In
te
r
n
ation
a
l Jou
rn
al
o
f Po
we
r
Elec
tron
ic
s an
d
D
r
ive S
y
stem
(IJ
PED
S
)
V
o
l.
11, N
o.
1, Mar
ch 20
20,
p
p.
359~
3
6
6
IS
S
N
: 2088-
86
94,
D
O
I
:
10.11
59
1
/ij
ped
s
.
v11
.
i
1.pp
3
59-
36
6
359
Jou
rn
a
l
h
o
me
pa
ge
:
ht
tp:
//i
j
p
eds.i
a
esco
re
.com
An ELM
-based single i
n
put r
ule m
o
dule and its applic
ation in
p
o
wer generation
C
h
on
g Ta
k
Y
a
w
1
,
Shen Yo
u
n
g
Wo
ng
2
, K
eem S
i
a
n
Yap
3
1,
3
Dep
a
rt
m
e
nt
o
f Electro
ni
cs
an
d
Co
m
mu
n
i
cati
o
n
En
gi
neerin
g
,
U
n
i
ver
s
i
ti T
e
nag
a
N
asi
o
n
a
l, M
a
l
ays
i
a.
2
Dep
a
rt
m
e
nt
o
f E
l
ectri
cal an
d
E
lect
ron
i
cs Eng
in
e
e
rin
g
, X
i
am
en U
n
ivers
i
t
y
M
al
aysia, M
al
aysi
a
Art
i
cl
e In
fo
ABSTRACT
A
r
tic
le hist
o
r
y
:
R
e
ce
i
v
e
d
Sep 7,
2019
Re
vise
d N
ov
1
2
,
201
9
A
c
c
e
pte
d
D
ec 8,
201
9
Ex
trem
e
Learni
n
g
M
ach
in
e
(ELM
)
is
w
id
ely
known
as
a
n
eff
ecti
v
e
learni
ng
alg
o
rithm
t
h
an
t
he
c
on
vent
io
nal
l
earni
ng
m
e
th
ods
f
rom
th
e
poi
n
t
o
f
learni
ng
speed
a
s
well
as
g
ener
al
izati
on.
I
n
traditional
f
u
zzy
i
nf
erence
m
etho
d
w
h
ich
was
th
e
"i
f-t
h
en"
ru
les
,
a
ll
t
h
e
i
np
ut
a
nd
o
u
t
put
o
bjects
w
e
re
as
si
gn
ed
t
o
ant
ecedent
and
c
o
ns
equ
e
nt
c
o
m
pon
ent
respect
iv
e
l
y
.
H
ow
eve
r
,
a
m
a
jor
di
le
m
m
a
w
as
t
hat
t
h
e
f
u
zzy
r
ules'
nu
m
b
er
k
ept
i
n
creasi
n
g
u
n
til
th
e
s
y
s
t
em
and
arrang
em
ent of
t
he ru
l
es
b
ecam
e c
o
m
p
l
i
cated.
T
h
eref
ore,
t
h
e
s
ingle inp
u
t
rul
e
m
o
d
u
l
es
c
on
nected
t
yp
e
f
u
zzy
i
n
f
erence
(S
IRM)
m
eth
o
d
wh
ere
c
o
nsoc
ia
te
d
th
e
ou
tp
ut
o
f
th
e
fu
z
z
y
r
u
l
e
s
m
od
ule
s
s
ig
ni
fic
a
n
t
l
y
.
In
t
his
paper,
w
e
p
u
t
f
o
r
w
a
r
d
a
n
o
v
e
l
s
i
n
g
l
e
i
n
p
u
t
r
u
l
e
m
o
d
u
l
e
s
b
a
s
e
d
o
n
e
x
t
r
e
m
e
l
earni
ng
m
achi
n
e
(d
eno
t
e
d
a
s
S
I
RM
-E
LM
)
f
o
r
s
o
lvin
g
dat
a
r
egr
e
s
s
io
n
pro
b
l
em
s.
I
n
thi
s
h
yb
r
i
d
mod
e
l,
t
h
e
c
on
c
e
p
t
of
S
IRM
is
a
pp
lie
d
a
s
hidd
e
n
n
e
u
rons
o
f
E
L
M
and
ea
ch
o
f
th
em
r
ep
resent
s
a
s
i
n
g
le
i
n
p
u
t
f
uzzy
r
ul
es.
Hen
ce,
t
h
e
nu
m
b
er
o
f
f
u
zzy
r
ul
e
and
the
n
u
m
b
er
o
f
h
i
d
d
en
n
euro
n
o
f
E
L
M
a
re
t
he
s
am
e.
T
h
e
ef
f
ecti
v
enes
s
of
p
rop
o
sed
S
I
RM
-EL
M
m
od
el
i
s
verifi
ed
u
s
i
ng
s
igm
oid
acti
v
at
ion
f
unc
tio
ns
b
ased
on
s
e
veral
b
e
nch
m
ark
d
a
tas
e
ts
a
n
d
a
NOx
emissi
on
of
pow
er
g
ener
ati
o
n
pl
an
t.
Exper
i
ment
al
r
esults
illus
tra
t
e
that
o
ur
pro
p
o
s
ed
S
IRM
-
ELM
m
o
del
is
capabl
e
o
f
achiev
i
n
g
s
m
a
l
l
r
oo
t
mean
s
q
u
a
r
e
error,
i
.e., 0.027448 f
o
r
predictio
n of
N
O
x
e
m
i
s
s
i
o
n
.
K
eyw
ord
s
:
D
a
ta
R
egre
ss
io
n
Ex
trem
e
Lear
n
i
n
g
M
ac
h
i
ne
(ELM)
NO
x
E
mission of
Power
G
e
ne
rati
o
n
P
l
a
nt
S
i
ng
le
I
np
u
t
R
ule
Mo
dule
(SIR
M)
Th
is
is a
n
o
p
en acces
s a
r
ti
cle u
n
d
e
r t
h
e
CC
B
Y
-S
A
li
cens
e
.
Corres
pon
d
i
n
g
Au
th
or:
S
h
e
n
Y
uo
ng
Wo
ng
D
e
pa
rtme
nt
o
f
El
e
c
t
rica
l
and
El
ect
ro
ni
c
s
Eng
in
e
e
ring
,
Xiam
en
U
ni
ve
rsity
M
a
l
ays
i
a
,
Jala
n
S
unsur
ia,
Ban
d
ar
S
unsu
r
ia,
439
00
S
e
p
a
ng, S
elang
o
r
,
Ma
l
a
y
s
ia
.
Em
ail:
she
n
y
u
o
n
g
.
w
on
g
@
x
m
u.
edu.m
y
1.
I
N
TR
OD
U
C
TI
O
N
La
te
ly,
E
x
tre
m
e
Le
arnin
g
M
ac
hine
(
EL
M)
h
as
b
e
e
n
a
c
know
le
d
g
e
d
a
s
a
n
e
ffe
c
t
i
v
e
lear
n
i
n
g
alg
o
ri
t
h
m
tha
n
t
he
c
o
nve
n
tio
na
l
le
arn
i
ng
m
e
th
o
d
s
from
th
e
persp
ect
i
v
e
of
g
en
era
l
i
z
at
ion
a
n
d
le
arn
i
ng
s
p
ee
d
[1-8]
.
T
he
i
ns
pi
r
a
ti
on
of
t
he
E
x
t
rem
e
L
ear
nin
g
M
ach
i
n
e
(ELM)
su
gg
est
e
d
by
H
u
a
ng
e
t
a
l
.
co
mes
f
r
o
m
bi
olo
g
i
ca
l
le
ar
ni
n
g
.
It
i
s
ap
pl
ica
b
l
e
f
or
s
ol
v
i
n
g
p
r
oble
m
s
per
t
ai
nin
g
to
b
ac
k-pro
p
a
ga
ti
on
(
B
P
)
l
ear
n
i
n
g
alg
o
ri
t
h
ms.
It
i
s
t
h
e
r
efor
e
c
o
n
j
e
c
t
u
red
t
h
a
t
c
ertai
n
p
ar
t
s
o
f
t
h
e
b
r
a
i
n
s
i
g
n
a
l
s
a
r
e
m
a
d
e
u
p
o
f
r
a
n
d
o
m
n
e
u
r
o
n
s
tha
t
a
re
i
n
d
ep
ende
n
t
o
f
t
h
e
i
r
en
vironm
ent
[1].
T
his
pr
ocess
is
k
n
o
w
n
a
s
E
LM
o
r
so
call
e
d
Si
ngle
Lay
e
r
F
e
edfor
w
ard
N
e
tw
ork
(S
LF
N
)
.
Its
c
o
rrespon
d
i
n
g
g
ene
r
a
l
a
rchi
tec
t
ur
e
w
a
s
il
lus
t
rate
d
i
n
F
ig
ure
1.
E
L
M
h
a
s
the c
a
pa
b
i
li
ty
t
o
ma
ke
unive
rs
al a
ppr
ox
i
m
at
i
on w
i
th ha
pha
z
a
r
d
b
iases
an
d in
p
u
t w
e
igh
t
s [
9
].
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N: 2
0
8
8
-
86
94
I
nt
J
P
o
w
E
l
e
c
&
D
r
i
S
yst
V
o
l.
11,
N
o.
1
,
Mar
202
0
:
359
–
36
6
36
0
F
i
gur
e
1.
A
r
c
hitec
t
ur
e
of
E
L
M
I
n
t
r
a
d
i
t
i
o
n
a
l
f
u
z
z
y
i
n
f
e
r
e
n
c
e
m
e
t
h
o
d
w
h
i
c
h
w
a
s
t
h
e
"
i
f
-
t
h
e
n
"
r
ule
s
,
al
l
the
inp
u
t
a
n
d
o
ut
put
o
b
j
ec
ts
w
e
r
e
a
ssigne
d
t
o
a
ntec
e
d
e
n
t a
nd c
onse
q
uen
t
c
om
po
ne
nt
r
es
pec
t
i
v
e
ly
.
Ho
wev
e
r,
a
m
ajo
r
d
i
l
emma was t
h
a
t
th
e
f
u
zz
y
r
u
le
s'
n
u
m
ber
ke
p
t
i
nc
r
e
a
s
i
ng
u
n
t
il
th
e
sys
t
em
a
n
d
a
r
r
ang
em
e
n
t
of
t
he
r
ule
s
b
eca
me
c
om
pl
i
c
ate
d
[
1
0
]
.
The
r
ef
or
e,
t
he
s
i
ngle
i
npu
t
r
u
le
m
od
u
l
es
c
o
nnec
t
e
d
t
y
p
e
f
u
zz
y
i
nf
er
e
n
ce
(
SI
R
M
)
m
e
th
od
w
h
e
r
e
co
ns
ocia
t
e
d
the
o
u
t
p
u
t
o
f
t
h
e
f
u
z
z
y
r
u
l
es
m
odule
s
s
i
g
nif
i
can
t
l
y
[
1
1-
16]
.
T
he
S
IR
M
met
hod
h
a
d
b
een
a
p
p
li
ed
t
o
con
t
ro
l
of
f
i
r
s
t
a
s
w
el
l
as
s
e
c
o
nd
or
de
r
l
a
g
s
y
stem
w
i
t
h
d
e
a
d
t
i
m
e
[
11-
1
2
]
,
n
o
n
lin
ea
r
f
u
nc
tio
n
id
en
ti
fi
c
a
ti
on
[
10
],
a
n
t
i
-
sw
in
g
c
o
n
t
r
o
l
a
nd
p
o
s
i
t
i
o
n
i
n
g
of
o
ver
h
ea
d
tr
ave
l
i
n
g
cr
a
n
e
[1
3]
,
s
t
a
bi
li
zat
ion
co
nt
rol
o
f
i
n
v
e
r
t
e
d
p
endu
l
u
m
sys
t
em
s
[14-
16
]
,
a
s w
e
ll
a
s
ot
her
s
,
o
f
w
h
ich
dece
n
t
r
e
s
ul
ts
w
e
r
e
ac
q
u
ir
ed [1
7
-
2
2].
Assume
t
hat
a
sys
t
em
c
ons
is
ts
o
f
n
i
n
p
u
t
s
our
ce
a
n
d
one
o
ut
p
u
t
s
our
c
e
.
H
o
w
e
ve
r
,
t
he
s
ys
tem
c
a
n
a
l
s
o
b
e
e
x
te
n
d
e
d
w
i
t
h
p
lur
a
l
ou
t
p
u
t
s
our
ces.
This
i
s
t
h
e
bas
i
c,
with
n
i
np
u
t
s
our
c
e
f
or
S
I
R
M
:
m
i
j
j
i
i
j
i
i
j
i
C
u
A
x
R
th
e
n
if
i
SIRM
1
:
:
(1
)
In
(1
)
, e
ac
h
S
I
R
M
ind
e
p
e
n
d
e
n
t
ly
c
o
r
r
e
spond
e
d
to
n
i
n
p
u
t
sour
ce
s.
The
S
I
R
M-
i
w
her
e
t
he
i
ref
e
rs
t
o
i
t
h
i
npu
t
so
u
r
c
e
,
i
s
t
h
e
j
t
h
r
u
l
e
i
n
t
h
e
S
I
R
M
-
i
,
r
e
f
e
r
s
t
o
t
h
e
i
th
i
np
ut
s
o
u
r
c
e
var
i
ab
le
i
n
t
h
e
pr
e
c
e
d
i
n
g
pa
r
t
,
and
△
i
s
t
h
e
va
r
i
ab
le
i
n
t
h
e
fol
l
ow
ing pa
r
t
o
f
the
S
I
RM-
i
.
a
nd
a
r
e
t
he
m
e
m
ber
s
hi
p
f
unc
t
i
o
n
s
o
f
the
w
h
e
r
e
a
s
△
i
s
t
h
e
j
t
h
r
u
l
e
i
n
t
h
e
S
I
R
M
-
i
.
A
d
d
iti
o
n
al
l
y
,
i
=
1
,
2
,
…
,
n
i
s
t
h
e
i
n
d
e
x
n
u
m
b
e
r
o
f
t
h
e
SI
R
M
w
h
e
re
b
y
j
=
1
,
2
,…
,
i
s
the
i
n
de
x
num
be
r
o
f
t
he
r
ules
i
n
t
h
e
S
I
RM
-
i
.
Th
is
p
a
p
er
p
r
o
po
s
e
s
an
E
L
M
-
b
ase
d
m
ode
l
by
usi
n
g
EL
M
hy
br
id
w
i
t
h
S
I
R
M
(
h
e
r
e
a
f
te
r
de
no
t
e
d
as
S
I
RM
-
ELM)
.
I
n
the
S
I
RM-
ELM,
the
r
e
i
s
o
nl
y
a
s
i
ng
l
e
i
n
put
t
ha
t
c
onne
cted
t
o
t
h
e
r
u
l
e
s
w
h
er
e
t
h
e
r
u
le
s
ar
e
t
h
e
h
i
dd
en
n
e
u
ro
n
s
o
f
E
L
M
an
d
ea
ch
o
f
t
h
em
r
e
p
re
se
n
t
s
a
si
ngl
e
inp
u
t
f
uzz
y
r
ul
es.
He
nce,
t
he
num
ber
of
f
u
zz
y
r
u
le
a
n
d
t
he
n
umbe
r
of
h
i
d
den
ne
ur
o
n
o
f
EL
M
a
r
e
equi
va
len
t.
The
pa
per
is
o
r
d
e
r
ed
a
s
bel
o
w
.
I
n
S
e
c
t
i
o
n
I
I
,
the
l
e
ar
n
i
ng
alg
o
r
i
t
h
ms
o
f
S
I
RM-
ELM
a
r
e
e
xpla
i
ne
d.
Aft
e
r th
a
t
,
S
e
ct
ion
III p
r
esent
s
th
e
r
esults o
f b
e
n
c
h
m
ark
r
e
gre
ss
ion
d
a
tasets
(
e.g
.
A
b
a
l
o
n
e
, B
allo
on
, Strike and
S
p
ace
-
g
a)
t
o
t
e
st
t
he
p
r
o
pose
d
m
ode
l's
per
f
o
r
m
ance
.
The
a
ppl
ica
ti
on
o
f
t
he
p
r
o
po
sed
m
odel
is
t
es
te
d
an
d
pr
esen
ted
in
S
e
c
tio
n
I
V
w
h
i
c
h
i
s
us
i
n
g
th
e
NO
x
e
m
i
s
s
i
o
n
i
n
a
p
o
w
e
r
g
e
n
e
r
a
t
i
o
n
p
l
a
n
t
.
L
a
s
t
l
y
,
S
e
c
t
i
o
n
V
pr
esen
ts
a
r
e
c
a
p
i
t
u
la
t
i
on
o
f
i
m
por
ta
nt
f
ind
i
ngs
w
i
t
h
s
ug
ge
st
i
o
n
f
or
f
ur
the
r
w
or
k.
2.
THE
AL
GORIT
H
MS
OF SIRM-ELM
T
h
e
st
ru
ct
u
r
e
o
f
S
IR
M
-
E
L
M
i
s
i
l
l
u
s
t
r
at
ed
i
n
Fi
gu
re
2
.
T
h
e
st
ep
w
i
s
e
t
r
ai
nin
g
p
r
o
toc
o
l
s
a
re
liste
d
a
s
be
low
.
Ref
er
t
o
F
i
gur
e
2
for
the
deta
ils
d
e
f
i
n
i
t
i
on
o
f
v
ar
ia
bl
e
s
and
par
a
me
te
r
s
.
Step
1
:
H
a
pha
z
a
r
d
l
y
s
et
t
he
i
np
u
t
w
e
i
gh
ts
j
i
a
,
a
s
w
ell
a
s
b
ias,
j
i
b
(f
o
r
i
=1
,
2
.
..
,
N
wher
e
as
f
or
j
=
1
,
2,
3
)
o
f
hi
d
den
ne
ur
on
s.
T
ake
i
n
to
a
c
c
o
un
t
t
h
at
j
i
a
a
nd
j
i
b
a
r
e
p
ar
a
m
e
t
er
s
of
m
e
m
be
r
s
hip fu
n
c
ti
o
n
f
or
S
I
R
M
,
j
i
A
.
The
w
e
igh
t
s
a
r
e
gene
r
a
ted
base
d
on
D
,
wh
ere
D
i
s
u
n
i
f
orm
d
i
st
ri
buti
on
f
u
n
c
t
i
o
n
t
h
a
t
r
an
do
mly
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t
J
P
o
w
Elec
&
D
r
i
S
y
st
I
S
S
N
:
2088-
86
94
An EL
M-base
d
sin
g
le
in
pu
t ru
le
mod
u
le a
n
d
i
t
s ap
p
lic
a
t
i
on
in p
o
w
e
r
ge
ne
r
a
ti
on (
C
ho
n
g
Tak
Ya
w)
36
1
ge
ner
a
t
e
a
num
ber
betw
e
e
n
0
t
o 1,
α
a
nd
ω
a
r
e
t
he
p
a
r
am
eter
s.
B
y
def
a
ul
t,
α
=
2
,
ω
=
1
.
A
s
t
h
e
r
e
s
u
l
t,
t
h
e
j
i
a
an
d
j
i
b
a
r
e
i
n
t
he
r
ange
o
f
-
1
t
o
+
1
.
F
i
gur
e
2.
O
ver
v
i
e
w
o
f
S
I
R
M-
ELM.
(
a
)
G
e
ner
a
l
of
S
I
R
M-
ELM
m
ode
l;
(
b
)
G
e
ner
a
l
de
t
a
ils
f
or
e
a
c
h
hi
d
den
ne
ur
on.
Step
2
:
Fo
r
th
e
t
r
aini
ng
p
a
i
r
(
x
pi
,
t
p
)
wh
ere
x
pi
i
s
i
th
f
e
a
t
ur
e
of
p
th
t
ra
in
ing
p
a
i
r
a
nd
t
p
i
s
ta
rg
e
t
o
u
t
pu
t
(f
o
r
p
=
1
,
2
,
.
..,
P
).
Calcu
la
te the
h
id
de
n
la
ye
r
ou
t
p
u
t
m
a
t
rix
H
base
d o
n
m
em
ber
s
hip fu
n
c
ti
o
n
)
,
(
j
i
pi
A
x
.
Fo
r
sim
p
lic
i
t
y,
t
he
m
em
ber
s
hip
f
u
nct
i
on
ca
n
b
e
d
eno
t
e
d
a
s
j
pi
)}
(
e
xp{
1
1
)
,
,
(
j
i
pi
j
i
j
i
j
i
pi
b
x
a
b
a
x
(
2
)
N
P
PN
PN
PN
P
P
P
P
P
N
N
N
N
N
N
3
3
2
1
2
2
1
2
3
1
2
1
1
1
3
2
2
2
1
2
2
22
1
22
3
21
2
21
1
21
3
1
2
1
1
1
2
12
1
12
3
11
2
11
1
11
..
.
..
.
..
.
H
(
3
)
St
e
p
3
:
Th
e
out
put
w
ei
ght
s
,
β
,
w
e
r
e
co
mp
ut
ed
.
Si
n
c
e
it
i
s
hi
g
h
p
o
s
si
b
ilit
y
th
a
t
H
is
a
n
on-
sym
m
e
try
m
a
trix,
t
h
e
i
n
v
e
rse
m
a
trix
c
a
n
n
o
t
b
e
reso
lv
ed.
To
c
irc
u
mve
n
t
t
h
is
p
r
o
b
l
e
m
,
a
m
oor
e
-
penr
ose
pse
u
d
o
i
n
v
e
r
s
e
m
a
tr
ix
m
e
t
h
od
i
s
u
t
i
l
i
zed,
he
nc
e
w
o
r
k
out
t
h
e
out
p
u
t
w
e
i
g
h
t
s
of
β
b
y
(4
),
T
H
H
H
β
T
T
1
)
(
(
4
)
wh
ere
T
i
s targ
et o
utp
u
t
m
at
r
i
x
,
i.e.
,
T
N
t
t
t
]
...
[
2
1
T
Step
4
:
Afte
r
the
o
u
tp
ut
w
e
i
gh
ts
o
f
SI
RM-ELM
w
e
r
e
c
a
l
c
ula
t
e
d
,
pred
ic
ti
o
n
o
f
a
s
e
t
o
f
n
e
w
a
n
d
un
la
be
l
e
d
sa
mple
s
z
ca
n
be
c
om
pute
d
,
i
.
e.
,
(.)
i
s
t
h
e
me
mb
e
r
s
h
i
p
f
un
ct
io
n
,
h
i
s
the
h
i
dde
n
l
a
yer
w
h
er
e
by
y
is
t
he
p
r
e
dic
t
i
o
n
ou
t
p
u
t
.
)}
(
ex
p
{
1
1
)
,
,
(
j
i
qi
j
i
j
i
j
i
qi
b
z
a
b
a
z
(
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-
8694
I
nt
J
P
ow
Elec
& Dr
i
S
y
st V
ol.
11,
N
o.
1
, Ma
r
202
0
:
359
–
36
6
36
2
N
Q
QN
QN
QN
Q
Q
Q
Q
Q
N
N
N
N
N
N
3
3
2
1
2
2
1
2
3
1
2
1
1
1
3
2
2
2
1
2
2
22
1
22
3
21
2
21
1
21
3
1
2
1
1
1
2
12
1
12
3
11
2
11
1
11
...
...
...
h
(
6
)
hβ
y
(7
)
wher
e
q
=
1
,
2
,
.
.
.
.
Q
a
nd
Q
is num
ber
of
t
e
s
t
sam
p
l
e
s.
St
e
p
5
:
Aft
e
r
co
mp
ut
e
t
h
e
o
u
tp
ut
o
f
ELM
fo
r
t
e
st
i
n
g
sa
mpl
e
s
,
c
al
cul
a
t
e
th
e
roo
t
m
ea
n
sq
ua
red
er
ror
(RMSE),
i.e.,
Q
d
y
RMSE
Q
q
q
q
test
1
2
)
(
(
8
)
wher
e
y
q
a
nd
d
q
w
ere
pre
d
i
c
t
i
o
n
a
n
d
a
c
t
ua
l
ou
tpu
t
r
espec
tive
to
z
q
.
F
l
ow
char
t
s
w
ere
deli
nea
t
e
d
i
n
F
i
g
u
r
e
3
and
F
i
g
u
re
4
to
sim
p
lify
t
h
e
proc
ed
ure
s
t
ake
n
by ste
p
w
i
se tr
a
in
ing
pr
ot
o
c
ols.
F
i
gure
3.
F
low
c
hart t
ha
t
repr
esen
ts
t
h
e
s
t
ep
1
t
o st
ep 3.
F
i
gure
4. F
low
c
har
t
t
hat re
prese
n
t
s
the
s
t
e
p
4
3.
RESULT
S
A
N
D
DISCU
SSIO
N
Th
e
ap
pli
c
ab
ili
t
y
o
f
th
e
SIR
M
-ELM
m
od
e
l
w
as
i
n
v
e
st
ig
at
ed
i
n
th
i
s
sect
io
n.
F
our
b
e
n
chm
a
rk
regr
ession
da
t
a
sets
from
the
U
C
I
m
a
c
h
in
e
re
posit
ory
(e.g.
A
b
a
l
o
n
e
,
Ba
l
l
o
on,
S
trike
and
S
p
ace
-ga)
w
ere
ut
iliz
e
d
f
or
p
e
rforma
n
ce
e
va
lua
t
io
n
of
S
IR
M-EL
M.
O
nly
A
d
d
i
ct
ive
S
ig
moid
h
id
de
n
n
e
uro
n
(
S
i
g
A
ct)
w
a
s
ut
iliz
e
d
i
n t
h
e anal
ys
is. A
ll a
n
al
ys
is w
er
e ru
n on
a perso
na
l
c
omp
u
t
er
e
qu
i
ppe
d w
i
t
h
In
t
el
(R) Core
(TM) i
7
2.
9
G
H
z
C
P
U
a
n
d
8
G
R
A
M
u
s
i
n
g
M
A
T
L
A
B
(
v
e
r
.
2
0
1
0
)
,
a
s
d
e
t
a
i
l
e
d
i
n
T
a
b
l
e
1
.
Ta
bl
e
2
l
i
st
ed
t
h
e
d
at
a
s
e
t
s
spec
ific
a
tio
ns
u
se
d in t
he
e
x
p
e
r
i
m
e
nts.
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
An
ELM-b
a
s
e
d
si
ng
l
e
inp
u
t
rul
e
mo
du
l
e
and
i
t
s
a
ppl
i
c
at
ion
i
n
powe
r
g
e
n
e
ra
ti
o
n
(Ch
ong
T
a
k
Y
a
w)
36
3
Ta
ble
1.
S
pe
cifica
ti
on o
f
per
sona
l
com
p
u
t
e
r
and s
o
ftw
a
re
p
acka
g
es u
til
i
zed
for
expe
r
i
m
e
nt
s a
ndc
om
paris
on.
Item
s
S
p
ec
i
f
ica
tion
P
e
rsona
l C
o
m
put
er
A
sus
Ope
r
a
ting
S
y
s
t
em
s
Wi
nd
ow
s
8.
1
C
P
U
Inte
l(R
)
C
or
e
(
T
M
)
i7 2.5GHz
RA
M
8
G
B
S
o
f
t
wa
r
e
M
a
t
l
a
b
7.
11.
0.
584
(R
2010b)
P
r
og
ram
m
ing
La
n
gua
ge
M
a
t
l
a
b
L
a
ngu
a
g
e
Ta
ble
2.
S
pe
cifica
ti
on o
f
be
n
c
h
ma
rk re
g
ression da
t
a
se
ts.
Data
s
e
ts
#
A
ttribute
s
#
T
r
a
i
ning
Sam
p
l
e
s
#
Te
st
ing
S
a
m
p
l
e
s
#
T
o
t
a
l
Sa
m
p
le
s
Ab
al
o
n
e
8
3000
1177
4177
Ba
lloon
2
1334
667
2001
St
r
i
ke
6
416
209
625
Spa
c
e
-
g
a
6
2071
1036
3107
I
n
a
l
l
e
xper
i
m
e
nt
s,
f
our
b
e
n
c
h
ma
rk
r
egr
e
ssion
d
a
tase
t
s
w
ith
t
ra
i
n
in
g
a
nd
va
li
dat
i
on
sa
mples
w
e
r
e
eva
l
ua
t
e
d
usi
n
g
the
t
r
ai
n-val
i
dat
i
on-
t
e
st
m
et
ho
d
a
s
s
u
g
g
e
s
t
e
d
b
y
l
i
tera
t
u
re
[
1].
The
n
u
m
ber
of
m
em
ber
s
hip
fu
nc
ti
o
n
o
f
an in
put a
t
t
ri
bute
is
teste
d for 1, 2
o
r 3, (i.e
.
,
j
= 1, 2, 3
)
for
a
l
l th
e
r
e
gre
s
si
on da
ta
sets. In
a
d
d
it
io
n,
the
RMS
E
i
s
b
a
sed
on
d
e
f
a
u
l
t
r
ange
f
or
j
i
a
a
n
d
j
i
b
for
a
ll
ru
les
(i.
e
.,
i
=
1,
2
.
..,3
N
).
N
ote
t
h
at
i
n
S
I
RM-ELM,
the
numbe
r
o
f
f
uz
zy
r
u
l
e
w
a
s
e
q
u
i
v
a
l
e
n
t
t
o
num
ber
o
f
h
i
dde
n
ne
u
ro
n
o
f
E
LM
.
F
o
r
ea
c
h
d
at
ase
t
,
t
h
e
expe
r
i
me
n
t
s
w
e
re
cond
uc
t
e
d for
50
tim
es w
ith r
an
dom
j
i
a
and
j
i
b
a
nd the
aver
a
g
e
results ar
e
r
e
c
orde
d.
The
resul
t
s
of
p
roposed SI
R
M-EL
M
were
a
l
s
o com
p
are
d
to re
s
u
lts
of other ELM-bas
ed
me
t
hod
s
.
A
s
seen
f
rom
Ta
ble
3,
t
he
R
M
S
E
of
S
IR
M-ELM
a
re
b
et
t
e
r
w
h
e
n
c
om
par
e
w
i
t
h
O
S
-
E
L
M
[
2
1
]
,
S
V
M
[
2
1
]
a
n
d
ELM
[
1].
Note
t
hat
S
I
RM-ELM
p
er
fo
rm
b
etter
than
O
S
-
ELM
f
or
A
bal
one
d
a
t
as
e
t
a
s
it
has
o
n
l
y
o
n
e
para
me
ters
a
s c
o
mpa
r
ed
t
o
O
S
-
ELM
t
hat ha
s
th
r
e
e
par
a
m
e
ters.
Table
3.
R
MS
E of
S
IRM-ELM,
ELM [
1]
,
SV
M [2
1 ]
and O
S
-ELM
[
2
1
]
A
l
gorithm
A
b
a
l
on
e
Ba
ll
oon
Strike
S
pa
ce
-
g
a
RM
S
E
R
MS
E
RMS
E
R
M
S
E
S
I
R
M
-
E
LM
0
.
07598
0
.
0443
2
0.
2656
0
.
0359
1
O
S
-E
L
M
[
21]
0
.
0771
-
-
-
S
V
M
[21]
0
.
0764
0.
059
0
.
2282
0
.
0648
EL
M
[1]
0.
0761
0.
0553
0
.
2985
0
.
0624
4.
NO
x
EMIS
S
IO
N
OF POWE
R
G
E
N
ERATIO
N
PL
A
N
T
Ni
t
r
og
en
o
cc
urred
n
a
tu
ra
lly
i
n
t
h
e
at
mo
sph
e
re
a
s
an
i
na
c
t
iv
e
g
a
s.
I
n
add
i
ti
on
,
ou
r
a
t
mo
sph
e
re
c
o
nt
ai
ns
j
u
s
t
a
b
out
7
8
%
N
2
by
v
o
l
u
m
e
i
n
the
a
i
r
.
T
he
N
O
x
w
as
r
e
f
er
ri
ng
to
n
i
t
ro
ge
n
oxi
de
s
but
m
os
tly
inc
l
ude
n
i
t
r
oge
n
mo
n
o
x
i
de,
al
so
i
de
nt
ifie
d
a
s
n
i
t
ric
o
x
i
d
e,
N
O
as
w
e
ll
as
n
i
t
ro
ge
n
di
o
x
ide
,
N
O
2
.
Ther
e
wer
e
als
o
o
ther
s
i
n
t
he
f
am
i
l
y
l
i
ke
l
au
g
h
i
n
g
ga
s
(kn
o
w
n
a
s
n
i
t
r
ous
o
xi
de,
N
2
O
)
,
nit
r
o
g
e
n
p
e
n
to
xi
de
(
N
2
O
5
)
and
ni
t
r
oge
n tetr
ox
ide
(N
2
O
4
).
Th
e
p
r
esen
c
e
o
f
NO
x
i
n
the
atm
o
sphe
re
p
ose
d
d
ire
c
t
a
n
d
i
nd
ire
c
t
e
f
fe
cts
o
n
h
um
an
h
ea
lt
h
an
d
ec
osystem
s
,
i.e.
a
nima
l
s
a
n
d
p
lan
t
s,
i
n
t
h
e
en
vir
onm
en
t.
N
O
x
r
e
a
c
t
e
d
w
i
t
h
co
mp
on
en
ts
s
uch
as
w
at
er,
o
x
y
g
en
and
o
t
he
r
c
h
e
m
i
c
als
t
o
f
orm
smog
a
n
d
ac
i
d
ic
p
o
l
l
u
tan
t
s
w
h
ic
h
le
ad
s
to
t
he
f
orm
a
ti
o
n
of
a
c
i
d
ra
in.
In
t
urn,
ac
i
d
r
ai
n,
toge
t
h
er
w
it
h
dry de
posi
t
i
o
n
an
d
c
l
o
u
d
, m
ay
c
a
u
se
da
ma
ges a
nd deter
i
or
a
tio
n
to
c
a
r
s a
nd bu
i
l
d
i
ngs.
NO
x
i
s
ma
inl
y
r
elea
se
d
duri
n
g
com
b
ust
i
o
n
proc
es
s
of
f
o
ssi
l
fue
l
s
l
ike
coa
l
,
oi
l
an
d
nat
u
ra
l
ga
s.
Ac
co
rd
ing
to
E
u
r
op
e
a
n
Envi
ro
n
m
e
n
t
Ag
ency
(
EEA)
t
e
chn
i
cal
r
e
p
o
r
t
(
19
90
-
201
3
)
,
2
1
%
o
f
t
h
e
NO
x
g
a
s
e
m
i
s
sio
n
s
i
n
E
u
r
op
e
a
n
U
ni
on
w
e
r
e
f
r
o
m
t
he
e
n
e
rgy
p
r
odu
ct
io
n
an
d
d
is
tr
ibut
i
on,
w
h
i
c
h
w
a
s
a
ppr
o
x
im
atel
y
1
,
6
0
0
ki
l
o
t
onne
[
23
,
24
].
H
o
w
ev
e
r
,
t
h
e
g
r
owt
h
o
f
po
wer
g
e
n
e
rat
i
on
i
ndu
st
ri
e
s
w
a
s
e
xp
ec
t
e
d
to
b
e
i
n
cre
a
s
i
ng
by 18.
7 g
i
gaw
a
t
t
s
(
G
W) in
t
h
e com
i
ng yea
r
s
,
201
6
- 20
1
8
, due t
o
p
ri
c
e
and
a
v
ail
a
bil
i
t
y
of
n
a
t
u
r
al
g
as.
He
n
c
e,
pred
ic
ti
on o
f
N
O
x
e
m
i
ssi
on is
v
i
t
a
l
to
t
h
e
p
o
w
er
gene
r
atio
n
sec
t
or a
nd i
t
shal
l
n
o
t be
t
ake
n
lig
ht
l
y
.
I
n
c
a
s
e
o
f
a
pp
lica
t
i
on,
t
he
N
O
x
e
m
i
ssi
on
o
f
an
o
p
e
n
c
y
cl
e
g
a
s
t
u
rbin
e
in
a
p
o
w
er
g
en
e
r
at
io
n
pl
a
nt
(loca
t
e
d
i
n
P
o
rt
D
i
c
ks
on,
M
a
l
ay
si
a
)
h
as
b
e
e
n
inve
s
t
i
g
at
ed
[
2
5
]
.
T
h
e
o
b
j
ect
i
v
e
was
t
o
d
ev
e
l
op
a
n
eu
ra
l
netw
ork
m
o
de
l
f
o
r
pre
d
i
c
t
i
o
n
o
f
N
O
x
e
m
i
s
s
i
o
n
.
T
h
e
r
e
a
r
e
1
5
0
i
n
p
u
t
a
t
t
r
i
b
u
t
e
s
t
a
k
e
n
f
r
o
m
t
h
e
p
a
r
a
m
e
t
ers
o
f
the
p
o
w
e
r
ge
n
e
rati
on
p
la
nt
s
uc
h
a
s
t
he
l
oad
i
n
g
o
f
the
g
a
s
t
u
rb
i
n
e
,
tem
p
era
t
u
r
e,
p
re
ssur
e
a
nd
e
t
c
.
The
ta
rgete
d
ou
tpu
t
i
s
the
q
u
an
t
ity
o
f
N
O
x
(
in
p
pm)
emissi
o
n
f
rom
the
ga
s turb
ine.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SSN: 2088-
8694
I
nt
J
P
ow
Elec
& Dr
i
S
y
st V
ol.
11,
N
o.
1
, Ma
r
202
0
:
359
–
36
6
36
4
A
tota
l
o
f
3
,4
05
data
s
am
p
l
e
s
h
ave
be
en
c
olle
c
t
e
d
f
or
t
r
a
in
ing
a
nd
t
e
s
t
i
ng
o
f
S
IRM-E
L
M.
O
ut
o
f
3,40
5
da
ta
s
a
m
ple
s
,
2,
2
70
w
e
re
u
sed
f
o
r
trai
nin
g
w
h
i
le
t
he
r
em
aini
ng
1
,
1
3
5
w
ere
us
ed
f
o
r
t
e
s
t
i
n
g
.
T
he
num
b
e
r
of
m
e
m
b
ersh
ip
f
uncti
on
o
f
an
i
np
ut
a
tt
rib
u
t
e
w
as
t
e
ste
d
f
or
1
,
2
or
3
,
(i.
e
.,
j
=
1
,
2,
3
)
an
d
the
r
e
sul
t
s
are
s
how
n
in Ta
b
le
4.
Base
d
o
n
t
he
r
e
s
u
lts
on
t
h
e
T
a
ble
4,
t
he
j
i
a
and
j
i
b
w
a
r
e
i
n
d
e
f
a
u
l
t
s
e
t
t
i
n
g
(
i
n
S
t
e
p
1
)
.
A
f
t
e
r
s
e
t
t
h
e
numbe
r
of
m
em
bershi
p
fu
nct
i
on
of
a
n
i
n
pu
t
att
r
i
b
u
t
e
a
s
1
,
in
o
r
de
r
to
g
e
t
t
he
l
ow
est
ro
o
t
m
ea
n
squa
re
d
er
ror
(RMSE),
t
h
e
j
i
a
a
n
d
j
i
b
nee
d
t
o
be
t
u
n
ed
i
n
d
i
ffe
re
nt
r
a
nges.
T
he
c
om
ple
t
e
tun
i
ng
r
esu
l
t
s
a
r
e
r
e
c
o
r
d
e
d
i
n
Table
5
.
Th
e
be
st
RM
S
E i
n
Ta
b
l
e
5
is 0
.
0
286
47
.
Ta
b
l
e
4.
Resu
lts
f
or
N
O
x
E
mission
o
f
S
IRM-ELM us
i
ng d
i
ffe
r
ence
s of
n
u
m
be
r of
me
mb
e
r
s
h
i
p
f
un
c
t
i
o
n
.
#
N
u
m
b
e
r
o
f
m
e
mbe
r
ship
f
unc
ti
on of
a
n
i
nput a
tt
r
i
but
e
RM
SE
1
0.
0303
58
2
0.
0564
54
3
0.
8051
05
Ta
bl
e 5
.
R
esul
ts fo
r
NO
x
Em
i
ssi
on o
f
S
IRM-E
L
M
us
ing
d
i
f
f
ere
n
t ra
n
g
es o
f
w
e
ig
hts.
Ran
g
e
RM
S
E
j
i
a
j
i
b
-1 to +
1
-1
t
o
+
1
0.
0303
58
-2 to 2
-1
t
o
+
1
0.
0325
02
0
to
+
1
-1
t
o
+
1
0.
0317
03
-1 to 0
-1
t
o
+
1
0.
0311
73
-1 to +
1
0
to
+
1
0.
0338
23
-1 to +
1
-1
t
o
0
0.
0332
94
-1 to +
1
0.
5
to
+
1
0.
0310
32
-1 to +
1
0.
5
0.
0286
47
I
n
t
h
e
e
x
p
e
r
i
m
e
n
t
o
f
u
s
i
n
g
E
L
M
,
2
/
3
o
f
t
h
e
d
a
t
a
s
a
m
p
l
e
s
w
e
r
e
u
t
il
i
zed
f
or
t
ra
in
in
g
w
h
ile
t
he
rem
a
ini
n
g
1
/
3 w
e
re
u
t
iliz
e
d
t
o veri
fy
t
he
m
os
t
s
u
i
t
a
b
le
n
u
m
ber
of
n
e
u
r
ons of t
h
e
pa
re
n
t
E
LM
(
i.e
.
,
L
)
t
h
ro
ugh
a
va
l
i
da
ti
on
p
r
oc
ess.
F
or
s
igm
o
id
a
c
t
i
v
a
t
i
o
n
fu
nc
ti
o
n
o
f
ELM,
t
rai
n
in
g
an
d
va
li
dat
i
on
proc
esse
s
s
t
art
b
y
set
tin
g
L
=
5
0 un
i
t
s an
d the
n
incr
ease
d
b
y a
n
i
ncr
e
m
e
nt of 50 uni
ts. A
s
a
n
exa
mple
, Tab
l
e
6 sh
ow
s
t
h
e t
e
stin
g
proce
s
se
s
ba
se
d
o
n
s
i
g
mo
id
a
cti
v
a
t
ion
func
ti
on.
B
ase
d
on
t
h
e
re
sults
o
f
R
M
S
E
i
n
F
i
gur
e
5,
t
he
b
e
s
t
R
M
S
E
i
s
0
.
0
270
86
.
Usin
g
th
e
resu
lt
i
n
F
i
gu
re
5
t
o
comp
a
r
e
wi
t
h
T
a
b
l
e
5
,
t
h
e
R
M
S
E
o
f
E
L
M
i
s
l
o
w
e
r
t
h
a
n
R
M
S
E
o
f
S
I
RM-ELM
d
ue
to the
com
p
le
xi
ty
o
f h
i
dde
n
neur
ons i
n
EL
M
.
F
i
gur
e 5.
RMS
E
o
f
N
O
x
em
i
ssion
for
ELM
.
Evaluation Warning : The document was created with Spire.PDF for Python.
Int J
P
o
w
E
l
e
c
&
D
ri S
yst
IS
S
N
:
2088-
86
94
An
ELM-b
a
s
e
d
si
ng
l
e
inp
u
t
rul
e
mo
du
l
e
and
i
t
s
a
ppl
i
c
at
ion
i
n
powe
r
g
e
n
e
ra
ti
o
n
(Ch
ong
T
a
k
Y
a
w)
36
5
4.
CONCL
U
S
ION
I
n
e
ssence
,
t
his
pa
per
pre
s
e
n
t
e
d
a
fram
e
w
o
r
k
o
f
Ex
t
r
em
e
Lea
r
ni
n
g
M
ach
in
e
with
S
i
ngl
e
In
put
R
ul
e
Mo
du
le
,
whic
h
was
deem
ed
a
s
i
g
n
i
fica
n
t
i
nn
o
v
at
i
on
i
n
E
LM
i
de
o
l
o
gy
(
here
a
fter
d
enoted
a
s
S
I
RM
-
ELM)
.
A
d
o
p
t
i
ng S
i
n
g
l
e
I
npu
t Ru
le Mod
u
l
e i
n
t
he
ELM
h
i
d
d
e
n la
ye
r c
a
n b
e a
g
o
od a
l
t
e
rnat
i
v
e to t
he c
omm
o
n
l
y use
d
ac
t
i
va
t
i
o
n
f
u
n
c
t
i
o
n,
i
.e.
,
S
i
g
m
o
id
(
S
i
gA
c
t
)
.
S
I
R
M-EL
M
has
bee
n
t
e
st
e
d
w
i
t
h
s
ig
moid
a
c
tiv
at
ion
fun
c
ti
o
n
s
ut
iliz
i
n
g
be
nc
hm
ar
k
regre
ssi
o
n
d
a
t
a
s
e
t
s,
i
nc
l
u
sive
o
f
A
b
al
one,
B
a
l
l
oo
n,
S
trike
a
n
d
S
p
ace
-ga.
T
h
e
expe
r
i
me
n
t
a
l
r
esu
lts
d
em
ons
tr
ated
t
ha
t
o
u
r
pro
pose
d
m
od
e
l
w
as
m
ore
super
i
or
c
om
pa
re
d
to
O
S
-
ELM
[
21
]
,
S
V
M
[21]
a
n
d
ELM
[
1],
as
s
how
n
in
T
ab
le
2
.
A
s
f
or
r
ea
l
w
o
rld
a
p
plica
tio
n,
t
he
i
m
p
le
me
nt
a
t
i
o
n
of
S
I
R
M-
ELM in the
pre
d
i
c
tio
n of
N
O
x
em
i
t
ted in
po
w
er
g
ene
r
ati
o
n
pla
n
t w
i
t
h
low
RM
S
E
su
g
g
es
ted pro
p
o
sed
met
h
od
is
a
p
p
l
i
ca
ble
i
n
power
gene
r
at
io
n.
ACKNOW
LEDG
E
MEN
T
S
Th
is
w
or
k
w
a
s
su
p
por
ted
b
y
U
n
i
ver
s
i
t
i
T
e
naga
N
asi
o
na
l
(J5
100
5
0
6
84)
a
n
d
X
i
a
me
n
U
n
i
v
ersi
ty
Ma
lays
ia (
IECE/00
0
1
).
REFE
RENCES
[1]
G.
B
.
Hu
a
n
g,
Q
.Y.
Zh
u
and
C.
K
.
Siew
,
"E
x
t
reme
l
earn
i
ng
m
ach
in
e
:
a
new
l
earni
ng
sch
e
m
e
o
f
f
e
edf
o
rward
neu
r
al
net
w
ork
s
,
"
IEEE
In
ter
nati
onal Joi
n
t
Co
nfer
ence
on
Neur
al
Net
w
o
r
ks
(IJ
CNN200
4),
vo
l.
2
,
Bu
d
a
pes
t
,
Hu
ng
a
r
y
,
pp
.
9
85-9
9
0
,
J
u
l
2
0
04.
[2]
K.
S
.
Y
a
p
and
H.J.
Y
ap
,
"
D
ail
y
m
axi
m
u
m
l
o
a
d
f
o
recas
ti
ng
of
c
ons
ecut
i
v
e
n
ati
o
n
a
l
ho
li
days
u
sin
g
O
S
E
LM
-bas
ed
mu
l
t
i-
a
g
e
n
ts
s
y
s
t
e
m
w
i
th
a
ve
r
a
g
e
s
t
r
a
t
e
g
y
,
"
Ne
ur
oc
ompu
tin
g
, vo
l
.
81
, pp
.
1
08
-1
12
,
2
0
1
2
.
[3]
G.
H
uang
,
G.
B
.
Hu
ang
,
S
.
S
o
ng
an
d
K.
Y
ou
,
"Tren
d
s
i
n
e
xt
rem
e
l
earnin
g
m
achi
n
e:
A
r
ev
iew
,
"
Neur
al
Net
w
or
k
s
,
vo
l.
61,
p
p
.
3
2-48,
2
01
5.
[4]
M
.
V
.
Hees
wijk
a
nd
Y.
M
i
c
he,
"Bin
ary/
ternary
ext
r
em
e
learni
ng
m
achin
e,"
Neur
ocom
p
u
t
i
n
g
,
vo
l.
149,
pp
.
1
87-1
9
7
,
2
0
1
5.
[5]
Y.
I
m
a
m
v
erdi
ye
v
an
d
L.
S
uk
ho
stst,
"An
o
m
a
ly
d
etect
io
n
in
n
etwo
rk
t
raffic
usin
g
ext
r
em
e
learn
i
ng
m
ach
in
e,"
IE
E
E
10
th
Int
e
rn
a
tional
Co
nf
erence
on
Ap
plica
t
i
o
n
of
In
formation
a
nd
Com
m
u
n
ica
t
i
o
n
T
echn
o
l
ogi
es (
A
ICT)
,
pp
.
1
4-16
,
2
0
1
6
.
[6]
G.
B
.
H
u
ang
,
H
.
Z
h
o
u
,
X.
D
in
g,
a
n
d
R
.
Zh
ang
,
"
E
x
t
r
eme
l
earnin
g
m
achi
n
e
f
o
r
reg
r
es
si
on
a
n
d
m
ul
ti
c
l
as
s
clas
sificati
o
n
"
,
I
E
EE
Tran
sa
c
tio
ns
o
n
Sy
ste
ms, M
a
n, an
d Cy
be
rn
e
t
ic
s, Pa
rt
B (
C
y
b
e
r
ne
t
i
c
s
)
,
vo
l.
4
2,
n
o
.
2
,
pp
.
5
13-5
2
9
,
2
0
1
2.
[7]
G
.
H
u
a
n
g
,
Z
.
B
a
i
,
L
.
K
a
s
u
n
,
a
n
d
C
.
V
o
n
g
,
"
L
o
c
a
l
r
e
c
e
p
t
i
v
e
b
a
s
e
d
extrem
e
l
earni
ng
m
achi
n
e",
IE
EE Com
put
atio
na
l
Int
e
ll
igen
ce M
a
g
a
z
in
e,
v
o
l.
1
0
,
no.
2
,
201
5.
[8]
J.
T
ang
,
C
.
Den
g
,
and
G.-B.
H
u
ang
,
"
Ext
r
em
e
learni
ng
m
ach
in
e
f
o
r
multil
a
yer
percep
tron
,
"
I
E
EE
transactions
on
n
e
ural ne
two
r
k
s
an
d le
arn
i
ng
sy
s
t
e
ms,
vol.
27
,
no.
4
,
p
p
.
8
0
9
-
82
1,
2
01
6.
[9]
M
.
V
.
H
eeswijk
,
Y.
M
i
c
he,
E.
O
ja
a
n
d
A
.
Len
d
asse,
"GP
U
-accel
erat
ed
a
n
d
p
aral
leli
zed
E
L
M
e
nsem
b
l
es
f
o
r
l
arge-
scal
e regress
i
o
n
,"
Ne
uroc
ompu
ti
ng
,
vo
l
.
74
,
no
. 1
6, pp
.
24
3
0
-
24
3
7
, 20
1
1
.
[10]
C.D.
L
i
,
L
.
J
.
L
.
Gao
,
J
.
Q
.
Yi
a
nd
G
.Q.
Zh
ang
,
"
Analy
s
is
a
nd
d
esi
gn
o
f
f
u
nctionall
y
wei
ghted
s
i
ngle-i
n
p
u
t-ru
le
-
m
o
d
u
les con
n
ect
ed f
u
zzy i
n
f
erence s
y
s
t
ems
,
"
IEEE
Trans. on
F
u
zzy System
s
,
vol.
26,
n
o
.
1
,
pp
.
5
6
-7
1,
2
01
8.
[11]
N.
Y
u
b
azak
i,
J
.
Yi
a
n
d
K
.
H
i
rota,
"
S
IRM
s
(
S
i
n
g
l
e
I
np
ut
R
ule
Mod
ules
)
co
nn
ec
t
e
d
f
u
zzy
i
n
f
erenc
e
m
odel
,
"
J. Ad
v
.
Comput. Intell.
I
n
telli
gent Inf.
,
v
o
l.
1
,
pp
. 2
3-3
0
,
1
99
7.
[12]
J.
Y
i,
N
.
Yub
azaki
an
d
K.
H
iro
t
a,
"
A
pro
p
o
s
al
o
f
S
I
RM
s
dyn
amica
l
l
y
co
nnect
ed
f
uzzy
i
n
f
eren
ce
m
o
del
f
o
r
plu
r
al
input
f
u
zzy contr
ol,
"
Fu
zz
y
Se
ts
S
y
st
.,
vol.
125
,
pp.
7
9
-
9
2
,
2
0
0
2
.
[13]
J.
Y
i,
N
.
Yubazaki
and
K
.
H
irot
a,
"
Anti
-
swing
a
nd
p
os
itioning
c
o
n
tro
l
o
f
o
v
erhead
t
rav
e
lin
g
cr
ane,
"
Inf
.
Sci
.,
vo
l.
155
,
p
p
.
19
-42,
2
0
0
2
.
[14]
J.
Y
i
,
N
.
Yu
bazaki
an
d
K.
H
i
r
ot
a
,
"
S
t
ab
il
ization
co
nt
rol
o
f
s
er
i
e
s
ty
pe
dou
bl
e
i
n
v
e
rt
ed
p
en
du
lum
syst
ems
u
s
i
n
g
t
h
e
S
I
RMs d
ynami
call
y
co
nnect
ed fu
zzy i
nf
eren
c
e
m
odel
,
"
Artif
. In
tell.
Eng
., v
ol
. 1
5
,
p
p.
29
7
-
30
8,
2
0
0
1
.
[15]
J.
Y
i,
N
.
Yub
azaki
and
K.
H
iro
t
a,
"
Up
sw
in
g
an
d
s
t
abili
zati
on
co
nt
rol
o
f
i
nv
erte
d
pend
ul
um
s
ystem
bas
e
d
o
n
t
he
S
I
RMs d
ynami
call
y
co
nnect
ed fu
zzy i
nf
eren
c
e
m
odel
,
"
F
u
zz
y
Se
t
s
S
y
st
.,
v
o
l
.
1
2
2,
p
p
.
1
39-1
5
2
, 2
00
1.
[16]
J.
Y
i,
N
.
Yu
ba
zak
i
an
d
K
.
H
iro
t
a,
"
A
new
fuzzy
c
on
tro
ller
f
o
r
s
tab
i
lizati
on
of
p
arall
e
l-t
ype
d
o
ubl
e
i
nverte
d
pen
d
u
l
um
s
ystem
,
"
Fu
zz
y
Se
ts
S
y
st
.
,
v
o
l
.
12
6,
pp
. 10
5
-1
19
,
20
02
.
[17]
H.
S
eki
and
M
.
M
uzu
m
oto,
"
S
I
RM
s
con
n
ected
f
u
zzy
i
n
f
eren
ce
m
e
th
o
d
a
d
op
ting
e
mph
a
si
s
a
n
d
su
pp
re
ssion
,
"
F
u
zzy
Set
s
a
nd Sys
t
em,
v
o
l
.
21
5
,
p
p
.
112-1
26,
2
0
0
3
.
[18]
C.
D
.
L
i
,
L.
W
a
n
g
,
G
.
Q.
Z
h
a
ng
,
H.
D
.
W
a
n
g
a
n
d
F
.
Sh
ang
,
"
Fu
n
c
t
io
nal
t
y
p
e
s
ingle
in
put
r
ul
e
mo
du
les
con
n
ect
ed
neu
r
al
f
u
zzy
s
y
s
tem
f
o
r
w
i
nd
sp
eed
p
redi
ction,
"
IEEE/CAA Jo
ur
nal
of A
u
t
o
m
a
ti
ca
S
i
nica
,
vo
l
.
4
,
n
o
.
4
,
pp
.
7
51-7
6
2
,
2
0
1
7.
[19]
H.
M
iy
ajim
a,
K
.
K
i
s
h
i
d
a,
N
.
S
h
igei
a
n
d
H
.
Miyaj
i
m
a
,
"Learni
ng
a
l
gorith
ms
f
or
f
uzzy
i
n
f
erence
sys
t
ems
co
mp
os
e
d
of
d
ou
b
l
e
an
d
s
i
n
g
le
i
np
u
t
r
u
l
e
mo
du
les,
"
Wo
rl
d Ac
ad
e
m
y
of S
c
ie
n
c
e
,
En
gine
e
r
in
g
an
d Te
c
h
n
o
l
o
g
y
Inte
rna
tion
a
l
Jou
r
n
a
l
o
f
Co
m
p
ut
er a
nd Inf
o
r
m
a
t
i
o
n
Engin
eeri
n
g
, v
o
l
. 1
0,
n
o.
3,
p
p
.
1
,
20
16
.
[20]
H.
M
i
y
a
j
im
a,
N
.
S
h
i
g
ei
a
n
d
H
.
M
i
y
a
jima,
"
S
I
RM
s
f
u
zzy
i
n
f
erence
m
o
d
e
l
w
i
t
h
linear
t
rans
form
ati
o
n
of
i
n
put
vari
ables
and
un
ivers
a
l
app
r
ox
im
atio
n,"
In
:
R
o
jas
I.,
Jo
y
a
G
.,
Catal
a
A
.
(eds)
Ad
van
ces in Com
p
u
t
atio
nal
Int
e
ll
igen
ce. IW
ANN 20
15.
L
ectu
r
e
No
tes in Co
m
p
u
t
er
Sc
i
e
nce
, 90
9
4
. Spr
ing
e
r,
C
ham, 2
01
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N: 2
0
8
8
-
86
94
I
nt
J
P
o
w
E
l
e
c
&
D
r
i
S
yst
V
o
l.
11,
N
o.
1
,
Mar
202
0
:
359
–
36
6
36
6
[21]
T
.
Y
o
s
him
u
ra,
"Di
r
ect
a
d
a
pt
ive fuzzy
b
ack
step
pi
ng
c
on
tro
l
f
o
r
u
ncertai
n
d
is
crete
tim
e
n
o
n
l
i
n
ear
s
ys
tem
u
s
i
ng no
is
y
m
eas
urem
ent
s
,"
In
tern
atio
nal Jour
nal o
f
S
y
st
e
m
s
Scien
c
e
,
v
ol.
48,
n
o.
4
,
201
7.
[22]
N
.
Y
.
L
iang
a
nd
G.
B.
H
uan
g
,
"A
f
ast
an
d
accurate
o
nli
n
e
seq
u
en
ti
al
l
earni
ng
a
l
gorithm
fo
r
feed
f
o
rw
ard
net
w
ork
s
,
"
IE
EE T
r
a
n
s.
on
Neur
al
Networks
,
vol.
1
7
,
n
o
.
6
,
pp.
1
4
11-1
4
2
3
,
2
0
06.
[23]
OECD
,
I
E
A. "
En
erg
y
an
d a
i
r
poll
u
tion
:
wo
r
l
d ener
gy o
u
t
l
o
o
k
s
p
ecia
l
repo
rt
,"
2
01
6.
[24]
A
s
m
e
las
h
,
a
n
d
H
e
no
k
Birh
anu
.
"
P
h
asi
n
g
ou
t
fos
s
il
f
u
el
s
ub
s
i
d
i
es
i
n
t
h
e
G
2
0
:
P
r
o
g
r
e
s
s
,
c
h
a
l
l
e
n
g
e
s
,
a
n
d
w
a
y
s
for
w
a
r
d
,
"
T
h
in
k
Pi
ece.
Gen
eva: In
ter
natio
nal Centr
e
fo
r T
r
ad
e
and S
u
sta
i
n
able Deve
l
o
p
m
en
t
(
I
CTS
D
)
, 20
1
7
.
[25]
I.B.
S
ai
fu
l,
C
.P
.
C
h
en
a
n
d
S
.K.
T
i
o
n
g
,
"
Predic
t
i
on
o
f
NO
x
u
si
ng
supp
ort
vect
or
m
ach
in
e
f
o
r
gas
t
u
rbi
n
e
emissio
n
a
t
p
u
t
r
ajaya
pow
er st
a
ti
on
,"
Jou
r
nal
of Ad
vanced
Science a
n
d
En
gi
ne
er
ing
R
e
s
e
ar
ch
, v
o
l
.
4,
no
.
1
,
pp
.
3
7
-4
6, 2
01
4
.
BIOGRAPHI
E
S
OF
AUT
HORS
Cho
ng
T
a
k
Y
a
w
w
a
s
b
o
rn
i
n
Ku
al
a
Lu
m
pur
on
F
e
b
r
uary
2
1
,
1984
.
He
r
ecei
ved
hi
s
Bachel
ors
deg
r
ee
f
ro
m
U
n
i
v
ersiti
T
enag
a
Nas
i
on
al
(
U
N
ITE
N
),
M
alay
si
a
wit
h
H
ono
rs
i
n
El
ectri
cal
a
nd
El
ec
t
r
onics
E
n
g
i
n
eerin
g
in
2
0
0
8
.
H
e
recei
ved
hi
s
M
a
ster
d
egree
f
r
om
U
ni
versiti
Tenaga
Nas
i
on
al
(
U
N
ITEN),
M
alaysia
wit
h
H
on
ors
in
E
l
ectrical
a
nd
E
l
ect
ron
i
c
s
E
n
g
in
e
e
r
in
g
in
2
01
2.
P
r
evio
usly
h
e
i
s
w
ork
i
ng
a
s
a
P
r
o
j
ect
E
ngin
eer
i
n
th
e
En
e
r
g
y
b
u
s
i
n
e
ss
a
nd
Techn
o
l
ogy
C
ent
r
e
(EBT
E
C
)
of
U
n
i
versiti
Tenaga
N
asi
o
nal
(UNI
TEN)
since
Januar
y
20
09
t
ill
2010.
In
t
he
y
ear
of
20
11
t
i
l
l
2
0
1
2
,
h
e
wo
rked
a
s
an
e
ng
in
eer
i
n
W
i
n
t
rad
Ind
u
s
t
i
r
es
(
s
wi
tc
hb
oa
rd
m
a
n
ufa
c
t
u
r
e
r).
H
e
ju
st
c
o
m
p
l
eted
h
is
P
hD
i
n
201
9
i
n
E
lectri
cal
E
n
g
ineeri
n
g
.
H
is
r
es
ea
rch
in
terest
s
i
n
cl
ude
n
eural
net
w
ork
s
,
su
pport
vecto
r
m
achi
n
es
a
nd
e
xt
reme
l
earni
ng
m
achi
n
e.
S
h
en
Y
uo
ng
W
on
g
receiv
e
d
h
e
r
Bach
el
or
D
eg
ree
and
M
.
S
c
.
o
f
E
lect
rical
a
n
d
E
lectro
n
i
c
Engineering
w
i
t
h
f
i
r
st
c
lass
hono
rs
f
ro
m
Universiti
T
e
naga
N
asi
o
n
a
l
,
M
a
l
a
y
s
i
a
i
n
2
0
1
0
,
a
n
d
2
0
1
2
re
sp
e
c
t
iv
e
l
y
.
S
he
r
e
c
e
i
v
e
d
he
r
Ph
D
d
e
g
r
ee
i
n
En
gine
e
r
in
g
f
rom
th
e
s
a
me
uni
vers
it
y
in
20
15.
H
er
r
esearch
i
nt
erests
i
n
c
lu
de
a
pp
li
cati
on
o
f
a
rtif
ic
ial
intel
l
i
g
ence,
p
att
e
rn
r
ecog
n
i
t
i
o
n
,
f
u
zzy
l
o
g
i
c
a
n
d
E
x
t
re
m
e
L
earni
ng
M
achi
n
es.
Curren
t
l
y
s
he
i
s
an
A
s
s
i
stan
t
P
r
of
ess
o
r
a
t
Dep
a
rt
m
e
nt of Elect
rical
and El
ect
ron
i
cs
En
g
i
n
eeri
ng, X
i
a
m
e
n U
n
i
v
ersity
M
alay
sia, M
alay
si
a.
Keem
S
iah
Yap
receiv
e
d
hi
s
B.
E
n
g
(
Electri
cal)
(Ho
n
s
.
)
and
M
.
S
c
(
E
l
ectri
cal
E
ngineeri
n
g
)
deg
r
ee
b
o
t
h
f
r
om
U
n
i
vers
iti
T
e
kn
ol
ogi
M
al
aysia
i
n
199
8
and
20
00
r
esp
ecti
v
el
y.
I
n
y
ear
2
010
,
h
e
received
P
h
D
i
n
E
lectro
ni
cs
E
n
g
i
n
eerin
g
f
r
o
m
U
ni
versiti
S
a
in
s
M
alay
sia.
H
is
r
e
s
earch
i
nt
e
r
es
ts
in
clud
e
th
eory
a
n
d
a
p
p
li
cati
o
n
s
o
f
art
i
fi
cial
i
nt
e
l
l
i
g
e
n
ce.
C
ur
ren
t
ly,
h
e
i
s
a
P
r
of
esso
r
at
C
olleg
e
of
E
ngin
e
eri
n
g
,
U
niv
e
rsi
t
i
Ten
a
ga
N
asi
onal,
M
al
aysi
a.
Evaluation Warning : The document was created with Spire.PDF for Python.