I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
,
p
p
.
1
0
9
8
~
1
1
1
1
I
SS
N:
2252
-
8
8
1
4
,
DOI
:
1
0
.
1
1
5
9
1
/ijaas
.
v
1
5
.
i
3
.
pp
1
0
9
8
-
1
1
1
1
1098
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
a
s
.
ia
esco
r
e.
co
m
A nov
el comb
ined
deep
l
ea
rning
-
ba
sed tec
hnique
f
o
r
elect
rici
ty
thef
t
de
tect
io
n
in
a
sma
rt
g
rid
E
t
ienne F
ra
nço
is
M
o
uck
o
m
e
y
1
,
J
a
cques
B
ik
a
i
2
,
Ca
m
ille F
ra
nk
lin
M
bey
1
,
F
elix
G
is
la
in
Yem
So
uh
e
1
,
Alex
a
nd
re
T
epla
ira
B
o
um
1
,
Vinn
y
J
un
io
r
F
o
ba
K
a
k
eu
1
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
EN
S
ET
,
U
n
i
v
e
r
si
t
y
o
f
D
o
u
a
l
a
,
D
o
u
a
l
a
,
C
a
mero
o
n
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
y
o
f
N
g
a
o
u
n
d
e
r
e
,
N
g
a
o
u
n
d
e
r
e
,
C
a
m
e
r
o
o
n
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
2
,
2
0
2
5
R
ev
is
ed
Au
g
8
,
2
0
2
6
Acc
ep
ted
Au
g
1
3
,
2
0
2
6
En
e
rg
y
th
e
ft
re
p
re
se
n
ts
a
m
a
jo
r
c
h
a
ll
e
n
g
e
fo
r
e
lec
tri
c
it
y
d
i
strib
u
ti
o
n
c
o
m
p
a
n
ies
wo
rl
d
wid
e
a
n
d
c
o
n
st
it
u
tes
a
k
e
y
c
o
m
p
o
n
e
n
t
o
f
n
o
n
-
tec
h
n
ica
l
lo
ss
e
s
(
NTLs
)
,
a
lo
n
g
si
d
e
fa
u
lt
y
m
e
ters
a
n
d
b
il
li
n
g
in
a
c
c
u
ra
c
ies
.
De
sp
it
e
th
e
ri
se
o
f
a
rt
ifi
c
ial
i
n
telli
g
e
n
c
e
t
e
c
h
n
iq
u
e
s
fo
r
e
lec
tri
c
it
y
th
e
ft
d
e
tec
ti
o
n
,
e
x
isti
n
g
a
p
p
ro
a
c
h
e
s
e
x
h
i
b
it
li
m
i
tatio
n
s
in
c
las
sifica
ti
o
n
a
n
d
d
a
t
a
a
n
a
ly
sis
wh
e
n
a
p
p
l
ied
to
lar
g
e
-
sc
a
le
d
a
tas
e
ts.
Th
e
se
sh
o
rtco
m
in
g
s
c
a
n
r
e
d
u
c
e
th
e
e
ffe
c
ti
v
e
n
e
ss
a
n
d
p
re
c
isio
n
o
f
t
h
e
th
e
ft
d
e
tec
ti
o
n
sy
ste
m
s.
Th
e
r
e
fo
re
,
th
is
a
rti
c
le
p
ro
p
o
se
s
a
b
i
d
irec
ti
o
n
a
l
l
o
n
g
sh
o
r
t
-
term
m
e
m
o
ry
(Bi
-
LS
TM
)
with
th
e
lo
o
k
a
h
e
a
d
a
d
a
p
ti
v
e
m
o
m
e
n
t
e
stim
a
ti
o
n
(LADA
M
)
o
p
ti
m
ize
r
a
n
d
a
su
p
p
o
rt
v
e
c
to
r
m
a
c
h
in
e
(S
VM)
with
a
b
id
irec
ti
o
n
a
l
g
a
ted
re
c
u
rre
n
t
u
n
it
(Bi
-
G
RU
)
fo
r
th
e
d
e
tec
ti
o
n
o
f
e
n
e
rg
y
t
h
e
ft
in
a
sm
a
rt
e
lec
tri
c
it
y
g
rid
to
a
ss
e
ss
e
n
e
rg
y
su
p
p
l
ier
c
o
m
p
a
n
ies
.
I
n
th
is
wo
rk
,
th
e
p
r
o
p
o
se
d
m
o
d
e
l
is
imp
lem
e
n
ted
on
th
e
S
tate
G
rid
C
o
rp
o
ra
ti
o
n
o
f
C
h
in
a
(S
G
CC)
d
a
tas
e
t.
In
th
e
p
ro
p
o
se
d
fra
m
e
wo
rk
,
th
e
Bi
-
LS
TM
is
u
se
d
fo
r
f
e
a
tu
re
e
x
trac
ti
o
n
,
wh
il
e
t
h
e
LADA
M
o
p
ti
m
ize
r
is
e
m
p
l
o
y
e
d
fo
r
fe
a
tu
r
e
se
lec
ti
o
n
a
n
d
n
o
rm
a
li
z
a
ti
o
n
.
T
h
e
re
fin
e
d
fe
a
tu
re
se
t
is su
b
se
q
u
e
n
tl
y
fe
d
i
n
t
o
th
e
S
VM
c
las
sifier t
o
d
isti
n
g
u
is
h
b
e
twe
e
n
th
e
ft
a
n
d
n
o
n
-
t
h
e
ft
i
n
sta
n
c
e
s.
T
h
e
o
b
tai
n
e
d
fin
d
i
n
g
s
h
i
g
h
li
g
h
t
t
h
e
su
p
e
ri
o
r
e
fficie
n
c
y
o
f
t
h
e
p
r
o
p
o
se
d
stra
teg
y
c
o
m
p
a
re
d
with
th
o
se
i
n
th
e
li
tera
tu
re
,
h
ig
h
li
g
h
ti
n
g
it
s
p
o
ten
t
ial
fo
r
p
ra
c
ti
c
a
l
d
e
p
lo
y
m
e
n
t
i
n
re
a
l
-
wo
rld
sm
a
rt
g
rid
sy
ste
m
s.
K
ey
w
o
r
d
s
:
ADAM
Bi
-
L
ADAM
-
SVM
-
G
R
U
m
o
d
el
Dee
p
lear
n
in
g
E
lectr
icity
th
ef
t
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
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
:
E
tien
n
e
Fra
n
ço
is
Mo
u
c
k
o
m
e
y
Dep
ar
tm
en
t o
f
E
lectr
ical
E
n
g
i
n
ee
r
in
g
,
E
NSET
,
Un
iv
er
s
ity
o
f
Do
u
ala
Do
u
ala,
C
am
er
o
o
n
E
m
ail: e
m
o
u
ck
o
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
g
lo
b
al
d
e
m
an
d
f
o
r
elec
t
r
ical
en
er
g
y
h
as
b
ee
n
in
cr
ea
s
in
g
s
tead
ily
d
u
e
to
p
o
p
u
latio
n
g
r
o
wth
,
in
d
u
s
tr
ial
ex
p
a
n
s
io
n
,
a
n
d
th
e
elec
tr
if
icatio
n
o
f
n
ew
s
ec
to
r
s
o
f
ac
tiv
ity
.
T
h
is
g
r
o
win
g
d
em
an
d
r
e
q
u
ir
es
p
o
wer
g
r
id
o
p
e
r
ato
r
s
to
en
s
u
r
e
a
co
n
tin
u
o
u
s
,
s
tab
le,
an
d
r
elia
b
le
elec
tr
icity
s
u
p
p
ly
to
co
n
s
u
m
er
s
[
1
]
.
Ho
wev
er
,
m
ain
tain
in
g
s
u
c
h
r
eliab
ilit
y
r
e
m
ain
s
a
ch
allen
g
e
d
u
e
to
v
ar
i
o
u
s
lo
s
s
es
o
cc
u
r
r
i
n
g
with
in
th
e
p
o
wer
g
r
i
d
,
w
h
ich
d
ir
ec
tly
im
p
ac
t
th
e
ef
f
icien
c
y
an
d
p
r
o
f
itab
ilit
y
o
f
elec
tr
icit
y
d
is
tr
ib
u
tio
n
s
y
s
tem
s
.
Po
wer
lo
s
s
es
in
elec
tr
ical
g
r
id
s
ca
n
b
e
b
r
o
a
d
ly
d
iv
id
ed
i
n
to
two
g
r
o
u
p
s
:
tech
n
ical
lo
s
s
es
an
d
n
o
n
-
tec
h
n
ical
lo
s
s
es
(
N
T
L
s
)
[
2
]
.
T
ec
h
n
ical
lo
s
s
es
ar
e
in
h
er
en
t
to
th
e
p
h
y
s
ical
ch
ar
ac
ter
is
tics
o
f
elec
tr
ical
co
m
p
o
n
en
ts
,
s
u
ch
as
r
esis
tiv
e
d
is
s
ip
atio
n
in
tr
an
s
m
i
s
s
io
n
lin
es,
tr
an
s
f
o
r
m
er
in
ef
f
icien
cies,
an
d
en
e
r
g
y
leak
ag
e
in
p
o
o
r
ly
m
ain
tain
e
d
in
f
r
astru
ctu
r
es.
I
n
co
n
tr
ast,
NT
L
s
r
e
p
r
esen
t
t
h
e
d
if
f
er
en
ce
b
etwe
en
th
e
to
tal
e
n
er
g
y
s
u
p
p
lied
th
r
o
u
g
h
t
r
an
s
m
is
s
io
n
lin
es
an
d
th
e
en
er
g
y
ac
tu
ally
b
illed
to
en
d
u
s
er
s
.
T
h
ese
lo
s
s
es
ar
e
ty
p
i
ca
lly
ca
u
s
ed
b
y
h
u
m
an
ac
tio
n
s
s
u
ch
as
m
eter
tam
p
er
in
g
,
illeg
al
co
n
n
ec
tio
n
s
,
an
d
f
alsi
f
ied
r
ea
d
in
g
s
[
3
]
.
N
T
L
s
r
ep
r
esen
t
a
m
ajo
r
ec
o
n
o
m
ic
an
d
o
p
e
r
atio
n
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
n
o
ve
l c
o
mb
i
n
ed
d
ee
p
lea
r
n
i
n
g
-
b
a
s
ed
tech
n
iq
u
e
fo
r
elec
tr
icity
th
eft
…
(
E
tien
n
e
F
r
a
n
ç
o
is
Mo
u
ck
o
mey
)
1099
ch
allen
g
e,
co
s
tin
g
p
o
wer
u
tili
ties
m
ill
io
n
s
o
f
d
o
llar
s
an
n
u
ally
[4
]
.
B
ey
o
n
d
th
e
f
in
a
n
cial
im
p
ac
t,
NT
L
s
ca
u
s
e
n
etwo
r
k
o
v
er
lo
ad
i
n
g
,
v
o
ltag
e
in
s
tab
ilit
y
,
an
d
u
n
f
air
tar
if
f
ad
ju
s
tm
en
ts
th
at
p
en
alize
co
m
p
lian
t
co
n
s
u
m
er
s
.
Am
o
n
g
th
ese
lo
s
s
es,
elec
tr
icit
y
th
ef
t
is
th
e
m
o
s
t
s
ig
n
if
ican
t
an
d
wid
esp
r
ea
d
is
s
u
e.
I
t
r
ef
e
r
s
to
u
n
au
th
o
r
ized
co
n
s
u
m
p
tio
n
o
f
elec
tr
ical
en
e
r
g
y
th
at
is
eith
e
r
u
n
m
eter
e
d
o
r
u
n
b
illed
,
lead
in
g
to
s
u
b
s
tan
t
ial
r
ev
en
u
e
d
ef
icits
f
o
r
elec
tr
icity
co
m
p
an
ies
[
5
]
.
Fo
r
in
s
tan
ce
,
elec
tr
icity
lo
s
s
es
ex
ce
ed
4
.
8
b
illi
o
n
r
u
p
ee
s
an
n
u
ally
in
I
n
d
ia
[
6
]
,
ab
o
u
t
0
.
8
9
b
illi
o
n
r
u
p
ee
s
in
P
ak
is
tan
[
7
]
,
a
n
d
b
etwe
en
8
a
n
d
1
2
b
illi
o
n
r
a
n
d
in
S
o
u
th
Af
r
ica
[
8
]
,
[
9
]
.
Usu
ally
,
elec
tr
icity
th
ef
t
ca
n
b
e
d
etec
t
ed
th
r
o
u
g
h
m
a
n
u
al
f
ield
v
e
r
if
icatio
n
s
,
wh
ich
a
r
e
co
s
tly
an
d
tim
e
-
co
n
s
u
m
in
g
.
Har
d
war
e
-
b
ased
a
p
p
r
o
ac
h
es
r
eq
u
ir
in
g
ad
d
itio
n
al
s
en
s
o
r
s
o
r
m
o
n
ito
r
in
g
eq
u
ip
m
en
t
ca
n
im
p
r
o
v
e
d
etec
t
io
n
ac
cu
r
ac
y
b
u
t
r
e
m
ain
ex
p
en
s
iv
e
to
d
e
p
lo
y
at
lar
g
e
s
ca
le
[
1
0
]
,
[
1
1
]
.
T
h
e
ad
v
e
n
t
o
f
s
m
ar
t
g
r
id
s
an
d
s
m
ar
t m
eter
s
h
as
o
p
en
e
d
n
ew
o
p
p
o
r
tu
n
ities
f
o
r
a
u
to
m
ated
a
n
d
d
ata
-
d
r
iv
en
th
e
f
t
d
etec
tio
n
.
T
h
ese
d
e
v
ices
g
en
er
ate
lar
g
e
v
o
lu
m
es
o
f
tim
e
-
s
er
ies
co
n
s
u
m
p
tio
n
d
ata,
en
ab
lin
g
th
e
ap
p
licatio
n
o
f
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
tech
n
iq
u
es to
id
e
n
tify
ab
n
o
r
m
al
co
n
s
u
m
p
tio
n
p
atter
n
s
an
d
d
etec
t f
r
au
d
u
le
n
t b
eh
a
v
io
r
m
o
r
e
ef
f
icien
tly
[
1
2
]
.
Ov
er
th
e
p
ast
d
ec
a
d
e,
s
ev
er
a
l
r
esear
ch
ef
f
o
r
ts
h
av
e
f
o
cu
s
ed
o
n
d
ev
elo
p
in
g
i
n
tellig
en
t
s
y
s
tem
s
to
r
ed
u
ce
NT
L
s
,
in
clu
d
i
n
g
elec
tr
icity
th
ef
t
an
d
f
alse
d
ata
i
n
jectio
n
b
ased
o
n
g
en
e
r
alize
d
r
o
b
u
s
t
m
etr
ics
,
m
u
lti
-
class
if
icatio
n
s
u
p
p
o
r
t
v
e
cto
r
m
ac
h
in
e
(
SVM)
[
1
3
]
,
q
u
a
n
tu
m
m
ac
h
in
e
lear
n
in
g
[
1
4
]
,
h
y
b
r
id
m
o
d
e
l
s
s
u
ch
as
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN
)
-
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
an
d
b
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
it
(
Bi
-
GR
U
)
-
b
id
ir
ec
tio
n
al
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
Bi
-
L
STM
)
[
1
5
]
,
a
n
d
C
NN
-
ad
ap
tiv
e
b
o
o
s
tin
g
(
Ad
a
B
o
o
st
)
[
1
6
]
.
Fo
r
in
s
tan
ce
,
Gu
n
d
u
z
an
d
Das
[
1
7
]
p
r
o
p
o
s
ed
an
ef
f
ec
tiv
e
h
y
b
r
id
C
NN
-
b
ased
ap
p
r
o
ac
h
f
o
r
d
etec
tin
g
en
er
g
y
th
ef
t
u
s
in
g
co
n
s
u
m
p
tio
n
p
atter
n
s
o
n
s
ix
d
if
f
er
en
t
d
atasets
.
B
ai
et
a
l.
[
1
8
]
in
tr
o
d
u
ce
d
a
C
atch
2
2
-
C
o
n
v
-
T
r
an
s
f
o
r
m
e
r
m
eth
o
d
f
o
r
d
etec
tio
n
o
f
el
ec
tr
icity
th
ef
t
o
n
a
d
ataset
o
f
th
e
State
Gr
id
C
o
r
p
o
r
atio
n
o
f
C
h
in
a
(
SGC
C
)
,
wh
ile
Nir
m
al
et
a
l.
[
1
9
]
s
u
g
g
ested
a
h
y
b
r
id
C
NN
-
Ad
a
B
o
o
s
t
m
o
d
el
f
o
r
elec
tr
icity
th
ef
t
d
etec
tio
n
in
th
e
s
m
ar
t
g
r
id
o
n
th
e
SGC
C
d
ataset.
Similar
ly
,
I
q
b
al
et
a
l.
[
2
0
]
p
r
esen
ted
C
NN
-
L
STM
an
d
B
i
-
G
R
U
-
B
i
L
STM
h
y
b
r
id
m
o
d
els
f
o
r
id
e
n
tific
atio
n
o
f
elec
tr
icity
th
ef
t
o
n
tem
p
o
r
al
s
eq
u
e
n
ce
d
ata
o
f
SGC
C
.
Mo
r
eo
v
er
,
Alg
ar
n
i
et
a
l.
[
2
1
]
p
r
esen
ted
C
NN
-
k
-
n
ea
r
est
n
eig
h
b
o
r
s
(
KN
N
)
an
d
C
NN
-
L
STM
h
y
b
r
id
m
o
d
els
f
o
r
elec
tr
icity
th
ef
t
d
etec
tio
n
o
n
th
e
S
GC
C
d
ata
s
et.
Dee
p
lear
n
in
g
m
eth
o
d
s
,
s
u
ch
as
g
r
ad
ien
t
-
weig
h
ted
class
ac
tiv
atio
n
m
ap
p
in
g
(
Gr
ad
C
AM
)
[
2
2
]
,
r
an
d
o
m
f
o
r
est
(
R
F)
an
d
KNN
[
2
3
]
,
p
ar
t
icle
s
war
m
o
p
tim
izatio
n
(
PS
O)
-
b
ased
ad
ap
tiv
e
g
e
n
er
ativ
e
ad
v
er
s
ar
ial
n
etwo
r
k
s
(
AGAN
)
[
2
4
]
,
an
d
h
y
b
r
id
C
NN
-
m
u
ltil
ay
er
p
er
ce
p
tr
o
n
(
MLP
)
m
o
d
el
[
2
5
]
h
av
e
also
b
ee
n
s
u
cc
ess
f
u
lly
ap
p
lied
to
d
e
tect
elec
tr
icity
th
ef
t
in
s
m
ar
t g
r
id
s
y
s
tem
s
.
Desp
ite
th
ese
ad
v
an
ce
s
,
ch
allen
g
es r
em
ain
r
eg
a
r
d
in
g
d
etec
tio
n
p
r
ec
is
io
n
a
n
d
ad
a
p
tab
ilit
y
to
ir
r
eg
u
lar
co
n
s
u
m
p
tio
n
b
eh
av
io
r
s
.
Ma
n
y
ex
is
tin
g
m
o
d
els
s
tr
u
g
g
le
to
ef
f
ec
tiv
ely
ca
p
tu
r
e
s
u
d
d
en
v
ar
i
atio
n
s
in
u
s
er
lo
ad
p
r
o
f
iles
an
d
to
m
ain
tain
r
o
b
u
s
tn
ess
ac
r
o
s
s
h
eter
o
g
en
eo
u
s
d
atasets
.
T
o
ad
d
r
ess
th
ese
lim
itatio
n
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
a
n
o
v
el
h
y
b
r
id
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
,
n
am
e
d
b
id
i
r
ec
tio
n
al
lo
o
k
ah
ea
d
a
d
ap
tiv
e
m
o
m
en
t
esti
m
atio
n
(
Bi
-
L
ADAM
)
-
SVM
-
g
ated
r
ec
u
r
r
en
t
u
n
it
(
GR
U
)
,
wh
ic
h
co
m
b
in
es
Bi
-
L
ADAM
,
SVM,
an
d
GR
U
,
d
esig
n
ed
to
en
h
an
ce
th
e
ac
c
u
r
ac
y
a
n
d
ef
f
icien
cy
o
f
elec
tr
icity
th
e
f
t
d
etec
tio
n
.
T
h
is
m
o
d
el
in
teg
r
ates
th
e
s
tr
en
g
th
s
o
f
s
ev
er
al
ar
ch
itectu
r
es:
a
B
i
-
L
STM
n
etwo
r
k
to
ca
p
tu
r
e
lo
n
g
-
t
er
m
tem
p
o
r
al
d
ep
en
d
en
cies,
a
n
ad
ap
tiv
e
m
o
m
en
t
esti
m
atio
n
(
ADAM
)
alg
o
r
ith
m
f
o
r
o
p
tim
al
h
y
p
er
p
ar
a
m
eter
tu
n
in
g
,
a
n
SVM
class
if
ier
f
o
r
r
o
b
u
s
t
p
atter
n
d
is
cr
im
in
atio
n
,
an
d
a
B
i
-
GR
U
f
o
r
m
o
d
elin
g
s
h
o
r
t
-
ter
m
d
y
n
a
m
ics.
T
h
e
m
ajo
r
c
o
n
tr
ib
u
tio
n
s
o
f
t
h
is
wo
r
k
ar
e
g
i
v
en
as f
o
llo
ws:
−
F
ir
s
t
p
r
esen
t
a
co
m
p
r
e
h
en
s
iv
e
ev
alu
atio
n
o
f
d
ee
p
lear
n
in
g
m
o
d
els
co
m
m
o
n
ly
u
s
ed
f
o
r
el
ec
tr
icity
th
ef
t
d
etec
tio
n
,
in
clu
d
in
g
L
STM
,
C
NN,
B
i
-
L
STM
,
B
i
-
G
R
U,
e
x
tr
em
e
g
r
ad
ie
n
t
b
o
o
s
tin
g
(
X
GB
o
o
s
t
)
,
an
d
d
ec
is
io
n
tr
ee
.
−
T
h
en
,
p
r
o
p
o
s
e
d
a
n
e
w
h
y
b
r
i
d
m
o
d
el
(
B
i
-
L
AD
AM
-
SVM
-
GR
U)
co
m
b
i
n
i
n
g
B
i
-
L
S
T
M
,
AD
AM
o
p
tim
iz
er
,
SVM,
a
n
d
B
i
-
GR
U
t
o
e
n
h
a
n
ce
te
m
p
o
r
a
l
f
e
at
u
r
e
e
x
t
r
ac
ti
o
n
a
n
d
im
p
r
o
v
e
class
i
f
i
ca
t
i
o
n
a
cc
u
r
ac
y
.
−
Su
b
s
eq
u
en
tly
,
th
is
s
tu
d
y
im
p
l
em
en
ted
an
ex
p
er
im
en
tal
v
ali
d
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
u
s
in
g
a
r
ea
l
co
n
s
u
m
p
tio
n
d
ataset
f
r
o
m
th
e
SGC
C
to
ass
es
s
its
p
r
ac
tical
ef
f
ec
tiv
en
ess
.
−
I
n
ad
d
itio
n
,
th
is
s
tu
d
y
ev
alu
ated
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el
th
r
o
u
g
h
m
u
ltip
l
e
s
tatis
t
ical
in
d
icato
r
s
s
u
ch
as
m
ea
n
ab
s
o
l
u
te
er
r
o
r
(
MA
E
)
,
m
ea
n
ab
s
o
lu
te
p
er
ce
n
tag
e
e
r
r
o
r
(
MA
PE)
,
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
,
ac
cu
r
ac
y
r
ate,
F1
-
s
co
r
e,
r
ec
a
ll,
a
r
ea
u
n
d
e
r
th
e
cu
r
v
e
(
AUC
)
,
R
-
s
q
u
ar
ed
(
R
2
)
,
an
d
ad
ju
s
ted
R
2
an
d
co
m
p
a
r
e
d
with
b
aselin
e
s
tr
ateg
ies
s
u
ch
as
C
NN,
ar
tific
ia
l
n
eu
r
al
n
etwo
r
k
(
ANN
)
,
ML
P,
d
ee
p
n
eu
r
al
n
e
two
r
k
(
D
NN
)
,
an
d
lin
ea
r
r
eg
r
ess
io
n
.
T
h
e
r
est
o
f
th
is
s
tu
d
y
is
o
r
g
an
ized
as
f
o
llo
ws
.
S
ec
tio
n
2
d
escr
ib
es
th
e
m
ater
ials
.
Sectio
n
3
d
escr
ib
es
th
e
m
eth
o
d
o
lo
g
ies
em
p
lo
y
ed
.
Sectio
n
4
p
r
esen
ts
an
d
d
is
cu
s
s
es
th
e
ex
p
er
im
en
tal
r
esu
lts
.
Fin
ally
,
s
ec
tio
n
5
co
n
clu
d
es t
h
e
wo
r
k
a
n
d
o
u
tlin
es p
er
s
p
ec
tiv
es f
o
r
f
u
tu
r
e
r
ese
ar
ch
.
2.
M
AT
E
R
I
AL
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
e
d
ataset
g
iv
en
in
T
ab
le
1
was
r
ec
o
r
d
ed
f
r
o
m
s
m
ar
t
m
ete
r
s
o
r
elec
tr
icity
co
n
s
u
m
er
s
en
s
o
r
s
o
v
er
a
p
er
io
d
o
f
3
4
m
o
n
th
s
f
r
o
m
SGC
C
.
SGC
C
o
f
f
er
s
a
d
ataset
co
n
s
is
tin
g
o
f
4
2
,
3
7
2
u
s
er
s
o
v
er
1
,
0
3
5
d
ay
s
f
r
o
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
0
9
8
-
1
1
1
1
1100
1
J
an
u
ar
y
to
3
1
Octo
b
er
2
0
1
6
.
T
h
er
e
ar
e
3
8
,
7
5
7
n
o
r
m
al
u
s
er
s
m
ar
k
ed
as
“0
”
a
n
d
3
,
6
1
5
ele
ctr
icity
th
ef
t
u
s
er
s
m
ar
k
ed
as “
1
”
;
t
h
e
elec
tr
icity
t
h
ef
t
u
s
er
s
ac
co
u
n
t f
o
r
8
.
5
% o
f
th
e
g
lo
b
al
d
ataset.
T
ab
le
1
.
Deta
iled
d
escr
ip
tio
n
o
f
th
e
d
ataset
Ty
p
e
o
f
d
a
t
a
En
e
r
g
y
c
o
n
su
m
p
t
i
o
n
R
e
s
o
l
u
t
i
o
n
1
d
a
y
N
u
mb
e
r
o
f
c
o
n
s
u
m
e
r
s
42
,
372
S
i
z
e
1
,
0
3
5
d
a
y
s
N
o
r
mal
u
s
e
r
s
38
,
757
El
e
c
t
r
i
c
i
t
y
t
h
e
f
t
3
,
6
1
5
P
r
e
p
r
o
c
e
ss
i
n
g
d
a
t
a
7
0
%
Te
st
i
n
g
d
a
t
a
3
0
%
I
mb
a
l
a
n
c
e
d
d
a
t
a
2
0
%
2
.
2
.
P
y
t
ho
n
Py
th
o
n
is
a
p
r
o
g
r
am
m
in
g
lan
g
u
ag
e
i
n
th
e
f
ield
o
f
m
ac
h
i
n
e
lear
n
in
g
an
d
d
ee
p
lear
n
in
g
.
W
ith
in
th
e
f
r
am
ewo
r
k
o
f
th
is
s
tu
d
y
,
it
s
er
v
ed
as
th
e
f
o
u
n
d
atio
n
f
o
r
d
e
v
elo
p
in
g
th
e
p
r
o
p
o
s
ed
d
ee
p
lea
r
n
in
g
m
o
d
el
.
U
s
in
g
its
k
ey
lib
r
ar
ies,
in
clu
d
in
g
Ma
tp
lo
tlib
,
Seab
o
r
n
,
Nu
m
Py
,
Pan
d
as,
an
d
s
cik
it
-
lear
n
.
2
.
3
.
M
AT
L
AB
MA
T
L
AB
is
a
s
p
ec
ialized
p
r
o
g
r
am
m
in
g
e
n
v
ir
o
n
m
en
t
u
s
ed
in
en
g
i
n
ee
r
in
g
a
n
d
s
cien
tific
ap
p
licatio
n
s
.
I
n
th
is
wo
r
k
,
tr
ain
in
g
a
n
d
e
v
alu
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
w
er
e
m
ad
e
p
o
s
s
ib
le
u
s
in
g
th
e
MA
T
L
AB
2
0
2
1
a
v
er
s
io
n
.
E
v
a
lu
atio
n
o
f
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
was a
ls
o
co
n
d
u
cted
u
s
in
g
th
e
s
am
e
en
v
ir
o
n
m
en
t.
2
.
4
.
Co
m
pu
t
er
All
e
x
p
e
r
i
m
e
n
ts
a
n
d
c
o
m
p
u
t
ati
o
n
s
f
o
r
t
h
is
s
t
u
d
y
w
e
r
e
p
e
r
f
o
r
m
e
d
o
n
a
DE
L
L
c
o
m
p
u
te
r
e
q
u
ip
p
e
d
wi
th
a
C
o
r
e
i
7
p
r
o
ce
s
s
o
r
(
2
.
2
0
G
Hz)
,
8
GB
o
f
R
A
M,
an
d
W
in
d
o
ws
1
0
o
p
e
r
a
ti
n
g
s
y
s
t
em
.
T
h
is
c
o
n
f
i
g
u
r
at
io
n
f
a
cili
tat
ed
t
h
e
ex
ec
u
ti
o
n
o
f
a
ll
i
m
p
le
m
e
n
t
e
d
d
e
e
p
lea
r
n
i
n
g
al
g
o
r
it
h
m
s
,
i
n
cl
u
d
i
n
g
L
ST
M
,
GR
U,
B
i
-
L
S
T
M
n
et
wo
r
k
s
,
Bi
-
G
R
U
n
e
t
w
o
r
k
s
,
S
V
M
,
C
N
Ns
,
X
GB
o
o
s
t
,
a
n
d
d
e
c
i
s
i
o
n
t
r
e
es
.
A
l
l
a
l
g
o
r
i
t
h
m
s
w
e
r
e
e
x
e
c
u
t
e
d
w
i
th
e
x
c
e
l
l
e
n
t
c
o
m
p
u
t
at
i
o
n
a
l
ti
m
e
.
3.
M
E
T
H
O
D
3
.
1
.
L
o
ng
s
ho
rt
-
t
er
m
m
emo
ry
L
STM
is
a
s
p
ec
ialized
v
ar
ian
t
o
f
th
e
r
ec
u
r
r
en
t
n
e
u
r
al
n
etwo
r
k
(
R
NN)
ar
ch
itectu
r
e,
d
esig
n
ed
to
m
o
r
e
ef
f
ec
tiv
ely
m
o
d
el
lo
n
g
-
r
an
g
e
t
em
p
o
r
al
d
e
p
en
d
e
n
cies.
A
k
ey
ad
v
an
tag
e
o
f
L
STM
is
its
ex
c
ep
tio
n
al
ca
p
ab
ilit
y
in
p
r
ep
r
o
ce
s
s
in
g
s
eq
u
en
tial
d
a
ta,
s
u
ch
as
tim
e
s
er
i
es.
T
h
er
e
ar
e
in
ter
ac
tio
n
s
b
etwe
en
th
e
v
ar
io
u
s
co
m
p
o
n
en
ts
;
ℎ
−
1
an
d
ℎ
in
d
icate
th
e
o
u
tp
u
t
o
f
th
e
p
r
ec
ed
in
g
a
n
d
cu
r
r
en
t
ce
ll,
r
esp
ec
tiv
ely
.
T
h
e
cu
r
r
e
n
t
u
n
it
in
p
u
t
is
s
y
m
b
o
lized
as
,
wh
ile
th
e
n
eu
r
al
n
etwo
r
k
lay
e
r
is
r
ep
r
esen
ted
b
y
a
b
o
x
.
3
.
2
.
Su
pp
o
rt
v
ec
t
o
r
ma
chine
T
h
e
SVM
alg
o
r
ith
m
was
d
ev
elo
p
ed
in
t
h
e
1
9
8
0
s
b
y
Vlad
im
ir
Vap
n
ik
u
s
in
g
m
at
h
em
atica
l
an
d
o
p
tim
izatio
n
ev
o
lu
tio
n
ar
y
to
o
l
s
,
co
n
s
id
er
in
g
h
ig
h
-
d
im
en
s
io
n
al
d
atasets
an
d
g
o
o
d
p
r
ac
tical
r
esu
lts
.
Pra
ctica
lly
,
th
e
SVM
is
m
ad
e
o
f
two
g
r
o
u
p
s
th
at
ar
e
s
ep
a
r
ated
by
a
h
y
p
er
p
lan
e
.
H
o
wev
er
,
th
er
e
i
s
an
o
p
tim
al
s
p
ac
e
am
o
n
g
t
h
e
s
ep
ar
atio
n
p
lan
es
,
wh
er
e
th
e
d
is
tan
ce
b
etwe
en
t
h
e
clo
s
est
h
y
p
er
p
lan
e
an
d
th
e
tr
ain
in
g
elem
e
n
t
is
o
p
tim
al.
T
h
e
p
r
o
b
lem
is
to
f
in
d
th
e
o
p
tim
al
s
ep
ar
atio
n
f
r
o
n
ti
er
o
f
th
e
s
am
p
les
b
etwe
en
t
h
e
two
p
ar
ts
.
So
lv
in
g
th
ese
p
r
o
b
lem
s
in
v
o
lv
es c
o
n
s
t
r
u
ctin
g
a
h
y
p
e
r
p
lan
e
ℎ
with
weig
h
ts
an
d
0
as p
r
esen
ted
in
(
1
)
.
ℎ
(
)
=
+
0
(
1
)
3
.
3
.
G
a
t
ed
re
curr
ent
u
nit
net
wo
rk
T
h
e
GR
U
n
etwo
r
k
is
a
k
i
n
d
o
f
R
NN
th
at
h
as
g
ar
n
er
ed
co
n
s
id
er
ab
le
in
te
r
est
f
o
r
its
ca
p
ac
ity
to
ef
f
ec
tiv
ely
ca
p
tu
r
e
an
d
m
o
d
e
l
lo
n
g
-
ter
m
r
elatio
n
s
h
ip
s
.
GR
U
n
etwo
r
k
s
d
is
tin
g
u
is
h
th
em
s
elv
es
f
r
o
m
L
STM
n
etwo
r
k
s
b
y
in
te
g
r
atin
g
t
h
e
c
ell
s
tate
an
d
h
id
d
en
s
tate
in
to
a
u
n
if
ied
v
a
r
iab
le,
d
e
n
o
ted
as
ℎ
,
wh
er
ea
s
L
STM
n
etwo
r
k
s
m
ain
tain
th
em
as
d
i
s
tin
ct
en
titi
es.
GR
U
n
etwo
r
k
s
em
p
lo
y
th
e
f
o
llo
win
g
m
at
h
e
m
atica
l
eq
u
atio
n
s
t
o
ca
r
r
y
o
u
t
t
h
is
m
er
g
in
g
.
I
n
(
2
)
an
d
(
3
)
,
in
p
u
ts
ar
e
c
o
llected
f
r
o
m
th
e
p
r
e
v
io
u
s
h
id
d
en
s
tat
e
an
d
th
e
cu
r
r
en
t
d
ata
p
o
in
t,
wh
ich
ar
e
th
en
p
r
o
ce
s
s
ed
with
r
esp
ec
tiv
e
weig
h
ts
.
=
(
[
ℎ
−
1
,
]
+
)
(
2
)
=
(
[
ℎ
−
1
,
]
+
)
(
3
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
n
o
ve
l c
o
mb
i
n
ed
d
ee
p
lea
r
n
i
n
g
-
b
a
s
ed
tech
n
iq
u
e
fo
r
elec
tr
icity
th
eft
…
(
E
tien
n
e
F
r
a
n
ç
o
is
Mo
u
ck
o
mey
)
1101
3
.
4
.
Co
nv
o
lutio
na
l
neura
l net
wo
rk
s
C
NNs
ar
e
a
s
p
ec
if
ic
ca
teg
o
r
y
o
f
ANNs
th
at
h
av
e
g
ar
n
er
e
d
co
n
s
id
er
ab
le
i
n
ter
est
b
ec
au
s
e
o
f
th
eir
o
u
ts
tan
d
in
g
ca
p
a
b
ilit
ies
in
tas
k
s
s
u
ch
as
im
ag
e
id
en
tific
atio
n
,
o
b
ject
d
etec
tio
n
,
an
d
m
e
d
i
ca
l
im
ag
e
an
aly
s
is
.
T
h
e
s
tr
u
ctu
r
e
o
f
C
NNs
g
en
er
ally
co
m
p
r
is
es
co
n
v
o
lu
tio
n
al
lay
er
s
,
p
o
o
lin
g
lay
er
s
,
an
d
f
u
lly
co
n
n
ec
ted
lay
er
s
.
C
o
n
v
o
lu
tio
n
al
la
y
er
s
f
ac
ilit
ate
th
e
n
etwo
r
k
in
ef
f
ec
tiv
el
y
ca
p
tu
r
in
g
s
p
atial
h
ier
ar
c
h
ies
an
d
p
atter
n
s
in
th
e
in
p
u
t
d
ata,
h
en
ce
m
ak
in
g
C
NNs
v
er
y
s
u
cc
ess
f
u
l
f
o
r
task
s
t
h
at
in
clu
d
e
v
is
u
al
in
f
o
r
m
atio
n
.
Mo
r
eo
v
er
,
C
NNs
h
av
e
p
r
o
v
en
to
b
e
ef
f
ec
tiv
e
i
n
a
wid
e
r
an
g
e
o
f
tim
e
s
er
ies
ap
p
licatio
n
s
,
s
u
ch
as
class
if
y
in
g
d
ata,
d
etec
tin
g
an
o
m
alies,
an
d
p
r
ed
ictin
g
o
u
tc
o
m
es.
3
.
5
.
XG
B
o
o
s
t
decisi
o
n t
re
e
XGBo
o
s
t
d
ec
is
io
n
tr
ee
is
em
p
lo
y
ed
i
n
s
u
p
er
v
is
ed
m
ac
h
in
e
lear
n
in
g
a
p
p
licatio
n
s
f
o
r
class
if
icatio
n
,
r
eg
r
ess
io
n
,
an
d
r
an
k
i
n
g
.
I
t
wo
r
k
s
b
y
iter
ativ
ely
tr
ain
i
n
g
d
ec
is
io
n
tr
ee
s
o
n
th
e
r
esid
u
al
er
r
o
r
s
o
f
th
e
p
r
ev
io
u
s
tr
ee
.
T
h
e
s
y
s
tem
em
p
lo
y
s
g
r
ad
ien
t
d
escen
t
o
p
tim
izatio
n
t
o
r
ed
u
ce
a
lo
s
s
f
u
n
ctio
n
,
w
h
ich
q
u
a
n
tifie
s
th
e
d
is
p
ar
ity
b
etwe
en
t
h
e
an
t
icip
a
ted
an
d
ac
tu
al
v
alu
es.
T
h
e
alg
o
r
ith
m
is
u
s
ed
f
o
r
s
u
p
er
v
is
ed
lear
n
in
g
p
r
o
b
lem
s
b
y
p
r
ed
ictin
g
th
e
tar
g
et
lab
el
with
tr
ain
in
g
d
ata
.
T
r
ain
in
g
lo
s
s
an
d
r
eg
u
lar
izatio
n
ter
m
s
co
m
p
r
is
e
th
e
o
b
jectiv
e
f
u
n
ctio
n
o
f
XGBo
o
s
t a
s
in
(
4
)
.
(
)
=
∑
(
,
̂
)
+
(
)
=
1
(
4
)
3
.
6
.
P
er
f
o
r
m
a
nce
m
e
t
rics
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
ev
alu
ated
u
s
in
g
t
h
e
f
o
llo
win
g
m
etr
ics.
i)
Acc
u
r
ac
y
:
wh
ich
d
escr
ib
es
t
h
e
o
v
er
all
p
er
f
o
r
m
a
n
ce
o
f
a
m
o
d
el
ac
r
o
s
s
all
class
e
s
.
T
h
e
ca
lcu
latio
n
in
v
o
lv
es
d
eter
m
i
n
in
g
t
h
e
p
r
o
p
o
r
tio
n
o
f
ac
cu
r
ate
p
r
e
d
ictio
n
s
r
elativ
e
to
th
e
o
v
er
all
n
u
m
b
e
r
o
f
f
o
r
ec
asts
m
ad
e
,
as sh
o
wn
in
(
5
)
.
=
+
+
+
+
(
5
)
ii)
Pre
cisi
o
n
:
wh
ich
ev
alu
ates th
e
ab
ilit
y
o
f
th
e
m
o
d
el
t
o
co
r
r
ec
tly
class
if
y
a
s
am
p
le
,
as d
ef
in
e
d
in
(
6
)
.
=
+
(
6
)
iii)
R
ec
all: w
h
ich
class
if
ies th
e
p
o
s
itiv
e
s
am
p
les
,
as g
iv
en
in
(
7
)
.
=
+
(
7
)
iv
)
F1
-
s
co
r
e:
wh
ich
g
iv
es th
e
r
ela
tio
n
s
h
ip
b
etwe
en
p
r
ec
is
io
n
an
d
s
en
s
itiv
ity
,
as p
r
esen
ted
in
(
8
)
.
1
−
=
2
×
(
2
×
+
+
)
(
8
)
W
ith
tr
u
e
p
o
s
itiv
e
(
T
P),
f
alse
p
o
s
itiv
e
(
FP
)
,
tr
u
e
n
eg
ativ
e
(
T
N)
,
an
d
f
alse
n
eg
ativ
e
(
FN)
.
T
h
ese
m
etr
ics
r
ep
r
esen
t th
e
p
o
s
s
ib
ilit
ies o
f
th
e
co
n
s
u
m
er
s
b
ein
g
eith
er
a
tr
u
e
o
r
a
f
alse c
o
n
s
u
m
er
.
v)
M
S
E
:
r
e
p
r
es
e
n
ts
m
e
a
n
o
f
s
q
u
a
r
e
d
i
f
f
e
r
e
n
c
e
b
e
t
w
ee
n
t
h
e
p
r
e
d
i
ct
e
d
v
a
l
u
e
s
a
n
d
r
e
al
v
a
l
u
e
s
,
a
s
d
e
f
i
n
e
d
i
n
(
9
)
.
=
1
∑
(
̃
−
)
2
=
1
(
9
)
v
i)
R
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
(
R
MSE
)
:
r
ep
r
esen
ts
th
e
s
tan
d
ar
d
d
ev
iatio
n
b
etwe
en
p
r
ed
icted
v
alu
es
an
d
r
ea
l
v
a
l
u
e
s
,
as g
iv
en
in
(
1
0
)
.
=
√
1
∑
(
̃
−
)
2
=
1
(
1
0
)
v
ii)
M
A
E
:
r
ep
r
esen
ts
th
e
ab
s
o
lu
te
v
alu
e
o
f
t
h
e
d
if
f
e
r
en
ce
b
etwe
en
th
e
p
r
e
d
icted
v
alu
es
an
d
th
e
r
ea
l
v
alu
es
,
as p
r
esen
ted
in
(
1
1
)
.
=
1
∑
|
̃
−
|
=
1
(
1
1
)
v
iii)
MA
PE:
r
ep
r
esen
ts
th
e
m
ea
n
o
f
ab
s
o
lu
te
p
er
ce
n
tag
e
p
r
ed
icti
o
n
er
r
o
r
s
,
as f
o
r
m
u
lated
in
(
1
2
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
0
9
8
-
1
1
1
1
1102
=
1
∑
|
̃
−
|
×
100%
=
1
(
1
2
)
ix
)
R
2
:
r
ep
r
esen
ts
th
e
p
r
o
p
o
r
tio
n
o
f
v
a
r
ian
ce
in
th
e
d
ep
en
d
en
t
v
ar
iab
le
th
at
is
p
r
e
d
ictab
le
f
r
o
m
th
e
in
d
ep
en
d
en
t
v
ar
iab
les,
as
ex
p
r
ess
ed
in
(
1
3
)
.
W
ith
th
e
r
ea
l
v
alu
e,
̃
th
e
p
r
ed
icted
v
alu
e
,
an
d
̄
th
e
m
ea
n
v
alu
e
o
f
th
e
r
ec
o
r
d
ed
f
e
atu
r
es.
2
=
1
−
∑
(
̃
−
)
2
=
1
∑
(
̃
−
̄
)
=
1
(
1
3
)
T
h
e
h
ig
h
est
ac
cu
r
ac
y
m
ea
n
s
th
at
th
e
f
r
au
d
u
len
t
co
n
s
u
m
er
h
as
b
ee
n
d
etec
ted
.
W
h
en
th
e
r
ec
all
a
n
d
F1
-
s
co
r
e
ar
e
h
ig
h
,
it m
ea
n
s
th
at
a
co
n
s
u
m
er
is
d
etec
ted
as f
r
au
d
u
len
t o
r
n
o
r
m
al.
3
.
7
.
P
r
o
po
s
ed
m
o
dels
3
.
7
.
1
.
L
o
o
ka
hea
d
a
da
ptiv
e
mo
m
ent
estima
t
io
n
mo
del
T
h
e
L
ADAM
m
o
d
el
is
a
h
y
b
r
id
alg
o
r
ith
m
th
at
co
m
b
in
es
th
e
L
STM
with
th
e
ADAM
a
lg
o
r
ith
m
f
o
r
o
p
tim
izin
g
n
eu
r
al
n
etwo
r
k
h
y
p
er
p
a
r
am
eter
s
.
ADAM
is
o
n
e
o
f
th
e
m
o
s
t
r
o
b
u
s
t
o
p
tim
izer
s
th
at
im
p
r
o
v
es
tr
ain
in
g
s
p
ee
d
s
in
DNN
to
ac
h
iev
e
co
n
v
e
r
g
en
ce
q
u
ick
ly
.
I
t
co
m
b
in
es
th
e
b
est
elem
en
ts
o
f
m
o
m
en
t
u
m
an
d
p
r
o
ce
s
s
in
g
.
Data
ar
e
p
r
ep
r
o
ce
s
s
ed
to
v
er
if
y
th
e
ab
s
en
ce
o
f
n
u
ll
an
d
m
is
s
in
g
v
alu
es
an
d
to
d
etec
t
an
y
o
u
tlier
s
.
Data
s
ca
lin
g
n
o
r
m
alize
s
th
e
r
an
g
e
o
f
v
ar
iab
les.
T
h
e
d
ata
ar
e
s
ep
ar
ated
in
to
t
r
ain
in
g
an
d
t
esti
n
g
s
ets
u
s
in
g
a
tr
ain
in
g
an
d
test
s
p
lit,
en
s
u
r
in
g
th
at
th
e
tem
p
o
r
al
o
r
d
er
is
p
r
eser
v
ed
.
T
h
e
p
r
e
p
ar
ed
d
ata
ca
n
b
e
f
u
r
th
e
r
u
tili
ze
d
f
o
r
m
ac
h
in
e
lear
n
i
n
g
p
u
r
p
o
s
e
s
by
em
p
lo
y
in
g
d
ee
p
L
STM
m
o
d
els
o
p
tim
ized
b
y
th
e
ADAM
alg
o
r
ith
m
,
as
d
ep
icted
in
Fig
u
r
e
1
.
T
h
e
m
o
d
els
wer
e
tr
ain
ed
u
s
in
g
a
r
ea
lis
tic
elec
tr
icity
co
n
s
u
m
p
tio
n
d
atab
ase
th
at
was
m
ad
e
av
ailab
le
b
y
th
e
SGC
C
.
Fig
u
r
e
1
.
Pro
p
o
s
ed
f
l
o
wch
ar
t f
o
r
th
e
L
ADAM
h
y
b
r
id
alg
o
r
it
h
m
3
.
7
.
2
.
P
ro
po
s
ed
m
et
ho
do
lo
g
y
f
ra
m
ewo
r
k
f
o
r
elec
t
ricit
y
t
h
ef
t
det
ec
t
io
n
I
n
th
is
s
tu
d
y
,
a
co
m
b
i
n
ed
s
ch
e
m
e
b
ased
o
n
L
ADAM
,
B
i
-
L
STM
,
B
i
-
GR
U
,
an
d
SVM
is
im
p
lem
en
ted
f
o
r
t
h
e
d
etec
tio
n
o
f
elec
tr
icity
th
ef
t
i
n
a
s
m
ar
t
p
o
wer
g
r
id
.
T
h
e
L
ADAM
m
o
d
el
,
m
a
d
e
o
f
L
STM
an
d
ADAM
o
p
tim
izer
,
is
d
ev
elo
p
ed
f
o
r
m
ea
n
in
g
f
u
l
f
ea
tu
r
e
e
x
tr
ac
tio
n
;
th
e
B
i
-
L
STM
is
u
s
ed
f
o
r
t
h
e
ex
tr
ac
tio
n
,
w
h
ile
L
ADAM
s
er
v
es
as
f
ea
tu
r
e
s
elec
tio
n
.
T
h
e
SVM
allo
ws
to
class
if
y
th
e
e
x
tr
ac
ted
f
ea
t
u
r
es
as
n
o
r
m
al
o
r
f
r
au
d
u
le
n
t
,
wh
ile
GR
U
i
s
u
s
ed
f
o
r
f
ea
tu
r
e
p
r
e
d
ictio
n
.
T
h
e
f
l
o
wch
ar
t
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
lo
g
y
f
r
am
ewo
r
k
f
o
r
elec
tr
icity
th
e
f
t
d
etec
tio
n
is
g
iv
en
in
Fig
u
r
e
2
.
T
h
is
f
lo
wch
ar
t
o
u
tlin
es
th
e
k
ey
s
ta
g
es,
in
clu
d
in
g
d
ata
co
llectio
n
,
d
ata
an
aly
s
is
an
d
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
ex
tr
a
ctio
n
an
d
s
elec
tio
n
,
f
ea
tu
r
e
class
if
icatio
n
an
d
p
r
ed
ictio
n
,
n
o
r
m
aliza
tio
n
,
ap
p
licatio
n
o
f
d
ee
p
lea
r
n
in
g
-
b
ase
d
t
ec
h
n
iq
u
es,
h
y
p
er
p
ar
am
eter
o
p
tim
izatio
n
,
an
d
p
er
f
o
r
m
an
ce
ev
alu
atio
n
m
etr
i
cs.
T
h
is
s
tr
u
ctu
r
ed
f
r
am
ewo
r
k
en
s
u
r
es
clar
ity
in
th
e
s
tep
s
f
o
llo
wed
f
o
r
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
n
o
ve
l c
o
mb
i
n
ed
d
ee
p
lea
r
n
i
n
g
-
b
a
s
ed
tech
n
iq
u
e
fo
r
elec
tr
icity
th
eft
…
(
E
tien
n
e
F
r
a
n
ç
o
is
Mo
u
ck
o
mey
)
1103
d
ev
elo
p
m
e
n
t
an
d
v
alid
atio
n
.
T
h
e
p
ar
am
eter
s
o
f
th
e
m
o
d
els
ar
e
d
escr
ib
ed
as
f
o
llo
ws:
4
c
o
n
v
o
lu
ti
o
n
lay
er
s
,
a
s
o
f
t
m
ax
L
STM
tr
an
s
f
er
f
u
n
ctio
n
,
7
m
i
n
im
u
m
b
atc
h
s
ize,
6
0
m
ax
im
u
m
ep
o
ch
,
1
0
0
iter
atio
n
s
,
a
lear
n
in
g
r
at
e
o
f
0
.
0
0
0
1
,
a
n
ADAM
o
p
tim
izer
,
a
3
0
%
d
r
o
p
o
u
t
r
ate,
an
d
a
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
e
m
eth
o
d
o
lo
g
y
f
o
r
d
ata
s
p
litt
in
g
,
cr
o
s
s
-
v
alid
atio
n
,
an
d
h
y
p
er
p
ar
am
et
er
o
p
tim
iza
tio
n
is
d
escr
ib
ed
as f
o
llo
ws:
i)
Fo
r
th
e
tr
ain
in
g
-
v
alid
atio
n
s
p
l
itti
n
g
m
eth
o
d
,
th
e
d
ataset
is
o
r
g
an
ized
b
y
ch
ar
ac
te
r
is
tics
an
d
d
ata
n
atu
r
e
,
en
s
u
r
in
g
t
h
at
ea
ch
g
r
o
u
p
co
n
tain
s
a
tim
e
s
er
ies.
T
h
is
g
r
o
u
p
in
g
s
tr
ateg
y
e
n
ab
les
co
n
t
r
o
lled
o
v
er
lap
h
an
d
lin
g
,
wh
ich
is
ess
en
tial
f
o
r
s
eq
u
e
n
tial
lear
n
i
n
g
m
o
d
els
.
T
h
e
d
ataset
is
p
a
r
titi
o
n
ed
as
f
o
llo
ws:
7
0
%
f
o
r
tr
ain
in
g
,
1
5
% f
o
r
v
alid
atio
n
,
an
d
1
5
% f
o
r
test
in
g
.
ii)
T
o
en
h
an
ce
g
en
er
aliza
tio
n
,
d
a
ta
s
eg
m
en
ts
f
r
o
m
all
s
m
ar
t
m
eter
s
ar
e
ag
g
r
eg
ated
to
f
o
r
m
c
o
m
p
r
eh
e
n
s
iv
e
g
lo
b
al
s
ets.
T
h
is
ag
g
r
eg
atio
n
allo
ws
f
o
r
ass
ess
in
g
tem
p
o
r
al
d
eg
r
ad
atio
n
an
d
f
o
r
cla
s
s
if
y
in
g
n
ew,
p
r
ev
io
u
s
ly
u
n
s
ee
n
cu
s
to
m
er
s
.
Ad
d
itio
n
ally
,
s
am
p
le
weig
h
tin
g
,
f
o
ca
l
lo
s
s
,
an
d
co
n
tr
o
lled
tem
p
o
r
al
o
v
er
s
am
p
lin
g
(
s
y
n
th
etic
m
in
o
r
ity
o
v
er
-
s
am
p
lin
g
tech
n
iq
u
e
(
SMOT
E
)
)
ar
e
em
p
lo
y
ed
t
o
b
alan
ce
th
e
d
ataset,
p
ar
ticu
lar
ly
b
y
au
g
m
e
n
tin
g
o
r
d
u
p
licatin
g
f
r
a
u
d
u
len
t
tim
e
win
d
o
ws.
iii)
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
co
m
p
r
is
es
4
2
,
3
7
2
s
m
ar
t
m
ete
r
s
in
s
talled
at
cu
s
to
m
e
r
p
r
em
i
s
es,
p
r
o
v
id
in
g
d
aily
o
b
s
er
v
atio
n
s
o
v
er
a
1
0
-
m
o
n
th
p
er
io
d
.
iv
)
Fo
r
th
e
cr
o
s
s
-
v
alid
atio
n
s
tr
ate
g
y
,
th
is
s
tu
d
y
ad
o
p
ts
a
tim
e
-
awa
r
e
g
r
o
u
p
ed
k
-
f
o
ld
a
p
p
r
o
ac
h
d
esig
n
ed
to
ev
alu
ate
tem
p
o
r
al
r
o
b
u
s
tn
ess
wh
ile
p
r
eser
v
in
g
g
r
o
u
p
in
teg
r
i
ty
b
y
m
eter
I
D.
T
h
e
t
o
tal
o
b
s
e
r
v
atio
n
p
e
r
io
d
is
d
iv
id
ed
in
to
K
co
n
tig
u
o
u
s
tim
e
b
lo
ck
s
.
Fo
r
ea
ch
f
o
ld
i,
th
e
m
o
d
el
is
tr
ain
ed
o
n
b
lo
c
k
s
1
to
i
an
d
v
alid
ated
o
n
b
lo
ck
i
+
1
.
C
o
n
s
eq
u
en
tly
,
th
e
tr
ain
in
g
s
et
e
x
p
an
d
s
with
ea
ch
iter
atio
n
,
e
m
u
latin
g
th
e
p
r
o
ce
s
s
o
f
r
ea
l
-
wo
r
ld
d
e
p
lo
y
m
en
t
an
d
e
n
ab
lin
g
t
h
e
m
o
d
el
to
p
r
ed
ict
f
u
tu
r
e
c
o
n
s
u
m
p
tio
n
with
o
u
t
d
at
a
leak
ag
e
f
r
o
m
p
r
i
v
ate
cu
s
to
m
er
in
f
o
r
m
atio
n
.
v)
R
eg
ar
d
in
g
h
y
p
e
r
p
ar
am
ete
r
tu
n
in
g
,
th
is
wo
r
k
em
p
lo
y
s
a
B
ay
esian
o
p
tim
izatio
n
s
tr
ateg
y
with
in
a
n
ested
s
ea
r
ch
f
r
am
ewo
r
k
to
m
itig
at
e
o
v
er
-
o
p
tim
is
m
an
d
en
h
an
c
e
g
en
er
aliza
tio
n
.
T
h
e
co
m
p
l
ete
m
o
d
elin
g
p
ip
elin
e
in
clu
d
es
th
r
ee
s
tag
es:
d
ata
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
en
g
in
ee
r
in
g
,
an
d
m
o
d
el
au
g
m
en
tatio
n
.
T
h
e
d
ef
in
ed
h
y
p
er
p
ar
a
m
eter
s
ea
r
c
h
s
p
ac
e
in
clu
d
es
th
e
f
o
llo
win
g
:
lear
n
in
g
r
ate
o
f
0
.
0
0
2
,
b
at
ch
s
ize
o
f
3
2
,
4
h
id
d
en
la
y
er
s
,
6
4
h
i
d
d
en
u
n
its
,
d
r
o
p
o
u
t
r
ate
o
f
0
.
3
,
weig
h
t
d
ec
ay
o
f
0
.
0
2
,
s
eq
u
e
n
ce
len
g
th
o
f
1
2
,
a
n
d
laten
t d
im
en
s
io
n
o
f
1
6
(
with
t
h
e
ADAM
o
p
tim
izer
)
.
T
h
e
B
ay
esian
s
ea
r
ch
iter
ativ
ely
ex
p
lo
r
es
th
e
p
ar
am
eter
s
p
a
ce
wh
ile
ter
m
in
atin
g
u
n
p
r
o
m
i
s
in
g
tr
ial
s
ea
r
ly
.
Fo
r
ea
ch
o
u
ter
iter
atio
n
,
th
e
b
est
-
p
er
f
o
r
m
in
g
s
et
o
f
h
y
p
er
p
a
r
am
eter
s
id
en
tifie
d
in
te
r
n
ally
is
s
elec
ted
to
r
etr
ain
th
e
m
o
d
el
o
n
th
e
en
tir
e
in
p
u
t
d
ataset
.
T
h
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
m
ay
b
e
u
n
d
e
r
esti
m
ated
if
f
r
au
d
u
len
t
b
eh
av
io
r
is
s
tr
o
n
g
ly
c
o
r
r
elate
d
with
m
etad
ata.
Fig
u
r
e
2
.
Pro
p
o
s
ed
m
et
h
o
d
o
lo
g
y
f
r
a
m
ewo
r
k
f
o
r
elec
tr
icity
t
h
ef
t d
etec
tio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
0
9
8
-
1
1
1
1
1104
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
f
o
r
e
lectr
icity
th
ef
t d
etec
tio
n
is
im
p
lem
en
te
d
as f
o
llo
ws:
i)
Step
1
:
d
ata
co
llectio
n
.
I
n
th
is
s
tep
,
th
e
d
ata
ar
e
c
o
llected
ev
er
y
d
ay
f
r
o
m
s
m
ar
t
m
eter
s
in
s
talled
to
co
n
s
u
m
er
s
.
I
n
th
is
wo
r
k
,
4
2
,
3
7
2
co
n
s
u
m
p
tio
n
s
h
a
v
e
b
ee
n
r
e
co
r
d
ed
d
u
r
i
n
g
1
,
0
3
5
d
ay
s
.
ii)
Step
2
:
d
ata
an
aly
s
is
an
d
p
r
ep
r
o
ce
s
s
in
g
.
T
h
e
co
llected
d
ata
f
r
o
m
elec
tr
ical
co
n
s
u
m
er
s
ar
e
an
aly
ze
d
an
d
p
r
ep
r
o
ce
s
s
ed
.
T
h
is
s
tep
allo
ws
to
av
o
id
m
is
s
in
g
d
ata
with
an
in
ter
p
o
latio
n
m
eth
o
d
u
s
in
g
d
ata
clea
n
in
g
an
d
d
ata
n
o
r
m
aliza
tio
n
.
T
h
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
s
tep
im
p
r
o
v
es
th
e
d
ata
ac
cu
r
ac
y
,
w
h
ich
d
ir
ec
tly
e
n
h
an
ce
s
th
e
ef
f
icac
y
o
f
th
e
m
o
d
el.
iii)
Step
3
:
d
ata
tr
ain
in
g
an
d
test
i
n
g
.
I
n
th
is
m
o
d
el,
th
e
tr
ain
in
g
s
tep
allo
ws
to
tr
ain
th
e
d
ataset
wh
ile
th
e
test
in
g
d
ata
ar
e
u
s
ed
to
v
alid
a
te
th
e
m
o
d
el
's
g
en
er
aliza
tio
n
.
T
h
e
o
v
er
all
d
ataset
is
d
iv
id
ed
in
to
tr
ain
in
g
d
ata
o
f
7
0
% a
n
d
test
in
g
d
ata
o
f
3
0
%.
iv
)
Step
4
:
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
s
elec
tio
n
.
I
n
th
is
s
tep
,
th
e
B
i
-
L
STM
allo
ws
f
ea
tu
r
e
ex
tr
ac
tio
n
,
wh
ile
th
e
L
ADAM
alg
o
r
ith
m
is
u
s
ed
to
s
elec
t a
n
d
n
o
r
m
alize
th
e
f
ea
tu
r
es o
b
tain
ed
af
ter
t
h
e
tr
ain
in
g
s
tag
e.
v)
Step
5
:
f
ea
tu
r
e
class
if
icat
io
n
an
d
p
r
ed
ictio
n
.
I
n
th
is
s
tag
e,
th
e
SVM
m
o
d
el
allo
ws
to
class
if
y
th
e
n
o
r
m
alize
d
f
ea
tu
r
es
with
th
e
aim
o
f
d
eter
m
in
in
g
if
th
e
u
s
e
r
is
s
tea
lin
g
en
er
g
y
.
I
n
th
is
wo
r
k
,
th
e
SVM
ex
am
in
es
th
e
d
ataset
an
d
d
iv
id
es
it
in
to
two
g
r
o
u
p
s
:
n
o
r
m
al
co
n
s
u
m
er
s
an
d
th
ef
t
c
o
n
s
u
m
er
s
.
SVM
wo
r
k
s
o
n
th
e
p
h
e
n
o
m
e
n
a
o
f
a
h
y
p
er
p
lan
e
an
d
d
iv
id
es
th
e
ex
tr
ac
ted
f
ea
tu
r
es
o
b
tain
ed
f
r
o
m
th
e
L
ADAM
m
o
d
el
in
to
th
ef
t
an
d
n
o
n
-
t
h
ef
t
u
s
er
s
.
T
h
en
,
th
e
GR
U
i
s
e
m
p
lo
y
ed
f
o
r
th
e
p
r
e
d
ictio
n
o
f
th
e
class
if
ied
f
ea
tu
r
es a
n
d
f
in
ally
id
en
tifie
s
t
h
e
p
o
ten
tial f
r
au
d
s
ter
s
.
v
i)
Step
6
:
v
alid
atio
n
o
f
th
e
m
o
d
el
.
T
h
is
s
tep
co
n
s
is
t
s
o
f
v
er
if
y
in
g
th
e
ef
f
icien
c
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
with
th
e
im
p
lem
en
tatio
n
o
f
p
er
f
o
r
m
a
n
ce
m
etr
ics
s
u
ch
as
MA
E
,
MSE
,
MA
PE,
R
2
,
F1
-
s
co
r
e,
p
r
ec
is
io
n
,
AUC
,
an
d
r
ec
all.
Fo
r
th
e
th
e
f
t
co
n
s
u
m
er
s
,
t
h
e
M
AE
,
MA
PE
,
an
d
R
MSE
s
h
o
u
ld
b
e
lo
w
,
w
h
ile
F1
-
s
co
r
e,
ac
cu
r
ac
y
,
r
ec
all
,
a
n
d
AUC
s
h
o
u
ld
b
e
h
ig
h
est
as
p
o
s
s
ib
le.
Fo
r
t
h
e
n
o
n
-
t
h
ef
t
c
o
n
s
u
m
e
r
s
,
th
e
MA
E
,
MA
PE
,
an
d
R
MSE
s
h
o
u
ld
b
e
h
ig
h
,
wh
ile
F1
-
s
co
r
e,
ac
c
u
r
ac
y
,
r
ec
all
,
an
d
AUC
s
h
o
u
ld
b
e
lo
west
as
p
o
s
s
ib
le.
W
h
en
th
e
s
u
s
p
icio
u
s
co
n
s
u
m
er
s
ar
e
id
en
tifie
d
u
s
in
g
th
e
p
r
o
p
o
s
ed
B
i
-
L
ADAM
-
SVM
-
GR
U
m
o
d
el,
th
e
en
er
g
y
s
u
p
p
lier
co
m
p
an
y
ca
n
th
er
ef
o
r
e
s
en
d
a
p
en
alty
to
th
e
f
r
au
d
u
len
t
co
n
s
u
m
er
to
co
m
p
en
s
ate
f
o
r
th
e
illeg
ally
co
n
s
u
m
ed
e
n
er
g
y
.
T
h
e
n
,
if
t
h
e
co
n
s
u
m
e
r
d
o
esn
’
t
p
ay
t
h
e
p
en
alty
,
th
e
co
m
p
an
y
ca
n
s
en
d
tech
n
ician
s
to
s
h
u
t d
o
w
n
th
e
co
n
s
u
m
er
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Ana
ly
s
is
o
f
cus
t
o
m
er
co
ns
um
ptio
n
T
h
e
co
llectio
n
o
f
d
ata
all
o
ws
an
aly
s
is
o
f
elec
tr
icity
co
n
s
u
m
p
tio
n
b
eh
a
v
io
r
o
f
u
s
er
s
with
d
em
o
g
r
a
p
h
ic,
s
o
cial
,
an
d
f
in
a
n
cial
tr
en
d
s
.
Data
ar
e
c
o
llected
o
v
e
r
1
,
0
3
5
d
ay
s
f
o
r
4
2
,
3
7
2
co
n
s
u
m
er
s
.
I
n
th
i
s
s
tu
d
y
,
co
n
s
u
m
er
s
h
a
v
e
in
s
tall
ed
s
m
ar
t
m
eter
s
at
th
eir
h
o
m
e
s
to
co
llect
r
ea
l
-
tim
e
elec
tr
icity
co
n
s
u
m
p
tio
n
.
I
n
th
is
ca
s
e,
th
is
s
t
udy
w
as
in
ter
ested
in
th
e
co
n
s
u
m
p
tio
n
o
f
c
u
s
to
m
er
3
5
.
T
h
e
an
al
y
s
is
o
f
its
co
n
s
u
m
p
tio
n
was
ca
r
r
ied
o
u
t o
v
er
th
e
th
r
ee
y
ea
r
s
f
o
r
wh
ich
o
u
r
c
o
n
s
u
m
p
tio
n
r
ec
o
r
d
s
wer
e
tak
en
.
4
.
1
.
1
.
F
o
r
t
he
y
ea
r
2
0
1
4
Fig
u
r
e
3
d
em
o
n
s
tr
ates
cu
s
to
m
er
3
5
th
r
o
u
g
h
o
u
t
2
0
1
4
,
w
ith
h
ig
h
en
er
g
y
co
n
s
u
m
p
tio
n
b
etwe
en
J
an
u
ar
y
an
d
Ma
r
ch
,
m
id
-
Ma
y
an
d
Sep
tem
b
er
,
a
n
d
m
i
d
-
No
v
em
b
er
a
n
d
Dec
em
b
er
.
H
o
wev
er
,
it
d
ec
lin
es
b
etwe
en
Ap
r
il
an
d
m
i
d
-
Ma
y
an
d
th
en
b
etwe
en
Sep
tem
b
er
an
d
No
v
em
b
er
.
I
t
ca
n
b
e
o
b
s
er
v
ed
th
at
t
h
e
co
n
s
u
m
p
tio
n
is
d
ep
en
d
in
g
to
th
e
p
er
io
d
s
o
f
th
e
y
ea
r
,
wh
eth
e
r
th
e
u
s
er
s
ar
e
at
h
o
m
e
o
r
at
v
a
ca
n
cy
.
Fig
u
r
e
3
.
E
n
er
g
y
co
n
s
u
m
p
tio
n
b
y
cu
s
to
m
e
r
3
5
i
n
2
0
1
4
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
n
o
ve
l c
o
mb
i
n
ed
d
ee
p
lea
r
n
i
n
g
-
b
a
s
ed
tech
n
iq
u
e
fo
r
elec
tr
icity
th
eft
…
(
E
tien
n
e
F
r
a
n
ç
o
is
Mo
u
ck
o
mey
)
1105
4
.
1
.
2
.
F
o
r
t
he
y
ea
r
2
0
1
5
Fig
u
r
e
4
d
em
o
n
s
tr
ates
th
e
ch
an
g
es
in
c
o
n
s
u
m
p
tio
n
f
o
r
cu
s
to
m
er
3
5
th
r
o
u
g
h
o
u
t
2
0
1
5
,
with
h
ig
h
en
er
g
y
co
n
s
u
m
p
tio
n
b
etwe
e
n
J
an
u
ar
y
an
d
th
e
f
ir
s
t
h
alf
o
f
Ma
r
ch
an
d
th
en
b
etwe
en
J
u
n
e
an
d
Sep
tem
b
er
.
I
t
ca
n
b
e
o
b
s
er
v
ed
a
h
ig
h
v
a
r
ia
b
ilit
y
o
f
co
n
s
u
m
p
tio
n
d
u
r
in
g
th
is
y
ea
r
.
T
h
is
ca
n
b
e
ex
p
lain
ed
with
a
r
eg
u
lar
en
h
an
ce
m
e
n
t o
f
p
eo
p
le'
s
life
s
t
y
les
d
u
r
in
g
t
h
is
y
ea
r
.
Fig
u
r
e
4
.
E
n
er
g
y
co
n
s
u
m
p
tio
n
b
y
cu
s
t
o
m
e
r
3
5
i
n
2
0
1
5
4
.
1
.
3
.
F
o
r
t
he
y
ea
r
2
0
1
6
Vo
ltag
e
r
is
es
ar
e
r
ec
o
r
d
ed
b
e
twee
n
J
an
u
ar
y
a
n
d
Ma
r
c
h
,
J
u
n
e
an
d
Oct
o
b
er
,
an
d
t
h
e
s
ec
o
n
d
h
alf
o
f
Sep
tem
b
er
,
as
illu
s
tr
ated
in
F
i
g
u
r
e
5
.
I
t
ca
n
b
e
o
b
s
er
v
e
d
a
g
r
ea
t
co
n
s
u
m
p
tio
n
o
f
ar
o
u
n
d
1
0
,
000
MW
b
etwe
en
J
u
ly
an
d
Sep
tem
b
e
r
,
a
n
d
a
lo
w
co
n
s
u
m
p
ti
o
n
o
f
ar
o
u
n
d
7
,
0
0
0
MW
b
etwe
en
J
an
u
a
r
y
an
d
Ma
y
.
Mo
r
eo
v
er
,
a
m
in
im
al
co
n
s
u
m
p
tio
n
is
r
ec
o
r
d
ed
at
ar
o
u
n
d
2
,
0
0
0
MW b
etw
ee
n
Octo
b
er
a
n
d
Dec
em
b
e
r
.
Fig
u
r
e
5
.
E
n
er
g
y
co
n
s
u
m
e
d
b
y
cu
s
to
m
er
3
5
in
2
0
1
6
Fo
r
th
e
d
ataset
an
aly
s
is
,
th
is
s
tu
d
y
u
s
e
L
ADAM
m
o
d
el
to
e
x
tr
ac
t
im
p
o
r
tan
t
f
ea
tu
r
es
b
y
e
lim
in
atin
g
n
o
is
y
d
ata.
Fro
m
th
e
o
b
tain
e
d
d
ata,
th
e
SVM
allo
ws
to
c
lass
if
y
th
e
co
n
s
u
m
e
r
s
as
n
o
r
m
al
co
n
s
u
m
e
r
s
o
r
ab
n
o
r
m
al
co
n
s
u
m
er
s
.
T
h
er
ef
o
r
e,
s
u
d
d
en
ch
an
g
es
in
t
h
e
c
o
n
s
u
m
p
tio
n
o
f
s
u
s
p
icio
u
s
c
o
n
s
u
m
er
s
allo
w
th
e
SVM
to
id
en
tify
f
r
a
u
d
th
at
m
a
y
o
cc
u
r
.
Fro
m
th
e
o
b
tain
ed
r
esu
lts
,
th
e
L
ADAM
-
SVM
m
o
d
el
allo
ws
f
o
r
id
en
tify
in
g
ar
o
u
n
d
4
6
2
f
r
a
u
d
u
le
n
t
co
n
s
u
m
er
s
,
wh
ich
r
ep
r
esen
ts
a
r
o
u
n
d
1
.
0
9
%
o
f
th
e
o
v
e
r
all
co
n
s
u
m
e
r
s
.
T
h
e
id
en
tific
atio
n
o
f
th
ese
f
r
au
d
s
ter
s
allo
ws
f
o
r
av
o
id
i
n
g
th
e
lo
s
s
o
f
m
illi
o
n
s
o
f
d
o
llar
s
d
u
r
in
g
th
is
p
er
io
d
f
o
r
th
e
elec
tr
icity
s
u
p
p
lier
co
m
p
an
ies.
4
.
2
.
Co
m
pa
ra
t
iv
e
s
t
ud
y
o
f
o
ptim
iza
t
io
n t
ec
hn
iqu
es
T
h
e
ch
o
ice
o
f
th
e
m
o
d
el
led
u
s
to
ca
r
r
y
o
u
t
a
co
m
p
a
r
ativ
e
s
tu
d
y
with
th
e
s
tate
-
of
-
th
e
-
a
r
t
o
p
tim
izer
s
.
T
h
ese
ar
e
d
if
f
er
en
t
f
o
r
m
s
o
f
g
r
ad
ien
t
d
escen
t
em
p
lo
y
ed
in
m
ac
h
in
e
lear
n
in
g
to
o
p
tim
ize
th
e
co
s
t
f
u
n
ctio
n
b
y
iter
ativ
ely
m
o
d
if
y
i
n
g
th
e
p
a
r
am
eter
s
to
war
d
s
th
e
n
eg
ati
v
e
g
r
ad
ien
t
d
i
r
ec
tio
n
to
d
is
co
v
er
th
e
o
p
tim
al
p
ar
am
eter
s
f
o
r
a
m
o
d
el.
T
h
e
aim
o
f
g
r
ad
ien
t
d
escen
t
is
t
o
f
in
d
t
h
e
s
et
o
f
p
a
r
am
eter
s
th
at
m
in
im
izes
th
is
d
ev
iatio
n
an
d
im
p
r
o
v
es
th
e
m
o
d
el’
s
p
er
f
o
r
m
a
n
ce
.
T
h
e
c
o
s
t
f
u
n
ctio
n
,
in
th
is
ca
s
e
th
e
MSE
,
q
u
an
tifie
s
th
e
d
is
p
ar
ity
b
et
wee
n
p
r
e
d
icted
co
n
s
u
m
p
tio
n
a
n
d
ac
tu
al
c
o
n
s
u
m
p
tio
n
b
y
th
e
m
o
d
el.
T
h
e
co
m
p
a
r
is
o
n
o
f
im
p
lem
en
ted
m
o
d
els is
g
iv
en
i
n
T
ab
le
2
.
T
h
e
m
o
m
en
tu
m
an
d
r
o
o
t m
ea
n
s
q
u
ar
e
p
r
o
p
a
g
atio
n
(
R
MSPr
o
p
)
alg
o
r
ith
m
s
h
av
e
b
ee
n
ass
o
ciate
d
with
th
e
cr
ea
tio
n
o
f
th
e
ADAM
o
p
t
im
izatio
n
alg
o
r
ith
m
.
T
h
e
c
o
s
t
f
u
n
ctio
n
f
o
r
t
h
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
6
.
4
5
e
-
09
.
T
h
e
Nester
o
v
alg
o
r
ith
m
,
a
n
e
x
ten
s
io
n
o
f
th
e
class
ic
g
r
ad
ien
t
d
escen
t
alg
o
r
ith
m
,
co
n
s
id
er
s
p
r
ev
io
u
s
g
r
ad
ien
t
u
p
d
ates.
I
ts
co
s
t f
u
n
ctio
n
f
o
r
t
h
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
3
.
7
2
e
-
05
.
T
h
e
m
o
m
en
tu
m
g
r
a
d
ien
t
d
escen
t
alg
o
r
ith
m
is
an
o
p
t
im
izatio
n
alg
o
r
ith
m
th
at
f
o
llo
ws
th
e
co
n
v
er
g
en
ce
o
f
g
r
ad
ie
n
t
d
esce
n
t
b
y
o
p
tim
izin
g
th
e
u
p
d
atin
g
o
f
th
e
weig
h
ts
.
T
h
e
im
p
u
ls
e
te
r
m
is
b
ased
o
n
th
e
p
r
ev
io
u
s
u
p
d
ate
o
f
th
e
weig
h
t
s
an
d
h
elp
s
th
e
alg
o
r
ith
m
g
ai
n
m
o
m
en
tu
m
as
it
m
o
v
es
d
o
wn
th
e
lo
s
s
f
u
n
ctio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
8
1
4
I
n
t J Ad
v
Ap
p
l Sci
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
2
0
2
6
:
1
0
9
8
-
1
1
1
1
1106
T
h
e
co
s
t
f
u
n
ctio
n
f
o
r
t
h
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
8
.
7
5
e
-
05
.
T
h
e
ad
ap
tiv
e
g
r
a
d
ien
t
(
Ad
aG
r
ad
)
alg
o
r
ith
m
is
a
g
r
ad
ien
t
-
b
ased
o
p
tim
izatio
n
a
lg
o
r
ith
m
.
I
ts
lear
n
in
g
r
ate
is
ad
ap
ted
to
th
e
h
y
p
er
p
ar
am
et
er
s
b
y
k
n
o
wled
g
e
in
co
r
p
o
r
atio
n
o
f
p
ast
o
b
s
er
v
at
io
n
s
.
I
ts
co
s
t
f
u
n
ctio
n
f
o
r
th
i
s
s
p
ec
if
ic
ca
s
e
is
MSE
=
1
.
8
2
e
-
04
.
T
h
e
R
MSPr
o
p
alg
o
r
ith
m
is
a
g
r
a
d
ien
t
-
b
ased
m
eth
o
d
in
wh
ich
its
co
s
t f
u
n
ct
io
n
f
o
r
th
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
5
.
7
9
e
-
04
.
T
h
e
m
in
i
-
b
atch
g
r
ad
ie
n
t
d
esc
en
t
alg
o
r
ith
m
is
a
v
ar
ian
t
o
f
t
h
e
g
r
a
d
ien
t
d
escen
t
alg
o
r
ith
m
f
r
eq
u
en
tly
us
ed
in
tr
ai
n
in
g
d
ee
p
lear
n
in
g
m
o
d
els.
T
h
e
b
asic
p
r
in
cip
le
o
f
th
is
alg
o
r
ith
m
is
to
s
ep
ar
ate
th
e
tr
ain
in
g
d
ata
t
o
p
r
o
ce
s
s
th
em
s
eq
u
en
tially
.
I
ts
co
s
t f
u
n
ctio
n
f
o
r
th
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
8
.
1
6
e
-
04
.
T
h
e
g
r
ad
ien
t d
escen
t w
ith
d
ec
ay
in
g
lear
n
in
g
r
ate
alg
o
r
it
h
m
,
a
m
o
d
er
n
n
eu
r
al
n
etwo
r
k
tr
ain
in
g
tech
n
iq
u
e,
s
tar
ts
b
y
t
r
ain
in
g
th
e
n
etwo
r
k
with
a
h
ig
h
lear
n
in
g
r
ate
a
n
d
th
en
s
lo
wly
r
ed
u
cin
g
it
u
n
til
a
lo
ca
l
m
in
im
u
m
is
o
b
tain
e
d
.
T
h
is
s
h
o
ws
th
at
an
in
itially
h
ig
h
lear
n
i
n
g
r
ate
p
r
e
v
en
ts
th
e
n
etwo
r
k
f
r
o
m
n
o
is
y
d
ata
m
em
o
r
izatio
n
,
w
h
ile
a
d
ec
r
ea
s
in
g
lear
n
i
n
g
r
ate
im
p
r
o
v
es
th
e
lear
n
in
g
o
f
co
m
p
lex
m
o
d
els.
T
h
e
c
o
s
t
f
u
n
ctio
n
f
o
r
th
is
s
p
ec
if
ic
ca
s
e
is
MSE
=
0
.
2
7
.
I
t
ca
n
also
b
e
o
b
s
er
v
ed
t
h
at
T
ab
le
2
s
u
m
m
ar
izes
th
e
co
s
t
f
u
n
ctio
n
o
p
tim
ized
r
esp
ec
tiv
ely
with
ADAM
,
Nester
o
v
ac
ce
ler
a
ted
g
r
ad
ien
t
d
escen
t,
m
o
m
en
tu
m
-
b
ased
g
r
a
d
ien
t
d
e
s
ce
n
t,
Ad
aGr
ad
,
R
MSPr
o
p
,
m
in
i
-
b
atch
g
r
ad
ien
t
d
escen
t,
an
d
g
r
a
d
ien
t d
escen
t
with
d
ec
ay
in
g
lear
n
i
n
g
r
ate.
T
h
is
s
h
o
ws th
at
th
e
L
ADAM
-
S
VM
o
p
tim
izer
,
with
a
co
s
t
f
u
n
ctio
n
(
MSE
)
eq
u
al
to
6
.
4
5
e
-
09
,
o
u
t
p
er
f
o
r
m
s
th
e
o
th
er
s
ix
o
p
tim
izer
s
.
T
h
e
r
ed
u
ctio
n
o
f
MSE
with
th
e
L
ADAM
m
o
d
el
p
r
o
v
es its
o
u
tp
er
f
o
r
m
an
ce
co
m
p
ar
e
d
to
th
o
s
e
in
th
e
liter
atu
r
e.
T
ab
le
2
.
C
o
m
p
a
r
is
o
n
o
f
im
p
le
m
en
ted
m
o
d
els
O
p
t
i
mi
z
e
r
s
Ep
o
c
h
s
Le
a
r
n
i
n
g
r
a
t
e
C
o
s
t
f
u
n
c
t
i
o
n
(
M
S
E)
P
r
o
p
o
se
d
m
o
d
e
l
(
LA
D
A
M
-
S
V
M
)
5
0
0
0
.
0
1
6
.
4
5
e
-
09
N
e
st
e
r
o
v
a
c
c
e
l
e
r
a
t
e
d
g
r
a
d
i
e
n
t
d
e
s
c
e
n
t
5
0
0
0
.
0
1
3
.
7
2
e
-
05
M
o
me
n
t
u
m
-
b
a
se
d
g
r
a
d
i
e
n
t
d
e
s
c
e
n
t
5
0
0
0
.
0
1
8
.
7
5
e
-
05
A
d
a
G
r
a
d
5
0
0
0
.
0
1
1
.
8
4
e
-
04
R
M
S
P
r
o
p
5
0
0
0
.
0
1
5
.
7
9
e
-
04
M
i
n
i
-
b
a
t
c
h
g
r
a
d
i
e
n
t
d
e
s
c
e
n
t
5
0
0
0
.
0
1
8
.
1
6
e
-
04
G
r
a
d
i
e
n
t
d
e
s
c
e
n
t
w
i
t
h
d
e
c
a
y
i
n
g
l
e
a
r
n
i
n
g
r
a
t
e
5
0
0
0
.
0
1
0
.
2
7
4
.
3
.
P
re
dict
ing
cus
t
o
m
er
co
ns
um
ptio
n
T
h
i
s
s
t
u
d
y
s
e
t
o
u
t
t
h
e
c
o
n
s
u
m
p
t
i
o
n
o
f
c
u
s
t
o
m
e
r
3
5
f
r
o
m
J
a
n
u
a
r
y
1
,
2
0
1
6
t
o
J
a
n
u
a
r
y
1
,
2
0
1
7
.
F
i
g
u
r
e
6
s
h
o
ws
th
e
elec
tr
icity
co
n
s
u
m
p
tio
n
m
ea
s
u
r
ed
af
ter
ea
ch
h
o
u
r
.
T
h
is
p
ap
er
ca
n
o
b
s
er
v
e
th
at
th
e
co
n
s
u
m
p
tio
n
is
g
r
o
win
g
ar
o
u
n
d
8
,
0
0
0
KW
d
u
r
in
g
No
v
em
b
er
2
0
1
6
an
d
Dec
em
b
er
2
0
1
6
.
Mo
r
e
o
v
er
,
th
e
co
n
s
u
m
p
tio
n
is
r
ed
u
ce
d
t
o
ar
o
u
n
d
3
,
0
0
0
KW
d
u
r
in
g
Octo
b
er
2
0
1
6
,
wh
ich
is
ca
u
s
ed
b
y
t
h
e
r
ed
u
ctio
n
o
f
liv
in
g
co
s
ts
.
Fig
u
r
e
6
.
Pre
d
ictio
n
o
f
cu
s
to
m
er
3
5
’
s
co
n
s
u
m
p
tio
n
in
2
0
1
6
4
.
4
.
Det
ec
t
ing
ener
g
y
t
hef
t
T
h
e
en
er
g
y
th
e
f
t
d
etec
tio
n
m
o
d
el
p
r
esen
ted
in
th
is
s
tu
d
y
c
o
m
p
r
is
es
m
u
ltip
le
s
tag
es,
with
th
e
in
itial
s
tag
e
b
ein
g
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
.
Su
b
s
eq
u
en
tly
,
th
e
s
eq
u
e
n
tial
d
iag
r
am
is
d
ev
elo
p
ed
,
as
d
etailed
in
th
e
f
o
r
th
co
m
in
g
s
u
b
s
ec
tio
n
.
T
h
e
s
u
b
s
eq
u
en
t
s
tag
e
in
v
o
lv
e
s
co
n
s
tr
u
ctin
g
th
e
en
er
g
y
t
h
ef
t
p
r
ed
ictio
n
m
o
d
el
u
tili
zin
g
th
e
L
STM
ap
p
r
o
ac
h
,
with
its
h
y
p
er
p
ar
a
m
eter
s
f
in
e
-
tu
n
ed
th
r
o
u
g
h
th
e
u
tili
za
tio
n
o
f
th
e
ADAM
alg
o
r
ith
m
.
I
t
is
d
ee
m
ed
s
u
s
p
icio
u
s
an
d
n
ec
ess
itates
m
o
n
it
o
r
in
g
b
y
th
e
en
e
r
g
y
d
is
tr
ib
u
ti
o
n
co
m
p
a
n
y
if
th
e
e
n
er
g
y
c
o
n
s
u
m
p
tio
n
is
less
th
an
1
0
%
o
f
th
e
an
ticip
ated
v
al
u
e
d
u
r
in
g
a
s
p
an
o
f
s
ix
m
o
n
th
s
.
T
h
e
u
tili
za
tio
n
o
f
f
o
r
ec
ast
m
o
d
el
e
m
p
o
wer
e
d
u
s
to
p
o
ten
tially
d
etec
t
d
ay
s
wh
en
th
e
cu
s
to
m
e
r
’
s
u
s
ag
e
was
q
u
esti
o
n
ab
le
an
d
d
ay
s
wh
en
it wa
s
ty
p
ical.
T
h
e
d
ate
s
u
s
p
ec
ted
o
f
en
er
g
y
th
ef
t
is
th
e
o
n
e
th
at
s
en
s
es a
n
in
cr
e
ase
in
co
n
s
u
m
p
tio
n
o
f
m
o
r
e
th
an
1
0
% o
f
its
p
r
ed
ic
ted
co
n
s
u
m
p
tio
n
.
T
h
e
co
n
s
u
m
p
tio
n
g
a
p
(
C
G)
is
d
ef
in
ed
as
(
1
4
)
.
(
%
)
=
100
×
−
(
1
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ad
v
Ap
p
l Sci
I
SS
N:
2252
-
8
8
1
4
A
n
o
ve
l c
o
mb
i
n
ed
d
ee
p
lea
r
n
i
n
g
-
b
a
s
ed
tech
n
iq
u
e
fo
r
elec
tr
icity
th
eft
…
(
E
tien
n
e
F
r
a
n
ç
o
is
Mo
u
ck
o
mey
)
1107
I
f
C
G(
%)
<
1
0
%,
th
e
cu
s
to
m
er
’
s
co
n
s
u
m
p
tio
n
is
co
n
s
id
er
e
d
n
o
r
m
al,
wh
e
r
ea
s
if
C
G(
%)
≥
1
0
%,
th
e
co
n
s
u
m
p
tio
n
is
class
if
ied
as
s
u
s
p
ec
t.
Her
e,
PC
d
en
o
tes
th
e
p
r
ed
icted
c
o
n
s
u
m
p
tio
n
an
d
R
C
th
e
r
ea
l
co
n
s
u
m
p
tio
n
.
Fu
r
t
h
er
m
o
r
e,
if
a
cu
s
to
m
e
r
’
s
co
n
s
u
m
p
tio
n
r
e
m
ain
s
u
n
d
e
r
1
0
%
o
f
th
e
p
r
e
d
icted
v
alu
e
f
o
r
s
ix
co
n
s
ec
u
tiv
e
m
o
n
th
s
,
a
n
in
s
p
e
ctio
n
team
will
b
e
d
is
p
atch
e
d
.
T
ab
le
3
s
u
m
m
ar
izes
th
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
o
b
tain
ed
f
o
r
elec
tr
icity
th
ef
t
d
etec
tio
n
u
s
in
g
th
e
im
p
lem
e
n
ted
d
ee
p
lear
n
in
g
s
tr
ateg
ies.
T
h
e
co
r
r
esp
o
n
d
in
g
co
n
f
u
s
io
n
m
atr
ices o
f
th
e
im
p
l
em
en
ted
d
ee
p
lear
n
in
g
m
o
d
els ar
e
d
ep
icted
i
n
Fig
u
r
e
7
.
T
ab
le
3
.
Per
f
o
r
m
an
ce
m
etr
ics
f
o
r
elec
tr
icity
th
ef
t
d
etec
tio
n
M
o
d
e
l
s
F1
-
sc
o
r
e
R
e
c
a
l
l
A
c
c
u
r
a
c
y
P
r
e
c
i
s
i
o
n
R
O
C
-
AUC
LSTM
0
.
8
3
0
.
8
0
0
.
8
1
0
.
9
1
0
.
8
1
G
R
U
0
.
8
7
0
.
8
3
0
.
8
6
0
.
9
3
0
.
8
3
LSTM
-
S
V
M
0
.
9
0
0
.
9
3
0
.
9
2
0
.
9
5
0
.
8
7
Bi
-
LST
M
0
.
9
1
0
.
9
5
0
.
9
4
0
.
9
5
0
.
9
1
Bi
-
G
R
U
0
.
9
1
0
.
9
6
0
.
9
5
0
.
9
7
0
.
0
1
Bi
-
LA
D
A
M
-
S
V
M
-
G
R
U
0
.
9
2
0
.
9
7
0
.
9
6
0
.
9
8
0
.
9
2
Fig
u
r
e
7
.
C
o
n
f
u
s
io
n
m
atr
ices
o
f
im
p
lem
en
te
d
d
ee
p
lear
n
in
g
m
o
d
els
Evaluation Warning : The document was created with Spire.PDF for Python.