I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
6
,
Decem
b
er
20
25
,
p
p
.
5
0
6
7
~
5
0
7
9
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
6
.
pp
5
0
6
7
-
5
0
7
9
5067
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Im
pa
ct
o
f
o
utli
er
detec
tion te
chniq
ues o
n t
ime
-
series
f
o
recas
ting
a
ccuracy
f
o
r mul
t
i
-
co
untry e
nerg
y
dema
nd predi
ctio
n
Sh
re
y
a
s
K
a
rnick
1
,
Sa
nja
y
L
a
k
s
hm
ina
ra
y
a
na
n
1
,
M
a
dh
u
P
a
la
t
i
1
,
P
ra
k
a
s
h R
2
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
El
e
c
t
r
o
n
i
c
s E
n
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
E
n
g
i
n
e
e
r
i
n
g
,
B
M
S
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
a
n
d
M
a
n
a
g
e
me
n
t
,
V
i
sv
e
sv
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
s
i
t
y
,
B
e
l
a
g
a
v
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
A
c
h
a
r
y
a
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
3
,
2
0
2
5
R
ev
is
ed
J
u
l 1
5
,
2
0
2
5
Acc
ep
ted
Sep
1
4
,
2
0
2
5
Ac
c
u
ra
te
e
n
e
rg
y
d
e
m
a
n
d
p
re
d
icti
o
n
is
c
r
u
c
ial
fo
r
e
fficie
n
t
g
ri
d
m
a
n
a
g
e
m
e
n
t
a
n
d
re
so
u
rc
e
o
p
t
imiz
a
ti
o
n
,
p
a
rti
c
u
larly
a
c
ro
ss
m
u
lt
ip
le
c
o
u
n
t
ries
with
v
a
ry
in
g
c
o
n
su
m
p
t
io
n
p
a
tt
e
r
n
s.
Ho
we
v
e
r,
re
a
l
-
wo
rld
e
n
e
rg
y
d
e
m
a
n
d
d
a
ta
o
ften
c
o
n
tai
n
s
o
u
tl
iers
t
h
a
t
c
a
n
d
ist
o
rt
f
o
re
c
a
stin
g
a
c
c
u
ra
c
y
.
T
h
is
stu
d
y
e
v
a
lu
a
tes
th
e
imp
a
c
t
o
f
fi
v
e
o
u
t
li
e
r
d
e
tec
ti
o
n
tec
h
n
iq
u
e
s
—
Z
-
S
c
o
re
,
d
en
s
ity
-
b
ased
s
p
atial
clu
s
ter
in
g
o
f
ap
p
licatio
n
s
with
n
o
is
e
(
DBSCAN
)
,
iso
latio
n
f
o
re
st
(IF
)
,
lo
c
a
l
o
u
tl
ier
fa
c
to
r
(LOF
)
,
a
n
d
o
n
e
-
c
las
s
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(
S
VM
)
—
o
n
t
h
e
p
e
rfo
rm
a
n
c
e
o
f
th
re
e
ti
m
e
-
se
ries
fo
re
c
a
stin
g
m
o
d
e
ls:
l
o
n
g
sh
o
r
t
-
term
m
e
m
o
ry
(L
S
TM
)
n
e
tw
o
rk
s,
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
(CNN
)
Au
t
o
e
n
c
o
d
e
r
s,
a
n
d
LS
TM
with
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
s
.
Th
e
m
o
d
e
ls
a
re
tes
ted
u
si
n
g
e
n
e
rg
y
d
e
m
a
n
d
d
a
ta
fr
o
m
fo
u
r
E
u
ro
p
e
a
n
c
o
u
n
tri
e
s
—
G
e
r
m
a
n
y
,
F
ra
n
c
e
,
S
p
a
in
,
a
n
d
It
a
ly
—
d
e
riv
e
d
fro
m
re
a
l
-
ti
m
e
c
o
n
su
m
p
ti
o
n
re
c
o
rd
s.
A
c
o
m
p
a
ra
ti
v
e
a
n
a
l
y
sis
b
a
se
d
o
n
ro
o
t
m
e
a
n
sq
u
a
re
d
e
rr
o
r
(RM
S
E)
d
e
m
o
n
stra
tes
t
h
a
t
i
n
c
o
rp
o
ra
ti
n
g
o
u
tl
ier
d
e
tec
ti
o
n
si
g
n
if
ica
n
tl
y
e
n
h
a
n
c
e
s
m
o
d
e
l
r
o
b
u
stn
e
ss
,
re
d
u
c
in
g
f
o
re
c
a
stin
g
e
rro
rs
c
a
u
se
d
b
y
a
n
o
m
a
lo
u
s
d
a
ta.
Th
e
fin
d
in
g
s
e
m
p
h
a
siz
e
th
e
im
p
o
rtan
c
e
o
f
se
lec
ti
n
g
a
p
p
r
o
p
ria
te
o
u
tl
ier
d
e
tec
ti
o
n
stra
teg
ies
t
o
imp
r
o
v
e
t
h
e
a
c
c
u
ra
c
y
a
n
d
re
li
a
b
il
it
y
o
f
e
n
e
rg
y
d
e
m
a
n
d
fo
re
c
a
stin
g
.
T
h
is
re
se
a
rc
h
p
ro
v
i
d
e
s
v
a
lu
a
b
le
i
n
sig
h
ts
in
t
o
th
e
trad
e
-
o
ffs
in
v
o
lv
e
d
i
n
o
u
tl
ier
re
m
o
v
a
l,
wit
h
imp
l
ica
ti
o
n
s
fo
r
p
o
li
c
y
a
n
d
o
p
e
ra
ti
o
n
a
l
p
ra
c
ti
c
e
s in
e
n
e
rg
y
m
a
n
a
g
e
m
e
n
t.
K
ey
w
o
r
d
s
:
E
n
er
g
y
d
em
an
d
p
r
e
d
ictio
n
L
o
ad
f
o
r
ec
asti
n
g
Ou
tlier
d
etec
tio
n
R
o
o
t m
ea
n
s
q
u
ar
e
d
er
r
o
r
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 B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Sh
r
ey
as Ka
r
n
ick
Dep
ar
tm
e
n
t
o
f
E
l
ec
t
r
i
ca
l
a
n
d
E
le
ct
r
o
n
i
cs
E
n
g
in
ee
r
i
n
g
,
Fa
cu
l
ty
o
f
E
n
g
i
n
ee
r
i
n
g
,
B
MS
I
n
s
t
it
u
te
o
f
T
e
ch
n
o
l
o
g
y
an
d
Ma
n
a
g
em
en
t,
Vis
v
es
v
ar
ay
a
T
ec
h
n
o
l
o
g
ica
l
Un
iv
er
s
it
y
B
elag
av
i 5
6
0
0
1
8
,
Kar
n
atak
a,
I
n
d
ia
E
m
ail:
s
h
r
ey
ask
ar
n
ick
2
0
2
2
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Acc
u
r
ate
en
er
g
y
d
em
an
d
f
o
r
e
ca
s
tin
g
is
a
cr
itical
co
m
p
o
n
en
t
o
f
m
o
d
er
n
p
o
wer
s
y
s
tem
m
a
n
ag
em
en
t.
I
t
f
ac
ilit
ates
o
p
tim
al
g
r
id
o
p
er
atio
n
,
en
s
u
r
es
th
e
r
eliab
ilit
y
o
f
p
o
wer
d
is
tr
ib
u
tio
n
,
a
n
d
s
u
p
p
o
r
ts
th
e
in
teg
r
atio
n
o
f
r
e
n
ewa
b
le
e
n
er
g
y
s
o
u
r
ce
s
b
y
p
r
ed
ictin
g
f
l
u
ctu
atio
n
s
in
e
n
er
g
y
c
o
n
s
u
m
p
tio
n
.
W
ith
th
e
i
n
cr
ea
s
in
g
co
m
p
lex
ity
o
f
g
lo
b
al
en
er
g
y
s
y
s
tem
s
,
f
o
r
ec
asti
n
g
m
o
d
els
m
u
s
t
ac
co
u
n
t
f
o
r
th
e
d
y
n
am
i
c
n
atu
r
e
o
f
en
er
g
y
d
em
an
d
ac
r
o
s
s
d
if
f
er
en
t
r
eg
i
o
n
s
,
in
f
lu
en
ce
d
b
y
f
ac
t
o
r
s
s
u
ch
as
clim
ate,
s
o
cio
-
ec
o
n
o
m
ic
ac
tiv
ities
,
an
d
tech
n
o
lo
g
ical
a
d
v
an
ce
m
en
ts
.
I
n
p
ar
ticu
lar
,
m
u
lti
-
co
u
n
tr
y
en
er
g
y
d
em
a
n
d
p
r
ed
ictio
n
p
r
esen
ts
a
u
n
iq
u
e
ch
allen
g
e,
as
ea
ch
co
u
n
tr
y
ex
h
ib
its
its
o
wn
co
n
s
u
m
p
tio
n
p
atter
n
s
an
d
r
esp
o
n
d
s
d
if
f
er
en
tly
to
v
ar
io
u
s
in
f
lu
en
cin
g
f
ac
to
r
s
.
E
f
f
ec
tiv
e
f
o
r
ec
asti
n
g
m
o
d
els
ar
e
th
er
ef
o
r
e
ess
en
tial
f
o
r
ac
h
iev
in
g
g
r
ea
ter
ac
cu
r
ac
y
an
d
ef
f
icien
cy
in
m
an
ag
in
g
en
e
r
g
y
r
eso
u
r
ce
s
at
th
e
r
e
g
io
n
al
an
d
i
n
ter
n
atio
n
al
lev
els.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
0
6
7
-
5
0
7
9
5068
On
e
s
ig
n
if
ican
t
ch
allen
g
e
in
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
is
th
e
p
r
esen
ce
o
f
o
u
tlier
s
,
wh
ich
ar
e
ab
n
o
r
m
al
d
ata
p
o
in
ts
th
at
d
ev
iate
s
ig
n
i
f
ican
tly
f
r
o
m
ex
p
ec
ted
p
atter
n
s
.
Ou
tlier
s
in
en
er
g
y
d
em
an
d
d
a
ta
m
ay
ar
is
e
d
u
e
to
m
ea
s
u
r
em
en
t
er
r
o
r
s
,
s
y
s
tem
f
ailu
r
es,
o
r
r
ar
e
ev
en
ts
s
u
ch
as
n
atu
r
al
d
is
aster
s
o
r
s
u
d
d
en
ch
an
g
es
in
co
n
s
u
m
p
tio
n
p
atter
n
s
.
T
h
ese
an
o
m
alo
u
s
v
alu
es,
if
lef
t
u
n
t
r
ea
ted
,
ca
n
s
ev
er
ely
im
p
ac
t
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
f
o
r
ec
asti
n
g
m
o
d
els,
lea
d
in
g
to
in
ac
cu
r
ate
p
r
ed
ictio
n
s
a
n
d
in
ef
f
icien
t
d
ec
is
io
n
-
m
ak
in
g
.
As
s
u
ch
,
ad
d
r
ess
in
g
o
u
tlier
s
is
a
v
ital
p
r
ep
r
o
ce
s
s
in
g
s
tep
f
o
r
en
h
an
cin
g
th
e
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
o
f
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
m
o
d
els.
Var
io
u
s
o
u
tlier
d
etec
t
io
n
tech
n
iq
u
es,
in
clu
d
in
g
s
tati
s
tical
m
eth
o
d
s
an
d
m
ac
h
in
e
le
ar
n
in
g
ap
p
r
o
ac
h
es,
h
av
e
b
ee
n
d
ev
el
o
p
ed
t
o
id
en
tif
y
an
d
m
itig
ate
th
e
ef
f
ec
ts
o
f
o
u
tlier
s
in
tim
e
-
s
er
ies d
ata.
I
n
r
ec
en
t
y
ea
r
s
,
th
e
ap
p
licat
io
n
o
f
m
ac
h
in
e
lear
n
in
g
m
o
d
els
s
u
ch
as
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
n
etwo
r
k
s
,
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN
s
)
,
an
d
atten
tio
n
m
ec
h
a
n
is
m
s
h
as
g
ain
ed
s
ig
n
if
ican
t
atten
tio
n
in
en
er
g
y
d
em
an
d
f
o
r
ec
asti
n
g
.
T
h
ese
m
o
d
els
h
a
v
e
d
em
o
n
s
tr
ated
th
e
ab
ilit
y
t
o
ca
p
tu
r
e
c
o
m
p
lex
tem
p
o
r
al
d
ep
e
n
d
en
cies a
n
d
p
r
o
v
id
e
h
ig
h
ly
ac
cu
r
ate
p
r
ed
icti
o
n
s
.
Ho
wev
er
,
th
e
p
e
r
f
o
r
m
an
c
e
o
f
th
ese
m
o
d
els is
h
ig
h
ly
s
en
s
itiv
e
to
th
e
q
u
ality
o
f
th
e
in
p
u
t
d
ata,
p
ar
ticu
lar
ly
in
th
e
p
r
esen
ce
o
f
o
u
tlier
s
.
Desp
ite
th
e
ad
v
an
ce
s
in
f
o
r
ec
asti
n
g
tech
n
iq
u
es,
th
e
im
p
ac
t
o
f
o
u
tlier
d
etec
tio
n
m
eth
o
d
s
o
n
m
o
d
el
p
er
f
o
r
m
an
ce
,
p
ar
ticu
lar
ly
f
o
r
m
u
lti
-
co
u
n
tr
y
en
er
g
y
d
em
a
n
d
p
r
ed
ictio
n
,
h
as
n
o
t
b
ee
n
e
x
ten
s
iv
ely
ex
p
lo
r
ed
.
I
n
th
is
co
n
tex
t,
th
e
p
r
esen
t
s
tu
d
y
aim
s
to
in
v
esti
g
ate
th
e
r
o
le
o
f
d
if
f
er
en
t
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es
in
im
p
r
o
v
in
g
t
h
e
ac
cu
r
ac
y
o
f
en
e
r
g
y
d
em
an
d
f
o
r
ec
asti
n
g
f
o
r
m
u
ltip
le
co
u
n
tr
ies.
T
h
is
s
tu
d
y
f
o
cu
s
es
o
n
ev
alu
atin
g
f
iv
e
p
o
p
u
lar
o
u
tlier
d
etec
t
io
n
tech
n
iq
u
es
—
Z
-
s
co
r
e,
d
e
n
s
ity
-
b
ased
s
p
atial
clu
s
ter
in
g
o
f
ap
p
licatio
n
s
with
n
o
is
e
(
DB
S
C
AN
)
,
is
o
latio
n
f
o
r
est
(
I
F)
,
lo
ca
l
o
u
tlier
f
ac
to
r
(
L
OF)
,
an
d
one
-
class
s
u
p
p
o
r
t
v
ec
to
r
m
a
ch
in
e
(
SVM
)
—
o
n
th
ei
r
ab
ili
ty
to
im
p
r
o
v
e
t
h
e
ac
cu
r
ac
y
o
f
en
er
g
y
d
em
an
d
p
r
ed
ictio
n
s
u
s
in
g
th
r
ee
s
tate
-
of
-
th
e
-
ar
t
tim
e
-
s
er
ies
f
o
r
ec
a
s
tin
g
m
o
d
els:
L
STM
,
C
NN
Au
to
en
co
d
e
r
s
,
an
d
L
STM
with
atten
tio
n
.
B
y
a
n
aly
zin
g
r
ea
l
-
tim
e
en
er
g
y
c
o
n
s
u
m
p
tio
n
d
ata
f
r
o
m
f
o
u
r
E
u
r
o
p
ea
n
c
o
u
n
tr
ies
—
Ger
m
an
y
,
Fra
n
ce
,
Sp
ain
,
an
d
I
taly
—
th
e
s
tu
d
y
p
r
o
v
id
es
in
s
ig
h
ts
in
to
th
e
ef
f
ec
tiv
en
ess
o
f
th
ese
o
u
tlier
d
etec
tio
n
m
eth
o
d
s
in
en
h
an
ci
n
g
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
T
h
e
r
esu
lts
o
f
th
is
s
tu
d
y
ar
e
ex
p
ec
ted
to
co
n
tr
ib
u
te
v
alu
ab
le
k
n
o
wled
g
e
to
th
e
f
ield
o
f
en
er
g
y
s
y
s
tem
s
,
o
f
f
e
r
in
g
a
b
etter
u
n
d
e
r
s
tan
d
in
g
o
f
th
e
r
elatio
n
s
h
ip
b
etwe
en
o
u
tlier
d
etec
tio
n
an
d
f
o
r
ec
asti
n
g
p
er
f
o
r
m
a
n
ce
,
with
im
p
licatio
n
s
f
o
r
b
o
th
ac
a
d
em
ic
r
esear
ch
an
d
p
r
ac
tical
ap
p
licatio
n
s
in
th
e
e
n
er
g
y
s
ec
to
r
.
T
h
e
p
ap
e
r
is
o
r
g
an
ized
as:
s
ec
tio
n
2
r
ev
iews
r
elate
d
wo
r
k
;
s
ec
tio
n
3
d
escr
ib
es
th
e
m
eth
o
d
o
lo
g
y
; sectio
n
4
p
r
esen
ts
an
d
d
is
cu
s
s
es
th
e
r
esu
lts
d
is
cu
s
s
io
n
s
;
s
ec
t
io
n
5
co
n
clu
d
es
th
e
p
ap
er
an
d
s
u
g
g
ests
f
u
tu
r
e
d
ir
e
ctio
n
s
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
Ou
tlier
d
etec
tio
n
in
tim
e
s
er
ie
s
d
ata
is
a
cr
itical
co
m
p
o
n
en
t
in
en
h
a
n
cin
g
t
h
e
ac
cu
r
ac
y
o
f
f
o
r
ec
asti
n
g
m
o
d
els,
p
ar
ticu
lar
l
y
in
a
p
p
lic
atio
n
s
lik
e
en
er
g
y
d
em
an
d
p
r
ed
ictio
n
.
Ma
n
y
s
tu
d
ies
em
p
h
asize
th
e
im
p
ac
t
o
f
s
elec
tin
g
ef
f
ec
tiv
e
o
u
tlier
d
et
ec
tio
n
tech
n
iq
u
es
o
n
th
e
p
er
f
o
r
m
an
ce
o
f
f
o
r
ec
asti
n
g
m
o
d
els.
T
h
e
s
tu
d
y
b
y
Am
alo
u
et
a
l.
[
1
]
p
r
esen
ts
t
h
e
f
ast
in
cr
em
en
tal
s
u
p
p
o
r
t
v
ec
to
r
d
ata
d
escr
ip
tio
n
(
FIS
VDD)
alg
o
r
ith
m
f
o
r
o
u
tlier
d
etec
tio
n
,
d
e
m
o
n
s
tr
ati
n
g
its
ef
f
ec
tiv
e
n
ess
in
en
e
r
g
y
tim
e
s
er
ies
f
o
r
ec
asti
n
g
.
T
h
e
r
esear
ch
h
ig
h
lig
h
ts
th
at
ch
o
o
s
in
g
th
e
a
p
p
r
o
p
r
iate
k
er
n
el
f
u
n
ctio
n
f
o
r
t
h
e
FISVDD
m
o
d
el
s
ig
n
if
ican
tly
im
p
r
o
v
es
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
T
h
is
im
p
r
o
v
em
en
t
i
s
v
alid
ated
u
s
in
g
th
e
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
ev
alu
atio
n
,
wh
ich
s
h
o
ws
th
at
FISVDD
o
u
tp
er
f
o
r
m
s
o
th
er
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es.
B
y
s
elec
tin
g
th
e
r
ig
h
t
k
er
n
el
f
u
n
ctio
n
,
t
h
e
m
eth
o
d
ef
f
ec
tiv
ely
h
an
d
les
ir
r
eg
u
lar
it
ies
in
en
er
g
y
co
n
s
u
m
p
tio
n
d
ata,
lead
in
g
to
s
u
p
er
i
o
r
r
esu
l
ts
in
m
u
lti
-
co
u
n
tr
y
en
er
g
y
d
em
an
d
f
o
r
ec
asti
n
g
.
B
an
d
h
an
an
d
Gan
a
p
ati
[
2
]
h
a
v
e
d
is
cu
s
s
ed
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es.
T
h
e
s
tu
d
y
class
if
ies
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es
in
to
f
iv
e
m
ajo
r
ca
teg
o
r
ies:
s
tatis
t
ical
m
eth
o
d
s
,
d
is
tan
ce
-
b
ased
ap
p
r
o
a
ch
es,
d
en
s
ity
-
b
ased
m
eth
o
d
s
,
clu
s
ter
in
g
-
b
ased
tec
h
n
iq
u
es,
an
d
en
s
em
b
le
m
et
h
o
d
s
.
E
ac
h
tech
n
iq
u
e
o
f
f
er
s
u
n
i
q
u
e
ad
v
a
n
tag
es
f
o
r
id
en
tify
in
g
an
o
m
al
o
u
s
d
ata
p
o
in
ts
.
T
h
e
ad
a
p
tab
ilit
y
o
f
th
e
m
eth
o
d
s
f
o
r
d
if
f
er
en
t
d
o
m
ain
s
o
f
th
e
s
tu
d
y
g
ain
s
in
ter
est
in
f
u
r
t
h
er
in
v
esti
g
atio
n
.
R
ich
ar
d
[
3
]
also
ex
p
l
o
r
es
t
h
e
v
ar
io
u
s
tech
n
iq
u
es
an
d
th
eir
ad
v
an
ta
g
es
an
d
lim
itatio
n
s
.
T
h
e
au
t
h
o
r
s
in
[
4
]
h
av
e
d
is
cu
s
s
ed
an
en
h
a
n
ce
d
tech
n
iq
u
e
,
ca
lled
u
n
s
u
p
er
v
is
ed
o
u
tlier
d
etec
tio
n
ar
ch
itectu
r
e
with
g
r
a
p
h
n
eu
r
al
n
etwo
r
k
(
UOSC
-
GNN)
.
T
h
e
au
th
o
r
s
o
f
r
e
f
er
en
ce
[
5
]
h
av
e
d
is
cu
s
s
ed
an
o
m
aly
d
etec
tio
n
tech
n
iq
u
es
an
d
co
m
p
ar
ed
t
h
e
m
eth
o
d
s
,
IF
,
g
a
u
s
s
ian
m
ix
tu
r
e
m
o
d
el
(
GM
M)
,
a
n
d
k
-
n
ea
r
e
s
t
n
eig
h
b
o
r
(
k
NN)
alg
o
r
ith
m
s
an
d
co
n
clu
d
es
th
at
IF
o
u
tp
er
f
o
r
m
s
b
o
th
GM
M
a
n
d
k
NN
in
e
f
f
ec
tiv
ely
is
o
lati
n
g
o
u
tlier
s
f
r
o
m
d
ata.
Ou
tlier
d
etec
tio
n
b
ased
o
n
lo
ca
l
d
en
s
ity
a
n
d
n
atu
r
al
n
eig
h
b
o
r
s
h
av
e
b
ee
n
d
is
cu
s
s
ed
in
[
6
]
,
wh
er
ein
a
k
n
o
wled
g
e
o
f
k
n
o
win
g
p
ar
a
m
eter
K,
f
o
r
ad
d
r
ess
in
g
ch
allen
g
es
in
ex
is
tin
g
m
eth
o
d
s
r
elate
d
to
p
a
r
am
eter
s
elec
tio
n
.
I
n
t
h
is
wo
r
k
,
m
an
u
al
p
ar
am
eter
s
ettin
g
r
eq
u
ir
e
d
f
o
r
n
eig
h
b
o
r
h
o
o
d
p
a
r
am
eter
K
is
n
o
t
r
eq
u
ir
ed
.
An
o
t
h
er
w
o
r
k
th
at
in
teg
r
ates
clu
s
ter
in
g
an
d
o
u
tlier
s
co
r
in
g
s
ch
em
es,
s
p
ec
if
ically
u
s
in
g
u
n
ce
r
tain
ty
s
o
f
t
clu
s
ter
in
g
b
ased
o
n
r
o
u
g
h
s
et
th
eo
r
y
is
r
ep
o
r
te
d
in
[
7
]
.
T
h
e
wo
r
k
d
is
cu
s
s
es
a
Ker
n
el
R
o
u
g
h
C
lu
s
ter
in
g
alg
o
r
ith
m
,
d
e
m
o
n
s
tr
atin
g
s
u
p
er
io
r
d
etec
tio
n
ac
cu
r
ac
y
co
m
p
ar
ed
t
o
f
iv
e
ex
is
tin
g
m
eth
o
d
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
I
mp
a
ct
o
f o
u
tlier
d
etec
tio
n
tec
h
n
iq
u
es o
n
time
-
s
eries
fo
r
ec
a
s
tin
g
a
cc
u
r
a
cy
fo
r
…
(
S
h
r
ey
a
s
K
a
r
n
ick
)
5069
A
s
in
g
le
d
en
s
ity
n
etwo
r
k
(
SD
N)
an
d
Z
-
s
co
r
e
f
o
r
o
u
tlier
d
et
ec
tio
n
in
a
n
alo
g
test
s
is
p
r
esen
ted
in
[
8
]
,
it
in
tr
o
d
u
ce
s
m
etr
ics
lik
e
s
elf
-
ex
clu
d
ed
f
ail
r
ate
(
SE
f
ail
r
ate)
an
d
n
o
r
m
alize
d
a
r
ea
u
n
d
er
cu
r
v
e
(
AUC)
to
q
u
an
tify
an
d
v
is
u
alize
ab
n
o
r
m
ality
ef
f
ec
tiv
ely
.
T
h
e
tech
n
iq
u
es
in
clu
d
e
I
Fs
,
wh
ich
u
tili
ze
b
in
ar
y
d
ec
is
io
n
tr
ee
s
to
is
o
late
an
o
m
alies,
cr
u
cial
f
o
r
v
ar
i
o
u
s
ap
p
licatio
n
s
[
9
]
,
[
1
0
]
.
I
n
[
1
1
]
,
an
elec
tr
icity
p
r
ice
f
o
r
ec
asti
n
g
o
f
Dan
is
h
elec
tr
icity
m
ar
k
et,
u
tili
zin
g
a
GM
M
-
lig
h
tweig
h
t
g
r
ad
ien
t
b
o
o
s
tin
g
m
ac
h
in
e
h
y
b
r
id
d
etec
to
r
a
n
d
L
STNe
t
-
k
er
n
el
d
en
s
ity
esti
m
atio
n
(
L
STNe
t
-
KDE
)
m
eth
o
d
,
wh
ich
en
h
an
ce
s
f
o
r
ec
asti
n
g
a
cc
u
r
ac
y
b
y
ef
f
ec
tiv
ely
is
o
latin
g
an
d
p
r
e
d
ictin
g
o
u
tlier
s
eq
u
en
ce
s
is
p
r
esen
ted
.
On
th
e
o
th
er
h
a
n
d
,
R
F
alg
o
r
it
h
m
f
o
r
o
u
tlier
d
etec
tio
n
is
p
r
e
s
en
ted
in
[
1
2
]
.
Sin
g
le
-
v
alu
e
d
m
etr
ic
p
r
e
d
ictio
n
is
p
r
esen
ted
in
[
1
3
]
,
wh
ich
en
h
a
n
ce
s
th
e
ac
cu
r
ac
y
o
f
tim
e
-
s
er
i
es
f
o
r
ec
asti
n
g
[
1
4
]
,
[
1
5
]
in
v
a
r
io
u
s
ap
p
licatio
n
s
,
in
clu
d
in
g
en
e
r
g
y
d
em
an
d
p
r
ed
ictio
n
.
An
o
th
er
alg
o
r
ith
m
,
ca
ll
ed
f
ast
in
cr
em
en
tal
FISVDD
i
s
d
is
cu
s
s
ed
in
[
1
6
]
f
o
r
en
h
an
cin
g
th
e
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
An
o
th
er
s
ig
n
if
ican
t
co
n
tr
ib
u
ti
o
n
to
th
is
f
ield
is
th
e
h
y
b
r
i
d
m
o
d
el
p
r
o
p
o
s
ed
b
y
So
n
g
h
u
a
[
1
7
]
,
wh
ich
co
m
b
in
es
I
F
with
o
u
tlier
r
ec
o
n
s
tr
u
ctio
n
(
OR
)
,
C
NN,
an
d
r
a
n
d
o
m
f
o
r
est
(
R
F)
f
o
r
en
er
g
y
d
em
an
d
f
o
r
ec
asti
n
g
.
T
h
is
m
o
d
el,
d
e
n
o
ted
as
I
F
-
OR
-
C
NN
-
R
F,
d
em
o
n
s
tr
ates
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
m
etr
ics,
s
u
ch
as
m
ea
n
ab
s
o
lu
te
er
r
o
r
(
MA
E
)
an
d
r
o
o
t
m
ea
n
s
q
u
ar
ed
e
r
r
o
r
(
R
MSE
)
,
co
m
p
ar
ed
t
o
o
th
e
r
C
NN
-
b
ased
m
o
d
els.
T
h
e
s
tu
d
y
u
n
d
er
s
co
r
es
th
at
in
teg
r
atin
g
o
u
tlier
d
etec
tio
n
m
eth
o
d
s
with
d
ee
p
lear
n
in
g
tech
n
iq
u
es
en
h
a
n
ce
s
th
e
r
o
b
u
s
tn
ess
o
f
f
o
r
ec
asti
n
g
m
o
d
els,
p
a
r
ticu
lar
ly
in
th
e
p
r
esen
ce
o
f
o
u
tl
ier
s
.
T
h
is
h
y
b
r
id
a
p
p
r
o
ac
h
a
d
d
r
ess
es
ch
allen
g
es
in
h
er
en
t
i
n
en
e
r
g
y
d
em
an
d
p
r
ed
ictio
n
b
y
m
itig
atin
g
th
e
i
n
f
lu
en
ce
o
f
ab
n
o
r
m
al
d
ata
p
o
in
ts
,
lead
in
g
to
m
o
r
e
r
eliab
le
an
d
ac
c
u
r
ate
f
o
r
ec
asts
.
I
n
a
s
im
ilar
v
ein
,
L
i
et
a
l.
[
1
8
]
p
r
o
p
o
s
es th
e
C
NN
-
g
ated
r
ec
u
r
r
en
t u
n
it (
C
NN
-
GR
U
)
m
eth
o
d
,
co
u
p
led
with
a
r
an
d
o
m
f
o
r
est
d
etec
tio
n
m
o
d
el
o
p
tim
ized
b
y
g
r
i
d
s
ea
r
ch
(
C
GA
-
R
F),
f
o
r
an
o
m
aly
d
etec
tio
n
in
en
er
g
y
co
n
s
u
m
p
tio
n
d
ata.
T
h
eir
s
tu
d
y
r
ev
ea
ls
s
ig
n
if
ican
t
im
p
r
o
v
e
m
en
ts
in
p
er
f
o
r
m
an
ce
m
etr
ic
s
s
u
ch
as
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
a
n
d
F1
s
co
r
e
,
co
m
p
a
r
ed
to
co
n
v
en
tio
n
al
m
eth
o
d
s
.
T
h
e
u
s
e
o
f
a
s
elf
-
atten
tiv
e
m
ec
h
an
is
m
in
th
e
C
NN
-
GR
U
m
o
d
el
h
elp
s
i
n
ca
p
tu
r
i
n
g
d
y
n
am
ic
ch
an
g
es
in
en
er
g
y
co
n
s
u
m
p
tio
n
,
wh
ile
th
e
r
an
d
o
m
f
o
r
est
m
o
d
el
ex
ce
ls
in
d
etec
tin
g
an
o
m
alies
in
r
esid
u
als,
u
ltima
tely
b
o
o
s
tin
g
f
o
r
ec
asti
n
g
ac
c
u
r
ac
y
.
T
h
e
a
u
th
o
r
s
em
p
h
asize
th
at
h
an
d
lin
g
a
n
o
m
alies
ef
f
ec
tiv
ely
is
cr
u
cial
f
o
r
en
h
an
cin
g
en
e
r
g
y
m
an
a
g
e
m
en
t
an
d
o
p
er
atio
n
al
ef
f
icien
cy
in
e
n
er
g
y
s
y
s
tem
s
.
Fu
et
a
l.
[
1
9
]
co
n
tr
i
b
u
te
to
th
i
s
ar
ea
b
y
p
r
esen
tin
g
a
tr
ee
-
b
ased
an
o
m
aly
d
etec
tio
n
m
o
d
el,
wh
ich
was
th
e
win
n
in
g
s
o
lu
tio
n
in
th
e
lar
g
e
-
s
ca
le
en
er
g
y
an
o
m
aly
d
ete
ctio
n
(
L
E
AD)
c
o
m
p
etitio
n
.
T
h
is
m
eth
o
d
ac
h
iev
ed
a
h
ig
h
R
OC
-
AUC
s
co
r
e
o
f
0
.
9
8
6
6
,
u
n
d
e
r
s
co
r
in
g
its
ef
f
icac
y
in
id
en
tify
in
g
o
u
tlier
s
in
en
er
g
y
tim
e
s
er
ies.
T
h
e
s
tu
d
y
em
p
h
asizes
th
e
im
p
o
r
t
an
ce
o
f
f
ea
tu
r
e
en
g
in
ee
r
in
g
,
p
ar
ticu
lar
ly
th
r
o
u
g
h
v
alu
e
-
ch
an
g
in
g
f
ea
tu
r
es
th
at
ca
p
tu
r
e
v
ar
iatio
n
s
in
tim
e
s
er
i
es
d
ata.
T
h
is
r
esear
ch
h
ig
h
lig
h
ts
th
e
n
ee
d
f
o
r
ef
f
ec
tiv
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
a
n
d
an
o
m
aly
d
etec
tio
n
to
e
n
s
u
r
e
th
e
ac
cu
r
ac
y
o
f
en
e
r
g
y
c
o
n
s
u
m
p
tio
n
f
o
r
ec
asti
n
g
m
o
d
els.
Simil
a
r
l
y
,
G
ao
e
t
a
l
.
[
2
0
]
ex
p
l
o
r
e
o
u
t
lie
r
d
et
ec
ti
o
n
t
h
r
o
u
g
h
c
o
r
r
elat
io
n
a
n
al
y
s
is
b
ase
d
o
n
g
r
ap
h
n
e
u
r
al
n
et
wo
r
k
s
(
GN
Ns)
.
T
h
ei
r
p
r
o
p
o
s
e
d
UOSC
-
GNN
a
r
c
h
it
ec
t
u
r
e
i
m
p
r
o
v
es
a
n
o
m
a
ly
d
et
ec
t
io
n
b
y
m
e
asu
r
i
n
g
t
h
e
v
a
r
i
an
ce
b
e
twe
e
n
e
x
p
ec
te
d
an
d
a
ct
u
a
l
d
at
a
s
ta
tes
,
s
h
o
w
in
g
i
m
p
r
o
v
e
m
e
n
ts
in
ac
cu
r
a
cy
an
d
s
e
n
s
iti
v
i
ty
.
Alt
h
o
u
g
h
th
e
s
t
u
d
y
d
o
es
n
o
t
d
i
r
e
ctl
y
a
d
d
r
ess
e
n
e
r
g
y
f
o
r
ec
a
s
tin
g
,
t
h
e
t
ec
h
n
iq
u
es
d
is
c
u
s
s
e
d
a
r
e
ap
p
l
ica
b
l
e
i
n
en
er
g
y
d
em
an
d
p
r
e
d
i
cti
o
n
,
es
p
ec
i
all
y
in
i
d
e
n
t
if
y
i
n
g
a
n
o
m
a
lo
u
s
p
a
tte
r
n
s
th
at
m
ay
i
n
f
l
u
e
n
c
e
f
o
r
ec
as
ti
n
g
m
o
d
e
ls
.
T
h
e
in
teg
r
atio
n
o
f
m
ac
h
in
e
le
ar
n
in
g
tech
n
iq
u
es
f
o
r
o
u
tlier
d
etec
tio
n
in
en
er
g
y
tim
e
s
er
ies
is
f
u
r
th
er
ex
p
lo
r
ed
in
th
e
w
o
r
k
o
f
I
s
m
a
ee
l
et
a
l.
[
2
1
]
,
wh
ich
i
n
v
esti
g
ates
th
e
s
cien
tific
c
o
m
p
u
tin
g
ass
o
ciate
s
(
SC
A)
s
tatis
t
ical
s
y
s
tem
f
o
r
o
u
tlier
d
etec
tio
n
in
th
e
c
o
n
tex
t
o
f
wat
er
v
o
lu
m
e
f
o
r
ec
asti
n
g
f
o
r
th
e
Do
h
u
k
Dam
.
W
h
ile
th
e
p
r
im
ar
y
f
o
cu
s
is
o
n
wate
r
v
o
lu
m
e
d
ata,
t
h
e
p
r
in
cip
les
o
f
o
u
tlier
c
o
r
r
ec
tio
n
in
tim
e
s
er
ies
an
aly
s
is
ca
n
b
e
ap
p
lied
to
e
n
er
g
y
f
o
r
ec
asti
n
g
[
2
2
]
.
T
h
e
p
ap
er
d
em
o
n
s
tr
ates
th
at
o
u
tlier
-
ad
ju
s
ted
f
o
r
ec
a
s
ts
p
er
f
o
r
m
b
etter
,
en
h
an
cin
g
th
e
ac
cu
r
ac
y
o
f
ti
m
e
-
s
er
ies
m
o
d
els
b
y
co
r
r
ec
ti
n
g
f
o
r
ab
n
o
r
m
al
d
ata
p
o
in
ts
t
h
at
wo
u
ld
o
th
e
r
wis
e
lead
to
f
o
r
ec
asti
n
g
er
r
o
r
s
.
Ky
o
[
2
3
]
h
av
e
p
r
esen
ted
a
m
u
lti
-
o
b
jectiv
e
o
p
tim
izatio
n
ap
p
r
o
ac
h
co
m
b
in
in
g
m
i
n
im
u
m
in
d
ex
o
f
s
y
m
m
etr
y
an
d
u
n
if
o
r
m
ity
(
I
SU)
an
d
m
ax
im
u
m
lik
elih
o
o
d
au
to
r
e
g
r
ess
iv
e
(
AR
)
m
o
d
elin
g
f
o
r
d
etec
tin
g
o
u
tlier
s
in
n
o
n
s
tatio
n
ar
y
tim
e
s
er
ies
to
d
ec
o
m
p
o
s
es
tr
en
d
an
d
s
tatio
n
ar
y
c
o
m
p
o
n
en
ts
wh
il
e
b
alan
cin
g
o
u
tlier
d
etec
tio
n
an
d
m
o
d
el
s
elec
tio
n
.
A
d
ee
p
lear
n
in
g
f
r
am
ewo
r
k
u
s
in
g
au
to
en
co
d
er
s
an
d
L
STM
n
etwo
r
k
s
to
d
etec
t
an
o
m
alies
in
tim
e
s
er
ies
d
ata
is
d
is
cu
s
s
ed
in
[
2
4
]
.
T
h
e
h
y
b
r
id
m
o
d
el
ca
p
tu
r
es
c
o
m
p
lex
tem
p
o
r
al
p
atter
n
s
th
r
o
u
g
h
r
ec
o
n
s
tr
u
ctio
n
er
r
o
r
s
,
en
h
a
n
cin
g
r
eliab
ilit
y
ac
r
o
s
s
ap
p
licatio
n
s
.
Ku
m
ar
et
a
l
.
[
2
5
]
d
ev
elo
p
ed
a
n
AR
I
MA
-
DC
GAN
s
y
n
er
g
y
th
a
t
lev
er
ag
es
AR
I
MA
’
s
lin
ea
r
m
o
d
elin
g
an
d
DC
GAN’
s
n
o
n
lin
ea
r
ca
p
a
b
ilit
ies
f
o
r
o
u
tlier
d
etec
tio
n
in
tim
e
s
er
ies.
T
h
is
ap
p
r
o
ac
h
o
u
t
p
er
f
o
r
m
s
ex
is
tin
g
m
eth
o
d
s
,
b
en
ef
it
in
g
ap
p
licatio
n
s
lik
e
f
r
au
d
d
etec
tio
n
an
d
p
r
ed
ictiv
e
m
ain
ten
an
ce
.
Da
n
i
et
a
l.
[
2
6
]
d
ev
el
o
p
ed
a
n
AR
I
MA
-
DC
GAN
s
y
n
er
g
y
th
at
lev
er
ag
es
AR
I
MA
’
s
l
in
ea
r
m
o
d
elin
g
an
d
DC
GAN’
s
n
o
n
lin
ea
r
ca
p
ab
ilit
ies
f
o
r
o
u
tlier
d
etec
tio
n
in
tim
e
s
er
ies.
T
h
is
ap
p
r
o
ac
h
o
u
tp
e
r
f
o
r
m
s
e
x
is
tin
g
m
eth
o
d
s
,
b
e
n
ef
itin
g
a
p
p
licatio
n
s
lik
e
f
r
au
d
d
etec
tio
n
a
n
d
p
r
ed
ictiv
e
m
ain
ten
an
ce
.
Dan
i
et
a
l
.
[
2
6
]
em
p
lo
y
s
p
r
in
cip
al
co
m
p
o
n
en
t
an
aly
s
is
(
P
C
A)
f
o
r
an
o
m
aly
d
etec
tio
n
in
tim
e
s
er
ies
b
y
r
ed
u
cin
g
d
im
en
s
io
n
ality
an
d
h
ig
h
lig
h
tin
g
d
ev
iatio
n
s
.
T
h
is
tech
n
iq
u
e
ai
d
s
in
ti
m
ely
r
is
k
m
itig
atio
n
an
d
in
f
o
r
m
ed
d
ec
is
io
n
-
m
a
k
in
g
in
o
r
g
a
n
izatio
n
al
co
n
tex
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
0
6
7
-
5
0
7
9
5070
C
u
r
r
en
t
liter
atu
r
e
ty
p
ically
f
o
cu
s
es
eith
er
o
n
f
o
r
ec
asti
n
g
m
o
d
els
(
lik
e
L
STM
o
r
C
N
Ns)
o
r
o
n
an
o
m
aly
d
etec
tio
n
m
eth
o
d
s
i
n
is
o
latio
n
.
Ou
r
wo
r
k
b
r
id
g
es
th
is
g
ap
b
y
d
em
o
n
s
tr
atin
g
h
o
w
o
u
tlier
d
etec
tio
n
d
ir
ec
tly
im
p
r
o
v
es
m
o
d
el
p
er
f
o
r
m
an
ce
a
n
d
q
u
an
tif
y
in
g
t
h
ese
im
p
r
o
v
em
en
ts
ac
r
o
s
s
m
o
d
els
an
d
co
u
n
tr
ies.
I
t
also
r
ev
ea
ls
m
o
d
el
-
s
p
ec
if
ic
s
en
s
itiv
ities
—
f
o
r
ex
am
p
le,
L
S
T
M
-
Atten
tio
n
m
o
d
els,
wh
ile
p
o
wer
f
u
l,
ar
e
m
o
r
e
s
en
s
itiv
e
to
o
u
tlier
s
,
a
n
u
an
ce
n
o
t
well
d
o
cu
m
e
n
ted
b
e
f
o
r
e.
T
h
u
s
,
th
is
m
a
n
u
s
cr
ip
t
ad
d
s
p
r
ac
tical
k
n
o
wled
g
e
f
o
r
b
o
t
h
r
esear
ch
er
s
an
d
p
r
a
ctitio
n
er
s
o
n
h
o
w
to
s
elec
t
an
d
co
m
b
i
n
e
an
o
m
al
y
d
etec
tio
n
an
d
f
o
r
ec
asti
n
g
m
eth
o
d
s
ef
f
ec
tiv
ely
,
co
n
t
r
ib
u
t
in
g
to
m
o
r
e
r
esil
ien
t a
n
d
a
cc
u
r
ate
en
er
g
y
s
y
s
tem
o
p
er
atio
n
s
.
3.
M
E
T
H
O
DO
L
O
G
Y
I
n
th
is
wo
r
k
,
o
u
r
m
eth
o
d
o
l
o
g
ical
f
r
am
ewo
r
k
in
v
o
l
v
es:
i)
r
eal
-
wo
r
ld
en
e
r
g
y
c
o
n
s
u
m
p
tio
n
d
atasets
f
r
o
m
f
o
u
r
E
u
r
o
p
ea
n
co
u
n
tr
ie
s
,
ii)
s
y
s
tem
atic
p
r
ep
r
o
ce
s
s
in
g
(
n
o
r
m
aliza
tio
n
,
in
ter
p
o
latio
n
,
o
u
tlier
r
em
o
v
al)
,
iii)
m
ath
em
atica
l
f
o
r
m
u
latio
n
an
d
im
p
lem
e
n
tatio
n
o
f
ea
c
h
f
o
r
ec
asti
n
g
an
d
o
u
tlier
d
etec
tio
n
tech
n
iq
u
e
a
n
d
iv
)
e
v
alu
atio
n
m
etr
ics
m
ea
n
a
b
s
o
lu
te
p
er
ce
n
tag
e
e
r
r
o
r
(
m
ea
n
a
b
s
o
lu
te
p
er
ce
n
tag
e
er
r
o
r
(
MA
PE
)
,
R
MSE
,
MA
E
)
to
q
u
an
tif
y
im
p
r
o
v
em
e
n
ts
in
p
r
ed
ictiv
e
ac
cu
r
ac
y
.
T
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
o
l
o
g
y
f
o
r
e
n
e
r
g
y
d
e
m
a
n
d
f
o
r
e
c
a
s
t
i
n
g
,
a
s
i
l
l
u
s
t
r
a
t
e
d
i
n
F
i
g
u
r
e
1
,
f
o
l
l
o
w
s
a
s
t
r
u
c
t
u
r
e
d
a
p
p
r
o
a
c
h
t
o
p
r
o
c
e
s
s
m
u
l
t
i
-
c
o
u
n
t
r
y
e
n
e
r
g
y
d
e
m
a
n
d
d
a
t
a
s
e
t
s
e
f
f
i
c
ie
n
t
l
y
.
T
h
e
p
r
o
c
e
s
s
b
e
g
i
n
s
w
i
t
h
d
at
a
p
r
e
p
r
o
c
e
s
s
i
n
g
,
w
h
e
r
e
n
o
r
m
a
l
i
za
t
i
o
n
a
n
d
t
r
a
i
n
-
t
e
s
t
s
p
l
i
tt
i
n
g
a
r
e
p
e
r
f
o
r
m
e
d
t
o
s
t
a
n
d
a
r
d
i
z
e
t
h
e
d
a
t
a
s
e
t
f
o
r
m
o
d
e
l
t
r
a
i
n
i
n
g
.
S
u
b
s
e
q
u
e
n
t
l
y
,
o
u
t
li
e
r
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
es
s
u
c
h
a
s
Z
-
s
c
o
r
e
,
D
B
SC
A
N
,
IF
,
L
O
F
,
a
n
d
S
V
M
a
r
e
a
p
p
l
i
e
d
t
o
i
d
e
n
t
i
f
y
a
n
o
m
a
l
o
u
s
d
a
t
a
p
o
in
t
s
.
A
d
e
c
is
i
o
n
-
m
a
k
i
n
g
s
t
e
p
d
et
e
r
m
i
n
e
s
w
h
e
t
h
e
r
o
u
t
l
ie
r
s
a
r
e
p
r
e
s
e
n
t
,
i
n
f
l
u
e
n
c
i
n
g
t
h
e
c
h
o
i
c
e
o
f
f
o
r
e
c
as
ti
n
g
m
o
d
e
l
s
.
I
f
o
u
tl
i
e
r
s
a
r
e
d
et
e
c
t
e
d
,
a
d
v
a
n
c
e
d
m
o
d
e
l
s
s
u
c
h
as
L
S
T
M
,
C
N
N
A
u
t
o
e
n
c
o
d
e
r
,
a
n
d
L
S
T
M
w
i
t
h
A
t
te
n
t
i
o
n
m
e
ch
a
n
i
s
m
a
r
e
u
t
i
li
z
e
d
t
o
e
n
h
a
n
c
e
f
o
r
e
c
a
s
t
i
n
g
ac
c
u
r
a
c
y
.
T
h
es
e
m
o
d
e
l
s
l
e
v
e
r
a
g
e
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
s
t
o
c
a
p
t
u
r
e
t
e
m
p
o
r
a
l
d
e
p
e
n
d
e
n
c
i
e
s
a
n
d
c
o
m
p
l
e
x
p
a
t
t
e
r
n
s
w
i
t
h
i
n
t
h
e
d
a
t
a
s
e
t
.
M
o
d
el
p
e
r
f
o
r
m
a
n
c
e
i
s
e
v
a
l
u
a
t
e
d
u
s
i
n
g
R
MS
E
,
MA
P
E
,
a
n
d
MS
E
t
o
e
n
s
u
r
e
r
o
b
u
s
t
p
r
e
d
i
c
ti
v
e
a
c
c
u
r
a
c
y
.
T
h
e
f
i
n
a
l
s
t
e
p
c
o
n
s
o
l
i
d
a
t
es
t
h
e
f
o
r
e
c
as
ti
n
g
r
e
s
u
l
t
s
,
p
r
o
v
i
d
i
n
g
i
n
s
i
g
h
t
s
i
n
t
o
e
n
e
r
g
y
d
e
m
a
n
d
t
r
e
n
d
s
a
c
r
o
s
s
m
u
lt
i
p
l
e
r
e
g
i
o
n
s
.
Fig
u
r
e
1
.
Flo
w
ch
a
r
t o
f
t
h
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
f
o
r
e
n
er
g
y
d
em
an
d
f
o
r
ec
asti
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
I
mp
a
ct
o
f o
u
tlier
d
etec
tio
n
tec
h
n
iq
u
es o
n
time
-
s
eries
fo
r
ec
a
s
tin
g
a
cc
u
r
a
cy
fo
r
…
(
S
h
r
ey
a
s
K
a
r
n
ick
)
5071
3
.
1
.
Da
t
a
co
llect
io
n
a
nd
p
re
pro
ce
s
s
ing
3
.
1
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
co
m
p
r
is
es
h
o
u
r
ly
en
e
r
g
y
c
o
n
s
u
m
p
tio
n
d
ata
f
r
o
m
m
u
ltip
le
E
u
r
o
p
ea
n
n
atio
n
s
,
s
p
ec
if
ically
Ger
m
an
y
(
DE
)
,
Fra
n
ce
(
FR
)
,
Sp
ain
(
E
S),
an
d
I
taly
(
I
T
)
.
T
h
e
d
ata
was
ac
q
u
ir
ed
f
r
o
m
th
e
o
p
en
p
o
wer
s
y
s
tem
d
ata
(
O
PS
D)
r
ep
o
s
ito
r
y
,
a
wid
ely
r
ec
o
g
n
ized
s
o
u
r
ce
f
o
r
en
er
g
y
-
r
elate
d
tim
e
-
s
er
ies
d
atasets
.
T
h
is
d
atase
t
p
r
o
v
id
es
h
is
to
r
ical
elec
tr
icity
d
em
an
d
v
alu
es
r
ec
o
r
d
ed
at
an
h
o
u
r
ly
r
eso
lu
tio
n
,
en
s
u
r
in
g
a
g
r
an
u
lar
u
n
d
e
r
s
tan
d
in
g
o
f
e
n
er
g
y
co
n
s
u
m
p
tio
n
tr
en
d
s
.
Giv
en
th
e
d
y
n
am
ic
n
atu
r
e
o
f
en
e
r
g
y
d
em
an
d
,
f
ac
t
o
r
s
s
u
ch
as
s
ea
s
o
n
al
v
ar
iatio
n
s
,
ec
o
n
o
m
ic
ac
tiv
ity
,
a
n
d
clim
atic
co
n
d
itio
n
s
in
f
lu
e
n
ce
co
n
s
u
m
p
tio
n
p
atter
n
s
.
T
h
u
s
,
an
in
-
d
e
p
th
ex
p
lo
r
atio
n
o
f
th
es
e
v
ar
iatio
n
s
is
n
ec
ess
ar
y
f
o
r
e
n
h
an
cin
g
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
an
d
r
o
b
u
s
tn
ess
.
3
.
1
.
2
.
Da
t
a
prepro
ce
s
s
ing
Han
d
lin
g
m
is
s
in
g
v
alu
es:
m
i
s
s
in
g
tim
estam
p
s
ar
e
f
illed
u
s
in
g
lin
ea
r
in
ter
p
o
latio
n
,
w
h
ile
en
er
g
y
d
em
an
d
v
alu
es
a
r
e
im
p
u
ted
u
s
in
g
f
o
r
wa
r
d
a
n
d
b
ac
k
war
d
f
illi
n
g
.
Data
n
o
r
m
aliza
tio
n
:
t
o
en
s
u
r
e
co
n
s
is
ten
cy
ac
r
o
s
s
d
if
f
er
en
t
co
u
n
tr
ies
an
d
elim
in
ate
s
ca
le
d
is
p
ar
ities
,
th
e
r
aw
tim
e
-
s
er
ies
d
ata
u
n
d
er
g
o
es
n
o
r
m
aliza
tio
n
.
T
h
e
Min
-
Ma
x
n
o
r
m
aliza
tio
n
tech
n
iq
u
e
is
ap
p
lied
,
wh
ich
s
c
ales
v
alu
es
b
etwe
en
0
an
d
1
,
th
er
eb
y
f
ac
ilit
atin
g
s
tab
le
co
n
v
er
g
en
ce
in
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
e
tr
an
s
f
o
r
m
atio
n
o
f
an
en
e
r
g
y
d
em
a
n
d
v
a
lu
e
is
co
m
p
u
te
d
as (
1
)
.
′
=
−
−
(
1
)
I
n
(
1
)
,
an
d
r
ep
r
esen
t
th
e
m
in
im
u
m
an
d
m
ax
im
u
m
v
alu
e
s
with
in
th
e
d
ataset,
r
esp
ec
tiv
ely
.
T
h
is
n
o
r
m
aliza
tio
n
m
itig
ates
th
e
i
m
p
ac
t
o
f
lar
g
e
-
s
ca
le
d
is
cr
ep
a
n
cies
am
o
n
g
d
if
f
er
en
t
co
u
n
tr
i
es
wh
ile
p
r
eser
v
in
g
th
e
r
elativ
e
m
ag
n
itu
d
e
o
f
f
lu
ct
u
atio
n
s
,
en
s
u
r
in
g
o
p
tim
al
m
o
d
el
p
er
f
o
r
m
an
ce
.
3
.
2
.
O
utlier
det
ec
t
i
o
n t
ec
hn
iqu
es
Ou
tlier
s
in
tim
e
-
s
er
ies
d
ata
ar
is
e
d
u
e
to
v
ar
io
u
s
f
ac
t
o
r
s
,
in
clu
d
in
g
s
en
s
o
r
m
alf
u
n
ctio
n
s
,
er
r
o
n
e
o
u
s
r
ec
o
r
d
in
g
s
,
g
r
id
f
ailu
r
es,
o
r
u
n
f
o
r
eseen
s
p
ik
es
in
en
er
g
y
co
n
s
u
m
p
tio
n
.
Failu
r
e
to
ad
d
r
ess
th
ese
an
o
m
alies
ca
n
lead
to
in
ac
cu
r
ate
p
r
ed
ictio
n
s
an
d
m
o
d
el
in
s
tab
ilit
y
.
T
h
i
s
s
tu
d
y
ex
am
in
es
f
iv
e
r
o
b
u
s
t
o
u
tlier
d
etec
tio
n
m
eth
o
d
s
Z
-
s
co
r
e
,
DB
SC
AN,
IF
,
L
OF,
an
d
one
-
class
SVM.
E
ac
h
tech
n
iq
u
e
id
en
tifie
s
a
n
o
m
alies
b
ased
o
n
d
is
tin
ct
m
ath
em
atica
l f
o
r
m
u
lat
io
n
s
an
d
u
n
d
er
ly
i
n
g
p
r
in
cip
les
.
3
.
2
.
1
.
Z
-
Sco
re
m
et
ho
d
T
h
e
Z
-
s
co
r
e
m
eth
o
d
is
a
s
tat
is
tical
tech
n
iq
u
e
th
at
q
u
an
tifie
s
th
e
d
ev
iatio
n
o
f
ea
ch
d
ata
p
o
i
n
t f
r
o
m
th
e
m
ea
n
in
ter
m
s
o
f
s
tan
d
a
r
d
d
ev
iatio
n
s
.
T
h
is
ap
p
r
o
ac
h
ass
u
m
e
s
th
at
en
er
g
y
co
n
s
u
m
p
tio
n
d
at
a
f
o
llo
ws
a
n
o
r
m
al
d
is
tr
ib
u
tio
n
,
allo
win
g
t
h
e
id
e
n
tific
atio
n
o
f
ex
tr
e
m
e
d
ev
iatio
n
s
.
T
h
e
Z
-
s
co
r
e
f
o
r
ea
c
h
v
alu
e
o
f
is
co
m
p
u
ted
as (
2
)
.
=
−
µ
(
2
)
I
n
(
2
)
,
µ
is
th
e
m
ea
n
an
d
is
th
e
s
tan
d
ar
d
d
e
v
iatio
n
.
Data
p
o
in
t
s
with
|
|
>
3
ar
e
class
if
ied
as o
u
tlier
s
.
3
.
2
.
2
.
DB
SCAN
DB
S
C
AN
i
s
a
clu
s
ter
in
g
-
b
ased
an
o
m
aly
d
etec
tio
n
tec
h
n
iq
u
e
th
at
d
is
tin
g
u
is
h
es
n
o
r
m
al
an
d
an
o
m
alo
u
s
p
o
i
n
ts
b
ased
o
n
d
ata
d
en
s
ity
.
A
p
o
in
t
is
co
n
s
id
er
ed
an
o
u
tlier
if
it
d
o
es
n
o
t
b
elo
n
g
to
an
y
h
ig
h
-
d
en
s
ity
clu
s
ter
.
T
h
e
alg
o
r
ith
m
r
elies o
n
a
n
eig
h
b
o
r
h
o
o
d
f
u
n
ct
io
n
(
3
)
.
(
)
=
{
∈
|
(
,
)
≤
}
(
3
)
I
n
(
3
)
,
(
,
)
d
en
o
tes
th
e
d
is
tan
ce
b
etwe
en
d
ata
p
o
in
ts
,
an
d
is
a
p
r
ed
ef
i
n
ed
th
r
esh
o
ld
.
Po
in
ts
with
f
ewe
r
th
an
m
in
s
am
p
les
n
eig
h
b
o
r
s
ar
e
lab
eled
as
o
u
tlier
s
.
T
h
is
m
eth
o
d
is
p
ar
ticu
lar
ly
ef
f
e
ctiv
e
f
o
r
d
etec
tin
g
an
o
m
alies in
d
atasets
ex
h
ib
itin
g
n
o
n
lin
ea
r
s
tr
u
ctu
r
es.
3
.
2
.
3
.
IF
I
F
is
an
en
s
em
b
le
lear
n
in
g
tech
n
iq
u
e
th
at
is
o
lates
an
o
m
alies
b
y
r
ec
u
r
s
iv
ely
p
ar
titi
o
n
i
n
g
d
a
ta
p
o
in
ts
.
Un
lik
e
tr
ad
itio
n
al
m
et
h
o
d
s
th
a
t
r
ely
o
n
d
is
tan
ce
m
etr
ics,
I
F
c
o
n
s
tr
u
cts
d
ec
is
io
n
tr
ee
s
wh
er
e
an
o
m
alo
u
s
p
o
in
ts
ar
e
id
en
tifie
d
th
r
o
u
g
h
s
h
o
r
ter
p
ath
len
g
th
s
.
T
h
e
an
o
m
aly
s
co
r
e
is
g
iv
en
b
y
(
4
)
.
(
,
)
=
2
−
(
ℎ
(
)
)
(
(
)
)
(
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
0
6
7
-
5
0
7
9
5072
I
n
(
4
)
,
ℎ
(
)
d
ep
en
d
s
o
n
in
th
e
f
o
r
est
an
d
(
)
is
th
e
av
er
ag
e
p
at
h
len
g
th
f
o
r
a
d
ataset
o
f
s
ize
.
I
F
i
s
co
m
p
u
tatio
n
ally
ef
f
icien
t a
n
d
h
ig
h
ly
ef
f
ec
tiv
e
f
o
r
h
ig
h
-
d
im
e
n
s
io
n
al
d
ata.
3
.
2
.
4
.
L
O
F
L
OF
ass
es
s
es
o
u
tlier
s
b
y
co
m
p
ar
in
g
t
h
e
d
e
n
s
ity
o
f
a
p
o
in
t
with
its
s
u
r
r
o
u
n
d
in
g
n
eig
h
b
o
r
s
.
A
lo
w
-
d
en
s
ity
p
o
in
t
r
elativ
e
to
its
n
ei
g
h
b
o
r
s
is
f
lag
g
ed
as a
n
o
u
tlier
.
T
h
e
L
OF sco
r
e
is
co
m
p
u
ted
a
s
(
5
)
.
(
)
=
∑
(
)
(
)
∈
(
)
|
(
)
|
(
5
)
I
n
(
5
)
,
(
)
r
ep
r
esen
ts
th
e
lo
ca
l
r
ea
ch
ab
ilit
y
d
e
n
s
ity
,
an
d
(
)
is
th
e
s
et
o
f
k
-
n
ea
r
est
n
eig
h
b
o
r
s
.
T
h
is
ap
p
r
o
ac
h
is
ad
v
an
ta
g
eo
u
s
f
o
r
d
etec
tin
g
s
u
b
tle
an
o
m
alies in
d
y
n
am
ic
e
n
v
ir
o
n
m
en
ts
.
3
.
2
.
5
.
O
ne
-
c
la
s
s
SVM
On
e
-
c
lass
SVM
co
n
s
tr
u
cts a
h
y
p
er
p
lan
e
th
at
d
if
f
e
r
en
tiates n
o
r
m
al
in
s
tan
ce
s
f
r
o
m
o
u
tlier
s
u
s
in
g
k
er
n
el
tr
an
s
f
o
r
m
atio
n
s
.
T
h
e
o
b
jectiv
e
f
u
n
ctio
n
f
o
r
an
o
m
aly
d
etec
tio
n
is
f
o
r
m
u
lated
as (
6
)
.
min
,
,
1
2
|
|
|
|
2
+
1
ν
∑
ξ
=
1
−
(
6
)
I
n
(
6
)
,
r
eg
u
lates
th
e
p
r
o
p
o
r
ti
o
n
o
f
o
u
tlier
s
.
one
-
class
SVM
is
p
ar
ticu
lar
ly
u
s
ef
u
l
f
o
r
d
atasets
w
ith
co
m
p
lex
d
is
tr
ib
u
tio
n
s
.
3
.
3
.
F
o
re
ca
s
t
ing
m
o
dels
3
.
3
.
1
.
L
T
S
M
T
h
e
L
STM
n
etwo
r
k
s
ar
e
an
ad
v
an
ce
d
v
a
r
ian
t
o
f
r
ec
u
r
r
e
n
t
n
eu
r
al
n
etwo
r
k
s
(
R
NNs)
s
p
ec
if
ically
d
esig
n
ed
to
ad
d
r
ess
th
e
v
an
i
s
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
th
at
h
in
d
er
s
tr
ad
itio
n
al
R
NNs
in
ca
p
tu
r
in
g
lo
n
g
-
ter
m
d
ep
en
d
e
n
cies.
L
STM
s
h
av
e
g
ain
ed
s
ig
n
if
ica
n
t
tr
ac
tio
n
in
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
,
n
atu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
,
an
d
s
eq
u
en
tial
d
a
ta
m
o
d
elin
g
d
u
e
to
th
eir
ab
ili
ty
to
r
etain
ess
en
tial
in
f
o
r
m
at
io
n
o
v
er
ex
ten
d
ed
tim
e
in
ter
v
als.
Un
lik
e
co
n
v
en
tio
n
al
R
NNs,
L
STM
s
u
tili
ze
m
em
o
r
y
ce
lls
an
d
s
p
ec
ialized
g
atin
g
m
ec
h
an
is
m
s
th
at
s
elec
tiv
ely
s
to
r
e
o
r
d
is
ca
r
d
in
f
o
r
m
atio
n
,
e
n
ab
lin
g
m
o
r
e
ef
f
ec
tiv
e
lear
n
in
g
f
r
o
m
lo
n
g
-
r
an
g
e
d
e
p
en
d
e
n
cies.
T
h
e
ar
c
h
itectu
r
e
o
f
an
L
STM
as
s
h
o
wn
in
Fig
u
r
e
2
co
n
s
is
t
s
o
f
m
e
m
o
r
y
ce
lls
r
e
g
u
lat
ed
b
y
th
r
ee
f
u
n
d
am
e
n
tal
g
ates:
th
e
f
o
r
g
et
g
ate,
wh
ich
d
eter
m
i
n
es
th
e
r
et
en
tio
n
o
f
p
ast
in
f
o
r
m
atio
n
;
th
e
in
p
u
t
g
ate,
wh
ich
co
n
tr
o
ls
th
e
in
teg
r
atio
n
o
f
n
e
w
in
f
o
r
m
atio
n
;
an
d
th
e
o
u
tp
u
t
g
ate,
wh
ich
d
ictates
th
e
tr
an
s
m
is
s
io
n
o
f
r
elev
an
t
in
f
o
r
m
atio
n
to
th
e
n
ex
t
tim
e
s
tep
.
T
h
ese
g
ates
co
llectiv
e
ly
m
an
ag
e
th
e
f
lo
w
o
f
in
f
o
r
m
atio
n
with
in
th
e
n
etwo
r
k
,
th
er
eb
y
m
itig
atin
g
is
s
u
es
ass
o
ciate
d
with
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies.
T
h
e
m
ath
em
atica
l
f
o
r
m
u
latio
n
o
f
th
ese
g
ates
en
s
u
r
es
th
at
th
e
m
o
d
el
lear
n
s
an
d
ad
a
p
ts
ef
f
e
ctiv
ely
to
s
eq
u
en
tial
p
atter
n
s
in
th
e
d
ata,
m
ak
in
g
L
STM
s
p
ar
ticu
lar
ly
s
u
itab
le
f
o
r
ap
p
licatio
n
s
in
v
o
lv
in
g
tem
p
o
r
al
d
ep
e
n
d
en
cies.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
L
ST
M
lay
er
F
o
r
g
e
t
g
a
t
e
:
t
h
e
f
o
r
g
e
t
g
at
e
r
e
g
u
l
a
t
e
s
w
h
e
t
h
e
r
i
n
f
o
r
m
a
t
i
o
n
f
r
o
m
p
r
e
v
i
o
u
s
t
i
m
e
s
t
e
p
s
s
h
o
u
l
d
b
e
r
e
t
a
i
n
e
d
o
r
d
i
s
c
a
r
d
e
d
b
a
s
e
d
o
n
t
h
e
c
u
r
r
en
t
i
n
p
u
t
a
n
d
t
h
e
p
r
e
v
i
o
u
s
h
i
d
d
en
s
t
a
t
e
.
M
a
t
h
e
m
at
i
c
al
l
y
,
i
t
i
s
d
ef
i
n
e
d
a
s
(
7
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
I
mp
a
ct
o
f o
u
tlier
d
etec
tio
n
tec
h
n
iq
u
es o
n
time
-
s
eries
fo
r
ec
a
s
tin
g
a
cc
u
r
a
cy
fo
r
…
(
S
h
r
ey
a
s
K
a
r
n
ick
)
5073
=
σ
(
.
[
ℎ
−
1
,
]
+
)
(
7
)
I
n
(
7
)
,
r
ep
r
esen
ts
th
e
f
o
r
g
et
g
ate
ac
tiv
atio
n
,
an
d
σ
is
th
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
t
h
at
co
n
s
tr
ain
s
v
alu
e
s
b
etwe
en
0
a
n
d
1
.
A
v
alu
e
clo
s
e
to
0
r
esu
lts
in
f
o
r
g
ettin
g
p
ast
in
f
o
r
m
atio
n
,
wh
ile
a
v
alu
e
n
e
ar
1
r
etain
s
it.
T
h
is
m
ec
h
an
is
m
en
s
u
r
es
th
at
ir
r
ele
v
an
t
in
f
o
r
m
atio
n
d
o
es
n
o
t
ac
c
u
m
u
late
in
th
e
m
em
o
r
y
ce
ll,
t
h
er
eb
y
im
p
r
o
v
in
g
m
o
d
el
ef
f
icien
c
y
.
I
n
p
u
t
g
ate:
T
h
e
in
p
u
t
g
ate
g
o
v
er
n
s
th
e
ex
ten
t
to
wh
ich
n
ew
i
n
f
o
r
m
atio
n
is
in
co
r
p
o
r
ated
in
t
o
th
e
ce
ll
s
tate.
I
t
o
p
er
ates
b
y
g
e
n
er
atin
g
a
ca
n
d
id
ate
m
em
o
r
y
v
alu
e
t
h
r
o
u
g
h
a
ℎ
ac
tiv
atio
n
f
u
n
ctio
n
an
d
s
ca
lin
g
it
u
s
in
g
a
s
ig
m
o
id
g
ate
to
r
e
g
u
la
te
its
in
f
lu
en
ce
.
T
h
e
c
o
r
r
esp
o
n
d
in
g
(
8
)
an
d
(
9
)
.
=
σ
(
.
[
ℎ
−
1
,
]
+
)
(
8
)
̃
=
ℎ
(
.
[
ℎ
−
1
,
]
+
)
(
9
)
I
n
(
8
)
–
(
9
)
,
r
ep
r
esen
ts
th
e
in
p
u
t
g
ate
ac
tiv
atio
n
,
a
n
d
̃
d
en
o
tes
th
e
ca
n
d
id
ate
m
em
o
r
y
co
n
ten
t.
T
h
is
co
n
tr
o
lled
u
p
d
ate
m
ec
h
a
n
is
m
en
s
u
r
es
th
at
o
n
ly
r
elev
a
n
t
n
ew
in
f
o
r
m
atio
n
is
ad
d
ed
to
th
e
m
em
o
r
y
,
p
r
ev
en
ti
n
g
u
n
n
ec
ess
ar
y
f
lu
ct
u
atio
n
s
in
th
e
lear
n
in
g
p
r
o
ce
s
s
.
C
ell
s
tate
u
p
d
ate
:
T
h
e
ce
ll
s
tate
s
er
v
es
as
th
e
m
em
o
r
y
o
f
an
L
STM
u
n
it
a
n
d
is
u
p
d
ate
d
b
y
co
m
b
in
in
g
r
etain
ed
p
ast in
f
o
r
m
atio
n
with
n
ew
in
p
u
ts
.
T
h
e
u
p
d
ate
eq
u
ati
o
n
is
g
iv
en
b
y
(
1
0
)
.
=
⊙
−
1
+
⊙
̃
(
1
0
)
I
n
(
1
0
)
,
⊙
r
ep
r
esen
ts
th
e
elem
en
t
-
wis
e
p
r
o
d
u
ct.
T
h
e
i
n
clu
s
io
n
o
f
th
e
f
o
r
g
et
g
ate
en
s
u
r
e
s
th
at
lo
n
g
-
ter
m
d
ep
en
d
e
n
cies
ar
e
p
r
eser
v
ed
w
h
ile
allo
win
g
n
ew,
r
elev
an
t
d
ata
to
b
e
in
co
r
p
o
r
ate
d
ef
f
ec
ti
v
ely
.
T
h
is
d
y
n
am
ic
b
alan
ce
b
etwe
en
m
em
o
r
y
r
ete
n
tio
n
a
n
d
u
p
d
ate
is
k
ey
to
L
STM
'
s
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
in
h
an
d
lin
g
s
eq
u
en
tial
d
ata.
Ou
tp
u
t
g
ate:
t
h
e
o
u
tp
u
t
g
ate
d
eter
m
in
es
h
o
w
m
u
c
h
o
f
th
e
u
p
d
ated
ce
ll
s
tate
co
n
tr
ib
u
tes
to
t
h
e
h
id
d
e
n
s
tate
an
d
,
co
n
s
eq
u
e
n
tly
,
th
e
f
in
al
o
u
tp
u
t o
f
th
e
n
etwo
r
k
.
T
h
i
s
p
r
o
ce
s
s
is
g
o
v
er
n
ed
b
y
(
1
1
)
an
d
(
1
2
)
.
=
σ
(
.
[
ℎ
−
1
,
]
+
)
(
1
1
)
ℎ
=
⊙
ℎ
(
)
(
1
2
)
I
n
(
1
1
)
an
d
(
1
2
)
,
ar
e
t
h
e
o
u
tp
u
t
g
ate
ac
tiv
atio
n
.
T
h
e
tan
h
a
ctiv
atio
n
en
s
u
r
es
th
at
th
e
o
u
tp
u
t
v
alu
es
r
e
m
ain
with
in
a
m
an
ag
ea
b
le
r
an
g
e,
th
er
eb
y
p
r
e
v
en
tin
g
e
x
tr
em
e
f
lu
ctu
atio
n
s
.
T
h
is
s
elec
tiv
e
in
f
o
r
m
atio
n
tr
a
n
s
f
er
en
h
an
ce
s
th
e
m
o
d
el’
s
ab
ilit
y
t
o
g
en
er
ate
m
ea
n
in
g
f
u
l r
e
p
r
esen
tatio
n
s
o
f
s
eq
u
e
n
tial d
ata.
3
.
3
.
2
.
CNN
Aut
o
enco
der
C
NN
Au
to
en
co
d
er
is
a
ty
p
e
o
f
n
e
u
r
al
n
etwo
r
k
th
at
le
ar
n
s
ef
f
icien
t
r
ep
r
esen
tatio
n
s
o
f
in
p
u
t
s
eq
u
en
ce
s
th
r
o
u
g
h
an
en
co
d
er
-
d
ec
o
d
e
r
s
tr
u
ctu
r
e
.
T
h
e
en
c
o
d
er
ex
tr
ac
ts
im
p
o
r
tan
t
te
m
p
o
r
a
l
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
tim
e
-
s
er
ies
d
ata
an
d
c
o
m
p
r
ess
es
th
em
in
to
a
lo
we
r
-
d
im
en
s
io
n
al
laten
t
s
p
ac
e
,
wh
ile
th
e
d
ec
o
d
er
r
ec
o
n
s
tr
u
cts
th
e
o
r
ig
in
al
in
p
u
t
f
r
o
m
th
is
co
m
p
r
ess
ed
r
ep
r
esen
tatio
n
.
Giv
en
a
tim
e
-
s
er
ies
s
eq
u
en
ce
=
{
1
,
2
,
3
,
…
,
}
,
th
e
en
co
d
er
ap
p
lies
a
s
er
ies
o
f
co
n
v
o
lu
tio
n
al
o
p
e
r
atio
n
s
to
g
en
er
ate
f
ea
t
u
r
e
m
a
p
s
.
T
h
e
co
n
v
o
l
u
tio
n
al
tr
an
s
f
o
r
m
atio
n
f
o
r
ea
ch
f
ilter
ℎ
is
g
iv
en
b
y
(
1
3
)
.
ℎ
=
σ
(
∗
+
)
(
1
3
)
I
n
(
1
3
)
,
r
ep
r
esen
ts
th
e
f
ilter
weig
h
ts
,
is
th
e
b
ias
ter
m
,
∗
d
en
o
tes
th
e
co
n
v
o
lu
tio
n
o
p
er
ati
o
n
,
an
d
σ
is
a
non
-
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
(
e.
g
.
,
R
eL
U)
.
T
h
e
en
co
d
e
d
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
ar
e
f
u
r
t
h
er
p
ass
ed
th
r
o
u
g
h
m
ax
-
p
o
o
lin
g
lay
e
r
s
to
r
ed
u
c
e
d
im
en
s
io
n
ality
wh
ile
p
r
ese
r
v
in
g
th
e
m
o
s
t
s
ig
n
if
ican
t
f
ea
tu
r
es.
T
h
e
laten
t
r
ep
r
esen
tatio
n
,
,
is
o
b
tain
ed
as
(
1
4
)
.
I
n
(
1
4
)
,
m
ax
-
p
o
o
lin
g
h
e
lp
s
to
r
etain
d
o
m
in
an
t
s
p
atial
-
tem
p
o
r
al
f
ea
tu
r
es
an
d
r
ed
u
ce
s
co
m
p
u
tatio
n
al
co
m
p
lex
ity
.
=
(
ℎ
)
(
1
4
)
T
h
e
d
ec
o
d
er
r
ec
o
n
s
tr
u
cts
th
e
in
p
u
t
s
eq
u
en
ce
f
r
o
m
th
e
laten
t
r
ep
r
esen
tatio
n
b
y
a
p
p
ly
in
g
t
r
an
s
p
o
s
ed
co
n
v
o
l
u
tio
n
(
d
ec
o
n
v
o
lu
tio
n
)
l
ay
er
s
,
en
s
u
r
in
g
th
at
th
e
lear
n
ed
f
ea
tu
r
es
ac
cu
r
ately
ca
p
tu
r
e
u
n
d
er
ly
in
g
tim
e
-
d
ep
en
d
e
n
t p
atter
n
s
.
T
h
e
r
ec
o
n
s
tr
u
cted
o
u
tp
u
t seq
u
en
ce
̂
is
g
en
er
ated
as (
1
5
)
.
̂
=
σ
(
∗
+
)
(
1
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
0
6
7
-
5
0
7
9
5074
I
n
(
1
5
)
,
an
d
ar
e
th
e
weig
h
ts
an
d
b
ias
o
f
t
h
e
d
ec
o
d
er
n
etwo
r
k
,
r
esp
ec
tiv
el
y
.
T
o
f
i
n
e
-
tu
n
e
th
e
m
o
d
el
f
o
r
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
,
th
e
f
i
n
al
lay
er
is
m
o
d
if
ied
to
p
r
ed
i
ct
th
e
f
u
tu
r
e
tim
e
s
tep
s
̂
b
ased
o
n
th
e
lear
n
ed
laten
t f
ea
tu
r
es is
g
iv
en
b
y
(
1
6
)
.
̂
=
(
)
(
1
6
)
I
n
(
1
6
)
,
th
e
d
en
s
e
lay
er
m
a
p
s
th
e
co
m
p
r
ess
ed
r
e
p
r
esen
tatio
n
to
th
e
o
u
tp
u
t
s
p
ac
e.
B
y
lev
er
ag
in
g
C
NN
-
b
ased
f
ea
tu
r
e
e
x
tr
ac
tio
n
,
th
e
au
to
en
co
d
er
im
p
r
o
v
es
f
o
r
ec
asti
n
g
ac
c
u
r
ac
y
b
y
ca
p
t
u
r
in
g
in
tr
icate
tem
p
o
r
a
l
d
ep
en
d
e
n
cies w
h
ile
ef
f
ec
tiv
el
y
h
an
d
li
n
g
n
o
is
e
an
d
o
u
tlier
s
in
th
e
d
ataset.
3
.
3
.
3
L
ST
M
wit
h
a
t
t
ent
io
n m
ec
ha
nis
m
T
h
e
L
STM
n
etwo
r
k
s
ar
e
wid
ely
u
s
ed
f
o
r
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
d
u
e
to
th
eir
ab
ilit
y
to
r
etain
lo
n
g
-
ter
m
d
ep
e
n
d
en
cies
wh
ile
m
iti
g
atin
g
th
e
v
an
is
h
in
g
g
r
a
d
ien
t
p
r
o
b
lem
.
Ho
wev
e
r
,
tr
a
d
itio
n
al
L
STM
s
tr
ea
t
all
tim
e
s
tep
s
with
eq
u
al
im
p
o
r
t
an
ce
,
wh
ich
ca
n
lea
d
to
s
u
b
o
p
tim
al
p
er
f
o
r
m
an
ce
i
n
co
m
p
lex
d
atasets
wh
er
e
ce
r
tain
p
ast
tim
e
s
tep
s
co
n
tr
ib
u
te
m
o
r
e
s
ig
n
if
ican
tly
to
f
u
tu
r
e
p
r
ed
ictio
n
s
.
T
h
e
atten
tio
n
m
ec
h
an
is
m
en
h
an
ce
s
L
STM
b
y
d
y
n
a
m
ically
weig
h
in
g
th
e
im
p
o
r
tan
ce
o
f
p
ast
o
b
s
er
v
atio
n
s
is
p
r
esen
ted
in
Fig
u
r
e
3
,
allo
win
g
th
e
m
o
d
el
to
f
o
cu
s
o
n
t
h
e
m
o
s
t
r
elev
an
t
tim
e
s
tep
s
.
Giv
en
an
i
n
p
u
t
s
eq
u
en
ce
=
{
1
,
2
,
3
,
…
,
}
,
th
e
L
STM
p
r
o
ce
s
s
es th
e
s
eq
u
en
ce
iter
ati
v
ely
u
s
in
g
th
e
(
1
7
)
.
=
σ
(
ℎ
−
1
+
+
)
=
ℎ
(
ℎ
−
1
+
+
)
=
⊙
−
1
+
⊙
̃
ℎ
=
⊙
ℎ
(
)
(
1
7
)
I
n
(
1
7
)
,
,
an
d
d
en
o
te
th
e
f
o
r
g
et,
in
p
u
t,
a
n
d
o
u
tp
u
t
g
ates
,
r
esp
ec
tiv
ely
,
is
th
e
ce
ll
s
tate,
ℎ
is
th
e
h
id
d
en
s
tate,
a
n
d
r
ep
r
esen
ts
th
e
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
cti
o
n
.
T
h
e
atten
tio
n
m
ec
h
an
is
m
is
th
en
ap
p
lied
to
en
h
an
ce
th
e
L
STM
’
s
ab
ilit
y
to
f
o
cu
s
o
n
cr
itical
tim
e
s
tep
s
.
T
h
e
atten
tio
n
s
co
r
e
$
\
alp
h
a
_
{t}$
is
co
m
p
u
ted
u
s
in
g
an
alig
n
m
en
t
f
u
n
ctio
n
t
h
at
d
eter
m
in
es
th
e
r
elev
an
ce
o
f
ea
ch
h
id
d
en
s
tate
ℎ
with
r
es
p
ec
t
to
th
e
tar
g
et
o
u
tp
u
t.
T
h
e
atten
tio
n
weig
h
ts
ar
e
co
m
p
u
ted
as (
1
8
)
–
(
2
0
)
.
=
ta
n
h
(
ℎ
+
)
(
1
8
)
α
=
ex
p
(
)
∑
ex
p
(
́
)
́
(
1
9
)
=
∑
α
ℎ
(
2
0
)
I
n
(
1
8
)
–
(
2
0
)
,
r
ep
r
esen
ts
th
e
at
ten
tio
n
s
co
r
e,
an
d
ar
e
lear
n
a
b
le
p
ar
am
eter
s
,
a
n
d
is
th
e
co
n
tex
t
v
ec
to
r
o
b
tain
ed
b
y
tak
in
g
th
e
weig
h
ted
s
u
m
o
f
h
id
d
e
n
s
tates.
T
h
e
f
in
al
o
u
tp
u
t is th
en
co
m
p
u
ted
a
s
(
2
1
)
.
=
(
[
;
ℎ
]
+
)
(
21)
B
y
in
co
r
p
o
r
atin
g
atten
tio
n
,
t
h
e
m
o
d
el
s
elec
tiv
ely
f
o
cu
s
es
o
n
in
f
o
r
m
ativ
e
tim
e
s
tep
s
,
lead
in
g
to
im
p
r
o
v
e
d
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
T
h
is
ap
p
r
o
ac
h
is
p
ar
ticu
la
r
ly
b
en
e
f
icial
f
o
r
en
er
g
y
d
em
an
d
p
r
e
d
ictio
n
,
wh
er
e
ex
ter
n
al
f
ac
to
r
s
s
u
ch
as
s
ea
s
o
n
al
v
ar
iatio
n
s
an
d
p
ea
k
d
em
a
n
d
p
er
io
d
s
ex
er
t
v
ar
y
in
g
lev
e
ls
o
f
in
f
lu
en
ce
o
n
f
u
tu
r
e
c
o
n
s
u
m
p
tio
n
.
T
h
e
atten
tio
n
-
en
h
a
n
ce
d
L
STM
p
r
o
v
i
d
e
s
g
r
ea
ter
in
ter
p
r
etab
ilit
y
an
d
a
d
ap
tab
ilit
y
,
m
ak
i
n
g
it a
r
o
b
u
s
t c
h
o
ice
f
o
r
tim
e
-
s
er
ies f
o
r
ec
asti
n
g
task
s
in
en
er
g
y
m
an
ag
em
en
t sy
s
tem
s
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
L
ST
M
with
atten
tio
n
m
ec
h
an
is
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
I
mp
a
ct
o
f o
u
tlier
d
etec
tio
n
tec
h
n
iq
u
es o
n
time
-
s
eries
fo
r
ec
a
s
tin
g
a
cc
u
r
a
cy
fo
r
…
(
S
h
r
ey
a
s
K
a
r
n
ick
)
5075
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
im
p
ac
t
o
f
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es
o
n
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
was
an
aly
ze
d
u
s
in
g
th
r
ee
d
ee
p
lear
n
i
n
g
m
o
d
els
—
L
STM
,
C
NN
-
Au
to
en
co
d
er
,
an
d
L
STM
with
a
tten
tio
n
—
ac
r
o
s
s
m
u
ltip
le
co
u
n
tr
ies,
in
clu
d
i
n
g
Ger
m
a
n
y
(
DE
)
,
Fra
n
ce
(
FR
)
,
Sp
ain
(
E
S),
an
d
I
taly
(
I
T
)
.
T
h
e
e
x
p
er
im
en
tal
r
esu
lts
r
ev
ea
led
t
h
at
o
u
tlier
r
e
m
o
v
al
s
ig
n
if
ican
tly
im
p
r
o
v
e
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
b
y
r
ed
u
ci
n
g
er
r
o
r
m
et
r
ics
s
u
ch
as
R
MSE
an
d
MA
E
.
Am
o
n
g
th
e
f
iv
e
o
u
tlier
d
etec
tio
n
tech
n
i
q
u
es
ap
p
lied
,
IF
an
d
L
OF
d
em
o
n
s
tr
ated
s
u
p
er
io
r
ca
p
ab
ilit
y
in
d
etec
tin
g
an
o
m
alo
u
s
p
atter
n
s
,
lead
in
g
to
th
e
m
o
s
t
n
o
ticea
b
le
im
p
r
o
v
e
m
en
ts
in
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
.
Sp
ec
if
ically
,
in
th
e
ca
s
e
o
f
Ger
m
an
y
,
t
h
e
R
MSE
f
o
r
L
STM
with
o
u
t
o
u
tlier
r
em
o
v
al
was
1
7
.
3
4
MW,
wh
ich
r
ed
u
ce
d
to
1
2
.
2
1
MW
af
ter
ap
p
ly
in
g
IF
.
Similar
ly
,
Sp
ain
ex
h
ib
ited
a
s
u
b
s
tan
tial
r
ed
u
ctio
n
in
f
o
r
ec
asti
n
g
er
r
o
r
,
wh
er
e
th
e
R
MSE
im
p
r
o
v
ed
f
r
o
m
1
9
.
8
7
to
1
4
.
0
2
MW
p
o
s
t
o
u
tlier
r
em
o
v
al.
T
h
ese
r
esu
lts
in
d
icate
th
at
elim
in
atin
g
o
u
tli
er
s
ef
f
ec
tiv
ely
m
itig
ates
n
o
is
e,
en
ab
lin
g
th
e
m
o
d
els
to
lear
n
m
o
r
e
r
ep
r
esen
tativ
e
en
er
g
y
c
o
n
s
u
m
p
ti
o
n
p
atter
n
s
.
T
h
e
co
m
p
ar
ativ
e
an
aly
s
is
o
f
v
ar
i
o
u
s
f
o
r
ec
asti
n
g
m
eth
o
d
s
with
an
d
with
o
u
t
o
u
tlier
d
etec
tio
n
tech
n
iq
u
es
d
em
o
n
s
tr
ates
s
ig
n
if
ican
t
v
ar
iatio
n
s
in
p
r
e
d
ictiv
e
ac
cu
r
ac
y
ac
r
o
s
s
d
if
f
er
en
t
co
u
n
tr
ies.
T
h
e
T
ab
le
1
p
r
esen
ts
MA
PE
v
alu
es
f
o
r
Ger
m
an
y
(
DE
)
,
Fra
n
ce
(
FR
)
,
Sp
ain
(
E
S),
an
d
I
taly
(
I
T
)
,
e
m
p
lo
y
in
g
d
i
f
f
er
en
t
f
o
r
ec
asti
n
g
ap
p
r
o
ac
h
es
s
u
ch
a
s
L
STM
,
C
NN
-
Au
to
en
co
d
er
,
an
d
L
STM
with
atten
tio
n
m
e
ch
an
is
m
.
W
h
en
n
o
o
u
tlier
d
etec
tio
n
m
eth
o
d
is
ap
p
lied
,
th
e
C
NN
-
Au
to
en
c
o
d
e
r
co
n
s
is
ten
tly
ex
h
ib
its
th
e
lo
west
MA
PE
v
alu
es
ac
r
o
s
s
all
r
eg
io
n
s
,
in
d
icatin
g
i
ts
r
o
b
u
s
tn
ess
in
h
an
d
lin
g
r
aw
d
ata.
C
o
n
v
e
r
s
ely
,
L
STM
-
Atten
tio
n
p
er
f
o
r
m
s
th
e
wo
r
s
t
am
o
n
g
th
e
th
r
ee
f
o
r
ec
as
tin
g
m
o
d
els,
y
ield
i
n
g
th
e
h
ig
h
est
MA
PE
v
alu
es,
p
ar
ticu
lar
ly
in
I
taly
(
1
9
.
7
0
%)
an
d
Sp
ain
(
1
4
.
0
8
%).
T
h
is
s
u
g
g
ests
th
at
wh
ile
atten
tio
n
m
ec
h
an
is
m
s
en
h
an
ce
L
STM
m
o
d
els
in
ce
r
tain
s
ce
n
ar
io
s
,
th
ey
m
ay
b
e
m
o
r
e
s
en
s
itiv
e
to
an
o
m
alies p
r
esen
t i
n
th
e
d
ataset.
T
h
e
im
p
lem
e
n
tatio
n
o
f
o
u
tlier
d
etec
tio
n
tech
n
i
q
u
es
lead
s
to
a
s
u
b
s
tan
tial
im
p
r
o
v
em
e
n
t
in
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
,
with
o
n
e
-
class
SVM
em
er
g
in
g
as
th
e
m
o
s
t
ef
f
ec
tiv
e
m
eth
o
d
f
o
r
n
o
is
e
r
ed
u
ctio
n
.
U
n
d
er
th
is
ap
p
r
o
ac
h
,
C
NN
-
Au
to
en
co
d
er
attain
s
th
e
lo
west
MA
PE
v
alu
es
ac
r
o
s
s
all
co
u
n
tr
ies,
p
a
r
ticu
lar
ly
in
Fra
n
ce
(
2
.
1
2
%)
an
d
I
taly
(
3
.
1
7
%),
u
n
d
er
s
co
r
i
n
g
its
ef
f
icien
cy
i
n
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
d
e
n
o
is
in
g
ca
p
ab
ilit
ies.
Similar
ly
,
th
e
ap
p
licatio
n
o
f
L
OF
an
d
I
F
also
im
p
r
o
v
es
p
r
ed
ictio
n
ac
cu
r
ac
y
,
alb
eit
to
a
s
lig
h
tly
less
er
ex
ten
t.
No
tab
ly
,
L
STM
'
s
p
er
f
o
r
m
an
c
e
s
ig
n
if
ican
tly
b
en
ef
its
f
r
o
m
t
h
ese
m
eth
o
d
s
,
with
MA
PE
v
a
lu
es
d
r
o
p
p
in
g
f
r
o
m
1
0
.
7
3
%
(
with
o
u
t
o
u
tlier
d
ete
ctio
n
)
to
as
lo
w
as
3
.
1
8
%
(
o
n
e
-
class
SVM)
in
Fra
n
ce
.
T
h
is
h
ig
h
lig
h
ts
th
e
im
p
o
r
tan
ce
o
f
o
u
tlier
h
an
d
lin
g
in
im
p
r
o
v
in
g
t
h
e
p
r
e
d
ictiv
e
r
e
liab
ilit
y
o
f
r
ec
u
r
r
en
t n
e
u
r
al
n
et
wo
r
k
s
.
Am
o
n
g
th
e
o
u
tlier
d
etec
tio
n
m
eth
o
d
s
,
DB
SC
AN
an
d
Z
-
Sco
r
e
f
ilter
in
g
also
ex
h
ib
it
p
r
o
m
is
i
n
g
r
esu
lts
,
th
o
u
g
h
th
eir
ef
f
ec
tiv
en
ess
v
ar
ies
b
y
f
o
r
ec
asti
n
g
m
o
d
el.
DB
SC
AN,
f
o
r
in
s
tan
ce
,
h
elp
s
r
e
d
u
ce
MA
PE
v
alu
es
in
L
STM
m
o
d
els
co
n
s
id
er
ab
l
y
,
b
r
in
g
in
g
th
em
d
o
wn
to
5
.
8
1
%
(
Ger
m
an
y
)
an
d
4
.
4
7
%
(
Fra
n
ce
)
.
L
ik
ewise,
C
NN
-
Au
to
en
co
d
er
b
e
n
ef
its
f
r
o
m
DB
SC
A
N,
attain
in
g
a
MA
PE
o
f
2
.
8
8
%
in
Sp
ain
,
wh
i
ch
is
a
co
n
s
id
er
ab
le
im
p
r
o
v
em
e
n
t
f
r
o
m
th
e
b
aselin
e.
Ho
wev
er
,
th
e
L
STM
-
Atten
tio
n
m
o
d
el,
d
esp
ite
s
o
m
e
im
p
r
o
v
em
en
ts
,
co
n
tin
u
es
to
e
x
h
ib
it
r
elativ
ely
h
ig
h
e
r
er
r
o
r
r
ates
ac
r
o
s
s
m
o
s
t
co
u
n
tr
ies,
s
u
g
g
esti
n
g
t
h
at
atten
tio
n
-
b
ased
ar
ch
itectu
r
es m
ig
h
t r
e
q
u
ir
e
m
o
r
e
s
o
p
h
is
ticated
an
o
m
aly
h
a
n
d
lin
g
tech
n
i
q
u
es f
o
r
o
p
tim
al
p
er
f
o
r
m
a
n
ce
.
Ov
er
all,
th
e
f
in
d
in
g
s
p
lace
cr
u
cial
r
o
le
o
f
o
u
tlier
d
etec
tio
n
in
en
h
an
cin
g
f
o
r
ec
asti
n
g
ac
cu
r
ac
y
,
with
one
-
class
SVM
an
d
C
NN
-
A
u
to
en
co
d
er
em
er
g
in
g
as
th
e
m
o
s
t
ef
f
ec
tiv
e
co
m
b
in
atio
n
.
W
h
ile
L
STM
-
b
ased
m
o
d
els
b
e
n
ef
it
f
r
o
m
an
o
m
aly
f
ilter
in
g
,
t
h
e
c
h
o
ice
o
f
f
o
r
ec
as
tin
g
m
o
d
el
a
n
d
o
u
tlier
d
etec
ti
o
n
m
et
h
o
d
m
u
s
t
b
e
tailo
r
ed
to
th
e
s
p
ec
if
ic
d
ataset
an
d
a
p
p
licatio
n
c
o
n
tex
t.
Fu
t
u
r
e
r
esear
ch
c
o
u
ld
e
x
p
lo
r
e
h
y
b
r
id
ap
p
r
o
ac
h
es
th
at
in
teg
r
ate
m
u
ltip
le
an
o
m
aly
d
e
tectio
n
s
tr
ateg
ies
o
r
lev
er
ag
e
ad
ap
tiv
e
f
ilter
in
g
m
ec
h
a
n
is
m
s
to
f
u
r
th
er
im
p
r
o
v
e
p
r
ed
ictiv
e
p
e
r
f
o
r
m
an
ce
in
tim
e
s
er
ies f
o
r
ec
asti
n
g
.
T
h
e
ef
f
ec
t
o
f
in
d
iv
id
u
al
o
u
tlie
r
d
etec
tio
n
m
eth
o
d
s
v
ar
ied
ac
r
o
s
s
co
u
n
tr
ies
d
u
e
to
d
if
f
er
en
ce
s
in
d
ata
ch
ar
ac
ter
is
tics
an
d
e
n
er
g
y
c
o
n
s
u
m
p
tio
n
tr
en
d
s
.
Z
-
Sco
r
e
a
n
d
one
-
class
SVM,
wh
ile
ef
f
ec
tiv
e
in
d
etec
tin
g
ex
tr
em
e
d
ev
iatio
n
s
,
s
tr
u
g
g
le
d
with
s
u
b
tle
an
o
m
alies p
r
esen
t
in
n
o
n
-
Gau
s
s
ian
d
is
tr
ib
u
tio
n
s
.
On
th
e
o
th
er
h
an
d
,
DB
S
C
AN,
wh
ich
clu
s
ter
s
d
ata
b
ased
o
n
d
en
s
ity
,
d
em
o
n
s
tr
ated
m
ix
ed
r
esu
lts
s
h
o
wn
in
Fig
u
r
es
4
,
5
an
d
6
p
er
f
o
r
m
in
g
well
in
s
tr
u
ctu
r
ed
d
atasets
lik
e
Fra
n
ce
b
u
t
u
n
d
e
r
p
er
f
o
r
m
in
g
in
I
taly
d
u
e
to
i
r
r
eg
u
lar
f
lu
ct
u
atio
n
s
in
d
em
an
d
p
atter
n
s
.
Ov
e
r
all,
th
e
s
tu
d
y
co
n
f
ir
m
s
th
at
s
elec
tin
g
an
a
p
p
r
o
p
r
iate
o
u
tlier
d
et
ec
tio
n
tech
n
iq
u
e
is
cr
u
cial
f
o
r
o
p
tim
izin
g
f
o
r
ec
a
s
tin
g
ac
cu
r
ac
y
,
a
n
d
th
e
b
est
ch
o
ice
o
f
ten
d
ep
e
n
d
s
o
n
t
h
e
u
n
d
er
l
y
in
g
d
ata
d
is
tr
ib
u
tio
n
.
T
h
e
f
in
d
in
g
s
also
h
ig
h
lig
h
t
th
e
n
ec
ess
ity
o
f
ad
ap
tiv
e
a
n
o
m
aly
d
etec
tio
n
s
tr
ateg
ies
th
at
ca
n
d
y
n
am
ically
a
d
ju
s
t to
s
ea
s
o
n
al
v
ar
iatio
n
s
an
d
lo
n
g
-
ter
m
tr
en
d
s
in
en
er
g
y
co
n
s
u
m
p
tio
n
.
T
h
e
o
v
er
all
im
p
r
o
v
em
en
ts
in
m
o
d
el
p
er
f
o
r
m
a
n
ce
p
o
s
t
o
u
tlier
r
em
o
v
al
r
ein
f
o
r
ce
th
e
im
p
o
r
tan
ce
o
f
d
ata
p
r
ep
r
o
ce
s
s
in
g
in
tim
e
-
s
er
ies
f
o
r
ec
asti
n
g
task
s
.
W
h
ile
d
ee
p
lear
n
in
g
ar
ch
itectu
r
e
ca
n
ca
p
tu
r
e
co
m
p
lex
tem
p
o
r
al
d
e
p
en
d
e
n
cies,
th
eir
ef
f
ec
tiv
en
ess
is
s
ig
n
if
ican
tly
in
f
lu
en
ce
d
b
y
d
ata
q
u
ality
.
T
h
is
s
tu
d
y
d
em
o
n
s
tr
ates
th
at
in
teg
r
atin
g
r
o
b
u
s
t
o
u
tlier
d
etec
tio
n
m
e
ch
an
is
m
s
ca
n
s
u
b
s
tan
tially
e
n
h
an
ce
f
o
r
ec
asti
n
g
r
eliab
ilit
y
,
m
ak
in
g
en
er
g
y
d
e
m
an
d
p
r
ed
ictio
n
m
o
d
els
m
o
r
e
ap
p
licab
le
f
o
r
r
ea
l
-
wo
r
l
d
en
er
g
y
m
an
ag
em
e
n
t
an
d
g
r
id
o
p
tim
izatio
n
.
Fu
tu
r
e
r
esear
ch
s
h
o
u
ld
e
x
p
lo
r
e
th
e
co
m
b
in
atio
n
o
f
m
u
ltip
le
a
n
o
m
aly
d
etec
tio
n
m
eth
o
d
s
u
s
in
g
en
s
em
b
le
tech
n
iq
u
es
a
n
d
in
v
esti
g
ate
th
e
im
p
ac
t
o
f
in
co
r
p
o
r
atin
g
ex
ter
n
al
f
ac
t
o
r
s
s
u
ch
as
wea
th
er
co
n
d
itio
n
s
an
d
ec
o
n
o
m
ic
in
d
ic
ato
r
s
to
f
u
r
t
h
er
r
ef
i
n
e
p
r
e
d
ictio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
6
,
Decem
b
e
r
20
25
:
5
0
6
7
-
5
0
7
9
5076
T
ab
le
1
.
C
o
m
p
a
r
ativ
e
r
esu
lts
f
o
r
v
ar
i
o
u
s
f
o
r
ec
asti
n
g
m
eth
o
d
s
with
an
d
with
o
u
t o
u
tlier
m
et
h
o
d
s
O
u
t
l
i
e
r
me
t
h
o
d
F
o
r
e
c
a
st
i
n
g
m
e
t
h
o
d
G
e
r
man
y
(
D
E)
M
A
P
E
(
%)
F
r
a
n
c
e
(
F
R
)
M
A
P
E
(
%)
S
p
a
i
n
(
ES)
M
A
P
E
(
%)
I
t
a
l
y
(
I
T)
M
A
P
E
(
%)
N
o
n
e
LSTM
1
0
.
7
3
6
.
9
6
1
0
.
0
0
1
5
.
4
9
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
5
.
0
3
4
.
6
9
4
.
1
3
7
.
2
3
LSTM
-
a
t
t
e
n
t
i
o
n
1
4
.
0
1
7
.
6
3
1
4
.
0
8
1
9
.
7
0
Z
-
S
c
o
r
e
LSTM
9
.
0
1
5
.
3
6
7
.
1
6
1
2
.
6
5
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
3
.
7
0
3
.
4
0
3
.
1
8
4
.
4
8
LSTM
-
a
t
t
e
n
t
i
o
n
7
.
9
9
6
.
3
8
9
.
4
5
1
1
.
1
9
D
B
S
C
A
N
LSTM
5
.
8
1
4
.
4
7
5
.
7
6
8
.
6
3
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
3
.
6
0
2
.
9
1
2
.
8
8
3
.
7
4
LSTM
-
a
t
t
e
n
t
i
o
n
7
.
8
8
5
.
3
8
8
.
5
9
1
0
.
6
3
IF
LSTM
5
.
7
9
3
.
8
5
5
.
4
9
9
.
1
7
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
3
.
4
8
2
.
7
8
3
.
4
4
5
.
4
9
LSTM
-
a
t
t
e
n
t
i
o
n
9
.
3
5
5
.
0
0
8
.
5
1
1
3
.
4
1
LO
F
LSTM
4
.
8
2
3
.
4
1
4
.
6
6
7
.
6
4
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
3
.
1
2
2
.
7
0
2
.
7
0
3
.
7
6
LSTM
-
a
t
t
e
n
t
i
o
n
6
.
4
8
4
.
2
1
6
.
3
9
9
.
0
3
O
n
e
C
l
a
ss SV
M
LSTM
5
.
7
6
3
.
1
8
4
.
7
1
8
.
5
0
C
N
N
-
A
u
t
o
e
n
c
o
d
e
r
2
.
8
0
2
.
1
2
2
.
5
1
3
.
1
7
LSTM
-
a
t
t
e
n
t
i
o
n
5
.
2
7
3
.
2
6
4
.
3
5
8
.
1
1
Fig
u
r
e
4
.
L
o
ad
f
o
r
ec
asti
n
g
u
s
i
n
g
L
STM
with
o
u
t
o
u
tlier
s
r
em
o
v
ed
Fig
u
r
e
5
.
L
o
ad
f
o
r
ec
asti
n
g
u
s
i
n
g
C
NN
au
to
en
co
d
er
with
o
u
t
o
u
tlier
s
r
em
o
v
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.