Indonesian J
ournal of Ele
c
trical Engin
eering and
Computer Sci
e
nce
Vol. 1, No. 2,
February 20
1
6
, pp. 300 ~
309
DOI: 10.115
9
1
/ijeecs.v1.i2.pp30
0-3
0
9
300
Re
cei
v
ed Se
ptem
ber 20, 2015; Revi
se
d Jan
uary 12,
2016; Accept
ed Ja
nua
ry 2
8
, 2016
Loss of Excitation Faults Detection in Hydro-
Generators Using an Adaptive Neuro Fuzzy Inference
System
M.S. Abdel Az
iz
*
1
, M. Elsamah
y
2
, M.A.
Moustafa Ha
ssan
3
, F. Be
ndar
y
4
1
Dar Al-Han
da
sah (Sha
ir an
d partners), Eg
ypt
2
Elec. Po
w
e
rand Computer Engineering D
ept., Universit
y
of Saskatche
w
a
n,
Canada
3
Elec. Po
w
e
r Dept., Facult
y
of
Engin
eering,
Cairo Universit
y
, Egy
p
t.
4
Elec. Po
w
e
r Dept., Facult
y
of
Engi
neering, Benha
Un
iversit
y
, Egy
p
t.
Corresp
on
din
g
author, e-Mai
l
: mohame
d
sal
a
h84
4@
ya
ho
o.com
A
b
st
r
a
ct
T
h
is
paper presents
a new
approach
for Loss of
Excitation (LOE) faults
detection in Hydro-
generators
using Adaptive Neuro
F
u
z
z
y
Inference System.
T
he pr
oposed scheme
w
a
s trained by data
from simulation of a 345kV system under
various f
aults conditions and tested for different
loading
conditions. Details of the design process and the re
sults of performance usi
ng the proposed technique are
discussed in the paper.
T
w
o different techniques
are di
scussed in
this article according
to the type
of
inputs
to the proposed ANF
I
S unit,
the generator te
rminal impedance measurements
(R and X) and the
generator RMS Line
to Line
voltage and
Phase current
(V
tr
m
s
and I
a
). T
he
tw
o proposed
techniques results
are
compared w
i
th each other
and are compared w
i
th
t
he traditional distance relay
response in addition to
other
techniques. The results
show
t
hat the
proposed Artificial Intelligent
based technique is
efficient in the
Loss of Excitation faults (LOE) detection
proce
ss and the
obtained
results
are very
promising.
Ke
y
w
ords
:
Adaptive Neuro F
u
z
z
y
Inference System,
Loss
of Excitation, Hydr
o-Generator, Dynamic
Performance, Simulation
Copy
right
©
2016 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduction
Loss of Excitation (LOE
) is a very wide
spr
ead fault in synch
r
o
nou
s machine
s
a
nd can
be ca
used b
y
short
circui
t of the field windin
g
, un
expecte
d field bre
a
ker o
p
en or L
o
ss
of
Excitation (L
OE) rel
a
y ma
l-ope
ratio
n
. Accordi
ng to t
he stati
s
tics i
n
Chi
na, the
gene
rato
r failure
due to
L
o
ss
of Excitation
(LOE
) a
c
cou
n
ts fo
r 6
9
.5%
of all
ge
nera
t
or failu
re
sa
s de
scrib
ed i
n
[1,
2]. Loss of Excitation (L
O
E
) may ca
use sh
arp
dam
age
s to both
gene
rato
r an
d system. F
o
r the
gene
rato
r; when Lo
ss of Excitation (L
OE) hap
pen
s,
a slip occu
rs whi
c
h m
a
y cau
s
e
rotor o
v
er
heating d
ue to the slip fre
q
uen
cy in rotor circui
ts. Also
, as the ma
ch
ine ope
rate
s
as an in
du
ction
machi
ne after Loss of Exci
tation (LOE
) con
d
ition
s
, large am
ount o
f
reactive po
wer
su
pplied
by
stator cu
rrent
is req
u
ire
d
a
nd
the stato
r
may
su
ffe
r o
v
er he
ating b
e
ca
use of thi
s
la
rge
curre
n
t.
On the oth
e
r hand, fo
r th
e syste
m
; its voltage de
cl
i
nes
after the
gene
rato
r lo
se its
excitati
on,
becau
se the
gene
rato
r op
erate
s
a
s
an
indu
ction
ma
chin
e and a
b
sorbs
rea
c
tive power fro
m
th
e
system.
F
o
r some wea
k
system,
the system
volt
age
may coll
ap
se
due to
the L
o
ss of Excitat
i
on
(LOE
) of
an i
m
porta
nt ge
n
e
rato
ra
s expl
ained
in [3
].
Also,
when
a
gen
erato
r
l
o
se
s its excitat
i
on,
other generat
o
rs i
n
the sy
stem
will increase their
reactive po
wer output. This may cause t
h
e
overloa
d
ing i
n
som
e
tra
n
smissi
on lin
es
or tra
n
sf
o
r
me
rs a
nd the
over-cu
r
rent rel
a
y may con
s
i
d
e
r
this overlo
adi
ng as a fault and isolate
the non-fa
ult equipme
n
t [4-14]. The
s
e
above rea
s
o
n
s
motivate this
research work
to s
o
lve for
this
problem.
In the mi
ddle
of the
20
th
ce
ntury, a
sin
g
le ph
ase offse
t
mho
relay
was
develo
ped
for th
e
high
spe
ed
d
e
tection
of L
o
ss of Excita
tion (L
OE) condition
s
in
synchro
nou
s gene
rato
rs.
T
h
is
distan
ce rela
y approa
ch
wa
s develop
ed to pr
ovid
e enhan
ce
d sele
ctivity between Lo
ss of
Excitation (L
OE)
con
d
itio
ns
and
othe
r
norm
a
l o
r
ab
norm
a
l o
pera
t
ing conditio
n
s
a
n
d
to p
r
ov
ide
the op
eratin
g
times n
e
cessary for optimu
m
protecti
o
n
of both th
e g
e
nerato
r
and
the
system
[1
5].
Over the yea
r
s, the offset mho relay ha
s be
en wi
del
y accepted fo
r loss of ex
citation protecti
on
and experi
e
n
c
e with
t
he relay
ha
s bee
n
a
c
cepted.
The relay ha
s
d
e
mo
nstrated
its capa
bil
i
ty
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Loss of Excitation Faults
Dete
ction in
Hydro
-
G
ene
rators
Using a
n
Adaptive
…
(M.S. Abdel Aziz)
301
detectin
g
different excitati
on system fail
ure
s
and
to di
scrimin
a
te be
tween
su
ch failure
s an
d other
operating
con
d
itions. Th
e relatively few
ca
se
s of
in
co
rre
ct op
eratio
n that have o
c
curred
ca
n
be
refered to in
corre
c
t rel
a
y conn
ectio
n
s (majo
r
ca
use), and bl
own potential transfo
rme
r
fu
se
s.
Reg
a
rdl
e
ss
of this a
c
ce
pted expe
rie
n
ce, the
r
e
has
been
some u
s
e
r
worry a
bout t
h
e
perfo
rman
ce
of dista
n
ce type of
relayin
g
for l
o
ss
of ex
citation
prote
c
tion. In
pa
rticula
r
, the
r
e
h
a
s
been
con
c
ern
over po
ssibl
e
in
co
rre
ct o
p
e
ration
of
th
e
rel
a
y when
o
peratin
g the
gene
rato
r in
the
unde
rexcite
d
regio
n
, du
rin
g
stabl
e tra
n
s
ient
swi
n
g
s
and d
u
ri
ng m
a
jor
system
d
i
sturb
a
n
c
e
s
t
hat
cau
s
e u
nde
r
freque
ncy
co
ndition
s. In view of this
co
ntinuing
con
c
ern ove
r
rel
a
y perform
an
ce,
a
gene
ral stu
d
y
was laun
ch
ed to review the perform
ance of the offset mho L
o
ss of Excitation
(LOE
) relay d
i
fferent syste
m
con
d
ition
s
. S
ubse
que
ntly, many approac
he
s and
algorith
m
s h
a
v
e
been a
ddressed to solve th
e gene
rato
rs
Loss
of Excitation (LOE
) p
r
oble
m
su
ch
as:
a)
Fuzzy inference mechanism based technique [16].
b)
ANN based technique [17].
c)
Adaptive Loss of Excitation relay
basedon time-derivatives of impedance[18].
d)
Adaptive loss of excitation
protection relay based on the steady-state stability limit [19].
e)
Technique based on
the derivativ
e
of the
terminal voltage and
t
he output
reactive power
of
the generator [20].
Thus
, the necess
ity for this
arti
cle
came i
n
to si
ght a
s
t
he d
e
ficien
cy
of Lo
ss of Excitation
(LOE
) di
stan
ce
relay
s
be
came
clea
r. M
o
reove
r
; the
s
e dist
ance
rel
a
ys b
ehavio
r
to different
L
o
ss
of Excitation
(LOE)
co
nditio
n
s i
s
totally
d
epen
ding
on t
he g
ene
rato
r
loadin
g
an
d t
he p
e
rcenta
g
e
loss of excitation and many
loss of excita
tion (L
OE)
co
ndition
s are n
o
t detected b
y
these relay
s
.
Therefore, th
e nee
d for d
e
veloping
an
Artificial
Inte
lligent (AI) b
a
sed
relay to
overcome th
ese
probl
em
s app
eare
d
.
This a
r
ticle p
r
esents t
w
o rece
nt optimization
algo
rith
ms ba
se
d on
Artificial Intelligen
ce
(AI) te
chniq
u
e
s. Th
e two
different te
ch
nique
s di
scu
s
sed i
n
this
article
cl
assifi
ed b
a
sed o
n
the
type of input
s to the p
r
o
p
o
s
ed
ANFIS u
n
it, t
he gen
erator te
rminal
i
m
peda
nce m
easure
m
ent
s
(R
and X
)
and
the g
ene
rato
r RMS
Lin
e
to
Line
voltage
and
Pha
s
e
current
(V
trms
and I
a
)
.
T
h
e tw
o
prop
osed te
chni
que
s re
sults are
co
mpared wi
th
each oth
e
r and are
co
mpared with
the
conve
n
tional
distan
ce
rela
y re
spo
n
se in
additio
n
to
o
t
her te
ch
niqu
es. T
he
re
sul
t
s sho
w
th
at the
proposed A
r
tificial Intelligent (AI) based techni
ques
are effici
ent in the Loss of
Excitation faults
(LOE
) detecti
on pro
c
e
s
s. The obtain
e
d
result
s ar
e
very promi
s
i
ng. The re
st of the paper is
orga
nized a
s
follows: Section 2 pre
s
e
n
ts the sy
stem u
nder
study, while Section 3
describ
es th
e
Adaptive Ne
u
r
o Fu
zzy Inferen
c
e System
techniq
ue,
o
n
the other
h
and, Sectio
n 4 illuminate
s
the
simulatio
n
en
vironme
n
t an
d finally, Section 5 pre
s
e
n
ts the re
sult
s and di
scussio
n
.
2. Sy
stem
Under
Study
The
system
use
d
in
the i
n
vestigatio
ns of this
pa
per is
sh
own in
Figure 1. It
consi
s
ts of
two hyd
r
o-ge
nerato
r
s which are c
onn
ected via tran
sf
orme
rs to an
infinite-bu
s
system thro
ug
h a
300 km, 34
5 kV tran
smissi
on line. The
system dat
a
are given in
Appendix
-
A, as given in [21].
The PSCAD/
E
MTDC
simu
lation pa
ckag
e is used for i
n
the simulati
on pro
c
e
s
s [22].
Figure 1. One
-
line Di
ag
ram
of the Simulation Model in
PSCAD
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 300 – 309
302
3.
Adaptiv
e
Neuro Fuzzy
Inference Sy
stem (ANFIS)
A Fuzzy Logi
c System
(FL
S
) ca
n be vi
ewe
d
a
s
a n
on-lin
ea
r ma
pping from th
e inpu
t
spa
c
e to the
output space.
A FLS con
s
i
s
ts of
five main c
o
mponents
:
Fuzz
y S
e
ts
, fuzz
ifiers
,
fuzzy rule
s, a
n
inferen
c
e e
ngine
an
d
d
e
fuzzifiers
. However fuzzy i
n
ference sy
st
em i
s
limited
in
its appli
c
ation
to only modeling ill defined
system
s.
These
syste
m
s h
a
ve rule
stru
ctu
r
e
wh
ich i
s
e
s
senti
a
lly pre
determined
by the
use
r
'
s
interp
retation
of the
cha
r
a
c
teri
stic
of th
e varia
b
le
s in
the mo
del. It ha
s be
en
co
nsid
ere
d
o
n
ly
fixed memb
ership fu
ncti
ons that
we
re
cho
s
e
n
arbitrarily. Howeve
r, in
some
mo
deli
n
g
situation
s
, it can
not be di
stingui
sh
ed
what the
me
mbershi
p
fun
c
tion
s sh
ould
look like sim
p
ly
from loo
k
in
g
at data. Rather tha
n
cho
o
si
ng the
para
m
eters asso
ciated
with a giv
en
membe
r
ship
function a
r
bitrarily, these paramet
e
r
s co
uld b
e
cho
s
e
n
so
as t
o
t
a
ilor
t
h
e
membe
r
ship f
unctio
n
s to
th
e input/outp
u
t data in
order to acco
unt fo
r the
s
e type
s
of variation
s
in the
data
va
lues. In
su
ch
ca
se th
e n
e
cessit
y of the
ANFIS be
co
mes obviou
s
.
Adaptive
Ne
uro
-
Fuzzy n
e
two
r
ks a
r
e
enh
an
ced
FLS
s
with lea
r
ni
n
g
, g
eneralization,
and
ad
aptive capa
bilities.
These net
wo
rks en
co
de t
he fuzzy if-th
en rul
e
s
into
a neu
ral net
work-like stru
cture
and th
e
n
use
ap
pro
p
riate lea
r
nin
g
alg
o
rithm
s
to
minim
i
ze th
e
out
put e
rro
r
b
a
se
d o
n
th
e
training/valid
ation data sets [23-2
8
].
Neuro-adaptive learning techniques
provide
a
method for the
fuzzy modeling
procedure
to
learn information
about a
data set. It
computes
the
membership function
parameters that best
allow the associated fuzzy inference to
track the given input/output data.
A
network-type structure simila
r
to that of
an Artificial Neural
Network (ANN) can be
used to interpret
the input/output map.
Theref
ore, it maps
inputs through input
membership
functions and
associated parameters,
and t
hen through
output membership
functions and
associated parameters to outputs. These par
ameters change through the learning process.
The used ANFIS is assumed to have
the following properties [27, 28]:
a)
It is
z
e
ro
th
order sugeno-type system.
b)
It has a single
output, obtained using we
ighted
average defuzzification. All
output
membership functions are constant.
c)
It has no rule sharing. Different rule
s do not share the same output membership
function;
the number of output membership
functions must be equal to the number
of rules.
d)
It has unity weight for each rule.
Figure
2 shows the architecture of
the AN
FIS, comprising by input, fuzzificaiton,
inference and
defuzzificaiton layers.
The network
c
an
be visualized
as consisting
of inputs, with
N
neurons in the input layer and F input membership functions for each input, with F * N
neurons in the
fuzzificaiton layer. There
are F^
N rules
with F^N neurons
in the inference
and
defuzzificaiton layers. It is assumed one neuron in the output layer.
Figure 2. The
Archite
c
ture
of the ANFIS
The propo
se
d ANFIS uni
t consi
s
t
s
of two neu
rons in the input layer i.e. N=2, six
Membe
r
ship
Functio
n
s
(M
F) for
ea
ch i
nput i.e.
M=6
and
con
s
tan
t
membershi
p
functio
n
for the
output layer, Appendix
-
B.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Loss of Excitation Faults
Dete
ction in
Hydro
-
G
ene
rators
Using a
n
Adaptive
…
(M.S. Abdel Aziz)
303
4. Simulation
Env
i
ronment
The sim
u
lati
on environm
ent based o
n
t
he MATL
AB software
packag
e
(T
he Math
Wo
rks, Natick, Ma
ssachu
setts,
USA) i
s
sele
cted a
s
the m
a
in e
ngine
erin
g to
ol for p
e
rfo
r
ming
modelin
g and
simulatio
n
of powe
r
syste
m
s an
d rel
a
ys. The PSCA
D
/EMTDC p
r
ogra
m
is u
s
e
d
for
detailed
mod
e
ling
of a
po
wer n
e
two
r
k
and
simul
a
tio
n
of inte
re
stin
g event
s. Sce
nario
setting
and
a relaying algorithm
will be implemented in t
he MATLAB program,
while
the dat
a generation
for
training and testing of this
algorithm w
ill
be executed
by the PSCAD/EMTDC program.
The u
s
ed trai
ning data to
train the ANFIS ar
e take
n at Loss of
Excitation (L
OE) fault
con
d
ition
s
an
d no-fault con
d
itions.
The fault con
d
itions a
r
e ca
rrie
d
out at di
fferent Lo
ss of
Excitation (L
OE) fault types:
a)
Partial Lo
ss o
f
Excitation (LOE) faults.
b)
Compl
e
te Lo
ss of Ex
c
i
tation (LOE
) faults
.
These fault condition
s are carrie
d out at diffe
rent gen
erato
r
s lo
adin
g
con
d
itions
(18.5%,
25%, 35%, 40%, 50%, 55%, 60%, 65%, 70% and 80
%) with incep
t
ion fault time T
f
= 5 se
c an
d
different Lo
ss of Excitation (LOE
) ca
se
s
(20%, 25%, 50%, 60%, 70%, 75%, 80% and 100%
).
Two
differe
nt propo
sed
m
e
thod
s a
r
e
compa
r
ed
wit
h
ea
ch
othe
r in thi
s
a
r
ticl
e for the
purp
o
se of Loss of Excitation (LOE) fa
ults
dete
c
tion
, one metho
d
is based o
n
the gene
ra
tor
terminal i
m
p
edan
ce
mea
s
ureme
n
ts
(R an
d X)
an
d the oth
e
r i
s
ba
se
d o
n
the ge
ne
rator RMS
Line to
Line
voltage an
d
Phase
curren
t (V
trms
and I
a
) me
asure
m
e
n
ts, the
obtai
ned
re
sults from
both schem
e
s
are b
e
tter than [29, 30].
Testing
data
are cho
s
en
rand
omly from the
data
that were in
clud
ed in the
training
pro
c
e
ss,
whil
e the validation data
a
r
e chosen at different condition
s data that n
o
t were incl
u
ded
in the training
process to e
n
su
re the p
r
o
posed metho
d
profici
e
n
c
y.
The se
que
nce of the prop
ose
d
(R a
nd
X) techni
que
is ba
sed o
n
tran
sform
a
tion
of the
gene
rato
r te
rminal volta
g
e
s
a
n
d
stato
r
curre
n
ts
mag
n
itude
s a
nd
angle
s
whi
c
h
are o
b
taine
d
by
Fouri
e
r Tran
sform to impe
dan
ce mea
s
u
r
eme
n
ts (R and X) whi
c
h are involved
in the trainin
g
,
testing an
d validating p
r
o
c
esse
s.
Figure 3 pre
s
ents the flowcha
r
t for the Loss of Excitation
(LOE
) detection p
r
o
c
edure of
the prop
osed
(R an
d X) pro
t
ection sch
e
m
e. On the
other ha
nd, Fig
u
re 4 de
pict
s the flowchart
for
the Loss of Excitation (LO
E
) detectio
n
pro
c
ed
ure of the other pro
posed (V
trms
and I
a
) prote
c
tion
scheme.
5.
Results and Discussion
The syste
m
wa
s simul
a
te
d using PSCAD/EMTDC a
s
well a
s
Mat
l
ab and the result
s of
simulatio
n
are explained i
n
this pap
er.
5.1.
The Propos
e
d
(R and X)
Protec
tion Scheme
The in
puts to
the ANFIS
u
n
it are
the g
e
ner
ato
r
te
rmi
nal imp
eda
nce mea
s
u
r
em
ents
(R
and X) which
are
obtain
e
d
from the
ge
nerato
r
te
rm
i
nal voltage
a
nd stato
r
cu
rrent value
s
, a
n
d
the testing
da
ta are
ch
ose
n
to have
dat
a from th
e tra
i
ning p
r
o
c
e
ss while
the vali
dation d
a
ta a
r
e
cho
s
e
n
to ha
ve data not includ
ed in the
training p
r
o
c
e
ss.
Table 1
pre
s
e
n
ts the testin
g data of the
pr
op
osed (R
and X) A
N
FIS sch
eme. T
he testin
g
data a
r
e in
clu
ded inth
e trai
ning p
r
o
c
e
s
s.Table
(2
) illu
strates th
e vali
dation d
a
ta o
f
the pro
p
o
s
e
d
(R a
nd X)
scheme. Th
e validation d
a
ta
are n
o
t incl
u
ded in th
e tra
i
ning p
r
o
c
e
s
s and a
r
e
cho
s
en
at different ge
nerato
r
loa
d
in
g and Lo
ss of Excitation (L
OE) co
ndition
s.
Table 1 an
d
2 depict the
promi
s
ing a
c
cura
cy of the prop
osed (R and X) A
N
FIS in
detectin
g
the
gene
rato
r L
o
ss of Excita
tion (L
OE
) fa
ults un
der dif
f
erent lo
adin
g
co
ndition
s
and
comp
are it wi
th the conve
n
tional di
stan
ce
rela
y
resp
onse. The te
sting time
col
u
mn
sho
w
s the
Loss of Excitation (LOE
) d
e
tection time
by the
propo
sed ANFIS rel
a
y, which
wh
en com
p
a
r
ed
to
the tradition
al
distan
ce rela
y trip time sh
ows t
he prom
ising effici
en
cy of the prop
ose
d
(R and
X)
ANFIS s
c
heme.
For
example,
the 1
st
ro
w i
n
Tabl
e 2
de
scribe
s
wh
en
the ge
ne
rat
o
r lo
sse
s
5
0
%
of its
excitation at
T
f
= 5
se
c while it wa
s lo
aded
by 80%
of it
s’ full lo
ad, the conv
entional
dista
n
ce
relay will detect the Loss of Excitation (LOE
) f
ault at “14.5 sec”,
while the
proposed (R and X )
ANFIS scheme will
detect this faul
t at
“6 sec” which
means that
the fault will
be
detected after its
ince
ption tim
e
by
“1
se
c” t
h
rou
g
h
the
calcul
ated i
n
d
e
x “I
R40
”
which is g
r
eate
r
t
han th
e th
re
shold
value “0.85
”
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 300 – 309
304
Figure 3. Flowchart for the
Loss of Excitation
(LOE
) Detection Procedure ba
sed
on (R a
nd X)
Figure 4. Flowchart for the
Loss of Exci
tation (LOE
) Detection Procedure ba
sed
on (V
trms
and I
a
)
Als
o
, the 4
th
row i
n
T
able
2
presents wh
en the
ge
ne
rator l
o
sse
s
7
5
% of its exci
tation at
T
f
= 5 sec while it was lo
aded by 70% of
its’ full load,
the co
nventional di
stance
relay will
detect
the Lo
ss of
Excitation
(LO
E
) fault at
“1
3.2 se
c”,
whil
e the
propo
sed
(R an
d X) ANFIS
sche
me
will dete
c
t it at “5.8 se
c” through the
cal
c
ulated ind
e
x “I
R40
”.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Loss of Excitation Faults
Dete
ction in
Hydro
-
G
ene
rators
Using a
n
Adaptive
…
(M.S. Abdel Aziz)
305
From th
e b
e
low T
able
s
(1) and
(2
)
calcu
l
ated indi
ce
s
(I
R40
) it is ea
sy to con
c
lu
de
that the
output of the prop
osed (R
and X) ANFI
S unit should
be fairly cho
s
en as:
a) I
R40
≥
0.85 for Loss of Exci
tation (LOE
) con
d
ition
s
an
d
b) I
R40
≤
0.2 for
no-fault condi
tions.
The propo
se
d (R a
nd X)
ANFIS relay
detect
s
the d
i
fferent Lo
ss
of Excitation (LOE
)
con
d
ition
s
wi
thin about
(3
00-1
400 m
s
e
c
) afte
r
the fault ince
ptio
n unde
r diffe
rent ge
nerator
loadin
g
co
ndi
tions from
(1
8.5% to 80%) whi
c
h is
m
o
re efficient th
an the co
nve
n
tional di
stan
ce
relay
.
5.2.
The Propos
e
d
(V
trms
and I
a
) Protec
tion
Scheme
On this
sche
me, the input
s to the ANFI
S unit are co
nsid
ere
d
to b
e
the gene
rat
o
r RMS
Line to Line v
o
ltage an
d Phase cu
rrent (V
trms
and I
a
).
Table 3 illustrates the test
i
ng data of the proposed (V
trms
and I
a
)
ANFIS s
c
heme, while
Table 4 d
epi
cts the valid
ation data of
the pr
op
ose
d
scheme. T
he validation
data are n
o
t
inclu
ded
in t
he trai
ning
p
r
ocess
and
are
ch
os
en
at different
g
enerator loa
d
ing a
nd
Loss of
Excitation (L
OE) co
ndition
s.
Table
3 an
d
4 sh
ow ho
w
much
the p
r
o
posed
(V
trms
a
nd I
a
) ANFIS scheme
is
accurate in
detectin
g
the
gen
erato
r
L
o
ss of Ex
citation (L
OE)
d
u
ring
ge
nerat
or h
eavy loa
d
ing
co
nditio
n
s
(more than 5
0
% of its rat
ed power) com
pare
d
to the (R and X) ANFIS sche
m
e.
For exampl
e, the 5
th
row in Table 3 sho
w
s whe
n
the generator losse
s
25
% of its
excitation at T
f
= 5 sec
while it was lo
aded by 50%
of its’ full lo
ad, the prop
o
s
ed (V
trms
an
d I
a
)
ANFIS will de
tect the Lo
ss
of Excitation (LOE) faul
t at “9.7 sec”, whi
c
h is m
o
re de
layed than th
e
prop
osed (R and
X
)
ANFIS
schem
e wh
ich will
d
e
tect
the fault
at “7.5se
c” (as p
r
esented
in
T
able
(1)), on the ot
her h
and, the
conventio
nal
distan
ce
rela
y will not det
ect the fault o
c
curren
ce. T
he
same
is fo
r t
he 13
th
ro
w i
n
Tabl
e 4, where
the
(V
trms
and I
a
) A
N
FIS will
det
ect the
Loss of
Exc
i
tation (LOE) fault at “6.5 s
e
c”, which is
mo
re th
a
n
pro
p
o
s
ed
(R an
d X) A
N
FIS sch
eme
whi
c
h
will dete
c
t it at “5.3 sec” (as de
scri
bed
in Tabl
e 2
)
, while the
co
n
v
entional rela
y will not detect
the fault occu
rre
nce.
On the
othe
r
hand,
whe
n
t
he g
ene
rator
is h
eavy load
ed, a
s
illu
stra
ted in th
e 1
st
, 4
th
an
d
16
th
rows on
Table (4
), it is cle
a
r that the (V
trms
and I
a
) ANFIS will detect the Loss of Excitation
(LOE
) faults faster th
an th
e other p
r
op
o
s
ed
(R a
nd X) ANFIS sche
me. For exa
m
ple, the 1
st
row
in Tabl
e 4
de
scribe
s
wh
en
the g
ene
rato
r lo
sses 50%
of its
excitati
on at T
f
= 5
sec
whil
e it
was
loade
d by 80% of its’
full load, the prop
osed (V
tr
ms
and I
a
) ANFIS will det
ect the Loss of
Excitation (L
OE) fault at
“5.6 sec”, whi
c
h i
s
fa
ster th
an the
pro
p
o
s
ed
(R an
d X
)
ANFIS
sche
me
whi
c
h
will
det
ect the
fault
at “6
sec”, o
n
the oth
e
r ha
nd, the t
r
aditi
onal
dista
n
ce
rel
a
y will
det
ect
the Loss of Excitation (LO
E
) fault at “14
.
5 sec”.
From th
e bel
ow T
able
3 a
nd 4
cal
c
ulat
ed indi
ce
s
(I
R40
) it is ea
sy
to con
c
lu
de t
hat the
output of the prop
osed (V
tr
ms
and I
a
) ANFIS unit shoul
d be ch
osen
as:
a) I
R40
≥
0.85 for Loss of Exci
tation (LOE
) con
d
ition
s
an
d
b) I
R40
≤
0.25 for no-fault co
nd
itions.
The p
r
op
ose
d
(V
trms
and
I
a
) ANFIS sch
e
me d
e
tect
s
the Lo
ss of E
x
citation (LO
E
) faults
within (5
00
-9
00 mse
c
) after the fault inceptio
n
wh
en
the generato
r
loadin
g
is m
o
re than 5
0
%
o
f
its’ rate
d p
o
wer fa
ster than
the oth
e
r
pro
pos
ed
(R an
d
X) A
N
FIS scheme
whi
c
h
detect
s
the
L
O
E
faults within (300-140
0 msec) in
wide
r
l
oadin
g
ran
ge
from (1
8.5% to 80%).
Thus, it i
s
o
b
v
ious that th
e
gene
rato
r te
rminal imp
eda
nce
mea
s
u
r
e
m
ents
(R an
d
X) an
d
the gene
rator RMS Line to Line voltage and Pha
s
e current (V
trms
and I
a
) as inpu
ts for the ANFIS
units give
su
perio
r results more
accu
ra
te than
the
convention
a
l d
i
stan
ce relay
s
an
d othe
r u
s
ed
techni
que
s
such
a
s
[29, 3
0
], and
are
very cl
ose to
t
he exp
e
cted
i
ndices. T
h
e
s
e indi
ce
s a
r
e
the
output valu
e
s
of
the A
N
FIS unit. Th
e expe
cted
value fo
r L
o
s
s of Excitat
i
on
(LOE
) fa
ult
con
d
ition
s
is 1, and the expecte
d value for no-fa
ul
t condition
s is 0
.
In addition,
whe
n
usin
g the
threshold val
ues a
s
refe
re
nce, the obtai
ned re
sult
s wi
ll lead to zero
erro
rs.
The tabulate
d
results in T
able (5
) illu
strate
that the Re
spo
n
se Time for the propo
se
d
Loss
of Excitation (LOE)
ANFIS rel
a
y based o
n
(V
trms
and I
a
) m
e
asu
r
em
ents is le
ss than
th
at of
Loss of Excitation (LOE
) ANFIS relay ba
sed o
n
(R
a
n
d
X) due to the ready me
asurem
ents in t
he
1
st
schem
e, while th
e 2
nd
scheme
nee
d
s
calculation
s
for (R an
d X
)
value
s
. Thi
s
mean
s that t
h
e
differen
c
e in the Re
sp
on
se
Time is due
to the requi
re
d time in calculation
s
of (R and X) value
s
.
Ho
wev
e
r, the
2
nd
schem
e covers
wide
r g
enerator lo
adi
ng co
ndition
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 300 – 309
306
Table 1. Te
sting Data Fo
r
The Prop
ose
d
(R an
d X) ANFIS Schem
e
%
G
e
nera
tor
Load
i
ng
%
LOE
LOE Faul
t
Incep
tio
n
Time
(sec)
Con
v
.
Distan
ce
Rela
y
tri
p
time
(sec)
Testi
ng
time
(sec)
R
v
a
l
u
e
X
v
a
lu
e
C
a
l
c
u
l
at
e
d
Index
“I
R40
”
Expecte
d
Index
%
Error
80%
75%
5
11.8
6
21.7024849
46
-.1887288
5161
0.874
1
12.6
80%
75%
5
11.8
8 14.4319610
29
-6.603402
62718
1.001
1
0.1
80%
75%
5
11.8
7 17.9859285
18
-4.819896
85374
1.003
1
0.3
80%
75%
5
11.8
6.5 19.9597008
91
-3.258635
35618
1.009
1
0.9
50%
25%
5
-
7.5 33.2204024
76
-10.38890
87234
0.865
1
13.5
50%
25%
5
-
8 31.7480263
35
-11.65382
45134
0.987
1
1.3
50%
25%
5
-
2 38.9802546
85
-1.900262
34583
0.166
0
16.6
50%
25%
5
-
3.5 39.0941175
15
-1.662984
60533
0.103
0
10.3
70%
25%
5
35
6.1 25.0072123
36
-2.155995
65675
0.853
1
14.7
70%
25%
5
35
7 23.3640844
26
-4.304839
97607
0.995
1
0.5
70%
25%
5
35
8.5 20.5596075
06
-6.882813
2688
1.002
1
0.2
70%
25%
5
35
3 26.0727466
56
-0.280338
13466
0.105
0
10.5
70%
25%
5
35
4 26.0714547
80
-0.269513
94394
0.1
0
10
Table 2. Valid
ation Data Fo
r The Pro
p
o
s
ed (R a
nd X)
ANFIS Sche
me
%
G
e
nera
tor
Load
i
ng
%
LOE
LOE Faul
t
Incep
tio
n
Time
(sec)
Con
v
.
Distan
ce
Rela
y
trip ti
me
(sec)
Testi
ng
time
(sec)
R
v
a
l
u
e
X
v
a
lu
e
C
a
l
c
u
l
at
e
d
Index
“I
R40
”
Expecte
d
Index
%
Error
80%
50%
5
14.5
6 21.7597036
56
-1.129444
27988
0.856
1
14.4
80%
50%
5
14.5
7.5 17.8027379
65
-5.163579
29149
1.002
1
0.2
80%
50%
5
14.5
4 22.9766573
26
1.06010676
291
0.143
0
14.3
70%
75%
5
13.2
5.8 24.9687738
74
-2.245064
21121
0.869
1
13.1
70%
75%
5
13.2
6 24.1873534
57
-3.163880
5022
0.97
1
3
70%
75%
5
13.2
8 16.0625793
30
-8.486558
70187
1.002
1
0.2
70%
75%
5
13.2
2.5 26.0739948
72
-0.292568
00175
0.11
0
11
70%
75%
5
13.2
3.5 26.0725376
08
-0.273972
05787
0.1
0
10
35%
75%
5
36
5.7 43.9462623
48
-16.85973
56813
0.882
1
11.8
35%
75%
5
36
6 40.4882549
79
-18.72519
25448
1.059
1
5.9
35%
75%
5
36
9 16.9469088
28
-20.80374
0385
0.996
1
4
35%
75%
5
36
4.5 47.1698782
56
-13.99305
97882
0.08
0
8
25%
80%
5
-
5.3 58.2303000
18
-28.21639
89537
0.934
1
6.6
25%
80%
5
-
6 46.2231542
1
-31.99589
15062
1.038
1
3.8
25%
80%
5
-
8.5 16.3102287
65
-26.80671
00233
0.993
1
7
25%
80%
5
-
4 59.2503547
43
-27.33679
62253
0.094
0
9.4
80%
20%
5
24
6.4 21.7871991
61
-1.217456
75159
0.865
1
13.5
80%
20%
5
24
7.5 20.0241848
86
-3.526453
76589
1.006
1
0.6
80%
20%
5
24
15 11.9442066
38
-7.434313
95269
0.9987
1
0.13
80%
20%
5
24
1.5 23.0485449
02
1.10041433
155
0.118
0
11.8
80%
20%
5
24
3 23.0552596
68
1.24793741
479
0.103
0
10.3
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Loss of Excitation Faults
Dete
ction in
Hydro
-
G
ene
rators
Using a
n
Adaptive
…
(M.S. Abdel Aziz)
307
Table 3. Te
sting Data Fo
r
The Prop
ose
d
(VtrmsandI
a) ANFIS Sch
e
me
%
G
e
nera
tor
Load
i
ng
%
LOE
LOE
Incep
tio
n
Time (sec)
Con
v
.
Distan
ce
Rela
y
tri
p
time (sec)
Testi
ng
time
(sec)
V
tr
ms
(kV)
I
a
(k
A
)
Calcula
t
ed
Index
“I
R40
”
Expecte
d
Index
%
Error
80%
75%
5
11.8
6 22.1039403
349
6.15473895
178
0.913
1
8.7
80%
75%
5
11.8
8 19.9499867
446
7.59776976
215
0.9917
1
0.83
80%
75%
5
11.8
7 20.9389505
073
6.79874267
771
0.962
1
3.8
80%
75%
5
11.8
6.5 21.5454928
37
6.39715738
451
0.936
1
6.4
50%
25%
5
-
9.7 22.8156947
143
4.26372233
932
0.86
1
14
50%
25%
5
-
2 24.1853845
047
3.55665724
573
0.22
0
22
50%
25%
5
-
3.5 24.1737469
291
3.55510879
725
0.23
0
23
70%
25%
5
35
6.1 22.5860386
167
5.26761383
15
0.913
1
8.7
70%
25%
5
35
7 22.1380556
938
5.47053009
617
0.929
1
7.1
70%
25%
5
35
8.5 21.5718305
263
5.82951955
847
0.952
1
4.8
70%
25%
5
35
3 23.1285410
774
5.10601022
447
0.13
0
13
70%
25%
5
35
4 23.1287487
203
5.10353080
899
0.129
0
12.9
Table 4. Valid
ation Data Fo
r The Pro
p
o
s
ed (Vtrm
s
an
d
I
a) ANFIS Scheme
%
G
e
nera
tor
Load
i
ng
%
LOE
LOE
Incep
tio
n
Time (sec)
Con
v
.
Distan
ce
Rela
y
tri
p
time (sec)
Testi
ng
time
(sec)
V
tr
ms
(kV)
I
a
(k
A
)
Calcula
t
ed
Index
“I
R40
”
Expecte
d
Index
%
Error
80%
50%
5
14.5
5.6 22.7325458
622
5.87024619
341
0.89
1
11
80%
50%
5
14.5
7.5 20.9698505
845
6.73924137
012
0.963
1
3.7
80%
50%
5
14.5
4 23.1301013
869
5.78595531
658
0.077
0
7.7
70%
75%
5
13.2
5.5 22.7699244
38
5.17262079
063
0.877
1
12.3
70%
75%
5
13.2
6 22.2419457
099
5.41025280
814
0.925
1
7.5
70%
75%
5
13.2
8 20.3443215
281
6.73790700
902
1.019
1
1.9
70%
75%
5
13.2
2.5 23.1311322
847
5.10582533
742
0.123
0
12.3
70%
75%
5
13.2
3.5 23.1284259
625
5.10345127
336
0.131
0
13.1
35%
75%
5
36
5.7 22.7114108
547
2.94652624
889
0.85
1
15
35%
75%
5
36
6 22.5310482
085
3.06901733
206
0.902
1
9.8
35%
75%
5
36
9 21.0039384
567
4.66437003
799
1.019
1
1.9
35%
75%
5
36
4.5 23.1500185
586
2.70620408
002
0.2
0
20
25%
80%
5
-
6.5 22.0860724
544
2.93699549
769
0.863
1
13.7
25%
80%
5
-
8.5 21.1905383
127
4.07455289
665
1.019
1
1.9
25%
80%
5
-
4 23.1659150
76
2.04201972
852
0.157
0
15.7
80%
20%
5
24
5.9 22.7645144
459
5.91737574
141
0.86
1
14
80%
20%
5
24
7.5 21.8267515
929
6.30115523
521
0.923
1
7.7
80%
20%
5
24
15 19.6016838
869
8.11584341
685
0.988
1
1.2
80%
20%
5
24
1.5 23.2802942
717
5.79235491
353
0.076
0
7.6
80%
20%
5
24
3 23.2447967
124
5.79561559
451
0.076
0
7.6
Table 5. Co
m
pari
s
on b
e
tween Different
Loss of Excitation (LOE
) T
e
ch
niqu
es
Tech
niq
u
e
Genera
tor Loa
d
i
ng
%
Respo
n
se Time (sec)
LOE ANF
I
S rela
y based on (V
trm
s
a
nd I
a
)
Higher than 5
0
%
(500-9
00 msec)
LOE ANF
I
S rela
y based on (R
and
X)
From 18.
5% to
8
0
%
(300-1
400 msec)
Conventional distance rela
y
From
18.
5% to
8
0
%
Minimum 7-8 sec.
Other
technique
based on “r
eactive po
w
e
r
measurements”[
29]
10% and
50%
Within 1120 msec
Other technique based
on
“R
-X w
i
th
directional element scheme” [30]
40% and
80%
6.931 and 4.
175
sec.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 25
02-4
752
IJEECS
Vol.
1, No. 2, February 201
6 : 300 – 309
308
6. Conclusion
This arti
cle
pre
s
ent
s a novel study of Hy
dro-gen
erato
r
s Lo
ss of Excitatio
n
(LOE)
scheme
usi
n
g Adaptive Neuro
Fu
zzy
In
feren
c
e Syste
m
(ANFIS). T
he ne
ce
ssity for thi
s
re
se
arch
work ap
pea
re
d du
e to
the i
naccu
rate
re
spon
se
of
the
distan
ce
Lo
ss of
Excitatio
n
(LOE) relays
whi
c
h i
s
com
p
letely de
pen
ding
on th
e g
enerator pe
rcentage
loa
d
in
g an
d the
lo
ss of
excitatio
n
percenta
ge.
The Pro
p
o
s
e
d
Artificial Intelligent
App
r
oach dem
on
strates
su
ccessful pe
rform
a
nce
for Lo
ss of E
x
citation (LO
E
) faults dete
c
tion. T
w
ova
r
ious algo
rith
ms a
r
e
u
s
ed
i
n
this arti
cle,
they
are ba
se
d on
the type of th
e inputs to th
e prop
os
ed A
N
FIS unit for Loss of Excitation (LOE
) fault
detectio
n
, the
gene
rato
r te
rminal im
ped
ance mea
s
u
r
ements (R a
nd X) a
nd th
e gen
erato
r
RMS
Line to Li
ne
voltage an
d
Phase
curren
t (V
trms
and I
a
) mea
s
ureme
n
ts. The
obta
i
ned results f
r
om
both
al
gorith
m
s are
comp
ared
with
e
a
c
h othe
r
a
n
d
co
mpa
r
ed
with the
conve
n
tional
dista
n
c
e
relay
re
spo
n
s
e i
n
a
dditio
n
to oth
e
r tech
niqu
es. It
wa
s fo
und
that the g
e
n
e
rato
r te
rmin
al
impeda
nce m
easure
m
ent
s (R an
d X) a
nd the ge
nerator RMS Li
n
e
to Line voltage an
d Pha
s
e
cur
r
e
n
t
(V
trms
and I
a
) play t
he e
s
sential
rule
in th
e L
o
s
s of Excitati
on
(LOE
) d
e
tection
p
r
ocess.
For fault dete
c
tion task, all the validation data
for the ANFIS sch
emes in the fault and no fault
con
d
ition
s
give the expe
cte
d
output
s. Th
e used
data f
o
r testin
g an
d
validation a
r
e of both ki
nd
s
of data: used
in training a
n
d
not use
d
in
trai
ning resp
ectively. Suggeste
d indi
ce
s for o
c
curre
n
ce
of the Loss of Excitation (LOE) condi
tions we
re in
trodu
ced. Th
e obtained
result
s are v
e
ry
brilliant.
References
[1]
Weijia
n Wan
g
. Princip
l
e
and
Applic
atio
n of
Electric Po
w
e
r Equ
i
pme
n
t Protection.
C
h
ina El
ectric
.
Power Press
. 2002.
[2]
Z
hanp
en
g Shi.
Investigati
on
on Gen
e
rator
Loss of
E
x
cita
tion Protecti
on
in Gener
ator
Protectio
n
Coor
din
a
tio
n
. Master T
hesis. Stockholm, Sw
e
d
en: Sc
ho
ol
of Electrical Engi
neer
in
g Ro
yal Institute o
f
T
e
chnolog
y; 2
010.
[3]
Mehrdad G
h
andhari. Dy
namic An
aly
s
is of
Po
w
e
r Sy
stems PART
II.
Royal Institute
of
T
e
chn
o
l
ogy
.
200
8.
[4]
Don
a
lt Reim
en
t. Protective Rela
yi
ng for Po
w
e
r
Gener
atio
n S
y
stems. Bo
ca
Raton: CR
C
Press. 2006.
[5]
P Kundur. Po
wer S
y
stem Sta
b
ilit
y a
nd Co
ntrol. McGra
w
-
Hil
l
Inc.
[6]
Gabrie
l Be
nm
ou
ya
l. T
he Impact of S
y
n
c
h
r
ono
us
Generators Ex
citati
o
n
Su
ppl
y o
n
Protection
a
n
d
R
e
l
a
y
s
. Sche
w
e
i
t
ze
r En
gi
ne
e
r
i
n
g
La
bo
ra
to
ries In
c.
[7]
T
G
Paithankar
, SR Bhide. F
und
ament
als o
f
Po
w
e
r
S
y
ste
m
Protection. Prentic
e-
Hal
l
o
f
India Priva
t
e
Limite
d. 200
3.
[8]
IEEE Std. . C37.102™.
IEEE Guide for AC
Generator Prot
ection
. 20
06.
[9]
JL Bl
ackbur
n,
T
J
Domin. Pro
t
ective R
e
la
yi
n
g
:
Princ
i
pl
es a
nd
A
ppl
icati
o
n
s
.
3rd editi
on. CRC
Pr
ess.
200
7.
[10]
W
Elmore. Protective Re
la
yi
ng
T
heor
y
and A
pplic
atio
ns. Se
cond e
d
iti
on. C
RC Press. 200
4.
[11]
D Reimert. Pro
t
ective Rel
a
yin
g
for
Po
w
e
r Ge
nerati
on S
y
ste
m
s. CRC Press. 2006.
[12]
CJ Mozin
a
, M Reich
a
rd, Z Bukha
l
a.
Co
ordi
natio
n of Gene
rator Protecti
o
n
w
i
th Generat
or Excitati
o
n
Contro
l a
n
d
Gener
ator C
a
p
a
b
ility.
Pr
ocee
di
ngs
of the
IEEE Po
w
e
r En
gin
eeri
n
g
Soci
et
y
G
ener
a
l
Meetin
g. T
a
mp
a, F
L
. 2007: 1-
17.
[13]
S Patel, K Stepha
n, M Bajpa
i
, R Das,
T
J
Domin,
E F
e
n
nell, JD Gard
e
ll, I Gibbs, C
Henvi
l
l
e
, PM
Kerrigan, HJ King, P Kumar, CJ Mozina, M Reicha
rd, J Uc
hi
yam
a
, S Usman, D Viers, D Wardlo
w
,
M
Yall
a. Performance of G
ener
ator Protection
durin
g
Major S
y
stem
Distur
b
ances.
IEEE Tr
ansactions
on
Pow
e
r Deliv
ery
. 2004; 19(
4): 1650-
166
2.
[1
4
]
Eb
ra
hi
mi
, Seye
d
Ya
se
r, Amir Gh
o
r
b
a
n
i
.
Pe
rfo
rma
n
c
e
co
mp
a
r
i
s
o
n
o
f
L
O
E p
r
o
t
e
c
ti
o
n
o
f
sy
n
c
h
r
o
nous
gen
erator in
th
e
pr
esenc
e of UPF
C
. Eng
i
ne
erin
g Sci
enc
e
and
T
e
chn
o
l
o
g
y
, an Inter
nati
ona
l Jo
urn
a
l.
201
5.
[15]
Berd
y
J. Loss
of excitatio
n
protec
tio
n
for moder
n s
y
nchr
ono
us gen
erat
ors.
IEEE Trans. On PAS.
197
5; 94(5): 14
57-1
463.
[16]
De M
o
rais
AP, Car
doso
G, M
a
riotto
L. An
i
n
nov
ativ
e l
o
ss-o
f
-excitati
o
n
pr
o
t
ection
bas
ed
on th
e fuzz
y
infere
nce mec
han
ism.
Pow
e
r Delivery, IEEE
T
r
ansactions
on.
201
0; 25(4)
: 2197-2
2
0
4
.
[17]
Sharaf AM, Li
e T
T
.
ANN based patter
n
cla
ssificati
o
n
of sync
hro
nous
ge
nerat
or sta
b
il
ity a
nd l
o
ss o
f
ex
citation.
Energy Conv
ersion, IEEE
Transac
tions on
. 1
994;
9(4): 753-7
59.
[18]
T
a
mba
y
SR, P
a
itha
nkar YG.
A new
ad
aptiv
e loss
of excita
tion re
lay
aug
me
nted
by rate
of cha
nge
o
f
reactanc
e
. Po
w
e
r Eng
i
ne
eri
n
g Societ
y
Gen
e
ral Me
eting, IEEE. 2005; 12:
1831-
18
35.
[19]
Liu YD, W
ang
Z
P
,
Z
heng T
,
T
u
LM, Su Y, W
u
Z
Q
.
A novel ad
aptiv
e lo
ss of excitatio
n
protectio
n
criterion
bas
e
d
on
stea
dy-state stabi
lity
li
mit
. In Po
w
e
r a
nd E
n
e
r
g
y
Eng
i
n
eeri
ng C
onfer
enc
e
(APPEEC), IEEE PES Asia-Pacific. 2013: 1-5.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJEECS
ISSN:
2502-4
752
Loss of Excitation Faults
Dete
ction in
Hydro
-
G
ene
rators
Using a
n
Adaptive
…
(M.S. Abdel Aziz)
309
[20]
Amini M, Dav
a
rpan
ah M, San
a
y
e-
Pas
a
n
d
M. A Novel A
ppr
o
a
ch to
D
e
tect the S
y
nc
hron
ou
s Generato
r
Loss of Exc
i
tati
on.
IEEE Transactions On Power Delivery.
2
015; 30(
3): 142
9.
[21]
Elsamahy
M, Fari
ed SO, Ramakrishna G.
Impact of
mi
dp
oi
nt ST
AT
COM on th
e co
ordi
n
a
tion
betw
e
e
n
gen
erator
dista
n
ce p
has
e b
a
c
k
up pr
ot
ectio
n
and
g
ener
ator
capa
bi
lity cur
v
es
. In Po
w
e
r
and
Ener
g
y
Societ
y
Gener
al Meeti
ng, IEEE. 2010: 1-7.
[22]
PSCAD/EMT
D
C
User’s Man
u
a
l. Manito
ba H
V
DC Res
earch
Centre. 200
3.
[23]
MS Abde
l Aziz
, MA Hassan,
EA Z
ahab.
A
p
plicati
ons
of ANF
I
S in Hig
h I
m
p
e
d
ance F
a
u
l
ts Detectio
n
and Cl
assific
a
tion in D
i
stributi
on Netw
orks.
Presente
d
at
T
he 8th
IEEE
Internati
o
n
a
l S
y
mposi
u
m
o
n
Diag
nostics
for
Electric
al M
a
c
h
in
es, Po
w
e
r
Electron
ics a
n
d
Driv
es, (SDE
MPED 2
011).
Bolo
gna,
Ital
y
.
201
1: 612-
619.
[24]
MS Abd
e
l
Aziz
, MA Hass
an,
EA Z
aha
b. A
n
Artifi
cial
Intell
ig
ence
Bas
e
d
A
ppro
a
ch
for H
i
gh Imp
e
d
anc
e
F
aults Anal
ys
i
s
in Distributi
on Net
w
o
r
ks.
Internatio
na
l Journ
a
l of Syst
em Dy
na
mics
Applic
atio
n
s
IJSDA
. 2012: 4
4
-59.
[25]
MS Abde
l Azi
z
, MA Hassa
n
,
EA. Z
ahab.
High-
im
p
eda
nc
e F
aults A
nal
ysis in
Distrib
uti
on N
e
t
w
ork
s
Using
an Ad
ap
tive Neur
o F
u
zz
y
Infere
nce S
y
stem.
Electric
Pow
e
r Compo
nents an
d Systems.
2
012
;
40:13
00-
131
8.
[26]
MS Abd
e
l
Aziz
, MA Hass
an,
EA Z
aha
b.
A
n
Artificial
Intell
ig
ence
Bas
e
d
A
ppro
a
ch
for H
i
gh I
m
p
e
d
anc
e
F
aults Ana
l
ysi
s in Distri
buti
o
n Netw
orks un
der Differ
ent L
oad
ing
Co
nditi
ons.
T
he 21st Internatio
na
l
confere
n
ce o
n
Comp
uter T
heor
y
a
nd Ap
pl
ications. Ale
x
an
dria, Eg
ypt. 20
11.
[27]
Kamel T
S
, MM
A Hassan, A El-Morshedy
. An Anfi
s Based D
i
stance R
e
la
y
Protection for
T
r
ansmission
Lines in EPS.
Internati
o
n
a
l Jo
urna
l of Innova
t
ions in El
ectric
al Pow
e
r systems
. 20
11.
[28]
Kamel T
S
, MM A Hassan, A
El–Mors
hed
y.
Advanc
ed d
i
stance
protectio
n
techn
i
qu
e b
a
s
ed o
n
multi
p
l
e
classifie
d
A
N
F
I
S consid
eri
ng
differe
nt l
oad
ing
co
nditi
ons for
lo
ng
transmissi
on
li
nes
in
EP
S.
Internatio
na
l Journ
a
l of Mod
e
llin
g, Identificat
ion a
nd C
ontrol
. Inderscienc
e. 201
2; 16(2): 10
8-12
1.
[29]
Omer Usta, MH Musa, M Ba
yrak, MA R
edf
ern. A Ne
w
R
e
lay
i
ng
Al
go
ri
thm to
D
e
tect Loss of Ex
citation
of S
y
nchro
n
o
u
s
Generat
ors
. T
u
rkish Jo
urna
l of Electric
al
Engi
neer
in
g a
nd C
o
mput
er
Scienc
e
. 20
07;
15(3): 33
9-3
4
9
.
[30]
Shi ZP, Wang
JP, Gajic Z, Sao
C, Gha
n
d
h
a
ri M. T
he c
o
mpariso
n
an
d
ana
l
y
sis for
l
o
s
s
of e
x
c
i
tatio
n
protectio
n
sche
m
es in ge
nerat
or protectio
n
. 2
012.
Evaluation Warning : The document was created with Spire.PDF for Python.