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.
16
,
No
.
5
,
Octo
b
er
20
26
,
p
p
.
2
3
6
5
~
2
3
7
8
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
16
i
5
.
pp
2
3
6
5
-
2
3
7
8
2365
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Tem
po
ra
l deep
r
e
presenta
tion
lea
r
ning
bas
ed remai
ning
useful
Lifecy
le pr
edictio
n of ele
ctric
vehic
le lit
hiu
m
-
io
n bat
terie
s
P
ra
dis
h Va
idy
a
S
.
1
,
K
ev
in De
nzil V
.
1
,
M
o
ni
s
h S
.
1
,
S.
Clo
u
din
1
,
S.
Sa
ra
dh
a
2
1
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
Ea
sw
a
r
i
E
n
g
i
n
e
e
r
i
n
g
C
o
l
l
e
g
e
,
C
h
e
n
n
a
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
S
R
M
I
n
st
i
t
u
t
e
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
C
h
e
n
n
a
i
,
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
J
u
n
1
5
,
2
0
2
6
R
ev
is
ed
Au
g
1
0
,
2
0
2
6
Acc
ep
ted
Au
g
1
9
,
2
0
2
6
In
re
c
e
n
t
y
e
a
rs,
ra
p
id
g
r
o
wth
o
f
re
n
e
wa
b
le
e
n
e
rg
y
sy
ste
m
s
a
n
d
e
lec
tri
c
v
e
h
icle
tec
h
n
o
l
o
g
ies
h
a
s
in
ten
si
fied
th
e
f
o
c
u
s
o
n
li
t
h
iu
m
-
i
o
n
b
a
tt
e
ries
a
s
c
rit
ica
l
e
n
e
rg
y
sto
ra
g
e
c
o
m
p
o
n
e
n
ts.
Co
n
se
q
u
e
n
t
ly
,
e
v
a
lu
a
t
in
g
t
h
e
ir
p
e
rfo
rm
a
n
c
e
a
n
d
li
fe
sp
a
n
h
a
s
b
e
c
o
m
e
a
sig
n
ifi
c
a
n
t
c
o
n
c
e
rn
in
b
o
t
h
sc
ien
ti
fic
re
se
a
rc
h
a
n
d
i
n
d
u
strial
a
p
p
li
c
a
t
io
n
s.
Th
e
a
c
c
u
ra
te
re
m
a
in
in
g
u
se
fu
l
li
fe
(RUL)
p
re
d
icti
o
n
o
f
li
t
h
i
u
m
-
io
n
b
a
tt
e
ries
a
re
m
a
jo
r
imp
o
rtan
c
e
t
o
in
c
re
a
se
e
n
e
rg
y
m
a
n
a
g
e
m
e
n
t
e
ffica
c
y
a
n
d
in
c
re
a
se
th
e
b
a
tt
e
ry
li
fe
ti
m
e
.
Re
c
e
n
tl
y
,
t
h
e
q
u
ick
a
d
v
a
n
c
e
m
e
n
t
o
f
m
a
c
h
in
e
lea
rn
in
g
(M
L)
a
n
d
a
rti
ficia
l
i
n
telli
g
e
n
c
e
(AI)
k
n
o
wle
d
g
e
,
d
a
ta
-
d
riv
e
n
a
p
p
r
o
a
c
h
e
s
fo
r
fo
re
c
a
stin
g
t
h
e
li
th
i
u
m
-
i
o
n
b
a
tt
e
r
y
d
u
ra
ti
o
n
a
re
in
v
o
l
v
e
d
e
x
ten
si
v
e
a
tt
e
n
ti
o
n
.
Nu
m
e
ro
u
s
re
se
a
r
c
h
e
rs
a
re
e
m
p
lo
y
i
n
g
d
e
e
p
lea
rn
in
g
(DL)
a
p
p
r
o
a
c
h
e
s
to
m
a
k
e
se
v
e
ra
l
tec
h
n
iq
u
e
s
fo
r
p
re
d
ictin
g
RUL,
n
a
m
e
ly
re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
r
k
(RNN
)
a
n
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
(CNN
).
Th
is p
a
p
e
r
in
tr
o
d
u
c
e
s
a
tem
p
o
ra
l
d
e
e
p
re
p
r
e
se
n
tatio
n
lea
rn
in
g
b
a
se
d
re
m
a
in
in
g
u
se
fu
l
Li
fe
c
y
le
p
re
d
ictio
n
(TRD
L
-
RULP
)
a
p
p
ro
a
c
h
f
o
r
e
lec
tri
c
v
e
h
icle
li
t
h
iu
m
-
io
n
b
a
tt
e
rie
s.
T
h
e
m
a
in
g
o
a
l
o
f
th
e
TRDL
-
RULP
fra
m
e
wo
rk
is
t
o
su
p
p
o
r
t
i
n
telli
g
e
n
t
b
a
tt
e
ry
m
a
n
a
g
e
m
e
n
t
m
e
th
o
d
s
a
n
d
c
o
n
tri
b
u
te
to
i
n
c
re
a
sin
g
th
e
o
p
e
ra
ti
o
n
a
l
sa
fe
ty
a
n
d
li
fe
c
y
c
le
m
a
n
a
g
e
m
e
n
t
o
f
e
lec
tri
c
v
e
h
icle
e
n
e
rg
y
re
se
rv
e
s
sy
ste
m
s.
In
it
i
a
ll
y
,
ra
w
b
a
tt
e
ry
d
e
g
ra
d
a
ti
o
n
d
a
ta
a
re
p
ro
c
e
ss
e
d
u
sin
g
d
a
ta
n
o
rm
a
li
z
a
ti
o
n
to
e
li
m
in
a
te
sc
a
le
v
a
riatio
n
s
a
n
d
imp
r
o
v
e
m
o
d
e
l
sta
b
il
i
ty
.
To
e
n
h
a
n
c
e
fe
a
tu
re
re
lev
a
n
c
e
a
n
d
re
d
u
c
e
d
ime
n
sio
n
a
l
it
y
,
a
s
n
a
k
e
o
p
ti
m
iza
ti
o
n
-
b
a
se
d
fe
a
tu
re
se
lec
ti
o
n
stra
teg
y
is
e
m
p
l
o
y
e
d
to
e
n
a
b
le
th
e
e
x
trac
ti
o
n
o
f
t
h
e
m
o
st
i
n
fo
rm
a
ti
v
e
d
e
g
ra
d
a
ti
o
n
i
n
d
ica
to
rs.
F
o
r
RU
L
p
re
d
icti
o
n
,
a
l
o
n
g
sh
o
rt
-
term
m
e
m
o
ry
a
u
to
e
n
c
o
d
e
r
h
a
s
b
e
e
n
e
x
p
l
o
it
e
d
t
o
c
a
p
t
u
re
n
o
n
l
in
e
a
r
tem
p
o
ra
l
d
e
p
e
n
d
e
n
c
ies
a
n
d
e
x
trac
t
r
o
b
u
st
late
n
t
re
p
re
se
n
tatio
n
s.
F
i
n
a
ll
y
,
t
h
e
tu
n
a
sw
a
rm
o
p
ti
m
iza
ti
o
n
a
l
g
o
ri
th
m
c
a
n
b
e
a
p
p
li
e
d
f
o
r
o
p
ti
m
a
l
o
p
ti
m
ize
p
a
ra
m
e
ters
to
imp
ro
v
e
c
o
n
v
e
rg
e
n
c
e
sp
e
e
d
a
n
d
p
re
d
ictiv
e
p
e
rfo
rm
a
n
c
e
.
T
h
e
sim
u
latio
n
stu
d
y
o
f
t
h
e
p
r
o
p
o
se
d
TRDL
-
RULP
a
lg
o
ri
th
m
is
c
o
n
d
u
c
ted
u
sin
g
a
b
e
n
c
h
m
a
rk
li
th
iu
m
-
io
n
b
a
tt
e
ry
d
e
g
ra
d
a
ti
o
n
d
a
tas
e
t
o
b
tain
e
d
fr
o
m
t
h
e
Ka
g
g
le
re
p
o
sito
r
y
.
E
x
ten
siv
e
c
o
m
p
a
ra
ti
v
e
re
su
lt
s
d
e
m
o
n
stra
te
th
e
su
p
e
rio
r
p
e
rfo
rm
a
n
c
e
o
f
t
h
e
TRDL
-
RULP
a
p
p
ro
a
c
h
o
v
e
r
re
c
e
n
t
m
e
th
o
d
o
l
o
g
ies
.
K
ey
w
o
r
d
s
:
Au
to
en
co
d
e
r
Dee
p
lear
n
in
g
E
lectr
ic
v
eh
icle
Featu
r
e
s
elec
tio
n
L
ith
iu
m
-
io
n
b
atter
ies
R
em
ain
in
g
u
s
ef
u
l
L
if
ec
y
le
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
:
S.
C
lo
u
d
in
C
o
m
p
u
ter
Scien
ce
an
d
E
n
g
i
n
e
er
in
g
,
E
aswar
i E
n
g
in
ee
r
in
g
C
o
lleg
e
C
h
en
n
ai,
I
n
d
ia
E
m
ail:
p
r
o
f
ess
o
r
clo
u
d
in
@
g
m
a
il.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
I
n
th
e
p
ast
f
ew
y
ea
r
s
,
th
e
p
o
p
u
lar
ity
o
f
elec
tr
ic
v
e
h
icles
(
E
Vs)
h
as
co
n
s
id
er
ab
l
y
in
cr
ea
s
e
d
o
win
g
to
ex
ten
d
ed
cr
u
is
e
r
a
n
g
e
a
n
d
d
e
cr
ea
s
ed
co
s
ts
o
f
o
n
-
b
o
ar
d
lith
iu
m
-
io
n
b
atter
ies.
C
o
n
v
e
r
s
ely
,
th
e
h
ig
h
er
en
er
g
y
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.
16
,
No
.
5
,
Octo
b
e
r
20
26
:
2
3
6
5
-
2
3
7
8
2366
d
en
s
ity
an
d
in
tr
icate
m
an
u
f
a
ctu
r
in
g
p
r
o
ce
d
u
r
es
co
u
ld
g
en
er
ate
f
au
lty
b
atter
y
ce
lls
,
wh
ich
h
av
e
s
h
o
r
t
life
cy
cles
o
r
lead
to
f
ir
e
ac
cid
en
ts
[
1
]
.
Nev
er
th
eless
,
h
ea
lth
y
b
atter
ies
an
d
ca
p
ab
ilit
ies
ar
e
ev
alu
ated
th
r
o
u
g
h
p
h
y
s
ical
an
d
ch
em
ical
test
s
,
d
estru
ctiv
e
ap
p
r
o
ac
h
es
f
r
eq
u
e
n
tly
r
en
d
e
r
th
e
b
atter
y
in
e
f
f
e
ctiv
e
an
d
p
r
o
d
u
ce
s
u
b
s
tan
tial
ec
o
n
o
m
ic
lo
s
s
f
o
r
au
to
m
a
k
er
s
an
d
co
n
s
u
m
e
r
s
[
2
]
.
T
o
m
o
n
ito
r
th
e
d
is
ch
ar
g
in
g
an
d
ch
a
r
g
in
g
p
r
o
ce
d
u
r
es
o
f
E
V
b
atter
ies
an
d
ex
am
in
e
th
e
b
atter
y
p
e
r
f
o
r
m
an
ce
an
d
its
h
ea
lth
h
as
b
ec
o
m
e
th
e
b
est
o
p
tio
n
in
test
in
g
.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
b
atter
y
d
ec
lin
es
o
v
er
tim
e
o
win
g
to
ca
len
d
ar
an
d
cy
cle
ag
ein
g
,
r
esu
ltin
g
i
n
n
u
m
er
o
u
s
d
eg
r
a
d
atio
n
e
v
en
ts
[
3
]
.
Ag
ein
g
am
p
lifie
s
o
p
er
ati
o
n
al
co
s
ts
,
d
im
in
is
h
es
th
e
life
s
p
an
o
f
eq
u
ip
m
en
t,
an
d
im
p
ac
ts
s
af
ety
.
Usu
ally
,
b
atter
y
ca
n
d
elib
er
ate
to
ac
h
iev
e
its
life
cy
cle
o
n
ce
its
ca
p
ab
il
ity
f
alls
to
8
0
%
o
f
its
in
itial
v
alu
e.
T
h
e
r
em
ai
n
in
g
u
s
ef
u
l
life
(
R
UL
)
d
ep
icts
th
e
p
r
ed
icted
o
p
er
atio
n
al
p
er
io
d
f
r
o
m
t
h
e
ex
is
tin
g
s
tate
u
n
til
th
e
b
atter
y
ac
co
m
p
lis
h
es
its
life
cy
cle
[
4
]
.
B
atter
y
lo
n
g
ev
ity
is
im
p
ac
ted
b
y
its
ar
r
an
g
e
m
en
t
an
d
th
e
in
ter
i
o
r
ch
em
ical
v
ar
i
atio
n
s
h
ap
p
en
s
in
c
h
ar
g
e
-
d
is
ch
ar
g
e
cy
cles.
T
h
is
ag
ein
g
p
r
o
ce
d
u
r
e
is
d
if
f
icu
lt
an
d
n
o
n
-
lin
ea
r
,
im
p
ac
ted
b
y
asp
ec
ts
s
u
ch
as
tem
p
er
atu
r
e
,
d
is
ch
ar
g
e
o
r
ch
a
r
g
e
r
ates,
an
d
en
v
ir
o
n
m
e
n
tal
f
ac
to
r
s
[
5
]
.
I
n
in
d
u
s
tr
ies,
ac
cu
r
ate
R
UL
p
r
ed
ictio
n
s
co
u
l
d
less
en
co
s
t
o
f
in
v
estme
n
t
a
n
d
e
n
h
an
ce
p
r
o
f
itab
ilit
y
.
T
h
ey
also
am
p
lif
y
e
n
er
g
y
s
to
r
ag
e
s
y
s
tem
s
,
lo
n
g
ev
ity
,
an
d
co
n
s
is
ten
cy
.
T
h
e
R
UL
p
r
ed
ict
io
n
ap
p
r
o
ac
h
es
ar
e
class
if
ied
in
to
th
r
ee
m
ajo
r
tech
n
iq
u
es:
d
ata
-
d
r
iv
en
,
m
o
d
el
-
b
ased
,
an
d
h
y
b
r
id
m
eth
o
d
s
[
6
]
.
T
h
e
m
o
d
el
-
b
ased
s
tr
ateg
y
ca
lcu
lates
a
m
ath
em
atica
l
d
ep
ictio
n
o
f
in
ter
n
al
b
a
tter
y
an
d
elec
tr
o
ch
e
m
ical
o
u
t
co
m
es,
p
r
o
d
u
cin
g
a
p
r
ed
ictiv
e
tech
n
i
q
u
e
th
at
ca
lc
u
lates
b
atter
y
s
tatu
s
.
Nev
er
th
eless
,
p
r
o
g
r
ess
in
th
ese
tech
n
iq
u
es
is
s
tati
s
tically
in
ten
s
iv
e
an
d
n
ec
ess
itates
co
m
p
r
eh
en
s
iv
e
p
a
r
am
eter
izatio
n
,
f
o
r
m
in
g
p
r
ac
tical
o
p
e
r
atio
n
a
l
ch
allen
g
es
[
7
]
.
On
th
e
o
th
er
h
an
d
,
d
ata
-
d
r
iv
en
m
o
d
els
em
p
lo
y
h
is
to
r
ic
in
f
o
r
m
a
tio
n
,
m
ak
in
g
th
em
m
o
r
e
f
ea
s
i
b
le
f
o
r
lith
iu
m
-
io
n
b
atter
y
a
p
p
licatio
n
s
o
win
g
to
th
e
in
tr
icac
y
o
f
b
atter
y
b
eh
a
v
io
r
.
T
h
e
p
r
e
d
ictio
n
o
f
E
V
b
atter
ies
i
s
v
ital
t
o
en
h
an
ce
b
o
th
s
af
ety
a
n
d
p
er
f
o
r
m
an
ce
.
B
esid
es
co
n
v
en
tio
n
al
s
tatis
tical
an
aly
s
i
s
,
d
ee
p
lear
n
in
g
(
DL
)
h
as
r
ea
ch
ed
n
o
tab
le
p
er
f
o
r
m
a
n
ce
s
in
th
e
s
am
e
task
s
[
8
]
.
T
h
e
p
r
o
s
p
er
ity
o
f
n
eu
r
al
n
etwo
r
k
tech
n
iq
u
es h
as r
ev
ea
led
th
e
en
o
r
m
o
u
s
d
em
an
d
f
o
r
d
ata
q
u
ality
a
n
d
q
u
an
tity
.
I
n
co
n
tr
ast,
th
e
o
b
tain
a
b
ilit
y
o
f
p
u
b
lic
lar
g
e
-
s
ca
le
b
atter
y
d
ata
f
o
r
E
Vs
is
in
ad
eq
u
ate.
T
h
e
cu
r
r
en
t
b
atter
y
d
atasets
u
s
u
ally
h
av
e
co
n
s
tr
ain
ts
lik
e
s
m
al
ler
s
ize,
ab
s
en
ce
o
f
d
iv
er
s
ity
,
o
r
n
ec
ess
itatin
g
s
y
n
th
etic
s
im
u
latio
n
s
th
at
co
n
f
i
n
e
th
e
ap
p
licatio
n
o
f
DL
to
r
ea
l
-
wo
r
ld
b
atter
y
s
y
s
tem
s
[
9
]
.
T
h
is
m
an
u
s
cr
ip
t
in
tr
o
d
u
ce
s
a
t
em
p
o
r
al
d
ee
p
r
ep
r
esen
tatio
n
l
ea
r
n
in
g
b
ased
r
em
ain
in
g
u
s
ef
u
l
L
if
ec
y
le
p
r
ed
ictio
n
(
T
R
DL
-
R
UL
P)
ap
p
r
o
ac
h
f
o
r
E
V
lith
iu
m
-
io
n
b
atte
r
ies
.
Prim
ar
ily
,
r
aw
b
atter
y
d
eg
r
ad
atio
n
d
ata
ca
n
b
e
tr
ea
ted
b
y
u
s
in
g
d
ata
n
o
r
m
aliza
tio
n
to
r
em
o
v
e
s
ca
le
v
ar
i
atio
n
s
an
d
en
h
an
ce
m
o
d
el
s
ta
b
ilit
y
.
Nex
t,
a
s
n
ak
e
o
p
tim
izatio
n
(
SO)
-
b
ased
f
ea
tu
r
e
s
elec
tio
n
(
FS
)
p
r
o
ce
s
s
was
d
ep
lo
y
e
d
to
allo
w
t
h
e
ex
t
r
ac
tio
n
o
f
t
h
e
m
o
s
t
u
s
ef
u
l
d
e
g
r
ad
atio
n
in
d
icato
r
s
.
Fo
r
R
UL
p
r
ed
ictio
n
,
a
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
au
to
e
n
co
d
er
(
L
STM
-
AE
)
h
as
b
ee
n
ex
p
lo
ited
t
o
ca
p
t
u
r
e
n
o
n
-
lin
ea
r
tem
p
o
r
al
n
ee
d
s
an
d
ex
t
r
ac
t
r
o
b
u
s
t
laten
t
r
ep
r
esen
tatio
n
s
.
E
v
en
tu
ally
,
th
e
tu
n
a
s
war
m
o
p
tim
izatio
n
(
T
SO)
alg
o
r
ith
m
was
ex
ec
u
ted
f
o
r
o
p
tim
u
m
p
ar
a
m
eter
tu
n
in
g
to
im
p
r
o
v
e
co
n
v
er
g
en
ce
s
p
ee
d
an
d
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
.
T
h
e
s
im
u
latio
n
an
aly
s
is
o
f
th
e
p
r
o
p
o
s
ed
T
R
DL
-
R
UL
P
alg
o
r
ith
m
is
co
n
d
u
cted
u
tili
zin
g
a
b
en
c
h
m
ar
k
lith
iu
m
-
io
n
b
atter
y
d
eg
r
ad
atio
n
d
ataset
o
b
tain
ed
f
r
o
m
th
e
Kag
g
le
r
ep
o
s
ito
r
y
.
Alth
o
u
g
h
a
g
r
ea
t
d
ea
l
o
f
r
ese
ar
ch
h
as
b
ee
n
d
o
n
e
o
n
th
e
p
r
o
g
n
o
s
tics
o
f
b
atter
ies
u
s
in
g
d
ata,
m
an
y
ch
allen
g
es
r
em
ain
to
m
ak
e
t
h
e
ex
is
tin
g
b
atter
y
R
UL
p
r
e
d
ictio
n
m
eth
o
d
s
p
r
ac
tically
u
s
ef
u
l.
T
r
ad
itio
n
al
m
ac
h
in
e
-
lear
n
i
n
g
a
p
p
r
o
ac
h
es
r
eq
u
i
r
e
ca
r
e
f
u
l
m
an
u
ally
ch
o
s
en
d
eg
r
ad
atio
n
m
ar
k
er
s
an
d
ca
n
lack
a
r
o
b
u
s
t
r
ep
r
esen
tatio
n
o
f
n
o
n
lin
ea
r
tem
p
o
r
al
r
elatio
n
s
h
ip
s
b
etw
ee
n
s
u
cc
ess
iv
e
ch
ar
g
e/d
is
ch
ar
g
e
cy
cles.
Dee
p
r
ec
u
r
r
en
t
m
o
d
els
ar
e
b
etter
at
m
o
d
elin
g
tem
p
o
r
al
b
e
h
av
io
r
,
b
u
t
p
r
e
d
ictio
n
ac
cu
r
ac
y
is
s
en
s
itiv
e
to
th
e
ch
o
ice
o
f
in
p
u
t
f
ea
tu
r
es
an
d
th
e
h
y
p
er
p
ar
am
eter
s
o
f
th
e
m
o
d
els.
Fu
r
th
er
m
o
r
e
,
th
e
m
o
r
e
r
e
d
u
n
d
an
t
th
e
d
eg
r
a
d
atio
n
v
ar
iab
les,
th
e
m
o
r
e
co
m
p
u
t
atio
n
ally
co
m
p
le
x
th
e
m
o
d
el
will
b
e,
with
o
u
t
g
ai
n
in
g
a
co
r
r
esp
o
n
d
in
g
im
p
r
o
v
em
e
n
t
in
p
r
ed
ictiv
e
ab
i
lity
.
I
n
ad
d
itio
n
,
cu
r
r
en
t
h
y
b
r
i
d
m
eth
o
d
s
th
at
ar
e
b
ased
o
n
o
p
tim
izatio
n
u
s
u
ally
u
s
e
a
s
in
g
le
o
p
tim
izatio
n
s
tag
e
f
o
r
f
ea
t
u
r
e
s
elec
tio
n
o
r
f
o
r
m
o
d
el
tu
n
i
n
g
,
an
d
th
e
o
p
ti
m
izatio
n
o
f
f
ea
tu
r
e
s
elec
tio
n
f
o
r
d
eg
r
ad
atio
n
an
d
t
h
e
o
p
tim
izatio
n
o
f
tem
p
o
r
al
r
e
p
r
esen
tatio
n
lear
n
i
n
g
ar
e
r
ec
ei
v
ed
co
m
p
ar
ativ
ely
less
atten
tio
n
.
T
h
e
lim
itatio
n
s
in
s
p
ir
e
th
e
d
esig
n
o
f
a
u
n
if
ie
d
f
r
am
ewo
r
k
to
id
en
tif
y
in
f
o
r
m
ativ
e
in
f
o
r
m
atio
n
o
f
b
atter
y
d
eg
r
ad
atio
n
in
d
icato
r
s
,
to
lear
n
n
o
n
lin
ea
r
tem
p
o
r
al
r
ep
r
esen
tatio
n
an
d
to
s
y
s
tem
atica
lly
o
p
tim
ize
th
e
p
r
ed
ictio
n
m
o
d
el
f
o
r
co
n
tin
u
o
u
s
b
atter
y
R
UL
esti
m
a
tio
n
.
An
ad
v
a
n
ce
d
b
atter
y
m
an
ag
em
en
t
s
y
s
tem
(
B
MS)
is
an
im
p
o
r
tan
t
f
u
n
cti
o
n
to
r
eliab
ly
esti
m
ate
b
atter
y
R
UL
f
r
o
m
a
n
elec
tr
ical
an
d
p
o
wer
e
n
g
in
ee
r
i
n
g
s
tan
d
p
o
in
t.
T
h
e
esti
m
ate
o
f
r
em
ain
in
g
o
p
er
atin
g
life
ca
n
b
e
in
co
r
r
ec
t,
r
esu
ltin
g
in
eith
er
p
r
em
atu
r
e
b
atter
y
r
ep
lace
m
en
t
o
r
u
n
ex
p
ec
ted
d
e
g
r
ad
atio
n
,
a
d
r
o
p
in
v
eh
icle
a
v
ailab
ilit
y
,
o
r
o
p
e
r
atio
n
o
f
th
e
b
atter
y
n
ea
r
u
n
s
af
e
ag
ein
g
c
o
n
d
itio
n
s
.
He
n
ce
,
co
n
tin
u
o
u
s
R
UL
esti
m
atio
n
ca
n
h
elp
with
th
e
p
lan
n
in
g
o
f
p
r
e
v
e
n
tiv
e
m
ain
ten
a
n
ce
,
b
atter
y
r
ep
lace
m
e
n
t,
m
an
a
g
e
m
en
t
o
f
c
h
ar
g
in
g
/d
is
ch
ar
g
in
g
,
g
r
ad
in
g
th
e
s
ev
er
ity
o
f
d
e
g
r
a
d
atio
n
an
d
s
af
e
u
s
e
o
f
av
ailab
le
b
atter
y
ca
p
ac
ity
.
Fo
r
an
elec
tr
ic
v
eh
icle
,
th
is
p
r
o
g
n
o
s
tic
in
f
o
r
m
atio
n
co
u
ld
b
e
u
s
ed
alo
n
g
s
id
e
o
th
er
B
MS
f
u
n
ctio
n
s
,
lik
e
s
tate
o
f
ch
ar
g
e
an
d
s
tate
o
f
h
ea
lth
esti
m
atio
n
,
an
d
ca
n
g
iv
e
a
n
E
V
u
s
er
a
lo
n
g
er
-
ter
m
in
d
icatio
n
o
f
b
atter
y
ag
ein
g
.
T
h
is
s
tu
d
y
h
as
a
n
u
m
b
e
r
o
f
i
m
p
o
r
ta
n
t
r
esu
lts
th
at
ca
n
b
e
h
ig
h
lig
h
ted
as
f
o
llo
ws:
An
SO
-
b
ased
f
ea
tu
r
e
s
elec
tio
n
m
ec
h
an
is
m
is
ad
d
ed
to
s
elec
t
in
f
o
r
m
ativ
e
d
eg
r
a
d
atio
n
in
d
icato
r
s
with
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
Temp
o
r
a
l d
ee
p
r
ep
r
esen
ta
tio
n
lea
r
n
in
g
b
a
s
ed
r
ema
in
in
g
u
s
efu
l Lifecy
le
…
(
P
r
a
d
is
h
V
a
id
ya
S
.
)
2367
th
e
in
ten
tio
n
o
f
r
ed
u
cin
g
th
e
r
ed
u
n
d
an
t
in
p
u
t
in
f
o
r
m
atio
n
.
Seco
n
d
,
an
L
STM
au
to
en
co
d
er
is
u
s
ed
to
lear
n
co
m
p
ac
t
laten
t
r
ep
r
esen
tatio
n
s
f
r
o
m
s
eq
u
en
tial
m
ea
s
u
r
em
e
n
t
o
f
b
atter
y
d
eg
r
ad
atio
n
an
d
to
m
o
d
el
th
e
n
o
n
-
lin
ea
r
tem
p
o
r
al
d
ep
e
n
d
en
c
y
r
elate
d
to
th
e
ag
ein
g
o
f
th
e
b
a
tter
y
.
T
h
ir
d
,
a
n
im
p
r
o
v
ed
m
o
d
el
co
n
f
ig
u
r
atio
n
is
o
b
tain
ed
th
r
o
u
g
h
th
e
in
teg
r
ati
o
n
o
f
a
p
a
r
am
eter
-
tu
n
in
g
s
tr
ateg
y
b
ased
o
n
a
T
SO,
with
th
e
tem
p
o
r
al
p
r
ed
ictio
n
m
o
d
el,
ac
co
r
d
in
g
to
a
n
o
b
ject
iv
e
f
u
n
ctio
n
b
ased
o
n
a
r
eg
r
es
s
io
n
ap
p
r
o
ac
h
.
L
astl
y
,
t
h
e
o
v
e
r
al
l
p
e
r
f
o
r
m
a
n
c
e
o
f
R
UL
p
r
e
d
ic
ti
o
n
is
q
u
an
tit
ati
v
ely
ass
ess
e
d
b
ase
d
o
n
s
ta
n
d
a
r
d
c
o
n
ti
n
u
o
u
s
r
e
g
r
ess
i
o
n
m
et
r
ics
s
u
c
h
as
m
e
a
n
s
q
u
a
r
e
d
e
r
r
o
r
(
MS
E
)
,
r
o
o
t
m
ea
n
s
q
u
a
r
e
d
e
r
r
o
r
(
R
MS
E
)
,
m
e
an
ab
s
o
l
u
t
e
e
r
r
o
r
(
M
AE
)
,
m
ea
n
a
b
s
o
lu
te
p
er
ce
n
t
ag
e
er
r
o
r
(
MA
P
E
)
,
a
n
d
t
h
e
c
o
e
f
f
i
ci
en
t
o
f
d
et
e
r
m
i
n
ati
o
n
(
R
2
)
w
it
h
th
e
c
o
m
p
let
e
T
R
D
L
-
R
U
L
P
f
r
a
m
ew
o
r
k
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
T
h
is
s
ec
tio
n
p
r
esen
ts
a
co
m
p
r
eh
en
s
iv
e
an
d
cr
itical
liter
atu
r
e
r
ev
iew
o
n
E
Vs
b
atter
y
life
cy
cle
p
r
ed
ictio
n
m
eth
o
d
s
.
Sh
i
et
a
l
.
[
1
0
]
p
r
esen
t
a
n
en
c
o
d
er
–
d
e
co
d
er
f
r
am
ewo
r
k
t
h
at
co
m
b
in
es
b
in
ar
y
atten
tio
n
m
o
d
u
les
with
b
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
it
(
Bi
-
GR
U
)
.
Du
r
in
g
th
e
e
n
co
d
in
g
p
r
o
ce
s
s
,
an
atten
tio
n
-
au
g
m
en
ted
B
i
-
GR
U
m
o
d
u
le
h
ig
h
lig
h
ts
v
ital
a
g
in
g
f
ac
to
r
s
.
An
atten
tio
n
m
ec
h
a
n
is
m
s
tr
en
u
o
u
s
ly
p
o
lis
h
es
th
e
o
r
d
er
o
f
i
n
p
u
t
b
y
ass
u
r
in
g
v
i
tal
d
ata
d
u
r
i
n
g
e
n
co
d
in
g
.
I
n
d
ec
o
d
in
g
p
r
o
ce
s
s
,
B
i
-
GR
U
n
etwo
r
k
clar
if
ies
an
ac
cu
r
ate
ca
p
ab
ilit
y
d
eg
r
a
d
atio
n
tr
ajec
to
r
y
.
Do
n
g
et
a
l.
[
1
1
]
p
r
o
ject
an
in
n
o
v
ativ
e
d
ee
p
n
e
u
r
al
n
etwo
r
k
(
DNN)
f
r
am
ewo
r
k
,
L
B
-
Def
o
r
m
Net,
wh
ich
in
co
r
p
o
r
ates
m
u
lti
-
h
ea
d
d
ef
o
r
m
ab
le
atten
tio
n
(
MH
DA)
with
a
lo
ad
-
b
alan
cin
g
r
estrictio
n
m
ix
tu
r
e
o
f
ex
p
er
ts
(
L
B
R
MO
E
)
m
o
d
u
le.
T
h
e
MH
DA
s
tr
o
n
g
ly
m
o
d
if
ies
its
r
ec
ep
tiv
e
r
eg
io
n
b
ased
o
n
d
eg
r
ad
atio
n
,
allo
win
g
th
at
m
eth
o
d
to
co
n
ce
n
tr
ate
v
ital
tem
p
o
r
al
ar
ea
s
th
at
co
n
s
id
er
th
e
ev
o
lu
tio
n
o
f
b
atter
y
h
ea
lth
.
C
o
n
cu
r
r
en
tly
,
th
e
L
B
R
MO
E
m
o
d
u
le
p
r
esen
ts
a
lo
a
d
-
b
al
an
cin
g
r
estra
in
t
b
y
ass
u
r
in
g
co
n
s
is
ten
t
ex
p
er
t
ac
tiv
atio
n
,
im
p
r
o
v
in
g
lear
n
in
g
ass
o
r
tm
en
ts
,
an
d
a
v
o
id
in
g
ex
p
er
t
co
llap
s
e.
Hu
et
a
l.
[
1
2
]
s
u
g
g
est
a
lig
h
tweig
h
t
d
y
n
am
ic
m
u
lti
-
teac
h
er
k
n
o
wle
d
g
e
d
is
till
atio
n
(
L
D
-
MT
KD)
m
eth
o
d
o
lo
g
y
with
s
o
m
e
m
ajo
r
n
o
v
elties.
Prim
ar
ily
,
a
d
y
n
am
ic
m
u
lti
-
teac
h
e
r
ch
o
ice
is
em
p
lo
y
ed
,
em
p
lo
y
in
g
d
ee
p
r
ein
f
o
r
ce
m
e
n
t
lear
n
in
g
(
DR
L
)
to
en
h
a
n
ce
m
o
d
el
s
elec
tio
n
d
ep
e
n
d
in
g
o
n
f
ea
tu
r
es
an
d
m
o
d
el
o
u
tco
m
es.
Af
ter
war
d
s
,
a
n
in
n
o
v
ativ
e
DAF
-
Mo
E
f
r
am
ew
o
r
k
in
teg
r
ates
a
m
ix
tu
r
e
o
f
ex
p
er
ts
(
Mo
E
)
an
d
d
y
n
am
ic
atten
tio
n
f
u
s
io
n
(
DAF)
f
o
r
im
p
r
o
v
in
g
f
lex
ib
ilit
y
an
d
d
is
s
im
ilar
ity
.
E
v
en
tu
ally
,
r
ec
o
m
m
en
d
an
i
n
ter
v
al
p
r
ed
ictio
n
s
y
s
tem
d
ep
e
n
d
o
n
q
u
an
tile r
eg
r
ess
io
n
,
o
f
f
e
r
in
g
a
n
ex
ten
s
iv
e
ev
alu
atio
n
o
f
b
atte
r
y
h
ea
lth
.
Mc
h
ar
a
an
d
R
aiss
i
[
1
3
]
r
ec
o
m
m
en
d
a
g
r
ea
ter
-
p
er
f
o
r
m
a
n
ce
with
a
h
y
b
r
id
p
r
o
g
n
o
s
tic
f
r
am
ewo
r
k
th
at
co
n
cu
r
r
en
tly
in
co
r
p
o
r
ates
th
e
m
u
ltip
le
-
r
eso
lu
tio
n
r
e
p
r
esen
ta
tio
n
s
th
r
o
u
g
h
d
is
cr
ete
wav
elet
tr
an
s
f
o
r
m
(
DW
T
)
,
th
en
th
e
lo
n
g
er
-
r
an
g
e
tem
p
o
r
al
m
o
d
el
o
v
er
a
tr
an
s
f
o
r
m
er
n
etwo
r
k
with
MH
SA,
an
d
n
o
n
-
lin
ea
r
r
esid
u
al
co
r
r
ec
tio
n
u
tili
zin
g
XGBo
o
s
t
.
T
o
en
s
u
r
e
o
p
tim
u
m
p
ar
am
eter
co
n
f
ig
u
r
atio
n
a
n
d
s
tr
o
n
g
co
n
v
er
g
e
n
ce
,
th
e
o
v
er
all
f
r
am
ew
o
r
k
ca
n
en
h
an
ce
u
tili
zin
g
th
e
ch
a
o
tic
b
illi
ar
d
s
o
p
tim
izatio
n
(
C
B
O)
,
s
u
p
p
lem
en
ted
b
y
lo
ca
l
r
ef
in
em
en
t
with
Ad
am
o
p
tim
izer
.
He
et
a
l.
[
1
4
]
in
t
r
o
d
u
ce
a
n
en
h
an
ce
d
GR
U
-
Ko
lm
o
g
o
r
o
v
-
Ar
n
o
ld
n
etwo
r
k
-
g
en
er
alize
d
C
au
ch
y
(
GR
U
-
KAN
-
GC
)
ap
p
r
o
ac
h
f
o
r
th
e
r
em
ain
in
g
life
s
p
an
o
f
lith
i
u
m
-
io
n
b
atter
ies.
I
t
ef
f
ec
tiv
ely
ac
q
u
ir
es
th
e
tim
e
s
eq
u
en
ce
s
asp
ec
ts
f
r
o
m
th
e
d
eter
io
r
atio
n
o
f
lith
iu
m
-
io
n
b
a
tter
y
b
y
p
r
esen
tin
g
GR
U
s
y
s
tem
,
wh
er
ea
s
en
h
a
n
cin
g
th
e
p
r
o
f
icien
cy
o
f
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
b
y
in
teg
r
atin
g
th
e
KAN
to
u
p
g
r
a
d
e
th
e
ex
ec
u
tio
n
.
Pra
th
e
eb
a
an
d
Su
k
u
m
ar
[
1
5
]
p
r
o
ject
ed
an
in
n
o
v
ativ
e
h
y
b
r
id
m
eth
o
d
o
lo
g
y
,
in
teg
r
atin
g
h
ier
ar
ch
ical
g
ated
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
(
R
NN)
an
d
s
an
d
ca
t
s
war
m
o
p
tim
izer
s
.
T
h
e
m
ain
o
b
jectiv
e
is
to
en
h
an
ce
th
e
R
UL
p
r
e
d
ictio
n
,
ch
ar
g
in
g
s
tate,
an
d
s
tate
o
f
h
ea
lth
f
o
r
lith
iu
m
-
io
n
b
a
tter
y
in
E
Vs.
T
h
e
h
ier
ar
ch
ical
g
ated
R
NN
h
as b
ee
n
em
p
lo
y
ed
f
o
r
r
em
ai
n
in
g
lif
esp
an
p
r
ed
ictio
n
.
3.
T
H
E
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
I
n
th
is
p
ap
er
,
we
h
av
e
in
tr
o
d
u
ce
d
a
n
o
v
el
ca
lled
T
R
DL
-
R
UL
P
m
eth
o
d
o
lo
g
y
f
o
r
E
V
lith
iu
m
-
io
n
b
atter
ies
.
T
h
e
m
ajo
r
o
b
jectiv
e
o
f
th
e
p
r
esen
ted
T
R
DL
-
R
UL
P
is
to
m
ain
tain
in
tellig
en
t
b
a
tter
y
m
an
ag
e
m
en
t
m
eth
o
d
s
an
d
ass
is
t
in
im
p
r
o
v
in
g
th
e
s
af
ety
o
f
o
p
er
atio
n
an
d
life
cy
cle
s
tr
ateg
y
o
f
E
V
en
er
g
y
s
to
r
ag
e
s
y
s
tem
s
.
Fig
u
r
e
1
d
ep
icts
th
e
en
tire
p
r
o
ce
s
s
f
lo
w
o
f
th
e
T
R
DL
-
R
UL
P
ap
p
r
o
ac
h
.
As
s
h
o
wn
in
Fig
u
r
e
1
,
th
e
f
r
am
ewo
r
k
co
n
tain
s
f
o
u
r
s
tag
es:
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
,
SO
-
b
ased
f
e
atu
r
e
r
e
d
u
ctio
n
s
tr
ateg
y
is
u
tili
ze
d
to
r
e
d
u
ce
d
im
en
s
io
n
ality
,
R
UL
p
r
ed
icti
o
n
u
s
in
g
a
h
y
b
r
id
L
STM
with
AE
n
etwo
r
k
,
an
d
T
SO
alg
o
r
ith
m
-
b
ased
f
in
e
-
tu
n
ed
o
p
tim
izer
p
r
o
ce
s
s
.
3
.
1
.
Da
t
a
no
r
m
a
liza
t
i
o
n
I
n
th
e
b
eg
in
n
i
n
g
s
tag
e,
r
aw
b
atter
y
d
eg
r
ad
atio
n
d
ata
ar
e
p
r
o
ce
s
s
ed
u
tili
zin
g
d
ata
n
o
r
m
aliza
tio
n
to
er
ad
icate
s
ca
le
d
if
f
er
en
ce
s
a
n
d
en
h
a
n
ce
m
o
d
el
co
n
s
is
ten
cy
.
T
h
e
d
ata
p
r
e‐
p
r
o
ce
s
s
in
g
i
s
ca
r
r
ied
o
u
t
u
s
in
g
‐
s
co
r
e
n
o
r
m
aliza
tio
n
to
en
h
a
n
ce
th
e
p
er
f
o
r
m
a
n
ce
o
f
m
o
d
el
[
1
6
]
.
T
h
is
ef
f
ec
tiv
ely
less
en
s
th
e
im
p
ac
ts
o
f
d
if
f
er
in
g
s
ca
les
an
d
h
i
n
d
er
s
t
h
e
f
ea
tu
r
es
with
g
r
ea
ter
v
al
u
es
b
y
d
o
m
in
atin
g
th
e
p
r
o
ce
s
s
o
f
lear
n
in
g
.
T
h
e
p
er
f
o
r
m
an
ce
an
d
co
n
v
er
g
e
n
ce
ar
e
also
en
h
a
n
ce
d
b
y
th
e
m
o
d
el,
a
n
d
it
ca
n
b
e
m
in
im
al
e
x
p
o
s
ed
t
o
o
u
tlier
s
.
C
ateg
o
r
ical
asp
ec
ts
ar
e
e
n
co
d
er
b
y
em
p
lo
y
in
g
a
o
n
e‐
h
o
t
en
c
o
d
er
o
r
lab
el
e
n
co
d
e
r
d
ep
en
d
o
n
th
e
r
e
q
u
ir
em
e
n
ts
o
f
th
e
m
o
d
el
a
n
d
co
n
s
eq
u
e
n
t
ly
tr
an
s
f
o
r
m
th
em
in
to
n
u
m
b
er
n
o
tatio
n
s
u
ited
f
o
r
m
o
d
el.
Ass
u
r
in
g
th
at
b
o
t
h
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.
16
,
No
.
5
,
Octo
b
e
r
20
26
:
2
3
6
5
-
2
3
7
8
2368
n
u
m
er
ic
an
d
ca
teg
o
r
ical
d
ata
ca
n
ef
f
icien
tly
b
e
em
p
l
o
y
ed
t
h
r
o
u
g
h
Z
‐
s
co
r
e
n
o
r
m
aliza
tio
n
b
y
co
n
v
er
tin
g
d
ata
in
to
u
n
its
with
0
m
ea
n
an
d
u
n
i
t
v
ar
ian
ce
.
I
t c
h
an
g
es
ea
ch
d
at
a
p
o
in
t.
,
in
d
ata
in
t
o
its
n
o
r
m
a
lized
v
alu
e
,
′
d
ep
en
d
o
n
(
1
)
.
,
′
=
,
−
(
1
)
Her
e,
an
d
r
ef
er
s
th
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
(
SD)
,
r
esp
ec
tiv
ely
.
I
t
lo
wer
s
th
e
ef
f
ec
ts
o
f
o
u
tlier
s
.
Alth
o
u
g
h
th
e
m
ea
n
an
d
S
D
v
alu
es
ar
e
ch
an
g
ed
b
y
tim
e,
th
eir
ef
f
ec
tiv
en
ess
to
m
ain
tain
co
n
s
tan
t
n
o
r
m
aliza
tio
n
o
v
er
d
ata
with
r
elate
d
u
n
its
d
im
in
is
h
es.
Fig
u
r
e
1
.
Ov
e
r
all
f
lo
w
o
f
T
R
DL
-
R
UL
P a
p
p
r
o
ac
h
3
.
2
.
F
e
a
t
ure
re
du
ct
io
n pro
ce
s
s
I
n
th
is
s
ec
tio
n
,
an
SO
-
b
ased
FS
ap
p
r
o
ac
h
was
u
tili
ze
d
to
im
p
r
o
v
e
f
ea
tu
r
e
r
elev
a
n
ce
a
n
d
r
ed
u
ce
d
im
en
s
io
n
ality
.
SO
is
a
m
o
d
el
th
at
d
ep
en
d
s
u
p
o
n
t
h
e
o
p
tim
u
m
s
im
u
latio
n
o
f
a
s
n
ak
e
[
1
7
]
.
T
h
e
p
o
p
u
latio
n
o
f
s
n
ak
es
h
as
b
ee
n
d
i
v
id
ed
in
t
o
5
0
p
e
r
ce
n
t
o
f
m
ales
an
d
5
0
p
er
ce
n
t
o
f
f
em
ales
th
at
p
r
ete
n
d
am
o
u
n
t
o
f
f
o
o
d
,
tem
p
er
atu
r
e,
an
d
m
o
v
e
to
wa
r
d
f
o
o
d
,
s
ea
r
ch
f
o
r
f
o
o
d
,
m
ati
n
g
,
f
ig
h
t,
an
d
eg
g
‐
lay
in
g
s
o
ci
o
-
s
ex
u
al
h
ab
its
.
SO
p
ar
titi
o
n
in
g
th
e
iter
atio
n
s
i
n
to
lo
ca
l
e
x
p
lo
itatio
n
a
n
d
g
lo
b
al
ex
p
lo
r
atio
n
th
at
ef
f
ec
tiv
el
y
b
alan
ce
s
th
e
p
ath
r
an
g
es.
T
h
e
ca
lc
u
latio
n
m
o
d
el
f
o
r
d
i
v
er
s
e
tech
n
iq
u
es is
s
p
ec
if
ied
.
{
=
e
xp
(
−
)
=
1
(
−
)
(
2
)
Her
e
r
ep
r
esen
ts
th
e
m
ax
im
al
iter
atio
n
n
u
m
b
er
,
1
=
0
.
5
,
in
d
icate
s
th
e
ex
is
tin
g
iter
atio
n
n
u
m
b
er
,
r
ep
r
esen
ts
tem
p
er
atu
r
e,
an
d
s
ig
n
if
ies
th
e
q
u
an
tity
o
f
f
o
o
d
.
W
h
ile
<
0
.
25
,
th
ey
en
ter
th
e
f
o
o
d
‐
s
e
ar
ch
in
g
way
.
{
=
(
−
,
,
)
,
(
+
1
)
=
,
(
)
±
2
×
×
(
(
m
ax
−
m
i
n
)
×
+
m
i
n
)
(
3
)
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
Temp
o
r
a
l d
ee
p
r
ep
r
esen
ta
tio
n
lea
r
n
in
g
b
a
s
ed
r
ema
in
in
g
u
s
efu
l Lifecy
le
…
(
P
r
a
d
is
h
V
a
id
ya
S
.
)
2369
{
=
(
−
,
,
)
,
(
+
1
)
=
,
(
)
±
2
×
×
(
(
m
ax
−
m
i
n
)
×
+
m
i
n
)
(
4
)
Her
e
,
,
,
,
in
d
icate
s
ar
b
itra
r
ily
g
e
n
er
ated
p
o
s
itio
n
s
o
f
f
em
ale
an
d
m
ale
s
n
ak
es,
,
r
ef
er
s
to
th
e
ca
p
ab
ilit
y
o
f
f
em
ale
an
d
m
ale
s
n
ak
es
to
f
in
d
f
o
o
d
;
,
,
,
s
ig
n
if
ies
th
e
lo
ca
tio
n
o
f
ℎ
m
ale
an
d
f
em
ale
s
n
ak
es;
m
ax
,
m
i
n
r
ef
er
s
to
th
e
u
p
p
e
r
a
n
d
lo
wer
b
o
u
n
d
s
,
an
d
s
ig
n
if
ies
an
ar
b
itra
r
y
v
alu
e
f
r
o
m
ze
r
o
to
o
n
e.
2
=
0
.
05
.
W
h
ile
≥
0
.
25
an
d
>
0
.
6
,
th
ey
m
o
v
e
to
war
d
s
th
e
n
u
tr
itio
n
p
h
ase.
{
,
(
+
1
)
=
±
3
×
×
×
(
−
,
(
)
)
,
(
+
1
)
=
±
3
×
×
×
(
−
,
(
)
)
(
5
)
No
w
r
ep
r
esen
ts
th
e
lo
ca
tio
n
o
f
th
e
o
p
tim
u
m
in
d
i
v
id
u
al,
3
=
2
,
an
d
an
ar
b
itra
r
y
n
u
m
b
e
r
f
r
o
m
ze
r
o
to
o
n
e
h
as
b
ee
n
cr
ea
te
d
b
y
.
I
n
co
n
tr
ast,
a
lik
elih
o
o
d
th
r
esh
o
l
d
o
f
0
.
6
p
er
m
its
th
e
ch
o
ice
o
f
m
atin
g
o
r
f
ig
h
tin
g
m
o
d
es.
W
h
ile
<
0
.
6
,
th
ey
e
n
ter
co
m
b
at.
{
=
(
−
,
)
,
(
+
1
)
=
,
(
)
±
3
×
×
×
(
×
,
−
,
(
)
)
(
6
)
{
=
(
−
,
)
,
(
+
1
)
=
,
(
)
±
3
×
×
×
(
×
,
−
,
(
)
)
(
7
)
Her
e
,
,
,
r
ef
er
s
to
f
itn
ess
v
alu
e
o
f
o
p
tim
u
m
in
d
iv
id
u
als
b
et
wee
n
m
ale
a
n
d
f
em
ale
s
n
a
k
es;
s
ig
n
if
ies
f
itn
ess
v
alu
e
o
f
in
d
i
v
id
u
al
an
d
,
r
ef
er
s
to
th
e
f
ig
h
tin
g
ca
p
a
b
ilit
y
o
f
f
em
ale
an
d
m
ale
s
n
ak
es;
W
h
ile
≥
0
.
6
,
th
e
s
n
ak
e
m
o
d
el
e
n
ter
s
th
e
m
atin
g
.
{
=
(
−
,
,
)
,
(
+
1
)
=
,
(
)
±
3
×
×
×
(
×
,
−
,
(
)
)
(
8
)
{
=
(
−
,
,
)
,
(
+
1
)
=
,
(
)
±
3
×
×
×
(
×
,
−
,
(
)
)
(
9
)
No
w
,
,
,
r
ef
er
s
to
v
alu
e
o
f
f
itn
ess
o
f
m
ale
an
d
f
em
ale
s
n
ak
es a
n
d
,
r
ep
r
esen
t m
atin
g
ca
p
a
b
ilit
y
o
f
f
em
ale
an
d
m
ale
s
n
ak
es.
W
h
ile
th
e
s
n
ak
e
m
o
d
el
g
o
es in
t
o
th
e
m
o
d
e
o
f
s
p
awn
in
g
{
,
=
m
i
n
+
×
(
m
ax
−
m
i
n
)
,
=
m
i
n
+
×
(
m
ax
−
m
i
n
)
(
1
0
)
Her
e
,
,
,
in
d
icate
th
e
p
o
o
r
in
d
i
v
id
u
als
with
a
p
o
p
u
latio
n
o
f
f
em
ale
an
d
m
ale
s
n
ak
es.
T
h
e
f
itn
ess
f
u
n
ctio
n
(
FF
)
em
p
lo
y
ed
in
th
e
p
r
esen
ted
m
eth
o
d
o
l
o
g
y
is
in
ten
d
e
d
to
h
a
v
e
a
b
a
lan
ce
am
o
n
g
s
t
th
e
ch
o
s
en
f
ea
tu
r
e
co
u
n
ts
in
all
s
o
lu
tio
n
s
(
m
in
im
u
m
)
an
d
th
e
cla
s
s
if
icatio
n
p
r
ec
is
io
n
(
m
ax
im
u
m
)
.
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.
16
,
No
.
5
,
Octo
b
e
r
20
26
:
2
3
6
5
-
2
3
7
8
2370
=
(
)
+
|
|
|
|
(
1
1
)
No
w
(
)
d
ep
icts
th
e
class
if
icatio
n
r
ate
o
f
er
r
o
r
o
f
a
s
tated
cla
s
s
if
ier
.
|
|
s
ig
n
if
ies
th
e
ca
r
d
in
ality
o
f
t
h
e
ch
o
s
en
s
u
b
s
et,
an
d
|
|
in
d
icate
s
th
e
o
v
er
all
n
u
m
b
e
r
o
f
f
ea
tu
r
es
in
th
e
d
atab
ase,
an
d
r
ef
er
to
2
p
ar
am
eter
s
.
3
.
3
.
H
y
brid L
ST
M
-
A
E
p
re
d
ict
io
n
Fo
r
R
UL
p
r
ed
ictio
n
,
a
n
L
ST
M
-
AE
h
as
b
ee
n
e
x
p
lo
ited
t
o
c
ap
tu
r
e
n
o
n
-
lin
ea
r
tem
p
o
r
al
d
e
p
en
d
en
cies
an
d
ex
tr
ac
t
r
o
b
u
s
t
laten
t
r
ep
r
esen
tatio
n
s
.
T
h
e
L
STM
‐
AE
m
o
d
el
ca
n
u
tili
ze
d
f
o
r
th
e
cl
ass
if
icatio
n
m
eth
o
d
[
1
6
]
.
T
h
is
tech
n
iq
u
e
ef
f
icien
tly
g
ain
s
co
n
s
ec
u
tiv
e
an
d
tem
p
o
r
al
d
ep
e
n
d
en
c
y
.
I
n
co
n
tr
ast
with
tr
ad
itio
n
al
class
if
ier
s
,
lo
n
g
‐
r
a
n
g
e
p
atter
n
s
an
d
laten
t
co
r
r
elatio
n
h
a
v
e
b
ee
n
s
u
cc
ess
f
u
lly
d
em
o
n
s
tr
ate
d
in
th
is
m
eth
o
d
to
in
cr
ea
s
e
th
e
f
o
r
ec
ast
ac
cu
r
ac
y
f
o
r
tim
e‐
d
e
p
en
d
e
n
t
o
r
s
eq
u
en
tial
d
ata.
T
h
e
f
latten
ed
r
ep
r
esen
tatio
n
o
f
in
p
u
t
d
ata
ca
n
b
e
ap
p
r
o
p
r
iately
ac
q
u
ir
ed
t
h
r
o
u
g
h
th
e
AE
m
o
d
el,
r
ed
u
cin
g
n
o
is
e
an
d
h
ig
h
lig
h
tin
g
v
ital
class
if
icatio
n
f
o
r
m
s
.
I
t
in
cr
ea
s
es
r
o
b
u
s
tn
ess
an
d
d
ea
ls
with
co
m
p
lex
n
o
n
‐
li
n
ea
r
co
r
r
elatio
n
s
,
an
d
less
en
s
o
v
er
f
itti
n
g
th
r
o
u
g
h
lear
n
in
g
a
u
s
ef
u
l
laten
t
s
p
ac
e
.
T
h
e
a
r
ch
itectu
r
e
is
d
ee
m
ed
m
o
r
e
s
u
itab
le
with
g
r
ea
ter
a
cc
u
r
ac
y
a
n
d
b
etter
g
en
er
aliza
tio
n
wh
en
co
m
p
ar
e
d
to
s
tan
d
ar
d
class
if
icatio
n
tec
h
n
iq
u
es.
An
L
STM
‐
AE
is
a
s
p
ec
ialized
tech
n
iq
u
e
p
r
ec
is
ely
d
ev
elo
p
e
d
f
o
r
c
o
n
s
ec
u
tiv
e
d
ata
p
r
o
ce
s
s
in
g
.
T
h
is
ca
n
b
e
u
tili
ze
d
as
an
en
co
d
er
‐
d
ec
o
d
er
m
o
d
el
co
m
p
o
s
ed
o
f
L
STM
.
T
h
is
f
r
a
m
ewo
r
k
c
o
m
p
r
ess
es
th
e
in
p
u
t
o
r
d
er
s
in
t
o
a
v
ec
t
o
r
o
f
r
e
g
u
l
ar
s
ize,
in
ce
s
s
an
tly
em
p
lo
y
ed
f
o
r
r
ec
r
ea
tin
g
th
e
n
ew
d
ata,
th
e
r
ef
o
r
e
en
s
u
r
i
n
g
t
h
e
k
ee
p
in
g
o
f
s
ig
n
if
ica
n
t
tim
e‐
r
elate
d
d
ata.
Su
ch
co
n
f
ig
u
r
atio
n
s
h
a
v
e
b
ee
n
ex
t
r
em
ely
ef
f
icien
t
f
o
r
task
s
wh
ich
ca
n
b
e
r
e
q
u
ir
e
d
co
m
p
r
eh
e
n
s
io
n
o
f
h
o
w
d
ata
is
d
ev
elo
p
e
d
with
r
esp
ec
t to
tim
e
,
o
f
ten
e
x
ce
ed
in
g
th
e
o
u
tco
m
e
s
o
f
tr
ad
itio
n
al
AE
s
in
th
ese
d
o
m
ain
s
.
T
h
e
L
STM
n
etwo
r
k
s
ar
e
d
et
er
m
in
ed
to
b
e
a
n
o
r
m
al
class
o
f
DL
tech
n
iq
u
es
th
at
wer
e
esp
ec
ially
d
ev
elo
p
e
d
f
o
r
allev
iatin
g
th
e
c
o
m
p
lex
p
r
o
b
lem
o
f
v
an
is
h
in
g
g
r
ad
ien
t,
a
m
ain
task
ca
m
e
ac
r
o
s
s
s
tan
d
ar
d
R
NN
.
T
h
is
m
em
o
r
y
ce
ll
is
p
er
f
o
r
m
ed
u
n
d
er
3
m
ajo
r
g
ates.
T
h
e
f
ir
s
t
o
n
e
is
th
e
i
n
p
u
t
g
at
e
th
at
wo
r
k
s
o
n
t
h
e
in
teg
r
atio
n
o
f
n
ew
d
ata.
T
h
e
s
ec
o
n
d
o
n
e
is
th
e
f
o
r
g
et
g
ate
th
at
p
o
s
s
ess
es
th
e
ac
co
u
n
ta
b
ilit
y
to
f
in
d
wh
at
t
o
co
n
tin
u
e
o
r
r
e
m
o
v
e.
Fin
ally
,
t
h
e
th
ir
d
o
n
e
is
o
u
tp
u
t
g
ate
th
a
t
ca
n
co
n
t
r
o
l
th
e
t
r
an
s
m
itted
d
ata
to
th
e
o
u
tco
m
e
o
v
er
th
e
n
etwo
r
k
.
T
h
e
m
ath
e
m
atica
l
f
o
r
m
is
r
ep
r
esen
ted
f
o
r
all
g
ates
as
d
en
o
ted
in
(
1
2
)
t
o
(
1
7
)
.
T
o
g
ain
th
e
in
p
u
t o
r
d
er
=
(
1
,
2
,
,
)
,
th
e
L
STM
ce
ll e
x
ec
u
tes th
e
f
o
llo
win
g
p
r
o
g
r
ess
io
n
at
ea
ch
tim
e
s
tep
z
,
=
(
⋅
[
ℎ
−
1
,
]
+
)
(
1
2
)
̃
=
ℎ
(
⋅
[
ℎ
−
1
,
]
+
)
(
1
3
)
=
⊙
−
1
+
⊙
̃
(
1
4
)
=
(
⋅
[
ℎ
−
1
,
]
+
)
(
1
5
)
=
(
⋅
[
ℎ
−
1
,
]
+
)
(
1
6
)
ℎ
=
⊙
ℎ
(
)
(
1
7
)
Her
e
th
e
p
ar
a
m
eter
d
ef
in
es
t
h
e
o
u
tp
u
t
g
ate,
in
d
icate
s
th
e
in
p
u
t
g
ate,
s
p
ec
if
ies
th
e
f
o
r
g
et
g
ate.
T
h
e
ter
m
̃
th
e
ca
n
d
id
ate
m
em
o
r
y
c
ell
v
ec
to
r
is
s
p
ec
if
ied
.
d
en
o
t
es
th
e
v
ec
to
r
o
f
ce
ll
la
y
er
.
ℎ
r
e
p
r
esen
ts
th
e
v
ec
to
r
o
f
h
id
d
e
n
lay
er
(
HL
)
.
T
h
e
s
y
m
b
o
l
σ
s
ig
n
if
ies
th
e
ac
tiv
atio
n
s
ig
m
o
id
f
u
n
ctio
n
.
T
h
e
s
y
m
b
o
l
⊙
r
ep
r
esen
ts
elem
en
t‐
wis
e
m
u
lti
p
licatio
n
.
T
h
e
p
a
r
am
eter
,
,
,
d
e
n
o
te
th
e
weig
h
t
m
atr
ices.
T
h
e
ter
m
s
,
,
,
an
d
0
r
em
ain
s
b
iased
v
ec
t
o
r
s
.
T
h
e
L
STM
‐
AE
is
g
en
er
ally
u
s
ef
u
l
an
d
ca
n
id
en
tify
lo
n
g
er
‐
r
an
g
e
p
atter
n
s
with
r
esp
ec
t
to
tim
e‐
s
eq
u
en
ce
s
d
ata.
I
t
ca
n
en
h
a
n
c
e
d
ep
en
d
a
b
ilit
y
an
d
wo
r
k
i
n
g
ef
f
icien
cy
.
T
h
is
ca
n
b
e
in
cl
u
d
ed
in
d
u
al
u
n
its
:
an
en
co
d
ed
tr
an
s
f
o
r
m
s
th
e
in
p
u
t
o
r
d
er
s
in
t
o
ce
r
tain
s
ize
v
ec
to
r
s
,
an
d
th
e
d
ec
o
d
ed
th
at
r
ef
o
r
m
s
th
e
o
r
d
e
r
f
r
o
m
th
o
s
e
v
ec
to
r
s
.
T
h
e
ar
ch
itectu
r
e
o
f
th
e
L
STM
‐
AE
is
ex
p
lain
ed
h
o
w
th
ese
m
o
d
els
ar
e
p
er
f
o
r
m
ed
f
o
r
p
r
o
v
id
in
g
s
tr
o
n
g
ef
f
ec
tiv
e
n
ess
co
r
r
esp
o
n
d
in
g
with
tim
e‐
s
er
ies.
3
.
4
.
F
ine
-
t
un
ed
m
o
del v
i
a
T
SO
Fin
ally
,
th
e
T
SO
alg
o
r
ith
m
ca
n
b
e
ex
ec
u
ted
f
o
r
b
etter
p
ar
am
eter
tu
n
in
g
to
im
p
r
o
v
e
co
n
v
er
g
en
c
e
s
p
ee
d
an
d
p
r
e
d
ictiv
e
r
esu
lts
.
T
u
n
as
ar
e
m
ar
in
e
p
r
ed
ato
r
f
is
h
th
at
ea
t
eith
er
s
u
r
f
ac
e
o
r
m
id
wate
r
f
is
h
.
T
h
e
ab
ilit
y
o
f
th
e
tu
n
as
to
co
n
s
tan
tly
s
wim
is
eith
er
ef
f
ec
tiv
e
o
r
ex
ce
p
tio
n
al.
T
h
ey
u
tili
ze
th
e
g
r
o
u
p
tr
av
el
tactic
f
o
r
p
r
e
d
ato
r
y
b
eh
a
v
io
r
.
T
o
f
in
d
an
d
h
u
n
t
th
eir
p
r
ey
,
th
ey
u
tili
ze
its
in
tellig
en
ce
[
1
8
]
.
I
n
a
d
d
itio
n
,
tu
n
as
h
av
e
d
ev
elo
p
e
d
a
ty
p
e
o
f
in
tellectu
al
an
d
ef
f
icien
t
h
u
n
tin
g
tactics,
co
m
p
r
is
in
g
s
p
ir
al
an
d
p
a
r
ab
o
lic
h
u
n
tin
g
.
T
o
r
elo
ca
te
th
eir
tar
g
et
ac
r
o
s
s
th
e
s
h
allo
w
wate
r
an
d
attac
k
with
o
u
t
ef
f
o
r
t,
th
e
y
u
tili
ze
th
e
s
p
ir
al
p
atter
n
th
ey
g
en
er
ate
wh
ile
s
wim
m
in
g
wit
h
in
th
eir
h
u
n
tin
g
s
tr
ateg
y
.
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
Temp
o
r
a
l d
ee
p
r
ep
r
esen
ta
tio
n
lea
r
n
in
g
b
a
s
ed
r
ema
in
in
g
u
s
efu
l Lifecy
le
…
(
P
r
a
d
is
h
V
a
id
ya
S
.
)
2371
3
.
4
.
1
.
I
nitia
liza
t
io
n
T
h
e
p
r
im
a
r
y
p
o
p
u
latio
n
s
in
t
h
e
s
ea
r
ch
in
g
r
eg
io
n
ar
e
r
an
d
o
m
ly
g
en
er
ate
d
b
y
th
e
T
SO
m
o
d
el
to
b
eg
i
n
th
e
o
p
tim
izer
p
r
o
ce
d
u
r
e.
=
ℜ
⋅
(
−
)
+
,
=
1
,
2
,
…
,
(
1
8
)
T
h
e
s
ea
r
ch
in
g
r
eg
io
n
s
’
m
ax
im
u
m
an
d
m
in
im
u
m
ar
e
in
d
ic
ated
b
y
an
d
,
th
e
ar
b
itra
r
y
v
ec
to
r
in
th
e
in
ter
v
al
o
f
ze
r
o
an
d
o
n
e
is
s
tated
as
ℜ
,
th
e
en
tire
tu
n
a
p
o
p
u
latio
n
s
ca
n
p
o
r
tr
ay
ed
as
,
an
d
th
e
ℎ
T
h
e
f
ir
s
t in
d
iv
id
u
al
is
s
y
m
b
o
li
ze
d
b
y
.
2
.
4
.
2
.
Sp
ira
l
f
o
ra
g
ing
W
h
ile
th
e
s
m
allest
s
ch
o
o
lin
g
f
is
h
m
ee
t
p
r
e
d
ato
r
s
,
th
e
y
all
g
en
er
ate
a
d
e
n
s
e
f
o
r
m
ati
o
n
,
wh
ich
co
n
tin
u
ally
alter
s
th
eir
s
wim
m
in
g
r
o
u
te.
T
h
er
e
f
o
r
e,
t
h
e
p
r
e
d
a
to
r
s
f
in
d
it c
h
allen
g
i
n
g
to
s
u
r
r
o
u
n
d
t
h
e
tar
g
et.
T
o
h
u
n
t
th
eir
v
ictim
,
th
e
tu
n
a
tr
o
o
p
g
en
e
r
ates
clo
s
ed
s
p
ir
al
s
tr
u
ctu
r
es.
L
ik
e
th
e
g
ath
er
in
g
o
f
f
is
h
th
at
s
wim
d
y
n
am
ically
alo
n
g
t
h
e
s
elec
ted
p
ath
,
n
ea
r
b
y
f
is
h
ch
a
n
g
e
t
h
eir
r
o
u
te
o
n
e
b
y
o
n
e,
cr
ea
te
a
lar
g
er
g
r
o
u
p
with
a
co
m
m
o
n
o
b
jectiv
e,
an
d
b
eg
in
f
o
r
ag
in
g
.
E
ac
h
tu
n
a
f
o
llo
ws
th
e
p
r
io
r
f
is
h
,
an
d
th
e
s
ch
o
o
ls
o
f
tu
n
a
in
ter
co
n
n
ec
t
with
o
n
e
an
o
th
er
.
+
1
=
{
1
⋅
(
+
⋅
|
−
|
)
+
2
⋅
,
=
1
1
⋅
(
+
⋅
|
−
|
)
+
2
⋅
−
1
,
=
2
,
3
,
…
,
(
1
9
)
1
=
+
(
1
−
)
⋅
m
ax
(
2
0
)
2
=
(
1
−
)
−
(
1
−
)
⋅
m
ax
(
2
1
)
=
e
xp
(
)
⋅
c
os
(
2
)
(
2
2
)
=
e
xp
(
3
c
os
(
(
(
m
ax
+
1
)
−
1
)
)
)
(
2
3
)
W
h
er
ea
s
m
ax
ch
ar
ac
te
r
izes
th
e
it
er
atio
n
b
o
u
n
d
ar
y
,
th
e
p
r
o
v
i
d
e
d
co
n
s
tan
t
ca
n
b
e
s
p
ec
if
ied
b
y
,
th
e
ℎ
an
in
d
iv
id
u
al
f
o
r
iter
atio
n
+
1
is
r
ep
r
esen
ted
b
y
+
1
,
th
e
weig
h
t
co
ef
f
icien
ts
ar
e
s
p
ec
if
ied
b
y
1
an
d
2
,
th
e
n
u
m
b
er
s
elec
ted
at
r
a
n
d
o
m
a
m
o
n
g
ze
r
o
a
n
d
o
n
e
is
s
tated
as
,
th
e
p
r
esen
t
to
p
i
n
d
iv
id
u
al
is
r
ep
r
esen
ted
b
y
,
an
d
th
e
p
r
esen
t
iter
atio
n
is
c
h
ar
ac
ter
ized
b
y
.
E
v
er
y
tu
n
a
ca
n
u
s
e
th
e
ar
ea
n
ea
r
its
p
r
ey
.
I
n
ad
d
itio
n
,
th
e
s
ea
r
ch
in
g
r
eg
io
n
cr
ea
tes
s
to
ch
asti
c
co
o
r
d
i
n
ates
f
o
r
s
p
ir
al
h
u
n
tin
g
th
at
aid
all
in
d
iv
id
u
al
s
in
lo
o
k
in
g
f
o
r
th
e
lar
g
est r
eg
io
n
a
n
d
g
i
v
e
th
e
T
S
O
th
e
ca
p
ab
ilit
y
o
f
d
is
co
v
er
in
g
th
e
co
m
p
lete
wo
r
ld
.
+
1
=
{
1
⋅
(
ℜ
+
⋅
|
ℜ
−
|
)
+
2
⋅
,
=
1
1
⋅
(
ℜ
+
⋅
|
ℜ
−
|
)
+
2
⋅
−
1
,
=
2
,
3
,
…
,
(
2
4
)
W
h
er
ea
s
ℜ
d
is
p
lay
s
th
e
r
an
d
o
m
ly
estab
lis
h
ed
r
ef
er
e
n
ce
p
o
i
n
t
with
in
th
e
p
o
s
s
ib
le
ar
ea
.
T
h
e
last
d
ev
elo
p
e
d
f
o
r
m
u
latio
n
f
o
r
th
e
s
p
ir
al
s
ea
r
ch
in
g
tactic
is
f
o
r
m
u
lated
as:
+
1
=
{
{
1
.
(
ℜ
+
∙
|
ℜ
+
|
)
+
2
∙
,
=
1
1
.
(
ℜ
+
∙
|
ℜ
+
|
)
+
2
∙
−
1
,
=
2
,
3
,
…
,
ℜ
<
m
ax
{
1
.
(
+
∙
|
+
|
)
+
2
∙
,
=
1
1
.
(
+
∙
|
+
|
)
+
2
∙
−
1
,
=
2
,
3
,
…
,
.
,
ℜ
<
m
ax
(
2
5
)
2
.
4
.
3
.
P
a
r
a
bo
lic
f
o
ra
g
ing
T
h
ey
im
p
r
o
v
e
p
a
r
ab
o
lic
f
o
r
m
atio
n
f
r
o
m
th
e
s
elec
ted
p
o
in
t,
i.e
.
,
f
o
o
d
.
Ad
d
itio
n
ally
,
th
ey
d
is
co
v
er
th
eir
en
v
ir
o
n
m
en
ts
.
As
a
r
esu
l
t,
eith
er
o
f
t
h
ese
ap
p
r
o
ac
h
es
i
s
u
tili
ze
d
s
im
u
ltan
eo
u
s
ly
,
an
d
a
5
0
%
p
r
o
b
ab
ilit
y
s
elec
tio
n
is
u
s
ed
f
o
r
eith
er
m
eth
o
d
.
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.
16
,
No
.
5
,
Octo
b
e
r
20
26
:
2
3
6
5
-
2
3
7
8
2372
+
1
=
{
+
ℜ
⋅
(
−
)
+
⋅
2
⋅
(
−
)
,
ℜ
<
0
.
5
⋅
2
⋅
,
ℜ
≥
0
.
5
(
2
6
)
=
(
1
−
m
ax
)
(
m
ax
)
(
2
7
)
An
ar
b
itra
r
y
n
u
m
b
er
i
n
th
is
s
am
p
le
d
r
o
p
s
i
n
th
e
in
ter
v
al
[
1
,
1
]
s
p
ec
if
ied
b
y
.
T
ill
th
e
e
n
d
co
n
d
itio
n
is
f
u
lf
illed
,
ev
e
r
y
T
SO
will
b
e
u
p
g
r
ad
ed
an
d
c
o
n
s
tan
tly
ca
lcu
lated
in
th
e
p
r
o
c
ed
u
r
e
o
f
c
o
m
p
lete
o
p
tim
izatio
n
.
Fit
n
ess
ch
o
ice
i
s
a
m
ajo
r
f
ea
tu
r
e
p
r
o
m
p
tin
g
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
T
SO
m
eth
o
d
o
l
o
g
y
.
T
h
e
p
ar
am
eter
s
elec
tio
n
m
eth
o
d
co
m
p
r
is
es
th
e
s
o
lu
tio
n
en
c
o
d
in
g
m
et
h
o
d
o
lo
g
y
to
e
v
alu
a
te
th
e
ef
f
icac
y
o
f
ca
n
d
id
ate
p
e
r
f
o
r
m
an
ce
s
.
T
h
e
T
SO
r
ef
lects
p
r
ec
is
io
n
as
t
h
e
m
ajo
r
s
tan
d
ar
d
to
s
tr
ateg
y
th
e
FF
,
wh
ich
is
r
ep
r
esen
ted
as e
x
p
o
s
ed
.
=
ma
x
(
)
(
2
8
)
=
+
(
2
9
)
No
w,
an
d
ep
ito
m
ize
th
e
tr
u
e
an
d
th
e
f
alse p
o
s
itiv
e
v
alu
es.
4.
P
E
RF
O
RM
A
NCE V
AL
I
DA
T
I
O
N
I
n
th
is
p
ar
t,
th
e
ex
p
er
im
en
tal
r
esu
lt
o
f
th
e
T
R
DL
-
R
UL
P
m
eth
o
d
co
u
ld
b
e
in
s
p
ec
ted
e
m
p
lo
y
in
g
th
e
lith
iu
m
-
io
n
b
atter
y
d
eg
r
ad
atio
n
d
ataset
[
1
9
]
.
T
h
is
d
atab
ase
is
m
o
d
elled
af
ter
th
e
NASA
Am
es
Pro
g
n
o
s
tics
C
en
ter
o
f
E
x
ce
llen
ce
lith
iu
m
-
io
n
b
atter
y
d
e
g
r
ad
e
d
d
atab
ase
s
.
I
t
s
u
g
g
ests
th
e
ch
ar
g
e
-
d
is
ch
ar
g
e
b
e
h
av
io
r
a
n
d
ag
in
g
p
r
o
ce
d
u
r
e
o
f
lith
iu
m
-
io
n
b
atter
ies
th
r
o
u
g
h
m
an
y
c
y
cl
es,
tak
in
g
r
ea
l
ten
d
en
c
y
in
b
at
ter
y
h
ea
lth
in
tim
e.
T
h
e
d
a
t
a
b
a
s
e
i
n
v
o
l
v
e
s
1
1
a
t
t
r
ib
u
t
e
s
:
B
C
t
,
a
n
d
R
U
L
,
S
OH
,
c
h
I
,
d
i
s
I
,
d
i
s
V
,
c
h
V
,
d
i
s
T
,
c
h
,
a
n
d
b
a
t
t
e
r
y
_
i
d
,
c
y
c
l
e
.
F
o
r
a
n
a
l
y
s
is
p
u
r
p
o
s
es
,
1
0
a
t
t
r
i
b
u
t
e
s
a
r
e
c
h
o
s
e
n
:
B
C
t
,
SO
H
,
R
U
L
,
d
i
s
I
,
c
h
I
,
d
is
V
,
c
h
V
,
d
i
s
T
,
c
h
T
,
a
n
d
c
y
c
l
e
.
Fig
u
r
e
2
s
h
o
ws
th
e
co
r
r
elatio
n
m
atr
ix
o
f
th
e
T
R
DL
-
R
UL
P
m
eth
o
d
b
etwe
en
b
atter
y
attr
ib
u
tes,
wh
er
e
v
alu
es n
ea
r
1
an
d
-
1
s
p
ec
if
y
s
tr
o
n
g
er
p
o
s
itiv
e
an
d
n
e
g
ativ
e
c
o
r
r
elatio
n
s
.
I
t is n
o
ted
th
at
SOH
an
d
B
C
t r
ev
ea
l a
r
o
b
u
s
t
p
o
s
itiv
e
co
r
r
elatio
n
wit
h
R
UL
,
em
p
h
asizin
g
th
eir
s
ig
n
if
ican
ce
in
p
r
ed
ictin
g
b
atter
y
life
.
Ho
wev
er
,
t
h
e
cy
cle
s
h
o
ws
a
r
o
b
u
s
t
n
eg
ativ
e
co
r
r
elatio
n
with
R
UL
,
im
p
ly
i
n
g
th
at
b
atter
y
life
r
e
d
u
ce
s
with
g
r
o
win
g
c
y
cles.
I
n
ad
d
itio
n
,
d
is
I
,
d
is
V,
d
is
T
,
ch
l,
ch
V,
an
d
ch
T
d
is
p
lay
wea
k
er
co
r
r
elatio
n
s
,
s
ig
n
if
y
i
n
g
th
eir
r
estricte
d
in
f
lu
en
ce
o
n
R
UL
ap
p
r
o
x
im
at
io
n
.
Fig
u
r
e
2
.
C
o
r
r
ela
ti
o
n
m
at
r
i
x
o
f
t
h
e
T
R
D
L
-
R
U
L
P
m
et
h
o
d
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
Temp
o
r
a
l d
ee
p
r
ep
r
esen
ta
tio
n
lea
r
n
in
g
b
a
s
ed
r
ema
in
in
g
u
s
efu
l Lifecy
le
…
(
P
r
a
d
is
h
V
a
id
ya
S
.
)
2373
Fig
u
r
e
3
r
ep
r
esen
ts
th
e
f
ea
tu
r
e
d
is
tr
ib
u
tio
n
o
f
th
e
T
R
DL
-
R
UL
P
m
eth
o
d
b
etwe
en
d
is
I
an
d
R
UL
.
I
t
is
n
o
ted
th
at
th
e
d
ata
p
o
in
ts
ar
e
r
an
d
o
m
l
y
d
is
tr
ib
u
ted
with
o
u
t
an
y
clea
r
g
r
o
win
g
o
r
d
e
clin
in
g
ten
d
en
c
y
.
T
h
is
s
p
ec
if
ies th
at
d
is
I
h
as a
wea
k
er
co
r
r
elatio
n
with
R
UL
an
d
h
a
s
m
in
im
u
m
in
f
lu
e
n
ce
o
n
b
atter
y
life
p
r
e
d
ictio
n
.
Fig
u
r
e
3
.
F
ea
tu
r
e
d
is
t
r
i
b
u
ti
o
n
wit
h
d
is
I
a
n
d
R
UL
Fig
u
r
e
4
d
e
p
icts
th
e
p
r
ed
ictio
n
p
er
f
o
r
m
an
ce
o
f
th
e
T
R
DL
-
R
UL
P
s
y
s
tem
at
v
ar
io
u
s
ep
o
c
h
s
(
2
0
,
3
0
,
4
0
,
a
n
d
5
0
)
f
o
r
b
o
th
test
in
g
a
n
d
tr
ain
i
n
g
d
atasets
.
I
t
is
m
o
n
ito
r
ed
t
h
at
at
less
er
ep
o
ch
s
,
th
er
e
is
a
lo
n
g
g
ap
b
etwe
en
th
e
ac
tu
al
an
d
p
r
ed
ic
ted
v
alu
es,
r
ep
r
esen
tin
g
m
in
i
m
al
p
r
ed
ictio
n
ac
cu
r
ac
y
.
As
th
e
ep
o
ch
s
r
is
e,
th
e
p
r
ed
icted
o
u
tc
o
m
es
ar
e
clo
s
er
to
ac
tu
al
o
u
tco
m
es,
ex
h
ib
it
in
g
en
h
an
ce
d
lear
n
in
g
ab
ilit
y
.
At
ep
o
ch
5
0
,
th
e
tech
n
iq
u
e
attain
s
b
est alig
n
m
e
n
t a
n
d
s
tead
in
ess
,
in
d
icatin
g
e
f
f
icien
t a
n
d
p
r
ec
is
e
R
UL
p
r
ed
i
ctio
n
.
Fig
u
r
e
4
.
P
r
ed
ict
io
n
r
esu
lt
o
f
t
h
e
T
R
D
L
-
R
UL
P
a
p
p
r
o
a
c
h
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.
16
,
No
.
5
,
Octo
b
e
r
20
26
:
2
3
6
5
-
2
3
7
8
2374
T
ab
le
1
p
o
r
tr
ay
s
th
e
R
UL
p
r
ed
ictio
n
r
esu
lt
o
f
th
e
T
R
DL
-
R
UL
P
ap
p
r
o
ac
h
with
s
ev
er
al
m
etr
ics
u
s
in
g
tr
ain
in
g
an
d
test
in
g
s
ets.
T
h
e
o
u
tco
m
e
in
d
icate
s
th
at
th
e
T
R
DL
-
R
UL
P
ap
p
r
o
ac
h
ex
a
ctly
id
en
tifie
d
th
e
s
am
p
les.
Un
d
er
tr
ain
in
g
s
et,
th
e
p
r
o
p
o
s
ed
T
R
DL
-
R
UL
P
t
ec
h
n
iq
u
e
p
r
o
v
id
es
r
esu
lts
wit
h
MSE
o
f
0
.
0
2
4
2
,
R
MSE
o
f
0
.
1
5
5
7
,
MA
E
o
f
0
.
1
3
1
9
,
MA
PE
o
f
7
.
3
1
9
1
,
a
n
d
R
2
Sco
r
e
o
f
0
.
6
5
6
2
,
co
r
r
esp
o
n
d
in
g
l
y
.
I
n
t
h
e
m
ea
n
tim
e,
u
n
d
e
r
test
in
g
s
et,
t
h
e
p
r
o
p
o
s
ed
T
R
DL
-
R
UL
P
m
o
d
el
g
iv
es
o
u
tco
m
es
with
MSE
o
f
0
.
0
2
5
7
,
R
MSE
o
f
0
.
1
6
0
3
,
MA
E
o
f
0
.
1
3
6
3
,
M
APE
o
f
1
.
8
4
9
6
,
an
d
R
2
Sco
r
e
o
f
0
.
6
5
0
3
,
co
r
r
esp
o
n
d
in
g
ly
.
T
a
b
le
1
.
R
UL
p
r
e
d
ic
ti
o
n
o
u
tc
o
m
es
o
f
t
h
e
T
R
D
L
-
R
U
L
P
m
o
d
e
l w
i
th
m
et
r
i
cs
M
e
t
r
i
c
s
Tr
a
i
n
i
n
g
S
e
t
T
e
s
t
i
n
g
S
e
t
M
S
E
0
.
0
2
4
2
0
.
0
2
5
7
R
M
S
E
0
.
1
5
5
7
0
.
1
6
0
3
M
A
E
0
.
1
3
1
9
0
.
1
3
6
3
M
A
P
E
7
.
3
1
9
1
1
.
8
4
9
6
R
2
S
c
o
r
e
0
.
6
5
6
2
0
.
6
5
0
3
Fig
u
r
e
5
illu
s
tr
ates
th
e
lo
s
s
cu
r
v
es
o
f
th
e
T
R
DL
-
R
UL
P
m
o
d
el
u
s
in
g
d
i
f
f
er
en
t
p
er
f
o
r
m
a
n
ce
m
etr
ics
lik
e
MSE
,
R
MSE
,
MA
E
,
MA
PE,
an
d
R
²
s
co
r
e
f
o
r
b
o
th
tr
ain
in
g
a
n
d
test
in
g
d
atasets
.
I
t
is
o
b
s
er
v
e
d
th
at
th
e
MSE
,
R
MSE
,
an
d
MA
E
v
al
u
es
d
ec
r
ea
s
e
s
tead
ily
with
i
n
cr
ea
s
in
g
ep
o
ch
s
,
i
n
d
icatin
g
i
m
p
r
o
v
e
d
p
r
ed
ictio
n
ac
cu
r
ac
y
an
d
ef
f
ec
tiv
e
lear
n
i
n
g
.
T
h
e
R
²
s
co
r
e
in
cr
ea
s
es
g
r
ad
u
ally
,
s
h
o
win
g
en
h
an
ce
d
m
o
d
el
p
er
f
o
r
m
a
n
ce
an
d
b
etter
f
it
to
th
e
d
ata.
Ad
d
itio
n
ally
,
th
e
tr
ain
in
g
an
d
test
in
g
cu
r
v
es
f
o
llo
w
a
s
im
ilar
tr
en
d
,
in
d
icatin
g
g
o
o
d
g
en
er
aliza
tio
n
with
o
u
t sig
n
if
ic
an
t o
v
er
f
itti
n
g
.
Fig
u
r
e
5
.
L
o
s
s
cu
r
v
e
o
f
t
h
e
T
R
DL
-
R
U
L
P
m
et
h
o
d
wit
h
v
a
r
i
o
u
s
m
et
r
ics
T
h
e
co
m
p
ar
ativ
e
i
n
v
esti
g
atio
n
o
f
th
e
T
R
DL
-
R
UL
P
m
eth
o
d
with
o
th
er
ex
is
tin
g
ap
p
r
o
ac
h
es
is
p
r
o
v
id
e
d
in
T
ab
le
2
a
n
d
Fig
u
r
e
6
[
2
0
]
,
[
2
1
]
.
T
h
e
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
im
p
lied
th
at
th
e
T
R
DL
-
R
UL
P
s
y
s
tem
h
as
r
ev
ea
led
th
e
b
est
o
u
tco
m
es.
T
h
e
p
r
o
p
o
s
ed
T
R
DL
-
R
UL
P
m
o
d
el
h
as
r
ea
ch
ed
t
h
e
lo
west
o
u
tco
m
es
with
MSE
o
f
0
.
0
2
4
,
R
MSE
o
f
0
.
1
5
6
,
MA
E
o
f
0
.
1
3
2
,
an
d
R
2
Sco
r
e
o
f
0
.
6
5
6
,
wh
er
ea
s
t
h
e
SVM
m
eth
o
d
h
as
s
lig
h
tly
h
ig
h
er
r
esu
lts
with
MSE
o
f
0
.
1
9
2
,
R
MSE
o
f
2
.
1
3
0
,
MA
E
o
f
1
.
8
3
0
,
an
d
R
2
Sco
r
e
o
f
2
.
7
8
4
.
I
n
ad
d
itio
n
,
NN,
SVR
,
L
STM
-
DSSN,
F
C
NN,
an
d
L
STM
m
eth
o
d
o
lo
g
ies
h
av
e
ac
h
ie
v
ed
m
a
x
im
al
p
er
f
o
r
m
an
ce
.
Evaluation Warning : The document was created with Spire.PDF for Python.