beginning of the fault inception which then decreased towards the end of
the signal.
In Figures 6.4(b) and 6.5(b), based on the result of scale-dependent en-
ergy distribution, we recognize the peak amplitude of the energy percent-
age (%) that corresponds to the dominant energetic scales within the sig-
nal. Hence, the scales obtained from the highlighted peak are 23 and
24 for voltage and current signals, respectively. From the scale-energy
distribution spectrum, we can conclude that, for this case of earth fault
simulated transient signal, the charge transient component is more dom-
inant than the discharge transient component in both signals. Extracted
CWT coefficient of voltage and current signal is depicted in Figures 6.4(c)
and 6.5(c), respectively, while its FFT frequency spectrum is presented in
Figures 6.4(d) and 6.5(d). The peak amplitude in FFT frequency spectrum
is corresponding to the dominant damped transient frequency of the sig-
nal. In this case the value is 703.1 Hz. Figures 6.4(e) and 6.5(e), show an
envelope of extracted CWT coefficient fitted using polynomial curve fitting
technique in MATLAB environment. Hence, from these points, with the
estimated value of damped transient frequency and the damping factor,
the undamped transient frequency can be estimated.
In order to show the effect of high fault resistance on transient signal as
well as to the algorithm, the earth fault was simulated for fault resistance
up to 400 Ω. Figures 6.6 and 6.7, shows the simulated current transient
signal and its CWT scalogram for fault distance of 10 km from the substa-
tion. The fault resistance was varied between 25 Ω and 400 Ω. From the
result, it can be seen that the transient signal becomes short and more
damped when the fault resistance is increased.
100
Voltage[V]
2000
0
−2000
Fault inception
50 100 150 200 250 300 350 400 450 500
40
37 0.12
34
31 0.1
Scales a
28
25 Charge transient component 0.08
22
19 0.06
16
13 Discharge transient component
10 0.04
7
4 0.02
1
50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(a)
7
X: 23 VEP
X: 14
6 Y: 5.68 Y: 6.85
5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b)
4
x 10
1.5
V
CWT
1
0.5
Voltage[V]
−0.5
−1
−1.5
0 50 100 150 200 250 300 350 400 450 500
Saample Number[k]
(c)
4500
V
4000 X: 703.1
Y: 3632
3500
Magnitude[FFT]
3000
2500
2000
1500
1000
500
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency (Hz)
(d)
10
V
HT
9.5
9
log(|V (t)|)
8.5
8
HT
7.5
6.5
5.5
0 50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(e)
Figure 6.4. Example of the output from pre-processing of voltage transient signals using
CWT and Hilbert transformation. The fault distance is 10 km from the sub-
station. (a) Simulated transient voltage (above) and its CWT scalogram (be-
low), (b) scale-dependent energy percentage distribution, (c) extracted CWT
coefficient, (d) FFT frequency spectrum of extracted CWT coefficient, (e) lin-
ear decaying of extracted CWT coefficient.
101
Current[A]
20
0
−20 Fault inception
50 100 150 200 250 300 350 400 450 500
40
37
34 0.15
31
Scales a
28
25 Charge transient component
22 0.1
19
16
13
10 0.05
7
4
1
50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(a)
9
I
EP
8 X: 24
Y: 8.182
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b)
200
I
CWT
150
100
Current[A]
50
−50
−100
−150
−200
0 50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(c)
80
I
70 X: 703.1
Y: 62.94
60
Magnitude[FFT]
50
40
30
20
10
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frequency (Hz)
(d)
5.5
I
HT
5
4.5
log(|I (t)|)
3.5
HT
2.5
1.5
1
0 50 100 150 200 250 300 350 400 450 500
Sample Number[k]
(e)
CWT and Hilbert transformation. The fault distance is 10 km from the sub-
station. (a) Simulated transient current (above) and its CWT scalogram (be-
low), (b) scale-dependent energy percentage distribution, (c) extracted CWT
coefficient, (d) FFT frequency spectrum of extracted CWT coefficient, (e) lin-
ear decaying of extracted CWT coefficient.
102
40 0.2
Current[A]
20
0.18
0
0.16
−20
0.14
−40
100 120 140 160 180 200 220 240 260 280 300
0.12
49
46
43
40 0.1
37
Scales a 34
31 0.08
28
25
22 0.06
19
16
13 0.04
10
7
4
1 0.02
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(a)
40 0.22
Current[A]
20 0.2
0
0.18
−20
0.16
−40
100 120 140 160 180 200 220 240 260 280 300
0.14
49
46
43 0.12
40
37
34 0.1
Scales a
31
28 0.08
25
22
19 0.06
16
13
10 0.04
7
4
1 0.02
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(b)
40 0.18
Current[A]
20
0.16
0
0.14
−20
100 120 140 160 180 200 220 240 260 280 300
0.12
49
46
43 0.1
40
37
34 0.08
Scales a
31
28
25
22 0.06
19
16
13 0.04
10
7
4
1 0.02
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(c)
CWT with the effect of fault resistance. The fault distance is 10 km from the
substation. (a) Simulated transient current (above) and its CWT scalogram
(below) with Rf =25 Ω, (b) Rf =50 Ω and (c) Rf =100 Ω.
103
50 0.14
Current[A]
0 0.12
−50 0.1
100 120 140 160 180 200 220 240 260 280 300
49
46
43 0.08
40
37
34
Scales a
31
28 0.06
25
22
19
16 0.04
13
10
7
4
1 0.02
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(a)
50 0.12
Current[A]
0.11
0
0.1
0.09
−50
100 120 140 160 180 200 220 240 260 280 300 0.08
49
46
43 0.07
40
37
34 0.06
Scales a
31
28 Scale = 25
0.05
25
22
19 0.04
16
13
10 0.03
7
4
1 0.02
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(b)
30 0.11
Current[A]
20 0.1
10
0.09
0
0.08
−10
100 120 140 160 180 200 220 240 260 280 300 0.07
49
46
43 0.06
40
37
34 0.05
Scales a
31 Scale = 28
28 0.04
25
22
19 0.03
16
13
10 0.02
7
4
1 0.01
100 120 140 160 180 200 220 240 260 280 300
Sample Number [k]
(c)
CWT with the effect of fault resistance. The fault distance is 10 km from the
substation. (a) Simulated transient current (above) and its CWT scalogram
(below) with Rf =200 Ω, (b) Rf =300 Ω and (c) Rf =400 Ω.
Figures 6.8 and 6.9 shows the scale-energy percentage distribution for
the simulated transient signal. Using the method for identification of
charge transient signal as explained in Chapter 4 Section 4.4, in this par-
ticular case the charge transient component can only be detected for earth
fault with fault resistance up to 200 Ω. However, if we observed the result
104
of CWT scalogram in Figure 6.7(b) and (c), we can make a guess estima-
tion what is the scale corresponding to the charge transient frequency by
inspecting the maxima of the energy. The maxima of the energy in Fig-
ure 6.7(b) is located between scale 22 and 28, while in Figure 6.7(c), it is
located between scale 22 and 37. The arrows in Figure 6.7(b) and (c) show
the area of the maxima of the energy. The purple line in Figure 6.7(b) and
(c), shows the guess estimation of the scales that correspond to the charge
transient frequency of the measured transient signal.
The calculated and estimated frequencies and damping factor of the
charge transient signal as a function of fault distance are shown in Fig-
ure 6.10, and the tabulated results are found in Appendix C Table C1.
Based on the result shown in Figure 6.10, the calculated and the esti-
mated values have a good agreement. The range of estimated damped
frequency is between 1132 Hz to 507 Hz. In Figure 6.10(a) and (b), the
graph shows that the damped and undamped charge transient frequency
is decreasing when the fault distance is increasing. The damping factor is
increased when the fault distance is increasing. The mean absolute error
(MAE) index of damped and undamped transient frequency is about 10
Hz, and 0.0006 for damping factor, as shown in Appendix C Table C1.
The results of fault distance estimation error using GM1 and GM2 algo-
rithms for simulated earth fault with different fault distances is given in
Figure 6.11, and the tabulated results are found in Appendix C Table C2.
As shown in Figure 6.11, basically the results from tested algorithms are
almost identical. In overall the distance calculation error is less than 1.5
km. The MAE index calculated for all algorithms is 0.36 km, as given in
Appendix C Table C2. The fault distance error is slightly higher when the
fault happens at the middle and at the end of the faulty feeder. From the
results, we found that the lowest fault distance estimation error is when
fault is simulated at 18 km from the substation. The reason why it has the
lowest fault distance estimation error is because based on the comparison
between the estimated (fdm ) and the calculated (fnm ) undamped charge
transient frequency, it has the lowest error (∆fn ) as given in Appendix C
Table C1.
105
4.5
4 X: 25
Y: 4.297
3.5
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(a)
4.5
4
X: 25
Y: 4.054
3.5
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b)
4
3.5 X: 26
Y: 3.675
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(c)
Figure 6.8. Example of scale-dependent energy percentage distribution. The fault dis-
tance is 10 km from the substation. (a) Scale-dependent energy percentage
distribution with Rf =25 Ω, (b) Rf =50 Ω and (c) Rf =100 Ω.
106
3.5
X: 31
3 Y: 3.253
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(a)
3.5
3
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(b)
4
3.5
2.5
1.5
0.5
0
0 5 10 15 20 25 30 35 40 45 50
Scales[a]
(c)
Figure 6.9. Example of scale-dependent energy percentage distribution. The fault dis-
tance is 10 km from the substation. (a) Scale-dependent energy percentage
distribution with Rf =200 Ω, (b) Rf =300 Ω and (c) Rf =400 Ω.
107
1200
fdc
fdm
1100
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(a) Damped transient frequencies as a function of fault distance
1200
fnc
fnm
1100
1000
Frequency [Hz]
900
800
700
600
500
2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(b) Undamped transient frequencies as a function of fault dis-

tance
0.05
ζc
ζm
0.04
0.03
0.02
0.01
0
2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(c) Damping factor as a function of fault distance
Figure 6.10. The comparison of simulation test result to the theory of earth fault tran-
sient. fdc , fnc and ζc are calculated using the GM1 equations. fdm , fnm
and ζm are estimated transient parameters from the simulated earth fault
transient waveforms. fdm = estimated damped transient frequency (Hz), fdc
= calculated damped transient frequency (Hz), ζm = estimated damping fac-
tor, ζc = calculated damping factor, fn = undamped transient frequency (Hz),
fnm = estimated undamped transient frequency (Hz) and fnc = calculated
undamped transient frequency (Hz)
108
2
GM1
fdm
1.5 GM1
fnm
GM2
1
Error[km]
0.5
−0.5
−1
2 4 6 8 10 12 14 16 18 20
Fault Distance[km]
Figure 6.11. Comparison of the distance estimates obtained using general model (GM)
algorithms for the simulated single line to earth faults with varies fault dis-
tance. GM1f dm denotes a distance calculation error using general model 1
(GM1) with damped transient frequency (fdm ), GM1f nm denotes a distance
calculation error using general model 1 (GM1) with undamped transient
frequency (fnm ) and GM2 denotes a distance calculation error using gen-
eral model 2 (GM2) algorithm.
Figures 6.12, 6.13, 6.14, 6.15 and 6.16 show the results of simulated
single line to earth fault transients with the effect of fault resistance,
and the tabulated results are found in Appendix C Table C3. In gen-
eral, referring to general model (GM) equivalent circuit, the transient
frequencies become lower when the resistance (R2 ) is increased. In Fig-
ure 6.12, the graphs show that the calculated and estimated frequencies
have small reduction when fault resistance is increased. The reduction
can be seen more clearly in estimated frequency when high fault resis-
tance (Rf = 100Ω) is used. In Figure 6.13, the estimated and calculated
frequency is almost identical except when fault resistance is 100 Ω.
Based on the results in Figure 6.13 (a), (b) and (c), for fault close to
the substation and for the certain range of the fault resistances, the gap
between estimated and calculated frequencies can be seen more clearly.
Based on the MAE index shown in Appendix C Table C3, the mean abso-
lute error (MAE) index of damped and undamped transient frequency is
about 22 Hz. The value is slightly higher compared to the results without
the effect of fault resistance. The reason is due to the effect of fault resis-
tance to the transient signal. The fault resistance increases the damping
of the estimated signal. The MAE of damping factor is 0.01, this is about
17 times higher than MAE of the damping factor without the effect of fault
resistance. When damping is higher the transient almost disappears and
subsequently the identification of the charge transient component will be
difficult.
109
1200
fdc−0Ω
1100 fdc−25Ω
fdc−50Ω
fdc−100Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(a) fdc for Rf = 0 - 100 Ω
1200
fdm−0Ω
1100 fdm−25Ω
fdm−50Ω
fdm−100Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(b) fdm for Rf = 0 - 100 Ω
1200
fnc−0Ω
fnc−50Ω
1100 fnc−25Ω
fnc−100Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(c) fnc for Rf = 0 - 100 Ω
1100 fnm−0Ω
fnm−25Ω
fnm−50Ω
1000 fnm−100Ω
Frequency [Hz]
900
800
700
600
2 4 6 8 10 12 14 16
Fault Distance [km]
(d) fnm for Rf = 0 - 100 Ω
Figure 6.12. Result of calculated and estimated damped (fdc , fdm ) and undamped
(fnc , fnm ) frequencies with the effect of fault resistance (Rf ). Fault resis-
tance (Rf )= 0, 25, 50, 100 Ω. For details of frequencies abbreviations see
Figure 6.10.
110
1200
fdc−0Ω
1100 fdm−0Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(a) fdc and fdm for Rf = 0 Ω
1200
fdc−25Ω
1100 fdm−25Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(b) fdc and fdm for Rf = 25 Ω
1200
fdc−50Ω
1100 fdm−50Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(c) fdc and fdm for Rf = 50 Ω
1200
fdc−100Ω
1100 fdm−100Ω
Frequency [Hz]
1000
900
800
700
600
500
2 4 6 8 10 12 14 16
Fault Distance [km]
(d) fdc and fdm for Rf = 100 Ω
Figure 6.13. Comparison of result of calculated (fdc ) and estimated (fdm ) damped fre-
quencies with the effect of fault resistance (Rf ). Fault resistance (Rf )= 0,
25, 50, 100 Ω. For details of frequencies abbreviations see Figure 6.10.
111
fnc−0Ω
1100 fnm−0Ω
Frequency [Hz]
1000
900
800
700
600
2 4 6 8 10 12 14 16
Fault Distance [km]
(a) fnc and fnm for Rf = 0 Ω
fnc−25Ω
1100 fnm−25Ω
1000
Frequency [Hz]
900
800
700
600
2 4 6 8 10 12 14 16
Fault Distance [km]
(b) fnc and fnm for Rf = 25 Ω
fnc−50Ω
1100 fnm−50Ω
Frequency [Hz]
1000
900
800
700
600
2 4 6 8 10 12 14 16
Fault Distance [km]
(c) fnc and fnm for Rf = 50 Ω
fnc−100Ω
1100 fnm−100Ω
Frequency [Hz]
1000
900
800
700
600
2 4 6 8 10 12 14 16
Fault Distance [km]
(d) fnc and fnm for Rf = 100 Ω
Figure 6.14. Comparison of result of calculated (fnc ) and estimated (fnm ) undamped fre-
quencies with the effect of fault resistance (Rf ). Fault resistance (Rf )= 0,
25, 50, 100 Ω. For details of frequencies abbreviations see Figure 6.10.
112
0.3
ζc−0Ω
ζc−25Ω
0.25 ζc−50Ω
ζc−100Ω
0.2
0.15
0.1
0.05
0
2 4 6 8 10 12 14 16
Fault Distance [km]
(a) ζc for Rf = 0 - 100 Ω
0.3
ζm−0Ω
ζm−25Ω
0.25 ζm−50Ω
ζm−100Ω
0.2
0.15
0.1
0.05
0
2 4 6 8 10 12 14 16
Fault Distance [km]
(b) ζm for Rf = 0 - 100 Ω
Figure 6.15. Result of calculated (ζc ) and estimated (ζm ) damping factor with the effect
of fault resistance (Rf ). Fault resistance (Rf )= 0, 25, 50, 100 Ω. For details
of frequencies abbreviations see Figure 6.10.
As shown in Figure 6.15, the increment of damping factor is significant

when an increased fault resistance. In comparison between estimated and
calculated damping factor as shown in Figure 6.16, the difference is small
for the fault resistance less than 25 Ω, however the gap can be noticed
clearly when fault resistance is higher than 25 Ω. In overall, with the
effect of fault resistance, the result of MAE of estimated and calculated
damped and undamped frequency, is two times higher than MAE of the
simulated earth fault without the effect of the fault resistance.
113
0.06
ζc−0Ω
ζm−0Ω
0.05
0.04
0.03
0.02
0.01
0
2 4 6 8 10 12 14 16
Fault Distance
(a) ζc and ζm for Rf = 0 Ω
0.08
ζc−25Ω
ζm−25Ω
0.075
0.07
0.065
0.06
0.055
0.05
2 4 6 8 10 12 14 16
Fault Distance [km]
(b) ζc and ζm for Rf = 25 Ω
0.12
ζc−50Ω
ζm−50Ω
0.115
0.11
0.105
0.1
0.095
0.09
0.085
0.08
2 4 6 8 10 12 14 16
Fault Distance [km]
(c) ζc and ζm for Rf = 50 Ω
0.4
ζc−100Ω
ζm−100Ω
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
2 4 6 8 10 12 14 16
Fault Distance [km]
(d) ζc and ζm for Rf = 100 Ω
Figure 6.16. Comparison of result of calculated (ζc ) and estimated (ζm ) damping factor
with the effect of fault resistance (Rf ). Fault resistance (Rf )= 0, 25, 50, 100
Ω. For details of frequencies abbreviations see Figure 6.10.
114
−0.7
GM1fdm
0.7
GM1fnm
Error[km]
0.6
GM2
0.3
0.1
−0.1
−0.3
−0.5
−0.7
2 4 10 16
Fault Distance[km]
(a) Fault location for Rf = 0 Ω

0.8
GM1fdm
0.6
GM1fnm
Error[km]
0.4
GM2
0.2
0
−0.2
−0.4
−0.6
−0.8
2 4 10 16
Fault Distance[km]
(b) Fault location for Rf = 25 Ω
4
GM1
3.5 fdm
GM1fnm
3
GM2
Error[km]
2.5
2
1.5
1
0.5
0
2 4 10 16
Fault Distance[km]
(c) Fault location for Rf = 50 Ω
4
GM1
3.5 fdm
GM1fnm
3
GM2
Error[km]
2.5
2
1.5
1
0.5
0
2 4 10 16
Fault Distance[km]
(d) Fault location for Rf = 100 Ω
Figure 6.17. Distance estimates with effect of fault resistance for general model net-
works.
115
Figure 6.17, shows the results of the estimated of fault location using
GM1 and GM2 algorithms with the effect of fault resistance, Appendix
C Table C4 shows the tabulated simulation results. Referring to Fig-
ure 6.17, we can see that result of the distance estimation error is almost
identical for the tested algorithms. In overall, the calculated MAE index
of all algorithms is less than 0.6 km, as given in Appendix C Table C4.
The fault distance error is lower when the fault is simulated at the mid-
dle of the faulty feeder with the fault resistance between 0 to 50 Ω. As
expected, the fault estimation error is higher when the fault is located at
the end of the line. In overall, the fault distance error is less than 2 km.
Appendix C, Table C5 shows the results of simulated single line to earth
fault with the effect of fault inception angle. The fault inception angles
used are 0, 45 and 90 ◦ . In overall, we found that the fault inception angle
does not affect the transient frequencies. Unfortunately, we found that
the process to locate the transient component becomes much challenging
since the lower fault inception angle will reduce the peak amplitude of
the transient, subsequently effecting the process of identification of the
charge transient component as shown in Figure 6.18.
In Figure 6.19, the results show estimates of fault location using GM1
algorithm. The fault distance estimation is calculated using a estimated
damped and undamped transient frequency. A single line to earth fault
was simulated with zero fault resistance. The MAE index of the simula-
tion result was 0.34 km.
Figure 6.20 shows results of calculated damped transient frequencies
with the effect of load and fault resistance. The frequency slightly in-
creases when the load is increased and slightly decreases when the fault
resistance is increased. The effect of load on damped transient frequency
is higher when the fault located close to the substation. Figure 6.21,
shows the estimated damped transient frequencies with the effect of load
and fault resistance. The estimated damped frequency shows significant
decrease when the fault resistance is increased.
116
Current[A]
0
−5
50 100 150 200 250 300
30 0.2
Scales [a]
25
20 0.15
15
0.1
10
5 0.05
50 100 150 200 250 300

Sample Number[k]
(a) Transient current (above) and its scalogram, fault inception angle 0◦
Current[A]
20
0
−20
−40
50 100 150 200 250 300
30
0.2
25
Scales [a]
20 0.15
15
0.1
10
5 0.05
50 100 150 200 250 300
Sample Number[k]
(b) Transient current (above) and its scalogram, fault inception angle 45 ◦
Figure 6.18. Comparison of simulated transient signal and its scalogram with the effect
of inception angle. The fault distance is 4 km and without fault resistance.
0.6
GM1fdm
0.4 GM1fnm
Error(km)
0.2
−0.2
−0.4
−0.6
490◦ 445◦ 40 ◦ 1690◦ 1645◦ 160◦
Fault Distance(km)Inception angle
Figure 6.19. Effect of fault inception angle on fault distance estimation. GM1f dm de-
notes a distance estimation error using GM1 with damped transient fre-
quency and GM1f nm denotes a distance estimation error using GM1 with
undamped transient frequency.
117
1100
fdc−0Ω−A
fdc−25Ω−A
1000 fdc−50Ω−A
Frequency [Hz]
fdc−100Ω−A
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(a) fdc for load A = 5 MVA Rf = 0 - 100 Ω
1100
fdc−0Ω−B
fdc−25Ω−B
1000 fdc−50Ω−B
fdc−100Ω−B
Frequency [Hz]
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(b) fdc for load B = 12 MVA Rf = 0 - 100 Ω
1100
fdc−0Ω−A
fdc−100Ω−A
1000 fdc−0Ω−B
fdc−100Ω−B
Frequency [Hz]
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(c) fdc for load A and B , Rf = 0 and 100 Ω
Figure 6.20. Comparison of calculated damped frequencies with the effect of load and
fault resistance
118
1000 fdm−0Ω−A
fdm−25Ω−A
fdm−50Ω−A
Frequency [Hz]
900 fdm−100Ω−A
800
700
600
4 6 8 10 12 14 16
Fault Distance [km]
(a) fdm for load A = 5 MVA Rf = 0 - 100 Ω
1100
fdm−0Ω−B
fdm−25Ω−B
1000 fdm−50Ω−B
Frequency [Hz]
fdm−100Ω−B
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(b) fdm for load B = 12 MVA Rf = 0 - 100 Ω
1100
fdm−0Ω−A
fdm−100Ω−A
1000 fdm−0Ω−B
Frequency [Hz]
fdm−100Ω−B
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(c) fdm for load A and B , Rf = 0 and 100 Ω
Figure 6.21. Comparison of estimated damped frequencies with the effect of load and
fault resistance
119
Figure 6.22, shows the effect of load and fault resistance on calculated
undamped transient frequency. It can be seen that, with the effect of 5
MVA and 12.5 MVA loads, the calculated undamped transient frequency
is almost not affected by fault resistance. Based on the result, it is found
that the undamped frequency with a higher load (12.5 MVA) has higher
frequency than the undamped frequency with a lower load (5 MVA). The
results shown in Figure 6.23, tell us that the estimated undamped fre-
quency with high load and lower fault resistance is greater than the esti-
mated undamped frequency having low load and high fault resistance.
In Figure 6.24(d), it can be seen that, the estimated and calculated
damping factor proved to be identical with small error. When the fault
resistance in increased the damping factor is increased as well. The re-
sult shows that, the damping factor with load B is lower than the damping
factor with load A.
The bar charts shown in Figures 6.25 and 6.26, present a comparison
of earth fault distance estimation error by GM algorithms when calculat-
ing earth fault distance in the simulated network. Appendix C, Table C8
shows tabulated overall distance estimation results. A single line to earth
fault was simulated at 4 km, 10 km and 16 km from substation for loads
5 MVA and 12.5 MVA. Earth faults were simulated with fault resistance
values of 0 Ω , 25 Ω, 50 Ω and 100 Ω.
In overall, based on the results, we found that the fault distance esti-
mation error is less than 1 km when the fault resistance is below 50 Ω.
However, when fault resistance is 100 Ω, the estimation error is increased.
In Figures 6.25 and 6.26, when fault happens at 4 km and with the fault
resistance less than 25 Ω, GM3 algorithm calculates the fault distance
with the lower error compared with the other algorithms. Unfortunately,
when the fault distance is simulated at 10 km, GM3 algorithm gives the
highest fault estimation error among all algorithms. It can be seen as
well that GM3 algorithm has the highest estimation error when the earth
fault happens at fault distance 16 km and with fault resistance 100 Ω.
The results from GM1, GM2 and GM4 algorithm are almost identical for
all cases of simulated earth faults. In overall, the MAE error of estimated
fault distance using proposed GM algorithms is less than 0.6 km, when
fault resistance is 50 Ω or less.
120
1100
fnc−0Ω−A
fnc−25Ω−A
1000 fnc−50Ω−A
Frequency [Hz]
fnc−100Ω−A
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(a) fnc for load A = 5 MVA Rf = 0 - 100 Ω
1100
fnc−0Ω−B
fnc−25Ω−B
1000 fnc−50Ω−B
Frequency [Hz]
fnc−100Ω−B
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(b) fnc for load B = 12 MVA Rf = 0 - 100 Ω
1100
fnc−0Ω−A
fnc−100Ω−A
1000 fnc−0Ω−B
Frequency [Hz]
fnc−100Ω−B
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(c) fnc for load A and B , Rf = 0 and 100 Ω
Figure 6.22. Comparison of calculated undamped frequencies with the effect of load and
fault resistance
121
1100
fnm−0Ω−A
fnm−25Ω−A
1000 fnm−50Ω−A
Frequency [Hz]
fnm−100Ω−A
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(a) fnm for load A = 5 MVA Rf = 0 - 100 Ω
1100
fnm−0Ω−B
fnm−25Ω−B
1000 fnm−50Ω−B
fnm−100Ω−B
Frequency [Hz]
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(b) fnm for load B = 12 MVA Rf = 0 - 100 Ω
1100
fnm−0Ω−A
fnm−100Ω−A
1000 fnm−0Ω−B
fnm−100Ω−B
Frequency [Hz]
900
800
700
600
500
4 6 8 10 12 14 16
Fault Distance [km]
(c) fnm for load A and B , Rf = 0 and 100 Ω
Figure 6.23. Comparison of estimated undamped frequencies with the effect of load and
fault resistance
122
ζc−0Ω−A
0.3 ζc−25Ω−A
ζc−50Ω−A
ζc−100Ω−A
0.25 ζc−0Ω−B
ζc−25Ω−B
0.2 ζc−50Ω−B
ζc−100Ω−B
0.15
0.1
0.05
4 6 8 10 12 14 16
Fault Distance[km]
(a) ζc for load A and B, Rf = 0 - 100 Ω
ζm−0Ω−A
ζm−25Ω−A
ζm−50Ω−A
0.3 ζm−100Ω−A
ζm−0Ω−B
ζm−25Ω−B
0.25 ζm−50Ω−B
ζm−100Ω−B
0.2
0.15
0.1
0.05
4 6 8 10 12 14 16
Fault Distance [km]
(b) ζm for load A and B, Rf = 0 - 100 Ω
ζc−0Ω−A
ζc−25Ω−A
ζc−50Ω−A
0.3 ζc−100Ω−A
ζm−0Ω−A
ζm−25Ω−A
0.25 ζm−50Ω−A
ζm−100Ω−A
0.2
0.15
0.1
0.05
4 6 8 10 12 14 16
Fault Distance [km]
(c) ζc and ζm for load A , Rf = 0 - 100 Ω
ζc−0Ω−B
ζc−25Ω−B
ζc−50Ω−B
0.3 ζc−100Ω−B
ζm−0Ω−B
ζm−25Ω−B
0.25 ζm−50Ω−B
ζm−100Ω−B
0.2
0.15
0.1
0.05
4 6 8 10 12 14 16
Fault Distance [km]
(d) ζc and ζm for load B , Rf = 0 - 100 Ω
Figure 6.24. Calculated and estimated damping factor (ζ) with the effect of load and fault
resistance.
123
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
Figure 6.25. Distance estimates with the effect of 5 MVA load and fault resistances for
general model networks.
124
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
2
GM1
GM2
1.5 GM3
GM4
1
Error[km]
0.5
−0.5
−1
4 10 16
Fault Distance[km]
Figure 6.26. Distance estimates with the effect of 12 MVA load and fault resistances for
general model networks.
125
In this section, we discuss the accuracy of the single line to earth fault lo-
cation algorithms that has been explained in Chapter 5, Section 5.2. The
fault location algorithms used in this section are the accurate model EM1
and for the comparison, the simple model GM2. The accuracy of the meth-
ods was tested using the simulation model and the network example with
overhead lines. The performance accuracy of EM1 and GM2 algorithm
was investigated in various fault conditions. For EM algorithm, the per-
formance accuracy was defined by the mean absolute error (MAEEM ) as
expressed as in Equation (6.3) below, while for GM algorithm, the perfor-
mance accuracy was defined as given in Equation (6.1).
n n
1 1
M AEEM = |EMi − li | = |∆EMi | (6.3)
n n
i=1 i=1
In 6.3, ∆EMi is the absolute error fault distance using EM algorithm,

EMi is estimated fault distance using EM algorithm, li is exact fault dis-
tance and n is numbers of observation of fault distance error.
The single line diagram of the simulated unearthed medium voltage (MV)
network and its simulation model is shown in Figure 6.27. The simula-
tion model is based on the sequence component networks for the exact-π
solution for the earth fault transients. The system is 21 kV overhead
lines unearthed MV radial network and the line parameters are as fol-
lows: Rl,p =0.6 Ω/km, Rl,0 =1.3 Ω/km, Ll,p =1.0 mH/km, Ll,0 = 5.0 mH/km,
Cl,p =10.71 nF/km, Cl,0 =6.12 nF/km, RT =0.0216 Ω, LT =2.8 mH. We assume
the negative sequence parameters to be equal to the positive sequence
ones.
126
1 17km
2 15km
3 22km
110/21kV
110 kV
Supply 4 24km
25MVA
5 20km
M
Zf
(a) A schematic diagram of 21 kV, 98 km, un-

earthed neutral MV network
F R0|5 L0|5 R0|j L0|j

M
C0|5 C0|5 C0|j C0|j "0"

2 2 2 2
3Zf
j=1...4
Rl|5 Ll|5 M R1|j L1|j
E L1T
Cl|5 Cl|5 C1|j C1|j "1"
2 2 2 2
R1T
j=1...4
R2|5 L2|5 M R2|j L2|j
L2T
C2|5 C2|5 C2|j C2|j "2"
2 2 2 2
R2T
j=1...4
(b) Simulation model based on symmetrical component equivalent circuit

for an earth fault located at the end of the faulty feeder
Figure 6.27. A schematic network diagram and its simulation model for a single phase
to earth fault in a distribution network. M denotes measurement point, F
refers to the fault location and Zf is the fault impedance. L0|5 , L1|5 , L2|5
are zero sequence, positive sequence and negative sequence inductance of
faulty line located in front of fault point (F ). L0|j , L1|j are zero sequence and
positive sequence inductance of sound line. LT and RT is the inductance and
resistance of the transformer, capacitances and resistances correspondingly.
The lines were modelled as a chain of exact π-model cells (1 cell per
10 km of the line). The network was 5 feeders and the earth fault was
simulated at feeder 5. The total length of the network is 98 km. All
faults with several different fault conditions were simulated with the soft-
ware package ATP (Alternative Transients Program), version of EMTP
127
program where the circuit was realized using ATPDraw. The sampling
frequency was 20 kHz. In the simulation analysis, the load was either
0.4 MW, 0.8 MW, 2 MW or zero.
Figure 6.28 shows the results of the simulation without the effect of load
and with the effect of fault distance, fault resistance and fault inception
angle. The tabulated results can be found in Appendix C Table C9, C10
and C11, respectively. In order to study the performance of the tested
algorithms, in Figure 6.28(a), an earth fault was simulated at 10 points
starting from 2 km from measuring point (M) with 2 kilometer steps. In
case of the effect of fault resistances as shown in Figure 6.28(b), 3 different
locations of earth fault was simulated which are 4 km, 10 km and 16 km
from the measuring point. Only two fault distances were simulated for
the case of fault inception angle variation, which are 4 km and 16 km, as
presented in Figure 6.28(c). Based on the MAE results given in Appendix
C Table C9, C10 and C11, it can be seen, that the EM1 algorithm provides
better results than the GM2 algorithm.
As shown in Figure 6.28(a), the estimation error is increased with the in-
crease of fault distance for both algorithms. GM2 algorithm recorded the
highest error, when earth fault was located at 20 km from the substation.
For the earth fault distance of 4km, 10km and 16km, the estimation error
is further increased when fault distance is increased with the increased
of fault resistance, as we seen in Figure 6.28(b).
In Figure 6.28(c), it seems that fault inception angle does not affect both
algorithms when fault happens at fault distance 4 km, however, when
fault is located at 16 km the error is increasing for EM1 algorithm but
decreasing for GM2 algorithm.
If we carefully observe the results in Figure 6.28, Table C10 and Ta-
ble C11 in Appendix C, in case of fault distance located at 16 km from
fault distance, for GM2 algorithm estimation result, indirectly the effect
of increased of fault resistance and fault inception angle at 0 degree is de-
creasing the estimated undamped charge transient frequency (fnm ) and
at the same time improving the algorithm accuracy.
128
2
EM1
GM2
1.5
0.5
Error(km)
0
−0.5
−1
−1.5
−2
2 4 6 8 10 12 14 16 18 20
Distance(km)
(a) Calculation error as a function of fault distance

2
EM1
GM2
1.5
1
Error(km)
0.5
−0.5
−1
−1.5
−2
410 4 4 4100 10 10 10 10 1610 16 16 16
30 50 10 30 50 100 30 50 100
Distance(km)Rf (Ω)
(b) Calculation error as a function of fault distance and fault resistance
2
EM1
GM2
1.5
1
Error(km)
0.5
−0.5
−1
−1.5
−2
490 445 40 1690 1645 160
Distance(km)Inception angle
◦
(c) Calculation error as a function of fault inception angle
Figure 6.28. The effect of fault distance, fault resistance and inception angle on fault
distance calulation. EM1 denotes the distance calculation error using ex-
act model 1 and GM2 denotes the distance calculation error using general
model 2.
129
Figure 6.29 shows the results of fault location performance with the
effect of resistive load located at the end of each feeder. The tabulated
results can be found in Appendix C Table C12. In this part of simulation,
each feeder has same amount of load with three different cases which
were 400 kW, 800 kW and 2 MW.
Based on the MAE results given in Appendix C Table C12, it can be
seen, that the EM1 algorithm again provides better results than the GM2
algorithm. In overall, both algorithms have an increase of total MAE
when resistive load is added to the simulated network.
As shown in Figure 6.29, the fault distance error seems to be increased
when the load is increased. The more far the fault happens from the
substation the higher is the fault distance error. In Figure 6.29, in some
cases corresponding to GM2, we can see that the fault distance error is
slightly decreased when fault resistance is increased. In overall, the fault
distance estimation error is less than 1.5 km.
Figure 6.30 shows the results of fault fault location performance with
the effect of inductive load located at the end of each feeder. The tabulated
results can be found in Appendix C Table C13. In this part of simulation,
each feeder has same amount of load with three difference cases which
were 400 kVAr, 800 kVAr and 2 MVAr. Corresponding to the overall MAE
result shown in Appendix C Table C13, EM1 algorithm provides better
results than the GM2 algorithm.
It is found that the performance accuracy for EM1 algorithm is increased
with the effect of inductive load compared to than when it is been affected
by resistive load. However, in case of GM2 algorithm, the performance
accuracy is decreased with the effect of inductive load. In Figure 6.30, in
overall, the results show that the fault distance estimation error is less
than 1.5 km.
130
1.5
EM1
GM2
0.5
Error(km)
0
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(a) 400 kW
1.5
EM1
GM2
1
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(b) 800 kW
1.5
EM1
GM2
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(c) 2 MW
Figure 6.29. Fault location performance with the effect of resistive load located at the
end of each feeder. EM1 denotes a distance calculation error using exact
model 1 and GM2 denotes a distance calculation error using general model
2.
131
1.5
EM1
GM2
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(a) 400 kVAr

1.5
EM1
GM2
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(b) 800 kVAr
1.5
EM1
GM2
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1030 1050 10100 1830 1850 18100
(c) 2 MVAr
Figure 6.30. Fault location performance with the effect of inductive load located at the
end of each feeder. EM1 denotes a distance calculation error using 1 and
GM2 denotes a distance calculation error using general model 2.
132
Figure 6.31 shows the results of fault fault location performance with
the effect of load for 0.9 and 0.5 power factor. The tabulated results can be
found in Appendix C Table C14 and Table C15, respectively. In this part
of simulation, each feeder has same amount of load with two difference
cases which were 400 KVA and 800 KVA.
Based on the MAE result given in Appendix C Table C14 and Table C14,
it can be seen, that the EM1 algorithm provides again better results than
the GM2 algorithm. By comparison of MAE results in Appendix C Ta-
ble C14 and Table C14, we found that both algorithms have a decrease of
MAE when low power factor is used.
In this work, the most influential source of error was resistive load.
Theoretically, the load can be compensated by adding the corresponding
impedance to the algorithm models. However, the data of load devices and
their transient impedances is difficult to get. Another way is to use the
curve fitting technique to compensate the error with known load condi-
tions.
Another source of error is inaccuracy in estimating the undamped com-
plex frequency. This error is caused by the inaccuracy in estimation of the
damping attenuation through Hilbert transformation method and also by
an inaccurate detection of charge transient component using CWT. When
the magnitude of fault resistance or resistive loads is increased, the tran-
sients become more damped, with the result that the estimation of the
charge transient frequency becomes more difficult which affects the over-
all computation of fault distance. In following section, the investigation of
performance accuracy of proposed algorithms is performed using JMarti
(frequency-dependent) line model which is more practical compared to the
exact-π solution that we presented in this section.
Considering which one of the algorithms is easy to be implement in real
operation, the general model algorithm should be much easier to be imple-
ment compare to the exact model algorithm. It is because its required a
simple iteration process to estimate the fault distance. While exact model
algorithm required a heavy computation to solve a higher order equation,
which generally may not be easy.
133
1.5
EM1
GM2
1
0.5
Error(km)
0
−0.5
−1
−1.5
2 250 2100 1830 1850 18100
30
(a) 400 kVA, 0.9 PF
1.5
EM1
GM2
1
0.5
Error(km)
−0.5
−1
−1.5
2 2 2100 1830 1850 18
30 50 100
(b) 800 kVA, 0.9 PF
1.5
EM1
GM2
0.5
Error(km)
−0.5
−1
−1.5
2 2 2100 1830 1850 18100
30 50
(c) 400 kVA, 0.5 PF
1.5
EM1
GM2
1
0.5
Error(km)
−0.5
−1
−1.5
230 250 2100 1830 1850 18
100
(d) 800 kVA, 0.5 PF
Figure 6.31. Fault location performance with the effect of load with 0.9 and 0.5 PF lo-
cated at the end of each feeder. EM1 denotes the distance calculation error
using exact model 1 and GM2 denotes the distance calculation error using
general model 2.
134
In this section, we investigate the proposed fault location algorithms as

described in Chapter 5, which use the earth fault transient signals mea-
sured at the secondary sides of MV and LV distribution transformers to
locate the single line to earth fault. The performance accuracy of the
wavelet transform (WT) version of the algorithms, is defined by mean ab-
solute error (MAEWT ) as expressed as in Equation (6.4) below:
n n
1 1
M AEW T = |W Ti − li | = |∆W Ti | (6.4)
n n
i=1 i=1
In 6.4, ∆W Ti is the absolute error fault distance using WT algorithm, W Ti

is estimated fault distance using WT algorithm, li is exact fault distance
and n is numbers of observation of fault distance error.
A 20 kV, 227.6 km overhead lines, unearthed neutral MV network was

modelled using the JMarti line constants of ATP/EMTP program taking
into account the real geometrical and electrical values. A schematic dia-
gram of the simulated unearthed neutral medium voltage (MV) network
and its ATPdraw circuit are shown in Figures 6.32 and 6.33, respectively.
The configuration of the feeders and the ATPDraw circuit are given in
Appendix B, Figure B1. The overhead lines were energized by AC sinu-
soidal waveform via 110/20 kV, delta/star unearthed neutral transformer.
An earth fault was simulated at 11 difference places as indicated in Fig-
ure 6.32. Each of the simulated earth faults were performed for fault
resistances of 0 Ω, 25 Ω and 50 Ω. For each case the fault inception angle
used were 45◦ and 90◦ . The sampling frequency used was 20 kHz.
135
N23
5.6km
N1 N2
44km 6
N21 N22 0.1MVA
N3 N4 1MVA
3.54km 5.28km Zone C
32km
N19 N20
0.1MVA
3.89km 2.45km
N5 N6 1MVA
25MVA
36km 4
0.1MVA 0.1MVA
N18
N7 N8 1MVA
3.98km
28km
5
1MVA 0.1MVA
N9 N11 N12 N13
2.8km
N10
3.5km 5.4km 3.8km 2.1km
N14
3.9km
N15
2.7km
N16 N17
Zone B
2.6km
1 3
0.1MVA 0.1MVA 0.1MVA 0.1MVA 0.1MVA 0.1MVA
2 0.1MVA
N24 N25 N29

3.88km 6.0km 5.48km
8 0.1MVA 9
0.1MVA
N26 N27 N28
7
Legend: 3.98km 2.68km 8.58km
Fault position
11
Zone A
x Fault number 0.1MVA
N30 N31
10
NX Node number X 0.1MVA 2.76km 2.7km
Zone
0.1MVA 0.1MVA
Figure 6.32. A schematic diagram of 20 kV, 227.6 km, unearthed neutral MV network.
Figure 6.33. ATPdraw circuit for the simulated network model 3.
136
The results of single line to earth fault estimation with WT algorithm are
shown in Tables 6.1 and 6.2. In overall, MAE of WT algorithm is 1.09 km.
In Tables 6.1 and 6.2, it is shown that the fault location error is increas-
ing when the fault is located far from the measuring point and the fault
inception angle is small. Also the fault location error is higher when the
fault is located far from the measuring point and has a high fault resis-
tance. In overall, the fault distance estimation error is less than 5.2 km
and 6.52 km for simulation with test angle of 90◦ and 45◦ , respectively. It
is found that the test number with the longest fault distance (test num-
ber 5 (26.19 km)) calculated the highest fault location error. From our
experience in analyzing the transient signal, referring to Tables 6.1 and
6.2, when the magnitude of the fault resistance is increased the transient
signals become more damped and the fault location is more difficult.
Figures 6.34, 6.35 , 6.36 , 6.37 and 6.38, show results of calculated CWT
FFT amplitudes of transient at MV/LV substation for fault at inception
angles of 90◦ and 45◦ which are used in correct path identification algo-
rithm. Based on the correct path identification algorithm, the highest
recorded FFT amplitude will indicate which MV/LV substation should we
locate in order to determine the correct fault location if there are several
branches in the network. Also, based on the test results, another assump-
tion is that if the fault distance estimation shows a large error in its cal-
culation, the correct path algorithm can still give us a good guess about
the fault position. However, this assumption cannot be true for all fault
positions. For instance, in Table 6.1, the fault is located at test number 1
which is 1.4 km from measuring point and the LV substation detected is
at substation S23. However, the nearest LV substation to the fault point
is substation S10. Similarly, for fault located at test number 2, 3, 4, 7
and 10. Therefore, the assumption can be used only if the fault happens
exactly at the section which located at the end of the feeder branches such
as the faults located at test number 5, 6, 8, 9 and 11.
In this work, the zones were defined based on the location of the MV/LV
substation located in the same main branches. For examples, it can be
seen that MV/LV substations number S25, S26, S28, S29, S30 and S31
located in zone A, MV/LV substations number S14, S15, S16, S17 and S18
is located at zone B and MV/LV substations number S19, S20, S22 and
S23 is located at zone C.
137
Table 6.1. Test results of single line to earth fault for 90◦ fault inception angle using the
WT algorithm.
Test Test
Substation detected Fault location error (km) MAE
number Section Candidates
(distance) / Node Rf 0 Rf 25 Rf 50 Rf 0 Rf 25 Rf 50 (km)
√
1 (1.4km) N9-N10 N9-N10 S23 S23 S23 -0.07 -0.43 -1.75 0.75
√
N11-N13
2 (11.7km) N12 N24-N25 S20 S20 S20 0.53 0.52 -0.02 0.36
N24-N26
√
N15-N16
3 (22.8km) N15-N16 S18 S18 S18 -0.07 -0.38 -1.44 0.63
N19-N21
N13-N14
√
N13-N19
4 (17.5km) N13-N19 N27-N28 S20 S20 S20 0.58 -1.46 -0.84 0.96
N27-N29
N30-N31
N16-N17
√
N16-N18
5 (26.19km) N16-N18 S18 S18 S18 -2.09 -3.09 -5.11 3.43
N21-N23
N21-N22
N21-N22
√
N21-N23
6 (25.83km) N21-N23 S23 S23 S23 -0.87 -1.24 -1.64 1.25
N16-N17
N16-N18
N11-N12
√
7 (10.18km) N24 N11-N25 S29 S25 S25 0.75 0.05 0.91 0.57
√
N11-N26
N12-N13
√
8 (12.53km) N24-N25 N24-N25 S25 S25 S25 0.64 0.58 0.25 0.49
N24-N26
N14-N15
N19-N20
9 (19.72km) N27-N29 N19-N21 S29 S29 S29 -0.6 -1.44 -2.16 1.4
N27-N28
√
N27-N29
N12-N13
√
N24-N27
10 (14.16km) N26 S28 S28 S28 0.52 0.37 -0.78 0.52
N24-N30
N24-N25
√
N27-N28
N27-N29
11 (21.72km) N27-N28 N16-N17 S28 S28 S28 0.34 0.18 -3.11 1.21
N16-N18
N19-N21
All simulation tests 1.05

√
*Note: Symbol( ) in the table indicate the correct section
138
Table 6.2. Test results of single line to earth fault for 45◦ fault inception angle
Test Test
Substation detected Fault location error (km) MAE
number Section Candidates
(distance) / Node Rf 0 Rf 25 Rf 50 Rf 0 Rf 25 Rf 50 (km)
√
1 (1.4km) N9-N10 N9-N10 S23 S23 S28 0.15 -1.27 -2.06 1.16
√
N11-N13
2 (11.7km) N12 N24-N25 S15 S15 S14 0.55 0.83 -0.23 0.54
N24-N26
√
N15-N16
3 (22.8km) N15-N16 S18 S18 S18 0.52 -0.08 -0.42 0.34
N19-N21
N13-N14
√
N13-N19
4 (17.5km) N13-N19 N27-N28 S20 S20 S20 -0.07 0.74 0.46 0.42
N27-N29
N30-N31
N16-N17
√
N16-N18
5 (26.19km) N16-N18 S18 S18 S18 -2 -2.26 -6.52 3.59
N21-N23
N21-N22
N21-N22
√
N21-N23
6 (25.83km) N21-N23 S23 S23 S23 -1.38 -1.9 0.03 1.1
N16-N17
N16-N18
N11-N12
7 (10.18km) N24 N11-N25 S31 S31 S31 0.96 0.45 -0.16 0.52
√
N11-N26
N12-N13
√
8 (12.53km) N24-N25 N24-N25 S25 S25 S25 0.68 0.92 0.11 0.57
N24-N26
N14-N15
N19-N20
9 (19.72km) N27-N29 N19-N21 S29 S29 S29 -0.67 -0.94 -3.84 1.82
N27-N28
√
N27-N29
N12-N13
√
N24-N27
10 (14.16km) N26 S29 S29 S29 -0.84 0.68 -2.32 1.28
N24-N30
N24-N25
√
N27-N28
N27-N29
11 (21.72km) N27-N28 N16-N17 S28 S28 S28 -0.19 -0.14 -2.44 0.92
N16-N18
N19-N21
All simulation tests 1.12

√
*Note: Symbol( ) in the table indicate the correct section
139
200
Rf=0
R =25
f
150 R =50
f
Amplitude[FFT]
100
50
0
S10 S12 S13 S14 S15 S16 S17 S18 S19 S20 S22 S23 S25 S26 S28 S29 S30 S31
Subtation
(a) 90◦ fault inception angle

120
R =0
f
100 R =25
f
Rf=50
Amplitude[FFT]
80
60
40
20
0
Substation
(b) 45◦ fault inception angle
Figure 6.34. CWT FFT amplitude of MV/LV substation for fault test number 1 with fault
inception angle of 90◦ and 45◦ . The arrow indicates the substation with the
highest amplitude.
From Figure 6.35, for example CWT FFT amplitude of MV/LV substa-
tions for test number 7, it can be seen that MV/LV substations number
S25, S26, S28, S29, S30 and S31 estimated almost the same CWT FFT
amplitude value, where substation S29 estimated the highest value. S25,
S26, S28, S29, S30 and S31 is actually placed inside zone A, as shown in
Figure 6.32. Therefore, if any one of these substations indicated the high-
est amplitude for test number 7 earth fault, we can make an assumption
that the correct path should go through that zone. Similar assumption
can be used for test number 8, 9, 10 and 11. The assumption also can
further be used for fault located at other zone, for instant test number 3,
referring to result in Figure 6.36, where it shows that CWT FFT ampli-
tude of MV/LV substations for S15, S16, S17 and S18 have close value and
fault is located at Zone B. Therefore, based from the earlier studies of the
simulation results, zoning the branches gives the first indication of the
correct path of fault passage in the network. However, this assumption
cannot be used for exact correct path of fault location since more branches
were located inside the zone. From the simulation results in Figures 6.34,
6.35, 6.36, 6.37 and 6.38, the correct path algorithm managed to identify
the correct LV substation to be used as a guidance of calculating the fault
distance from the measuring point towards the fault position. Figures
6.39, 6.40, 6.41, 6.42 and 6.43 show an example of schematic diagram of
correct path identification for test number 2, 3, 4, 6, and 8, respectively.
140
200
R =0
f
180 Rf=25
R =50
160 f
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation
(a) Test number 7 (10.18 km)

200
Rf=0
180 Rf=25
Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation
(b) Test number 2 (11.70 km)

200
Rf=0
180 Rf=25
Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation
(c) Test number 8 (12.53 km)

220
Rf=0
200 R =25
f
180 Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Subtation
(d) Test number 10 (14.16 km)

250
Rf=0
Rf=25
Rf=50
200
Amplitude [FFT]
150
100
50
0
Substation
(e) Test number 4 (17.50 km)
Figure 6.35. CWT FFT amplitude of MV/LV substation for fault test number 7, 2, 8, 10
and 4 with fault inception angle of 90◦ . The arrow indicates the substation
with the highest amplitude.
141
220
Rf=0
200 Rf=25
180 Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation
Rf=0
250 R =25
f
Rf=50
Amplitude [FFT]
200
150
100
50
0
Substation

300
Rf=0
Rf=25
250 Rf=50
Amplitude [FFT]
200
150
100
50
0
Substation

250
RF=0
RF=25
200 RF=50
Amplitude [FFT]
150
100
50
0
Substation
Rf=0
300
Rf=25
Rf=50
250
Amplitude [FFT]
200
150
100
50
0
Substation
142
200
Rf=0
180 Rf=25
R =50
160 f
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

200
Rf=0
180 R =25
f
Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

200
Rf=0
180 Rf=25
R =50
160 f
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

200
Rf=0
180 Rf=25
Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

220
Rf=0
200 Rf=25
180 Rf=50
Amplitude [FFT]
160
140
120
100
80
60
40
20
0
Substation
143
220
Rf=0
200 Rf=25
180 Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

250
Rf=0
Rf=25
Rf=50
200
Amplitude [FFT]
150
100
50
0
Substation

250
Rf=0
R =25
f
Rf=50
200
Amplitude [FFT]
150
100
50
0
Substation

200
Rf=0
180 Rf=25
Rf=50
160
Amplitude [FFT]
140
120
100
80
60
40
20
0
Substation

250
Rf=0
Rf=25
Rf=50
200
Amplitude [FFT]
150
100
50
0
Substation
144
N1 N2
44km
N3 N4 1MVA N23
32km 5.6km
N5 N6 1MVA N21 0.1MVA

25MVA N22
36km 3.54km 5.28km
N7 N8 1MVA N19 N20 0.1MVA N18

28km 3.89km 2.45km 3.98km
N9 N10 1MVA N11 N12 N13 N14 0.1MVA N15 0.1MVA N16 N17 0.1MVA
2.8km 3.5km 5.4km 3.8km 2.1km 3.9km 2.7km 2.6km
0.1MVA 0.1MVA 0.1MVA 0.1MVA 0.1MVA 0.1MVA

2 0.1MVA
6.0km 5.48km
N24 N25 N29
3.88km
N26 0.1MVA N27 N28 0.1MVA
3.98km 2.68km 8.58km
Legend:
N30 N31 0.1MVA
Fault position
0.1MVA 2.76km 2.7km
x Fault number
Correct path
NX Node number X 0.1MVA 0.1MVA
Figure 6.39. Test number 2 (11.70 km) correct path identification.
N1 N2
44km
N3 N4 1MVA N23
32km 5.6km
N5 N6 1MVA N21 N22 0.1MVA

25MVA
36km 3.54km 5.28km
N7 N8 1MVA N19 N20 0.1MVA N18

28km 3.89km 2.45km 3.98km
0.1MVA 3
0.1MVA N24 0.1MVA N25 0.1MVA N29 0.1MVA 0.1MVA
0.1MVA
3.88km 6.0km 5.48km
N26 0.1MVA N27 N28 0.1MVA

3.98km 2.68km 8.58km
Legend:
N30 N31 0.1MVA
Fault position
0.1MVA 2.76km 2.7km
x Fault number
Correct path
NX Node number x 0.1MVA 0.1MVA
145
N1 N2
44km
N3 N4 1MVA N23
32km 5.6km
25MVA
N5 N6 1MVA N21 N22 0.1MVA
36km 3.54km 5.28km
N7 N8 1MVA N19 N20 0.1MVA N18

28km 3.89km 2.45km 3.98km
4
0.1MVA N24 0.1MVA N25 0.1MVA 0.1MVA N29 0.1MVA 0.1MVA

0.1MVA
3.88km 6.0km 5.48km
N26 0.1MVA N27 N28 0.1MVA

3.98km 2.68km 8.58km
Legend:
N30 N31 0.1MVA
Fault position
0.1MVA 2.76km 2.7km
x Fault number
Correct path
N1 N2
44km
N3 N4 1MVA N23
32km 5.6km
6 0.1MVA
N5 N6 1MVA N21 N22
25MVA
36km 3.54km 5.28km
N7 N8 1MVA N19 N20 0.1MVA N18

28km 3.89km 2.45km 3.98km

0.1MVA
3.88km 6.0km 5.48km
N26 0.1MVA N27 N28 0.1MVA

3.98km 2.68km 8.58km
Legend:
N30 N31 0.1MVA
Fault position
0.1MVA 2.76km 2.7km
x Fault number
Correct path
146
N1 N2
44km
N3 N4 1MVA N23
32km 5.6km
N5 N6 1MVA N21 0.1MVA

25MVA N22
36km 3.54km 5.28km
N7 N8 1MVA N19 N20 0.1MVA N18

28km 3.89km 2.45km 3.98km

0.1MVA
3.88km 6.0km 5.48km
N26 0.1MVA N27

8
N28 0.1MVA
3.98km 2.68km 8.58km
Legend:
N30 N31 0.1MVA
Fault position 0.1MVA 2.76km 2.7km
x Fault number
Correct path
147
In this section, we examine the performance of the methods based on mul-

tiple regression analysis (MRA), neural network (NN), GM algorithm and
EM algorithm with the simulation network model 4. The description of
the network model is given in a following section. Single line to earth
faults were simulated in a 20 kV unearthed neutral radially operated
MV distribution network. The examined algorithms used the transient
voltage signal measured from the secondary side of MV/LV distribution
transformer to estimate the fault distance. The aim of the simulation is
to evaluate the proposed algorithms with alternative approaches as de-
scribed in Chapter 5, Section 5.4.
A schematic diagram and ATPDraw circuit of the simulated network is

illustrated in Figure 6.44 and Figure 6.45, respectively. The data of the
simulation network is presented in Appendix B Table B1. The network
consists of a total of 158.51 km overhead line (OH) network and 6 feeders.
The length of the faulty feeder is 22.51 km. The feeder OH lines are repre-
sented with the frequency dependent JMarti line model and its ATPDraw
line configuration is given in Appendix B, Figure B2.
1 2
f1
DT25 Legend:
3 4 F1 a −Length a (km)
f2
A −Load A (kVA)
DT26 DTX −Distribution
5 6 F2 23 transformer(DT) X
f3
l
25MVA DT27 DT12
110kV
Supply 7 8 F3 L
f4
17 18 19 20 21 22 24 25 26 27 28 29
DT28 g h i j k m n o p q r
f
9 10 F4 DT6 DT7 DT8 DT9 DT10
DT11
DT13 DT14 DT15 DT16 DT17 DT18
f5
F G H I J K M N O P Q R
DT29
F5
11 12 13 14 15 16 s 30 31 32 33 34 35
a b c d e t u v w x
DT1 DT2 DT3 DT4 DT5 DT19 DT20 DT21 DT22 DT23 DT24
A B C D E S T U V W X
Figure 6.44. Schematic diagram of 20 kV isolated neutral MV network.
148
Figure 6.45. ATPdraw circuit for the simulated network model 4.
149
In this work, the transient signal is measured from the secondary side
of MV/LV substation number 24 (DT24), as shown in Figure 6.44. In order
to examine the performance of the algorithms, a large number of data sets
is required. The variation of the data sets used for formula construction
and testing in this work is given in Table 6.3 and Table 6.4, respectively .
In Table 6.3, the network model was simulated with 5 different load con-
ditions. For each load condition given in Table 6.3, the fault distance, fault
resistance and fault inception angle were varied. Similarly in Table 6.4,
the network model was simulated with load conditions as in Table 6.3 but
with different fault distance.
Table 6.3. Data generation for MRA and NN formulation. Fault resistances are 0, 25 and
50 Ω. Inception angles are 90, 60 and 30 ◦ for load 1 and load 2. 90 and 45 ◦
for load 3 , 4 and 5.
Load number 1 2 3 4 5
Number of variations
Fault distance 13 13 2 2 2
Fault resistance 3 3 3 3 3
Inception angle 3 3 2 2 2
Total 117 117 12 12 12
Data sets 270
Table 6.4. Data generation for testing the MRA and NN algorithm. Fault resistances are
0, 25 and 50 Ω. Inception angles are 90 and 45 ◦ .
Load number 1 2 3 4 5
Number of variations
Fault distance 2 2 1 1 1
Fault resistance 2 2 2 2 2
Inception angle 3 3 3 3 3
Total 12 12 6 6 6
Data sets 42
Using the data in Table 6.3, the formula obtained for MRA algorithm,
which corresponds to Equation (5.33), is given as:
l = −15.4735 + 0.0052A + 1.0511S + 0.0621L (6.5)
where dependent and independent factors are: l is fault distance, A is

FFT amplitude of extracted CWT coefficient, S is scale of extracted CWT
coefficient and L is total load (MVA) of the network prior to the single
line to earth fault. In this work, the random error e in MRA model is ne-
glected. In case of ANN algorithm, the formula based on Equation (5.34)
constructed with 3 observed factors in this simulation can be expresses
150
as:
l = AN N (A, S, L) (6.6)
where, Equation (6.6) is a non-linear relation between l and A, S and L

which cannot be expressed explicitly, therefore it has been designed as a
’knowledge base’. A detailed architecture of proposed ANN used in this
section is explained in Chapter 5, Section 5.4.3. In this simulation study,
the performance accuracy of MRA, NN, GM and EM algorithm has been
investigated with different fault conditions as given in Table 6.4. For MRA
and NN algorithms, the performance accuracy is defined by mean abso-
lute error (MAEMRA ) and (MAENN ) as expressed as in Equation (6.7) and
Equation (6.8) below , while for GM and EM algorithms, the performance
accuracy is defined as given in Equation (6.1) and Equation (6.3), respec-
tively.
n n
1 1
M AEM RA = |M RAi − li | = |∆M RAi | (6.7)
n n
i=1 i=1
n n
1 1
M AEN N = |N Ni − li | = |∆N Ni | (6.8)
n n
i=1 i=1
where ∆M RAi and ∆N Ni is the absolute error fault distance using MRA
and NN algorithm, M RAi and N Ni is estimated fault distance using MRA
and NN algorithm, li is exact fault distance and n is numbers of observa-
tion of parameter error.
An example of the comparison of FFT amplitude of extracted CWT coef-

ficient measured from 4 distribution transformers (DTs) is presented in
Figures 6.46, 6.47 and 6.48. The results show a relationship between esti-
mated FFT amplitude of extracted CWT coefficient and the fault distance,
plotted as a linear function with a positive gradient. Clearly, the FFT am-
plitude of the extracted CWT coefficient is considerably increasing with
an increase of fault distance. In Figures 6.46, 6.47 and 6.48, we can no-
tice a separation of the straight line between DTs, but the separation is
reduced when the fault resistance is increased and fault inception angle
is decreased. In the graph, the linear functions of DT12, DT17 and DT24
have almost similar gradient, and the differences become small when high
fault resistance and low inception angle were used.
151
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(a) Rf =0 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT12 DT17 DT24 DT5
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance[km]
(b) Rf =25 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18
Fault Distance[km]
(c) Rf =50 Ω
Figure 6.46. A comparison of estimated FFT amplitudes of extracted CWT coefficient at 4

different MV/LV distribution transformers (DT). A single line to earth fault
is simulated with varied fault distance and fault resistance. An inception
angle of 90◦ is used with Load 1 condition. DT5, DT12, DT17 and DT24
represent the distribution transformer number 5, 12, 17 and 24.
152
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20
DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(a) Rf =0 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(b) Rf =25 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT5 DT12 DT17 DT24

0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance [km]
(c) Rf =50 Ω

153
200
180
Amplitude [FFT] 160

140
120
100
80
60
40
20 DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance[km]
(a) Rf =0 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20
DT5 DT12 DT17 DT24
0
0 2 4 6 8 10 12 14 16 18 20
Fault Distance[km]
(b) Rf =25 Ω
200
180
160
Amplitude [FFT]
140
120
100
80
60
40
20 DT5 DT12 DT17 DT24

0
0 2 4 6 8 10 12 14 16 18 20
Fault Location[km]
(c) Rf =50 Ω

154
Figures 6.49 and 6.50 show the results of fault location performance by
using MRA and NN algorithms at fault inception angles of 90 and 45 de-
grees. The tabulated results can be found in Appendix C, Table C16 and
C17, respectively. In comparison of calculated MAE of both algorithms, it
shows that NN algorithm performed slightly better than MRA algorithm.
Both algorithms have fault location error less than 1 km. It can be seen
that, fault resistance and fault inception angle have a small effect on the
result of the algorithms. It seems that the result of fault distance estima-
tion from simulated earth fault at 45 degrees of fault inception angle is
slightly lower than in the case of 90 degrees of fault inception angle.
Figures 6.51 and 6.52 show the results of fault location performance
by using GM(GM1, GM2, GM4) algorithms at fault inception angles of
90 and 45 degrees. The tabulated results can be found in Appendix C
Table C18 and C19, respectively. The results show that GM2 algorithm
was a bit better than GM1 and GM4 algorithms for fault inception angle
of 90 degree, with calculated MAE 0.88 km. In case of fault inception
angle of 45 degree, GM1 performed slightly better than the other two GM
algorithms. It is found that the results of MAE for all GM algorithms
worked better at fault inception angle of 45 degrees. In case of simulated
fault with inception angle of 90 degrees, the distance estimation error is
highest at fault distances of 7.86 km and 13.56 km. However, in the case
of fault simulated with inception angle of 45 degrees, the error of fault
distance 15.59 km is the highest.
Comparison of distance estimation errors calculated using EM (EM1,
EM2, EM3) algorithms with the effect of fault inception angle and fault
resistance are shown in Figures 6.53 and 6.54. Calculated MAE of EM
algorithms can be found in Appendix C, Table C20 and C21. From the
results, with the effect of fault inception angle of 90 degrees, it shows
that the MAE of EM3 algorithm is the lowest compared with the other
EM algorithms. But EM2 algorithm shows a slightly better performance
when fault inception angle of 45 degrees is used. EM algorithms distance
estimated error is less than 2 km when simulated earth fault is not af-
fected by the fault resistance. In overall, EM algorithms fault distance
estimation error was less than 4 km.
155
1.5
MRA
NN
1
0.5
Error(km)
−0.5
−1
−1.5
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
1.5
MRA
NN
1
0.5
Error(km)
−0.5
−1
−1.5
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
1.5
MRA
NN
1
0.5
Error(km)
−0.5
−1
−1.5
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.49. Comparison of the distance estimate errors obtained from multiple regres-
sion (MRA) and neural network (NN) algorithm with the effect of fault resis-
tance (Rf ). Fault inception angle is 90◦ . MRA denotes a distance calculation
error using multiple regression and NN denotes a distance calculation error
using neural network algorithm.
156
1
MRA
NN
0.5
Error(km)
0
−0.5
−1
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
1
MRA
NN
0.5
Error(km)
−0.5
−1
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
1
MRA
NN
0.5
Error(km)
−0.5
−1
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.50. Comparison of the distance estimate errors obtained from multiple regres-
sion (MRA) and neural network (NN) algorithm with the effect of fault resis-
tance (Rf ). Fault inception angle is 45◦ . MRA denotes a distance calculation
error using multiple regression and NN denotes a distance calculation error
using neural network algorithm.
157
4
GM1
GM2
3 GM4
Error(km) 2
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
4
GM1
GM2
3 GM4
2
Error(km)
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
4
GM1
GM2
3 GM4
2
Error(km)
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.51. Comparison of the distance estimate errors obtained from general model
(GM1,GM2,GM4) algorithms with the effect of fault resistance (Rf ). Signal
measured from MV/LV substation and fault inception angle is 90◦ . GM1
denotes a distance calculation error using general model 1, GM2 denotes
a distance calculation error using general model 2 and GM4 denotes a dis-
tance calculation error using general model 4.
158
2
GM1
1.5 GM2
GM4
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
2
GM1
1.5 GM2
GM4
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
2
GM1
1.5 GM2
GM4
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.52. Comparison of the distance estimate errors obtained from general model
(GM1,GM2,GM4) algorithms with the effect of fault resistance (Rf ). Signal
measured from MV/LV substation and fault inception angle is 45◦ . GM1
denotes a distance calculation error using general model 1, GM2 denotes
a distance calculation error using general model 2 and GM4 denotes a dis-
tance calculation error using general model 4.
159
4
EM1
EM2
EM3
3
Error(km) 2
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
4
EM1
EM2
EM3
3
2
Error(km)
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
4
EM1
EM2
EM3
3
2
Error(km)
−1
−2
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.53. Comparison of the distance estimate errors obtained from exact model
(EM1-EM3) algorithms with the effect of fault resistance (Rf ). Signal mea-
sured from MV/LV substation and fault inception angle is 90◦ . EM1 denotes
a distance calculation error using exact model 1, EM2 denotes a distance
calculation error using exact model 2 and EM3 denotes a distance calcula-
tion error using exact model 3.
160
2
EM1
EM2
1.5 EM3
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(a) Rf =0 Ω
2
EM1
1.5 EM2
EM3
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(b) Rf =25 Ω
2
EM1
1.5 EM2
EM3
1
0.5
Error(km)
0
−0.5
−1
−1.5
−2
−2.5
−3
1.84 2.34 4.45 7.86 8.07 13.56 15.59
Distance(km)
(c) Rf =50 Ω
Figure 6.54. Comparison of the distance estimate errors obtained from exact model
(EM1-EM3) algorithms with the effect of fault resistance (Rf ). Signal mea-
sured from MV/LV substation and fault inception angle is 45◦ . EM1 denotes
a distance calculation error using exact model 1, EM2 denotes a distance
calculation error using exact model 2 and EM3 denotes a distance calcula-
tion error using exact model 3.
161
In overall, the proposed algorithms are able to perform well with the
use of transient signal measured from the secondary side of the MV/LV
distribution transformer.
Considering the fault distance estimation results in this simulation,
NN algorithm shows a better performance compared with the other al-
gorithms. The lowest MAE calculated was 0.34 km and the highest was
1.18 km. The proposed fault location algorithms proved much more accu-
rate in this simulation model compared with previous simulation model
(model 3). We noticed that the number of branches in this simulation net-
work model is less than in the previous simulation network model. Also,
the length of the branches in simulation network model 4 is short com-
pared with the simulation network model 3. In this section, we did not
investigate the performance of WT algorithm because the transient sig-
nals that we used were measured from the secondary side of the MV/LV
distribution substation transformer.
It is worth to mentioned here that, based on our experience when pro-
cessing the measured transient signals, we found that it is more difficult
to analyze those measured transient signals when the fault distance is
far from the measuring MV/LV substation (DT24). In addition, with high
resistance values and low fault inception angles, the task is more critical.
This is because the damping will be increased and the transient signal is
damped much faster. In the following chapter, the proposed algorithms
are investigated using transient signals measured in real networks.
162
7. Performance Evaluation of Proposed
Fault Location Algorithms Using
Transients Recorded in Real
Networks
Three real case studies are presented in this chapter. These consist of iso-
lated, compensated and partially compensated neutral MV distribution
networks. Two of the cases were conducted in Finnish distribution net-
works and the third is performed in a Czech distribution network. The
recorded data is used to test selected fault location algorithms that we
have proposed in Chapter 5. Due to the limitation of information of line
parameters in real networks, only GM2 and CWT algorithm were tested
using real recorded data.
7.1 Case study 1
In this case, an artificial single line to earth fault was carried out in a
21 kV Finnish overhead line network. The case consists of unearthed and
compensated neutral MV distribution networks in Vaasan Sähkö Oy. The
data was recorded with sampling frequency of 20 kHz. In this case study,
CWT and GM2 algorithms were used to estimate the fault distance. The
procedures used to estimate the fault distance based on the measured
transient signals are explained in Chapter 5 of this dissertation. The per-
formance accuracy of the selected algorithms was measured using MAE
as given in Equation (6.1) and Equation (6.4).
7.1.1 Description of Real Network Case 1
The parameters of the networks used in this analysis are given in Ta-
ble 7.1. The network diagrams of the field tests are shown in Figures 7.1,
7.2 and 7.3.
163
Performance Evaluation of Proposed Fault Location Algorithms Using Transients Recorded in Real Networks
Figure 7.1. Compensated neutral network test diagram for test 1-2, 0.76 km [40].
164
Figure 7.2. Compensated neutral network test diagram for test 3-4, 10.40 km [40].
165
Figure 7.3. Isolated neutral network test diagram for test 5-8, 14.20 km [40].
166
Table 7.1. Case 1 network parameters [40].
Test number 1-2 3-4 5-8

Network Compensated neutral Compensated neutral Isolated neutral
Voltage 110/21 kV 110/21 kV 110/21 kV
Transformer 25 MVA 10% 25 MVA 10% 16 MVA 10%
Arc supression 90 A 90 A
Substation load 12 MW 12 MW 2.7 MW
Llp,n 1 mH/km 1 mH/km 1 mH/km
Ll0 5 mH/km 5 mH/km 5 mH/km
Cp 10.1 μF 8.5 μF 2.1 μF
C0 9.7 μF 8.1 μF 1.2 μF
Earth resistance 0.32 Ω 0.32 Ω 0.20 Ω
Soil resistivity 1000 Ωm 1000 Ωm 1000 Ωm
The recorded fault voltages and currents are shown in Figures 7.4, 7.5,
7.6 and 7.7. As an example of pre processing used in this work, one single
phase to ground fault from test number 1 is presented in Figure 7.8 and
the output from fault distances algorithm based on CWT is depicted in
Figure 7.9. The results of estimated fault distance using CWT algorithm
for test number 1, 3, 5 and 7 are depicted in Figure 7.10.
The summary of the results are given in Table 7.2. In order to compare
the performance accuracy of both algorithms, we used similar information
and parameters determined from the measured earth fault transient sig-
nals for both algorithms. Based on the results, CWT algorithm produced
good results compared to GM2 algorithm. GM2 algorithm performed well
in compensated neutral network compared to its result in isolated neutral
network. The MAE calculated was 0.52 km. In case of GM2 algorithm,
the MAE calculated was 3.80 km. GM2 algorithm gave good results when
the fault distance is short and close to the measuring point and when the
effect of fault resistance was low. Unfortunately, when fault distance is
far from the measuring point, the fault distance estimation error is high.
The result is of no use if the fault distance and fault resistance is to high.
Table 7.2. Calculation results using CWT and GM2 algorithm.
Test
l Rf CWT ∆CWT GM2 ∆GM2
number
1 0.76 0 0.38 -0.38 0.10 -0.66
2 0.76 0 0.17 -0.59 0.10 -0.66
3 10.40 0 9.92 -0.48 10.00 -0.40
4 10.40 0 9.81 -0.59 9.90 -0.50
5 14.20 0 13.57 -0.63 16.70 2.50
6 14.20 0 14.77 0.57 21.20 7.00
7 14.20 50 14.12 -0.08 20.63 6.43
8 14.20 50 13.39 -0.81 26.46 12.26
MAE 0.52 3.80
167
4
x 10
2
U
1.5
Voltage [V] 1
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(a) Test number 1 phase voltage
1500
I
1000
Current [A]
500
−500
−1000
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(b) Test number 1 phase current

4
x 10
2
U
1.5
1
Voltage [V]
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(c) Test number 2 phase voltage
1500
I
1000
Current [A]
500
−500
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(d) Test number 2 phase current
Figure 7.4. Recorded signals of single line to earth fault at 0.46 km.
168
4
x 10
2
U
1.5
1
Voltage [V]
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
300
I
200
Current [A]
100
−100
−200
−300
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(b) test number 3 phase current

4
x 10
2
U
1.5
1
Voltage [V]
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
300
I
200
Current [A]
100
−100
−200
−300
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
Figure 7.5. Recorded signals of single line to earth at 10.40 km.
169
4
x 10
2
V
1.5
Voltage [V] 1
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
100
I
50
Current [A]
−50
−100
−150
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]

4
x 10
2
V
1.5
1
Voltage [V]
0.5
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
150
I
100
Current [A]
50
−50
−100
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
Figure 7.6. Recorded signals of single line to earth fault at 14.20 km. Test number 5 and
6.
170
4
x 10
2
V
1.5
1
Voltage [V]
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
80
I
60
40
Current [A]
20
0
−20
−40
−60
−80
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]

4
x 10
2
V
1.5
1
Voltage [V]
0.5
0
−0.5
−1
−1.5
−2
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
80
I
60
40
Current [A]
20
−20
−40
−60
−80
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
Figure 7.7. Recorded signals of single line to earth fault at 14.20 km. Test number 7 and
8.
171
80 80
60
60 70
50
40 60
Scales[a]
20 40 50
X= 20
80 Y= 33 X: 33 40
Level= 0.11133 30 Y: 3.954
60 30
20
40 20
10
20 10
0 50 100 150 200 5 4 3 2 1 00

Sample Number[k] Energy percentage[%]
(a) Voltage CWT scalogram (left) and Scale-dependent energy percentage

distribution(right)
80 0.1
60
0.09
40
60 0.08
20
0.07
1
Scales[a]
0.06
40
80
0.05
X= 34 X: 34
Y= 34 Y: 3.918 60
0.04
Level= 0.10308
20
0.03 40
0.02 20
1 0.016
0 50 100 150 200 5 4 3 2 1 01
Sample Number[k] Energy percentage[%]
(b) Current CWT scalogram (left) and Scale-dependent energy percent-

age distribution(right)
4
x 10
2
V
CWT
Voltage[V]
−1
−2
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
4000
I
CWT
2000
Current[A]
−2000
−4000
−6000
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
(c) Extracted CWT coefficients

4000
V
I
3500 X: 468.8
Y: 3672
3000
Magnitude [FFT]
2500
2000
1500
X: 468.8
Y: 929.6
1000
500
0
0 500 1000 1500 2000 2500 3000 3500 4000
Frequency [Hz]
(d) Voltage and Current FFT frequency spectrum of extracted CWT
Figure 7.8. Example of output from pre-processing of test number 1 (0.76 km) transient
signal using CWT analysis.
172
4
x 10
2
U [V]
0
v
−2
0 20 40 60 80 100 120 140 160 180 200
5000
U [A]
0
I
−5000
0 20 40 60 80 100 120 140 160 180 200
−3
x 10
4
L(H)
2
0
0 20 40 60 80 100 120 140 160 180 200
0.4
X: 86
xstd
0.2 Y: 0.06533
0
0 20 40 60 80 100 120 140 160 180 200
2
lf(km)
X: 86
1 Y: 0.3747
0
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
Figure 7.9. Example of output from fault distances algorithm based on CWT for test
number 1.
100
80
l [km]
60
f
X: 51
40 Y: 30.34
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
100
80
lf[km]
60
40 X: 14
Y: 25.78
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
(a) Test number 1 (above) and test number 3 (below)
60
40
X: 13
Y: 13.57
lf[km]
20
−20
−40
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
100
50
l [km]
0 X: 11
f
Y: 14.12
−50
−100
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
(b) Test number 5 (above) and test number 7 (below)
Figure 7.10. Results of fault distance estimation (lf ) using CWT algorithm.
173
7.2 Case study 2
In this section an artificial single line to earth fault was carried out in
a 22 kV Finnish overhead line network. The field test was recorded in
Lounais-Suomen Sähkö Oy. The network is partially compensated neu-
tral network. The data was recorded with sampling frequency of 20 kHz.
In this case study, CWT and GM2 algorithms were used to estimate the
fault distance. The procedure used to estimate the fault distance was
based on the measured transient signal as is explained in Chapter 5 of
this dissertation. The performance accuracy of the selected algorithms
were measured using MAE as given in Equation (6.1) and Equation (6.4).
The parameters of the network are given in Table 7.3. The network dia-
grams of the field test are shown in Figure 7.11.
7.14 and 7.15. Figure 7.16 and 7.17 shows an example of output from
pre-processing of test number 4 transient signal using CWT analysis and
Hilbert transformation. The results of estimated fault distance using
CWT algorithm for test number 1, 3, 6 and 8 is depicted in Figure 7.18.
The summary of the results are given in Table 7.4. By comparison of MAE
for both algorithms, CWT algorithm performed slightly better compared
to GM2 algorithm. There are 2 results showing unacceptable accuracy
performed by GM2 algorithm where MAE increased significantly. The
lowest fault estimation error was 0.1 km and the highest was 12.40 km.
In this case GM2 algorithm gave good results when the fault distance
is far from the measuring point. In comparison with the result of GM2
algorithm in previous case study for fault distance far from the measur-
ing point, it seems that the algorithm has performed well in compensated
neutral network compared to its result in isolated neutral network.
174
Table 7.3. Case 2 network parameters [40].
Test number 1-5 6-8

Network Partially compensated
Voltage 110/22 kV
Transformer 25 MVA 10.6%
Arc supression 8.5 A
Compensation capacitor 1800 kVAr
Substation load 5.3 MW 6.9 MW
Llp,n 1.1 mH/km
L0 4.36 mH/km
Cp 18.7 μF 21.5 μF
C0 2.5 μF 3.6 μF
Earth resistance 0.6 Ω
Ground resistivity 200 Ωm
(a) Network diagram for test 1-5, 25.40 km
(b) Network diagram for test 6-8, 36.00 km

175
Figure 7.11. Lounais-Suomen Sahko OY network test diagram [40].
4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
150
I
100
Current [A]
50
−50
−100
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
150
I
100
Current [A]
50
−50
−100
−150
0 500 1000 1500 2000 2500 3000 3500 4000
Sample Number [k]
Figure 7.12. Recorded signals of single line to earth fault at 25.40 km. Test number 1
and 2.
176
4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 500 1000 1500 2000
Sample Number [k]
150
I
100
50
Current [A]
0
−50
−100
−150
−200
−250
0 500 1000 1500 2000
Sample Number [k]

4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 500 1000 1500 2000
Sample Number [k]
200
I
100
Current [A]
−100
−200
−300
0 500 1000 1500 2000
Sample Number [k]
Figure 7.13. Recorded signals of single line to earth fault at 25.40 km. Test number 3
and 4.
177
4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 500 1000 1500 2000
Sample Number [k]
300
I
250
200
Current [A]
150
100
50
0
−50
−100
−150
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
(b) test number 5 phase current

4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number [k]
400
I
300
Current [A]
200
100
−100
−200
0 500 1000 1500 2000
Sample Number [k]
Figure 7.14. Recorded signals of single line to earth fault at 25.40 km (test number 5)
and 36.0 km (test number 6).
178
4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 500 1000 1500 2000 2500 3000 3500 4000
Sample Number [k]
300
I
200
Current [A]
100
−100
−200
−300
−400
0 500 1000 1500 2000
Sample Number [k]

4
x 10
4
U
3
2
Voltage [V]
1
0
−1
−2
−3
−4
0 500 1000 1500 2000 2500 3000
Sample Number [k]
400
I
300
Current [A]
200
100
−100
−200
0 500 1000 1500 2000 2500 3000
Sample Number [k]
Figure 7.15. Recorded signals of single line to earth fault at 36.0 km.
179
4
x 10
2
Voltage[V}
V
0
−2
−4
20 40 60 80 100 120 140 160 180 200
115 0.03
109
103
97 0.025
91 X= 51
85 Y= 72
79 Level= 0.029502
Scales a
73 0.02
67
61
55
49 0.015
43
37
31
25 0.01
19
13
7
1 0.005
20 40 60 80 100 120 140 160 180 200
Sample Number[k]
(a) Voltage CWT scalogram(below) and measured signal(above)
2.5 X: 72
Y: 2.308
1.5
0.5
0
0 20 40 60 80 100 120
Scales[a]
(b) Voltage scale-dependent energy percentage distribution
200
I
Current[A]
100
0
−100
−200
−300
20 40 60 80 100 120 140 160 180 200
0.03
115
109
103
97 0.025
91
85
79
Scales[a]
73 0.02
67 X= 72
61 Y= 76
55 Level= 0.029556
49 0.015
43
37
31
25 0.01
19
13
7
1 0.005
20 40 60 80 100 120 140 160 180 200
Sample Number[k]
(c) Current CWT scalogram(below)and measured signal(above)

2.5
X: 76
Y: 2.1
2
1.5
0.5
0
0 20 40 60 80 100 120
Scales[a]
(d) Current scale-dependent energy percentage distribution
Figure 7.16. Example of CWT scalogram and energy percentage distribution of test num-
ber 4 (25.4 km) transient signal.
180
5
x 10
1
V
CWT
0.5
Voltage[V]
0
−0.5
−1
0 20 40 60 80 100 120 140 160 180 200
Sample Number[k]
1500
I
CWT
1000
Current[A]
500
0
−500
−1000
−1500
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
(a) Extracted CWT coefficients

4
x 10
4.5
V
X: 234.4 I
4 Y: 4.3e+004
3.5
3
Magnitude[FFT]
2.5
1.5
0.5 X: 234.4
Y: 451.9
0
0 1000 2000 3000 4000 5000 6000
Frequency[Hz]
(b) FFT frequency spectrum of extracted CWT

11.4
V
WT
11.3
11.2
11.1
log(|V (t)|)
11
WT
10.9
10.8
10.7
10.6
10.5
10.4
100 110 120 130 140 150 160 170 180 190 200
Sample Number[k]
(c) Voltage linear decaying of extracted CWT coefficient
6.8
I
WT
6.7
6.6
6.5
log(|IWT(t)|
6.4
6.3
6.2
6.1
5.9
100 110 120 130 140 150 160 170 180 190 200
Sample Number[k]
(d) Current linear decaying of extracted CWT coefficient
Figure 7.17. Example of CWT extracted coefficients, FFT frequency spectrum of ec-
tracted coefficent and, voltage and current linear decaying of extracted CWT
coefficient of test number 4 (25.4 km) transient signal.
181
100
80
lf[km]
60
X: 51
40 Y: 30.34
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
100
80
lf[km]
60
40 X: 14
Y: 25.78
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
100
80
lf[km]
60
X: 12
Y: 39.53
40
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
100
80
lf[km]
60 X: 12
Y: 38.46
40
20
0 20 40 60 80 100 120 140 160 180 200
Sample Number [k]
Figure 7.18. Fault distances calculated using CWT algorithm.
Table 7.4. Calculation results using CWT and GM2 algorithm.
Test
l Rf CWT ∆CWT GM2 ∆GM2
number
1 25.4 0 30.34 4.94 25.30 -0.10
2 25.4 0 29.82 4.42 25.30 -0.10
3 25.4 0 25.78 0.38 36.70 11.30
4 25.4 0 21.02 -4.38 25.30 -0.10
5 25.4 0 24.19 -1.21 37.80 12.40
6 36.0 0 39.53 3.53 36.70 0.70
7 36.0 0 38.70 2.70 36.10 0.10
8 36.0 0 38.46 2.46 35.50 -0.50
MAE 3.00 3.16
182
7.3 Case study 3
In this section an artificial single line to earth fault was carried out in a
22 kV Czech compensated neutral network. The field test was recorded
by David Topolanek and his team from Brno University. The data was
used during his research visit in autumn 2012 at Aalto University. The
measured signal was recorded from MV substation and secondary side of
MV/LV distribution transformer. Due to the limitation of information, the
data was tested only with CWT algorithm described in Section 5.3.
The experiment was conducted in a compensated neutral distribution net-

work. Initially, there were 29 units of LV distribution transformers scat-
tered along the faulty feeder. However only 4 of them were selected for
the experiment. The network consisted of a mix of underground cables
and overhead lines. The average line parameters are as follows: positive
sequence inductance is 1.1459 mH per km and zero sequence inductance
is 4.4551 mH per km. In this case, we assumed the negative sequence
parameters to be equal to the positive ones. During the experiment, the
digital recorder units were installed in four selected 22/0.4 kV LV dis-
tribution transformers and at the 110/22 kV MV substation as shown in
Figure 7.19.
OTHER FEEDERS
110/22kV .
.
.
.
.
.
.
.
.
.
Rp Lc
Fault distance = 13.9km
880m 9398m 3513m 3258m
912m 695m
22/0.4kV 22/0.4kV 8.5m 22/0.4kV 15m 22/0.4kV
Legend:
Measuring point
Fault position
Domanin Olesna Nove Mesto Pohledec
Figure 7.19. Schematic diagram of Czech distribution network.
183
(a)
(b)
Figure 7.20. (a) Photo of one of the low voltage (LV) switch-boards and the digital oscil-
loscope used during the field test. (b) Connection of measuring probe at the
back of the LV distribution panel
The recorder sampling rate at MV substation, and at the LV substations

“Domanin”, “Olesna”, “Nove Mesto” and “Pohledec” were 50 kHz, 50 kHz,
20 kHz, 50 kHz and 28 kHz respectively. For the purpose of normaliza-
tion in calculation, all the recorded data were resampled to 32 kHz. The
sampling frequency used in this case study is slightly higher than the
previous case sampling frequency. Unfortunately, in this work, we did
not study the effect of sampling frequency on the performance of the algo-
rithm. Perhaps, in order to see its influence to the algorithms correctness,
it shall be investigate in future.
In order to verify the algorithm used in this paper, an earth fault was
simulated about 13.90 km from measuring point located at MV substa-
tion as shown in Figure 7.19. Three of the LV distribution substations
were located in front of the fault location, while the fourth LV distribu-
tion substation was located approximately 3 km behind the fault. During
the experiment a basic metallic earth fault was conducted so that affected
184
phase was directly connected to the grounding system of the substation.

The experiment was also performed for the three basic levels of com-
pensation of the distribution system. The levels of compensation were
full compensation state, under compensation by about 15% and over com-
pensation by about 15%. The parameters and the states of individual
experiments conducted are given in Table 7.5.
Table 7.5. Description of actual single line to earth fault experiment.
Experiment Type of State of IL IC

number earth fault compensation (A) (A)
1 Metallic Ideal 267 267
4 Metallic Undercompensated 220 267
5 Metallic Overcompensated 305 267
*Note: IL = current at Peterson coil, IC =capacitive current of the network
7.23, 7.24 and 7.25. The fault location accuracy is measured with mean
average error (MAE) in km as in Equation (6.4). Table 7.6 shows the
results of the proposed scheme tested on the real recorded data. In overall,
the MAE for fault distance is 0.65 km. The fault distance for test number
1 , 3, 4 and 5 calculated using CWT algorithm is depicted in Figure 7.26.
The summary of the results are given in Table 7.6.
In Table 7.6, the results show that by using the proposed algorithm we
managed to identify the correct LV substation to be used as indicator for
selecting a correct path towards the fault position. Based on the results,
the detected LV substation is located in front of the fault position. The dis-
tance between the detected LV substation and the fault location is about
118.5 meters, which is very close. The results from test number 1 to test
number 4 show that substation “Pohledec” record the second highest peak
amplitude. The substation is located 3.148 km behind the fault position.
In this experiment result, it seems that the first and second highest peak
amplitude of the LV substations gives us an indication that the fault is
located in the section between these two substations.
185
4
x 10
4
Ua 400 Ia
3
Ub Ib
2 200 Ic
Current [A]
Voltage [V]
Uc
1 0
0
−200
−1
−2 −400
−3
−600
−4
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4 Sample Number [k] x 10
4
(a) MV phase voltages (b) MV phase currents

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Ub Uc
Voltage [V]
Voltage [V]
200 200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
4
(c) LV DOM phase voltages (d) LV NOVE phase voltages

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Uc Uc
Voltage [V]
Voltage [V]
200 200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4 4
Sample Number [k] x 10 Sample Number [k] x 10
(e) LV OLE phase voltages (f) LV POH phase voltages
Figure 7.21. Recorded signals of experiments 1.
4
x 10
5 400 Ia
4 Ua Ib
3 Ub 200 Ic
Current [A]
Uc
Voltage [V]
2
0
1
0 −200
−1
−400
−2
−3 −600
−4 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Uc Uc
Voltage [V]
Voltage [V]
200 200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
4

500
400 Ua 500
300 Ub 400 Ua
Uc Ub
200 300
Voltage [V]
Uc
Voltage [V]
100 200
0 100
0
−100
−100
−200
−200
−300
−300
−400
−400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
Sample Number [k] x 10 Sample Number [k] 4
x 10
186
4 400
x 10
5
Ia
4 Ua Ib
200
Ub Ic
3 Uc
Current [A]
Voltage [V]
2 0
1
−200
0
−1 −400
−2
−3 −600
−4 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 4

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Uc Uc
200 200
Voltage [V]
Voltage [V]
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4

500 500
400 Ua 400 Ua
Ub Ub
300 300
Uc Uc
200
Voltage [V]
200
Voltage [V]
100 100
0 0
−100 −100
−200 −200
−300
−300
−400
−400
−500
−500 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 4
Sample Number [k] 4 Sample Number [k] x 10
x 10
4
x 10 600 Ia
4 Ua Ib
Ub 400 Ic
Current [A]
Voltage [V]
2 Uc
200
0
0
−2
−200
−4
−400
−6 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 x 10
4
Sample Number [k] 4 Sample Number [k]
x 10

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Voltage [V]
Uc Uc
Voltage [V]
200 200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4 4

500 500
400 Ua
400 Ua
Ub
300 Ub 300
Uc
Uc
Voltage [V]
Voltage [V]
200 200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
4
187
4
x 10 600
4 Ia
3 Ua Ib
Ub 400 Ic
Current [A]
2
Voltage [V]
Uc
200
1
0 0
−1
−200
−2
−3 −400
−4 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 4
Sample Number [k] 4 Sample Number [k] x 10
x 10

500 500
400 Ua
400 Ua
300 Ub
300 Ub
Uc
Voltage [V]
200 Uc
Voltage [V]
200
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
Sample Number [k] x 10 4

500 500
400 Ua 400 Ua
300 Ub 300 Ub
Uc Uc
Voltage [V]
200 200
Voltage [V]
100 100
0 0
−100 −100
−200 −200
−300 −300
−400 −400
−500 −500
1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2 1.5 1.55 1.6 1.65 1.7 1.75 1.8 1.85 1.9 1.95 2
4
4
60
lf[km]
40
X: 172
20 Y: 14.04
0
0 50 100 150 200 250
Sample Number[k]
40
lf[km]
X: 166
20 Y: 14.44
0
0 50 100 150 200 250
Sample Number[k]
40
lf[km]
X: 232
20 Y: 15.6
0
0 50 100 150 200 250
Sample Number[k]
50
40
l [km]
30 X: 164
20 Y: 14.52
f
10
0
0 50 100 150 200 250
Sample Number[k]
Figure 7.26. Fault distances calculated using CWT algorithm.
188
Table 7.6. Test result of single line to earth fault using recorded signal.
Test
1 2 3 4 5
number
State of Under Over MAE

Ideal Ideal Ideal
Compensation compensated compensated (km)
Fault
14.01 14.15 14.44 15.6 14.52
Location (km)
Error (km) -0.14 -0.25 -0.54 -1.7 -0.62 0.65
Secondary Amp. Amp. Amp. Amp. Amp.

substation |FFT| |FFT| |FFT| |FFT| |FFT|
Pohledec 1150.04 1187.78 1134.6 1106.2 1070.23

Nove
1163.74 1195.16 1135.3 1111.4 1067.95
Mesto
Olesna 1038.6 1007.38 939.44 908.58 895.97
Domanin 693.82 631.67 458.22 455.28 479.76
Detected Nove Nove Nove Nove

Pohledec
substation Mesto Mesto Mesto Mesto
189
190
8. Conclusions and Future Work
This thesis presented fault location algorithms for earth faults in un-
earthed and compensated neutral networks using measured transient sig-
nals. The initial transients of the earth faults are important especially in
unearthed and compensated neutral networks. In this work, five types
of fault location algorithms based on initial transients of the earth faults
have been developed. Two of the algorithms were developed using the
measured transient signals from secondary side of the MV/LV distribu-
tion transformers. Five types of fault location algorithms developed in
this thesis are known as general model (GM), exact model (EM), contin-
uous wavelet transform (CWT), multiple regression analysis (MRA) and
artificial neural network (ANN) algorithms.
Considering the results from the simulated network model 1, we can
conclude that the damped and undamped charge transient frequency and
damping factor from the simulation agreed with the result calculated
by theoretical formula of the general model algorithms, with acceptable
small error. Based on the fault distance result, it was shown that GM1
and GM2 algorithms have almost identical result. It shows that, in order
to estimate the fault distance, by using the undamped charge transient
frequency, GM2 algorithm managed to compensate the effect of fault re-
sistance as well as the unknown resistive parameters of the circuit.
We compared the performance of EM1 and GM2 algorithms using se-
quence component network in simulation model 2. In overall, we conclude
that EM1 algorithm has performed much better than GM2 algorithm. The
main reason why EM1 algorithm performs much better than GM2 is that
the EM1 circuit model has a similar structure as the simulation network,
except that we purposely neglected the resistive part in EM1 model. On
the other hand, GM2 circuit model is more robust and simple. Both al-
gorithms used same signal pre processing technique to estimate the un-
191
Conclusions and Future Work
damped charge transient frequency.

Based on the results in simulated network model 3, CWT algorithm
works well in estimating fault distance using both voltage and current
of the earth fault transient signals. The performance of CWT algorithm is
slightly effected when the magnitude of the fault resistance is increased.
In this simulation case, by using the measured transient signal from sec-
ondary side of the MV/LV distribution transformers, we able to identified
the correct patch of fault location using the proposed algorithm.
MRA and NN algorithms have shown a good performance in estimating
fault distance in simulated network model 4. The network is typical MV
distribution network used in [144] with some modification on its data. In
order for MRA and NN algorithm to perform well, we need to developed
a database for having a relationship between the fault location and the
information gathered from the measured transient signals. In this simu-
lation case, the length of each feeder was almost the same (32 km, 20 km,
25 km, 40 km, 19 km and 22.511 km (faulty feeder). Based on the results,
GM and EM algorithms have shown an acceptable performance with this
kind of MV network. In overall, for GM and EM algorithms, estimated
fault location error is less than 3.5 km. In conclusion, it was proved that
using the recorded transient signal from the secondary side of the MV/LV
distribution transformer it is possible to present a good result of fault dis-
tance estimation.
The feasibility of the proposed algorithms was investigated using the
data from simulations and field experiments with fault resistance setting
up to 50 Ω. Due to the limitation of real network, only CWT and GM2 algo-
rithms were evaluated. In overall, CWT has shown a better performance
compared with the GM2 algorithm. However, in terms of the accuracy the
difference is not so much. In some cases GM2 algorithm gave better result
than the CWT algorithm. Since GM2 can be implemented with only one
measurement it is a good option to install a transient recorder in some
small network with balanced length of its feeders, for instance in a rural
area.
It is worth to mention here that, since the proposed algorithms in this
thesis are based on earth fault transient signals, the limitations of the
proposed algorithms are related to the earth fault resistance, sampling
frequency, loads and also network configuration. The error is also caused
by the inaccuracy in the identification of charge transient component by
CWT and in addition by inaccuracy in the estimation of the damping at-
192
tenuation through Hilbert transformation method. Whenever the magni-

tude of the fault resistance and the resistive load is increased, the tran-
sients become more damped, with the result that the estimation of the
charge transient frequency becomes more difficult, which affects the over-
all computation of the fault distance.
There are several things that can be done in the future to evaluate and
to improve the proposed algorithms. In this thesis we have managed to
test only two algorithms using real system data. Therefore it is important
in the future to evaluate the other algorithms with the real network mea-
sured data as well. The GM and EM algorithms’ performance is affected
by the configuration of the network. Therefore, it is important to study the
correction factor required to reduce the error due to unbalanced lengths
of the feeders with many branches. In this work, the simulation used only
OH lines, in the future we should consider developing the simulation net-
work with underground cables as well. The studies should consider the
effect of capacitances of underground cables which normally have much
higher value those of the OH line. It seems that measuring the transient
frequency from the secondary side of the MV/LV distribution transformer
opens new possibilities for the fault location algorithms. Therefore, in-
stead of using only one transient signal, it is worth to study in the future
the overall behavior of the voltages and currents at secondary side of the
MV/LV distribution transformers during the earth faults in the MV net-
work in order to develops new fault location and detection algorithms.
193
194
References
[1] S. Hänninen and M. Lehtonen, “Characteristics of earth faults in electri-

cal distribution networks with high impedance earthing,” Electric Power
Systems Research, vol. 44, no. 3, pp. 155–161, Mar 1998.
[2] K. Winter, “Null point analysis - new method for detecting high resistance
earth faults,” Elektricitetens Rationella Användning (ERA), vol. 61, pp. 18–
20,23–24, 1988.
[3] Sähköenergialiitto ry SENER, Keskeytystilasto 2001 (Interruption Statis-

tics 2001). Helsinki: Adato Energia Oy, 2002.
[4] P. Heine, J. Pitkänen, and M. Lehtonen, “Voltage sag characteristics of

covered conductor feeders,” in 38th International Universities Power En-
gineering Conference (UPEC2003). Thessaloniki Greece, Sept 2003, pp.
1–3.
[5] G. Eberl and P. Schegner, “Earth fault location based on short time
low ohmic neutral earthing in cabled medium voltage networks of the
DREWAG - Stadtwerke Dresden GmbH,” in Seminar on Methods and
Techniques for Earth Fault Detection, Indication and Location. Espoo
Finland, Feb 2011, pp. 135–141.
[6] P. Järventausta, “Feeder fault management in medium voltage electricity

distribution networks,” Ph.D. dissertation, Tampere University of Technol-
ogy, Tampere, 1995.
[7] M. Lehtonen and T. Hakola, Neutral earthing and power system protection:
earthing solutions and protective relaying in medium voltage distribution
networks. ABB Transmit Oy, 1996.
[8] O. Chaari, M. Meunier, and F. Brouaye, “Wavelets: a new tool for the res-
onant grounded power distribution systems relaying,” IEEE Transactions
on Power Delivery, vol. 11, no. 3, pp. 1301–1308, July 1996.
[9] A. Nikander and P. Järventausta, “Safety aspects and novel technical solu-
tions for earth fault management in MV electricity distribution networks,”
in IET Conference Proceedings of 3rd IEE International Conference on Re-
liability of Transmission and Distribution Networks (RTDN 2005), vol. 4.
London UK, Jan 2005, pp. 207–211.
[10] J. Altonen and A. Wahlroos, “Novel algorithm for earth-fault location in

compensated MV-networks,” in 22nd International Conference and Exhi-
195
References
bition on Electricity Distribution (CIRED 2013). Stockholm Sweden, June

2013, pp. 1–4.
[11] B. Yong, C. Wei, L. Jiansheng, D. Lei, L. Qingdong, and Y. Ning, “Single

phase to earth fault location method in distribution network based on sig-
nal injection principle,” in 4th International Conference on Electric Utility
Deregulation and Restructuring and Power Technologies (DRPT). Shan-
dong China, July 2011, pp. 204–208.
[12] K. Jia, D. Thomas, and M. Sumner, “Impedance-based earth fault location

for a non-directly grounded distribution systems,” IET Generation, Trans-
mission Distribution, vol. 6, no. 12, pp. 1272–1280, Dec 2012.
[13] G. Achleitner, C. Obkircher, L. Fickert, M. Sakulin, and C. Raunig, “An

earth fault distance location algorithm in compensated networks with ad-
ditional estimation of the fault impedance and fault current,” in Power
Quality and Supply Reliability Conference (PQ 2008). Parnu Estonia,
Aug 2008, pp. 193–198.
[14] G. Eberl, S. Hänninen, M. Lehtonen, and P. Schegner, “Comparison of ar-

tificial neural networks and conventional algorithms in ground fault dis-
tance computation,” in IEEE Power Engineering Society Winter Meeting,
vol. 3. Singapore, Jan 2000, pp. 1991–1996.
[15] L. V. Bogdashova and V. E. Kachesov, “Parametric on-line fault location

methods for distribution MV networks,” in IEEE Power Tech Conference.
St. Petersburg Russia, June 2005, pp. 1–7.
[16] P. Imriš, “Transient based earth fault location in 110 kV subtransmission

networks,” Ph.D. dissertation, Helsinki University of Technology, 2006.
[17] A. Girgis, C. Fallon, and D. Lubkeman, “A fault location technique for rural
distribution feeders,” IEEE Transactions on Industry Applications, vol. 29,
no. 6, pp. 1170–1175, Nov 1993.
[18] R. Aggarwal, Y. Aslan, and A. Johns, “New concept in fault location

for overhead distribution systems using superimposed components,” IEE
Proceedings-Generation, Transmission and Distribution, vol. 144, no. 3,
pp. 309–316, May 1997.
[19] M. Lehtonen, S. Pettissalo, and J.-H. Etula, “Calculational fault location

for electrical distribution networks,” in Third International Conference on
Power System Monitoring and Control. London UK, June 1991, pp. 38–43.
[20] P. Imriš and M. Lehtonen, “Transient based ground fault location using
wavelets,” in 4th IASTED International Conference on Power and Energy
Systems. Rhodes, Greece, 28-30 June 2004, pp. 507–511.
[21] A. Borghetti, S. Corsi, C. Nucci, M. Paolone, L. Peretto, and R. Tinarelli,

“On the use of continuous-wavelet transform for fault location in distribu-
tion power systems,” International Journal of Electrical Power and Energy
Systems, vol. 28, no. 9, pp. 608 – 617, Nov 2006.
[22] F. H. Magnago and A. Abur, “A new fault location technique for radial
distribution systems based on high frequency signals,” in IEEE Power En-
gineering Society Summer Meeting, vol. 1. Edmonton, Alberta, Canada,
18-22 July 1999, pp. 426–431.
196
References
[23] M. R. Adzman and M. Lehtonen, “A method for ground fault distance

computation in unearthed medium voltage distribution network,” Interna-
tional Review on Modelling and Simulations, vol. 5, no. 4, pp. 1765–1772,
Aug 2012.
[24] A. Borghetti, M. Bosetti, C. Nucci, M. Paolone, and A. Abur, “Integrated

use of time-frequency wavelet decompositions for fault location in distri-
bution networks: Theory and experimental validation,” IEEE Transactions
on Power Delivery, vol. 25, no. 4, pp. 3139–3146, Oct 2010.
[25] M. El-Hami, L. L. Lai, D. J. Daruvala, and A. Johns, “A new travelling-

wave based scheme for fault detection on overhead power distribution feed-
ers,” IEEE Transactions on Power Delivery, vol. 7, no. 4, pp. 1825–1833, Oct
1992.
[26] H. Hizman, P. Crossley, P. Gale, and G. Bryson, “Fault section identifica-

tion and location on a distribution feeder using travelling waves,” in IEEE
Power Engineering Society Summer Meeting, vol. 3. Chicago USA, July
2002, pp. 1107–1112.
[27] S. Navaneethan, J. J. Soraghan, W. Siew, F. McPherson, and P. F. Gale,

“Automatic fault location for underground low voltage distribution net-
works,” IEEE Transactions on Power Delivery, vol. 16, no. 2, pp. 346–351,
Apr 2001.
[28] P. Gale, “Cable-fault location by impulse-current method,” Proceedings of

the Institution of Electrical Engineers, vol. 122, no. 4, pp. 403–408, Apr
1975.
[29] A. Dan and D. Raisz, “Towards a more reliable operation of compensated

networks in case of single phase to ground faults,” in 14th International
Conference on Harmonics and Quality of Power (ICHQP). Bergamo Italy,
Sept 2010, pp. 1–4.
[30] Y.-Y. Hsu et al., “An expert system for locating distribution system faults,”
IEEE Transactions on Power Delivery, vol. 6, no. 1, pp. 366–372, Jan 1991.
[31] P. Järventausta, P. Verho, and J. Partanen, “Using fuzzy sets to model the
uncertainty in the fault location process of distribution networks,” IEEE
Transactions on Power Delivery, vol. 9, no. 2, pp. 954–960, Apr 1994.
[32] Y. D. Xue, B. Y. Xu, and Z. H. Wang, “The fault location technology us-
ing transient signals for single phase earth fault in non-solidly earthed
network,” in 22nd International Conference and Exhibition on Electricity
Distribution (CIRED 2013). Stockholm Sweden, June 2013, pp. 1–4.
[33] W.-R. Chen, Q.-Q. Qian, and X.-R. Wang, “Wavelet neural network based
transient fault signal detection and identification,” in Proceedings of 1997
International Conference on Information, Communications and Signal
Processing (ICICS), vol. 3. Singapore, 9-12 Sept 1997, pp. 1377–1381.
[34] X.-H. Zhang, S.-X. Fan, H. Duan, and B.-Y. Xu, “Analysis of transient dom-
inant frequency signal for single-phase earthed fault based on Prony algo-
rithm,” in China International Conference on Electricity Distribution (CI-
CED). Guangzhou China, 10-13 Dec 2008, pp. 1–6.
197
References
[35] S. Huang and X. G. Wang, “A fault location scheme based on spectrum char-
acteristic of fault-generated high-frequency transient signals,” in IEEE
Power Energy Society General Meeting (PES ’09). Calgary Canada, July
2009, pp. 1–5.
[36] R. Willheim and M. Waters, Neutral grounding in high-voltage transmis-

sion, 1st ed. Elsevier Science Ltd, New York, 1956.
[37] G. Druml, A. Kugi, and O. Seifert, “A new directional transient relay for
high ohmic earth faults,” in CIRED 17th International Conference on Elec-
tricity Distribution. Barcelona Spain, 12-15 May 2003, pp. 1–6.
[38] H. Pundt, “Untersuchung der ausgleichsvorgänge bei erdschluß in en-

ergieversorgungsnetzen,” Energietechnik, vol. 15, pp. 469–477, 1965.
[39] G. Herold, Elektrische Energieversorgung III.: Drehstrommaschinen,

Sternpunktbehandlung, Kurzschlußströme, ser. Elektrische Energiever-
sorgung. Schlembach Fachverlag, 2002.
[40] M. Lehtonen, Transient Analysis for Ground Fault Distance Estimation in

Electrical Distribution Networks, ser. VTT Publication 115. Espoo: Espoo
: Technical Research Centre of Finland, 1992.
[41] P. Schegner and G. Eberl, “Overview of methods for earth fault distance
calculation,” in Seminar on Methods and Techniques for Earth Fault De-
tection, Indication and Location. Espoo Finland, Feb 2011, pp. 36–49.
[42] M. Janssen, Lokalisatie van eenfase-aardfouten in middenspanningsnetten

met niet-geaard sterpunt, ser. TU Eindhoven. Fac. Elektrotechniek : afs-
tudeerverslagen ; 7591. Eindhoven : Technische Universiteit Eindhoven,
Aug 1998.
[43] S. Hänninen, Single phase earth faults in high impedance grounded net-
works: characteristics, indication and location, ser. VTT Publications 453.
Espoo: Espoo : Technical Research Centre of Finland, 2001.
[44] S. Hänninen, M. Lehtonen, T. Hakola, and R. Rantanen, “Comparison of

wavelet and differential equation algorithms in earth fault distance com-
putation,” in Proceedings of 13th Power Systems Computations Conference
(PSCC’99), vol. 2. Trondheim Norway, June 1999, pp. 801–807.
[45] A. G. Phadke and J. S. Thorp, Computer relaying for power systems, 2nd ed.
John Wiley & Sons Ltd, England, 2009.
[46] P. Schegner, “Digitaler erdschlussuniversalschutz: Konzept und erste re-

alisierung,” Ph.D. dissertation, Saarland University, Saarbrucken, 1989.
[47] M. Lehtonen, “Method for distance estimation of single-phase-to-ground

faults in electrical distribution networks with an isolated or compensated
neutral,” European Transactions on Electrical Power, vol. 5, no. 3, pp. 193–
198, May 1995.
[48] M. Igel, “Neuartige verfahren für den erdschlussdistanzschutz in isoliert

und kompensiert betriebenen netzen: Signale und algorithmen im fre-
quenzbereich,” Ph.D. dissertation, Saarland University, Saarbrucken,
1990.
198
References
[49] T. W. Stringfield, D. Marihart, and R. Stevens, “Fault location methods

for overhead lines,” IEEE Transactions of Power Apparatus and Systems,
vol. 76, no. 3, pp. 518–529, Apr 1957.
[50] P. Crossley and P. McLaren, “Distance protection based on travelling

waves,” IEEE Transactions of Power Apparatus and Systems, vol. PAS-
102, no. 9, pp. 2971–2983, Sept 1983.
[51] Z. Q. Bo, G. Weller, F. Jiang, and Q. X. Yang, “Application of GPS based

fault location scheme for distribution system,” in Proceedings of Interna-
tional Conference on Power System Technology (POWERCON ’98), vol. 1.
Beijing China, Aug 1998, pp. 53–57.
[52] M. Aurangzeb, P. Crossley, and P. Gale, “Fault location using the high fre-
quency travelling waves measured at a single location on a transmission
line,” in Seventh International Conference on Developments in Power Sys-
tem Protection. Amsterdam Netherlands, Apr 2001, pp. 403–406.
[53] J. Liang, S. Elangovan, and J. Devotta, “Application of wavelet transform

in travelling wave protection,” International Journal of Electrical Power &
Energy Systems, vol. 22, no. 8, pp. 537 – 542, Nov 2000.
[54] L. V. Bewley, “Traveling waves on transmission systems,” Transactions of

the American Institute of Electrical Engineers, vol. 50, no. 2, pp. 532–550,
June 1931.
[55] A. M. Elhaffar, “Power transmission line fault location based on current

traveling waves,” Ph.D. dissertation, Helsinki University of Technology,
Helsinki, 2008.
[56] M. Gilany, D. Ibrahim, and E. Eldin, “Traveling-wave-based fault-location

scheme for multiend-aged underground cable system,” IEEE Transactions
on Power Delivery, vol. 22, no. 1, pp. 82–89, Jan 2007.
[57] A. Borghetti, M. Bosetti, M. Di Silvestro, C. A. Nucci, and M. Paolone,

“Continuous-wavelet transform for fault location in distribution power net-
works: definition of mother wavelets inferred from fault originated tran-
sients,” IEEE Transactions on Power Systems, vol. 23, no. 2, pp. 380–388,
Apr 2008.
[58] G. Krzysztof, R. Kowalik, and D. Rasolomampionona, “Regular paper trav-

eling wave fault location in power transmission systems: An overview,”
Journal of Electrical Systems (JES), vol. 7, no. 3, pp. 287–296, 2011.
[59] Z. Q. Bo and A. T. Johns, “Transient based protection-a new concept in

power system protection,” in International conference on Power Systems
transients (IPST97). Seattle USA, June 1997, pp. 401–406.
[60] S. Hänninen and M. Lehtonen, Earth fault distance computation with fun-
damental frequency signals based on measurements in substation supply
bay, ser. VTT tiedotteita - Research Notes 2153. Espoo : Technical Re-
search Centre of Finland, 2002.
[61] S. Lin, Z. He, X. Li, and Q. Qian, “Travelling wave time-frequency

characteristic-based fault location method for transmission lines,” IET
Generation, Transmission and Distribution, vol. 6, no. 8, pp. 764–772, Aug
2012.
199
References
[62] P. Gale, P. Crossley, X. Bingyin, Y. Ge, B. Cory, and J. R. G. Barker, “Fault

location based on travelling waves,” in Fifth International Conference on
Developments in Power System Protection. York UK, Mar 1993, pp. 54–
59.
[63] C.-H. Kim et al., “A study on the fault indentification of underground cable
using neural networks,” in Proceedings of International Conference on En-
ergy Management and Power Delivery (EMPD ’95), vol. 2. Singapore, Nov
1995, pp. 571–576.
[64] A. S. Bretas, L. Pires, M. Moreto, and R. H. Salim, “A BP neural network

based technique for HIF detection and location on distribution systems
with distributed generation,” in Computational Intelligence: Lecture Notes
in Computer Science, ser. 4114. Springer Berlin Heidelberg, Aug 2006,
pp. 608–613.
[65] Y. Assef, P. Bastard, and M. Meunier, “Artificial neural networks for single
phase fault detection in resonant grounded power distribution systems,”
in Proceedings of Transmission and Distribution Conference. Los Angeles
USA, Sep 1996, pp. 566–572.
[66] A. Rafinia and J. Moshtagh, “A new approach to fault location in three-

phase underground distribution system using combination of wavelet
analysis with ANN and FLS,” International Journal of Electrical Power
& Energy Systems, vol. 55, pp. 261 – 274, Feb 2014.
[67] S. Ekici, S. Yildirim, and M. Poyraz, “Energy and entropy-based feature

extraction for locating fault on transmission lines by using neural net-
work and wavelet packet decomposition,” Expert Systems with Applica-
tions, vol. 34, no. 4, pp. 2937 – 2944, May 2008.
[68] Z. Yujun and W. Xiaowei, “Research on wavelet neural network for fault lo-
cation in power distribution network,” in Electrical, Information Engineer-
ing and Mechatronics 2011, ser. Lecture Notes in Electrical Engineering,
X. Wang, F. Wang, and S. Zhong, Eds. Springer London, 2012, vol. 138,
pp. 1049–1057.
[69] H. Zayandehroodi, A. Mohamed, M. Farhoodnea, and M. Mohammadjafari,

“An optimal radial basis function neural network for fault location in a
distribution network with high penetration of DG units,” Measurement,
vol. 46, no. 9, pp. 3319 – 3327, Nov 2013.
[70] R. Mahanty and P. D. Gupta, “Application of RBF neural network to fault

classification and location in transmission lines,” IEE Proceedings - Gener-
ation, Transmission and Distribution, vol. 151, pp. 201–212, Mar 2004.
[71] H. Zayandehroodi, A. Mohamed, H. Shareef, and M. Mohammadjafari,

“Performance comparison of MLP and RBF neural networks for fault loca-
tion in distribution networks with DGs,” in IEEE International Conference
on Power and Energy (PECon). Kuala Lumpur Malaysia, Nov 2010, pp.
341–345.
[72] D. Thukaram, H. P. Khincha, and H. P. Vijaynarasimha, “Artificial neural

network and support vector machine approach for locating faults in radial
distribution systems,” IEEE Transactions on Power Delivery, vol. 20, no. 2,
pp. 710–721, Apr 2005.
200
References
[73] R. Salat and S. Osowski, “Accurate fault location in the power transmission
line using support vector machine approach,” IEEE Transactions on Power
Systems, vol. 19, no. 2, pp. 979–986, May 2004.
[74] K. Gayathri and N. Kumarappan, “Accurate fault location on EHV lines

using both RBF based support vector machine and SCALCG based neural
network,” Expert Systems with Applications, vol. 37, no. 12, pp. 8822–8830,
Dec 2010.
[75] R. Aggarwal, M. Joorabian, and Y. Song, “Fuzzy neural network approach

to accurate transmission line fault location,” Engineering Intelligent Sys-
tems for Electrical Engineering and Communications, vol. 5, no. 4, pp. 251–
258, Dec 1997.
[76] F. Chunju, K. Li, W. Chan, Y. Weiyong, and Z. Zhaoning, “Application of

wavelet fuzzy neural network in locating single line to ground fault (SLG)
in distribution lines,” International Journal of Electrical Power & Energy
Systems, vol. 29, no. 6, pp. 497–503, July 2007.
[77] P. Dash, A. Pradhan, and G. Panda, “A novel fuzzy neural network based
distance relaying scheme,” IEEE Transactions on Power Delivery, vol. 15,
no. 3, pp. 902–907, July 2000.
[78] S. Hänninen and M. Lehtonen, “Earth fault distance computation with

artificial neural network trained by neutral voltage transients,” in IEEE
Power Engineering Society Summer Meeting, vol. 2. Vancouver Canada,
July 2001, pp. 1187–1192.
[79] S. Haykin, Neural networks: a comprehensive foundation, 2nd ed. Pren-

tice Hall PTR Upper Saddle River, New Jersey USA, 1999.
[80] T. ElAli, Discrete Systems and Digital Signal Processing with MATLAB,
Second Edition, 2nd ed. Boca Raton : CRC Press, Dec 2011.
[81] H. Darwish, M. Hesham, A.-M. Taalab, and N. Mansour, “Close accord on

DWT performance and real-time implementation for protection applica-
tions,” IEEE Transactions on Power Delivery, vol. 25, no. 4, pp. 2174–2183,
Oct 2010.
[82] D. Robertson, O. Camps, J. Mayer, and W. B. Gish, “Wavelets and electro-

magnetic power system transients,” IEEE Transactions on Power Delivery,
vol. 11, no. 2, pp. 1050–1058, Apr 1996.
[83] W. Cochran et al., “What is the fast Fourier transform?” Proceedings of the
IEEE, vol. 55, no. 10, pp. 1664–1674, Oct 1967.
[84] G. Heydt, “A new method for the calculation of subtransmission and distri-
bution system transients based on the FFT,” IEEE Transactions on Power
Delivery, vol. 4, no. 3, pp. 1869–1875, July 1989.
[85] L. Penghui, Z. Lijie, B. Haijun, and Z. Yanhua, “Power quality monitoring

of power system based on spectrum analysis,” in International Conference
on E-Product E-Service and E-Entertainment (ICEEE). Henan China,
Nov 2010, pp. 1–4.
201
References
[86] L. Mandache and K. Al-Haddad, “New high precision harmonic analysis

method for power quality assessment,” in Canadian Conference on Electri-
cal and Computer Engineering. Saskatoon Canada, May 2005, pp. 1958–
1961.
[87] R. Melicio, V. M. F. Mendes, and J. P. S. Catalao, “Electrical grid integra-

tion and power quality studies of a variable-speed wind energy conversion
system,” in IEEE PowerTech. Bucharest Romania, June 2009, pp. 1–6.
[88] R. Haddad and E. Strangas, “Fault detection and classification in per-

manent magnet synchronous machines using Fast Fourier Transform
and Linear Discriminant Analysis,” in 9th IEEE International Sympo-
sium on Diagnostics for Electric Machines, Power Electronics and Drives
(SDEMPED). Valencia Spain, Aug 2013, pp. 99–104.
[89] K. Musasa, W. Siti, and J. Jordaan, “Evaluating the contribution in the

harmonic distortion level based on Digital Fault Recorder (DFR) analysis,”
in IEEE AFRICON. Livingstone Zambia, Sept 2011, pp. 1–6.
[90] Z. Radojevic, V. Terzija, G. Preston, S. Padmanabhan, and D. Novosel,

“Smart overhead lines autoreclosure algorithm based on detailed fault
analysis,” IEEE Transactions on Smart Grid, vol. 4, no. 4, pp. 1829–1838,
Dec 2013.
[91] M. S. Mamiş, M. Arkan, and C. Keleş, “Transmission lines fault location

using transient signal spectrum,” International Journal of Electrical Power
& Energy Systems, vol. 53, pp. 714 – 718, Dec 2013.
[92] M. Gilany, E. El Din, M. Abdel Aziz, and D. Ibrahim, “An accurate scheme
for fault location in combined overhead line with underground power ca-
ble,” in IEEE Power Engineering Society General Meeting, vol. 3. San
Francisco USA, June 2005, pp. 2521–2527.
[93] Hasniaty, A. Mohamed, and A. Hussain, “Automating power quality distur-

bance analysis using the IPQDA software tool,” in 4th Student Conference
on Research and Development (SCOReD). Selangor Malaysia, June 2006,
pp. 211–214.
[94] C. Moo, Y. Chang, and P. Mok, “A digital measurement scheme for

time-varying transient harmonics,” IEEE Transactions on Power Delivery,
vol. 10, no. 2, pp. 588–594, Apr 1995.
[95] J. Goswami and A. Chan, Fundamentals of Wavelets: Theory, Algorithms,

and Applications, 2nd ed., ser. Wiley Series in Microwave and Optical En-
gineering. John Wiley & Sons , Hoboken, New Jersey, 2011.
[96] R. Rao and A. Bopardikar, Wavelet transforms: introduction to theory and

applications, 1st ed. Reading, MA : Addison-Wesley, 1998.
[97] C. Zhao, M. He, and X. Zhao, “Analysis of transient waveform based on

combined short time Fourier transform and wavelet transform,” in Inter-
national Conference on Power System Technology (PowerCon 2004), vol. 2.
Singapore, Nov 2004, pp. 1122–1126.
[98] L. Satish, “Short-time Fourier and wavelet transforms for fault detection
in power transformers during impulse tests,” IEE Proceedings - Science,
Measurement and Technology, vol. 145, no. 2, pp. 77–84, Mar 1998.
202
References
[99] B. Xu, H. Wang, L. Sun, and F. Yang, “Detection methods of broken rotor
bar fault in squirrel cage induction motor with light-load,” in International
Conference on Electrical Machines and Systems ( ICEMS 2008). Wuhan
China, Oct 2008, pp. 759–762.
[100] J.-Y. Lee, “Variable short-time Fourier transform for vibration signals with
transients,” Journal of Vibration and Control, vol. 1, pp. 1–15, July 2013.
[101] V. Climente-Alarcon et al., “Diagnosis of stator short-circuits through

Wigner-Ville transient-based analysis,” in 35th Annual Conference of IEEE
Industrial Electronics (IECON ’09). Porto Portugal, Nov 2009, pp. 1097–
1102.
[102] N. Mishra and G. V. Anand, “Transient signal detection using Wigner-Ville

distribution and wavelet denoising,” in IEEE Region 10 Conference TEN-
CON. Hongkong, Nov 2006, pp. 1–4.
[103] G. Andria, E. D’Ambrosio, M. Savino, and A. Trotta, “Application of

Wigner-Ville distribution to measurements on transient signals,” in IEEE
Instrumentation and Measurement Technology Conference (IMTC/93).
Irvine California USA, May 1993, pp. 612–617.
[104] J. Rosero, L. Romeral, J. Ortega, and E. Rosero, “Short circuit fault de-
tection in PMSM by means of Empirical Mode Decomposition (EMD) and
Wigner Ville Distribution (WVD),” in IEEE Twenty-Third Annual Applied
Power Electronics Conference and Exposition (APEC 2008). Austin Texas
USA, Feb 2008, pp. 98–103.
[105] J. Fonollosa and C. Nikias, “Analysis of transient signals using higher-

order time-frequency distributions,” in IEEE International Conference on
Acoustics, Speech, and Signal Processing (ICASSP-92), vol. 5. San Fran-
cisco USA, Mar 1992, pp. 197–200.
[106] W. Liu and X. Guo, “Detection of transient power quality disturbances

based EMD combined with Choi-Williams distribution,” in IEEE Interna-
tional Conference on Automation and Logistics (ICAL). Zhengzhou China,
Aug 2012, pp. 588–591.
[107] M. Prieto, A. Espinosa, J.-R. Ruiz, J. Urresty, and J. Ortega, “Feature ex-
traction of demagnetization faults in permanent-magnet synchronous mo-
tors based on box-counting fractal dimension,” IEEE Transactions on In-
dustrial Electronics, vol. 58, no. 5, pp. 1594–1605, May 2011.
[108] N. Elkalashy, M. Lehtonen, H. Darwish, M. Izzularab, and A.-M. Taalab,

“DWT-based investigation of phase currents for detecting high impedance
faults due to leaning trees in unearthed MV networks,” in IEEE Power
Engineering Society General Meeting. Tampa, Florida, USA, June 2007,
pp. 1–7.
[109] F. Jiang, Z. Bo, P. Chin, M. Redfern, and Z. Chen, “Power transformer

protection based on transient detection using discrete wavelet transform
(DWT),” in IEEE Power Engineering Society Winter Meeting, vol. 3. Sin-
gapore, Jan 2000, pp. 1856–1861.
203
References
[110] A. Quinquis and C. Dugal, “Using the wavelet transform for the detec-
tion of magnetic underwater transient signals,” in Proceedings Oceans En-
gineering for Today’s Technology and Tomorrow’s Preservation (OCEANS
’94), vol. 2. Brest, France, Sep 1994, pp. 538 – 543.
[111] F. Truchetet and O. Laligant, “Wavelets in industrial applications: a re-

view,” in Proceedings of the SPIE, vol. 5607. France, Nov 2004, pp. 1–14.
[112] C. H. Kim and R. Aggarwal, “Wavelet transforms in power systems. I. Gen-

eral introduction to the wavelet transforms,” Power Engineering Journal,
vol. 14, no. 2, pp. 81–87, Apr 2000.
[113] C. H. Kim and R. Aggarwal, “Wavelet transforms in power systems. Part

2: Examples of application to actual power system transients,” Power En-
gineering Journal, vol. 15, no. 4, pp. 193–202, Aug 2001.
[114] S. Chen and H. Y. Zhu, “Wavelet transform for processing power quality
disturbances,” EURASIP Journal on Advances in Signal Processing, vol.
2007, pp. 1–20, Feb 2007.
[115] L. Angrisani, P. Daponte, M. D’apuzzo, and A. Testa, “A measurement

method based on the wavelet transform for power quality analysis,” IEEE
Transactions on Power Delivery, vol. 13, no. 4, pp. 990–998, Oct 1998.
[116] M. Karimi, H. Mokhtari, and M. R. Iravani, “Wavelet based on-line dis-

turbance detection for power quality applications,” IEEE Transactions on
Power Delivery, vol. 15, no. 4, pp. 1212–1220, Oct 2000.
[117] S. Santoso, E. Powers, W. Grady, and P. Hofmann, “Power quality assess-

ment via wavelet transform analysis,” IEEE Transactions on Power Deliv-
ery, vol. 11, no. 2, pp. 924–930, Apr 1996.
[118] R. Shariatinasab, M. Akbari, and B. Rahmani, Application of Wavelet

Analysis in Power Systems, ser. Advances in Wavelet Theory and Their
Applications in Engineering, Physics and Technology, D. Baleanu, Ed. Ri-
jeka : InTech, Corp., 2012, vol. 2, no. 221-244.
[119] M. Mirzaei, M. Kadir, E. Moazami, and H. Hizam, “Review of fault location

methods for distribution power system,” Australian Journal of Basic &
Applied Sciences, vol. 3, no. 3, 2009.
[120] R. M. de Castro Fernández and H. N. D. Rojas, “An overview of wavelet

transforms application in power systems,” in Proceedings of the fourteenth
Power Systems Computation Conference (PSCC 14). Seville Spain, June
2002, pp. 1–8.
[121] F. H. Magnago and A. Abur, “Fault location using wavelets,” IEEE Trans-
actions on Power Delivery, vol. 13, no. 4, pp. 1475–1480, Oct 1998.
[122] Z. Wenhai, X. Xianyong, W. Ying, Z. Tianwen, Z. Hong, and W. Henggui,

“Charging-transient based SLG fault location in neutral-unearthed distri-
bution system,” in Asia-Pacific Power and Energy Engineering Conference
(APPEEC). Wuhan China, Mar 2011, pp. 1–4.
[123] A. R. Sedighi, M. R. Haghifam, O. P. Malik, and M. H. Ghassemian, “High

impedance fault detection based on wavelet transform and statistical pat-
tern recognition,” IEEE Transactions on Power Delivery, vol. 20, no. 4, pp.
2414–2421, Oct 2005.
204
References
[124] T. Lobos, P. Kostyla, J. Pospieszna, and M. Jaroszewski, “Location of

faults on transmission lines using wavelet transforms,” in International
Conference on High Voltage Engineering and Application (ICHVE 2008).
Chongqing China, Nov 2008, pp. 633–636.
[125] A. Borghetti, S. Corsi, C. Nucci, M. Paolone, L. Peretto, and R. Tinarelli,

“On the use of continuous-wavelet transform for fault location in distri-
bution power networks,” in 15th Power Systems Computation Conference
(PSCC05). Liege Belgium, 22-26 Aug 2005, pp. 1–7.
[126] M. Misiti, Y. Misiti, G. Oppenheim, and J.-M. Poggi, Wavelet toolbox for use
with MATLAB, 1st ed. Natick, Mass. : Math Works, 1997.
[127] R. Bracewell, The Fourier Transform and Its Applications, 3rd ed., ser.
McGraw-Hill international editions. New Delhi : Tata McGraw-Hill, 2000.
[128] A. Oppenheim and R. Schafer, Discrete-Time Signal Processing: Pearson

New International Edition, 3rd ed. Harlow : Pearson Education Limited,
2013.
[129] M. R. Adzman and M. Lehtonen, “The correlation of fault distance and

transient frequency,” in Seminar on Methods and Techniques for Earth
Fault Detection, Indication and Location. Espoo Finland, Feb 2011, pp.
123–134.
[130] H. Pundt, “Untersuchungen der ausgleichsvorgänge bei erdschluß in

hochspannungsnetzen mit isoliertem sternpunkt und induktiver stern-
punkterdung als grundlage zur selektiven erdschlußerfassung: Beilagen-
band,” Ph.D. dissertation, Dresden University of Technology, 1963.
[131] A. Agarwal and J. Lang, Foundations of Analog and Digital Electronic Cir-
cuits, 3rd ed., ser. The Morgan Kaufmann Series in Computer Architecture
and Design. Amsterdam ; Boston : Elsevier : Morgan Kaufman Publish-
ers, 2005.
[132] K. Winter and A. Ljungberg, Jordfelslokalisering I 10-70 kV nat Vattenfall

FUD - rapport U 1988/39. Ärvkarleby: Vattenfall, 1988.
[133] T. Welfonder, “Localisation de défauts monophasés dans les réseaux de

distribution à neutre compensé,” Ph.D. dissertation, Grenoble Institute of
Technology, 1998.
[134] A. Chuang. (2008, October) Epri smart grid R&D overview. Pre-
sentation at HP sustainability innovation workshop. EPRI. [Online].
Available: http://www.hpl.hp.com/open_innovation/workshops/microgrid%
20and%20distributed%20power/HP_Chuang_102008_post.pdf
[135] P. Järventausta, S. Repo, A. Rautiainen, and J. Partanen, “Smart grid

power system control in distributed generation environment,” Annual Re-
views in Control, vol. 34, no. 2, pp. 277 – 286, Dec 2010.
[136] D. Topolanek and P. Toman, “Possible ground fault location using data
recorded on the secondary side of distribution transformers,” in The 12th
international Scientific Conference on Electric Power Engineering (EPE12).
Kouty nad Desnou, Czech Republic, May 2011, pp. 209–212.
205
References
[137] M. R. Adzman, D. Topolanek, M. Lehtonen, and P. Toman, “An earth fault

location scheme for isolated and compensated neutral distribution sys-
tems,” International Review of Electrical Engineering, vol. 8, no. 5, pp.
1520–1531, Sept 2013.
[138] P. Heine, “Voltage sags in power distribution networks,” Ph.D. dissertation,

Helsinki University of Technology , Helsinki, Nov 2005.
[139] M. R. Adzman and M. Lehtonen, “Fault location in MV unearthed distri-

bution network using the undamped frequency of the transient signal,” in
Proceedings of the 3rd International Conference on Energy, Environment,
Devices, Systems, Communications, Computers (INEEE ’12). Rovaniemi
Finland, Apr 2012, pp. 165–170.
[140] S. H. Brown, “Multiple linear regression analysis: a matrix approach with

MATLAB,” Alabama Journal of Mathematics, vol. 34, pp. 1–3, 2009.
[141] J. E. Dayhoff, Neural Network Architectures: An Introduction. New York

USA: Van Nostrand Reinhold Co., 1990.
[142] M. Danyek, P. Handl, and D. Raisz, “Comparison of simulation tools atp-

emtp and matlab-simulink for time domain power system transient stud-
ies,” in EEUG Meeting. Sopron, Hungary, Dec 2002, pp. 1–14.
[143] H. Høidalen, L. Prikler, and J. Hall, “ATPDraw-Graphical preprocessor

to ATP, Windows version,” in International Conference on Power Systems
Transients (IPST1999). Budapest Hungary, June 1999, pp. 20–24.
[144] S. Kazemi, “Reliability evaluation of smart distribution grids,” Ph.D. dis-

sertation, Aalto University, Oct 2011.
206
Appendices
A. Mathematical Derivations and
MATLAB Scripts
A.1 Derivation of Equation for General Model 1 Damped Natural

Frequencies
R2 L2 C2
L1
E i2 i1 C1
R1
Figure A1. The equivalent circuit for ground fault transient model 1
The equations for voltages around the loops of the circuit are in Laplace-
domain as follows
1
i1 (s)(R1 + SL1 + ) − i2 (s)(R1 + SL1 ) = 0 (1)
SC1
1 E
−i1 (s)(R1 + SL1 ) + i2 (s)(R1 + R2 + SL1 + SL2 + )= (2)
SC2 S
The current i2 (s) can be solved by rearranging i1 (s) from equation (1) and
substituting it into equation (2). After the substitutions and rearranging
the terms, equation (2) becomes:
i2 (s)Q(s) = P (s) (3)
where,
P (s) = C2 E(SR1 C1 + S 2 L1 C1 + 1) (4)
209
Mathematical Derivations and MATLAB Scripts
and
Q(s) = S 4 L2 C2 L1 C1 + S 3 (R2 C2 L1 C1 + L2 C2 R1 C1 )
+S 2 (R2 C2 R1 C1 + L1 C2 + L1 C1 + L2 C2 )
+S(R1 C2 + R2 C2 + R1 C1 ) + 1 (5)
A.2 Derivation of Equation for General Model 2 Undamped Natural

Frequencies
L2 C2
E i2 i1 C1
L1
domain as follows
1
i1 (s)(SL1 + ) − i2 (s)(SL1 ) = 0 (1)
SC1
1 E
−i1 (s)SL1 + i2 (s)(SL1 + SL2 + )= (2)
SC2 S
The current i2 (s) can be solved by rearranging i1 (s) from equation (1)
and substituting it into equation (2). After the substitutions and rear-
ranging the terms, equation (2) becomes,
i2 (s)Q(s) = P (s) (3)
where,
P (s) = (S 2 L1 C1 + 1)EC2 (4)
and
Q(s) = S 4 L2 C2 L1 C1 + S 2 (L1 C2 + L2 C2 + L1 C1 ) + 1 (5)
210
A.3 Derivation of Equation for General Model 3 Damped Natural

Frequencies
R2 L2 C2 R3 L3
L1
E i2 i1 C1 i3 L4 i4 R4
R1
domain as follows
1 1
i1 (s)(R1 + SL1 + ) − i2 (s)(R1 + SL1 ) − i3 (s) =0 (1)
SC1 SC1
1 E
−i1 (s)(R1 + SL1 ) + i2 (s)(R2 + SL1 + SL2 + )= (2)
SC2 S
1 1
−i1 (s) + i3 (s)(R3 + SL3 + SL4 + ) − i4 (s)SL4 = 0 (3)
SC1 SC1
−i3 (s)SL4 + i4 (s)(R4 + SL4 ) = 0 (4)
From which the current i2 (s) can be solved in the following steps, First
Solving i4 (s) from equation (4) and substituting it into equation (3). Then,
solving i3 (s) from equation (3) and substituting it into equation (1). Next,
solving i1 (s) from equation (4) and substituting it into equation (2). After
the substitutions and rearranging the terms, equation (2) becomes:
i2 (s)Q(s) = P (s) (5)
where,
P (s) = EC2 (S 4 L1 L3 C1 L4 + S 3 (L1 R3 C1 L4 + L1 L3 C1 R4
+ R1 L3 C1 L4 + L1 L4 C1 R4 ) + S 2 (L1 R3 C1 R4 + R1 L4 C1 R4 + L4 L1
+ R1 R3 C1 L4 + L3 L4 + R1 L3 C1 R4 ) + S(R3 L4 + R1 R3 C1 R4 + L1 R4
+ L4 R1 + L4 R4 + L3 R4 ) + R1 R4 + R3 R4 ) (6)
211
and,
Q(s) = S 6 (L3 L1 L2 C2 L4 C1 ) + S 5 (C1 L4 L3 L1 R2 C2 + C1 L3 R4 L1 L2 C2
+ C1 L4 L3 R1 L2 C2 + R4 L4 L1 C1 L2 C2 + C1 L4 R3 L1 L2 C2 )+
S 4 (C1 L4 R3 R1 L2 C2 + L4 L1 L2 C2 + C1 L4 R3 L1 R2 C2
+ C 1 L 4 L 3 R1 R 2 C 2 + R4 L 4 R 1 C 1 L 2 C 2 + R 4 L 4 L 1 C 1 R 2 C 2
+ C 1 L 3 R4 L 1 R 2 C 2 + L 4 L 3 L 2 C 2 + C 1 R3 R 4 L 1 L 2 C 2
+ C1 L3 R4 R1 L2 C2 + L3 L4 L1 C2 + L1 L3 C1 L4 )+
S 3 (R4 L4 R1 C1 R2 C2 + C1 L3 R4 R1 R2 C2 + C1 R3 R4 R1 L2 C2
+ C 1 L 4 R 3 R 1 R2 C 2 + C 1 R 3 R 4 L 1 R 2 C 2 + L 4 R 3 L 2 C 2
+ L 4 R 4 L 2 C 2 + L 4 L 1 R2 C 2 + L 4 L 3 R2 C 2 + R 4 L 1 L 2 C 2
+ L 3 R 4 L 2 C 2 + L 4 R 1 L 2 C 2 + L 1 L 3 C 1 R4 + L 1 L 4 C 1 R4
+ L 1 R3 C 1 L 4 + L 3 R4 L 1 C 2 + L 4 R4 L 1 C 2 + R 3 L 4 L 1 C 2
+ R1 L3 C1 L4 + L3 L4 R1 C2 ) + S 2 (L4 R4 R2 C2 + L4 R3 R2 C2
+ L 4 R1 R 2 C 2 + R4 L 1 R 2 C 2 + R3 R 4 L 2 C 2 + L 3 R 4 R2 C 2 + L 1 L 4
+ R4 R 1 L 2 C 2 + C 1 R 3 R 4 R 1 R2 C 2 + L 3 R 4 R 1 C 2 + R 3 L 4 R 1 C 2
+ R3 R 4 L 1 C 2 + L 1 R 3 C 1 R4 + L 4 R 4 R1 C 2 + R 1 L 3 C 1 R4
+ R1 L4 C1 R4 + L3 L4 + R1 R3 C1 L4 ) + S(R3 R4 R2 C2 + R4 R1 R2 C2
+ L 4 R 4 + L 3 R 4 + R3 L 4 + L 1 R4 + R 3 R 4 R1 C 2 + R 1 R3 C 1 R 4
+ L4 R1 ) + R1 R4 + R3 R4 (7)
212
A.4 Derivation of Equation for General Model 4 Undamped Natural

Frequencies
L2 C2
E i2 L1 i1 C1 i3 L4
domain as follows
1 1
i1 (s)(SL1 + ) − i2 (s)SL1 − i3 (s) =0 (1)
SC1 SC1
1 E
−i1 (s)SL1 + i2 (s)(SL1 + SL2 + )= (2)
SC2 S
1 1
−i1 (s) + i3 (s)(SL4 + )=0 (3)
SC1 SC1
From which the current i2 (s) can be solved in the following steps, First
Solving i3 (s) from equation (3) and substituting it into equation (1). Next,
solving i1 (s) from equation (1) and substituting it into equation (2). After
the substitutions and rearranging the terms, equation (2) becomes:
i2 (s)Q(s) = P (s) (4)
where,
P (s) = (S 2 L1 L4 C1 + L1 + L4 )EC2 (5)
and
Q(s) = S 4 L2 C2 L1 L4 C1 + S 2 (L1 C2 L4 + L2 C2 L4 + L2 C2 L1 + L1 L4 C1 )
+L1 + L4 (6)
213
A.5 Example of Matlab Script for MRA model
load X; % Independent variables

load Y1; % Dependent variable
load B1; % FFT amplitude of extracted CWT coefficient
load B2; % Scale of extracted CWT coefficient
load B3; % Total loads prior to earth fault
A=X′∗ X;
K=(X′∗ X)∧ -1;
B=K∗ X′∗ Y1;
Y=B(1,1)+(B(2,1)∗ B1)+(B(3,1)∗ B2)+(B(4,1)∗ B3)
214
B. ATPDraw OH Line Configurations
and Load Variation
B.1 ATPDraw OH Line Geometrical and Electrical Configurations
1.3m 1.3m
Al/Fe 34/6 sparrow

d=6.56 mm
r=0.847Ω/km
soil resistivity=2300Ωm
8.0m
Figure B1. Simulation network model 3 ATPDraw OH configurations
1.1m 1.1m
1.1m
Al/Fe 34/6 sparrow

d=10.11 mm
r=0.536Ω/km
8.1m soil resistivity=2300Ωm
Figure B2. Simulation network model 4 ATPDraw OH configurations
215
ATPDraw OH Line Configurations and Load Variation
B.2 Load Variation for Simulation Model 4
Table B1. Simulation model 4 data.
Length Dist. Load 1 Load 2 Load 3 Load 4 Load 5

Section (kVA) (%) Load Pf
(km) Trans. (DT) kVA kVA kVA kVA kVA
a 0.63 1.00 100.00 4.00 A 14.30 20.02 18.59 10.01 22.17 0.91
b 0.72 2.00 200.00 4.00 B 94.00 131.60 122.20 65.80 145.70 0.95
c 1.90 3.00 315.00 4.00 C 170.00 238.00 221.00 119.00 263.50 0.90
d 0.73 4.00 100.00 4.00 D 35.70 49.98 46.41 24.99 55.34 0.92
e 0.94 5.00 100.00 4.00 E 40.20 56.28 52.26 28.14 62.31 0.93
f 0.35 6.00 200.00 4.00 F 101.00 141.40 131.30 70.70 156.55 0.92
g 1.28 7.00 100.00 4.00 G 29.20 40.88 37.96 20.44 45.26 0.94
h 1.25 8.00 100.00 4.00 H 42.70 59.78 55.51 29.89 66.19 0.91
i 2.21 9.00 100.00 4.00 I 36.30 50.82 47.19 25.41 56.27 0.96
j 1.40 10.00 100.00 4.00 J 11.40 15.96 14.82 7.98 17.67 0.97
k 0.78 11.00 200.00 4.00 K 91.40 127.96 118.82 63.98 141.67 0.98
l 0.38 12.00 100.00 4.00 L 32.90 46.06 42.77 23.03 51.00 0.99
m 0.47 13.00 100.00 4.00 M 23.60 33.04 30.68 16.52 36.58 0.94
n 1.35 14.00 200.00 4.00 N 118.00 165.20 153.40 82.60 182.90 0.95
o 0.50 15.00 100.00 4.00 O 63.60 89.04 82.68 44.52 98.58 0.98
p 0.49 16.00 315.00 4.00 P 154.00 215.60 200.20 107.80 238.70 0.99
q 0.74 17.00 100.00 4.00 Q 2.43 3.40 3.16 1.70 3.77 0.91
r 0.50 18.00 500.00 4.00 R 256.00 358.40 332.80 179.20 396.80 0.93
s 0.72 19.00 200.00 4.00 S 117.00 163.80 152.10 81.90 181.35 0.94
t 1.00 20.00 315.00 4.00 T 167.00 233.80 217.10 116.90 258.85 0.95
u 0.80 21.00 100.00 4.00 U 45.20 63.28 58.76 31.64 70.06 0.96
v 0.44 22.00 100.00 4.00 V 25.90 36.26 33.67 18.13 40.15 0.97
w 0.71 23.00 100.00 4.00 W 49.40 69.16 64.22 34.58 76.57 0.98
x 2.23 24.00 100.00 4.00 X 21.10 29.54 27.43 14.77 32.71 0.92
f1 32.00 25.00 2000.00 6.00 F1 1000.00 1000.00 1000.00 1000.00 1000.00 0.90
f2 20.00 26.00 2000.00 6.00 F2 1000.00 1000.00 1000.00 1000.00 1000.00 0.90
f3 25.00 27.00 2000.00 6.00 F3 1600.00 1600.00 1600.00 1600.00 1600.00 0.90
f4 40.00 28.00 2000.00 6.00 F4 1200.00 1200.00 1200.00 1200.00 1200.00 0.90
f5 19.00 29.00 2000.00 6.00 F5 1100.00 1100.00 1100.00 1100.00 1100.00 0.90
216
C. Tables of Simulation Results
C.1 Tabulated Results of Simulated Network Model 1
Table C1. Comparison of simulation test result to the theory of earth fault transient.
fdc , fnc and ζc are calculated using the GM1 equations. fdm , fnm and ζm
are estimated transient parameters from the simulated earth fault transient
waveforms.
l fd fn
ζ
(km) (Hz) (Hz)
fdm fdc ∆ fd ζm ζc ∆ζ fnm fnc ∆ fn
2 1132.81 1164.22 -31.41 0.0038 0.0039 -0.0001 1132.86 1164.27 -31.41

4 976.56 985.82 -9.25 0.0093 0.0095 -0.0002 976.71 985.96 -9.26
6 859.38 862.41 -3.04 0.0165 0.0166 0.0000 859.61 862.66 -3.05
8 781.25 773.77 7.48 0.0224 0.0233 -0.0009 781.56 774.10 7.46
10 703.13 706.95 -3.83 0.0297 0.0292 0.0005 703.51 707.35 -3.83
12 664.06 654.51 9.55 0.0345 0.0341 0.0005 664.51 654.97 9.54
14 625.00 612.04 12.96 0.0389 0.0383 0.0006 625.49 612.55 12.94
16 585.94 576.77 9.17 0.0407 0.0421 -0.0014 586.47 577.33 9.14
18 546.88 546.90 -0.03 0.0445 0.0455 -0.0009 547.47 547.51 -0.04
20 507.81 521.19 -13.38 0.0491 0.0486 0.0006 508.48 521.85 -13.37
MAE 10.01 0.0006 10.00
Notes: l = Actual fault distance (km), fd = Damped transient frequency (Hz), fdm = Estimated
damped transient frequency (Hz), fdc = Calculated damped transient frequency (Hz), ∆ fd = Error
of damped transient frequency (Hz), ζ = Damping factor, ζm = Estimated damping factor, ζc = Cal-
culated damping factor, ∆ζ = Error of damping factor, fn = Undamped transient frequency (Hz),
fnm = Estimated undamped transient frequency (Hz), fnc = Calculated undamped transient fre-
quency (Hz), ∆ fn = Error of undamped transient frequency (Hz), MAE = Mean average error.
217
Tables of Simulation Results
Table C2. Comparison of the distance estimates obtained using GM algo-

rithms for the simulated single line to earth faults with varies
fault distance.
l (km) GM1f dm ∆GM1f dm GM1f nm ∆GM1f nm GM2 ∆GM2
2 2.30 0.30 2.30 0.30 2.30 0.30

4 4.13 0.13 4.13 0.13 4.13 0.13
6 6.06 0.06 6.06 0.06 6.06 0.06
8 7.81 -0.19 7.81 -0.19 7.81 -0.19
10 10.13 0.13 10.13 0.13 10.13 0.13
12 11.60 -0.40 11.60 -0.40 11.60 -0.40
14 13.35 -0.65 13.35 -0.65 13.35 -0.65
16 15.45 -0.55 15.45 -0.55 15.45 -0.55
18 18.00 0.00 18.00 0.00 18.00 0.00
20 21.16 1.16 21.16 1.16 21.16 1.16
MAE (km) 0.36 0.36 0.36
Notes: l = Actual fault distance (km), GM1f dm = Estimated fault distance using
general model 1 and damped transient frequency (km), ∆GM1f dm = Error of esti-
mated fault distance using general model 1 and damped transient frequency (km),
GM1f nm = Estimated fault distance using General model 1 and undamped tran-
sient frequency (km), ∆GM1f nm = Error of estimated fault distance using general
model 1 and undamped transient frequency (km), GM2f nm = Estimated fault dis-
tance using General model 2 and undamped transient frequency (km), ∆GM2 = Er-
ror of estimated fault distance using general model 2 and undamped transient fre-
quency (km), MAE = Mean average error.
218
Table C3. Comparison of simulation test result to the theory of earth fault transient with the
effect of fault resistance. fdc , fnc and ζc are calculated using GM1 equations. fdm
, fnm and ζm are estimated transient parameters from the simulated earth fault
transient waveforms.
l Rf fd fn
ζ
(km) (Ω) (Hz) (Hz)
0 1132.81 1164.22 -31.41 0.01 0.01 0.00 1132.86 1164.27 -31.41

25 1132.81 1164.53 -31.71 0.0536 0.0538 -0.0002 1134.44 1166.22 -31.77
2
50 1132.81 1165.30 -32.49 0.0980 0.0987 -0.0007 1138.29 1171.02 -32.73
100 1093.75 1168.32 -74.57 0.1650 0.1905 -0.0254 1108.95 1190.11 -81.16
0 976.56 985.82 -9.25 0.0171 0.0174 -0.0003 976.71 985.96 -9.26

25 976.56 984.95 -8.38 0.0594 0.0605 -0.0011 978.29 986.75 -8.46
4
50 976.56 983.10 -6.53 0.1013 0.1037 -0.0024 981.61 988.42 -6.81
100 937.50 976.11 -38.61 0.1594 0.1912 -0.0318 949.64 994.45 -44.81
0 703.13 706.95 -3.83 0.0333 0.0334 -0.0001 703.51 707.35 -3.83

25 703.13 705.87 -2.75 0.0663 0.0668 -0.0005 704.67 707.45 -2.78
10
50 703.13 704.08 -0.95 0.0995 0.1001 -0.0006 706.63 707.63 -1.00
100 664.06 698.26 -34.20 0.1387 0.1669 -0.0282 670.55 708.20 -37.65
0 585.94 576.77 9.17 0.0428 0.0441 -0.0013 586.47 577.33 9.14

25 585.94 575.88 10.05 0.0693 0.0717 -0.0023 587.35 577.37 9.98
16
50 546.88 574.57 -27.69 0.1024 0.0992 0.0032 549.76 577.42 -27.65
100 546.88 570.64 -23.77 0.1403 0.1543 -0.0140 552.34 577.56 -25.22
MAE 21.59 0.01 22.73
Notes: l = Actual fault distance (km), fd = Damped transient frequency (Hz), fdm = Estimated damped
transient frequency (Hz), fdc = Calculated damped transient frequency (Hz), ∆ fd = Error of damped tran-
sient frequency (Hz), ζ = Damping factor, ζm = Estimated damping factor, ζc = Calculated damping factor,
∆ζ = Error of damping factor, fn = Undamped transient frequency (Hz), fnm = Estimated undamped tran-
sient frequency (Hz), fnc = Calculated undamped transient frequency (Hz), ∆ fn = Error of undamped
transient frequency (Hz), Rf = Fault resistance (Ω), MAE = Mean average error.
219
Table C4. Comparison of the distance estimates obtained of the GM algorithm meth-
ods with effect of fault resistance.
l (km) Rf (Ω) GM1f dm ∆GM1f dm GM1f nm ∆GM1f nm GM2 ∆GM2
0 2.30 0.30 2.30 0.30 2.30 0.30

25 2.30 0.30 2.29 0.29 2.29 0.29
2
50 2.30 0.30 2.25 0.25 2.25 0.25
100 2.70 0.70 2.55 0.55 2.55 0.55
0 4.13 0.13 4.13 0.13 4.13 0.13

25 4.13 0.13 4.11 0.11 4.10 0.10
4
50 4.13 0.13 4.06 0.06 4.06 0.06
100 4.70 0.70 4.52 0.52 4.52 0.52
0 10.13 0.13 10.13 0.13 10.13 0.13

25 10.13 0.13 10.09 0.09 10.09 0.09
10
50 10.13 0.13 10.02 0.02 10.02 0.02
100 11.60 1.60 11.36 1.36 11.36 1.36
0 15.45 -0.55 15.45 -0.55 15.45 -0.55

25 15.45 -0.55 15.39 -0.61 15.39 -0.61
16
50 18.00 2.00 17.84 1.84 17.84 1.84
100 18.00 1.65 17.65 1.65 17.65 1.65
MAE (km) 0.59 0.53 0.53
Notes: l = Actual fault distance (km), GM1f dm = Estimated fault distance using general
model 1 and damped transient frequency (km), ∆GM1f dm = Error of estimated fault distance
using general model 1 and damped transient frequency (km), GM1f nm = Estimated fault dis-
tance using General model 1 and undamped transient frequency (km), ∆GM1f nm = Error of
estimated fault distance using general model 1 and undamped transient frequency (km), GM2
= Estimated fault distance using General model 2 and undamped transient frequency (km),
∆GM2 = Error of estimated fault distance using general model 2 and undamped transient fre-
quency (km), Rf = Fault resistance (Ω), MAE = Mean average error.
220
Table C5. Comparison of simulation test result to the theory of ground fault transient
with effect of fault inception angle. fdc , fnc and ζc are calculated using the
GM1 equations. fdm , fnm and ζm are estimated transient parameters from
the simulated earth fault transient waveform.
l Rf fd fn
ζ
(km) (Ω) (Hz) (Hz)
90 976.56 985.82 -9.25 0.0171 0.0174 -0.0003 976.71 985.96 -9.26

4 45 976.56 985.82 -9.25 0.0171 0.0174 -0.0003 976.71 985.96 -9.26
0 976.56 985.82 -9.25 0.0172 0.0174 -0.0002 976.71 985.96 -9.26
90 585.94 576.77 9.17 0.0428 0.0441 -0.0013 586.47 577.33 9.14
16 45 585.94 576.77 9.17 0.0429 0.0441 -0.0012 586.48 577.33 9.14
0 585.94 576.77 9.17 0.0431 0.0441 -0.0010 586.48 577.33 9.15
MAE 9.21 0.0007 9.20
Notes: l = Actual fault distance (km),fd = Damped transient frequency (Hz), fdm = Estimated
damped transient frequency (Hz), fdc = Calculated damped transient frequency (Hz), ∆ fd = Er-
ror of damped transient frequency (Hz), ζ = Damping factor, ζm = Estimated damping factor, ζc
= Calculated damping factor, ∆ζ = Error of damping factor, fn = Undamped transient frequency
(Hz), fnm = Estimated undamped transient frequency (Hz), fnc = Calculated undamped tran-
sient frequency (Hz), ∆ fn = Error of undamped transient frequency (Hz), Rf = Fault resistance
(Ω), MAE = Mean average error.
Table C6. Comparison of the distance estimates obtained of the GM

algorithm methods with the effect of fault inception angle.
l (km) Angle (◦ ) GM1f dm ∆GM1f dm GM1f dm ∆GM1f nm
90 4.13 0.13 4.13 0.13

4 45 4.13 0.13 4.13 0.13
0 4.13 0.13 4.13 0.13
90 15.45 -0.55 15.45 -0.55
16 45 15.45 -0.55 15.45 -0.55
0 15.45 -0.55 15.45 -0.55
MAE (km) 0.34 0.34
Notes: l = Actual fault distance (km), GM1f dm = Estimated fault distance

using general model 1 and damped transient frequency (km), ∆GM1f dm =
Error of estimated fault distance using general model 1 and damped transient
frequency (km), GM1f nm = Estimated fault distance using General model
1 and undamped transient frequency (km), ∆GM1f nm = Error of estimated
fault distance using general model 1 and undamped transient frequency (km),
MAE = Mean average error.
221
Table C7. Comparison of simulation test result to the theory of ground fault transient with the effect
of composite loads. fdc , fnc and ζc are calculated using the general model 3 equations. fdm
, fnm and ζm are estimated transient parameters from the simulated earth fault transient
waveforms.
l Load Rf fd fn
ζ
(km) (MVA) (Ω) (Hz) (Hz)
0 1015.63 1020.08 -4.46 0.0380 0.0388 -0.0009 1016.36 1020.85 -4.49

25 1015.63 1016.61 -0.99 0.0828 0.0853 -0.0025 1019.12 1020.33 -1.20
5.0
50 976.56 1011.57 -35.00 0.1311 0.1318 -0.0007 985.06 1020.47 -35.41
100 976.56 996.34 -19.78 0.2093 0.2256 -0.0163 998.68 1022.71 -24.03
4
0 1015.63 1030.09 -14.46 0.0284 0.0288 -0.0004 1016.04 1030.52 -14.48
25 1015.63 1027.84 -12.22 0.0743 0.0754 -0.0011 1018.44 1030.77 -12.34
12.5
50 1015.63 1024.01 -8.39 0.1194 0.1221 -0.0027 1022.94 1031.73 -8.79
100 976.56 1011.21 -34.64 0.1839 0.2163 -0.0324 993.51 1035.72 -42.21
0 703.13 714.78 -11.65 0.0438 0.0435 0.0003 703.80 715.45 -11.65

25 703.13 713.07 -9.95 0.0779 0.0777 0.0002 705.27 715.23 -9.96
5.0
50 703.13 710.56 -7.44 0.1106 0.1119 -0.0012 707.47 715.05 -7.58
100 664.06 703.08 -39.01 0.1592 0.1803 -0.0210 672.64 714.79 -42.14
10
0 703.13 719.16 -16.03 0.0403 0.0399 0.0005 703.70 719.73 -16.03
25 703.13 717.74 -14.61 0.0748 0.0741 0.0007 705.10 719.72 -14.61
12.5
50 703.13 715.51 -12.38 0.1097 0.1084 0.0013 707.39 719.75 -12.36
100 664.06 708.58 -44.52 0.1773 0.1770 0.0003 674.75 719.95 -45.20
0 585.94 579.92 6.02 0.0492 0.0505 -0.0013 586.65 580.66 5.99

25 585.94 578.76 7.18 0.0761 0.0784 -0.0022 587.64 580.55 7.10
5.0
50 585.94 577.15 8.78 0.1027 0.1062 -0.0035 589.05 580.44 8.61
100 546.88 572.57 -25.70 0.1454 0.1620 -0.0166 552.75 580.24 -27.49
16
0 585.94 582.53 3.41 0.0477 0.0488 -0.0011 586.61 583.22 3.39
25 585.94 581.48 4.46 0.0749 0.0767 -0.0018 587.59 583.20 4.39
12.5
50 585.94 579.99 5.95 0.1017 0.1047 -0.0030 588.99 583.19 5.80
100 546.88 575.63 -28.75 0.1722 0.1607 0.0115 555.17 583.20 -28.04
MAE 15.66 0.0052 16.39
Notes: l = Actual fault distance (km), fd = Damped transient frequency (Hz), fdm = Estimated damped transient
frequency (Hz), fdc = Calculated damped transient frequency (Hz), ∆ fd = Error of damped transient frequency (Hz),
ζ = Damping factor, ζm = Estimated damping factor, ζc = Calculated damping factor, ∆ζ = Error of damping factor,
fn = Undamped transient frequency (Hz), fnm = Estimated undamped transient frequency (Hz), fnc = Calculated
undamped transient frequency (Hz), ∆ fn = Error of undamped transient frequency (Hz), Rf = Fault resistance (Ω),
MAE = Mean average error.
222
Table C8. Comparison of the distance estimates obtained of the GM algorithm methods with
effect of composite loads.
l (km) MVA Rf (Ω) GM1 ∆GM1 GM2 ∆GM2 GM3 ∆GM3 GM4 ∆GM4
0 3.60 -0.40 3.60 -0.40 4.05 0.05 3.63 -0.37

25 3.57 -0.43 3.57 -0.43 4.02 0.02 3.59 -0.41
5.0
50 4.01 0.01 4.01 0.01 4.44 0.44 4.04 0.04
100 3.83 -0.17 3.83 -0.17 4.27 0.27 3.85 -0.15
4
0 3.61 -0.39 3.61 -0.39 4.17 0.17 3.63 -0.37
25 3.58 -0.42 3.57 -0.43 4.14 0.14 3.60 -0.40
12.5
50 3.52 -0.48 3.52 -0.48 4.09 0.09 3.55 -0.45
100 3.90 -0.10 3.90 -0.10 4.45 0.45 3.92 -0.08
0 10.12 0.12 10.12 0.12 10.39 0.39 10.14 0.14

25 10.07 0.07 10.07 0.07 10.34 0.34 10.09 0.09
5.0
50 10.00 0.00 9.99 -0.01 10.26 0.26 10.02 0.02
100 11.27 1.27 11.27 1.27 11.52 1.52 11.29 1.29
10
0 10.13 0.13 10.12 0.12 10.53 0.53 10.17 0.17
25 10.08 0.08 10.08 0.08 10.48 0.48 10.12 0.12
12.5
50 10.00 0.00 10.00 0.00 10.41 0.41 10.05 0.05
100 11.19 1.19 11.19 1.19 11.59 1.59 11.24 1.24
0 15.44 -0.56 15.44 -0.56 15.64 -0.36 15.45 -0.55

25 15.38 -0.62 15.38 -0.62 15.59 -0.41 15.39 -0.61
5.0
50 15.30 -0.70 15.29 -0.71 15.50 -0.50 15.31 -0.69
100 17.63 1.63 17.62 1.62 17.82 1.82 17.64 1.64
16
0 15.44 -0.56 15.44 -0.56 15.80 -0.20 15.46 -0.54
25 15.38 -0.62 15.38 -0.62 15.74 -0.26 15.40 -0.60
12.5
50 15.30 -0.70 15.30 -0.70 15.66 -0.34 15.32 -0.68
100 17.46 1.46 17.45 1.45 17.80 1.80 17.47 1.47
MAE (km) 0.51 0.51 0.54 0.51
Notes: l = Actual fault distance (km), GM1 = Estimated fault distance using General model 1 and
undamped transient frequency (km), ∆GM1 = Error of estimated fault distance using general model 1
and undamped transient frequency (km), GM2 = Estimated fault distance using General model 2 and un-
damped transient frequency (km), ∆GM2 = Error of estimated fault distance using general model 2 and
undamped transient frequency, GM3 = Estimated fault distance using General model 3 and undamped
transient frequency (km), ∆GM3 = Error of estimated fault distance using general model 3 and undamped
transient frequency (km), GM4 = Estimated fault distance using General model 4 and undamped tran-
sient frequency (km), ∆GM4 = Error of estimated fault distance using general model 4 and undamped
transient frequency (km), MAE = Mean average error.
223
Table C9. Comparison of the distance estimates obtained be-

tween EX1 and GM2 algorithm. fdm , fnm and ζ are
estimated transient parameters from the simulated
earth fault transient waveform.
l fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
2 1250.00 0.015 1250.13 2.23 0.23 2.88 0.88

4 1083.33 0.020 1083.56 3.91 -0.09 4.21 0.21
6 955.88 0.025 956.19 5.69 -0.31 5.70 -0.30
8 855.26 0.029 855.63 7.61 -0.39 7.37 -0.63
10 773.81 0.034 774.26 9.68 -0.32 9.20 -0.80
12 706.52 0.036 706.99 11.90 -0.10 11.22 -0.78
14 650.00 0.041 650.55 14.25 0.25 13.42 -0.59
16 625.00 0.043 625.58 15.48 -0.52 14.58 -1.42
18 580.36 0.047 580.99 18.03 0.03 17.04 -0.96
20 560.34 0.048 560.99 19.35 -0.65 18.34 -1.66
MAE 0.29 0.82
Notes:
l = Actual fault distance (km), fdm = Estimated damped transient fre-
quency (Hz), ζ = Damping factor, fnm = Estimated undamped transient
frequency (Hz), EM1 = Estimated fault distance using exact model 1 (km)
, ∆EM1 = Error of estimated fault distance using exact model 1 (km), GM2
= Estimated fault distance using General model 2, ∆GM2 = Error of es-
timated fault distance using general model 2 (km), MAE = Mean average
error.
224
Table C10. Comparison of the distance estimates obtained between

EM1 and GM2 algorithm with the effect of fault resistance.
fdm , fnm and ζ are estimated transient parameters from
l Rf fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
10 1083.33 0.0370 1084.08 3.90 -0.10 4.21 0.21

30 1083.33 0.0705 1086.04 3.88 -0.12 4.19 0.19
4
50 1083.33 0.1041 1089.26 3.84 -0.16 4.16 0.16
100 1015.63 0.1978 1036.09 4.51 0.51 4.71 0.71
MAE 0.22 0.32
10 773.81 0.0456 774.61 9.67 -0.33 9.20 -0.80

30 773.81 0.0707 775.75 9.64 -0.36 9.17 -0.83
10
50 773.81 0.0955 777.36 9.59 -0.41 9.12 -0.88
100 738.64 0.1634 748.69 10.46 0.46 9.91 -0.09
MAE 0.39 0.65
10 625.00 0.0529 625.88 15.46 -0.54 14.56 -1.44

30 625.00 0.0737 626.70 15.42 -0.58 14.52 -1.48
16
50 601.85 0.0980 604.76 16.61 0.61 15.66 -0.34
100 601.85 0.1498 608.72 16.38 0.38 15.45 -0.55
MAE 0.53 0.95
Overall MAE (km) 0.38 0.64
Notes: l = Actual fault distance (km), fdm = Estimated damped transient

frequency (Hz), ζ = Damping factor, fnm = Estimated undamped transient fre-
quency (Hz), EM1 = Estimated fault distance using exact model 1 (km) , ∆EM1
= Error of estimated fault distance using exact model 1 (km), GM2 = Estimated
fault distance using General model 2, ∆GM2 = Error of estimated fault distance
using general model 2 (km), MAE = Mean average error.
Table C11. Comparison of the distance estimates obtained between EM1 and
GM2 algorithm with the effect of fault inception angle. fdm , fnm
and ζ are estimated transient parameters from the simulated earth
fault transient waveform.
l Angle (◦ ) fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
90.00 1083.33 0.020 1083.56 3.91 -0.09 4.21 0.21

4 45.00 1083.33 0.020 1083.56 3.91 -0.09 4.21 0.21
0.00 1083.33 0.020 1083.56 3.91 -0.09 4.21 0.21
90.00 625.00 0.042 625.56 15.48 -0.52 14.58 -1.42

16 45.00 625.00 0.042 625.56 15.48 -0.52 14.58 -1.42
0.00 601.85 0.043 602.42 16.74 0.74 15.79 -0.21
MAE (km) 0.34 0.61
Notes: l = Actual fault distance (km), fdm = Estimated damped transient frequency (Hz),
ζ = Damping factor, fnm = Estimated undamped transient frequency (Hz), EM1 = Estimated
fault distance using exact model 1 (km) , ∆EM1 = Error of estimated fault distance using
exact model 1 (km), GM2 = Estimated fault distance using General model 2, ∆GM2 = Error
of estimated fault distance using general model 2 (km), MAE = Mean average error.
225
Table C12. Comparison of the distance estimates obtained between EM1 and GM2 algorithm with
the effect of resistive loads. The loads are located at the end of each feeders. fdm ,
fnm and ζ are estimated transient parameters from the simulated earth fault transient
waveform.
l Load Rf fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
30 1250.00 0.095 1255.70 2.19 0.19 2.85 0.85

400 kW 50 1250.00 0.136 1261.80 2.14 0.14 2.81 0.81
100 1160.71 0.219 1189.53 2.78 0.78 3.31 1.31
30 1250.00 0.120 1259.16 2.16 0.16 2.83 0.83

2.00 800 kW 50 1250.00 0.173 1269.23 2.07 0.07 2.76 0.76
100 1160.71 0.223 1190.72 2.77 0.77 3.30 1.30
30 1354.17 0.107 1362.05 1.35 -0.65 2.23 0.23

2 MW 50 1250.00 0.158 1265.94 2.10 0.10 2.78 0.78
100 1250.00 0.206 1277.52 2.01 0.01 2.71 0.71
MAE 0.32 0.84
30 773.81 0.084 776.58 9.62 -0.38 9.14 -0.86

400 kW 50 738.64 0.114 743.46 10.63 0.63 10.06 0.06
100 738.64 0.177 750.51 10.40 0.40 9.86 -0.14
30 773.81 0.096 777.42 9.59 -0.41 9.12 -0.88

10.00 800 kW 50 738.64 0.127 744.67 10.59 0.59 10.03 0.03
100 738.64 0.189 752.16 10.35 0.35 9.81 -0.19
30 773.81 0.118 779.24 9.54 -0.46 9.08 -0.92

2 MW 50 773.81 0.143 781.88 9.46 -0.54 9.01 -0.99
100 738.64 0.204 754.52 10.28 0.28 9.74 -0.26
MAE 0.45 0.48
30 580.36 0.091 582.79 17.92 -0.08 16.93 -1.07

400 kW 50 580.36 0.111 584.00 17.84 -0.16 16.86 -1.14
100 560.34 0.167 568.29 18.85 0.85 17.85 -0.15
30 580.36 0.104 583.55 17.87 -0.13 16.89 -1.11

18.00 800 kW 50 580.36 0.125 584.91 17.78 -0.22 16.80 -1.20
100 560.34 0.180 569.62 18.77 0.77 17.76 -0.24
30 580.36 0.138 585.94 17.72 -0.28 16.74 -1.26

2 MW 50 580.36 0.158 587.78 17.61 -0.39 16.63 -1.37
100 560.34 0.215 573.81 18.49 0.49 17.49 -0.51
MAE (km) 0.37 0.89
Notes: l = Actual fault distance (km), fdm = Estimated damped transient frequency (Hz), ζ = Damping factor,
fnm = Estimated undamped transient frequency (Hz), EM1 = Estimated fault distance using exact model 1 (km)
, ∆EM1 = Error of estimated fault distance using exact model 1 (km), GM2 = Estimated fault distance using Gen-
eral model 2, ∆GM2 = Error of estimated fault distance using general model 2 (km), MAE = Mean average error.
226
Table C13. Comparison of the distance estimates obtained between EM1 and GM2 algorithm
with the effect of inductive loads. The loads are located at the end of each feeders.
fdm , fnm and ζ are estimated transient parameters from the simulated earth fault
transient waveform.
l Load Rf fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
30 1250.00 0.069 1253.03 2.21 0.21 2.86 0.86

400 kVAr 50 1250.00 0.106 1257.02 2.18 0.18 2.84 0.84
100 1250.00 0.210 1278.56 2.00 0.00 2.70 0.70
30 1250.00 0.070 1253.05 2.21 0.21 2.86 0.86

2.00 800 kVAr 50 1250.00 0.106 1257.07 2.18 0.18 2.84 0.84
100 1250.00 0.174 1269.44 2.07 0.07 2.76 0.76
30 1250.00 0.070 1253.10 2.21 0.21 2.86 0.86

2 MVAr 50 1250.00 0.107 1257.21 2.17 0.17 2.84 0.84
100 1250.00 0.176 1269.87 2.07 0.07 2.76 0.76
MAE 0.14 0.81
30 773.81 0.071 775.77 9.64 -0.36 9.17 -0.83

400 kVAr 50 738.64 0.101 742.41 10.66 0.66 10.09 0.09
100 738.64 0.147 746.74 10.52 0.52 9.96 -0.04
30 773.81 0.071 775.74 9.64 -0.36 9.17 -0.83

10.00 800 kVAr 50 738.64 0.100 742.37 10.66 0.66 10.09 0.09
100 738.64 0.165 748.85 10.46 0.46 9.90 -0.10
30 773.81 0.071 775.74 9.64 -0.36 9.17 -0.83

2 MVAr 50 773.81 0.096 777.37 9.59 -0.41 9.12 -0.88
100 738.64 0.165 748.95 10.45 0.45 9.90 -0.10
MAE 0.47 0.42
30 580.36 0.089 582.67 17.92 -0.08 16.94 -1.06

400 kVAr 50 580.36 0.096 583.02 17.90 -0.10 16.92 -1.08
100 580.36 0.143 586.41 17.69 -0.31 16.71 -1.29
30 580.36 0.076 582.03 17.96 -0.04 16.98 -1.02

18.00 800 kVAr 50 580.36 0.096 583.02 17.90 -0.10 16.92 -1.08
100 580.36 0.143 586.41 17.69 -0.31 16.71 -1.29
30 580.36 0.076 582.03 17.96 -0.04 16.98 -1.02

2 MVAr 50 580.36 0.095 583.02 17.90 -0.10 16.92 -1.08
100 580.36 0.144 586.43 17.69 -0.31 16.71 -1.29
MAE 0.15 1.14
Notes: l = Actual fault distance (km), fdm = Estimated damped transient frequency (Hz), ζ = Damping
factor, fnm = Estimated undamped transient frequency (Hz), EM1 = Estimated fault distance using exact
model 1 (km) , ∆EM1 = Error of estimated fault distance using exact model 1 (km), GM2 = Estimated fault
distance using General model 2, ∆GM2 = Error of estimated fault distance using general model 2 (km), MAE
= Mean average error.
227
Table C14. Comparison of the distance estimates obtained between EM1 and GM2 algo-
rithm with the effect of loads of 0.9 power factor. The loads are located at the
end of each feeders. fdm , fnm and ζ are estimated transient parameters from
l PF (MVA) Rf fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
30.00 1250.00 0.096 1255.86 2.19 0.19 2.85 0.85

400 50.00 1250.00 0.140 1262.46 2.13 0.13 2.80 0.80
100.00 1160.71 0.261 1202.36 2.66 0.66 3.21 1.21
2.00 0.90 30.00 1250.00 0.120 1259.03 2.16 0.16 2.83 0.83
800 50.00 1250.00 0.164 1267.13 2.09 0.09 2.77 0.77
100.00 1160.71 0.294 1214.23 2.55 0.55 3.13 1.13
MAE 0.30 0.93
30.00 580.36 0.090 582.72 17.92 -0.08 16.94 -1.06

400 50.00 580.36 0.110 583.90 17.85 -0.15 16.86 -1.14
100.00 560.34 0.165 568.15 18.86 0.86 17.86 -0.14
18.00 0.90 30.00 580.36 0.101 583.37 17.88 -0.12 16.90 -1.10
800 50.00 580.36 0.122 584.70 17.80 -0.20 16.82 -1.18
100.00 560.34 0.177 569.32 18.79 0.79 17.78 -0.22
MAE 0.37 0.81
Table C15. Comparison of the distance estimates obtained between EM1 and GM2 algorithm with the
effect of loads of 0.5 power factor. The loads are located at the end of each feeders. fdm
, fnm and ζ are estimated transient parameters from the simulated earth fault transient
waveform
l PF (MVA) Rf fdm ζ fnm EM1 ∆EM1 GM2 ∆GM2
30.00 1250.00 0.084 1254.40 2.20 0.20 2.86 0.86

400 50.00 1250.00 0.122 1259.47 2.16 0.16 2.82 0.82
100.00 1250.00 0.202 1276.18 2.02 0.02 2.72 0.72
2.00 0.50 30.00 1250.00 0.098 1256.05 2.18 0.18 2.85 0.85
800 50.00 1250.00 0.135 1261.63 2.14 0.14 2.81 0.81
100.00 1250.00 0.203 1276.46 2.01 0.01 2.72 0.72
MAE 0.12 0.79
30.00 580.36 0.081 582.28 17.95 -0.05 16.96 -1.04

400 50.00 580.36 0.101 583.34 17.88 -0.12 16.90 -1.10
100.00 560.34 0.156 567.25 18.92 0.92 17.92 -0.08
18.00 0.50 30.00 580.36 0.087 582.59 17.93 -0.07 16.94 -1.06
800 50.00 580.36 0.107 583.73 17.86 -0.14 16.87 -1.13
100.00 560.34 0.162 567.85 18.88 0.88 17.88 -0.12
MAE 0.37 0.75
Notes: l = Actual fault distance (km), fdm = Estimated damped transient frequency (Hz), ζ = Damping factor,
fnm = Estimated undamped transient frequency (Hz), EM1 = Estimated fault distance using exact model 1 (km) ,
∆EM1 = Error of estimated fault distance using exact model 1 (km), GM2 = Estimated fault distance using General
model 2, ∆GM2 = Error of estimated fault distance using general model 2 (km), MAE = Mean average error.
228
Table C16. Comparison of the distance estimates obtained from

multiple regression (MRA) and neural network (NN)
algorithm with the effect of fault resistance (Rf ).
Fault inception angle is 90◦ .
l Load (MVA) Rf (Ω) MRA ∆MRA NN ∆NN
0 1.07 0.77 1.51 0.33

1.84 7.55 25 2.03 -0.19 2.28 -0.44
50 2.01 -0.16 2.22 -0.37
0 2.14 0.2 2.32 0.02

2.34 6.93 25 2.02 0.33 2.18 0.16
50 3.04 -0.7 3.02 -0.67
0 4.34 -0.11 4.28 0.17

4.45 6.97 25 4.17 -0.28 4.2 0.25
50 5.17 0.72 5.01 -0.56
0 8.71 -0.84 8.28 -0.42

7.86 6.93 25 8.5 -0.63 8.22 -0.35
50 8.46 -0.58 8.08 -0.22
0 8.63 -0.55 8.28 -0.2

8.07 7.55 25 8.47 -0.39 8.14 -0.06
50 8.43 -0.35 8.01 0.07
0 14.1 0.54 12.99 0.58

13.56 5.51 25 13.83 0.26 14 -0.43
50 13.74 0.17 13.8 -0.23
0 16.31 0.71 16.72 -1.13

15.59 7.78 25 16.07 0.47 16.22 -0.62
50 15.97 0.38 15.84 -0.25
MAE (km) 0.44 0.36
Notes: l = Actual fault distance (km), MRA = Estimated fault distance

using multiple regression algorithm (km), ∆MRA = Error of estimated
fault distance using multiple regression algorithm(km), NN = Estimated
fault distance using neural network algorithm (km), ∆NN = Error of esti-
mated fault distance using neural network algorithm (km), MAE = Mean
average error (km).
229
Table C17. Comparison of the distance estimates obtained from

multiple regression (MRA) and neural network (NN)
algorithm with the effect of fault resistance (Rf ).
Fault inception angle is 45◦ .
l Load (MVA) Rf (Ω) MRA ∆MRA NN ∆NN
0 2.04 -0.19 1.53 0.31

1.84 7.55 25 1.98 -0.14 2.23 -0.38
50 1.96 -0.12 2.16 -0.31
0 2.04 0.31 1.97 0.38

2.34 6.93 25 1.95 0.39 1.73 0.61
50 1.93 0.41 1.66 0.69
0 4.21 0.24 3.9 0.55

4.45 6.97 25 4.12 0.34 3.71 0.75
50 4.09 0.36 3.61 0.84
0 8.66 -0.8 7.86 0

7.86 6.93 25 8.5 -0.64 7.89 -0.02
50 8.45 -0.59 7.76 0.11
0 8.6 -0.53 8.23 -0.15

8.07 7.55 25 8.47 -0.39 8.33 -0.25
50 8.42 -0.34 8.27 -0.19
0 14.04 -0.48 13.25 0.32

13.56 5.51 25 13.81 -0.24 13.51 0.06
50 13.73 -0.16 13.5 0.07
0 16.19 -0.59 15.75 -0.16

15.59 7.78 25 16.02 -0.42 16.09 -0.5
50 15.96 -0.37 16.03 -0.44
MAE (km) 0.38 0.34
Notes: l = Actual fault distance (km), MRA = Estimated fault distance

using multiple regression algorithm (km), ∆MRA = Error of estimated
fault distance using multiple regression algorithm(km), NN = Estimated
fault distance using neural network algorithm (km), ∆NN = Error of esti-
mated fault distance using neural network algorithm (km), MAE = Mean
average error (km).
230
Table C18. Comparison of the distance estimates obtained from general model
(GM1,GM2,GM4) algorithms with the effect of fault resistance
(Rf ). Signal measured from MV/LV side and fault inception an-
gle is 90◦ .
l Load (MVA) Rf (Ω) GM1 ∆GM1 GM2 ∆GM2 GM4 ∆GM4
0.00 1.64 -0.20 1.59 -0.25 1.65 -0.12

1.84 7.55 25.00 2.80 0.96 2.47 0.63 2.52 0.74
50.00 2.80 0.96 1.91 0.07 1.97 0.20
0.00 2.80 0.46 2.74 0.40 2.79 0.50
2.34 6.93 25.00 2.80 0.46 2.48 0.14 2.53 0.24
50.00 4.17 1.83 3.40 1.06 3.45 1.15
0.00 4.17 -0.28 4.12 -0.34 4.15 -0.25
4.45 6.97 25.00 4.17 -0.28 3.86 -0.59 3.90 -0.51
50.00 5.94 1.49 5.31 0.86 5.34 0.93
0.00 7.30 -0.56 7.24 -0.62 7.27 -0.56
7.86 6.93 25.00 11.15 3.29 10.57 2.71 10.59 2.76
50.00 11.15 3.29 10.42 2.56 10.45 2.62
0.00 8.38 0.31 7.23 -0.84 7.26 -0.77
8.07 7.55 25.00 8.38 0.31 7.98 -0.09 8.01 -0.03
50.00 8.38 0.31 7.90 -0.17 7.93 -0.11
0.00 12.01 -1.56 11.96 -1.60 11.99 -1.55
13.56 5.51 25.00 16.40 2.84 15.92 2.36 15.95 2.40
50.00 16.40 2.84 15.31 1.75 15.34 1.79
0.00 16.40 0.81 16.30 0.71 16.33 0.77
15.59 7.78 25.00 16.40 0.81 15.92 0.33 15.94 0.38
50.00 16.40 0.81 16.07 0.48 16.09 0.54
MAE 1.17 0.88 0.90
Notes: l = Actual fault distance (km), GM1 = Estimated fault distance using general
model 1 (km), ∆GM1 = Error of estimated fault distance using general model 1 (km),
GM2 = Estimated fault distance using general model 2 (km), ∆GM2 = Error of estimated
fault distance using general model 2 (km), GM4 = Estimated fault distance using general
model 4 (km), ∆GM4 = Error of estimated fault distance using general model 4 (km),
MAE = Mean average error (km).
231
Table C19. Comparison of the distance estimates obtained from general model
(GM1, GM2,GM4) algorithms with the effect of fault resistance (Rf ).
Signal measured from MV/LV side and fault inception angle is 45◦ .
l Load (MVA) Rf (Ω) GM1 ∆GM1 GM2 ∆GM2 GM4 ∆GM4
0.00 1.86 0.02 1.81 -0.03 1.87 0.10

1.84 7.55 25.00 1.86 0.02 1.84 0.00 1.90 0.13
50.00 1.86 0.02 1.74 -0.10 1.80 0.03
0.00 1.86 -0.48 1.81 -0.53 1.87 -0.41
2.34 6.93 25.00 1.86 -0.48 1.82 -0.52 1.87 -0.41
50.00 1.86 -0.48 1.74 -0.60 1.80 -0.48
0.00 3.59 -0.87 3.54 -0.91 3.58 -0.83
4.45 6.97 25.00 3.59 -0.87 3.40 -1.05 3.44 -0.96
50.00 3.59 -0.87 2.71 -1.74 2.76 -1.64
0.00 8.38 0.52 8.31 0.45 8.34 0.51
7.86 6.93 25.00 8.38 0.52 8.01 0.15 8.04 0.21
50.00 8.38 0.52 7.73 -0.13 7.76 -0.07
0.00 8.38 0.31 8.30 0.23 8.33 0.29
8.07 7.55 25.00 8.38 0.31 8.18 0.11 8.21 0.18
50.00 9.65 1.58 8.83 0.76 8.86 0.83
0.00 13.98 0.42 13.92 0.36 13.94 0.40
13.56 5.51 25.00 13.98 0.42 13.58 0.02 13.60 0.06
50.00 13.98 0.42 13.06 -0.50 13.08 -0.46
0.00 13.98 -1.61 13.89 -1.70 13.92 -1.64
15.59 7.78 25.00 13.98 -1.61 13.55 -2.04 13.58 -1.98
50.00 13.98 -1.61 13.09 -2.50 13.12 -2.44
MAE (km) 0.66 0.69 0.67
Notes: l = Actual fault distance (km), GM1 = Estimated fault distance using general model
1 (km), ∆GM1 = Error of estimated fault distance using general model 1 (km), GM2 = Esti-
mated fault distance using general model 2 (km), ∆GM2 = Error of estimated fault distance
using general model 2 (km), GM4 = Estimated fault distance using general model 4 (km),
∆GM4 = Error of estimated fault distance using general model 4 (km), MAE = Mean average
error (km).
232
Table C20. Comparison of the distance estimates obtained from exact model
(EM1-GM3) algorithms with the effect of fault resistance (Rf ). Sig-
nal measured from MV/LV side and fault inception angle is 90◦ .
l Load (MVA) Rf (∆) EM1 ∆EM1 EM2 ∆EM2 EM3 ∆EM3
0.00 1.24 -0.60 1.20 -0.64 1.19 -0.65

1.84 7.55 25.00 2.41 0.57 2.28 0.44 2.26 0.42
50.00 1.70 -0.14 1.63 -0.21 1.62 -0.22
0.00 2.74 0.40 2.58 0.24 2.56 0.22
2.34 6.93 25.00 2.42 0.08 2.29 -0.05 2.27 -0.07
50.00 3.51 1.17 3.30 0.96 3.26 0.92
0.00 4.31 -0.14 4.05 -0.40 3.99 -0.46
4.45 6.97 25.00 4.02 -0.43 3.78 -0.67 3.73 -0.72
50.00 5.63 1.18 5.31 0.86 5.19 0.74
0.00 7.73 -0.13 7.35 -0.51 7.12 -0.74
7.86 6.93 25.00 11.29 3.43 10.98 3.12 10.44 2.58
50.00 11.14 3.28 10.83 2.97 10.29 2.43
0.00 7.72 -0.35 7.34 -0.73 7.11 -0.96
8.07 7.55 25.00 8.53 0.46 8.15 0.08 7.86 -0.21
50.00 8.44 0.37 8.06 -0.01 7.78 -0.29
0.00 12.77 -0.79 12.55 -1.01 11.83 -1.73
13.56 5.51 25.00 16.90 3.34 17.16 3.60 15.78 2.22
50.00 16.27 2.71 16.43 2.87 15.18 1.62
0.00 17.29 1.70 17.62 2.03 16.17 0.58
15.59 7.78 25.00 16.89 1.30 17.15 1.56 15.78 0.19
50.00 17.05 1.46 17.34 1.75 15.93 0.34
MAE (km) 1.14 1.18 0.87
Notes: l = Actual fault distance (km), EM1 = Estimated fault distance using exact model 1
(km), ∆EM1 = Error of estimated fault distance using exact model 1 (km), EM2 = Estimated
fault distance using exact model 2 (km), ∆EM2 = Error of estimated fault distance using Ex-
act model 2 (km), EM3 = Estimated fault distance using exact model 3 (km), ∆GM3 = Error
of estimated fault distance using exact model 3 (km), MAE = Mean average error (km).
233
Table C21. Comparison of the distance estimates obtained from exact model
(EM1-GM3) algorithms with the effect of fault resistance (Rf ). Sig-
nal measured from MV/LV side and fault inception angle is 45◦ .
l Load (MVA) Rf (Ω) EM1 ∆EM1 EM2 ∆EM2 EM3 ∆EM3
0.00 1.56 -0.28 1.49 -0.35 1.49 -0.35

1.84 7.55 25.00 1.61 -0.23 1.54 -0.30 1.53 -0.31
50.00 1.46 -0.38 1.40 -0.44 1.40 -0.44
0.00 1.56 -0.78 1.50 -0.84 1.49 -0.85
2.34 6.93 25.00 1.57 -0.77 1.50 -0.84 1.50 -0.84
50.00 1.47 -0.87 1.41 -0.93 1.40 -0.94
0.00 3.66 -0.79 3.44 -1.01 3.40 -1.05
4.45 6.97 25.00 3.51 -0.94 3.30 -1.15 3.26 -1.19
50.00 2.70 -1.75 2.55 -1.90 2.53 -1.92
0.00 8.88 1.02 8.51 0.65 8.19 0.33
7.86 6.93 25.00 8.56 0.70 8.18 0.32 7.89 0.03
50.00 8.26 0.40 7.88 0.02 7.61 -0.25
0.00 8.87 0.80 8.49 0.42 8.18 0.11
8.07 7.55 25.00 8.75 0.68 8.37 0.30 8.06 -0.01
50.00 9.45 1.38 9.08 1.01 8.71 0.64
0.00 14.82 1.26 14.80 1.24 13.79 0.23
13.56 5.51 25.00 14.46 0.90 14.40 0.84 13.44 -0.12
50.00 13.92 0.36 13.81 0.25 12.93 -0.63
0.00 14.79 -0.80 14.77 -0.82 13.76 -1.83
15.59 7.78 25.00 14.44 -1.15 14.37 -1.22 13.42 -2.17
50.00 13.96 -1.63 13.85 -1.74 12.96 -2.63
MAE (km) 0.85 0.79 0.80
Notes: l = Actual fault distance (km), EM1 = Estimated fault distance using exact model 1
(km), ∆EM1 = Error of estimated fault distance using exact model 1 (km), EM2 = Estimated
fault distance using exact model 2 (km), ∆EM2 = Error of estimated fault distance using Ex-
act model 2 (km), EM3 = Estimated fault distance using exact model 3 (km), ∆GM3 = Error
of estimated fault distance using exact model 3 (km), MAE = Mean average error (km).
234
De
par
tme
nto
fEl
ect
ric
alE
ngi
nee
ringa
ndAut
oma
ti
on
T his disse rtat io n fo c use so n singl eph aset o

Eart
hFaul tD ist
anc e
Mo
al
t
eart hfaul ts in me dium vo ltage( MV)
h
-
D
d Rafi Adz
Computation Me tho ds
D1
dist ribut io n fe e de rs.Singl eph aset oe art h
89
faul tis th emo stc o mmo n faul tt ypein po w er
/
20
dist ribut io n ne tw o rks.I nF inland and in
1
4
Basedo
nT ransients in
mo sto fc o unt rie s in Euro pet heMV
man
ne tw o rks areo pe rat edw ithne ut raliso l
ated
or re so nante art h e d syst e ms.T h ew ay the
PowerDistribution Syste
ms
Eart
ne ut ralis c o nne c te d is impo rtantsinc eit
w illaffe c tth et ypeo f pro t e ction syst em
hFaul
ne ede d and po w e r syst e mc o mpo ne nt
tD
se l
e c t
io n.I n iso lat e d ne ut ralne tw orks e arth
ist
faul tc urre nt s arere lative l y smal lc o mpare d
anc
w ithe arth e d syst e ms.T h einit ialt ransie nts
eCo
ofe art hfaul t s areimpo rt antfo r faul t Mo
hdRa
fiAd
zma
n
mput
locat io ne spe c ially in une art h e d and
co mpe nsat ed ne ut ralne tw o rk.T h e
atio
elec tro magne tict ransie ntpro videt h e
n Me
avail abl einfo rmat io n fo re st imat ing faul t
t
locat io n.T his disse rtat io n pro po se s fault
ho
ds Base
locat io n algo rit h ms base do ne art hfaul t
transie ntsignal s me asure d atse co ndary
sideo f MV and L V dist ribut io nt ransfo rme rs
donT
in orde rt ol oc at ea singl el inet oe art hfaul t
ransie
in une arth e d and c o mpe nsat ed ne ut ralMV
dist ribut io n ne tw o rks.
nts in P
owe
rDist
ribut
ion Syst
ems
I
S BN978-
95 2
-60-
5973-0(pri
nted) BUSI
N E
S S+
9HSTFMG*afjhda+
I
S BN978-
95 2
-60-
5974-7(pdf) ECONOM Y
I
S SN-
L1799-4934
I
S SN1799-
4934(pri
nted) ART+
I
S SN1799-
4942(pdf) DESI
G N+
ARCH I
TECT
U RE
A
Aa
lt
oU ni
versi
ty
a
Sc
hoolofElect
ric
alE ngi
ne e
ring SCI
ENCE+
lt
oU
Depa
rtmentofElect
rica
lE ngi
neeri
nga
ndAut
oma
ti
on T
EC HNOL
O GY
www.
aalt
o.f
i
ni
CROSSOV
ER
ve
DOCTORAL
r
DOCTORAL DI
S SE
R T
ATI
O NS
sit
DI
S SE
R T
ATI
O NS
y

Earth Fault Distance Computation Methods

Uploaded by

Copyright:

Available Formats

Earth Fault Distance Computation Methods

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Earth Fault Distance Computation Methods

Uploaded by

Copyright:

Available Formats

De

T his disse rtat io n fo c use so n singl eph aset o

EarthF aultD ist

Opp one nts

I owe special thanks to Dr. David Topolanek for a useful discussions

I would like to express my appreciation for the ﬁnancial support I re-

I am personally want to thank all of my Malaysian friends and their

families for their outstanding support, encouragement and concerns during

Last but not least, I want to express my heartfelt thanks to my beloved

Vantaa, November 10, 2014,

Mohd Raﬁ Adzman

Contribution of the author 7

2. Transients of Single Phase to Earth Fault 39

3. Fault Location Methods Based on Earth Fault Transient Sig-

3.6 Traveling Wave method . . . . . . . . . . . . . . . . . . . . . . 52

4. Determination of Transient Components with Signal Pro-

5. Fault Distance Estimation Using Transient Signals 65

6. Performance Evaluation of Proposed Fault Location Algo-

6.3.1 Description of the simulation and the network model 135

7. Performance Evaluation of Proposed Fault Location Algo-

8. Conclusions and Future Work 191

Appendix A. Mathematical Derivations and MATLAB Scripts 209

Appendix B. ATPDraw OH Line Configurations and Load Vari-

Appendix C. Tables of Simulation Results 217

C.3 Tabulated Results of Simulated Network Model 4 . . . . . . 229

5.1 Description of the composite components. . . . . . . . . . . . 67

7.1 Case 1 network parameters. . . . . . . . . . . . . . . . . . . . 167

B1 Simulation model 4 data. . . . . . . . . . . . . . . . . . . . . . 216

C1 Comparison of simulation test result to the theory of earth

C6 Comparison of the distance estimates obtained of the GM

C20 Comparison of the distance estimates obtained from EM al-

2.1 An example of the earth fault transient signal in MV over-

3.1 Propagation of traveling wave in power network due to earth

4.1 Example of the ouput of pre-processing of transient signal

5.1 The general model for earth fault transients. . . . . . . . . . 67

5.5 The equivalent circuit for earth fault transient of general

6.1 A schematic diagram and ATPdraw simulated network model

6.5 Example of the output from pre-processing of current tran-

6.22 Comparison of calculated undamped frequencies with the

6.39 Test number 2 (11.70 km) correct path identiﬁcation. . . . . 145

6.54 Comparison of the distance estimate errors obtained from

7.24 Recorded signals of experiments 4. . . . . . . . . . . . . . . . 187

A1 The equivalent circuit for ground fault transient model 1 . . 209

B1 Simulation network model 3 ATPDraw OH conﬁgurations . 215

Ll,n Inductance per unit length of negative sequence system

Uv Analytic form of the extracted wavelet coefﬁcient of the voltage transient

ωdm Estimated damped transient frequency (rad/s)

OLE ’Olesna’ distribution substation

1.1 Background and Problem Statement

In Finland and in most of countries in Europe the medium voltage (MV)

1.2 Research Objective

ing a single line to earth fault in unearthed and compensated neutral MV

1.3 Research Contribution

1.4 Organization of Thesis

The thesis is organized into the following chapters:

Chapter 2 presents general knowledge of earth fault transient signals.

Chapter 3 presents some known fault location methods based on earth

Chapter 4 presents the determination of transient components with a

Chapter 5 presents the proposed fault distance estimation algorithms

Chapter 6 presents performance evaluation of the proposed fault loca-