I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l
E
ng
ineering
a
nd
Co
m
p
u
t
er
Science
Vo
l.
22
,
No
.
3
,
J
u
n
e
202
1
,
p
p
.
161
1
~
161
8
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ee
cs.v
2
2
.i
3
.
pp
161
1
-
161
8
1611
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
A hybrid
deep l
ea
rning
m
o
d
el f
o
r
air qua
lity t
i
m
e ser
ies
prediction
Sa
m
it
B
ha
nja
1
,
Abhi
s
he
k
Da
s
2
1
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
,
G
o
v
e
rn
m
e
n
t
G
e
n
e
ra
l
De
g
re
e
C
o
ll
e
g
e
,
Ho
o
g
h
ly
,
In
d
ia
2
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
,
A
li
a
h
Un
iv
e
rsity
,
Ko
lk
a
ta,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
4
,
2
0
2
0
R
ev
i
s
ed
Ma
y
2
7
,
2
0
2
1
A
cc
ep
ted
J
u
n
13
,
2
0
2
1
A
ir
q
u
a
li
ty
(
m
a
in
l
y
P
M
2
.
5
)
f
o
re
c
a
stin
g
p
lay
s
a
n
i
m
p
o
rtan
t
r
o
le
i
n
t
h
e
e
a
rly
d
e
tec
ti
o
n
a
n
d
c
o
n
tro
l
o
f
a
ir
p
o
ll
u
t
io
n
.
In
re
c
e
n
t
ti
m
e
s,
n
u
m
e
ro
u
s
d
e
e
p
lea
rn
in
g
-
b
a
se
d
m
o
d
e
ls
h
a
v
e
b
e
e
n
p
ro
p
o
se
d
t
o
f
o
re
c
a
st
a
ir
q
u
a
li
ty
m
o
re
a
c
c
u
ra
tel
y
.
T
h
e
su
c
c
e
s
s o
f
th
e
se
d
e
e
p
l
e
a
rn
in
g
m
o
d
e
ls h
e
a
v
il
y
d
e
p
e
n
d
s o
n
th
e
tw
o
k
e
y
f
a
c
to
rs
v
iz.
p
ro
p
e
r
re
p
re
se
n
tatio
n
o
f
th
e
in
p
u
t
d
a
ta
a
n
d
p
r
e
se
rv
a
ti
o
n
o
f
te
m
p
o
ra
l
o
rd
e
r
o
f
th
e
in
p
u
t
d
a
t
a
d
u
ri
n
g
th
e
f
e
a
tu
re
’s
e
x
tra
c
ti
o
n
p
h
a
se
.
He
re
w
e
p
ro
p
o
se
a
h
y
b
rid
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
(HD
NN
)
f
r
a
m
e
w
o
rk
to
fo
re
c
a
st
th
e
P
M
2
.
5
b
y
in
teg
ra
ti
n
g
tw
o
p
o
p
u
lar
d
e
e
p
lea
rn
in
g
a
rc
h
it
e
c
t
u
re
s,
v
iz.
Co
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
(CNN
)
a
n
d
b
i
d
irec
ti
o
n
a
l
lo
n
g
sh
o
rt
-
term
m
e
m
o
r
y
(BD
L
S
T
M
)
n
e
t
w
o
rk
.
H
e
re
w
e
b
u
il
d
a
3
D
i
n
p
u
t
ten
so
r
s
o
th
a
t
CNN
c
a
n
e
x
trac
t
th
e
tren
d
s
a
n
d
sp
a
ti
a
l
f
e
a
tu
re
s
m
o
re
a
c
c
u
ra
tel
y
w
it
h
in
th
e
i
n
p
u
t
w
in
d
o
w
.
He
re
w
e
a
lso
in
tro
d
u
c
e
a
li
n
k
in
g
lay
e
r
b
e
t
w
e
e
n
CNN
a
n
d
BDL
S
T
M
to
m
a
in
tain
th
e
tem
p
o
ra
l
o
rd
e
ri
n
g
o
f
f
e
a
tu
re
v
e
c
to
rs.
In
t
h
e
e
n
d
,
o
u
r
p
ro
p
o
se
d
HD
NN
f
r
a
m
e
w
o
rk
is
c
o
m
p
a
re
d
w
it
h
th
e
sta
te
-
of
-
th
e
-
a
rt
m
o
d
e
ls,
a
n
d
w
e
sh
o
w
th
a
t
HD
NN
o
u
tr
u
n
s
o
t
h
e
r
m
o
d
e
ls i
n
term
s o
f
p
re
d
ictio
n
a
c
c
u
r
a
c
y
.
K
ey
w
o
r
d
s
:
A
ir
q
u
al
it
y
C
NN
Dee
p
lear
n
in
g
L
ST
N
T
im
e
-
s
er
ies
f
o
r
ec
asti
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC
BY
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
A
b
h
is
h
ek
Da
s
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
i
n
ee
r
in
g
A
lia
h
U
n
i
v
er
s
it
y
I
I
A
/2
7
,
Ne
w
T
o
w
n
,
Ko
lk
ata
7
0
0
1
6
0
,
I
n
d
ia
E
m
ail:
ad
as
@
alia
h
.
ac
.
in
1.
I
NT
RO
D
UCT
I
O
N
A
ir
p
o
ll
u
tio
n
is
o
n
e
o
f
th
e
s
er
io
u
s
i
s
s
u
es
d
u
r
in
g
t
h
i
s
u
r
b
an
iz
atio
n
er
a,
an
d
P
M2
.
5
is
o
n
e
o
f
t
h
e
m
o
s
t
i
m
p
o
r
tan
t
p
o
llu
ta
n
ts
t
h
at
in
f
lu
en
ce
air
p
o
llu
tio
n
.
L
o
n
g
-
ti
m
e
ex
p
o
s
er
in
t
h
e
en
v
ir
o
n
m
en
t
w
ith
a
h
ig
h
co
n
ce
n
tr
atio
n
o
f
P
M2
.
5
ca
u
s
es
s
er
io
u
s
p
u
b
lic
h
ea
lth
i
s
s
u
e
s
,
w
h
ic
h
in
cl
u
d
e
asth
m
a,
ch
r
o
n
ic
b
r
o
n
ch
itis
,
a
n
d
h
ea
r
t
d
is
ea
s
e.
So
,
th
e
f
o
r
ec
as
tin
g
o
f
P
M2
.
5
h
as
a
s
ev
er
e
im
p
ac
t
o
n
o
u
r
liv
i
n
g
en
v
ir
o
n
m
en
t
an
d
also
th
e
p
h
y
s
ical
h
ea
l
th
o
f
h
u
m
an
b
ei
n
g
s
.
A
ir
q
u
ali
t
y
d
ep
en
d
s
o
n
v
ar
io
u
s
f
ac
to
r
s
,
th
e
s
e
f
ac
to
r
s
ar
e
v
er
y
m
u
c
h
d
y
n
a
m
ic
an
d
co
m
p
lex
in
n
at
u
r
e.
T
h
ese
f
ac
to
r
s
v
ar
y
o
v
er
ti
m
e.
T
h
u
s
,
air
q
u
alit
y
d
ata
is
ti
m
e
-
s
er
ies
d
ata.
T
im
e
-
s
er
ies
d
ata
is
a
p
r
o
g
r
es
s
io
n
o
f
d
ata
p
o
in
ts
r
ec
o
r
d
ed
in
ti
m
e
r
eq
u
e
s
ts
.
T
i
m
e
s
er
ies
in
f
o
r
m
at
io
n
m
a
y
b
e
u
n
i
v
ar
iate
o
r
m
u
lti
v
ar
iate.
I
n
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ies
(
MT
S)
in
f
o
r
m
atio
n
,
a
b
u
n
c
h
o
f
v
ar
iab
les
m
o
v
e
s
i
n
ti
m
e
o
r
d
er
.
Sin
ce
th
e
air
q
u
ali
t
y
d
ata
ar
e
co
llect
ed
f
r
o
m
v
ar
io
u
s
s
en
s
o
r
s
o
v
er
f
ix
ed
ti
m
e
i
n
ter
v
al
s
,
s
o
air
q
u
al
it
y
f
o
r
ec
asti
n
g
i
s
a
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ies
f
o
r
ec
asti
n
g
(
MT
SF
)
p
r
o
b
lem
.
I
n
th
e
la
s
t d
ec
ad
e,
m
ac
h
in
e
lea
r
n
in
g
ac
h
iev
e
s
g
r
ea
t
s
u
cc
e
s
s
i
n
ti
m
e
-
s
er
ies p
r
ed
ictio
n
,
o
n
l
y
b
ec
au
s
e
o
f
th
e
a
v
ailab
ilit
y
o
f
m
as
s
iv
e
d
ata
s
a
m
p
le
s
a
n
d
s
i
g
n
i
f
ican
t
i
m
p
r
o
v
e
m
e
n
t
o
f
co
m
p
u
ti
n
g
p
o
w
er
.
So
m
e
o
f
th
e
p
o
p
u
lar
m
ac
h
i
n
e
lear
n
in
g
al
g
o
r
ith
m
s
ar
e
s
u
p
p
o
r
t
v
ec
to
r
m
ac
h
in
e
s
(
SVM
)
,
Nai
v
e
B
a
y
es,
k
-
m
ea
n
s
,
a
n
d
r
an
d
o
m
f
o
r
est
.
A
ll
o
f
th
e
s
e
m
ac
h
in
e
lear
n
i
n
g
alg
o
r
ith
m
s
s
u
f
f
er
f
r
o
m
th
e
f
o
llo
w
i
n
g
ch
a
llen
g
es
v
iz.
1
.
lo
n
g
d
esig
n
ti
m
e
2
.
d
o
m
ain
ex
p
er
t
’
s
k
n
o
w
led
g
e
is
r
eq
u
ir
ed
f
o
r
f
ea
tu
r
e
d
esig
n
in
g
.
R
ec
en
tl
y
,
d
ee
p
lear
n
in
g
m
et
h
o
d
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
22
,
No
.
3
,
J
u
n
e
2
0
2
1
:
1
6
1
1
-
1
6
1
8
1612
h
av
e
b
ec
o
m
e
a
p
o
p
u
lar
to
o
l
f
o
r
m
u
lti
v
ar
iate
ti
m
e
s
er
ies
d
ata
p
r
o
ce
s
s
in
g
[1
]
-
[
4
]
.
I
t
is
a
s
u
b
s
et
o
f
m
ac
h
in
e
lear
n
in
g
m
et
h
o
d
s
,
an
d
it
s
o
b
j
e
ctiv
e
i
s
to
lear
n
th
e
m
o
d
el
p
ar
a
m
eter
s
f
r
o
m
th
e
r
ep
r
ese
n
tatio
n
o
f
t
h
e
d
ata.
Dee
p
n
eu
r
al
n
et
w
o
r
k
s
(
D
NNs)
ar
e
th
e
b
ac
k
b
o
n
e
f
o
r
th
e
i
m
p
le
m
en
tatio
n
o
f
d
ee
p
lear
n
i
n
g
ar
ch
itect
u
r
e.
A
n
e
u
r
al
n
et
w
o
r
k
w
it
h
m
o
r
e
th
a
n
o
n
e
h
id
d
en
la
y
er
i
s
ca
lled
a
DNN,
w
h
er
e
t
h
e
o
u
tp
u
t
o
f
a
la
y
er
t
u
r
n
s
i
n
to
t
h
e
i
n
p
u
t
o
f
th
e
s
u
b
s
eq
u
en
t
la
y
er
.
Du
e
to
t
h
is
la
y
er
ed
ar
ch
itect
u
r
e,
th
e
DNNs
b
ec
o
m
e
a
p
o
p
u
lar
f
ea
tu
r
e
lear
n
in
g
to
o
l
f
o
r
ex
tr
ac
ti
n
g
f
ea
t
u
r
es
f
r
o
m
t
h
e
h
i
s
to
r
ical
d
ataset.
I
n
r
ec
en
t
ti
m
e
s
,
th
e
t
w
o
m
o
s
t
p
o
p
u
lar
d
ee
p
l
ea
r
n
in
g
tec
h
n
iq
u
e
s
ar
e
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
t
wo
r
k
(
C
NN)
[
5
]
an
d
r
ec
u
r
r
en
t
n
eu
r
al
n
et
w
o
r
k
(
R
NN)
[
6
]
,
[
7
]
.
T
h
ese
n
et
w
o
r
k
s
h
av
e
r
ec
eiv
ed
in
ter
es
t
in
d
if
f
e
r
en
t
ap
p
licatio
n
ar
ea
s
,
lik
e
co
m
p
u
ter
v
i
s
io
n
,
n
at
u
r
al
lan
g
u
a
g
e
p
r
o
ce
s
s
in
g
,
an
d
ti
m
e
s
er
ie
s
f
o
r
ec
ast
in
g
.
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
ST
M)
[
8
]
,
[
9
]
r
ec
u
r
r
en
t
n
eu
r
al
n
et
w
o
r
k
s
ar
e
th
e
m
o
d
if
ied
v
er
s
io
n
o
f
t
h
e
R
NN
s
,
t
h
at
o
v
er
co
m
e
s
s
o
m
e
o
f
t
h
e
s
h
o
r
tf
all
s
o
f
t
h
e
R
N
Ns.
C
NN
[
1
0
]
,
[
1
1
]
h
as
p
r
o
d
u
ce
d
a
g
r
ea
t
d
ea
l
o
f
s
u
cc
ess
in
i
m
a
g
e
r
ec
o
g
n
itio
n
,
an
d
i
m
ag
e
cla
s
s
i
f
icatio
n
.
W
h
er
ea
s
,
L
ST
M
ac
h
iev
e
s
en
o
r
m
o
u
s
s
u
cc
e
s
s
i
n
m
ac
h
i
n
e
tr
an
s
latio
n
,
n
at
u
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
i
n
g
,
an
d
s
p
ee
ch
r
ec
o
g
n
itio
n
.
I
n
co
n
tr
ast
w
it
h
th
e
MT
S
d
ata,
C
NNs
ar
e
p
o
p
u
lar
f
o
r
t
h
eir
f
ea
tu
r
e
lear
n
i
n
g
ca
p
ab
ilit
ies
w
it
h
in
a
n
in
p
u
t
w
i
n
d
o
w
b
u
t
ar
e
n
o
t
u
s
e
f
u
l
f
o
r
ex
p
lo
r
in
g
t
h
e
te
m
p
o
r
al
f
ea
tu
r
es.
O
n
t
h
e
o
th
er
h
an
d
,
L
ST
Ms
ar
e
o
u
t
s
tan
d
i
n
g
to
m
o
d
el
th
e
lo
n
g
-
ter
m
d
ep
en
d
en
cies
an
d
te
m
p
o
r
al
f
e
atu
r
es
ex
tr
ac
tio
n
f
r
o
m
u
n
i
v
ar
iate
ti
m
e
s
er
ies
d
ata
,
b
u
t
n
o
t
s
u
itab
le
f
o
r
ex
tr
ac
tin
g
th
e
co
m
p
le
x
an
d
s
p
atial
f
e
atu
r
es
f
r
o
m
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ies
d
ata.
I
n
th
e
last
f
e
w
y
ea
r
s
,
d
if
f
er
e
n
t
m
o
d
el
s
[
1
2
]
-
[
1
8
]
is
p
r
o
p
o
s
ed
b
y
co
m
b
i
n
i
n
g
t
w
o
o
r
m
o
r
e
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
es
f
o
r
MT
SF
p
r
o
b
lem
s
,
an
d
th
ese
m
o
d
els
h
a
v
e
s
h
o
w
n
e
x
ce
llen
t
r
e
s
u
l
ts
.
D
u
et
a
l
.
[
1
5
]
p
r
o
p
o
s
ed
a
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
air
h
ea
lt
h
m
o
n
ito
r
i
n
g
.
T
h
is
m
o
d
el
is
co
n
s
tr
u
cted
b
y
co
m
b
i
n
i
n
g
1
D
-
C
N
Ns
a
n
d
b
id
ir
ec
tio
n
al
L
ST
Ms.
T
h
e
au
th
o
r
s
s
h
o
w
t
h
at
th
e
co
m
b
in
ed
m
o
d
el
p
r
o
d
u
ce
d
b
etter
r
esu
lts
co
m
p
ar
ed
to
th
e
s
h
allo
w
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
m
o
d
el
s
.
T
h
e
tr
an
s
f
er
r
ed
b
i
-
d
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
T
L
-
B
L
ST
M)
m
o
d
el
i
s
p
r
o
p
o
s
ed
in
[
1
9
]
p
r
ed
ict
th
e
air
q
u
ali
t
y
.
Her
e
th
e
au
t
h
o
r
u
s
ed
th
e
b
id
ir
ec
tio
n
al
L
ST
M
to
ex
tr
ac
t
th
e
te
m
p
o
r
al
d
ep
en
d
en
cies
o
f
P
M2
.
5
an
d
u
s
e
tr
an
s
f
er
lear
n
i
n
g
to
tr
an
s
f
er
th
e
k
n
o
w
led
g
e
f
r
o
m
th
e
s
m
aller
te
m
p
o
r
al
w
i
n
d
o
w
to
t
h
e
lar
g
er
te
m
p
o
r
al
w
i
n
d
o
w
.
B
u
t
t
h
ese
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
el
s
d
o
n
o
t
co
n
ce
n
tr
ate
o
n
t
h
e
p
r
o
p
er
r
ep
r
esen
tatio
n
o
f
th
e
i
n
p
u
t
d
ata
a
n
d
d
o
n
o
t
m
ai
n
tai
n
t
h
e
p
r
o
p
er
tem
p
o
r
al
o
r
d
er
in
g
o
f
t
h
e
e
x
tr
ac
ted
f
ea
t
u
r
es b
y
t
h
e
f
ir
s
t p
h
ase
o
f
t
h
e
m
o
d
els.
So
d
ev
iati
n
g
f
r
o
m
t
h
e
r
ec
en
t
r
esear
ch
w
o
r
k
,
h
er
e
w
e
p
r
o
p
o
s
e
a
h
y
b
r
id
d
ee
p
n
e
u
r
al
n
et
w
o
r
k
(
HDNN
)
f
r
a
m
e
w
o
r
k
b
y
co
m
b
i
n
in
g
t
w
o
m
o
s
t
p
o
p
u
lar
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
es
s
u
c
h
as
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
e
t
w
o
r
k
(
C
NN)
an
d
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
B
DL
ST
M)
r
e
cu
r
r
en
t
n
eu
r
al
n
et
wo
r
k
f
o
r
air
q
u
alit
y
f
o
r
ec
asti
n
g
.
I
n
t
h
i
s
f
r
a
m
e
w
o
r
k
,
w
e
p
r
o
p
o
s
e
a
3
D
ten
s
o
r
f
o
r
m
atio
n
s
c
h
e
m
e
to
co
n
v
er
t
t
h
e
m
u
l
tiv
ar
iate
t
i
m
e
-
s
er
ies
d
ata
to
i
m
a
g
e
lik
e
3
D
d
ata
as
in
p
u
t
o
f
th
e
C
NN
m
o
d
u
le.
W
e
also
in
tr
o
d
u
ce
a
lin
k
in
g
la
y
er
b
et
w
ee
n
th
e
C
NN
m
o
d
u
le
a
n
d
th
e
L
ST
M
m
o
d
u
le
to
m
ain
tain
th
e
te
m
p
o
r
al
o
r
d
e
r
o
f
th
e
f
ea
tu
r
es
e
x
tr
ac
ted
b
y
t
h
e
C
NN
m
o
d
u
le.
T
h
e
r
est
o
f
t
h
e
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
w
s
:
I
n
s
ec
tio
n
2
,
w
e
d
esc
r
ib
e
t
h
e
r
ese
ar
ch
m
et
h
o
d
o
lo
g
y
.
T
h
e
ex
p
er
i
m
e
n
tal
p
r
o
ce
s
s
is
p
r
esen
ted
in
s
ec
tio
n
3
.
Sectio
n
4
i
n
cl
u
d
es
d
is
c
u
s
s
io
n
ab
o
u
t
th
e
d
etailed
r
e
s
u
l
ts
ac
h
iev
ed
,
an
d
t
h
e
co
n
cl
u
s
io
n
i
s
d
r
a
w
n
in
s
ec
tio
n
5.
2.
RE
S
E
ARCH
M
E
T
H
O
D
O
L
O
G
Y
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n
a
nd
co
rr
ela
t
io
n
a
na
ly
s
is
I
n
th
is
w
o
r
k
,
w
e
u
s
e
th
e
air
p
o
llu
tan
t
an
d
th
e
m
eteo
r
o
lo
g
y
d
atasets
o
f
S
y
d
n
e
y
,
Au
s
tr
alia
an
d
Delh
i,
I
n
d
ia.
T
h
e
d
etail
d
escr
ip
tio
n
o
f
th
e
d
atasets
is
r
ep
r
esen
ted
in
th
e
T
a
b
le
1.
I
n
th
is
s
t
u
d
y
,
w
e
co
n
s
id
er
th
e
P
M2
.
5
as
o
u
r
tar
g
et
o
u
tp
u
t
t
o
p
r
ed
ict
th
e
air
q
u
alit
y
,
a
n
d
o
th
er
air
p
o
llu
ta
n
ts
an
d
m
et
eo
r
o
lo
g
ical
d
ata
ar
e
tr
ea
ted
as
th
e
i
n
p
u
t
o
f
t
h
e
m
o
d
el.
T
o
r
ed
u
ce
th
e
n
u
m
b
er
o
f
in
p
u
t
p
ar
a
m
eter
s
w
i
th
o
u
t
d
eg
r
ad
in
g
th
e
o
v
er
al
l
i
m
p
ac
t o
n
th
e
tar
g
et
o
u
tp
u
t,
h
e
r
e
th
e
co
r
r
elatio
n
a
n
al
y
s
is
is
c
o
n
d
u
cted
o
n
t
h
e
i
n
p
u
t
d
ataset
u
s
i
n
g
t
h
e
P
ea
r
s
o
n
’
s
co
r
r
elatio
n
co
ef
f
ic
ien
t
.
W
e
r
e
m
o
v
e
o
n
e
o
f
t
h
e
p
ar
a
m
e
ter
s
f
r
o
m
a
p
air
o
f
th
e
p
ar
a
m
eter
,
w
h
ic
h
ar
e
h
i
g
h
l
y
co
r
r
elate
d
.
T
ab
le
1
.
Data
s
et
d
escr
ip
tio
n
L
o
c
a
t
i
o
n
P
o
l
l
u
t
a
n
t
s
M
e
t
e
o
r
o
l
o
g
i
c
a
l
F
a
c
t
o
r
s
P
e
r
i
o
d
s
N
o
.
o
f
r
o
w
s
F
r
e
q
u
e
n
c
y
S
y
d
n
e
y
C
O
,
N
O
,
N
O
2
,
S
O
2
O
Z
O
N
E,
P
M
1
0
.
P
M
2
.
5
T
e
mp
e
r
a
t
u
r
e
,
W
i
n
d
D
i
r
e
c
t
i
o
n
H
u
mi
d
i
t
y
,
W
i
n
d
S
p
e
e
d
0
1
.
0
1
.
2
0
1
8
2
4
:
0
0
H
o
u
r
s
-
3
1
.
1
2
.
2
0
1
9
2
3
:
0
0
H
o
u
r
s
1
7
4
0
3
H
o
u
r
D
e
l
h
i
C
O
,
N
O
,
N
O
2
,
S
O
2
O
Z
O
N
E,
P
M
1
0
.
P
M
2
.
5
T
e
mp
e
r
a
t
u
r
e
,
W
i
n
d
D
i
r
e
c
t
i
o
n
W
i
n
d
S
p
e
e
d
,
H
u
m
i
d
i
t
y
,
P
r
e
ssu
r
e
0
1
.
0
1
.
2
0
1
5
0
3
:
0
0
H
o
u
r
s
-
2
4
.
0
4
.
2
0
1
7
2
3
:
0
0
H
o
u
r
s
2
0
2
7
7
H
o
u
r
2
.
2
.
Da
t
a
prepro
ce
s
s
ing
Sin
ce
th
e
air
p
o
llu
tio
n
d
ata
is
co
llected
f
r
o
m
d
if
f
er
en
t
air
p
o
llu
tio
n
s
en
s
o
r
s
,
s
o
th
er
e
is
a
p
o
s
s
ib
ilit
y
o
f
m
i
s
s
i
n
g
an
d
n
o
is
y
d
ata
[
2
0
]
.
T
h
e
p
e
r
f
o
r
m
a
n
ce
o
f
a
n
y
d
ee
p
lear
n
in
g
m
o
d
els
h
ea
v
il
y
d
ep
en
d
s
o
n
th
e
q
u
alit
y
o
f
th
e
in
p
u
t
d
ata.
So
,
to
p
r
o
d
u
ce
h
ig
h
-
q
u
alit
y
d
ata
f
r
o
m
it,
d
ata
p
r
ep
r
o
ce
s
s
in
g
is
an
i
m
p
o
r
t
an
t
p
ar
t
o
f
an
y
d
ee
p
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
h
yb
r
id
d
ee
p
lea
r
n
in
g
m
o
d
el
fo
r
a
ir
q
u
a
lity time
s
eries
p
r
ed
ictio
n
(
S
a
mit B
h
a
n
ja
)
1613
lear
n
in
g
al
g
o
r
ith
m
.
I
n
th
is
e
x
p
er
im
e
n
t,
w
e
r
ep
lace
e
v
er
y
m
is
s
i
n
g
v
al
u
e
b
y
th
e
m
ea
n
v
al
u
e
o
f
j
u
s
t
b
e
f
o
r
e
an
d
af
ter
t
h
e
m
is
s
in
g
v
al
u
e.
I
n
t
h
e
n
ex
t
s
tep
,
w
e
n
o
r
m
alize
th
e
r
ed
u
ce
d
d
ataset
b
y
m
i
n
-
m
ax
n
o
r
m
aliza
tio
n
tech
n
iq
u
e.
T
h
e
d
ata
n
o
r
m
aliza
tio
n
[
2
1
]
is
an
i
m
p
o
r
tan
t
s
tep
o
f
d
ata
p
r
ep
r
o
ce
s
s
in
g
b
ec
au
s
e
th
e
ti
m
e
-
s
er
ies d
at
a
ca
n
v
ar
y
o
v
er
a
lar
g
e
r
an
g
e
an
d
also
p
r
o
d
u
ce
a
g
o
o
d
q
u
alit
y
o
f
d
ata
f
r
o
m
it,
t
h
e
d
ata
m
u
s
t
b
e
s
ca
le
-
d
o
w
n
to
a
ce
r
tain
r
an
g
e.
2
.
3
.
I
np
ut
t
ens
o
r
f
o
r
m
a
t
io
n
I
n
th
is
s
ec
tio
n
,
w
e
r
ep
r
esen
t
th
e
in
p
u
t
ten
s
o
r
f
o
r
m
atio
n
s
ch
e
m
e.
C
NN
ac
h
ie
v
es
a
g
r
ea
t
d
ea
l
o
f
s
u
cc
e
s
s
i
n
i
m
a
g
e
cla
s
s
i
f
icat
io
n
,
an
d
i
m
a
g
e
r
ec
o
g
n
it
io
n
.
I
m
ag
e
d
ata
i
s
d
if
f
er
e
n
t
f
r
o
m
th
e
ti
m
e
-
s
er
ies
d
ata.
Hen
ce
,
to
f
ee
d
th
e
t
i
m
e
-
s
er
ies
d
ata
in
to
t
h
e
C
NN
m
o
d
el
w
e
ap
p
ly
t
h
e
i
m
a
g
e
-
li
k
e
te
n
s
o
r
f
o
r
m
atio
n
s
ch
e
m
e
to
en
co
d
e
th
e
m
u
ltid
i
m
e
n
s
io
n
al
t
i
m
e
-
s
er
ies
d
ata.
E
ac
h
r
ed
-
g
r
ee
n
-
b
l
u
e
(
R
GB
)
i
m
a
g
e
co
n
s
is
t
s
o
f
th
r
ee
2
D
ar
r
ay
s
o
f
p
ix
els,
w
h
ic
h
m
ea
n
s
ea
ch
R
GB
i
m
ag
e
ca
n
b
e
r
ep
r
esen
te
d
as
a
3
D
ten
s
o
r
o
f
th
e
d
im
en
s
io
n
ℎ
×
×
3
,
h
er
e
3
r
ep
r
esen
t
th
e
d
ep
th
o
f
th
e
t
en
s
o
r
.
B
y
ap
p
l
y
i
n
g
t
h
is
en
co
d
in
g
m
eth
o
d
,
w
e
tr
a
n
s
f
o
r
m
o
u
r
m
u
lti
v
ar
iat
e
ti
m
e
-
s
er
ies d
ata
in
to
3
D
ten
s
o
r
,
w
h
er
e
th
e
n
u
m
b
er
o
f
s
e
n
s
o
r
s
r
ep
r
esen
t
s
th
e
d
ep
th
o
f
th
e
te
n
s
o
r
.
T
h
e
in
p
u
t
ten
s
o
r
f
o
r
m
atio
n
s
ch
e
m
e
co
n
s
i
s
ts
o
f
t
w
o
-
p
h
a
s
e
as
d
ep
icted
in
Fig
u
r
e
1
.
I
t
r
e
p
r
esen
t
s
th
e
ten
s
o
r
f
o
r
m
atio
n
tech
n
iq
u
e
f
o
r
f
o
u
r
s
e
n
s
o
r
s
(
in
p
u
t)
s
i
g
n
a
ls
(
a,
b
,
c
an
d
d
)
w
h
er
e
t
h
e
le
n
g
t
h
o
f
t
h
e
s
lid
in
g
w
i
n
d
o
w
1
=
3
.
I
n
th
e
f
ir
s
t
p
h
ase,
w
e
co
n
s
tr
u
ct
a
te
n
s
o
r
o
f
th
e
d
i
m
en
s
io
n
3
×
1
×
1
f
r
o
m
ev
er
y
s
in
g
le
s
en
s
o
r
s
ig
n
al
w
it
h
in
th
e
s
li
d
in
g
w
in
d
o
w
a
n
d
th
e
n
s
u
c
h
f
o
u
r
3
D
te
n
s
o
r
s
o
f
t
h
at
s
l
id
in
g
w
in
d
o
w
ar
e
co
n
ca
ten
a
ted
in
t
h
e
d
ir
ec
tio
n
o
f
ax
is
to
f
o
r
m
a
3
D
ten
s
o
r
o
f
th
e
d
i
m
e
n
s
io
n
3
×
1
×
4
,
as
d
e
p
icted
in
th
e
Fig
u
r
e
1
(
a
)
.
I
n
th
is
w
a
y
,
w
e
c
o
n
s
tr
u
ct
a
s
er
ie
s
o
f
3
D
te
n
s
o
r
s
b
y
s
tr
id
i
n
g
th
e
s
lid
i
n
g
w
i
n
d
o
w
w
it
h
th
e
s
tr
id
i
n
g
len
g
th
1
(
1
>
1
≤
1
)
f
r
o
m
t
h
e
m
u
lti
v
ar
iat
e
ti
m
e
-
s
er
ies
d
ata
s
et.
I
n
t
h
e
n
ex
t
p
h
a
s
e,
all
th
r
ee
co
n
s
ec
u
tiv
e
3
D
ten
s
o
r
s
o
f
t
h
e
s
lid
in
g
w
i
n
d
o
w
(
2
=
3
)
ar
e
co
n
ca
ten
ated
to
th
e
d
ir
ec
tio
n
o
f
ax
i
s
to
f
o
r
m
an
in
p
u
t
ten
s
o
r
o
f
th
e
s
ize
3
×
3
×
4
,
as
d
ep
icted
in
th
e
Fig
u
r
e
1
(
b
)
.
I
n
s
u
c
h
a
w
a
y
,
w
e
co
n
s
tr
u
c
t
a
s
er
ies
o
f
3
D
in
p
u
t te
n
s
o
r
s
b
y
s
tr
id
in
g
th
e
s
lid
in
g
w
i
n
d
o
w
w
it
h
t
h
e
s
tr
i
d
in
g
le
n
g
th
2
,
w
h
er
e
1
<
2
≤
2
.
(
a)
(
b
)
Fig
u
r
e
1
.
I
n
p
u
t te
n
s
o
r
f
o
r
m
ati
o
n
:
(
a)
p
h
ase
o
n
e
an
d
(
b
)
p
h
ase
t
w
o
2
.
4
.
F
o
re
ca
s
t
ing
m
o
del c
o
ns
t
ruct
io
n
Her
e
w
e
d
escr
ib
e
o
u
r
p
r
o
p
o
s
ed
f
o
r
ec
asti
n
g
m
o
d
el
(
HDNN
)
.
Fig
u
r
e
2
r
e
p
r
esen
ts
th
e
o
v
er
all
ar
ch
itect
u
r
e
o
f
th
e
HDNN
f
o
r
ec
asti
n
g
m
o
d
el.
T
h
is
f
o
r
ec
as
tin
g
m
o
d
el
is
co
n
s
tr
u
cted
b
y
t
w
o
p
o
p
u
lar
d
ee
p
lear
n
in
g
ar
c
h
itect
u
r
es
-
C
NN
a
n
d
L
ST
M.
Her
e
o
u
r
HDNN
f
o
r
ec
asti
n
g
m
o
d
el
i
s
d
iv
id
ed
in
to
f
o
u
r
m
o
d
u
les
-
C
NN
Mo
d
u
le,
li
n
k
in
g
m
o
d
u
le
,
L
ST
M
Mo
d
u
le,
an
d
f
u
l
l
y
co
n
n
ec
ted
m
o
d
u
le
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
22
,
No
.
3
,
J
u
n
e
2
0
2
1
:
1
6
1
1
-
1
6
1
8
1614
Fig
u
r
e
2
.
Net
w
o
r
k
ar
c
h
itect
u
r
e
o
f
p
r
o
p
o
s
ed
HDNN
m
o
d
el
2
.
4
.
1
.
CNN
m
o
du
le
C
NN
is
a
d
ee
p
lear
n
i
n
g
ar
ch
itectu
r
e
t
h
at
is
s
p
ec
iall
y
d
esi
g
n
ed
to
p
r
o
ce
s
s
i
m
a
g
e
d
ata.
C
NNs
ar
e
w
id
el
y
u
s
ed
in
t
h
e
ar
ea
o
f
i
m
ag
e
r
ec
o
g
n
itio
n
,
i
m
a
g
e
class
i
f
icatio
n
s
,
an
d
o
b
j
ec
t
d
etec
tio
n
.
A
s
r
ec
en
t
r
esear
ch
w
o
r
k
[
2
2
]
,
[
2
3
]
s
h
o
w
s
t
h
at
th
e
d
ee
p
lear
n
in
g
m
o
d
el
w
ith
t
h
e
co
m
b
in
atio
n
o
f
C
NN
an
d
L
ST
M
ar
ch
itectu
r
es
p
r
o
d
u
ce
s
m
o
r
e
ac
cu
r
ate
r
esu
lt
s
f
o
r
MT
SF
p
r
o
b
lem
s
co
m
p
ar
ed
to
th
e
s
h
allo
w
d
ee
p
lear
n
i
n
g
m
o
d
els.
T
h
u
s
t
h
e
ex
tr
ac
ted
h
ig
h
er
-
o
r
d
er
f
ea
tu
r
e
s
b
y
C
NN
co
n
s
tr
u
ct
h
elp
s
th
e
o
u
ter
lev
el
L
ST
M
co
n
s
tr
u
ct
in
o
u
r
p
r
o
p
o
s
ed
w
o
r
k
f
o
r
m
o
r
e
ac
cu
r
ate
p
r
ed
ictio
n
.
I
n
th
e
p
r
o
p
o
s
ed
m
o
d
el,
th
e
C
NN
m
o
d
u
le
i
s
co
n
s
is
t
s
o
f
s
tac
k
in
g
o
f
t
w
o
C
N
N
la
y
er
s
w
h
er
e
th
e
o
u
tp
u
t
o
f
o
n
e
C
NN
la
y
er
i
s
co
n
s
id
er
ed
as
th
e
in
p
u
t
o
f
t
h
e
n
ex
t
C
NN
la
y
er
a
s
s
h
o
w
n
i
n
F
ig
u
r
e
2
.
E
ac
h
C
N
N
la
y
er
p
er
f
o
r
m
s
t
h
r
ee
d
i
f
f
er
e
n
t
o
p
er
atio
n
s
o
n
t
h
e
i
n
p
u
t
d
ata
-
co
n
v
o
lu
tio
n
al
,
R
e
L
U
[
2
4
]
,
an
d
p
o
o
lin
g
.
T
h
e
in
p
u
ts
o
f
th
e
C
NN
m
o
d
u
le
ar
e
th
e
3
D
in
p
u
t
ten
s
o
r
s
o
f
t
h
e
n
o
r
m
aliz
ed
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ie
s
d
ata
as
il
lu
s
tr
ated
i
n
s
ec
tio
n
2
.
3
.
T
h
e
o
p
er
atio
n
s
o
f
th
e
C
NN
la
y
er
ar
e
p
r
esen
ted
in
Fi
g
u
r
e
3
.
Her
e,
is
th
e
3
D
in
p
u
t
ten
s
o
r
o
f
s
ize
6
×
6
w
ith
d
ep
th
3
an
d
is
t
h
e
f
e
atu
r
e
d
etec
to
r
o
f
s
ize
3
×
3
w
it
h
=
4
ch
a
n
n
els
(
f
ilter
s
)
.
T
h
e
co
n
v
o
lu
tio
n
o
p
er
atio
n
w
it
h
a
s
tr
id
e
len
g
t
h
o
f
1
w
ill
g
e
n
er
at
io
n
4
f
ea
tu
r
e
m
ap
s
o
f
t
h
e
d
i
m
en
s
io
n
4
×
4
.
T
h
en
th
e
n
o
n
-
li
n
ea
r
ac
ti
v
atio
n
f
u
n
ctio
n
s
R
e
L
U
is
ap
p
lied
o
n
t
h
e
f
ea
t
u
r
e
m
ap
.
T
h
e
f
o
r
m
u
la
o
f
t
h
e
R
e
L
U
i
s
as
f
o
llo
w
s
:
(
)
=
{
,
>
0
0
,
≤
0
(
1
)
Af
ter
t
h
at,
w
e
ca
r
r
y
o
u
t
th
e
m
ax
-
p
o
o
lin
g
o
p
er
atio
n
to
p
o
o
l
th
e
m
a
x
i
m
u
m
o
v
er
ea
ch
in
t
er
v
al
as
i
ts
o
u
tp
u
t.
I
f
t
h
e
m
ax
-
p
o
o
lin
g
s
i
ze
is
2
×
2
th
e
n
it,
r
ed
u
ce
d
to
ea
ch
f
ea
t
u
r
e
m
atr
ix
o
f
s
ized
2
×
2
as
s
h
o
w
n
in
Fig
u
r
e
3
.
I
n
th
is
w
a
y
,
th
e
f
ir
s
t
C
NN
la
y
er
co
m
p
r
es
s
ed
t
h
e
i
n
p
u
t
f
ea
tu
r
e
m
atr
i
x
.
T
h
is
co
m
p
r
ess
ed
o
u
tp
u
t
o
f
th
e
f
ir
s
t
C
NN
la
y
er
i
s
t
h
e
lo
c
al
f
ea
t
u
r
e
v
ec
to
r
s
o
f
.
T
h
ese
l
o
ca
l
f
ea
t
u
r
e
v
ec
to
r
s
ar
e
t
h
e
n
f
ed
as
th
e
in
p
u
t
o
f
th
e
s
ec
o
n
d
C
NN
la
y
er
,
an
d
in
th
is
w
a
y
,
t
h
e
s
ec
o
n
d
C
N
N
la
y
er
g
en
er
ates t
h
e
h
ig
h
er
-
o
r
d
er
f
ea
tu
r
e
v
ec
to
r
s
.
Fig
u
r
e
3
.
Op
er
atio
n
o
f
C
NN
la
y
er
2
.
4
.
2
.
L
in
k
ing
o
f
CNN
m
o
d
ule
w
it
h L
ST
M
m
o
du
le
T
h
e
o
u
tp
u
t
o
f
th
e
C
NN
m
o
d
u
le
is
th
e
f
ea
t
u
r
e
v
ec
to
r
s
,
ex
p
r
ess
ed
as
w
h
er
e
=
1
,
2
,
.
.
.
˙
,
an
d
ea
ch
f
ea
tu
r
e
v
ec
to
r
h
a
s
t
h
e
le
n
g
t
h
$
L
$
,
s
o
w
e
h
a
v
e
=
{
,
1
,
,
2
,
.
.
.
.
˙
,
,
}
.
Her
e
ea
ch
f
ea
tu
r
e
v
ec
to
r
is
th
e
h
ig
h
er
-
o
r
d
er
f
ea
tu
r
e
o
f
t
h
e
o
r
ig
i
n
al
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ie
s
d
ataset
a
lo
n
g
t
h
e
ti
m
e
a
x
i
s
.
An
alo
g
o
u
s
to
t
h
e
b
en
ch
m
ar
k
C
NN,
w
h
er
e
th
e
f
latten
i
n
g
la
y
er
is
u
s
ed
to
f
latt
en
th
e
f
ea
t
u
r
e
v
ec
to
r
s
as
s
h
o
w
n
i
n
Fi
g
u
r
e
4
(
a)
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
h
yb
r
id
d
ee
p
lea
r
n
in
g
m
o
d
el
fo
r
a
ir
q
u
a
lity time
s
eries
p
r
ed
ictio
n
(
S
a
mit B
h
a
n
ja
)
1615
h
er
e
w
e
i
n
tr
o
d
u
ce
d
a
lin
k
i
n
g
m
o
d
u
le.
I
n
th
i
s
m
o
d
u
le,
w
e
r
e
d
is
tr
ib
u
ted
th
e
co
m
p
o
n
e
n
ts
o
f
th
e
f
ea
t
u
r
e
v
ec
to
r
s
alo
n
g
th
e
ti
m
e
a
x
is
a
s
s
h
o
w
n
in
Fi
g
u
r
e
4
(
b
)
.
T
h
ese
r
eg
e
n
er
ated
f
ea
tu
r
e
v
ec
to
r
s
ar
e
r
ep
r
esen
ted
i
n
a
m
o
r
e
r
elev
an
t
ti
m
e
o
r
d
er
co
m
p
ar
ed
to
th
e
f
latte
n
i
n
g
la
y
er
o
f
b
en
ch
m
ar
k
C
NN.
T
h
ese
n
e
w
l
y
g
en
er
ated
f
ea
t
u
r
e
v
ec
to
r
s
ar
e
u
s
ed
as t
h
e
in
p
u
t o
f
th
e
L
ST
M
m
o
d
u
le.
(
a)
(
b
)
Fig
u
r
e
4
.
T
h
ese
f
ig
u
r
es a
r
e;
(
a)
Flatten
i
n
g
la
y
er
an
d
(
b
)
lin
k
i
n
g
m
o
d
u
le
2
.
4
.
3
.
L
ST
M
m
o
du
le
I
n
t
h
is
L
ST
M
m
o
d
u
le,
w
e
cr
ea
te
t
w
o
la
y
er
s
o
f
b
id
ir
ec
tio
n
al
L
ST
Ms
(
B
DL
ST
Ms)
an
d
ea
ch
B
DL
ST
M
lay
er
ca
n
b
e
u
n
r
o
ll
ed
to
L
n
u
m
b
er
o
f
B
DL
ST
M
u
n
i
ts
alo
n
g
th
e
ti
m
e
ax
i
s
as
p
r
esen
ted
in
Fi
g
u
r
e
2
.
C
o
n
v
en
t
io
n
al
L
ST
M
n
et
w
o
r
k
ca
n
o
n
l
y
p
r
o
ce
s
s
in
t
h
e
f
o
r
w
ar
d
d
ir
ec
tio
n
a
n
d
as
a
r
esu
lt,
i
t
ca
n
m
is
s
s
o
m
e
o
f
th
e
u
s
e
f
u
l
in
f
o
r
m
at
io
n
d
u
r
i
n
g
th
e
ex
tr
ac
tio
n
o
f
t
h
e
te
m
p
o
r
al
f
ea
t
u
r
es.
B
u
t
t
h
e
B
DL
ST
M
n
et
w
o
r
k
ca
n
p
r
o
ce
s
s
in
b
o
th
th
e
f
o
r
w
ar
d
a
n
d
b
ac
k
w
ar
d
d
ir
ec
tio
n
,
s
o
it
is
e
f
f
icie
n
t
to
e
x
tr
ac
t
th
e
te
m
p
o
r
al
f
ea
t
u
r
es
co
m
p
ar
ed
to
t
h
e
co
n
v
e
n
tio
n
al
L
ST
M
n
et
w
o
r
k
.
T
h
is
is
th
e
r
ea
s
o
n
w
h
y
w
e
ch
o
o
s
e
th
e
B
DL
ST
M
[
2
5
]
,
[
2
6
]
o
v
er
L
ST
M.
Du
e
to
th
is
b
id
ir
ec
tio
n
al
p
r
o
p
er
t
y
,
ea
ch
B
D
L
ST
M
u
n
it
ac
c
u
m
u
late
s
t
h
e
f
ea
tu
r
e
in
f
o
r
m
at
io
n
f
r
o
m
b
o
th
s
id
es
o
f
t
h
e
m
u
lti
v
ar
iate
ti
m
e
-
s
er
ies d
ata
w
it
h
i
n
it
s
ti
m
e
f
r
a
m
e.
I
n
th
is
w
a
y
,
ea
ch
u
n
i
t e
x
tr
ac
t
s
te
m
p
o
r
al
f
ea
t
u
r
es.
2
.
4
.
4
.
F
ull
y
co
nn
ec
t
ed
m
o
du
le
I
t
is
th
e
last
m
o
d
u
le
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
T
h
is
m
o
d
u
le
c
o
n
s
is
ts
o
f
t
w
o
la
y
er
s
–
a
d
r
o
p
o
u
t
la
y
er
an
d
a
d
en
s
e
la
y
er
.
T
h
e
d
r
o
p
o
u
t
la
y
er
r
an
d
o
m
l
y
s
elec
t
s
s
o
m
e
o
f
th
e
n
o
d
es
a
n
d
d
r
o
p
o
u
t
t
h
e
m
.
T
h
is
is
a
r
eg
u
lar
izatio
n
tec
h
n
iq
u
e
t
h
at
i
s
u
s
ed
to
d
ela
w
it
h
th
e
o
v
er
f
itt
in
g
p
r
o
b
le
m
.
T
h
e
n
e
x
t
la
y
er
o
f
t
h
is
m
o
d
u
le
i
s
t
h
e
d
en
s
e
la
y
er
.
T
h
is
la
y
er
co
n
n
ec
t
s
th
e
o
u
tp
u
t
v
ec
to
r
s
f
r
o
m
all
t
h
e
B
DL
ST
M
u
n
its
o
f
t
h
e
s
ec
o
n
d
B
DL
ST
M
la
y
er
to
a
s
in
g
le
no
de
.
Fin
al
l
y
,
t
h
e
o
u
tp
u
t o
f
th
i
s
la
y
er
i
s
th
e
f
o
r
ec
asted
r
esu
lt o
f
o
u
r
p
r
o
p
o
s
ed
w
o
r
k
.
3.
E
XP
E
R
I
M
E
NT
3
.
1
.
E
x
peri
m
ent
a
l
s
et
up
I
n
t
h
i
s
w
o
r
k
,
al
l
th
e
ex
p
e
r
im
en
ts
a
r
e
c
o
n
d
u
c
t
e
d
i
n
P
y
t
h
o
n
p
r
o
g
r
am
m
in
g
la
n
g
u
ag
e
w
it
h
Ke
r
a
s
as
th
e
d
e
e
p
l
e
a
r
n
in
g
l
i
b
r
a
r
y
.
He
r
e
w
e
f
o
r
e
c
as
t
th
e
n
ex
t
h
o
u
r
PM
2
.
5
b
a
s
e
d
o
n
th
e
p
r
ev
i
o
u
s
8
h
o
u
r
s
p
o
ll
u
t
an
ts
c
o
n
c
en
t
r
at
i
o
n
(
C
O
,
N
O
2
,
an
d
S
O
2
)
a
n
d
th
e
m
et
e
o
r
o
l
o
g
i
c
a
l
f
a
c
t
o
r
s
(
t
em
p
e
r
at
u
r
e
,
w
in
d
d
i
r
ec
t
i
o
n
,
h
u
m
i
d
ity
,
a
n
d
w
in
d
s
p
e
e
d
)
.
W
e
h
av
e
u
s
e
d
7
0
%
o
f
th
e
d
a
t
a
f
o
r
t
r
a
in
i
n
g
p
u
r
p
o
s
e
s
,
1
5
%
o
f
d
a
t
a
f
o
r
t
h
e
v
al
i
d
a
t
i
o
n
p
u
r
p
o
s
e
,
an
d
w
e
t
e
s
t
th
e
p
e
r
f
o
r
m
an
c
e
o
f
th
e
m
o
d
e
l
b
y
th
e
r
em
a
in
in
g
1
5
%
o
f
d
a
t
a
.
I
n
th
e
in
p
u
t
t
en
s
o
r
f
o
r
m
a
ti
o
n
s
ch
em
e
,
f
o
r
t
h
e
f
i
r
s
t
p
h
as
e
,
w
e
s
e
t
th
e
l
en
g
th
o
f
th
e
s
li
d
in
g
w
in
d
o
w
1
=
5
,
a
n
d
s
t
r
i
d
in
g
-
l
en
g
th
o
f
th
e
s
l
i
d
in
g
w
in
d
o
w
1
=
1
.
I
n
t
h
e
s
e
c
o
n
d
p
h
as
e
,
w
e
s
e
t
th
e
l
en
g
th
o
f
th
e
s
li
d
in
g
w
in
d
o
w
2
=
4
,
a
n
d
s
t
r
i
d
in
g
-
l
en
g
th
2
=
2
.
T
h
e
r
e
f
o
r
e
,
th
e
d
im
en
s
i
o
n
o
f
th
e
c
o
n
s
t
r
u
c
t
e
d
3
D
i
n
p
u
t
t
en
s
o
r
s
b
e
c
o
m
es
5
×
4
×
7
.
T
h
e
s
e
l
e
c
ti
o
n
o
f
t
h
e
h
y
p
e
r
p
a
r
am
et
e
r
p
l
ay
s
a
c
r
u
ci
a
l
r
o
l
e
i
n
t
h
e
p
e
r
f
o
r
m
an
c
e
o
f
an
y
d
e
ep
l
e
a
r
n
in
g
m
o
d
e
l
.
A
f
t
e
r
th
e
f
in
e
-
t
u
n
i
n
g
o
f
o
u
r
f
o
r
e
c
a
s
t
in
g
m
o
d
el
,
w
e
s
e
t
th
e
h
y
p
e
r
p
a
r
am
e
t
e
r
s
a
r
e
as
f
o
l
l
o
w
s
:
F
o
r
t
h
e
f
i
r
s
t
C
N
N
l
ay
e
r
,
th
e
n
u
m
b
e
r
o
f
f
i
l
te
r
s
1
=
16
,
t
h
e
f
i
l
t
e
r
s
i
z
e
i
s
4
,
s
t
r
i
d
i
n
g
l
en
g
th
1
=
1
,
a
n
d
t
h
e
w
in
d
o
w
s
i
z
e
o
f
th
e
m
ax
-
p
o
o
l
in
g
i
s
2
.
F
o
r
th
e
s
e
c
o
n
d
C
NN
l
ay
e
r
,
th
e
n
u
m
b
e
r
o
f
f
i
l
te
r
s
2
=
12
,
t
h
e
f
i
lt
e
r
s
i
z
e
i
s
3
,
s
t
r
i
d
i
n
g
l
en
g
th
2
=
1
,
a
n
d
t
h
e
w
in
d
o
w
s
iz
e
o
f
th
e
m
ax
-
p
o
o
l
in
g
i
s
2
.
F
o
r
th
e
f
i
r
s
t
B
DL
S
T
M
l
ay
e
r
,
th
e
n
u
m
b
e
r
o
f
n
eu
r
o
n
s
i
s
6
4
.
a
n
d
in
th
e
s
ec
o
n
d
l
ay
e
r
,
n
u
m
b
e
r
o
f
n
eu
r
o
n
s
i
s
3
2
.
T
h
e
t
r
a
i
n
in
g
b
a
t
ch
s
i
z
e
i
s
1
6
,
t
h
e
m
o
d
e
l’
s
o
p
t
im
i
z
e
r
i
s
s
t
o
ch
as
t
i
c
o
p
t
im
i
z
a
t
i
o
n
(
A
d
am
)
,
an
d
th
e
n
u
m
b
e
r
o
f
i
t
e
r
a
ti
o
n
is
1
0
0
.
T
h
e
d
r
o
p
o
u
t
r
e
g
u
la
r
i
z
a
ti
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
22
,
No
.
3
,
J
u
n
e
2
0
2
1
:
1
6
1
1
-
1
6
1
8
1616
t
e
c
h
n
i
q
u
e
i
s
u
s
e
d
t
o
r
e
d
u
ce
th
e
o
v
e
r
f
i
tt
in
g
p
r
o
b
l
em
o
f
th
e
m
o
d
e
l
.
A
f
t
e
r
t
h
e
f
in
e
-
tu
n
in
g
o
f
o
u
r
m
o
d
e
l
,
w
e
s
e
t
th
e
d
r
o
p
o
u
t
r
a
t
e
t
o
0
.
6
,
an
d
th
is
d
r
o
p
o
u
t
l
ay
e
r
i
s
a
d
d
e
d
b
ef
o
r
e
th
e
f
i
n
al
d
e
n
s
e
l
ay
e
r
o
f
th
e
f
u
l
ly
c
o
n
n
e
c
t
e
d
m
o
d
u
l
e
.
3
.
2
.
E
v
a
lua
t
i
o
n
T
o
an
al
y
ze
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el,
h
er
e
w
e
co
n
s
id
er
th
r
ee
p
er
f
o
r
m
a
n
ce
m
a
tr
ics,
v
iz.
R
o
o
t
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
R
MSE
)
,
m
ea
n
ab
s
o
l
u
te
p
er
ce
n
tag
e
er
r
o
r
(
MA
P
E
)
,
an
d
s
y
m
m
e
tr
ic
m
ea
n
ab
s
o
l
u
t
e
p
er
ce
n
tag
e
er
r
o
r
(
SMA
P
E
)
.
T
h
ese
er
r
o
r
s
ar
e
ca
lcu
lated
u
s
in
g
th
e
f
o
llo
w
i
n
g
f
o
r
m
u
la.
=
√
1
∑
(
−
^
)
2
=
1
(
2
)
=
1
∑
(
|
−
^
|
)
=
1
×
100
(
3
)
=
1
∑
|
−
^
|
(
|
+
^
|
)
2
⁄
=
1
(
4
)
W
h
er
e
is
ac
tu
al
v
alu
e
a
n
d
^
is
t
h
e
p
r
ed
icted
v
alu
e
at
ti
m
e
-
s
ta
m
p
t
.
I
n
th
i
s
s
t
u
d
y
,
w
e
co
m
p
ar
e
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
o
u
r
p
r
o
p
o
s
ed
HDNN
w
it
h
t
h
e
f
o
u
r
p
o
p
u
lar
s
h
allo
w
m
o
d
el
s
an
d
th
r
ee
s
tates
-
of
-
t
h
e
-
ar
t
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
e
s
h
allo
w
m
o
d
els
ar
e
SVR
,
A
u
to
R
eg
r
es
s
i
v
e
I
n
teg
r
ated
Mo
v
in
g
Av
er
ag
e
(
AR
I
M
A
)
,
C
NN,
an
d
B
D
L
ST
M,
an
d
t
h
e
h
y
b
r
id
s
m
o
d
els
ar
e
DAQFF[
1
5
]
an
d
T
L
-
B
L
ST
M[
1
9
]
.
T
h
e
p
ar
am
eter
s
o
f
ea
ch
o
f
th
e
m
o
d
els ar
e
s
h
o
w
n
i
n
T
ab
l
e
2
.
T
ab
le
2
.
P
ar
am
eter
s
o
f
d
i
f
f
er
en
t
m
o
d
els
M
o
d
e
l
P
a
r
a
me
t
e
r
s
I
n
p
u
t
S
V
R
k
e
r
n
e
l
=
r
b
f
,
C
=
1
0
0
,
g
a
mm
a
=
0
.
1
P
M
2
.
5
T
i
me
-
se
r
i
e
s
A
R
I
M
A
p
=
8
,
d
=
1
,
q
=
1
P
M
2
.
5
T
i
me
-
se
r
i
e
s
C
N
N
f
i
l
t
e
r
s=6
4
,
k
e
r
n
e
l
=
3
×
3
,
M
a
x
P
o
o
l
i
n
g
=
2
×
2
M
u
l
t
i
v
a
r
i
a
t
e
T
i
me
-
se
r
i
e
s
B
D
L
S
T
M
2
l
a
y
e
r
s o
f
B
D
L
S
T
M
,
L
a
y
e
r
1
:
8
0
N
e
u
r
o
n
s,
L
a
y
e
r
2
:
3
2
N
e
u
r
o
n
s
M
u
l
t
i
v
a
r
i
a
t
e
T
i
me
-
se
r
i
e
s
D
A
Q
F
F
[
1
5
]
2
l
a
y
e
r
s o
f
1
D
-
C
N
N
a
n
d
t
w
o
l
a
y
e
r
s o
f
L
S
T
M
M
u
l
t
i
v
a
r
i
a
t
e
T
i
me
-
se
r
i
e
s
TL
-
B
L
S
T
M
[
1
9
]
3
l
a
y
e
r
s o
f
B
D
L
S
T
M
M
u
l
t
i
v
a
r
i
a
t
e
T
i
me
-
se
r
i
e
s
4.
RE
SU
L
T
AND
DI
SCUS
SI
O
N
I
n
Fig
u
r
e
s
5
an
d
6
,
w
e
g
r
ap
h
icall
y
r
ep
r
esen
t
(
zo
o
m
ed
f
o
r
5
0
0
h
o
u
r
s
)
th
e
p
r
ed
ict
ed
P
M2
.
5
co
n
ce
n
tr
atio
n
f
o
r
th
e
b
o
th
t
h
e
test
i
n
g
d
ataset
s
.
Fro
m
t
h
ese
f
i
g
u
r
es,
it
is
q
u
ite
clea
r
th
at
th
e
H
DNN
s
u
cc
e
s
s
f
u
ll
y
p
r
ed
icts
t
h
e
f
l
u
ct
u
atio
n
o
f
P
M2
.
5
co
n
ce
n
tr
atio
n
an
d
p
r
o
d
u
ce
s
a
s
tab
le
p
er
f
o
r
m
an
ce
.
T
h
e
r
es
u
lt
s
o
f
th
e
co
m
p
ar
ati
v
e
a
n
al
y
s
i
s
ar
e
p
r
esen
ted
in
T
ab
le
s
3
an
d
4
.
Fro
m
t
h
ese
tab
les,
w
e
c
an
o
b
s
er
v
e
t
h
at
t
h
e
s
h
allo
w
m
ac
h
i
n
e
lear
n
i
n
g
m
o
d
els
SV
R
a
n
d
A
R
I
M
A
p
r
o
d
u
ce
s
al
m
o
s
t
t
h
e
s
a
m
e
r
es
u
lt,
wh
er
ea
s
t
h
e
s
h
al
lo
w
d
ee
p
lear
n
in
g
m
o
d
els
B
DL
S
T
M
an
d
C
NN
p
r
o
d
u
ce
s
b
etter
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
to
t
h
e
s
h
allo
w
m
ac
h
i
n
e
lear
n
in
g
m
o
d
el
s
in
ter
m
s
o
f
R
MSE
an
d
M
A
P
E
.
On
th
e
o
th
er
h
a
n
d
,
test
in
g
er
r
o
r
s
o
f
all
th
e
h
y
b
r
id
d
ee
p
lear
n
in
g
m
o
d
els
ar
e
s
i
g
n
i
f
ica
n
tl
y
lo
w
co
m
p
ar
ed
to
th
e
s
h
all
o
w
d
ee
p
lear
n
i
n
g
m
o
d
els.
Fro
m
t
h
ese
tab
les,
w
e
ca
n
also
s
ee
t
h
at
o
u
r
p
r
o
p
o
s
ed
f
r
a
m
e
w
o
r
k
HD
NN
y
ield
s
th
e
lo
w
est
te
s
ti
n
g
er
r
o
r
s
co
m
p
ar
ed
to
all
o
th
er
m
o
d
el
s
.
T
h
e
r
esu
lt
s
co
n
f
ir
m
t
h
at
HDN
N
n
o
t
o
n
l
y
ca
p
tu
r
es
t
h
e
f
ea
tu
r
es
o
f
lo
ca
l
tr
e
n
d
s
a
n
d
also
th
e
lo
n
g
ter
m
te
m
p
o
r
al
d
ep
en
d
en
cies
f
r
o
m
t
h
e
h
i
s
to
r
ical
d
ataset
p
o
llu
ta
n
t
co
n
ce
n
tr
atio
n
s
an
d
m
eteo
r
o
lo
g
ical
f
ac
to
r
s
.
Fig
u
r
e
5
.
Fo
r
ec
asti
n
g
r
esu
l
ts
o
f
P
M2
.
5
co
n
ce
n
tr
atio
n
o
f
S
y
d
n
e
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4752
A
h
yb
r
id
d
ee
p
lea
r
n
in
g
m
o
d
el
fo
r
a
ir
q
u
a
lity time
s
eries
p
r
ed
ictio
n
(
S
a
mit B
h
a
n
ja
)
1617
Fig
u
r
e
6
.
Fo
r
ec
asti
n
g
r
esu
l
ts
o
f
P
M2
.
5
co
n
ce
n
tr
atio
n
o
f
Del
h
i
T
ab
le
3
.
T
h
e
p
er
f
o
r
m
an
ce
co
m
p
ar
i
s
o
n
o
f
d
i
f
f
er
en
t
m
o
d
el
s
f
o
r
Sy
d
n
e
y
d
ataset
M
o
d
e
l
S
V
R
A
R
I
M
A
B
D
L
S
T
M
C
N
N
D
A
Q
F
F
TL
-
B
L
S
T
M
HDNN
R
M
S
E
2
1
.
7
5
3
1
2
0
.
1
9
0
4
1
4
.
2
0
9
1
1
5
.
3
7
1
1
1
1
.
0
7
5
3
1
1
.
3
5
7
4
9
.
3
0
4
2
M
A
P
E
3
5
.
7
8
1
6
3
4
.
0
8
7
3
2
8
.
2
4
3
1
2
9
.
6
1
0
8
1
9
.
3
2
8
1
1
9
.
4
3
8
4
1
7
.
8
5
5
1
S
M
A
P
E
0
.
3
0
1
2
5
0
.
2
6
5
3
0
.
1
8
1
0
0
.
1
8
7
5
0
.
1
4
2
6
0
.
1
3
9
7
0
.
1
1
7
3
T
ab
le
4
.
T
h
e
p
er
f
o
r
m
an
ce
co
m
p
ar
i
s
o
n
o
f
d
i
f
f
er
en
t
m
o
d
el
s
f
o
r
Delh
i d
ataset
M
o
d
e
l
S
V
R
A
R
I
M
A
B
D
L
S
T
M
C
N
N
D
A
Q
F
F
TL
-
B
L
S
T
M
HDNN
R
M
S
E
3
2
.
1
9
2
0
3
1
.
0
8
4
7
2
6
.
9
9
2
1
2
8
.
3
0
0
8
2
4
.
3
5
2
0
2
4
.
1
0
8
3
2
3
.
8
1
7
1
M
A
P
E
1
8
.
5
4
9
0
1
5
.
1
0
0
8
1
3
.
2
8
7
1
1
3
.
9
8
1
2
8
.
0
9
3
1
8
.
1
1
5
9
7
.
0
8
4
6
S
M
A
P
E
0
.
3
8
2
9
0
.
2
5
4
6
0
.
2
4
0
9
0
.
2
5
7
1
0
.
1
5
0
9
0
.
1
5
3
2
0
.
1
4
8
1
5.
CO
NCLU
SI
O
N
A
ir
q
u
al
it
y
f
o
r
ec
asti
n
g
i
s
o
f
u
t
m
o
s
t
i
m
p
o
r
tan
ce
f
o
r
th
e
ea
r
l
y
d
etec
tio
n
o
f
air
p
o
llu
tio
n
.
I
n
t
h
is
s
tu
d
y
,
to
p
r
ed
ict
th
e
air
q
u
alit
y
m
o
r
e
ac
cu
r
atel
y
,
w
e
p
r
o
p
o
s
e
a
h
y
b
r
id
d
ee
p
lear
n
in
g
f
r
a
m
e
w
o
r
k
.
I
n
th
i
s
f
r
a
m
e
w
o
r
k
,
w
e
in
tr
o
d
u
ce
a
n
in
n
o
v
at
iv
e
3
D
in
p
u
t te
n
s
o
r
f
o
r
m
atio
n
tec
h
n
iq
u
e
to
co
n
v
er
t t
h
e
ti
m
e
-
s
er
ie
s
d
ata
to
a
3
D
i
m
a
g
e
lik
e
d
ata
f
o
r
f
ee
d
in
g
it
to
th
e
C
NN
m
o
d
u
le.
Her
e
w
e
al
s
o
in
tr
o
d
u
ce
a
lin
k
in
g
la
y
er
b
et
w
ee
n
t
h
e
C
N
N
m
o
d
u
le
an
d
th
e
L
ST
M
m
o
d
u
le.
T
h
e
ex
p
er
i
m
e
n
tal
r
es
u
lts
s
h
o
w
th
a
t
th
e
p
r
o
p
o
s
ed
f
r
am
e
w
o
r
k
o
u
ts
h
in
ed
o
th
er
s
tate
-
of
-
th
e
-
ar
t
m
o
d
els
in
ter
m
s
o
f
d
if
f
er
en
t
test
i
n
g
er
r
o
r
s
.
T
h
e
p
r
o
p
o
s
ed
f
r
a
m
e
w
o
r
k
also
p
r
o
v
id
es
s
tab
le
p
er
f
o
r
m
a
n
ce
,
esp
ec
iall
y
i
n
th
e
p
er
io
d
s
o
f
wav
e
p
ea
k
an
d
w
a
v
e
v
alle
y
.
S
o
,
it
ca
n
b
e
co
n
clu
d
ed
th
at
th
e
p
r
o
p
e
r
in
p
u
t
d
ata
r
ep
r
esen
tatio
n
p
la
y
s
a
v
ital
r
o
le
in
th
e
p
er
f
o
r
m
a
n
ce
o
f
an
y
d
ee
p
lear
n
in
g
m
o
d
el
s
.
T
h
er
ef
o
r
e,
th
is
f
r
a
m
e
w
o
r
k
w
il
l
h
elp
th
e
ad
m
in
i
s
tr
ati
v
e
au
th
o
r
it
y
f
o
r
co
n
tr
o
llin
g
th
e
air
p
o
llu
tio
n
w
ell
in
ad
v
an
ce
a
n
d
a
b
le
to
w
ar
n
th
e
citizen
.
Her
e
th
e
f
r
a
m
e
w
o
r
k
h
as
ap
p
lied
f
o
r
o
n
l
y
th
e
s
in
g
le
-
ti
m
e
-
s
tep
a
h
ea
d
p
r
ed
ictio
n
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
f
r
a
m
e
w
o
r
k
h
as
n
o
t
b
ee
n
an
al
y
ze
d
f
o
r
th
e
m
u
lti
-
t
i
m
e
-
s
te
p
ah
ea
d
p
r
ed
ictio
n
.
I
n
th
e
f
u
t
u
r
e,
w
e
w
ill
lik
e
to
ap
p
ly
o
u
r
m
o
d
el
f
o
r
m
u
lti
-
ti
m
e
-
s
tep
ah
ea
d
p
r
ed
ictio
n
.
RE
F
E
R
E
NC
E
S
[1
]
J.
F
.
T
o
rre
s,
A
.
G
a
li
c
ia,
A
.
T
ro
n
c
o
so
,
a
n
d
F
.
M
a
rti
n
e
z
-
Á
lv
a
re
z
,
“
A
sc
a
l
a
b
le
a
p
p
r
o
a
c
h
b
a
se
d
o
n
d
e
e
p
lea
rn
in
g
f
o
r
b
ig
d
a
ta
t
im
e
se
ries
f
o
re
c
a
sti
n
g
,
”
In
teg
r.
Co
m
p
u
t
.
A
id
e
d
.
En
g
.
,
v
o
l
.
2
5
,
n
o
.
4
,
p
p
.
3
3
5
–
3
4
8
,
2
0
1
8
,
d
o
i:
1
0
.
3
2
3
3
/ICA
-
1
8
0
5
8
0
.
[2
]
J.
C.
B.
G
a
m
b
o
a
,
“
De
e
p
lea
rn
in
g
f
o
r
ti
m
e
-
se
ries
a
n
a
l
y
sis,”
a
rXiv P
re
p
r.
a
rXiv1
7
0
1
.
0
1
8
8
7
,
2
0
1
7
.
[3
]
B.
S
.
F
re
e
m
a
n
,
G
.
T
a
y
lo
r,
B.
G
h
a
ra
b
a
g
h
i,
a
n
d
J.
T
h
é
,
“
F
o
re
c
a
stin
g
a
ir
q
u
a
li
ty
ti
m
e
se
ries
u
sin
g
d
e
e
p
lea
rn
in
g
,
”
J
.
Ai
r
&
W
a
ste
M
a
n
a
g
.
Ass
o
c
.
,
v
o
l.
6
8
,
n
o
.
8
,
p
p
.
8
6
6
–
8
8
6
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
8
0
/1
0
9
6
2
2
4
7
.
2
0
1
8
.
1
4
5
9
9
5
6
.
[4
]
S
.
Bh
a
n
ja
a
n
d
A
.
Da
s,
“
De
e
p
Ne
u
ra
l
Ne
t
w
o
rk
f
o
r
M
u
lt
iv
a
riate
T
i
m
e
-
S
e
ries
F
o
re
c
a
stin
g
,
”
in
Pro
c
e
e
d
in
g
s
o
f
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Fr
o
n
ti
e
rs
in
Co
m
p
u
ti
n
g
a
n
d
S
y
ste
ms
,
2
0
2
0
,
p
p
.
2
6
7
–
27
7.
[5
]
Y.
L
e
Cu
n
,
e
t
a
l.
,
“
Ha
n
d
w
rit
ten
d
ig
it
re
c
o
g
n
it
io
n
w
it
h
a
b
a
c
k
-
p
ro
p
a
g
a
ti
o
n
n
e
tw
o
rk
,
”
in
Ad
v
a
n
c
e
s
in
n
e
u
ra
l
in
fo
rm
a
ti
o
n
p
ro
c
e
ss
in
g
sy
ste
ms
,
1
9
9
0
,
p
p
.
3
9
6
-
4
0
4
.
[6
]
D.
E.
Ru
m
e
lh
a
rt,
G
.
E.
Hin
to
n
,
R.
J.
W
il
li
a
m
s,
a
n
d
o
th
e
rs,
“
L
e
a
rn
in
g
re
p
re
se
n
tati
o
n
s
b
y
b
a
c
k
-
p
ro
p
a
g
a
ti
n
g
e
rro
rs,”
Co
g
n
.
M
o
d
e
l
.
,
v
o
l.
5
,
n
o
.
3
,
p
.
1
,
1
9
8
8
.
[7
]
J.
L
.
El
m
a
n
,
“
F
in
d
i
n
g
stru
c
t
u
re
in
ti
m
e
,
”
Co
g
n
it
ive
S
c
ien
c
e
,
v
o
l.
1
4
,
n
o
.
2
,
p
p
.
1
7
9
-
2
1
1
,
1
9
9
0
,
d
o
i:
1
0
.
1
0
1
6
/0
3
6
4
-
0
2
1
3
(9
0
)9
0
0
0
2
-
E
.
[8
]
S
.
Ho
c
h
re
it
e
r
a
n
d
J.
S
c
h
m
id
h
u
b
e
r
,
“
L
o
n
g
sh
o
r
t
-
term
m
e
m
o
r
y
,
”
Ne
u
ra
l
C
o
mp
u
t.
,
v
o
l.
9
,
n
o
.
8
,
p
p
.
1
7
3
5
-
1
7
8
0
,
1
9
9
7
,
d
o
i:
1
0
.
1
1
6
2
/n
e
c
o
.
1
9
9
7
.
9
.
8
.
1
7
3
5
.
[9
]
M
.
Rh
a
n
o
u
i,
S
.
Yo
u
sf
i,
M
.
M
ik
r
a
m
,
a
n
d
H.
M
e
riza
k
,
“
F
o
re
c
a
stin
g
f
in
a
n
c
ial
b
u
d
g
e
t
ti
m
e
s
e
ries
:
ARIMA
ra
n
d
o
m
w
a
l
k
v
s
L
S
T
M
n
e
u
ra
l
n
e
tw
o
rk
,
”
IAE
S
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
A
rtif
icia
l
In
telli
g
e
n
c
e
,
v
o
l.
8
,
n
o
.
4
,
pp
.
3
1
7
-
3
2
7
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jai.
v
8
.
i4
.
p
p
3
1
7
-
3
2
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4752
I
n
d
o
n
esia
n
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
22
,
No
.
3
,
J
u
n
e
2
0
2
1
:
1
6
1
1
-
1
6
1
8
1618
[1
0
]
N.
A
.
Ra
h
m
a
d
,
N.
A
.
J.
S
u
f
ri,
N.
H.
M
u
z
a
m
il
,
a
n
d
M
.
A
.
A
s’
a
ri,
“
Ba
d
m
in
to
n
p
lay
e
r
d
e
tec
ti
o
n
u
si
n
g
f
a
ste
r
re
g
io
n
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
,
”
In
d
o
n
e
si
a
n
J
o
u
rn
a
l
o
f
E
l
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
Co
mp
u
ter
S
c
ien
c
e
,
v
o
l.
1
4
,
n
o
.
3
,
p
p
.
1
3
3
0
–
1
3
3
5
,
2
0
1
9
,
d
o
i:
0
.
1
1
5
9
1
/i
jee
c
s.v
1
4
.
i
3
.
p
p
1
3
3
0
-
1
3
3
5
.
[1
1
]
N.
Ka
sim
,
N.
Ra
h
m
a
n
,
Z.
Ib
ra
h
i
m
,
a
n
d
N.
N.
A
.
M
a
n
g
sh
o
r
,
“
Ce
l
e
b
rit
y
F
a
c
e
Re
c
o
g
n
it
io
n
u
si
n
g
D
e
e
p
L
e
a
rn
in
g
,
”
In
d
o
n
e
sia
n
J
o
u
r
n
a
l
o
f
E
lec
trica
l
En
g
i
n
e
e
rin
g
a
n
d
Co
mp
u
ter
S
c
ien
c
e
,
v
o
l.
1
2
,
n
o
.
2
,
p
p
.
4
7
6
-
4
8
1
,
2
0
1
8
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
1
2
.
i2
.
p
p
4
7
6
-
4
8
1
.
[1
2
]
S
.
Du
,
T
.
L
i,
Y.
Ya
n
g
,
a
n
d
S
.
-
J
.
Ho
rn
g
,
“
De
e
p
a
ir
q
u
a
li
ty
f
o
re
c
a
st
in
g
u
si
n
g
h
y
b
rid
d
e
e
p
lea
rn
i
n
g
f
ra
m
e
w
o
rk
,
”
a
rXiv
Pre
p
r.
a
rXiv1
8
1
2
.
0
4
7
8
3
,
2
0
1
8
.
[1
3
]
G
.
Ya
n
g
,
H.
L
e
e
,
a
n
d
G
.
L
e
e
,
“
A
H
y
b
rid
De
e
p
L
e
a
rn
in
g
M
o
d
e
l
to
F
o
re
c
a
st P
a
rti
c
u
late
M
a
tt
e
r
Co
n
c
e
n
tratio
n
L
e
v
e
ls
in
S
e
o
u
l,
S
o
u
th
Ko
re
a
,
”
At
mo
s
p
h
e
re
(
Ba
se
l).
,
v
o
l.
1
1
,
n
o
.
4
,
p
.
3
4
8
,
2
0
2
0
,
d
o
i:
1
0
.
3
3
9
0
/atm
o
s1
1
0
4
0
3
4
8
.
[1
4
]
Z.
Ha
n
,
J.
Z
h
a
o
,
H.
L
e
u
n
g
,
K.
F
.
M
a
,
a
n
d
W
.
W
a
n
g
,
“
A
re
v
ie
w
o
f
d
e
e
p
lea
rn
in
g
m
o
d
e
ls
f
o
r
ti
m
e
se
r
ies
p
re
d
ictio
n
,
”
IEE
E
S
e
n
s.
J
.
,
v
o
l
.
2
1
,
n
o
.
6
,
p
p
.
7
8
3
3
-
8
4
,
2
0
1
9
,
d
o
i:
1
0
.
1
1
0
9
/JS
E
N.2
0
1
9
.
2
9
2
3
9
8
2
.
[1
5
]
S
.
Du
,
T
.
L
i,
Y.
Ya
n
g
,
a
n
d
S
.
-
J.
Ho
rn
g
,
“
De
e
p
a
ir
q
u
a
li
ty
f
o
re
c
a
st
in
g
u
sin
g
h
y
b
rid
d
e
e
p
lea
rn
i
n
g
f
ra
m
e
w
o
rk
,
”
IEE
E
T
ra
n
s.
K
n
o
wl
.
Da
t
a
E
n
g
.
,
v
o
l
.
3
3
,
n
o
.
6
,
p
p
.
2
4
1
2
-
2
4
2
4
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/T
KD
E.
2
0
1
9
.
2
9
5
4
5
1
0
.
[1
6
]
S
.
Bh
a
n
ja
a
n
d
A
.
Da
s,
“
De
e
p
L
e
a
rn
in
g
-
b
a
se
d
In
teg
ra
ted
S
tac
k
e
d
M
o
d
e
l
f
o
r
th
e
S
t
o
c
k
M
a
rk
e
t
P
re
d
ictio
n
,
”
In
t.
J
.
En
g
.
Ad
v
.
T
e
c
h
n
o
l.
,
v
o
l.
9
,
n
o
.
1
,
p
p
.
5
1
6
7
-
5
1
7
4
,
Oc
t.
2
0
1
9
,
d
o
i:
1
0
.
3
5
9
4
0
/i
jea
t.
A
1
8
2
3
.
1
0
9
1
1
9
.
[1
7
]
Q.
S
u
n
,
Y
.
Z
h
u
,
X
.
C
h
e
n
,
A
.
X
u
,
a
n
d
X.
P
e
n
g
,
“
A
h
y
b
rid
d
e
e
p
le
a
rn
in
g
m
o
d
e
l
w
it
h
m
u
lt
i
-
so
u
rc
e
d
a
ta
f
o
r
P
M
2
.
5
c
o
n
c
e
n
tratio
n
f
o
re
c
a
st,”
Air
Qu
a
l
it
y
,
At
m
o
sp
h
e
re
&
He
a
lt
h
,
v
o
l.
1
4
,
n
o
.
4
,
p
p
.
1
–
1
1
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s
1
1
8
6
9
-
020
-
0
0
9
5
4
-
z
.
[1
8
]
B.
G
h
o
se
a
n
d
Z.
Re
h
e
n
a
,
“
A
De
e
p
L
e
a
rn
in
g
A
p
p
ro
a
c
h
f
o
r
P
re
d
ictin
g
A
ir
P
o
ll
u
ti
o
n
i
n
S
m
a
rt
Cit
ies
,
”
i
n
Co
mp
u
t
a
ti
o
n
a
l
I
n
telli
g
e
n
c
e
a
n
d
M
a
c
h
in
e
L
e
a
r
n
in
g
,
S
p
ri
n
g
e
r,
2
0
2
0
,
p
p
.
2
9
-
38
,
d
o
i:
1
0
.
1
0
0
7
/9
7
8
-
9
8
1
-
15
-
8
6
1
0
-
1
_
4
.
[1
9
]
J.
M
a
,
J.
C.
P
.
C
h
e
n
g
,
C.
L
in
,
Y
.
T
a
n
,
a
n
d
J.
Z
h
a
n
g
,
“
Im
p
ro
v
in
g
a
ir
q
u
a
li
ty
p
re
d
ictio
n
a
c
c
u
ra
c
y
a
t
l
a
rg
e
r
te
m
p
o
ra
l
re
so
lu
ti
o
n
s
u
si
n
g
d
e
e
p
lea
rn
in
g
a
n
d
tran
sf
e
r
lea
rn
in
g
tec
h
n
iq
u
e
s,
”
At
mo
s.
En
v
iro
n
.
,
v
o
l.
2
1
4
,
p
.
1
1
6
8
8
5
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/j
.
a
tm
o
se
n
v
.
2
0
1
9
.
1
1
6
8
8
5
.
[2
0
]
N.
H.
A
b
d
Ra
h
m
a
n
a
n
d
M
.
H.
L
e
e
,
“
A
rti
f
i
c
ial
n
e
u
ra
l
n
e
tw
o
r
k
f
o
re
c
a
stin
g
p
e
rf
o
rm
a
n
c
e
w
it
h
m
issin
g
v
a
lu
e
im
p
u
tatio
n
s,”
IAE
S
I
n
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
Arti
fi
c
i
a
l
I
n
tel
li
g
e
n
c
e
,
v
o
l.
9
,
n
o
.
1
,
pp
.
33
-
39
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
a
i.
v
9
.
i
1
.
p
p
3
3
-
39
.
[2
1
]
S
.
B
h
a
n
ja
a
n
d
A
.
Da
s,
“
Im
p
a
c
t
o
f
Da
ta
No
rm
a
li
z
a
ti
o
n
o
n
De
e
p
Ne
u
ra
l
Ne
tw
o
rk
f
o
r
T
i
m
e
S
e
ries
F
o
re
c
a
stin
g
,
”
a
rXiv P
re
p
r.
a
rXiv1
8
1
2
.
0
5
5
1
9
,
2
0
1
8
.
[2
2
]
I.
L
iv
ieris,
E.
P
in
tela
s,
a
n
d
P
.
P
in
tela
s,
“
A
CN
N
--
L
S
T
M
m
o
d
e
l
f
o
r
g
o
ld
p
ri
c
e
ti
m
e
-
s
e
ries
f
o
re
c
a
stin
g
,
”
Ne
u
ra
l
Co
mp
u
t
.
A
p
p
l
.
,
v
o
l.
3
2
,
n
o
.
2
3
,
p
p
.
1
7
3
5
1
–
1
7
3
6
0
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s0
0
5
2
1
-
0
2
0
-
0
4
8
6
7
-
x
.
[2
3
]
T
.
L
i,
M
.
Hu
a
,
a
n
d
X.
W
u
,
“
A
h
y
b
rid
CNN
-
L
S
T
M
m
o
d
e
l
f
o
r
f
o
re
c
a
stin
g
p
a
rti
c
u
late
m
a
tt
e
r
(P
M
2
.
5
)
,
”
IEE
E
Acc
e
ss
,
v
o
l.
8
,
p
p
.
2
6
9
3
3
–
2
6
9
4
0
,
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/A
CCES
S
.
2
0
2
0
.
2
9
7
1
3
4
8
.
[2
4
]
V
.
Na
ir
a
n
d
G
.
E.
Hin
to
n
,
“
Re
c
ti
f
ied
li
n
e
a
r
u
n
it
s
im
p
ro
v
e
re
stricte
d
b
o
lt
z
m
a
n
n
m
a
c
h
in
e
s,”
in
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
7
t
h
i
n
ter
n
a
t
io
n
a
l
c
o
n
fer
e
n
c
e
o
n
ma
c
h
in
e
le
a
rn
in
g
(
ICM
L
-
10)
,
2
0
1
0
,
p
p
.
8
0
7
–
8
1
4
.
[2
5
]
X
.
M
a
,
J.
Z
h
a
n
g
,
B.
Du
,
C.
Di
n
g
,
a
n
d
L
.
S
u
n
,
“
P
a
ra
ll
e
l
a
rc
h
it
e
c
tu
re
o
f
c
o
n
v
o
lu
ti
o
n
a
l
b
i
-
d
irec
ti
o
n
a
l
L
S
T
M
n
e
u
ra
l
n
e
tw
o
rk
s
f
o
r
n
e
tw
o
rk
-
w
id
e
m
e
tro
ri
d
e
rsh
ip
p
re
d
icti
o
n
,
”
IEE
E
T
ra
n
s.
In
tell
.
T
ra
n
sp
.
S
y
st.
,
v
o
l.
2
0
,
n
o
.
6
,
p
p
.
2
2
7
8
-
2
2
8
8
,
2
0
1
8
,
d
o
i:
1
0
.
1
1
0
9
/T
IT
S
.
2
0
1
8
.
2
8
6
7
0
4
2
.
[2
6
]
A
.
S
a
g
h
e
e
r
a
n
d
M
.
Ko
tb
,
“
T
ime
se
ries
f
o
re
c
a
stin
g
o
f
p
e
tro
leu
m
p
r
o
d
u
c
t
io
n
u
sin
g
d
e
e
p
L
S
T
M
re
c
u
r
re
n
t
n
e
tw
o
rk
s,”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
3
2
3
,
p
p
.
2
0
3
-
2
1
3
,
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/j
.
n
e
u
c
o
m
.
2
0
1
8
.
0
9
.
0
8
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.