Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 2
,
A
p
r
il
201
6, p
p
.
87
0
~
87
6
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
2.9
572
8
70
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJECE
New Cl
assifier Design for Stat
i
c
S
ecu
rity Evalu
a
ti
on using
Artificial Intelligence Techniques
Ibrahim
Saeh, M.W.
Mustaf
a, Nasir
A. Al-geelani
Facult
y of Ele
c
tr
ica
l
Eng
i
ne
ering
,
Univer
si
ti
Tekn
ologi Ma
la
ysia
(
U
TM), Mal
a
y
s
ia
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Aug 12, 2015
Rev
i
sed
No
v
25
, 20
15
Accepted Dec 16, 2015
This
paper prop
os
es
evaluat
i
on and clas
s
i
fi
cat
io
n clas
s
i
fier for s
t
ati
c
s
ecuri
t
y
evalu
a
tion (SSE) and classifica-tion. Da
ta are generated on (30,
57, 118 and
300) bus IEEE test s
y
stem
s used to de
sign the classifiers. Th
e
im
plem
entat
i
on decision tr
ee m
e
thods on
several
IEEE test s
y
st
e
m
s involved
appropriateness
SSE and cl
assi-fication b
y
usin
g four algorith
ms of
DT’s.
Em
pirica
ll
y, wi
t
h
the pres
ent of
F
S
A
, the im
plem
entation r
e
s
u
lt
s
indicat
e tha
t
these c
l
assifi
ers have
the
cap
a
b
ilit
y
for s
y
s
t
e
m
securit
y
ev
al
uation
and
clas
s
i
fi
cat
ion.
L
a
s
t
l
y
, F
S
A
is
effici
ent and
effe
c
tive
approach fo
r real-
tim
e
evalu
a
tion
and
c
l
as
s
i
fic
a
tion
cl
as
s
i
fier d
e
s
i
gn.
Keyword:
Artificial In
tellig
en
ce
Classification
Classifier desi
gn
Feature
Selection
Static Secu
rity
Ev
alu
a
tion
Copyright ©
201
6 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
Ib
rahim
Saeh,
Depa
rtm
e
nt of
Electrical an
d Electronic
Engineering,
Facu
lty of Electrical Eng
i
n
eerin
g
,
Un
iv
ersiti Tekn
o
l
o
g
i
Mal
a
ysia,
Jo
hor
Bah
r
u
81
310
, Joho
r
,
Malaysia.
Em
a
il: ib
rah
i
msaeh
@yah
oo
.co
m
1.
INTRODUCTION
Th
e electric mark
et co
m
p
etitio
n
fo
rces gen
e
ratin
g
en
tities an
d
system o
p
e
rators to
op
erate the
syste
m
with
in
th
eir secu
rity lev
e
l. Lo
ad
and
o
p
e
rating
c
ons
t
r
ai
ns ar
e t
w
o s
e
t
s
of
p
o
w
e
r sy
st
em
operat
i
o
n
co
n-
strain
s
[1
].
The lo
ad con
s
trai
n
t
is an equ
a
ti
o
n
con
s
trai
n
t
wh
ich
sets t
h
e to
tal g
e
n
e
rati
o
n
equ
a
l to to
t
a
l lo
ad
p
l
u
s
t
o
tal po
wer
lo
sses. Th
e
o
p
e
r
a
ting
con
s
tr
ain
t
s ar
e
up
per
and
/
or
low
e
r
li
m
its o
f
syste
m
's v
a
r
i
ab
les. Long
t
e
rm
pl
anni
n
g
or e
v
en i
n
op
e
r
at
i
onal
,
m
a
ki
ng sec
u
ri
t
y
de
ci
si
ons an
d a
conce
p
t
u
al
bas
i
s pro
v
i
d
e
by
sy
st
e
m
ope
rat
i
n
g st
at
e
s
.
Und
e
r con
tingen
t
con
d
ition
,
secu
rity can
b
e
d
e
fi
n
e
d
as th
e
ab
ility o
f
th
e
po
wer system
to
rem
a
in
in
a
se-cu
r
e st
at
e [
2
]
.
Secu
ri
t
y
assessm
ent
i
nvol
ves est
i
m
at
i
on of t
h
e rel
a
t
i
v
e
securi
t
y
l
e
vel
of t
h
e cu
rre
nt
ope
rat
-
in
g
co
nd
ition
of th
e system
u
s
in
g
av
ailab
l
e data
m
easu
r
em
e
n
ts. Th
e task
of security assessm
en
t is p
e
r-form
e
d
in
th
ree m
o
d
e
s - static, tran
sien
t an
d
d
y
n
a
mic [3
]. Mo
re
specifically the static security
is the stea
dy state
sy
st
em
beha
vi
ou
r
u
nde
r a
s
p
eci
fi
ed c
ont
i
n
g
e
ncy
,
wh
ereas
t
h
e t
r
a
n
si
e
n
t
securi
t
y
i
s
dea
l
i
ng
wi
t
h
e
v
al
uat
i
n
g
ro
t
o
r ang
l
e
o
s
cillatio
n
s
un
d
e
r a tran
sien
t
d
i
stu
r
b
a
n
c
e.
Dyna
m
i
c secu
rity d
eals
with
th
e
lo
ng
term
b
e
hav
i
ou
r
from
the instant of the system
transi
ently sec
u
re
to t
h
e insta
n
t of t
h
e
syste
m
will reaches
steady state.
Al
l
t
h
e t
h
ree m
odes need t
o
be seque
nt
i
a
l
l
y
perf
orm
e
d on-l
i
n
e
.
In case
of i
n
sec
u
ri
t
y
i
n
any
m
ode of
as-sessm
ent, an alarm
is signalled for the
operat
or t
o
tak
e
an
ap
pr
opr
iate r
e
m
e
d
i
al acti
o
n. Th
ro
ugh
si
m
u
la-
tio
n
,
Static Secu
rity Ev
alu
a
tion
(SSE)
as
si
st
s ope
rat
o
rs t
o
d
e
t
ect
fol
l
o
wi
ng
a gi
ve
n l
i
s
t
of
cont
i
n
ge
nci
e
s suc
h
as a vol
t
a
g
e
out
-o
f-l
i
m
it
or pot
e
n
t
i
a
l
a sy
st
em
bran
ch ove
rloa
de
d.
Due t
o
the la
rge system
size and
dere
gulated power system
,
a
steady-stat
e se
curity analysis
bec
o
m
e
s an im
po
ssible task due to the a
s
s
o
ciated
com
put
at
i
on b
u
r
d
e
n
.
In
SSE
, t
h
e c
ont
i
n
ge
nci
e
s s
e
veri
t
y
i
s
ju
d
g
ed
on
scal
e
per
f
o
r
m
a
nce i
nde
x
(P
I)
b
a
si
s. I
n
[
1
-
3
]
,
num
erous P
I
based m
e
thods
have
bee
n
re
ported.
Artific
ial intelligence (AI) ca
n be
divi
ded into t
w
o ty
pes
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 2, A
p
ri
l
20
16
:
87
0 – 8
7
6
87
1
t
ech-
n
i
q
ues, cl
ust
e
ri
n
g
t
e
c
hni
que
s a
n
d
cl
ass
i
fi
cat
i
on t
ech
ni
que
s, a
n
d i
t
s
p
o
we
rf
ul
of
re
d
u
ci
n
g
t
h
e
dat
a
com
-
p
l
ex
ity, m
a
d
e
it to
use in v
a
ri
o
u
s
area
s like
medical and enginee
r
ing [4, 5].
1.
1. St
ati
c
Sec
u
ri
ty
E
v
al
uati
on I
ndices
Selection
Po
wer sy
st
em
net
w
or
ks are
r
e
qui
red t
o
o
p
er
at
e wi
t
h
securi
t
y
l
i
m
i
t
s
. Securi
t
y
i
s
defi
ned as
pr
om
i
s
i
ng
th
e con
tin
uou
s o
p
e
ratio
n
o
f
a p
o
wer system cap
ab
ility u
n
d
e
r no
rm
al o
p
e
ration
ev
en
n
e
x
t
so
m
e
i
m
p
o
rtan
t
cont
i
n
ge
ncy
[
6
]
.
In
t
h
e literatu
re, sev
e
ral k
e
ys h
a
v
e
b
e
en
sugg
ested
as standard
s
fo
r static
secu
rity classificatio
n
and
eval-uation
[2, 7-10] includ
e
lines overl
o
a
d
ed
or \ and
bus
voltage
s collapse which l
e
t the syste
m
deviate
fr
om
norm
a
l
operat
i
n
g st
at
e l
i
m
i
t
s
. H
o
we
ve
r,
vi
ol
at
i
o
ns a
r
e n
o
t
i
n
t
h
e sa
m
e
l
e
vel
of
t
h
e
sam
e
si
gni
fi
cance.
In
t
h
e assessmen
t pro
cess
o
f
static secu
rity, it is
ev
al
u
a
ted
for sev
e
ral
feasib
le con
ting
e
n
c
ies v
i
a
so
lv
i
n
g
p
o
wer flo
w
no
n
lin
ear equ
a
tion
s
. Th
ese con
ting
e
ncies p
o
ssi
b
l
y will co
n
t
ain
ou
tag
e
o
f
a
g
e
neratin
g
u
n
it
o
r
N-1 tran
sm
issio
n
lin
e
o
r
a tran
sform
e
r.
For
num
ero
u
s di
st
ur
ba
nces, t
h
e l
o
ad fl
ow i
s
sim
u
l
a
t
e
d and t
h
e securi
t
y
l
i
m
i
t
a
t
i
ons are g
a
uge
d. T
h
e
ope
r-at
i
n
g
st
at
e of
p
o
w
er
sy
st
em
i
s
cat
egori
zed as
st
at
i
c
secure
(S
S-B
i
na
r
y
1) i
f
t
w
o t
h
e
l
i
m
i
t
a
t
i
ons i
n
equa
-
tio
n
s
(1
), and
(2
) are fu
l
f
illed
.
In
case of on
e
li
mitatio
n
is id
en
tified
su
b
s
equ
e
n
t
a con
ting
e
n
c
y, th
e state of th
e
syste
m
is categ
o
r
ized
as
static in
secu
re
(SI-B
in
ary
0
)
.
There
f
ore, i
t
i
s
com
pul
sory
t
o
devel
op a
n
effi
ci
ent
m
e
t
hods t
o
deal
ab
out
t
h
e com
p
l
e
xi
t
y
of dat
a
[10]. T
h
e tradi
tional elem
ent accounts for
coachi
ng t
h
e device unde
rsta
ndi
ng m
e
thod
s
for classification
of
static secu
rity ev
alu
a
tion
con
t
en
ts.
2.
ARTIFICIAL
INTELLIGE
NCE TE
CHNIQUES
Gen
e
rally, m
o
st o
f
th
e artifi
c
ial in
tellig
en
ce tech
n
i
q
u
e
s ap
pro
ach
es assess in
fo
rm
atio
n th
ro
ugh
th
e
data-base.
Nowadays
, database becom
e
s la
rge
r
in size, an
d
as resu
lt, it i
s
v
e
ry d
i
fficu
l
t
to
in
terpret com
p
lex
d
a
ta. Th
erefore, it is co
m
p
u
l
so
ry to
d
e
v
e
lop
efficien
t m
e
t
h
od
s t
o
d
eal ab
ou
t th
e co
m
p
lex
ity o
f
d
a
ta
[10
]
.
Mu
lti-layer feed
forward ar
tificial n
e
u
r
al
n
e
t
w
ork (MLFFN) and
rad
i
al b
a
sis
fun
c
tion
n
e
twork
(RBFN)
are
pr
o
pose
d
t
o
i
m
pl
em
ent
t
h
e onl
i
n
e
m
odul
e
fo
r
po
wer
sy
st
em
st
at
i
c
securi
t
y
assessm
ent
[
11]
.
The
s
ecuri
t
y
cl
assi
fi
cat
i
on,
cont
i
n
ge
ncy
se
l
ect
i
on an
d r
a
nki
ng a
r
e
do
n
e
base
d o
n
t
h
e
com
posi
t
e
securi
t
y
i
nde
x
wh
i
c
h i
s
capable
of acc
urately differe
n
tiating
the se
cure a
n
d non-s
ecure case
s
. For eac
h c
ontingency case as
well a
s
for b
a
se case co
nd
itio
n, th
e co
m
p
o
s
ite secu
rity
in
d
e
x
is com
p
u
t
ed
u
s
ing
th
e fu
ll Newton Rap
h
s
o
n
l
o
ad flo
w
an
alysis. Th
e
p
r
op
o
s
ed
artificial
n
e
ural n
e
t
w
ork (ANN)
m
o
d
e
ls tak
e
lo
ad
i
n
g cond
itio
n and
th
e pro
b
a
b
l
e
cont
i
n
ge
nci
e
s
as t
h
e i
n
put
a
n
d asse
ss t
h
e sy
st
em
securi
t
y
b
y
scree
n
i
n
g t
h
e
cre
d
i
b
l
e
c
o
nt
i
nge
nci
e
s a
n
d
r
a
nki
n
g
th
em
in
th
e order
o
f
sev
e
rity based
on
co
m
p
o
s
ite secu
rity ind
e
x.
The tra
d
itional ele
m
ent accoun
ts for coac
hi
ng the
de
vice
understa
ndi
ng methods
for
classification of
static secu
rity
ev
alu
a
tion
conten
ts. Fig
u
re 1 p
r
esen
ts th
e meth
o
d
o
l
og
y fo
r static secu
ri
ty ev
alu
a
tio
n
co
n
t
en
t
classificatio
n
ap
pro
ach b
a
sed
u
pon
th
e artificial in
tellig
en
ce tech
n
i
q
u
e
s.
The m
e
t
hodol
o
g
y
i
s
at
t
a
i
n
ed t
h
r
o
ug
h f
o
u
r
p
h
a
ses:
dat
a
set
col
l
ect
i
on,
dat
a
set
prep
r
o
cessi
ng
, t
r
ai
ni
n
g
pha
se, a
nd cl
assi
fi
er e
v
al
ua
t
i
on wi
t
h
t
e
st
i
ng
dat
a
. C
o
n
s
eque
nt
l
y
, t
h
e
st
at
i
c
securi
t
y
eval
uat
i
o
n c
a
n be
m
a
naged
ba
sed
o
n
t
h
e
t
r
ai
ne
d
machine learning classifier.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
New Cla
ssifier Desi
g
n
fo
r
S
t
atic S
ecu
rity Eva
l
ua
tio
n using
Artificia
l In
-tel
lig
en
ce Techn
i
q
u
e
s
(
I
br
a
h
i
m
S.)
87
2
Fig
u
re
1
.
Artificial In
tellig
en
ce Techn
i
qu
es
pro
c
edu
r
e fo
r static
secu
rity
evalu
a
tio
n
and
cl
assificatio
n
2
.
1
.
Ra
w
Da
taset
C
o
llect
io
n
NRLF an
alysis is u
s
ed
b
e
fo
re i
m
p
l
e
m
en
tatio
n
of
decisi
on t
r
ee to s
o
lve al
gebraic equation
which is
n
on-lin
ear to
th
e syste
m
u
s
ed
, an
d
co
llected
d
a
ta o
f
all lin
e
flow an
d
v
o
ltag
e
s of all b
u
s
es. Th
ese d
a
ta co
llect-
ed
will u
s
e as i
n
pu
t
v
ector fo
r train
i
ng
an
d testin
g
the
algo
ri
th
m
s
. Th
u
s
, test d
a
taset; wh
ich
is
d
i
ssimilar cases
from
the training dataset shoul
d ke
e
p
gett
ing a
n
accepta
ble accuracy re
sults.
NRLF were devel
ope
d
via
m
a
t
powe
r
3.
0
b
4
p
r
o
g
r
a
m
[12
]
and use
d
t
h
ro
ug
h t
h
i
s
st
u
d
y as a
m
a
trix
fo
rm
. In
th
is p
r
ogram
, th
e resu
lts can
be
s
h
ow
n by
usi
n
g
t
h
e
c
o
m
m
a
nd ru
n
p
f ('
case Z'
), wher
e Z i
s
t
h
e bu
ses num
ber. T
h
e l
i
s
t
of at
t
r
ibut
e
s
(features
)
used for t
h
e
pattern vector
for stat
i
c
security ev
al
u
a
tio
n is as
fo
llo
ws b
e
l
o
w.
X
SSE
= {| V|i ,
θ
i , SGi , SLi ,
Sij
}
(1)
Th
e con
ting
e
ncies can
in
clud
e in
terrup
tion o
n
a tran
sfo
r
mer o
r
th
e tran
sm
issio
n
lin
e o
r
m
a
yb
e
a
g
e
n
e
ra-t
o
r
. Perform
i
n
g
lo
ad fl
o
w
will ch
eck
all th
e bu
s
vo
ltag
e
s and
lin
e t
h
erm
a
l p
o
wer li
m
i
ts; (1
) vo
ltag
e
at
all buses
m
u
st be
within t
h
eir range
(0.94-1.06) p.u. [13,
14
], and
(2) all
lines are
not e
x
ceedi
n
g their
powe
r
range as
well
(S< Sm
ax.).
2.
2.
T
r
ai
ni
n
g
Da
ta
set Prep
a
r
ati
o
n
To
b
e
ab
le to
p
u
t
tog
e
th
er wo
rk
ing
ou
t in
fo
rm
atio
n
arrang
ed, th
e sp
ecified
o
p
tion
s
th
at co
m
e
with
t
h
e act
ual
sy
st
em
t
e
nd t
o
be
obt
ai
ne
d
fr
om
t
h
e act
ual
read
y
t
r
ack
d
o
cum
e
nt
s. T
h
e
key
f
unct
i
o
ns
o
f
t
h
e
p
o
we
r
sy
st
em
net
w
or
k are e
x
t
r
act
ed
i
n
or
de
r t
o
pre
p
are t
h
e t
r
ai
ni
n
g
dat
a
set
.
T
h
e
s
e fu
nct
i
o
ns t
e
nd t
o
be t
r
ans
f
orm
e
d
in
to
th
e actu
a
l
in
pu
t/o
u
t
p
u
t
d
a
taset o
r
ev
en
co
ach
i
n
g d
e
signs n
e
ed
ed
in th
e co
ach
i
ng
stage.
Wh
en
th
e instru
ctio
n
d
a
taset is read
y wh
ile describ
e
d
prev
i
o
u
s
ly, th
e actu
a
l
d
a
taset will b
e
stab
ilized
ap-
p
r
o
pri
a
t
e
l
y
i
n
va
ri
et
y
[0
,
1
]
by
ap
pl
y
i
ng
e
quat
i
o
n
(
2
).
Param
e
ters tun
i
ng
Start
Dataset pre-processing
Tr
ai
n
i
ng
d
a
t
a
s
e
t
p
r
ep
ar
at
i
o
n
Dataset co
llectio
n
fro
m
Power s
y
st
em
net
w
or
k
Train
i
ng
of Art
i
ficial
In
tellig
en
ce
Tech
ni
q
u
es t
e
c
hni
que
(
C4
.5
)
Evalu
a
tion
with
testing
dat
a
set
?
Security Status
Secure -
i
nsec
ure
End
O
k
No
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 2, A
p
ri
l
20
16
:
87
0 – 8
7
6
87
3
mi
n
ma
x
m
i
n
V
VV
v
v
(2
)
wh
ere v
is
th
e attrib
u
t
e V o
r
i
g
in
al v
a
lu
e,
v
is th
e attribu
t
e V
n
o
rm
alized
v
a
lu
e, and
mi
n
V
and
max
V
a
r
e
t
h
e mi
n
i
mu
m a
n
d
ma
x
i
mu
m a
t
t
r
i
b
u
t
e
V
v
a
l
u
e
s
.
2.
3. C
4
.
5
Cl
as
si
fi
er
T
r
ai
ni
n
g
C
4
.
5
deci
si
on
t
r
ee i
s
o
n
e
of t
h
e m
o
st
br
oadl
y
used
an
d r
e
a
l
-w
orl
d
ap
pr
oa
ch
es. In
C4.5, the
learne
d
classifier is rep
r
esen
ted b
y
a DT as sets
o
f
if-th
e
n
ru
les to
h
u
m
an
read
ab
ility i
m
p
r
o
v
e
men
t
. Th
erefo
r
e, t
h
e
d
ecision
tree is si
m
p
le
to
b
e
un
d
e
rstood
and
in
terpreted. Besid
e
s, it can
h
a
n
d
l
e no
m
i
n
a
l a
n
d
categorical d
a
ta
an
d p
e
rfo
r
m
well with
larg
e data set in
a sh
ort ti
m
e
[1
5
]
.
In C4
.5
trai
n
i
ng
,
th
e d
ecisi
o
n
t
r
ee is bu
ilt in
a
to
p
-
down recursive
way.
Learning
w
o
r
k
s
of
C4.5 as f
o
llows:
Prim
aril
y, all t
r
ain
i
ng
patterns fix
e
d
at roo
t
. Th
es
e
patterns are di
vide
d base
d on feat
ures selected
b
a
sed
on
an
im
p
u
r
ity fu
n
c
tio
n
in
recursiv
e rou
tin
e. Di
v
i
din
g
con
tinu
e
s till a
ll train
i
n
g
p
a
ttern
s
for a certai
n
no
de
bel
o
n
g
t
o
t
h
e si
m
i
l
a
r cl
ass. T
h
e
param
e
t
e
rs an
d t
h
ei
r
set
t
i
ngs
val
u
es
were
use
d
i
n
WE
KA a
s
s
h
o
w
n
i
n
Tabl
e 1.
Tabl
e
1.
Param
e
t
e
rs set
t
i
ngs
f
o
r
C
4
.
5
t
r
ai
ni
n
g
Para
m
e
ter
Description
Value
Conf
i
d
enceFactor
The conf
idence f
actor used f
o
r pr
unin
g
(
s
m
a
ller
values incur
m
o
r
e
pr
uning
)
.
0.
25
m
i
n
N
u
m
Obj
The
m
i
ni
m
u
m
nu
m
b
er of
instances
per leaf
.
2
Unpr
uned
W
h
ether
pr
uning is per
f
orm
e
d or
not
False
2.
4. Perf
orm
a
nce
E
v
al
u
a
ti
o
n
Th
e correct classificatio
n
rate
(CCR) can
b
e
d
e
fi
n
e
d
as
o
n
e
a k
e
y g
a
ug
e em
p
l
o
y
ed
fo
r analyzin
g
on
e
p
a
rticu
l
ar o
r
ev
en
classifier. Nev
e
rth
e
less, CCR o
n
l
y can
b
e
in
ad
equ
a
te reg
a
rd
i
n
g
g
a
ugin
g
a fun
c
tion
a
lity
o
f
th
e classifier
fo
r a static security in
d
e
x
d
a
ta set. An
d
so
, th
e tru
e
n
e
g
a
tive rate (TNR
) an
d tru
e
p
o
s
itive rate
(TPR) we
re us
ed
to
e
v
aluate the
cl
assifier
perform
a
nce. M
o
re
over, ge
om
etric
m
ean (GM) was a
d
di-tionally
u
tilized
in
th
is research
to
assess th
e actu
a
l o
v
e
rall p
e
rforman
ce reg
a
rd
i
n
g
d
e
v
i
ce studyin
g
tech
-n
iques, as
sho
w
n i
n
Ta
bl
e 2.
Tabl
e
2. T
h
e
p
r
oce
d
ures
em
ploy
ed
f
o
r a
ssess
i
ng t
h
e e
ffi
ci
en
cy
of
m
achi
n
e l
earni
n
g
t
e
c
hni
que
s
Measures na
m
e
F
o
rm
ula
Correct classification
rate (CC
R
)
(%
)
True positive rat
e
(TPR
)
(
%
)
True negative rat
e
(TNR
)
(
%
)
Geo
m
etri
c m
e
an (
G
M)
(
%
)
whe
r
e:
TP (t
ru
e
po
sitiv
e): t
h
e
n
u
m
b
e
r po
sitiv
e
samp
les classified
co
rrectly, FP
(false po
sitiv
e):
th
e
n
u
m
b
e
r
nega
-tive
sam
p
les classified i
n
correc
tly, TN (tru
e
n
e
g
a
tiv
e): th
e nu
m
b
er
n
e
g
a
tiv
e sam
p
les classified correctly
an
d FN (false
n
e
g
a
tiv
e): th
e
n
u
m
b
e
r
po
sitiv
e sam
p
les classified
in
co
rrectly
After
we in
itialize a p
a
ttern
vecto
r
(X
SSE
) fro
m
d
a
ta co
llec
tio
n
and
data pre-p
r
o
cessi
n
g
, we in
itialize
fea-ture vect
or (Z
SSE
) fr
om
cr
oss val
i
d
at
i
on
and
num
ber o
f
i
n
st
ances.
Dat
a
sam
p
l
e
s gen
e
rat
e
d are r
a
n
d
o
m
l
y
sp
lit in
train
i
ng
and
testin
g
p
r
o
cess in
approx
im
ate
l
y p
r
op
ortion
of 75% an
d
25
% resp
ectiv
ely. A t
r
ain
i
ng
pattern
(Z
SSE
v
ect
or)
t
a
kes
t
h
e
f
o
rm
at
<x1 ,
x
2
,
x
3
,
x4
,…
……,
x
n
>
w
h
er
e x1
,
x2
, x3
,
x
4
,………, xn
d
e
no
te
th
e in
pu
t v
ect
or
and
d
e
no
tes th
e secur
ity statu
s
ou
tpu
t
vector (ta
r
get). This t
r
aining
pattern call
e
d insta
n
ces
(row) wh
ile the in
pu
ts ar
e featu
r
ed
or attrib
-u
tes
(col
um
n). The
po
we
r sy
stem
con
d
ition is
, in
fact, k
n
o
w
n
as
‘Static Secure
’ (S
S-Bina
ry
o
n
e)
wh
en
-eve
r
all the
l
i
m
i
t
a
t
i
ons
m
e
nt
i
one
d
i
n
3.
1
are oft
e
n
sat
i
s
fi
ed fo
r
al
m
o
st
any
p
r
o
v
i
d
e
d
back
u
p
. Wh
en
s
o
m
e
body
i
ssues
brea
k '
i
s i
d
ent
i
f
i
e
d pe
rf
orm
i
ng a pr
o
b
l
e
m
,
the de
vi
ce si
t
u
at
i
on i
s
goi
ng t
o
be k
n
o
w
n
as
‘St
a
t
i
c
Insecu
re’ (
S
I
-
B
i
nary
zer
o)
.
TP
TN
CCR
T
P
FP
FN
T
N
TP
TPR
TP
F
N
TN
TN
R
TN
F
P
*
GM
T
P
R
T
N
R
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
New Cla
ssifier Desi
g
n
fo
r
S
t
atic S
ecu
rity Eva
l
ua
tio
n using
Artificia
l In
-tel
lig
en
ce Techn
i
q
u
e
s
(
I
br
a
h
i
m
S.)
87
4
Engineeri
n
g c
o
mm
on sense
occasionally may decide
on the actual e
n
t
e
r attributes.
Howe
ver, this
ki
n
d
of
ch
oi
ce
s i
s
g
o
i
n
g t
o
b
e
ve
ry
su
b
j
ect
i
v
e
usi
n
g
t
h
e c
h
ance
of esse
ntial factor
s
o
b
t
ai
n
i
ng
turn
ed dow
n.
A
typ
-
ical ap
pro
a
ch
to
feat
u
r
e selectio
n
will be a co
n
s
ec
u
tive featu
r
e cho
i
ce, co
m
p
o
s
ed
o
f
two
elem
en
ts
- a tar-
get
f
unct
i
o
n
kn
ow
n as c
r
i
t
e
ri
o
n
an
d al
s
o
a co
nsec
ut
i
v
e i
n
ve
st
i
g
at
i
on
fo
rm
ul
a. T
h
e r
eal feature factors
chose
n
through SFS te
chni
que ca
n se
rve as a
n
input
data source
re
garding creating the actual
cla
ssifier form
ula. The
SFS technique
utilized in the curre
nt f
uncti
on
begi
ns with an e
m
pty group
of feat
ures
and als
o
enc
o
urages
pr
os
pect
i
v
e cl
i
e
nt
f
u
nct
i
o
n
s
ubs
et
s wi
t
h
t
h
e hel
p
of
o
n
e
at
t
r
i
but
e e
v
ery
t
i
m
e
. Fo
r eac
h
pr
os
pect
i
v
e
cl
i
e
nt
p
e
rf
or
m
co
m
p
o
n
e
n
t
,
SFS op
er
ates th
e act
u
a
l
10
-f
o
l
d
co
m
b
i
n
e au
tho
r
ization
thr
ough
fr
equ
e
n
tly con
t
actin
g th
e
actu
a
l qu
alifyin
g
criterion
operate.
3.
RESULTS
A
N
D
DI
SC
US
S
I
ON
W
i
t
h
in
th
is research, C4
.5
mo
d
e
ls
were
p
r
op
er
l
y
t
r
ai
ne
d b
y
usi
ng a
WE
KA t
o
ol
.
WE
KA i
s
t
r
ul
y
a
work-b
en
ch
d
e
sig
n
e
d
t
o
help
th
e u
s
e of m
ach
in
e learn
i
n
g
ap
pro
ach
es to
variou
s actu
a
l
difficu
lties.
WEKA is
tru
t
hfu
lly a to
tally released
an
d
also
free co
d
e
d
e
ve
lop
e
d in
Jav
a
. In
WEKA, th
e machine learning algo-
rith
m
s
ten
d
to
b
e
realists o
r
g
a
n
i
zed
in
t
o
p
r
ogram
s, to
allo
w th
em to
b
eco
me efficien
tly b
r
o
ugh
t in
and
besid
e
s
appl
i
e
d
i
n
Ja
v
a
'
s
code. R
i
g
h
t
aft
e
r t
h
e t
r
ai
ni
n
g
, t
h
e p
r
o
p
e
rl
y
t
r
ai
ned
d
e
si
gns
ha
d
be
en st
o
r
ed
j
u
st
as t
h
e
doc
um
ent
s
bei
n
g
ap
pl
i
e
d i
n
e
nha
nci
n
g t
h
e s
t
at
i
c
securi
t
y
st
age d
u
r
i
n
g t
h
e
t
e
st
st
age.
A
b
out
a
p
pl
y
i
ng
WE
KA
classifiers in Ja
va'
s
code,
WE
KA
gu
id
e are av
ailab
l
e in [1
6
]
.
In
t
h
e stead
y
-state, th
e SSE li
m
i
tatio
n
s
are
th
e bu
s
vo
ltage m
a
g
n
itu
d
e
(
V
k
) an
d th
e li
n
e
th
erm
a
l
po
we
r (
S
)
an
d can
b
e
written
as:
1
.
09
>
V
k
>
0.91
an
d
S
<
S
ma
x
.
The o
u
t
c
om
es of i
n
f
o
rm
at
i
on b
u
i
l
d
i
n
g an
d sh
ow c
hoi
ce
st
ages of st
at
i
c
securi
t
y
eval
uat
i
on are
shown in Tabl
e 3. The
data sa
m
p
les in
m
-
dimensional
feat
ure space are random
ly split i
n
to traini
ng a
n
d test
sets.
Table 3. Data gene
ration
a
n
d
feature
se
lection of
differe
n
t
IEEE test syste
m
s
Syste
m
siz
e
Operating
scenarios
Static Se
cure
(SS)
Static Insec
u
re
(SI)
No
. o
f
pa
t
t
ern
variables (
X
SS
E
)
No.
of
feat
ures
selected
(
Z
SS
E
)
Di
m
e
nsionality
reduction
30 Bus
860
595
265
170
25
14.
70%
57 Bus
950
630
320
185
27
14.
59%
118 Bus
1100
750
350
210
29
13%
300 Bus
1330
760
570
220
26
11.
81%
Fr
o
m
th
is tab
l
e, 30
, 57
, 118 and
3
0
0
I
E
EE bu
s
syste
m
s ar
e
u
s
ed
i
n
t
h
is
p
a
p
e
r
,
t
h
e op
er
ation
scen
ar
i
o
s
ar
e 86
0, 9
5
0
,
11
00
an
d
133
0
r
e
sp
ectiv
ely.
A
ll
th
e
s
e scena
r
ios
are classified
either
static secure (SS)
o
r
static in
secure (SI). The imp
act of
th
e feat
u
r
e selection
ap
pro
ach
u
s
ed
in
th
is research work
is m
e
n
tio
n
e
d
in
th
e tab
l
e as
d
i
m
e
n
s
io
n
a
lity
red
u
c
tion
wh
ich
is
d
e
sign
ati
n
g b
y
bo
ld
v
a
l
u
es.
In
order t
o
eva
l
uate the perform
ance of a static s
ecu
rity ev
alu
a
tion
appro
ach, it is v
e
ry i
m
p
o
r
tan
t
to
measure its perform
a
nce. T
h
ere
f
or
e,
som
e
com
m
on pe
rf
orm
a
nce m
e
asur
e
s
are
us
ed to e
v
al
uate the
per
f
o
r
m
a
nce of
a pa
rt
i
c
ul
ar
se
curi
t
y
st
at
us i
n
dex
com
p
are
d
wi
t
h
ot
he
r a
p
p
r
oache
s
.
Fo
ur
di
ffe
re
nt
al
gori
t
hm
s of DT
’s wi
t
h
sam
e
t
r
ai
n da
t
a
set
s
and t
e
s
t
dat
a
set
s
are used i
n
a
co
m
p
ariso
n
. Th
is co
m
p
arison was in
term
s
o
f
CCR, TNR
,
TPR, GM and co
m
p
u
t
atio
n
ti
m
e
an
d
p
r
esen
ted
in
tab
l
e 4
.
Fo
r
extra kno
wled
g
e
reg
a
rd
i
n
g th
e
artificial in
te
lli
g
e
n
c
e techn
i
ques algo
rith
m
s
u
s
ed
in
t
h
is st
u
d
y
is
p
r
esen
ted in
[17
]
.
Tabl
e 4 s
h
ows
t
h
e com
p
ari
s
o
n
bet
w
een t
h
e
per
f
o
r
m
a
nce'
s
m
easures
of
pr
op
ose
d
C
4
.5 a
nd
ot
he
r f
o
u
r
vari
ous
D
T
’s
t
echni
que
s
fo
r t
h
e t
w
o
net
w
o
r
k
dat
a
set
s
(
5
7
an
d
1
1
8
IE
EE
t
e
st
sy
st
em
s)
i
n
b
o
t
h
t
r
ai
ni
n
g
a
n
d
testin
g
d
a
ta set
s
. In
Tab
l
e
4
,
th
e
b
e
st an
d the wo
rst
v
a
lu
es o
f
th
e m
easu
r
es are
h
i
gh
lighted
in
bo
ld
fo
nt an
d
un
de
rl
i
n
e f
o
nt
,
res
p
ect
i
v
el
y
.
I
n
t
r
ai
ni
n
g
phas
e
(
5
7
b
u
s
syste
m
), BF Tree,
Stum
p Tree, J
48 T
r
ee a
n
d J
48
gra
f
t
attain
ed
arou
nd
94
.70
%
, 95
.4
%, 93
.7
0%, 9
4
.60
%
of
CCR resp
ectiv
ely
,
wh
ile C4
.5
Tree attain
ed
o
f
CCR
aro
u
nd
9
8
.
6
4
%
. I
n
t
e
st
i
ng
pha
se, B
F
T
r
e
e
, St
um
p Tree,
J 48
Tree a
n
d
J 48
g
r
aft
at
t
a
i
n
ed a
r
o
u
nd
93
.5
0%
,
9
1
.2%, 92
.5
0%, 9
3
.40
%
of CCR
resp
ectively,
wh
ile C
4
.5 Tree attain
ed
aroun
d 97
.44
%
of CCR.
I
n
tr
ain
i
n
g
p
h
a
se (
1
1
8
b
u
s
sy
ste
m
)
,
BF Tr
ee, Stu
m
p
Tr
ee, J 4
8
Tr
ee and J 4
8
gr
af
t attain
ed
around
9
4
.50
%
, 95
.2%, 93
.5
0
%
, 94.20
% of CCR resp
ectiv
ely, wh
ile
C4
.5
Tree attain
ed
arou
nd
98
.4
4% o
f
C
CR. In
testing phase,
BF Tree, Stum
p Tree, J 48 Tree and J
48
gr
af
t attain
ed
aro
und
93
.80
%
,
9
1
.5
%, 92
.8
0%,
9
3
.70
%
of CC
R resp
ectiv
ely, wh
ile C4
.5
Tree attain
ed
arou
nd
97
.7
4%
o
f
CCR.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 2, A
p
ri
l
20
16
:
87
0 – 8
7
6
87
5
Table 4. Performance
of
C4.5 classi
fier for st
atic security evaluation
Proposed Classifi
er
Decision Tree
classifiers (
D
TC’s)
C4
.5
Tree
BF Tre
e
Stu
m
p
Tr
ee
J 4
8
Tree
J 4
8
g
r
af
t
IEEE 57 bus
T
o
tal sam
p
les
,
950
Train set
Sam
p
les
630
CCR (%)
98.
64
94.
70
95.
4
93.
70
94.
60
TPR (%
)
96.
30
93.
90
95.
1
93.
20
94.
00
TNR (%
)
97.
21
94.
30
95.
00
93.
30
94.
20
GM (%)
96.
75
94.
09
95.
049
93.
25
94.
10
Ti
m
e
(s)
0.
0001
0.
001
0.
02
0.
01
0.
03
Test set
Sam
p
les
320
CCR (%)
97.
44
93.
50
91.
2
92.
50
93.
40
TPR (%
)
95.
90
93.
60
95.
5
93.
70
94.
90
TNR (%
)
97.
21
94.
15
95.
70
93.
30
94.
20
GM (%)
96.
55
93.
87
95.
59
93.
49
94.
55
Ti
m
e
(s)
0.
0001
0.
003
0.
04
0.
02
0.
05
IEEE 118 bus
T
o
tal sam
p
les
,
1100
Train set
Sam
p
les
750
CCR (%)
98.
44
94.
50
95.
20
93.
50
94.
20
TPR (%
)
96.
80
94.
30
95.
2
93.
70
94.
80
TNR (%
)
97.
5
95.
00
94.
90
93.
10
94.
10
GM (%)
97.
14
94.
65
95.
049
93.
39
94.
45
Ti
m
e
(s)
0.
0001
0.
001
0.
052
0.
01
0.
05
Test set
Sa
m
p
les
350
CCR (%)
97.
74
93.
80
91.
50
92.
80
93.
70
TPR (%
)
97.
10
93.
75
94.
7
94.
10
94.
10
TNR (%
)
96.
90
94.
30
94.
90
93.
20
94.
30
GM (%)
96.
99
94.
02
94.
79
93.
64
94.
19
Ti
m
e
(s)
0.
001
0.
002
0.
055
0.
015
0.
08
B
o
l
d
val
u
e
val
i
dat
e
s t
h
at
C
4
.
5
pr
o
v
i
d
es
g
r
e
a
t
cor
r
ect
cl
ass
i
fi
cat
i
on
rat
e
a
n
d
m
i
nim
u
m
com
put
at
i
on
tim
e
to other DTC’s classifiers.
Fi
nal
l
y
, fo
r t
r
a
i
n m
ode an
d t
e
st
m
ode, t
a
bl
e 4 al
so
dem
onst
r
at
es t
h
e
co
m
put
at
i
on t
i
m
e
i
n
sec
o
n
d
s
.
St
ro
n
g
l
y
, i
t
can
be
o
b
ser
v
e
d
t
h
at
fo
r
bot
h sy
st
em
s used
, C
4
.
5
g
o
t
m
i
nim
u
m
com
put
at
i
on t
i
m
e
(0.
0
00
1)
se
con
d
fo
r t
r
ai
ni
ng a
n
d t
e
st
i
ng
pha
se
s. Fu
rt
he
rm
ore, fo
r t
h
e
recal
l
(t
est
)
p
h
ase
wh
ere C
4
.
5
got
c
o
m
put
at
i
on t
i
m
e of 0
s. an
d 0.001
s.
f
o
r
tr
ai
n
i
ng
and
testing
p
h
a
se r
e
sp
ectiv
ely.
4.
CO
NCL
USI
O
N
The res
u
l
t
s
an
d di
sc
ussi
o
n
s
of
usi
n
g C
4
.
5
and
ot
he
r deci
sion tree class
i
fiers for SSE
the electric
po
we
r has
pre
s
ent
e
d.
Al
s
o
, t
h
e resul
t
s
an
d
d
i
scussi
o
n
s
of
u
s
i
ng
feat
u
r
e se
l
ect
i
on f
o
r
desi
gni
ng cl
assi
fi
ers f
o
r
SSE the electric power
gri
d
has prese
n
ted. T
h
e im
ple
m
en
tation of feature
selection
invol
v
ed a
p
propriateness
data reduction. The im
ple
m
entation dec
i
sion tree
m
e
thods on se
veral IEEE
tes
t
syste
m
s
involve
d
appropriatenes
s SSE and cla
ssi
fi
cat
i
on by
usi
n
g f
o
u
r
al
g
o
ri
t
h
m
s
of
DT’s. Fro
m
th
is
research
wo
rk
, it is
obs
erved t
h
at
all these algorithm
s
prom
ise
success
f
ul
a
n
d alternative tec
hni
que
s for large scale powe
r
grid
SSE.
98
.7
%
of CCR and
0.0
001
seco
nd
of co
m
p
u
t
atio
n
ti
m
e
mad
e
C4.5
is
v
e
ry
well fit in
th
e
real-ti
m
e
powe
r system
s
SSE. Mentioned techniques
can e
ffectivel
y
be i
m
pl
em
ent
e
d
fo
r S
S
E
wi
t
h
hi
g
h
acc
urac
y
rat
e
.
ACKNOWLE
DGE
M
ENTS
Th
e au
tho
r
s
wo
u
l
d
lik
e to
exp
r
ess t
h
eir ap
precia
tio
n
to
Un
iv
ersiti Tek
n
o
l
og
i Malaysia
(UTM
) for
th
e facilities and
su
ppo
rt.
REFERE
NC
ES
[1]
Singh, S. and
S. Srivastava, I
m
proved
contin
gency
selection
algorithm for
vo
ltag
e
security analy
s
is.
Electric
machines and
po
wer s
y
stems,
19
98. 26(8): p. 855
-871.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
New Cla
ssifier Desi
g
n
fo
r
S
t
atic S
ecu
rity Eva
l
ua
tio
n using
Artificia
l In
-tel
lig
en
ce Techn
i
q
u
e
s
(
I
br
a
h
i
m
S.)
87
6
[2]
Ejeb
e, G. and B.
W
o
llenberg, Au
tom
a
tic con
tinge
nc
y
se
lec
tion
.
Power Apparatus and S
y
ste
m
s,
IEE
E
Tra
n
sac
t
ions
on, 1979(1): p. 9
7
-109.
[3]
Verm
a, K. and K. Niazi
, S
upervis
ed learn
i
ng a
pproach
to online conting
e
ncy
screening and r
a
nking in power
s
y
stems. International J
ournal of
Electrical Power
&
Energ
y
S
y
stems, 2012.
[4]
Camara, F., et al., Privacy
Pr
eser
ving RFE-
SVM for Distributed
Ge
ne Selection.
2012.
[5]
Jun, S., A Clustering Method
of
Highly
Dimensional Patent Data
Using Bay
e
sian
Approach. 2012.
[6]
Morison, K., L. Wang,
and
P. Kundur,
Power s
y
stem
secur
i
ty
assessm
ent. Power and
Energ
y
Magazine, I
E
EE,
2004. 2(5): p. 30
-39.
[7]
Albu
y
e
h
,
F., A
.
Bose,
and B
.
Heath,
Re
act
ive
power
consider
ations i
n
au
tom
a
ti
c con
tingen
c
y sele
ction
.
Pow
e
r
Apparatus and
S
y
stems, I
E
EE Tr
an
sactions on, 1
982(
1): p. 107-1
12.
[8]
Wehenkel, L.A., Automatic learn
i
ng
techniqu
es in power s
y
s
t
ems. 1998:
Kluwer
Academ
ic Publishers.
[9]
Marsadek, M., et al. R
i
sk based
static
secu
rity
assessment in a practical
in
ter
c
onnected power
s
y
stem. 2008: IEEE.
[10]
Mori, H. State-o
f-the-ar
t
ov
erview on da
ta minin
g
in power
s
y
s
t
ems. 2006: IEEE.
[11]
Sunitha R, R. Sreerama Kumarand Abraham
T. Mathew
” O
n
line Static Security
Assessment Module Usin
g
Artificial Neural Networks” I
E
EE Tr
ansactions
On
Power S
y
stems, VOL. 28, NO. 4, November
2013.
[12]
12.
Momoh, J.A., Y. Xia, an
d
G.D. B
o
swell.
Location
a
l Marg
inal Prici
ng for r
eal and
r
eactiv
e power
. 2
008: IEEE.
[13]
Stott, B. and O.
Alsac, Fast deco
upl
ed load flow.
Power Apparatus and Sy
st
ems, IEEE Tr
ansaction
s
on, 1974(3): p.
859-869.
[14]
Zhou, W., Y. Peng, and H. Sun. Proba
bilisti
c wind power penetr
ation of pow
er sy
stem
using nonlinear predictor-
corrector pr
imal-
dual
inter
i
or-point method. 2008: IEEE.
[15]
Huang, C.J., et
al., Applications of
machine
learning techniques to a senso
r-ne
t
work-bas
ed pro
s
thes
is
train
i
ng
s
y
stem. Applied
Soft Compu
ting, 2011. 11(3): p
.
3229-3237.
[16]
Bouckaer
t, R.R
.
, et al
., WEKA Manual for Version 3-6-
1. 2009, The Univer
sit
y
of Waikato
:
Ham
ilton, New
Zealand.
[17]
Witten, I.H
.
and
E. Frank
,
D
a
ta
Mi
ning: Practi
cal m
achin
e
learn
i
ng tools
and
tech
niques. 2005
: M
o
rgan Kaufm
a
nn
.
BIOGRAP
HI
ES OF
AUTH
ORS
Dr. Ibrahim Saeh receiv
e
d his
BSc. (
E
lectrica
l & Electronics) in (1997) fr
om Bright Star
University
of
Technolog
y
-
Lib
y
a. MSc, (Electri
cal Power) in (2
009) and Doctor
of Philosoph
y
(Electri
cal Power Eng.) in (20
14) from
Unive
r
siti Teknolog
i Malay
s
ia (UTM
). His research
inter
e
sts include power sy
stem analy
s
is, deregu
la
ted power s
y
s
t
e
m
, power s
y
s
t
e
m
s
ecurit
y
and
Artificial in
telligence
techn
i
ques. He has publis
hed as authored and co-autho
red a sever
a
l
res
earch p
a
pers
in num
erous
tec
hnica
l journa
ls
and confer
enc
e
proceed
ings
Dr. Ibrahim
is
a
director of
Environmental R
e
s
e
arch &
Cl
ean Energ
y
C
e
ntre
(ERCE) and
Th
e
Int
e
rnat
ional
S
o
ciet
y
of Oc
ea
n, M
ech
ani
cal
a
nd Aeros
p
ac
e –s
cien
ce
and
engin
eering
(IS
OM
Ase) m
e
m
b
er.
M.W. Mustafa
he receiv
e
d his
B.Eng d
e
gr
ee (1988), M.Sc
(1993) and PhD (1997) from
Universit
y
of S
t
rath
cl
yde
.
His
resear
ch in
ter
e
st includ
es po
wer s
y
stem
st
a
b
ilit
y,
FACTS,
wireless power
transmission an
d power s
y
s
t
em
distribution au
tomation,
de-r
egulated power
s
y
s
t
em
, et
c. He
is
currentl
y
a P
r
ofes
s
o
r and Deput
y
De
an (Acad
em
ic) at F
a
cult
y of Elec
tri
c
a
l
Engine
ering,
Universiti
Tekno
lo
gi Mala
ysi
a
. Dr
. Mustafa
is als
o
a m
e
m
b
er of
Institut
i
on of
Engineers Malaysia (IEM)
and a m
e
m
b
er
of
IEE
E
.
Dr. Nas
i
r Ahm
e
d Algeel
ani r
e
c
e
i
ved th
e B.
E. d
e
gree in
el
ec
tric
al
power s
y
s
t
em
from
Univers
i
t
y
of Aden, Yemen, Aden,
in 199
7, the M
.
E. degr
ee in
el
ectr
i
c
a
l
power s
y
s
t
em
e
ngineer
ing from
University
Tech
nolog
y
Malay
s
ia in 2009 and the
Ph.D. degree
in
high voltag
e
en
gineer
ing from
Universit
y
Tech
nolog
y
Mal
a
y
s
ia in 2014. He wa
s
a Lecturer wit
h
Industrial Technical Institut
e
(ITI) for 25
y
e
ars, where h
e
is
currently
a senior
le
cturer
of Hi
gh Voltag
e
Eng
i
neering
.
At
the
present he is a
postdoctoral can
didate at high
voltag
e
engineering depa
rtm
e
nt
at Univers
i
t
y
Techno
log
y
Malay
s
ia. He has published as auth
or
ed and co-authored more than 30 papers in
various techn
i
cal journals and
conferen
ce proc
eedings. His research interests
include high-
voltag
e
instrum
e
nta
tion, p
a
rti
a
l
discharge
,
det
ect
ion and war
n
ing s
y
stem
s a
nd condition
monitoring of h
i
gh power
equip
m
ent.
Evaluation Warning : The document was created with Spire.PDF for Python.